Home

InfoPACK User Guide. Version 2.2

Contents

1. InfoPACK User Guide Version 2 2 NASoftware Limited 1 Prospect Road Birkenhead Merseyside CH42 8LE United Kingdom Tel 44 0 151 609 1911 29 Mar 10 NASoftware Ltd Contents 11 Introduction 2 1 1 Installing and Running InfoPACK o o ee eee 3 12 The Lene c ooste a a a o a ee ee a ane d 4 L3 Howto ContactUs ce be a os OO Se he sacas a a SS 5 14 ACKHOWIEUSENIGME o ooo ee os o ele ee ee Oe oe a ie a 6 Synthetic Aperture Radar Images 7 2 1 Interpretation of SAR Images o o e ee a 7 2 2 The Formation of SAR Images oc cs a ceca dada ed ee eee 8 2 3 Prequency Bands 2 22 ise hdav eee vy bd pede ee ba eee Me bed 9 2 POSO o a ere wee aoe a iaa om Wb ern a oe ee a i e td ae ts A 9 2 5 The Number of Looks in an Image aaa o e 10 2 6 The Number of Looks and InfoPACK o e e eee 11 2 2 Multt dimenstonal Images occiso seat ea ae ae ee 11 28 Correlated Images ee ee aa Ga a Oe a a 12 2 9 Reducins the Image Sue o cada da eae ba ee BEERS GS 12 2 10 Sources Of SAR Data o o e o a ee ee RA ee 12 Image File Formats 15 3 1 The NASoftware format oe secu ee ee a a 15 3 2 Built In File Conversion Facilities 2 0 0 ee ee ee 15 3 2 1 File Import Conversion to NetCDF Format 15 322 PUG EXPO or eck anios ee eR ie ae bE oR A ke we a 17 So General Elle Foriats 226 6 ea ek owes Re EE a a 17 SA Ussa aua ME Ac ace hae Aa aa eek ee es 17
2. 158 NASoftware Ltd 29 Mar 10 B 2 IMAGE MANIPULATION ROUTINES Parameters mtpol integer default 0 Polarimetric data is stored as layers of complex data however this is the same file format as multi temporal polarimetric data To distinguish the two the mt pol parameter is used For instance 5 days of 3 channel polarimetric data would be stored as a 15 layer complex image and the mtpol 3 parameter would be used The default value of mt po1 0 uses all of the layers as a single polarimetric image tilesize integer default 512 A tilesize of 512 x 512 is used tmpdir directory default None Intermediate results are written to a system temporary directory B 2 19 polyscan Purpose Draws a closed polygon on an image Command Line Script polyscan py image_in image out polygon value tilesize tmpdir Python Procedure array_out polyscan array_in polygon value Parameters polygon list This is a list of x y pixel coordinates of the vertices For example polygon 10 10 100 10 10 100 Note no spaces are allowed around the commas polygon float Default O The enclosed area of the polygon is filled with this value tilesize integer default 512 A tilesize of 512 x 512 is used tmpdir directory default None Intermediate results are written to a system temporary directory B 2 20 resize Purpose Resize an existing image The resulting image is resized to the new width and height using a
3. Parameters tilesize integer default 512 A tilesize of 512 x 512 is used tmpdir directory default None Intermediate results are written to a system temporary directory B 2 22 scalar divide Purpose The result is the individual pixel values from the input image divided by a specified constant Command Line Script scalar_divide py image_in image_out scalar tilesize tmpdir Python Procedure array_out scalar_divide array_in scalar Parameters tilesize integer default 512 A tilesize of 512 x 512 is used tmpdir directory default None Intermediate results are written to a system temporary directory B 2 23 scalar_multiply Purpose The result is the individual pixel values from the input image multiplied by a specified constant Command Line Script scalarmultiply py image_in image_out scalar tilesize tmpdir 29 Mar 10 NASoftware Ltd 161 APPENDIX B LOW LEVEL AND UTILITY ROUTINE DOCUMENTATION Python Procedure array_out scalar multiply array in scalar Parameters tilesize integer default 512 A tilesize of 512 x 512 is used tmpdir directory default None Intermediate results are written to a system temporary directory B 2 24 scanedge Purpose Scanning edge detector This edge detector works by calculating the significance of every possible edge position within a moving window of size width by height Since the result is determined by the accuracy of the stati
4. 0 2712 0 1809 0 1447 0 0956 2 0 0203 0 1898 0 2280 0 2234 0 1338 0 1211 3 0 0020 0 0782 0 1525 0 1659 0 1646 0 1436 4 0 0 0 0248 0 0866 0 1300 0 1555 0 1258 5 0 0 0 0105 0 0919 0 2266 0 3653 0 4815 Table 6 1 2 b Confusion Matrix for Classification using a 3 by 3 Filter 58 NASoftware Ltd 29 Mar 10 6 1 SEGMENTATION AND CLASSIFICATION OF INTENSITY IMAGES Figure 6 1 2 b Classification using a 3 by 3 filter Figure 6 1 2 c Classification using segmentation 29 Mar 10 NASoftware Ltd 59 CHAPTER 6 CASE STUDIES Figure 6 1 2 d Classification from the Ground Truth Assigned class 0 1 2 3 4 5 True class Prob 0 8196 0 0870 0 0059 0 0 0 0 0 0 0 1236 0 7941 0 2886 0 0614 0 0457 0 0032 0 0466 0 1177 0 5413 0 32440 0 2057 0 0442 0 0095 0 0012 0 1514 0 4286 0 2815 0 1627 0 0007 0 0 0 0128 0 1629 0 2998 0 2491 0 0 0 0 0 0 0 0227 0 2011 0 5408 BY ol DP Re oO Table 6 1 2 c Confusion Matrix for Classification using Segmentation Assigned class 0 1 2 3 4 5 True class Prob 0 9243 0 0472 0 0329 0 0222 0 0307 0 0099 0 0277 0 8908 0 1205 0 0258 0 0579 0 0099 0 0345 0 0501 0 7894 0 0490 0 0 0 0385 0 0122 0 0010 0 0495 0 8928 0 0 0 0005 0 0014 0 0003 0 0033 0 0039 0 9096 0 0 0 0 0 0005 0 0042 0 0052 0 0018
5. None provided Parameters nbins integer default 100 The number of data bins used to store the results layer integer default None If specified select this layer from the input image sd float default 2 The number of bins is spread over the interval u sd x o u sd x o where u and o are the mean and standard deviation of the data respectively Any data values falling outside this range are placed into the bin with either the lowest or highest data values as appropriate min float default None If specified this sets the minimum histogram bin max float default None If specified this sets the maximum histogram bin norm float default None If this parameter is specified normalize the area under the histogram to this value cmpl2real string default abs Chooses from a set of methods for converting the complex image to a real one abs absolute value mag magnitude of complex number abs x abs phase Phase of complex number real Real part imag Imaginary part B 4 4 histogram Purpose Provides a graphical display of the data distribution of a given image file For multi dimensional images the histograms for individual layers can be viewed separately Command Line Script histogram py image_in nbins sdrange norm Python Procedure None provided 29 Mar 10 NASoftware Ltd 181 APPENDIX B LOW LEVEL AND UTILITY ROUTINE DOCUMENTATION Parameters nbins
6. Numeric log10 array_in Parameters tilesize integer default 512 A tilesize of 512 x 512 is used tmpdir directory default None Intermediate results are written to a system temporary directory B 2 14 loge Purpose The result is the natural logarithm of each individual pixel value from the input image Command Line Script loge py imagel_in image_out tilesize tmpdir Python Procedure array_out Numeric loge array in Parameters tilesize integer default 512 A tilesize of 512 x 512 is used tmpdir directory default None Intermediate results are written to a system temporary directory 156 NASoftware Ltd 29 Mar 10 B 2 IMAGE MANIPULATION ROUTINES B 2 15 median Purpose Median filter For each pixel the median value in the surrounding window is returned Command Line Script median py image in image_out width height tilesize overlap tmpdir Python Procedure array_out filter array_in method med width height Parameters width integer default 3 The width of the filter window height integer default 3 The height of the filter window tilesize integer default 512 A tilesize of 512 x 512 is used overlap integer default 48 To prevent obvious joins between tiles an overlap of 48 is used tmpdir directory default None Intermediate results are written to a system temporary directory B 2 16 multiply Purpose Multiplies two image files The re
7. b Result of change_detection_position between Figures a and b c Result of change_detection_position between Figures b and c d Result of change_detection_position between Figures c and d 29 Mar 10 NASoftware Ltd 33 CHAPTER 4 IMAGE PROCESSING WITH InfoPACK 4 4 Detecting Edges Detecting edges and region boundaries is a fundamental requirement for automated image interpre tation A segmentation algorithm divides up an image into regions An edge detector finds lines which separate areas with different characteristics these edges do not necessarily join up to form boundaries of closed regions An edge map is a binary image in which each pixel indicates whether or not an edge is present It can be produced directly from the raw data by an edge detection algorithm or usually much more reliably by identifying the region boundaries in a segmentation 4 4 1 The scanedge routine The scanedge function detects edges in a raw SAR image via a combination of a scanning window and a scanning edge The algorithm used has been shown theoretically 2 to give optimal detection of edges in a raw SAR image for the window size specified The sensitivity of the detector can be altered by setting the probability of false alarm pfa the probability of finding edges where none exist Smaller pfas lead to fewer edges being found The scanedge algorithm can be performed upon images with more than one image layer scanedge can also be
8. e The ratio of the original intensity to the despeckled version does not correspond to the ex pected speckle both in apparent ENL and the speckle correlation properties Only weak residual correlations on a scale of one pixel remain e The effects of correlated speckle can be removed by averaging the intensity over an area con sistent with the speckle correlation lengths and then subsampling at this spacing to remove sampling correlations e The correlation lengths can be obtained directly from the quoted instrumental resolution and sample spacing Alternatively it can be derived from the ACF of either the complex SLC or intensity images e This technique offers the means to despeckle intensity data with any speckle correlation prop erties in an optimum fashion optimising both the retention of information by speckle reduc tion through averaging the intensity and correlation removal by resampling 29 Mar 10 NASoftware Ltd 97 CHAPTER 6 CASE STUDIES Figure 6 17 An Example DLR ESAR Image 6 7 An Example Segmenting and Despeckling a Large Image Figure 6 17 shows an original intensity DLR ESAR image after it has been averaged and resampled 3x3 to remove correlations as discussed above We shall first segment this image and then despeckle it In each case we show the residual ratio to indicate how effective the image interpretation process is at removing speckle The quality of reconstruction will be indicated by the absence of
9. seedl seed2 tilesize tmpdir Python Procedure array out fill gammarand width height layers mean order seed1 seed2 29 Mar 10 NASoftware Ltd 151 APPENDIX B LOW LEVEL AND UTILITY ROUTINE DOCUMENTATION Parameters layers integer default 1 mean float default 1 order float default 1 seed1 integer default 0 seed2 integer default 0 tilesize integer default 512 A tilesize of 512 x 512 is used tmpdir directory default None Intermediate results are written to a system temporary directory B 2 8 fill_gaussrand Purpose Creates a real valued image of given dimensions width height and the number of layers filled with random samples These samples have a gaussian distribution with the given mean and standard deviation Seeds seed1 and seed2 may be specified to give a reproducible sequence of random numbers Width and height parameters must be specified Command Line Script fill gaussrand py image out width height layers mean sd seedl seed2 tilesize tmpdir Python Procedure array out fill gaussrand width height layers mean sd seed1 seed2 Parameters layers integer default 1 mean float default 0 sd float default 1 seed1 integer default 0 seed2 integer default 0 1152 NASoftware Ltd 29 Mar 10 B 2 IMAGE MANIPULATION ROUTINES tilesize integer default 512 A tilesize of 512
10. 0 9405 MB WD N el o Table 6 1 2 d Confusion Matrix for Classification using Ground Truth 60 NASoftware Ltd 29 Mar 10 6 1 SEGMENTATION AND CLASSIFICATION OF INTENSITY IMAGES Assigned class 0 1 2 3 4 5 True class Prob 0 0 0 0 0 0 0 0 0 9982 0 0287 0 0 0 0 0 0 0 0 0 0018 0 9713 0 0 9899 0 0002 0 0 0 0 0 0 0 0 1 0 0095 0 9998 0 0 0 0 0 0 0 0 2 0 0007 0 0 1 0000 0 0 0 0 0 3 0 0 0 0 0 0 1 0000 0 0 0 0 4 5 6 1 3 Conclusions It is apparent from this comparison that it is pointless attempting classification on individual inten sity pixels in SAR This example was for a 3 look system with single look data the result would be much worse Averaging over a 3x3 window improves the result The accuracy obtained in this way could be further improved by increasing the size of the window However there is then a problem that small regions will get averaged out and disappear What is required is a method which adapts the region size to the data this is segmentation Segmentation followed by classification yields a considerable improvement since it provides a filter whose window size and shape is adapted to the data The fact that the segments are not matched exactly to the ground truth regions is a consequence of speckle noise In this instance we know that the simulation was of a scene with uniform RCS within each region
11. Add two image files The result is the sum of corresponding pixel values of the input images Command Line Script add py imagel_in image2_in image_out tilesize tmpdir Python Procedure array_out Numeric add arrayl_in array2_in Parameters tilesize integer default 512 A tilesize of 512 x 512 is used tmpdir directory default None Intermediate results are written to a system temporary directory B 2 2 between Purpose Image pixels that lie between the upper and lower limits are set to a specified true value otherwise they are set to a false value Upper and lower limits must be given Command Line Script between py image_in image_out upper lower true false tilesize overlap tmpdir Python Procedure array_out between array in upper lower true false Parameters true float default 1 false float default 0 tilesize integer default 512 A tilesize of 512 x 512 is used 148 NASoftware Ltd 29 Mar 10 B 2 IMAGE MANIPULATION ROUTINES overlap integer default 48 To prevent obvious joins between tiles an overlap of 48 is used tmpdir directory default None Intermediate results are written to a system temporary directory B 2 3 clip Purpose Clips the pixels in an image with given minimum 0 and maximum 1 values Command Line Script clip py image_in image_out min max tilesize tmpdir Python Procedure array_out clip array_in min
12. Parameters image nc This can be either a single or a multiple layered real or complex valued image segment internally converts single layered complex images into single layered intensity images and processes them as with real images A polarimetric image should be given as a layered complex valued image output nc The coloured region map output image is of the same dimensionality as the input im age if the original image was real valued The coloured region map output image is single layered in the case of a complex valued original image 29 Mar 10 NASoftware Ltd 121 APPENDIX A InfoPACK HIGH LEVEL ROUTINE DOCUMENTATION quality Chooses from a predefined set of segmentation parameters quality default sensitive or fast default default These choices set the following parameter values Parameter Fast Default Sensitive Shape 0 000001 0 000001 0 000001 Stopping 0 03 0 01 0 003 Cooling 30 10 1 Regionsize 1 1 1 Looks 1 1 1 Maxiters 30 90 1000 Intensity 0 0 0 Any of these values can be over ridden by subsequent values on the command line The values fast default sensitive are provided as generally convenient settings but because of the iterative nature of the segmentation process and its data dependence some initial parameter searching may be worthwhile for best results shape float default 0 000001 The objective function minimised by segment includes a shape term designed to inh
13. The characteristic SAR speckle spoils this and is the major hurdle to be overcome in handling SAR images The Upper Limit as one would expect provides the highest statistical accuracy There are two important caveats to note however 1 The result is based on the assumption that the RCS within the region is indeed homogeneous In real applications this will not generally be the case 2 The ground truth can only be used where it exists ie the true region boundaries are known Segmentation on the other hand can use the training set to enable it to classify over as large an image as is required Thus the overall conclusion is that segmentation provides an essential precursor to intensity classification This is the approach taken by the classification routines in InfoPACK 29 Mar 10 NASoftware Ltd 61 CHAPTER 6 CASE STUDIES 6 2 Segmentation and Classification of Multiple Intensity Images 6 2 1 Introduction In the previous case study we demonstrated the capability of segmentation to improve the classifica tion of a simulated 3 look SAR image In some applications a sequence of such images is available We assume as prior knowledge that every image in the sequence has fundamentally the same regions and boundaries with only variation in the RCS within each region Under these conditions we may perform joint segmentation over every image in the sequence If we do this the common boundaries reinforce one another and help to overcom
14. Your Licence The licence to use the software does not include a licence to resell all or any part of it either in the form in which you received it or as a subsystem of any other software package Please contact NASoftware Ltd for details on run time licences Warranty NASoftware Ltd gives no warranty that all errors have been eliminated from this manual or from the software supplied with it Nor does NASoftware Ltd give any warranty or representation as to the fitness of the software for any particular purpose We strongly recommend that before using the software for any valued purpose you test its suitability for that purpose NASoftware Ltd disclaims all responsibility for any direct indirect or consequential losses damages costs expenses or claims of any nature resulting from any deficiency in either the documentation or the software 4 NASoftware Ltd 29 Mar 10 1 3 How to Contact Us Getting Help If you have problems installing our software then please contact us We can also provide suggestions on how to use InfoPACK to get the best results from your images To request help via e mail send a message to support nasoftware co uk It may also be worth checking the information on our website http www nasoftware co uk Bug Reports If you have any problems or find bugs in this software please inform us Give the following infor mation e Version of the software run any command line routine with the v option e Platf
15. d Upper Limit Table 6 4 2 a Single Pixel Assigned class True Class 0 1 2 3 4 5 0 0 9993 0 0 0007 0 0 0 1 0 0009 0 9404 0 0215 0 0245 0 0072 0 0055 2 0 0017 0 0210 0 8290 0 0706 0 0761 0 0017 3 0 0 0170 0 0660 0 9059 0 0105 0 0007 4 0 0 0072 0 1193 0 0199 0 7811 0 0723 5 0 0 0104 0 0042 0 0026 0 0439 0 9389 Table 6 4 2 b 3 by 3 Window Assigned class True Class 0 1 2 3 4 5 0 0 8581 0 0351 0 0689 0 0372 0 0007 0 1 0 0002 0 9984 0 0012 0 0 0002 0 2 0 0006 0 0014 0 9802 0 0 0178 0 3 0 0 0060 0 0060 0 9880 0 0 4 0 0 0 0823 0 0 9177 0 5 0 0 0 0005 0 0 0179 0 9816 29 Mar 10 NASoftware Ltd 83 CHAPTER 6 CASE STUDIES Table 6 4 2 c Joint Segmentation Assigned class True Class 0 1 2 3 4 5 0 0 9723 0 0041 0 0122 0 0115 0 0 1 0 0012 0 9960 0 0015 0 0007 0 0003 0 0002 2 0 0003 0 0019 0 9972 0 0003 0 0 0003 3 0 0 0013 0 0026 0 9961 0 0 4 0 0 0018 0 0018 0 0 9964 0 5 0 0005 0 0010 0 0010 0 0 0010 0 9963 Table 6 4 2 d Upper Limit Assigned class True Class 0 1 2 3 4 5 0 1 0000 0 0 0 0 0 1 0 0 9998 0 0002 0 0 0 2 0 0 0003 0 9997 0 0 0 3 0 0 0 1 0000 0 0 4 0 0 0 0 1 0000 0 5 0 0 0 0 0 1 0000 6 4 3 Conclusions 84 Joint segm
16. default 48 To prevent obvious joins between tiles an overlap of 48 is used tmpdir directory default None Intermediate results are written to a system temporary directory 124 NASoftware Ltd 29 Mar 10 A 4 HIGH LEVEL ROUTINES A4 3 texseg Purpose Segments the given input image according to the Blacknell Tough measure Tlog I I log 1 where I denotes the intensity image values This measure accounts for variation of texture within the image Joint segmentation is performed when there is more than one image layer Command Line Script texseg py image nc output nc quality shape stopping cooling looks maxiters regionsize intensity tilesize overlap tmpdir Python Procedure output_array texseg image_array quality shape stopping cooling looks maxiters regionsize intensity Parameters image nc The file to be segmented This can be either a single or a multiple layered real or complex valued image In this routine all complex valued layers of the image are converted to real valued layers before processing output nc The output file containing the segmented image with regions coloured according to the Blacknell Tough measure The texseg region map output image is the same size as the input image quality Chooses from a predefined set of segmentation parameters quality default a good compromise sensitive preserve detail or fast quick resu
17. in Figure 6 29 d As already noted there is little correspondence between the segment boundaries and the original scene structure This mixes the properties of the different classes and results in very poor classification The average fraction of correctly classified categories is only 0 373 which is worse than all the others Obviously the span measure is not worth pursuing Finally the Upper Limit provided by the ground truth itself leads to the results shown in Figure 6 29 e The average probability of correct classification is now 0 996 with only a few pixels in class O magenta incorrectly assigned Indeed the confusion matrix shown in Table 6 1 d reveals very little ambiguity 112 NASoftware Ltd 29 Mar 10 6 8 SEGMENTATION AND CLASSIFICATION OF POLARIMETRIC DATA Figure 6 29 b dh Figure 6 29 c Figure 6 29 d Figure 6 29 e Figure 6 29 Classification based on a single pixel b 3x3 window c ML polarimetric segmen tation d span segmentation e Upper Limit 29 Mar 10 NASoftware Ltd 113 CHAPTER 6 CASE STUDIES Assigned class True Class 0 1 2 3 4 5 0 0 4087 0 0946 0 0662 0 1108 0 1345 0 1851 1 0 1180 0 3069 0 0517 0 1300 0 2070 0 1865 2 0 0119 0 0039 0 9106 0 0628 0 0053 0 00
18. layers as a single polarimetric image tilesize integer default 512 A tilesize of 512 x 512 is used overlap integer default 48 To prevent obvious joins between tiles an overlap of 48 is used tmpdir directory default None Intermediate results are written to a system temporary directory Further Comments If you are interested solely in whether there are changes over the whole sequence of layers in the input image use the routine change_detection 138 NASoftware Ltd 29 Mar 10 A 4 HIGH LEVEL ROUTINES A 4 10 despeckle Purpose despeckle takes a speckled image and estimates the underlying radar cross section using a vari ant of the simulated annealing optimisation method The resulting despeckled image is smoothly varying yet preserves detail Command Line Script despeckle py image nc output nc quality cooling looks maxiters intensity tilesize overlap tmpdir Python Procedure output_array despeckle image_array quality cooling looks maxiters intensity Parameters image nc The image to be smoothed This can be either a single or a multiple layered real image output nc The smoothed output image is of the same dimensionality as the input image yet is always real valued quality Chooses from a predefined set of segmentation parameters quality default sensitive or fast default default These choices set the following parameter values Parameter Fa
19. looks pfa intensity Parameters image nc This can be either a single or a multiple layered real valued image output nc The coloured region map output image is of the same dimensionality as the input im age if the original image was real valued The coloured region map output image is single layered in the case of a complex valued original image width integer default 3 The width of the filter window height integer default 3 The height of the filter window looks float default 1 The looks measure of the image data If looks is negative use 1 the routine will compute this parameter from the data pfa float default 5 Specifies the Probability of False Alarm 107 4 if pfa gt 1 to be used in determining whether changes are significant The larger pfa is the less chance there will be of reporting insignificant changes intensity integer default 0 Assume the data is amplitude square root intensity tilesize integer default 512 A tilesize of 512 x 512 is used overlap integer default 48 To prevent obvious joins between tiles an overlap of 48 is used tmpdir directory default None Intermediate results are written to a system temporary directory 144 NASoftware Ltd 29 Mar 10 A 4 HIGH LEVEL ROUTINES Further Comments None 29 Mar 10 NASoftware Ltd 145 Appendix B Low Level and Utility Routine Documentation B 1 Documentation Conventions In this chapter we document all of the low level
20. 232 Using a VINE Re o ie Moe eB aerate all eau eo a hoe 17 333 ile Inipett URN odo isa Pe ed ee he ew ee A 18 Image Processing with InfoPACK 20 Al Segmentation 29k ko eo OA SOE he wa ee 20 41 1 Thesegment routing 0 5 48 65 44 4684 Sa bebe eee ee RS 21 412 Theeasyeeq rove sa boia a A dae Bae ae dde 24 4 1 3 Themergeroutine 2 2 o 25 414 The texseg routine a 26 44 5 The 61S6 TOUS oscuras o e A A a a Bw 26 4 1 6 Thedefleckroutine sec ecset aa 0 00 00 a a a 27 NASoftware Ltd 29 Mar 10 42 Classification ecce ee dhe Sk we he dh i ew we ee de aa 424 The classifi yiouune a osas seo na de ea aah ash a eA 42 2 Themask Stats Trout once ee Seca tha ate ah dee HB aoe 43 Aven Change Detection s cosa e ca coi eae a ee eee ee a ee 4 3 1 The change_detectionroutine 2 0 00 ee 4 3 2 The change_detection_positionroutine 44 D techne Edges ca cos Ba AED RESO A eo Ea ek ee we ad 44 1 Thescanedgeroutine s eas q a eaa 0 002 ee eee 44 2 Theedgeroutine e 45 Speckle Reduchom cor ss eosin a ads D ai eee eee eee ae be 4 5 1 The despeckle routine 0 0 0 0 0 2 ee eee eee 4 6 Point Object Detechion 2 cs es emro BSH Rw Roe Re eA ha Ree ae HO THE Gear TOUS oe le Ade ke ha we A ea Rd del BERENI 65 ea a AAA Be aa Ox BIDUOSTAPES ce a a BE ee a ds ee eee Sa eh 5 Using the InfoPACK GUI SA o 1 AS eee eee eed bebe bd eee RE Ee Sl Stating The InfoPACK GUL
21. 27 CHAPTER 4 IMAGE PROCESSING WITH InfoPACK 4 2 Classification A classification algorithm is similar to a segmentation algorithm in that it produces a cartoon image in which each pixel is assigned to one of a set of predefined classes The number of regions in a segmentation may be quite high but these segments may usually be classified into a small number of categories for example fields roads hedges forest clearings potatoes wheat ploughed grass For many purposes the classification is more useful than the segmentation 4 2 1 The classify routine classify performs Maximum Likelihood classification of intensity images Maximum likeli hood classification is based upon the assumption that there exist statistical models describing the distribution of the classes in the attribute space Given these models the class of a new object is determined by calculating which of the models is more likely to describe that object In other words the model with maximum likelihood is selected Starting with a segmentation of the original image which has the correct statistical model for the data and a plausible boundary model classify will allocate each region to one of a number of classes The output is a single image with the regions labelled with the class number ie 0 1 2 InfoPACK supports both supervised and unsupervised classification Unsupervised Classification In unsupervised classification the class values will be derived from un
22. 5 days of 3 channel polarimetric data would be stored as a 15 layer complex image and the mtpol 3 parameter would be used The default value of mt po1 0 uses all of the layers as a single polarimetric image tilesize integer default 512 A tilesize of 512 x 512 is used overlap integer default 48 To prevent obvious joins between tiles an overlap of 48 is used tmpdir directory default None Intermediate results are written to a system temporary directory Further Comments Except for the det measure each of these measures is computed layer by layer With the exception of det and mean the data must be real 130 NASoftware Ltd 29 Mar 10 A 4 HIGH LEVEL ROUTINES A 4 6 defleck Purpose Given an original image file and a region map of this image de fleck will remove small segments Command Line Script defleck py region_map nc image nc output nc tilesize overlap tmpdir size mtpol Python Procedure output_array defleck region_map_array image_array size mtpol Parameters region_map nc The file containing the region map image This image should be one of either e A single layered real valued image of dimensions equal to the dimensions of the first layer of the original image image nc e A real valued image of exactly the same dimensionality as image nc In the case of a multiple layered region map the algorithm takes the intersection of the given layers to be the region map on which to pe
23. Figure 6 23 is compared with the predicted gamma PDF of mean 1 0 and order 4 0 This is illustrated in Figure 6 24 It shows that both the order of the distribution and its shape differ from the ideal theory 6 7 3 Conclusions e We have illustrated the effect of removing speckle by invoking two different models the cartoon model for segmentation and the local similarity model for despeckling e Segmentation yields the result that is closest to the speckle model in that the residual ratio resembles speckle most closely e Segmentation shows evidence of many small segments These are retained when the data is merged indicating that they are statistically significant 102 NASoftware Ltd 29 Mar 10 6 7 AN EXAMPLE SEGMENTING AND DESPECKLING A LARGE IMAGE Figure 6 23 Residual ratio for the despeckled reconstruction in Figure 6 22 0 018 0 016 0 014 0 012 0 01 P R 0 008 0 006 0 004 0 002 Ratio R Figure 6 24 Comparison of observed residual ratio for despeckled data shown in Figure 6 22 with the theoretical PDF for ENL 4 0 full curve having mean 1 0 and order 4 0 29 Mar 10 NASoftware Ltd 103 CHAPTER 6 CASE STUDIES e Despeckling presents a more continuous fluctuation in the background inherent in the local similarity model e Despeckling shows strong correlations around structures in the scene as a consequence of the 3x3 construction window e Despeckling yields a poorer fit to th
24. Simulated 3 look SAR data 6 1 2 Comparison of classification methods We can now compare the results obtained classifying the data in the test image using the following methods Upper Limit classifying the average intensity in each ground truth region Segmentation classifying the average intensity in each segment 3x3 window classifying the average intensity over a 3x3 window Lower Limit classifying the intensity in each pixel We expect the Upper Limit results unachievable on real data to be best the Lower Limit results to be the worst The major interest is in seeing how much better are the segmentation based results than those achieved with a traditional local 3 by 3 filter In each case we calculate the likelihood that each region is a member of a gamma distributed PDF of defined mean depending on the class The region is then assigned to that class which has the maximum likelihood We assume for this comparison that the order parameter determined by the speckle is uniform across the image as would be expected in practice The classification results are shown in Figure 6 1 2 Classification of a single pixel shown in Figure 6 1 2 a is clearly very poor as expected Indeed the average probability of correctly classifying each category is only 0 364 and the associated confusion matrix shown in Table 6 1 2 a reveals how poorly the intermediate classes are assigned Classification over a 3x3 window shown in Figure
25. The process can be understood in terms of a simple multiplication noise speckle model it an Ne t eng NA w Ke IN Figure 6 5 2 c Figure 6 5 2 d 6 5 2 a A detail from the original image shown in Figure 6 13 b The segmentation c The residual speckle d The expected residual speckle Suppose that the observed intensity is represented as a speckle noise model such that I 0oxn where a is the underlying cross section and n the multiplicative noise which is gamma distributed with order L where L is the ENL Let us suppose that the segmentation succeeds in representing the 86 NASoftware Ltd 29 Mar 10 6 5 HANDLING CORRELATED IMAGES underlying cross section exactly Under these conditions the variance of the intensity normalised by the cross section is given by var I o var n 1 lt Ifo gt lt n gt L Thus if we take the ratio of the original intensity to the derived segmentation the result should have the properties of random speckle with the expected correlation properties and relative variance In Figure 6 5 2 c we show the residual speckle derived from Figures 6 5 2 a and 6 5 2 b In Figure 6 5 2 d we show a simulated speckle image having approximately the correct correlation properties The correlation in 6 5 2 d is assumed to correspond to a square averaging window no attempt has been made to reproduce the correct correlation function It is clear that 6 5 2 c fails to show the expected speckle c
26. accessed via the filter function use method scan Full Python script and function syntaxes for scanedge are given in section A 4 11 Figure 4 12 shows a raw DLR ESAR amplitude image and edge maps produced from it for several combinations of filter size and pfa It is clear that the dominating parameter is the filter size we recommend using a size of 7 by 7 4 4 2 The edge routine Alternatively by using the edge utility you can produce an edge map of the region boundaries from a segmentation An edge map overlaid on its corresponding cartoon may show up region boundaries you would otherwise overlook two adjacent regions may have similar intensities Do not use scanedge to convert a cartoon image the output of a segmentation function into an edge map Use edge to perform this task An example of the use of edge is shown in Figure 4 12 f The edge algorithm can be per formed upon images with more than one image layer Full Python script and function syntaxes for edge are given in section B 2 5 34 NASoftware Ltd 29 Mar 10 4 4 DETECTING EDGES Figure Figure Figure Figure Figure Figure Figure 4 12 a Raw ESAR Image b scanedge output with filter size 3 by 3 and pfa 1075 c 7 by 7 d 10 by 10 e 7 by 7 with pfa 1077 f Edges found from the segmented image using edge 29 Mar 10 NASoftware Ltd 35 CHAPTER 4 IMAGE PROCESSING WITH InfoPACK 4 5 Speckle Reduction Normally it is not necessar
27. ae ares et eae eg ile 159 B221 BCaler Add neue a beck el he h ee we ew a ee lh i 160 B 2 22 scalar divide erica dada eA ea ed eae Rede WO 161 A Scalar multiply e sa ah a a BA ee Re A ee aA 161 B22 SCAnedge o 6 eor ee ek RE Lae Le bs 162 B225 AE eo Ae ee Rs 163 B220 S VaTrE lidad ee al aa dob ee a 163 Big SUBLCICE eaaa a a ee ade a Beers amp BS BS 164 Bi2 28 SUM oca ead eda Oo bo Seda eawdiwea da edd eda we 164 P229 threshold cocina Ga eae ee eae be de Oe Bae mee a a 165 B 3 File Handling Facilities oscars dr do ob ba de 166 Bal A ae ee wt tee ee E G E Y 166 Bolt GOME ie ed eR Oe ah ae ae EM da 166 29 Mar 10 NASoftware Ltd v Bad SRETACE o 2 4 ibe ee ee ee ke be wd be dk ee ee 168 B34 GAL ELA Leite sozi e de a ee eee ae eM ee ee ed 168 Bio Gis IGG isis 242 bee Peed EARS Redo ba de aoe Hs 169 B30 CUESTE oi a wu A BL ae we A ES 169 Bio AMSG oe ale a BE ee ee ae a ete a 170 B383 l yerconbine err ceka ara a es eee ar E we 170 B39 Layer splat ic na dda da eb GOS Re A ode ee we Sw dd 171 Bad T 4 oh ba adobe aa tol amp Malik es oh ad 171 B 3 11 makebitmap oia daa 00 bed ba eda eh a Eo 172 BS Po e ee cee Hae a ee ek we Gk ed he a a 173 Bods meeeGedd idad AA 173 Biddle DEBES ser ead eda Oo doe Bada paer ea ea ew hav da OS 174 Bale mumbles ea idk hone Oe AE A oe ON a RE AN 174 E ee we a a aL a 175 LT o bo a EEE EE ce oe a we E 176 BS COPE 66 seat es he as Sacha ob a wee a a 177 Bad BORER las ba Roe eee ida ide RE TEE ES 1
28. applying average and shrink to the intensity of 2 pixels in range x and 3 in azimuth y The resampled intensity is shown in Figure 6 5 3 c with the corresponding segmentation in 6 5 3 d This segmentation has the expected form with large areas of uniform cross section and very few speckle sized segments The ratio image for the original data 6 5 3a divided by the seg mentation 6 5 3b is shown in Figure 6 5 3 e The speckle blobs are clearly not of the correct size for the expected correlations indicating that much of the speckle fluctuation is appearing in the segmented version The residual is merely the contributions which are uncorrelated on the scale of a pixel This is confirmed by calculating the ENL which is 1 70 instead of the theoretical value of 1 0 Figure 6 5 3 f shows the corresponding ratio after the original intensity data was resampled Speckle correlations are now on a much larger scale as expected and the calculated ENL is 1 03 Thus this resampling scheme has indeed removed the effect of speckle correlations to a large extent Thus we conclude this section by noting that the effects of speckle correlations in the intensity data can be removed by estimating the required resampling from the complex field or intensity ACFs The former is more accurate the latter can lead to complications since it includes the ACF of the cross section 6 5 4 Resampling applied to the original intensity image Let us conclude this
29. at half height is 1 4 pixels range and 1 9 pixels azimuth which is consistent with the defined correlations Now we construct ACF of the detected intensity image shown in Figure 6 15 In line with the discussion above we break the intensity ACF into terms corresponding to those proportional to the correlation of the instrument function the second and fourth terms above the flat background already subtracted and the third term arising from the incoherent imaging of the cross section variations Since the intensity derived from the SLC data has L 1 the instrumental profile term can be separated from the total as shown by the red lines in Figure 6 15 It is apparent that the intensity ACF is consistent with the measured instrumental ACF in Figure 6 14 and the expression above The next step is to derive the effective resampling values The instrumental part of the intensity ACF has a maximum value of 2 2 at zero lag A fraction of 4 of this would correspond to the 20 of the field ACF in Figure 6 14 This yields resampling values of 3 and 2 for azimuth and range lines respectively This section has demonstrated how the speckle correlations introduced by the imaging instru ment function appear in the complex field and intensity ACFs The next stage is to use the infor mation from these ACFs to remove the effects of correlation from the SAR intensity images Resampling to remove speckle correlations Let us initially consider how to
30. but will not square root the regional intensity gamma order parameters since that would make them meaningless A 2 3 Memory Management Options All command line InfoPACK scripts with very few exceptions perform file to file operations by splitting the infile into tiles of size tilesize These tiles are processed and then recombined into outfile An overlap can be used to remove the effect of any joins Note that the default tile sizes and overlaps may need altering to better reflect the sizes of the features of interest in the pro cessing chain For example when aiming to produce large regions with texseg or when running segment on polarimetric images tile and overlap sizes will typically need to be increased to avoid any tiling artifacts appearing in the final image By processing files in this manner arbitrarily large images can be processed entirely in RAM Due to the file to file action it is not possible to create outfile with the same name as infile By default intermediate results are written to a temporary directory as defined by system environ ment variables Typically this could be C WINNT TEMP for a Windows system or tmp under Linux Other directories can be used if preferred by changing the tmpdir parameter The following parameters can be set to manage this process tilesize integer default 512 A tilesize of 512 x 512 is used overlap integer default 48 To prevent obvious joins betwe
31. cyan 3 red 4 blue 5 green 6 yellow 7 and white 8 29 Mar 10 NASoftware Ltd 75 CHAPTER 6 CASE STUDIES Table 6 3 2 Probability of detecting changes at the correct position within a sequence The average probability is calculated over all 8 possible positions The specific results for the known change positions are also listed Method Average Position 4 Position 5 Position 6 Single Pixel 0 8618 0 9004 0 5123 0 5640 3 by 3 window 0 9433 0 9782 0 7322 0 8970 Segmentation 0 9925 0 9944 0 9729 0 9897 Upper Limit 0 9996 0 9998 0 9991 0 9995 The single pixel results confirm the fact that only the strong change 4 is reasonably detected The average false classification probability is 14 The 3x3 window gives better response for all three positions of change with the results graded in order of step strength 4 6 and 5 and an aver age false classification probability of 5 7 Again there is a considerable improvement using joint segmentation with an average false classification of only 0 75 The individual detection probabil ities are again in order of step strength Finally the Upper Limit is almost exact with an average false classification probability of only 0 04 Once again the individual detection probabilities are in order of step strength In making these comparisons we observe the advantage of using defined edge maps for the regions of interest where these are available and can
32. file but all layers must have the same dimensions A complex valued array is declared as variable_name num_rows num_cols_complex Again it can have any of the standard NetCDF types The array contains num_cols_complex elements but there are only half that many num_cols complex elements they are stored with the real and imaginary parts interleaved In summary all integer types can be signed or unsigned any type can be real or complex D 3 Attributes Many of the attributes correspond directly to those used in the GDAL interface These are indicated below Where possible the NetCDF tag used to represent the attribute is standard and all except the Conventions tag are optional D 3 1 Global Attributes NetCDF Tag GDAL Member Description Conventions A character array containing the string InfoSAR ConventionsURL A character array containing the string href http www nasoftware co uk geotransform GeoTransform An array of six doubles corresponding to the GDAL geotransform parameters projection Projection A character array containing a projection description in one of the formats recog nised by GDAL 198 NASoftware Ltd 29 Mar 10 D 3 ATTRIBUTES D 3 2 Variable Attributes NetCDF Tag GDAL Member Description long_name Description A character array containing a description of the data missing_value NoDataValue Any pixels in the data that have this
33. file from a classification For each class labelled 1 2 3 the mean and order parameter is calculated from the input image Colour Classes Use this routine to give false colour to your classified image allowing you to view all classes together in one image instead of the maximum of four allowed with OpenEV Classmap2Layers A classified image has regions labelled 1 2 3 This routine splits the classification into layers or masks so that the first layer is a binary 0 1 mask of those region in class 1 The second layer is a mask of class 2 and so on Layers2Classmap This routine converts a set of binary 0 1 masks into a classification The pixels in the first layer are converted to class 1 The pixels in layer 2 become class 2 and so on Display Class Means Given a classification means files this routine prints out the class means in an easy to understand format Combine Matrix The purpose of this routine is to combine a set of classification means files into a single file Use this routine on a set of means files as you would use Layer Combine in section B 3 8 on images Split Matrix If a classification means file is generated from a multi temporal image this routine splits the means file into seperate files as Layer Split would split the original multi temporal image into layers 48 NASoftware Ltd 29 Mar 10 5 3 THE MAIN InfoPACK MENUS 5 3 5 The Change Detect Menu Change Detection Detect change
34. h s x S where the summation ranges over the scatterers a s and s are the amplitude and phase respec tively of the signal received from scatterer s and h is the instrument or point spread function The value of h is near 1 when s is in or near the resolving cell corresponding to pixel x and near 8 NASoftware Ltd 29 Mar 10 2 3 FREQUENCY BANDS zero otherwise If we assume that A is translation invariant does not depend on x then it can be written as a one parameter function h s x Thus the detected field E is an array of complex numbers The square of the modulus of the field at x is called the detected intensity at x the square root of the intensity is called the envelope or the amplitude Note that this is not the same as the amplitude of the received signal because the received field is perturbed by the instrument function The amplitude of the received signal a s is called the reflectivity and its square is called the surface cross section It is this that we try to recreate when reducing the speckle Datasets for SAR images can be stored as either intensity or amplitude data It is important that you know what type of data you are using Many images are stored as amplitude since the speckle is less obtrusive when these images are displayed 2 3 Frequency Bands The microwave part of the electromagnetic spectrum 1 30cm is used by radar It was broken up into bands which were given arbitrary labels during Wo
35. in the segment Note that the default overlap size for this method is set to zero Use of a non zero overlap is likely to lead to unpredicatble behaviour due to the poor estimates for this measure from small sample sizes This measure can only be prescribed for real images size Size of segments Each pixel will have a value of the number of pixels in the region This measure can only be prescribed for real images region Region number of the segment This measure can only be prescribed for real images Also this measure is not supported under the tiling scheme and will thus not run in a memory efficient fashion with arbitrarily large images var The variance of the data in the segment This measure can only be prescribed for real images det The determinant of the M by M covariance matrix For M gt 1 complex images this is defined as det det X where E cov zi j i 1 2 M is the covariance of x and x This measure can only be prescribed for an M layer M gt 1 complex image looks float default 1 The looks measure of the image data If looks is negative use 1 the routine will compute this parameter from the data intensity integer default 0 Assume the data is amplitude square root intensity mtpol integer default 0 Polarimetric data is stored as layers of complex data however this is the same file format as multi temporal polarimetric data To distinguish the two the mt pol parameter is used For instance
36. in the textured segmentation routine texseg gorder Gamma order parameter of segments The estimated order of the gamma distribution best fitting the data in the window This measure can only be prescribed for real images korder K order parameter of segments The estimated order of the K distribution best fitting the data in the window This measure can only be prescribed for real images The parameter looks can be used with this method enl Equivalent number of looks ENL calculated from the squared mean divided by the variance This measure can only be prescribed for real images 29 Mar 10 NASoftware Ltd 141 APPENDIX A InfoPACK HIGH LEVEL ROUTINE DOCUMENTATION var The variance of the data in the window This measure can only be prescribed for real images med The median of the data in the window This measure can only be prescribed for real images The filter determines the median for the data given i e it ignores the intensity flag det The determinant of the M by M covariance matrix For M gt 1 complex images this is defined as det det X where Xij COU Di j 4 1 2 M is the covariance of x and x This measure can only be prescribed for an M layer M gt 1 complex image cfar Single pixel CFAR target detector The CFAR algorithm works by calculating the likelihood of the central pixel in the window coming from the background distribution as represented by the rest of the window The classification of
37. max Parameters minimum minimum value 0 maximum maximum value 1 tilesize integer default 512 A tilesize of 512 x 512 is used tmpdir directory default None Intermediate results are written to a system temporary directory B 2 4 divide Purpose Divides two image files The result is the division of corresponding pixel values of the input images Command Line Script divide py imagel_in image2_in image_out tilesize tmpdir Python Procedure array_out Numeric divide arrayl_in array2_in 29 Mar 10 NASoftware Ltd 149 APPENDIX B LOW LEVEL AND UTILITY ROUTINE DOCUMENTATION Parameters tilesize integer default 512 A tilesize of 512 x 512 is used tmpdir directory default None Intermediate results are written to a system temporary directory B 2 5 edge Purpose Given a region map edge will scan the segments and construct an edge map Command Line Script edge py region_map nc output nc tilesize overlap tmpdir Python Procedure output_array edge region map array Parameters region map nc The file containing the region map image This should be a single or multiple layered real valued image output nc The output file This is a real valued image of the same dimensionality as the input region map tilesize integer default 512 A tilesize of 512 x 512 is used overlap integer default 48 To prevent obvious joins between tiles an overlap of 48 is used tmpdir directory de
38. meth ods pfa float default 5 Specifies the Probability of False Alarm 107 4 if pfa gt 1 to be used in determining whether changes are significant The larger pfa is the less chance there will be of reporting insignificant changes This estimate can be used with some of the methods intensity integer default 0 Assume the data is amplitude square root intensity mtpol integer default 0 Polarimetric data is stored as layers of complex data however this is the 142 NASoftware Ltd 29 Mar 10 A 4 HIGH LEVEL ROUTINES same file format as multi temporal polarimetric data To distinguish the two the mt pol parameter is used For instance 5 days of 3 channel polarimetric data would be stored as a 15 layer complex image and the mtpol 3 parameter would be used The default value of mt po1 0 uses all of the layers as a single polarimetric image tilesize integer default 512 A tilesize of 512 x 512 is used overlap integer default 48 To prevent obvious joins between tiles an overlap of 48 is used tmpdir directory default None Intermediate results are written to a system temporary directory Further Comments Conventient wrappers to this function are available in the routines filter_bt filter_det filter_enl filter_gorder filter_korder filter mean filter var median and scanedge A4 12 filter_bt filter_det filter_enl filter_gorder filter_korder filter mean filter var Purpose Uses a
39. o acct oe a a a ee Oe S21 WINS ea a a a Raye Wa al A O A A eS 5 2 The Mata InfOPACK Menus o ss rh a de a a 33 1 The Ple Memo c cci e c oieee ad id Ch SEER 223 2 The Tools MENU qire eaaet eh e a E a ee Be 2 3 3 The Segmentation Menu 2244 44 84 60 ee eed ee ws Sod The Classity Memi i cies amo eb oad eA A we eae ee dae dS Soo The Change Detect Mena si n a a de e i e D O a a S20 The Help Ment is aaa ada isa ala ae ai eS 34 The OpenEY Image Viewer iii ee hee wae a a ee 5 4 1 Classification Images 2 ee a casa eee eG 5 5 Using the InfoPACK Shell Window o ee eee 53 1 Syntax Highlighting s es oh ap ew SL pdeMaho lt roec cec Ga ksa eae ba ade ala O we aS 233 Command History s s s eoa oe a AAA a Ea a 55 4 Function Call Prompting e s4 aoma sd ek nas a i 55 3 Class Member Promp ng sc osc sesadit greetae drk ren 5 6 Developing InfoPACK Scripts 0 0 eee ee es 5 7 The IDLE Integrated Development Environment 6 Case Studies 6 1 6 2 Segmentation and Classification of Intensity Images GLI CIEGO y s ereda bbe ea da Sard wales Ee oe Pe he OS 6 1 2 Comparison of classification methods o o 6 1 3 C nelusio s ociosa da Segmentation and Classification of Multiple Intensity Images OZIL INTO MICA oc a ra A a A A e 29 Mar 10 NASoftware Ltd 41 41 41 41 41 41 43 44 46 47 49 49 50 50 50 51 51 51 52 52 5
40. objects with a strong vertical component such as walls and waves For larger surface features the changes are unpredictable However the relative intensities in a patchwork of fields can be quite different under different polarisations allowing field See Section 2 4 for a description of this notation 29 Mar 10 NASoftware Ltd 7 CHAPTER 2 SYNTHETIC APERTURE RADAR IMAGES boundaries invisible in one image to be detected in another or providing the means to distin guish between various classes of ground cover much more reliably than if only unpolarised data is available Topography of the ground Ground which is sloping towards the antenna will appear brighter than ground sloping away This effect is known as layover It is also possible to get radar shadows dark areas in an image that were not illuminated Water content of the imaged area The scattering of radar pulses depends on the dielectric properties of the scatterers These are affected by water drier objects appear darker than wetter ones except that as noted previously still surface water appears dark Incidence angle The incidence angle is the angle between the radar beam and a perpendicular on the imaged area A small incidence angle gives high backscatter larger angles give weaker signals but larger shadows so increasing the texture in the image Images taken at large incidence angles are also less susceptible to layover Direction of the flight p
41. remove speckle correlations when SLC data is available This is not so in many instances Therefore it is necessary also to show how the same effect can be achieved when only intensity data is available Consider first the instrument ACF shown in Figure 6 14 above The effect of correlations is insignificant for lag values of 3 pixels in range and 4 in azimuth If the intensity image were resampled by taking every 3rd pixel in range and 4th in azimuth we would expect the samples to be essentially uncorrelated This somewhat crude recipe is easy to apply and has the desired effect Unfortunately it is found to lose some of the information contained in the images for two 88 NASoftware Ltd 29 Mar 10 6 5 HANDLING CORRELATED IMAGES ACF of instrument function Figure 6 14 ACF of the complex field for the SLC image Azimuth response full line slant range response dashed line Intensity autocorrelation function Figure 6 15 Intensity ACF for the SLC data Azimuth response full lines range dashed Terms proportional to instrument response red cross section ACF green 29 Mar 10 NASoftware Ltd 89 CHAPTER 6 CASE STUDIES reasons Firstly the actual instrumental correlation lengths are expected to be 1 6 and 2 1 pixels respectively so that further separation of samples is losing uncorrelated information and degrades the resolution Secondly if we take averages of the intensity over the resampling inte
42. response if the cross section were uniform In this case the intensity ACF would be give by 1 gr k 1 7 ge k This has the same shape as the square modulus of the complex field ACF defined above Thirdly there is a term corresponding to incoherent imaging of the cross section fluctuations The fourth term represents coherent interference between random scatterers modulated by the cross section 29 Mar 10 NASoftware Ltd 87 CHAPTER 6 CASE STUDIES fluctuations This approximates to the square modulus of the complex field ACF if the cross section correlation function is approximately constant over the decay length of the instrument correlations The total intensity ACF can then be approximated by oi k 214 5 glk P Halk 1 1 5 lal PS Thus the instrumental component of the intensity ACF near zero lag has the form y 92 0 P go 0 1 where v is the order parameter of the cross section fluctuations whereas the total value is 1 1 1 1 L z i v Let us now demonstrate the complex field correlation properties using SLC data from the same SAR radar as the intensity image above In this mode the DLR ESAR has resolution of 4 m in slant range and 0 72 m in azimuth The sample spacing is 2 5 m in slant range and 0 34 m in azimuth so that this data is over sampled by 1 6 in slant range and 2 1 in azimuth Figure 6 14 shows the derived ACF of the complex field The ENL is 1 0 From Figure 6 14 we note that the width
43. split py image_in image_out tilesize tmpdir Python Procedure array_out layer split array_in Parameters tilesize integer default 512 A tilesize of 512 x 512 is used tmpdir directory default None Intermediate results are written to a system temporary directory Further Comments The python procedure returns a list of arrays B 3 10 load Purpose Loads an image file and its GIS information if available Load returns a list of length two The first element is the GDAL file as a Numeric array while the second element is a class containing the GIS information Load loads the subimage of dimensions width by height starting at the pixel location given by woffset and hoffset Command Line Script None provided Python Procedure array out gis class load image_in woffset hoffset width height 29 Mar 10 NASoftware Ltd 171 APPENDIX B LOW LEVEL AND UTILITY ROUTINE DOCUMENTATION Parameters woffset integer default 0 hoffset integer default 0 width integer default None read whole image in height integer default None read whole image in B 3 11 makebitmap Purpose Converts a Numeric array to an RGB array The array can then be saved in a variety of formats suitable for display in common graphics packages For each layer of the image values that lies in the range min to max are mapped into the range 0 to 255 If min or max are not specified then t
44. the estimated ENL 4 03 rather than 4 0 there is actually a slight departure from the form of a gamma PDF One way to eliminate differences between neighbouring segments that are relatively insignifi cant is to apply the merge algorithm Rather than providing a global solution to the cartoon model implying uniform regions segments this examines pairs of adjoining segments and tests the statis tical significance of their separation This can be expressed in terms of the false alarm probability Pfa Which describes the probability that a single segment could be split into two separate segments by speckle noise For example Pfa 1075 would indicate that there was a probability of 10 that a segment containing a uniform RCS could be split by speckle noise If a split of this probability is 29 Mar 10 NASoftware Ltd 99 CHAPTER 6 CASE STUDIES 5 3 R Rato N o o x N rc rc rc o o o o o o o o o o o dd Figure 6 19 Residual ratio derived from figures 6 17 and 6 18 dotted curve for segmented data shown in o ti 0 full curve hav dual ra i f observed res 1 PDF for ENL 1son O Figure 6 20 Compar 1 0 and order 4 0 ing mean 4 lca Figure 6 19 with theoret 29 Mar 10 NASoftware Ltd 100 6 7 AN EXAMPLE SEGMENTING AND DESPECKLING A LARGE IMAGE Figure 6 21 Result of merging segments with Pf 1075 not significant then such regions shoul
45. tilesize of 512 x 512 is used overlap integer default 48 To prevent obvious joins between tiles an overlap of 48 is used tmpdir directory default None Intermediate results are written to a system temporary directory Further Comments The means nc and orders nc files can be generated by either e Uniformly distributed class means can be calculated from the mask_classes py script Entering the values in a text file and using the ascii2nc py utility The values could be selected by using histogram py to study the distribution of the data By running the mask_stats py utility on a set of given a priori class masks e By running the mask_stats py utility on a set of given class masks made by hand see the images directory for an example of this For polarimetric data there is amask_ polstats py which estimates the class means from the polarimetric covariance In any case the means nc and orders nc files can be edited further to provide iterative supervised classification See Section B 5 for more routines on dealing with classifications 134 NASoftware Ltd 29 Mar 10 A 4 HIGH LEVEL ROUTINES A4 8 change detection Purpose For each region in a given region map change_detection looks for significant change over a series of original speckled images The routine detects large scale changes between two or more images large scale implies that it is not tailored to detecting differences in point targets between images
46. user callable facilities provided in InfoPACK For each facility both function available from Python or from the InfoPACK console by importing module infopack and script available from a command line versions are listed for a discussion of the differences see sections 1 1 and A 1 Also see section A 2 for a description of the common parameters Image Parameters e For the script versions the image files are denoted image These parameters must be the full path name of an image file For the function version the data files are denoted array These parameters must be the name of Numeric Arrays containing the image data Input parameters are given suffix _in Output parameters are given suffix _out For the script version the specified file is written before exit For the function version the specified image is returned as the or a result of the function B 1 1 pydoc Purpose The Python documentation tool Command Line Script pydoc routine 146 NASoftware Ltd 29 Mar 10 B 1 DOCUMENTATION CONVENTIONS Python Procedure pydoc help routine Parameters routine Python comes with extensive on line help pydoc infopack gives an overview of the Python scripts available in InfoPACK while for instance pydoc infopack segment gives details of the particular routine 29 Mar 10 NASoftware Ltd 147 APPENDIX B LOW LEVEL AND UTILITY ROUTINE DOCUMENTATION B 2 Image Manipulation Routines B 2 1 add Purpose
47. value should be regarded as undefined and not part of the data It is stored as a double classnames CategoryNames A character array containing a list of names that can be used as labels for the classes in dexed by the image The names are sepa rated in the string by a newline n char acter colourmap ColorTable An array of 4 byte integers one per colour in the colour map Each integer should be interpreted as four unsigned bytes these represent red green blue and opacity in the default RGBA interpretation This can be changed by setting the palette tag The colour map need only hold as many values as are indexed by the data However images translated from other for mats may have a 256 entry table even though much of it is unused palette A character array of length 4 contain ing one of RGBA CMYK HLSA or Gray These describe how the colour map should be interpreted The interpretation defaults to RGBA if this tag is not present display ColorInterpretation A character array of length 1 indicating how the layer should be shown on display hardware The character stored is for greyscale K for black or the first letter of one of the following terms Red Green Blue Alpha Hue Saturation Lightness Cyan Magenta Yellow This tag should not be used when a colour map is present 29 Mar 10 NASoftware Ltd 199
48. x 512 is used tmpdir directory default None Intermediate results are written to a system temporary directory B 2 9 f1ill_unirand Purpose Creates a real valued image of given dimensions width height and the number of layers filled with random samples These samples have a uniform distribution between O and 1 Seeds seed1 and seed2 may be specified to give a reproducible sequence of random numbers Width and height parameters must be specified Command Line Script fill_unirand py image_out width height layers seedl seed2 tilesize tmpdir Python Procedure array_out fill_unirand width height layers seedl seed2 Parameters layers integer default 1 seed1 integer default 0 seed2 integer default 0 tilesize integer default 512 A tilesize of 512 x 512 is used tmpdir directory default None Intermediate results are written to a system temporary directory B 2 10 fill polrand Purpose Creates a complex valued image of given dimensions width height filled with random samples These samples are derived from the Cholesky decomposition of the covar input and so a3 x 3 input matrix will produce a 3 layered complex image Seeds seed1 and seed2 may be specified 29 Mar 10 NASoftware Ltd 153 APPENDIX B LOW LEVEL AND UTILITY ROUTINE DOCUMENTATION to give a reproducible sequence of random numbers The width height and covar parameters must be specifi
49. 0 4 0 01 0 1 0 2j 0 4 0 01j 0 18 0j 4 140j 0 73 0 73 0 092 0 046 0 73 0 73 2 1 0j 0 23 0 15j 0 092 0 046j 0 23 0 15j 0 08 0j 5 140j 1 2 0 2 0 05 0 05j 1 2 0 2j 2 6 0j 0 05 0 1j 0 05 0 05j 0 05 0 1j 0 11 0j 6 1 0j 0 3 0 6 0 1 0 2j 0 3 0 6j 1 5 0j 0 09 0 02 0 1 0 2j 0 09 0 02 0 14 0j Table 6 8 1 Values of the covariance matrix elements for the 6 classes 29 Mar 10 NASoftware Ltd 105 CHAPTER 6 CASE STUDIES Figure 6 25 a Figure 6 25 b Figure 6 25 c Figure 6 25 d Figure 6 25 a Simulation classes b ground truth c training mask d test mask The classes are denoted by 0 magenta 1 blue 2 cyan 3 green 4 yellow and 5 red 106 NASoftware Ltd 29 Mar 10 6 8 SEGMENTATION AND CLASSIFICATION OF POLARIMETRIC DATA Figure a Figure b Figure c Figure d Figure 6 26 Amplitude of simulated HH a HV b VV c and span d images An example of simulated imagery for the HH HV and VV intensities is shown in Figure The HH image in Figure a shows no structure as expected There is some evidence of struc ture however in the HV and VV images in Figures b and c respectively Since there is visible structure we might expect that these images could be segmented jointly in order to deter mine the structure within the scene We shall show later that the sensitivity of this measure is poor when compared with full polarised segmentation The sum of these three image
50. 0 6543 Table 6 2 2 b Confusion Matrix for Classification using 3 by 3 filter 29 Mar 10 NASoftware Ltd 63 CHAPTER 6 CASE STUDIES Figure 6 2 2 b Classification using 3 by 3 filter Figure 6 2 2 c Classification using joint segmentation 64 NASoftware Ltd 29 Mar 10 6 2 SEGMENTATION AND CLASSIFICATION OF MULTIPLE INTENSITY IMAGES Figure 6 2 2 d Classification using segmentation of the averaged images Figure 6 2 2 d Classification using Ground Truth segments 29 Mar 10 NASoftware Ltd 65 CHAPTER 6 CASE STUDIES Assigned class 0 1 2 3 4 5 True class Prob 0 0 8291 0 1020 0 0628 0 0070 0 0 0 0 1 0 0059 0 9769 0 0171 0 0 0 0 0 0 2 0 0 0 1004 0 7574 0 1394 0 0028 0 0 3 0 0 0 0060 0 1883 0 5374 0 2670 0 0013 4 0 0 0 0037 0 0640 0 2505 0 3766 0 3053 5 0 0 0 0 0 0047 0 0421 0 1864 0 7667 Table 6 2 2 c Confusion Matrix for C lassification using Joint Segmentation Assigned class 0 1 2 3 4 5 True class Prob 0 0 9777 0 0176 0 0047 0 0 0 0 0 0 1 0 0064 0 9873 0 0062 0 0002 0 0 0 0 2 0 0006 0 0102 0 9823 0 0053 0 0014 0 0003 3 0 0 0 0020 0 0078 0 9830 0 0059 0 0013 4 0 0 0 0 0 0036 0 0054 0 9837 0 0072 5 0 0 0 0 0 0 0 0031 0 0084 0 9885 Table 6 2 2 d Confusion Matrix for Classification using Segmenta
51. 2 53 55 55 55 56 61 62 62 lil iv 6 2 2 Comparison of classification methods osoo o 62 62 3 Conclusions o cia eae boa ee be edo ed bea eS 66 6 3 Change Detection between Intensity Images 68 A o A 68 6 3 2 Comparison of classification methods 4 68 6303 CONCIOSIONS es s Fee oe Oe See we ee ia 76 6 4 Temporal Signatures of Intensity Images 78 GAA Introduction s e soes eca esa aua a edb alee gw ead Be ee S 78 6 4 2 Classification methods ge ge ee ee 79 64 3 Conclusions occurred A g a 84 6 5 Handling Correlated Images coo oao cma tacsa s ea Ba es be eae a 85 63 1 Tntrod ctio s s 46406 e amoa a Dae ee aa we a S 85 6 5 2 Correlation and Segmentation 86 6 5 3 Speckle correlation properties o s ca casero esst asss 87 6 5 4 Resampling applied to the original intensity image 91 6 5 5 Segmentation Conclusions 0 e e eee ee eee 93 6 6 Correlated Images and Despeckling o oo 95 6 6 1 Resampling to remove speckle correlations 96 6 6 2 COMClUSIONS ce cisma ee ha bd bd EA a ed be Ee eS 97 6 7 An Example Segmenting and Despeckling a Large Image 98 OLI Segmentation e ok a ek ee ee ee a ye a Ee Ye 98 Gha IDOSPECKMIIE pucca da Oa d4 Bada Saws ea dae o Bade OS 101 63 CONCIURIONE ok fee kk eee d e ee ek A a 102 6 8 Segmentation and
52. 29 Mar 10 NASoftware Ltd 159 APPENDIX B LOW LEVEL AND UTILITY ROUTINE DOCUMENTATION specified method of interpolation Command Line Script resize py image in image_out width height wscale hscale method Python Procedure array_out resize array_in width height wscale hscale method Parameters width integer default None The new image width height integer default None The new image height wscale integer default 1 The width parameter can be also be specified as a scaling of the original image size with the pa rameters wscale For example wscale 0 5 would produce an image half the width of the original image hscale integer default 1 The height parameter can be also be specified as a scaling of the origi nal image size with the parameters hscale method Choose from the following methods of interpolation resample Nearest neighbour average Averages neighbouring pixels when shrink nearest neighbour when expand ing image linear Linear fft Fourier transform sinc The default method is resample B 2 21 scalar_add Purpose The result is the individual pixel values from the input image plus a specified constant Command Line Script scalar_add py image_in image_out scalar tilesize tmpdir 160 NASoftware Ltd 29 Mar 10 B 2 IMAGE MANIPULATION ROUTINES Python Procedure array out scalar_add array in scalar
53. 55 3 0 0274 0 0209 0 1234 0 7596 0 0431 0 0255 4 0 0705 0 0886 0 0579 0 1284 0 5533 0 1013 5 0 06663 0 0371 0 0397 0 0522 0 0527 0 7520 a Single Pixel Assigned class True Class 0 1 2 3 4 5 0 0 8858 0 0858 0 0034 0 0101 0 0149 0 1 0 0995 0 8687 0 0 0002 0 0283 0 0033 2 0 1902 0 0959 0 6801 0 0053 0 0284 0 3 0 1135 0 1095 0 0 7690 0 0080 O 4 0 0713 0 1846 0 0 0037 0 7367 0 0037 5 0 1169 0 0384 0 0 0 0 8447 b 3x3 window Assigned class True Class 0 1 2 3 4 5 0 0 9108 0 0358 0 0149 0 0365 0 0020 O 1 0 0646 0 9263 0 0009 0 0015 0 0067 0 033 2 0 0445 0 0003 0 9552 0 0003 0 0 3 0 0105 0 0059 0 0026 0 9811 0 0 4 0 1049 0 0018 0 0 0054 0 8879 0 5 0 0094 0 0021 0 0052 0 0 0 9833 c Segmentation Assigned class True Class 0 1 2 3 4 5 0 0 9757 0 0101 0 0014 0 0014 0 0055 0 0061 1 0 1 0000 0 0 0 0 2 0 0 1 0000 0 0 0 3 0 0 0 1 0000 0 0 4 0 0 0 0 1 0000 0 5 0 0 0 0 0 1 0008 d Upper Limit Table 6 1 Confusion matrix for the different classification methods The fraction of the true class assigned to each category is given 114 NASoftware Ltd 29 Mar 10 6 8 SEGMENTATION AND CLASSIFICATION OF POLARIMETRIC DATA 6 8 4 Conclusion It is apparent from this comparison that it is pointless attempting classification on individual polarised pixels in SAR Averaging over a 3x3 window im
54. 6 1 2 b shows considerable improvement However the average fraction of correctly 56 NASoftware Ltd 29 Mar 10 6 1 SEGMENTATION AND CLASSIFICATION OF INTENSITY IMAGES Figure 6 1 1 b Test data Ground Truth The classes are represented by 0 magenta 1 blue 2 cyan 3 green 4 yellow 5 red 29 Mar 10 NASoftware Ltd 57 CHAPTER 6 CASE STUDIES Figure 6 1 2 a Classification using single pixels classified pixels in each class is still only 0 571 and the associated confusion matrix in Table 6 1 2 b reveals considerable spread in the intermediate classes The result of classification following segmentation shown in Figure 6 1 2 c is much closer to the ground truth with the average fraction of correctly classified categories rising to 0 891 and the confusion matrix showing much less ambiguity between classes Finally the Upper Limit provided by the ground truth itself leads to the results shown in Figure 6 1 2 d The average probability of correct classification is now 0 993 and the confusion matrix shows very little ambiguity The conclusions are clear if you have an accurate region map ground truth use it If not segment the image and classify the computed segmentation Table 6 1 2 a Confusion Matrix for Single Pixel Classification Assigned class 0 1 2 3 4 5 True class Prob 0 0 8000 0 3445 0 1699 0 0732 0 0362 0 0324 1 0 1777 0 3522
55. 78 B 4 Image Information Routines 2 2 00 2 eee eee ees 180 BAA GGWnbeee co ee kee eh eee Eee ee eee OL Eee 180 BAZ Uat ea esc Ga Sate Pee bake aia Beda Bs 180 BAS ASES cisco Yu ee ee ee de Ke oe ee 180 Bao MASON A ee alee Grea A eR Ge we da 181 Daa ons Sa ep ears cas rag Bk ao ee de Ba aa a ae on pee a Ge 182 BAG graphtoOGL sisas Ce hada Wed waste dod 183 B 5 Classification Routines ee 186 Boil ECO TELAS en ea be ek Ge we e Se ee ee Gk we ate a he 186 BL GlasSSGarsS acu wn ee ht A E A 186 Bis lassm ap layers 2 ee ko haa Re Ee OE a a 187 B54 Layers2classes co ati je Beha Parke ee E RR dw BS 188 Boo MaSk Sbaets cirio bes ed ERR EES 188 BSGO o eorr ee OW Re eR Meee a Se ia 190 B37 MEWS CLAS SSE ak as co a a a a a we A E 190 Boo MAtELM COMBIAS 2 4 ca chad acacia bana aka dual scans 191 BSD Mattie split oea he bw mr EE Ew e EEE 192 BS e Ea beg et he eS a A See ee a we aai ea a 192 C Library Interface 194 GI SampleCode o s coi ecsedi Bega ee bee eee dor A 194 CL EXCORION cia eke ee ee e a he Ae ee ee eR eed EE ed 194 Co Cintertade is oak ee ee OR RS Bee bo Re Samak de aca ee 195 C31 Progress Function usos e Ae BA ee eG 195 oe Complex Types oos edea A a dw 196 vi NASoftware Ltd 29 Mar 10 D NASoftware Conventions for NetCDF 197 DI Dimensions s cewek ae e LO a Be He He Gh wy GS 197 D2 Varnes ca heb a dora es oh es ee hn de a a ee ee i 197 PER aoo s Boh hol s he a eS eR EE PA
56. A eS eRe eS 198 D3 41 GlosabAtiviblites occiso ca 2 dee ee wale Oe he G4 198 Di32 Variable Attributes s es 445404048 da Be bad eR ERED HO 199 29 Mar 10 NASoftware Ltd 1 Chapter 1 Introduction InfoPACK is a set of tools designed to help in the interpretation of SAR imagery The toolset incor porates advanced SAR processing algorithms developed by Professor Chris Oliver and colleagues at NASoftware Ltd These tools use a knowledge of the image formation techniques to remove or take account of the speckle and are we believe technically well in advance of any other such facilities InfoPACK includes routines for segmentation classification speckle reduction edge detection and target detection These routines accept single or multiple images including polarimetric and time sequenced image sets They are fully automatic allowing reproducible results to be obtained independent of the user and provide unparalleled facilities for extracting the maximum information from SAR imagery There are several easy to use methods for using InfoPACK routines ranging from the Graphi cal User Interface to constructing complex processing chains in an integrated development environ ment This manual provides e Asummary of the different ways to use InfoPACK routines see this chapter e A brief overview of SAR image theory see chapter 2 e Information on the native InfoPACK image format see chapter 3 e Details and commentary on the main Inf
57. AR images Segmentation provides an adaptive function by maximising the segment size and matching the shape of the largest possible homogeneous region consistent with the data We introduced the prior knowledge that every image in the sequence has fundamentally the same regions and boundaries with only variation in the RCS within each region Under these conditions we may perform joint segmentation over every image in the sequence The common boundaries reinforce one another and help to overcome speckle noise Thus the properties of each segment represent the most accurate estimates that can be obtained In some applications a sequence of images is available but we believe that although the un derlying scene structure segments have not changed the RCS will may have changed between individual images An example is a set of fields with growing crops we want to monitor the growth between images Then we are concerned with detecting those regions that have changed in inten sity Joint segmentation is ideally matched to this process since it constructs a set of segments that are common to each image It is therefore possible to study the temporal development of identical regions of these images Two methods of change detection are considered here neither of which requires any prior knowledge Thus these methods are both unsupervised methods The first considers the sequence of images and detects whether the intensity within each region is constant th
58. Classification of Polarimetric Data 105 6 8 1 Introduction gt ss eee eed ee ga eG bane eae a wes 105 6 8 2 Segmentation performance s o ss sos ecs aca e E e 109 6 8 3 Comparison of classification methods osooso o 109 0 94 Conclusion lt s ere seasis do eed mE ee Bead a a eS oe ws 115 InfoPACK High Level Routine Documentation 116 Bl User Intere oc tol ee te a Re i Be are ek a od eed 116 Ali Scripts aad PUDCUO S se a e ss ee ee A ee 116 Ads GIS Information gt ss ep a6 45 e aia a ee ee a 117 ALs CTO oa se wr eee a Bel eee kN ears ald aod eck oh Bo 117 ALA Online Help coccion a ee a ee ee we ee a 117 AZ Common Parameters gt o o cc eed Ge GES ae be ds ER AR bw PEER WE 118 ALI Image IMEASIO S oo se ae ew Ee A ae a 118 A22 The Intensity Flag o cera da ed be E 118 A 2 3 Memory Management Options 2000 119 A 3 Documentation Conventions s se lt e a asia a a E eee ee ee ee 119 AA High Level Rontin s o o saa qa Be dd ee eG 121 AE SE o S da di a ida eee ee hae ee Re 121 Adl o E 124 A43 CERS cias dt de eS eee ks 125 o eoep a hs a ar Re a de ws a E eT 127 BAS COOLS iia BERS RR EAR Ea eh as PERE SR 129 NASoftware Ltd 29 Mar 10 PAD CEU oe ca ca ce hc te ke we wh ee a es do eed Seve Bk a a Bh ewe 131 AAT CLASSE oaasi baie eee ee ee hd be OS PS eee i 133 AAS CHANGES OSESSCELOM cc o a am we we KES SN 135 A4 9 change detection position 137 A410 Ges
59. ESSAGE 1 29 Mar 10 NASoftware Ltd 195 APPENDIX C LIBRARY INTERFACE where INFOSAR PUR MEMERROR This is a warning message to report that the InfoPACK routine is un able to malloc enough memory INFOSAR_PUR_ERROR A fatal error has occured in the InfoPACK routine INFOSAR_PUR_WARNING A warning message INFOSAR_PUR BLIP This is used to let the user know that the routine is progressing normally INFOSAR PUR MESSAGE A debugging or information message If you do not wish to use this feature then NULL or 0 can be passed in instead of progress C 3 2 Complex Types The C standard has only recently introduce a complex data type and so some C compilers do not fully supprort complex data If you do not need complex data then you can define INFOPACK_NOCOMPLEX before including infopack h to omit those definitions that use complex data If you wish you call InfoPACK routines that accept complex data then you may need to include the lt complex h gt header file and to change the line You may change this to be your complex type typedef _ complex__ float komplex If your compiler is ISO C99 complient this could be changed to include lt complex h gt typedef _Complex float komplex 196 NASoftware Ltd 29 Mar 10 Appendix D NASoftware Conventions for NetCDF The NASoftware file format used as the default file format with InfoPACK is based on the Unidata NetCDF http www unidata u
60. FujiBAS Fuji BAS Scanner Image FIT FIT image USGSDEM USGS Optional ASCII DEM GXF GeoSoft Grid Exchange Format The default format is InfoSAR B 3 18 tojpeg Purpose Converts image data to a JPEG bitmap suitable for display in common graphics packages For each layer of the image values that lies in the range min to max are mapped into the range O to 255 If min or max are not specified then the parameter sd is used with the following mapping p sdxo gt 0 u sd x o 255 where y is the mean and is the variance of the data A value of sd 0 has the special meaning of minimum 0 maximum 255 Complex data cannot be directly displayed as a bitmap and so must be converted to real data first This is done by the option cmp12real Command Line Script tojpeg py image_in image_out jpg cmpl2real sd min max 29 Mar 10 NASoftware Ltd 177 APPENDIX B LOW LEVEL AND UTILITY ROUTINE DOCUMENTATION verbose tilesize tmpdir Python Procedure S procedure makebitmap Parameters cmpl2real string default abs Chooses from a set of methods for converting the complex image to a real one abs absolute value mag magnitude of complex number abs x abs phase Phase of complex number real Real part imag Imaginary part sd float default 2 0 min list default None This can either be a single value or a list eg min 0 3 7 2 Note no spaces allow
61. INTENSITY IMAGES Figure 6 2 2 a Classification using single pixels the poor performance is due to the averaging filter This means that each change in cross section is surrounded by a band of intermediate value which are classified into other categories The result of classification following segmentation shown in Figure 6 2 2 c is very close to the ground truth with the average fraction of correctly classified categories rising to 0 984 and the confusion matrix showing much less ambiguity between classes The Higher Limit result shown in Figure 6 2 2 d yields an identical value for the average probability of correct classification The fact that there is no advantage over joint segmentation reveals that the latter is capable of extracting all the information within the set of identical images Finally the Upper Limit provided by the ground truth itself leads to the results shown in Figure 6 2 2 e The average probability of correct classification is now 0 992 with the confusion matrix showing very little ambiguity Table 6 2 2 a Confusion Matrix for Classification using Single Pixels Assigned class 0 1 2 3 4 5 True class Prob 0 0 9696 0 0304 0 0 0 0 0 0 0 0 1 0 0346 0 8409 0 1236 0 0 0 0 0 0 2 0 0073 0 1455 0 6641 0 1821 0 0063 0 0 3 0 0 0 0085 0 2299 0 5036 0 2260 0 0320 4 0 0 0 0 0 0325 0 2767 0 4069 0 2839 5 0 0 0 0 0 0037 0 0695 0 2726
62. OUTINES A 4 11 filter Purpose Uses a moving window of size width by height to filter the image by computing an output pixel value from within each window according to the specified method Command Line Script filter py image nc output nc method width height looks pfa intensity tilesize overlap tmpdir Python Procedure output_array filter image_array method width height looks pfa intensity Parameters image nc This can be either a single or a multiple layered real or complex valued image de pending on the chosen method A polarimetric image should be given as a layered complex valued image output nc The coloured region map output image is of the same dimensionality as the input im age if the original image was real valued The coloured region map output image is single layered in the case of a complex valued original image method This string specifies the filter function to be used default mean The following methods are available mean This measure may be applied on either real or complex images For real images the mean value is given as the intensity mean or the square root of the intensity mean depending on whether the data given is intensity or amplitude respectively bt Blacknell Tough measure within each segment bt I log I I log Z where I denotes the intensity values This measure can only be prescribed for real images This measure is used
63. S No Yes USGS ASCII DEM dem USGSDEM No Yes X11 Pixmap xpm XPM Yes No Table 3 1 File Formats supported by GDAL and InfoPACK 16 NASoftware Ltd 29 Mar 10 3 3 GENERAL FILE FORMATS and for the converted filename default suffix nc The input file format is taken from the input filename suffix If the format is recognised the data will be converted to NetCDF format in the output file You will not normally need to import non NetCDF files before processing them they will be loaded provided that their format is recognised 3 2 2 File Export NetCDF files can be converted to other standard image formats using the Export Image item on the Tools menu see section 5 3 1 for details Alternatively the routine ygdal translate may be run from a command line Specific rou tines to output raw data nc2raw py ascii nc2ascii py and TIFF bitmap totiff py and makebitmap files are also provided Details of these routines and in particular the supported file formats are given in Appendix B 3 3 General File Formats If your file format is not supported by InfoPACK but you can describe how the data is laid out within the file then GDAL supports two different methods for reading the data 3 3 1 Using a aux file The first method is a partial implementation of the PCI aux labelled format For this a text file with the file extension aux is created which describes the image layout for example AuxilaryTarget ou
64. SIFICATION ROUTINES Command Line Script make_classes py image_in image_out num_classes sd minimum maximum cmpl2real intensity verbose tilesize overlap tmpdir Python Procedure mean_array make_classes array_in num_classes sd minimum maximum cmpl2real verbose intensity Parameters num_classes integer default 5 sd float default 2 0 minimum list default None maximum list default None cmpl2real string default abs Chooses from a set of methods for converting the complex image to a real one abs absolute value mag magnitude of complex number abs x abs phase Phase of complex number real Real part imag Imaginary part intensity integer default 0 Assume the data is amplitude square root intensity verbose integer default O If set to 1 prints out the class means tilesize integer default 512 A tilesize of 512 x 512 is used overlap integer default 48 To prevent obvious joins between tiles an overlap of 48 is used tmpdir directory default None Intermediate results are written to a system temporary directory B 5 8 matrix_combine Purpose Combines classification means files held in sets of images into a single file Classification can be performed on either single or layered images As single images can be com 29 Mar 10 NASoftware Ltd 191 APPENDIX B LOW LEVEL AND UTILITY RO
65. TY ROUTINE DOCUMENTATION B 3 File Handling Facilities B 3 1 ascii2nc Purpose Converts an ASCII file into an NASoftware Ltd NetCDF file This can be useful for generating means and orders files for use in supervised classification Each line in the ASCII file corresponds to one row of the array Numbers within a line are separated by white space Command Line Script ascii2nc py ascii in image out Python Procedure None provided B 3 2 convert Purpose A utility to convert files between different data types and various GDAL Geospatial Data Abstrac tion Library formats The Python procedure provides data type translation the command line script also allows for specifying a GDAL output format Command Line Script convert py image_in image_out method format Python Procedure array_out convert array in method Parameters method string default None None Copy the data useful for filetype conversion byte Convert image to byte data Numeric Int0 short Convert image to short data Numeric Int16 int Convert image to int data Numeric Int32 float Convert image to float data Numeric Float32 double Convert image to double data Numeric Float 166 NASoftware Ltd 29 Mar 10 B 3 FILE HANDLING FACILITIES complex Convert image to float complex data Numeric Complex32 dcomplex Convert image to double complex data Numeric Complex real Real part of complex data imag Imagina
66. The images to be scanned for changes must be provided as layers in a single multi layer image if your images are separate use layer combine to build them into a single 3 D image The routine uses the segmentation of each layer to carry out its analysis and this must be provided too In most cases the joint segmentation produced by segmenting the multilayer image is what will most conveniently be used The routine detects changes across the whole series of provided layers the ordering of the layers in the multiple layer image file is not important The output is a single layer image specifying where any significant changes were found over the whole sequence If you are interested in the individual changes found between adjacent pairs of layers in the input image use routine change_detection_position Command Line Script change detection py region_map nc image nc output nc pfa mlthresh looks intensity mtpol tilesize overlap tmpdir Python Procedure output_array change_detection regionmap array image array pfa mlthresh looks intensity mtpol Parameters region_map nc The file containing the region map image This image should be one of either e A single layered real valued image of dimensions equal to the dimensions of the first layer of the original image image nc e A real valued image of exactly the same dimensionality as image nc In the case of a multiple layered region m
67. This al lows temporal changes to be monitored Multi frequency images The images are formed with different frequency radars Since details smaller than the wavelength of the beam are invisible a multi frequency image shows struc tures on different length scales Multi polar images Like light radar can be polarised into horizontal H and vertical V planes Transmitting H or V polarised radiation and receiving H or V gives four possible images of the scene Multi polar images are also called polarimetric images Multi sensor images Remote sensing images cover a large range of the electromagnetic spectrum visible infra red thermal radiation and microwaves Images from several bands can be selectively combined to discriminate between particular features of interest InfoPACK makes 1t straightforward to carry out data fusion processing in this way 29 Mar 10 NASoftware Ltd 11 CHAPTER 2 SYNTHETIC APERTURE RADAR IMAGES 2 8 Correlated Images All of the routines in InfoPACK assume that the pixel values in an image are statistically uncorre lated If this requirement is not satisfied you are using ERS 1 PRI data for example the perfor mance of the routines can be affected Two major sources of correlations are e Oversampling the signal during the image formation process making the pixel size less than the spatial resolution e Transforming the image so that the pixel lattice is aligned with latitude and longitude lines
68. UTINE DOCUMENTATION bined into a layered image with layer_combine the class means files can be combined into a single file with mat rix_combine Command Line Script matrix_combine py matrixl_in matrix2_in matrix3_in matrix_out Python Procedure None provided Parameters B 5 9 matrix_split Purpose Splits the classification means held in means nc into a series of files Results are written into the files output_000 nc output_001 nc Classification can be performed on either single or layered images As layered images can be split into a set of single images with layer_split the class means files can be split into a set of files with matrix_split Command Line Script matrix_split py means nc output nc Python Procedure None provided Parameters B 5 10 matrix_print Purpose Displays the data in the means file used by classi fy in an easy to read form by class Command Line Script 192 NASoftware Ltd 29 Mar 10 B 5 CLASSIFICATION ROUTINES matrix_print py means_in Python Procedure None provided Parameters tilesize integer default 512 A tilesize of 512 x 512 is used overlap integer default 48 To prevent obvious joins between tiles an overlap of 48 is used tmpdir directory default None Intermediate results are written to a system temporary directory 29 Mar 10 NASoftware Ltd 193 Appendix C Library Interface If your application supports external libr
69. Y ROUTINE DOCUMENTATION Parameters tilesize integer default 512 A tilesize of 512 x 512 is used tmpdir directory default None Intermediate results are written to a system temporary directory Further Comments The individual layers of the output image can be obtained with the routine split_layers B 54 layers2classes Purpose Converts a set of 0 1 masks into a classification image with regions labelled 0 1 2 The pixels in layer O will be given the label 0 the pixels in layer 1 the label 1 and so on Command Line Script layers2classes py image_in image_out tilesize tmpdir Python Procedure array_out layers2classes array_in Parameters tilesize integer default 512 A tilesize of 512 x 512 is used tmpdir directory default None Intermediate results are written to a system temporary directory B 5 5 mask_stats Purpose Calculates the mean and order parameters for the masked regions of the original image The mean and order parameters can then be used as supervised inputs for the classify routine The masked image is a multi layered real valued image with number of layers equal to the number of desired classes Each layer has pixels coloured either 0 or 1 the latter denoting the spacial location of the training data for that class 188 NASoftware Ltd 29 Mar 10 B 5 CLASSIFICATION ROUTINES The original image either a single or multi layered real valued image The width and height of th
70. a with complete homogeneity within each region classification can be carried out by averaging the intensity within each ground truth region This represents what we shall term the Upper Limit of classification capability However this method can only classify images with known regions it has no ability to deal with raw imagery Furthermore in many situations the contents of the regions in an image whether it s region boundaries are known or not are not uniform Crops for example grow at different rates in different parts of a field and or may not actually fill the specified region Segmentation provides a tool for deriving homogeneous segments within the data based on the information within the data The intensity within each segment then represents the most reliable information about the cross section within that region Inhomogeneities arising from any cause will be detected In this study we compare the results of several possible approaches to the classification of regions including segmentation To ensure that we have ground truth to compare the results obtained we use a simulated three look SAR image containing a known pattern of fields This is representative of the SAR imagery obtained with the ERS1 and ERS satellites Ground truth is available for this simulated image and is divided into a training and test set illustrated in Figures 6 1 1 a and 6 1 1 b 29 Mar 10 NASoftware Ltd 55 CHAPTER 6 CASE STUDIES Figure 6 1
71. age tostring Figure 5 2 The IDLE Editor Window For those who need more control over their application they can access the InfoPACK com mands directly from Python The names of these scripts often end with a py These can be prepared using any text editor but InfoPACK also provides access to IDLE This Open Source IDE provides a syntax sensitive text editor and Python shell which is ideal for rapid development of Python scripts The editor is described in section 5 7 Python scripts are typically chains of InfoPACK commands but may contain any Python or Numeric Python commands Full documentation is accessible from the He 1p button 5 7 The IDLE Integrated Development Environment IDLE consists of the following integrated modules e Python Shell Window gives you access to the Python interactive mode e File Editor lets you create new or browse through and edit existing Python source files e Path Browser for searching through the path of available module source files e Class Browser for finding the methods of classes e Debugger for the symbolic debugging of Python programs Documentation for IDLE is available at http python org idle doc 29 Mar 10 NASoftware Ltd 53 CHAPTER 5 USING THE INFOPACK GUI 54 Python 2 2 1 Apr 12 2002 15 29 57 GCC 2 96 20000731 Red Hat Linux 7 2 2 96 109 on linuxZ Type copyright credits or license for more information IDLE 0 8 press F1 for help gt g
72. and gt 1 lt SourceBand gt lt SrcRect xOff 0 yOff 0 xSize 512 ySize 512 gt lt DstRect xOff 0 yO0ff 0 xSize 512 ySize 512 gt lt SimpleSource gt lt VRTRasterBand gt lt VRIDataset gt 3 3 3 File Import Utility If the built in import facilities do not cover your file format you can use the general conversion rou tine raw2nc py provided to convert your images into NetCDF format This routine is described in detail in Section B 3 16 There is also an export routine nc2raw py To use raw2nc py you will need to specify In The number of image layers default 1 tC the type of data in the raster file There are eight recognised types option t option t 1 option t s option t 1 option t d option t f option t F option t D b Byteswap default integers signed Int0 Int8 signed 2 byte integers Int16 signed 4 byte integers Int32 signed Float Float64 Float0 Float8 Float16 Float32 Complex0 Complex8 Complex 16 Complex32 Complex Complex64 Depending on how your images are constructed it is possible that you will need to reverse the order of the bytes To see whether this is the case we suggest you load the converted image into the viewer and use the Image Statistics menu button to check that the values it reports look sensible if not you probably need to swap the byte order If you do need to swap the byte order include the b flag option s reverse t
73. ap the algorithm takes the intersection of the given layers to be the region map on which to perform the change detection image nc This image must be a multiple layered i e more than 1 layer real or complex valued image output nc The output file The output is real valued single layer image with dimensionality equal to the first layer of the original image Pixels are either coloured with value 1 indicating that they belong to regions for which there has been significant change or with value O that they don t pfa float default 6 Specifies the Probability of False Alarm 10 if pfa gt 1 to be used 29 Mar 10 NASoftware Ltd 135 APPENDIX A InfoPACK HIGH LEVEL ROUTINE DOCUMENTATION in determining whether changes are significant The larger pfa is the less chance there will be of reporting insignificant changes pfa float default 0 If this is non zero threshold the likelihood directly rather than calculating the probability The pfa option will be ignored looks float default 1 The looks measure of the image data If looks is negative use 1 the routine will compute this parameter from the data intensity integer default 0 Assume the data is amplitude square root intensity mtpol integer default 0 Polarimetric data is stored as layers of complex data however this is the same file format as multi temporal polarimetric data To distinguish the two the mt pol parameter is used For instance 5 days
74. are written to a system temporary directory Further Comments This routine is actually a wrapper to the filter function of Section A 4 11 B 2 25 sqrt Purpose The result is the square root of individual pixel values from the input image Command Line Script sqrt py image_in image_out tilesize tmpdir Python Procedure array_out sqrt array_in Parameters tilesize integer default 512 A tilesize of 512 x 512 is used tmpdir directory default None Intermediate results are written to a system temporary directory B 2 26 square Purpose The result is the square of individual pixel values from the input image Command Line Script square py image_in image_out tilesize tmpdir 29 Mar 10 NASoftware Ltd 163 APPENDIX B LOW LEVEL AND UTILITY ROUTINE DOCUMENTATION Python Procedure array_out square array_in Parameters tilesize integer default 512 A tilesize of 512 x 512 is used tmpdir directory default None Intermediate results are written to a system temporary directory B 2 27 subtract Purpose Subtracts two image files The result is the subtraction of corresponding pixel values of the input images Command Line Script subtract py imagel_in image2_in image_out tilesize tmpdir Python Procedure array_out Numeric subtract arrayl_in array2_in Parameters tilesize integer default 512 A tilesize of 512 x 512 is used tmpdir directory default None Intermedi
75. aries you can call InfoPACK routines directly By using the library you can use your own file formats but since these routines operate on arrays in memory none of the tiling code will be available C 1 Sample Code In the library directory there is a sample C program segment c which shows how to call the three different segmenation routines in InfoPACK and link in the library libinfolib a Under linux use gcc I segment c L linfolib lm where 1m is the mathematics library 1 says the header file infopack h is in the current directory and L says the library 1ibinfolib a is also in the current directory C 2 Execution To use the library you need to set the environment variable InfoPACK to point to the directory where the licence file infopack 1ic is For instance if the licence file was in the same directory as the executable in DOS you could use set InfoPACK or in bash export InfoPACK This may have already been done when InfoPACK was installed If it works when you run the executable you should get Segment Polsegment 194 NASoftware Ltd 29 Mar 10 C 3 C INTERFACE If you get the message Segment segment failed with error 5 then the environment variable InfoPACK has not been set properly C 3 C interface All of the available C function calls are listed in the header file infopack h The routines accept natural arguments int double float pointers so that there will n
76. ata one would expect to need this parameter to be increased Smoother data such as that encountered on averaging speckle eg via multi looking should allow a reduced value of the stopping parameter 3 The second parameter in the annealing method is the cooling parameter which controls the rate at which the optimisation converges on the final solution If this is made too large the solution may freeze out away from the optimal result 29 Mar 10 NASoftware Ltd 21 CHAPTER 4 IMAGE PROCESSING WITH InfoPACK Figure 4 1 Simulated 3 look SAR data The shape parameter is most fundamentally related to the quality of the reconstruction since 1t controls the smoothness of the segment boundaries which affects the detail that can be retained in the reconstruction in the presence of speckle noise This parameter is the one that has the most direct relationship with the properties of the original image An image with considerable small scale structure generally requires a large value of the shape parameter whereas a scene with little structure 1s best suited to a small value In Figures 4 2 to 4 4 we illustrate the effect of the shape parameter on the result of segmenting Figure 4 1 From Figure 4 2 we observe that small details of scene structure are visible with a larger param eter value However there are also a number of small flecks in the reconstruction that are associated with speckle noise This is known because the simulated images had uni
77. ate results are written to a system temporary directory B 2 28 sum Purpose Adds two or more images The result is the sum of corresponding pixel values of the input images Command Line Script sum py imagel_in image2_in imagen_in image_out tilesize tmpdir 164 NASoftware Ltd 29 Mar 10 B 2 IMAGE MANIPULATION ROUTINES Python Procedure array_out listsum array in image2_in arrayn_in Parameters tilesize integer default 512 A tilesize of 512 x 512 is used tmpdir directory default None Intermediate results are written to a system temporary directory Further Comments The dimensions layers width height of the input images must be the same The output image will have the same dimensions B 2 29 threshold Purpose Image pixels that lie above the threshold value are set to a specified true value otherwise they are set to a false value The threshold value must be given Command Line Script threshold py image_in image_out threshold true false tilesize tmpdir Python Procedure array_out threshold array_in threshold true false Parameters threshold float must be specified true float default 1 false float default 0 tilesize integer default 512 A tilesize of 512 x 512 is used tmpdir directory default None Intermediate results are written to a system temporary directory 29 Mar 10 NASoftware Ltd 165 APPENDIX B LOW LEVEL AND UTILI
78. ath The appearance of linear features such as walls hedges and ocean waves will depend on their orientation relative to the radar beam If buildings are aligned to the beam it is possible to get a double bounce of the radar once off the building and once off the ground This produces very bright pixels in the image To a lesser degree this also happens from tree trunks along the edge of a wood These effects are also present in satellite SAR imagery 2 2 The Formation of SAR Images A SAR system illuminates a scene with microwaves and records both the amplitude and the phase of the back scattered radiation making it a coherent imaging process An inherent feature of SAR imaging is speckle which arises because each resolution cell contains many scatterers the phases of the return signals from these scatterers are randomly distributed and speckle is caused by the resulting interference The received signal is sampled and converted into a digital image Note that the pixel spacing in the final image and the resolution of the system may not be the same If a system with a resolution of 5m is set to oversample the signal and produces an image in which each pixel corresponds to 3m you do not get a more detailed image the information is just smeared out over several pixels In this case the image is correlated and is much more difficult to process The field recorded at pixel x denoted E x can be written E x d a s exp id s
79. ation of a c ratio of a to b d corresponding ratio for original data from Fig ure 6 5 2 c 6 5 5 Segmentation Conclusions In summary this study has demonstrated the following aspects of applying segmentation to corre lated data e If segmentation is applied directly to SAR intensity images that have been sampled at spacing less than the speckle correlation width which is determined by the resolution of the system the resulting segmentation is dominated by speckle sized segments e The ratio of the original intensity to the segmented version does not correspond to the ex pected speckle both in apparent ENL and the speckle correlation properties Only residual correlations on a scale of one pixel remain 29 Mar 10 NASoftware Ltd 93 CHAPTER 6 CASE STUDIES 94 If the SLC image is available the speckle correlations can be evaluated directly from the complex field ACF The resampling factors then correspond to the lag where the field ACF falls to 20 of its value at zero lag Similar information can be derived from the intensity ACF though more indirectly The component due to the instrumental response can be separated from that due to the underly ing cross section fluctuations The resampling factors then correspond to the lags where the instrumental part of the ACF falls to 4 of its value at zero lag Since the resample factor needs only be accurate to the nearest pixel it is often adequate to identify where
80. ation of each layer to carry out their analysis and this must be provided too In most cases the joint segmentation produced by segmenting the multi layer image is what will be conveniently used but if for example you are interested in changes in field boundaries then segment each layer separately and merge the individual segmentations into a multi layer image before calling the change detection routine You would use Change_detection if you have an unknown sequence of several changes The algorithm basically detects whether the mean remains constant or varies throughout the sequence Change position detection introduces further prior knowledge namely that there is one change between constant levels Since the prior knowledge is stronger the result is better If however you try to use it where there are several changes such as in agricultural monitoring then the results will not be of use 4 3 1 The change_detection routine change_detection looks for significant change over a series of original speckled images The routine detects changes across the whole series of provided image layers the ordering of the layers in the multiple layer image file is not important The output is a single layer image specifying where any significant changes were found over the whole sequence If you are more interested in the individual changes found between adjacent pairs of layers in the input image use routine change_detection_position Figure 4 10
81. ayer Operations This provides access to operations for generating a 3 D layered image from a set of single layer images and for extracting individual layers from a 3 D image Two options are provided e Split Layers Splits a 3 D image into a numbered set of the individual layers e Combine Layers Combine a set of individual files into a single 3 D image Full function specifications are given in Appendix B Extract Sub Image Extract a subimage from a given image You will be prompted for the subimage size width and height and for the w width and h height offsets Full function specification is given in Ap pendix B Resize Image Resize an existing image You will be prompted for the new image size width and height and for the method of interpolation Full function specification is given in Appendix B 44 NASoftware Ltd 29 Mar 10 5 3 THE MAIN InfoPACK MENUS Histogram Image Produce a histogram showing the distribution of amplitudes intensities of an image For multi layered images the histogram for individual layers can be viewed as well as for the whole image data This utility can be useful when planning to carry out supervised classification of an image Full function specification is given in Appendix B Image Statistics Provides a set of statistics on the specified image e Image size and datatype e Minimum and maximum pixel values e Mean and standard deviation of pixel values e The looks meas
82. be prompted for the file to be imported 29 Mar 10 NASoftware Ltd 15 CHAPTER 3 IMAGE FILE FORMATS Long Format Name Code Creation Georeferencing Arc Info ASCII Grid AAIGrid Yes Yes Arc Info Binary Grid adf AIG No Yes Microsoft Windows Device Independent Bitmap bmp BMP Yes Yes BSB Nautical Chart Format kap BSB No Yes CEOS Spot for instance CEOS No No First Generation USGS DOQ doq DOQ1 No Yes New Labelled USGS DOQ doq DOQ2 No Yes Military Elevation Data dt0 dt1 DTED No Yes Eosat Fast Format EFF No No ERMapper Compressed Wavelets ecw ECW Yes Yes ESRI hdr Labelled EHdr No Yes ENVI hdr Labelled Raster ENVI No Yes Envisat Image Product n1 Envisat No No EOSAT FAST Format FAST No Yes FITS fits FITS Yes No Graphics Interchange Format gif GIF Yes No Arc Info Binary Grid adf GIO Yes Yes GRASS Rasters GRASS No Yes TIFF GeoTIFF tif GTiff Yes Yes Hierarchical Data Format Release 4 HDF4 HDF4 Yes Yes partially Erdas Imagine img HFA Yes Yes NASoftware NetCDF Images nc NASoftware Yes Yes Atlantis MFF2e HKV Yes Yes Japanese DEM mem JDEM No Yes JPEG JFIF jpg JPEG Yes Yes JPEG2000 jp2 j2k JPEG Yes Yes NOAA Polar Orbiter Level 1b Data Set AVHRR LIB No Yes Atlantis MFF MFF Yes Yes NITF NITF Yes Yes OGDI Bridge OGDI No Yes PCI aux Labelled PAux Yes No Portable Network Graphics png PNG Yes No Netpbm ppm pgm PNM Yes No USGS SDTS DEM CATD DDF SDTS No Yes SAR CEOS SAR_CEO
83. be trusted However it must be noted that agricultural regions such as fields are very unlikely to be completely homogeneous The Upper Limit offers no capability for detecting these inhomogeneities Furthermore they will tend to de grade so called Upper Limit results whereas they will have much less impact on joint segmentation Perhaps most important is the fact that depending on existing edge maps excludes any possibility of extrapolating beyond available map data With segmentation extrapolation is automatic Finally it is instructive to compare the detection probabilities in the two methods of change detection From this comparison one would deduce that the detection of change together with its position is more sensitive than the simple change detection process Since positional change detection offers additional information as well as being more sensitive this is the technique that should be preferred 6 3 3 Conclusions e Joint segmentation offers a valuable precursor to change detection because it adapts the region size to the data and presents a common set of boundaries for the whole sequence of images e The Upper Limit provides the most sensitive method for detecting changes However it suffers from three disadvantages It cannot detect inhomogeneities within the defined regions The sensitivity is degraded by any inhomogeneities It only process within defined regions with no extrapolation capability such as that po
84. bsite http www nasoftware co uk Bibliography 1 Speckle reduction 40 I McConnell R G White C J Oliver and R Cook Radar cross section estimation of SAR images Proc EUROPTO Paris 1995 SPIE Vol 2584 Synthetic Aperture Radar and Passive Microwave Sensing Edge detection C J Oliver D Blacknell I McConnell and R G White Optimum edge detection in SAR Proc EUROPTO Paris 1995 SPIE Vol 2584 Synthetic Aperture Radar and Passive Mi crowave Sensing Segmentation texseg C J Oliver I McConnell and D Stewart Optimum texture segmentation of SAR clutter Proc EUSAR Conference K nigswinter 1996 Segmentation segment D Stewart D Blacknell A Blake R Cook and C J Oliver An optimal approach to SAR image segmentation and classification IEE Proc Radar Sonar Nav 2000 NASoftware Ltd 29 Mar 10 Chapter 5 Using the InfoPACK GUI 5 1 Introduction The InfoPACK GUI provides easy access to the InfoPACK Image Understanding Library by means of a set of menus and dialog boxes Routines are run within the main Python Shell Window Scripts Python commands and InfoPACK functions can also be run from within the shell window Also supplied is the Open Source Python Integrated Development Environment IDLE which provides a language sensitive editor and debugger for rapid development of Python scripts and complete processing chains 5 2 Starting The InfoPACK GUI 5 2 1 Windows Navigate
85. calculated The stats procedure returns a class containing the results The script version prints the results to screen Command Line Script stats py image_in whole intensity width height cmpl2real calctex tilesize overlap tmpdir Python Procedure s stats array_in whole intensity width height cmpl2real calctex print s 182 NASoftware Ltd 29 Mar 10 B 4 IMAGE INFORMATION ROUTINES print Minimum s min print Maximum s max print Mean s mean print SD s3sdai print Looks Measure s enl print Order s order Parameters whole integer default 0 If whole is 1 then the looks measure and order parameter are calculated for the whole image rather than as the average of local estimates intensity integer default 0 Assume the data is amplitude square root intensity width integer default 5 Window width size for calculating local estimates of the looks measure and order parameters height integer default 5 Window height size for calculating local estimates of the looks measure and order parameters cmpl2real string default abs Chooses from a set of methods for converting the complex image to a real one abs absolute value mag magnitude of complex number abs x abs phase Phase of complex number real Real part imag Imaginary part calctex integer default 1 Ca
86. can be calculated so it is possible to judge whether the transfer of a pixel improves the segmentation The calculation also takes into account the straightness of the region boundaries the amount of crinkle allowed is set by the user This process of transferring pixels is governed by simulated annealing and it will eventually find the global optimal solution for the segmentation of the image into a specified number of regions An optional post processing stage is available routine merge which merges regions which are statistically similar Full Python script and function syntaxes are given in section A 4 1 An Example Segmentation Figure 4 1 shows a simulated 3 look SAR image this image is included with the InfoPACK distri bution in the images directory An important issue with applying segmentation is the selection of operating parameters There are three main ones that control the process 1 The shape parameter attaches a penalty to excessive curvature in the reconstruction Large values of the parameter allow many changes of direction of the edge Small values bias the reconstruction towards smoother edges Since speckle yields very crinkled edges to regions of similar intensity it is important to adopt a large enough curvature penalty i e a small enough shape parameter 2 The optimisation method has two further parameters attached The first is the stopping pa rameter to assess when the reconstruction is complete For noisier d
87. car edu packages netcdf file format NetCDF is a very flexible format that would allow something as simple as a single image to be described in many different ways It would not be practicable for InfoPACK to be able to interpret all of these different descriptions and so we have defined a simplified format based on NetCDF known as the NASoftware format The NASoftware format allows for a single image to be stored in a file along with geographic GIS information This single image may be comprised of several layers of images but the type e g short float complex of each layer must be the same A NetCDF file is divided into four sections dimensions variables attributes and data The properties of the first three of these will now be described D 1 Dimensions The NASoftware Ltd convention uses the following dimension names e num_rows e num_cols e num_cols_complex NetCDF does not provide direct support for complex data types To get around this the names of the dimensions are used to indicate whether the data are real or complex If both the columns dimensions are present they must satisfy num_cols_complex 2 x num_cols D 2 Variables A real valued array is declared as variable_name num_rows num_cols 29 Mar 10 NASoftware Ltd 197 APPENDIX D NASoftware CONVENTIONS FOR NetCDF The variable can have any of the standard NetCDF types byte short int float or double Multiple layers can be stored in a single
88. caused by speckle noise The average probability 29 Mar 10 NASoftware Ltd 79 CHAPTER 6 CASE STUDIES Figure 6 11 a Ground truth training masks for the simulated data The classes are represented by 0 magenta 1 blue 2 cyan 3 green 4 yellow 5 red Figure 6 12 b Ground truth test masks for the simulated data 80 NASoftware Ltd 29 Mar 10 6 4 TEMPORAL SIGNATURES OF INTENSITY IMAGES Figure 6 4 2 a Classification using single pixels of correct classification is 0 899 After averaging over a 3x3 window the result in b shows that these spots have been averaged out However the correlations introduced by the filter leave a ring of misclassified regions around each large area Thus although the probability of correct classifi cation is reasonable within large areas of homogeneous RCS the average for this test scene 0 954 is not as good as one might expect Indeed an image with more structure would yield an even worse result The result after joint segmentation c is considerably better with all classes being separated including most of the narrow sections between large areas The average probability of correct classification has risen considerably to 0 992 Finally the Upper Limit yields essentially perfect classification Figure 4 Classified results from a single pixels b 3x3 window c joint segmentation d Upper Limit Table 6 4 2 Average probability of correct classification for t
89. cedure stats_class class_stats maskarray_in imagearray_in cmpl2real intensity verbose Parameters intensity integer default 0 Assume the data is amplitude square root intensity cmpl2real string default abs Chooses from a set of methods for converting the complex image to a real one abs absolute value mag magnitude of complex number abs x abs phase Phase of complex number real Real part imag Imaginary part verbose integer default O If set to 1 prints out the class statistics tilesize integer default 512 A tilesize of 512 x 512 is used overlap integer default 48 To prevent obvious joins between tiles an overlap of 48 is used tmpdir directory default None Intermediate results are written to a system temporary directory Further Comments Regions labelled 1 in the mask file are assigned to class 0 regions labelled 2 are assigned to class 1 and so on Regions labelled 0 are taken as the background and are not assigned to a class See also mask_stats B 5 3 classmap2layers Purpose Convert a classification image with regions labelled 0 1 2 to layers of 0 1 masks Hence the pixels labelled 2 will be marked with a 1 in layer 2 of the output image Command Line Script classmap2layers py image_in image_out tilesize tmpdir Python Procedure array_out classmap2layers array_in 29 Mar 10 NASoftware Ltd 187 APPENDIX B LOW LEVEL AND UTILIT
90. classification with regions labelled 0 1 2 to a colour image for easier interpretation Command Line Script colclass py image_in image_out format tilesize tmpdir Python Procedure array_out colclass array_in format Parameters format string default NASoftware Ltd NetCDF The other useful format is GTi ff tilesize integer default 512 A tilesize of 512 x 512 is used tmpdir directory default None Intermediate results are written to a system temporary directory B 5 2 classstats Purpose Calculates the mean and order parameters for the masked regions of the original image The mean and order parameters can then be used as supervised inputs for the classify routine The masked image is a single layer real valued image with the region labels corresponding to the training data for that class The class means should be a single layered real valued image of height equal to the number of layers The number of classes is then assumed to be equal to the width of the given image The class orders should be a single layered real valued image of height equal to the number of layers The number of classes is then equal to the width of the class orders image Command Line Script classstats py maskimage_in image_in meanimage_out orderimage_out cmpl12real intensity verbose tilesize overlap tmpdir 186 NASoftware Ltd 29 Mar 10 B 5 CLASSIFICATION ROUTINES Python Pro
91. d be merged and this is what merge does The result of applying this merge process to the data of Figure 6 17 and the segmentation of Figure 6 18 is shown in Figure 6 21 There is no appreciable difference between this and the orig inal segmentation in Figure 6 18 indicating that the many small regions visible are very unlikely to be a consequence of speckle fluctuations and hence are indeed significant within the cartoon segmentation model used 6 7 2 Despeckling Another method for removing speckle is to use the despeckling algorithm implemented in despeckle Rather than impose the cartoon model of uniform RCS within each segment this implies a local similarity over a neighbourhood of 3x3 pixels It identifies which possible arrangement of 3 cor related pixels out of the 12 possible is most consistent with the data and speckle It too depends on simulated annealing to determine the overall optimum despeckled solution The consequence of applying this algorithm to the data of Figure 6 17 is shown in Figure 6 22 Since despeckle does not impose uniform regions with consequent sharp discontinuities edges the result appears to yield a gently varying background This is likely to be more consistent with the actual real data than the cartoon model One might deduce that this would be a better algorithm to use for image interpretation However the main weakness of this method is that it merely generates a continu ously varying intensity largely f
92. d by entering 29 Mar 10 NASoftware Ltd 117 APPENDIX A InfoPACK HIGH LEVEL ROUTINE DOCUMENTATION Figure A 1 InfoPACK naming conventions for image dimensions pydoc infopack routine_namel on the command line A 2 Common Parameters Some parameters are common to all InfoPACK routines and are listed here A 2 1 Image dimensions Figure A 1 shows the InfoPACK naming conventions as regards width height wof fset and hoffset A 2 2 The Intensity Flag Datasets for SAR images can be stored as either intensity modulus or amplitude squared or am plitude square root intensity data It is important that you know what type of data you are using Many images are stored as amplitude since the speckle is less obtrusive when these images are dis played InfoPACK routines assume by default that images are stored as amplitude If they are not then the flag intensity 1 needs to be set in the function or script call in order to indicate the image is in fact intensity data 118 NASoftware Ltd 29 Mar 10 A 3 DOCUMENTATION CONVENTIONS Internally InfoPACK routines work on intensity data If a routine is given amplitude data it will typically convert the data to intensity data process the data and then convert the data back to amplitude It is always worth checking the individual function specifications for exact behaviour though For example col seg run on amplitude data will return the square root of regional intensity mean values
93. e or a complex po larised image or a sequence eg a multi temporal set of images Ideally a segmentation algorithm would divide up an image into regions each of which represented a feature of interest a wood town or field for example In practice they respond to qualities in the image such as variations in intensity or texture which may or may not correspond with your areas of interest The segment segmentation algorithm produces cartoon images it looks for regions of con stant underlying RCS allowing for speckle and produces an output image in which each region in the resulting segmentation is coloured with the mean of the data values in that region or determinant values in the case of polarimetric images Although the data values in a cartoon form a discrete set a segmentation is not the same as a classification the number of regions may be in the thousands but these regions could normally be classified into a small number of categories The segment algorithm accepts one or more images multiple images are used together to produce a jointly optimal set of segments The algorithm starts with a rectangular segmentation of the image s The initial size of the rectangles is determined by the user setting the average region size in pixels overall the default start with single pixels is suitable Pixels located at the edges of regions are transferred from one to another The likelihood that the image data fits a given segmentation
94. e original and masked images must be identical though they can differ in the number of layers The class means should be a single layered real valued image of height equal to the number of layers The number of classes is then assumed to be equal to the width of the given image The class orders should be a single layered real valued image of height equal to the number of layers The number of classes is then equal to the width of the class orders image Command Line Script mask_stats py maskimage_in image_in mean_out order_out intensity cmpl2real verbose tilesize overlap tmpdir Python Procedure stats_class mask_stats maskarray_in array_in intensity cmpl2real verbose Parameters intensity integer default 0 Assume the data is amplitude square root intensity cmpl2real string default abs Chooses from a set of methods for converting the complex image to a real one abs absolute value mag magnitude of complex number abs x abs phase Phase of complex number real Real part imag Imaginary part verbose integer default O Tf set to 1 prints out the class statistics tilesize integer default 512 A tilesize of 512 x 512 is used overlap integer default 48 To prevent obvious joins between tiles an overlap of 48 is used tmpdir directory default None Intermediate results are written to a system temporary directory Further Comments See a
95. e SAR instruments Whenever more than one polarisation is available it is likely that improved results can be achieved by processing both all polarisations jointly 29 Mar 10 NASoftware Ltd 9 CHAPTER 2 SYNTHETIC APERTURE RADAR IMAGES 2 5 The Number of Looks in an Image The number of looks in an intensity image is a measure of the statistical fluctuations introduced by speckle resulting from the interference between randomly positioned scatterers It is purely a property of the imaging process and depends on the number of independent speckle estimates that have been combined to form the final image For example ERS PRI data is described as 3 look This means that the final image has been formed by adding the intensities from three independent single look complex SLC components To reduce speckle levels even further many people sacrifice the resolution of a single image and produce a smaller one in which each pixel is formed by averaging a clump of four pixels in the original say If the image is uncorrelated this process increases the number of looks by a factor of four Unfortunately there is usually little redundant information in a SAR image unlike the situation in optical images because objects such as trees vehicles and buildings typically occupy only at most one or two pixels for currently available or envisaged satellite SAR systems The interaction between the ENL the pixel spacing and the resolution must be understood L
96. e methods tend to blur edges and other small features reducing the sharpness of the detail and hence the usefulness of the processed image The algorithms provided in InfoPACK use the known statistical properties of SAR data derived from a theoretical analysis of the image formation process By using a physically correct model they are able to preserve the detail and resolution of the original image while producing a smooth real valued output All the speckle reduction algorithms in InfoPACK work in a similar way they solve an opti misation problem to find the best fit of a given model to the image data They differ in the kind of model which is fitted or the kind of data single or multi dimensional images accepted The most common need is not only to remove speckle it is to remove it in order to split the image into homogeneous regions segmentation or classes classification Segmentation yields a number of regions consistent with the statistical uncertainty the con straint and the data model Classification is a much simplified representation in terms of useful categories such as field with crop Segmentation can thus provide a wealth of information about how the measured parameter varies within a single class Here we describe how to use the main segmentation facilities provided 20 NASoftware Ltd 29 Mar 10 4 1 SEGMENTATION 4 1 1 The segment routine The routine segment segments a single intensity amplitude or complex imag
97. e region boundaries are known It cannot extrapolate beyond the ground truth Segmentation on the other hand can use the training set to enable it to classify over as large an image as is required Thus the overall conclusion is that segmentation provides an essential precursor to polarised classification 29 Mar 10 NASoftware Ltd 115 Appendix A InfoPACK High Level Routine Documentation In this chapter we provide full specification for all the major high level InfoPACK routines A 1 User Interface A 1 1 Scripts and Functions As described in section 1 1 InfoPACK routines can be accessed as Python functions Python scripts or via the GUI The latter is simply a front end for the Python scripts The distinction between the two fundamental ways of calling the routines is as follows Python Functions the routine is called as a Python function The functions are contained in the Python module infopack and have a user interface designed to be as convenient as possible to use from within the InfoPACK or Python console Operations on images are in general carried out in memory and so do not provide a suitable environment for handling arbitrarily large images Python Scripts py files These are designed for users who prefer to use a command line inter face with or without Unix shell scripts Windows batch files Python scripts work file to file and are memory optimised in order to handle arbitrarily large images efficiently Th
98. e resolution is 4 m the azimuth resolution is 1 8 m the pixel spacing is 1 25 m in range and 0 68 m in azimuth The effec tive number of looks ENL is specified to be 4 0 Based on this information we note that the image is over sampled with 3 2 samples per resolution length in range and 2 6 in azimuth The effect of this oversampling is to make the speckle which reflects the coherent interference between random scatterers within each resolution cell subtend approximately 3x3 pixels However as noted previously the segmentation process implicitly assumes that neighbouring pixels are not correlated Speckle correlation will therefore impact on the quality of the segmenta tion reconstruction In this case study we demonstrate the effect of such correlations on segmenta tion and despeckling and show how the problem can be overcome 29 Mar 10 NASoftware Ltd 85 CHAPTER 6 CASE STUDIES 6 5 2 Correlation and Segmentation Previous studies have determined suitable sets of parameters for intensity segmentation In Figure 6 5 2 a we show a detail from Figure 6 13 together with its segmentation using default parameters in Figure 6 5 2 b It is immediately apparent that the segmented image has far too many segments which appear to follow the speckle fluctuations This is indeed precisely the effect that correlated speckle has on segmentation A region of high speckle intensity subtending several pixels is re constructed as a single segment
99. e scripts have a very similar syntax to the function form so that switching between them should you need to do this is straightforward Here is an example Command line call segment py image_file output_file quality sensitive InfoPACK Python function array_out segment array_in quality sensitive In the latter call array_in is the previously loaded image array the result will be stored in the Numeric array array_out Section A 1 3 discusses file I O for InfoPACK functions in greater detail 116 NASoftware Ltd 29 Mar 10 A l USER INTERFACE A 1 2 GIS Information InfoPACK import facilities preserve any GIS information in the original file and the imported NetCDF files can be loaded processed and saved while continuing to preserve the information Alternatively you can ignore it totally if it is not relevant to your application The way in which this is done is outlined in the next section Section A 1 3 A 1 3 File VO File VO is handled behind the scenes when using the Python Script versions of the routines Python Function calls however work via calls to and from the Python Numeric arrays associated with each image file Loading and saving of NetCDF files is done by calling the utilities load save The load function returns a list with two elements in it the first is an array containing the image in Numeric Python format and the second is a GIS object For example from the console a gis load file img re
100. e shall call this method the High Limit since it would be expected to be lower than the Upper Limit but since it is a maximum likelihood algorithm higher than joint segmentation In each case we calculate the likelihood that each region is a member of a gamma distributed PDF of defined mean depending on the class The region is then assigned to that class which has the maximum likelihood We assume for this comparison that the order parameter determined by the speckle is uniform across the image as before The classification results are shown in Figure 6 2 2 Classification using a single pixel shown in Figure 6 2 2 a is still very poor Indeed the average probability of correctly classifying each category is only 0 673 and the associated confusion matrix shown in Table 6 2 2 a reveals major ambiguities in assigning the intermediate classes Classification over a 3x3 window shown in Figure 6 2 2 b shows less improvement than might be expected with the average fraction of correctly classified pixels in each class only 0 707 The associated confusion matrix in Table 6 2 2 b again reveals considerable spread in the intermediate classes Closer study of Figure 6 2 2 b suggests that This latter expectation is not realised we see below that joint segmentation provides the same performance which is why there is not a specific routine for this case in InfoLIB 62 NASoftware Ltd 29 Mar 10 6 2 SEGMENTATION AND CLASSIFICATION OF MULTIPLE
101. e speckle model 104 NASoftware Ltd 29 Mar 10 6 8 SEGMENTATION AND CLASSIFICATION OF POLARIMETRIC DATA 6 8 Segmentation and Classification of Polarimetric Data 6 8 1 Introduction In many remote sensing applications important additional information can be derived from multipo larised imagery The different polarisations respond in different ways to the orientation and shape of the objects from which scattering takes place It is obviously important to have tools that make full use of this information and in this Case Study we look at the steps needed As previously we simulate data of the required type in this case HH HV and VV complex imagery The underlying class structure in the simulated image is shown in Figure 6 25 a The specified ground truth is illustrated in Figure 6 25 b while the training and test masks are shown in Figures 6 25 c and d respectively In order to simulate polarised imagery we specify the class covariance matrix for each of the six classes shown in Figure 6 25 The covariance matrix has the form EmuEBgg EngEpy EnukEyy EmvEgp EnvEpy EnmvEyy EvvEyg EvvEqy EvvEyy The values of the elements for each class are listed in Table 6 8 1 Note that the element 0 0 is identical for each class This means that the HH image will show no structure l 140 Oj Oj Oj 1 0j Oj Oj Oj 0 1 0j 2 140j 0 68 0j 0 21 0j 0 68 0j 1 8 0j 0 12 0j 0 21 0j 0 12 0 01j 0 15 0j 3 140j 0 43 0 06 0 1 0 2j 0 43 0 06 1 2 0j
102. e speckle noise One would therefore expect this to be reflected in improved segmentation and classification performance To demonstrate this we generate another 9 simulated images with the same underlying segments as those introduced in the previous study These then differ only in terms of speckle they simulate a set of images obtained by repeated passes of a SAR over the same scene A similar analysis to that performed previously can be carried out on this set of intensity images The ground truth training and test masks are identical to the previous analysis 6 2 2 Comparison of classification methods We now compare classification of the test image using the same methods as before e Upper Limit classifying the average intensity in each ground truth region e Joint Segmentation classifying the average intensity in each segment e 3x3 window classifying the average intensity over a 3x3 window e Lower Limit classifying the intensity in each pixel Note that the averages are now taken over each pixel within each segment and over all im ages in the sequence However there is another option available for this particular case where the boundaries and underlying cross section is identical in each image The Maximum Likelihood re construction is obtained by averaging all the intensities i e by summing the images in the first place Only a single image is then needed for segmentation rather than joint segmentation over the complete set W
103. ed Command Line Script fill_unirand py image_out width height covar seedl seed2 tilesize tmpdir Python Procedure array out fill polrand width height covar seedl seed2 Parameters covar The covariance matrix can either be specified on the command line For instance covar 1 0 0 0 1 33 0 0 0 0 1 23 Note no spaces around commas or the covariance matrix can be in a file as with covar covmat nc For a 3 x 3 covariance matrix the output image with have 3 complex layers seed1 integer default 0 seed2 integer default 0 tilesize integer default 512 A tilesize of 512 x 512 is used tmpdir directory default None Intermediate results are written to a system temporary directory B 2 11 georef Purpose Returns a geocoded image by shifting pixels according to x and y deltas These deltas must be stored as short 16 bit and the output image will be the same size as the two delta images Command Line Script polyscan py image_in image out xaverage yaverage intensity blank xoffset yoffset tilesize tmpdir Python Procedure array_out georef array_in xaverage yaverage intensity blank xoffset yoffset 154 NASoftware Ltd 29 Mar 10 B 2 IMAGE MANIPULATION ROUTINES Parameters xaverage integer default 1 Averages this many pixels horizontally yaverage integer default 1 Averages this many pixels vertically blank float default 0 Va
104. ed around the comma or in python min 0 3 7 2 max list default None As min this can be a single value or a list of values verbose integer default 0 If set to 1 then the estimated min and max values are printed out tilesize integer default 512 A tilesize of 512 x 512 is used tmpdir directory default None Intermediate results are written to a system temporary directory B 3 19 totiff Purpose Converts image data to a TIFF bitmap suitable for display in common graphics packages For each layer of the image values that lies in the range min to max are mapped into the range O to 255 If min or max are not specified then the parameter sd is used with the following mapping p sdxo 0 p sdxo gt 255 where yu is the mean and a is the variance of the data A value of sd 0 has the special meaning of minimum 0 maximum 255 Complex data cannot be directly displayed as a bitmap and so must be converted to real data first This is done by the option cmp12real 178 NASoftware Ltd 29 Mar 10 B 3 FILE HANDLING FACILITIES Command Line Script totiff py image_in image_out tif cmpl2real sd min max verbose tilesize tmpdir Python Procedure S procedure makebitmap Parameters cmpl2real string default abs Chooses from a set of methods for converting the complex image to a real one abs absolute value mag magnitude of complex number abs x abs phase Phase of com
105. ed by segment can in theory be of any statistical significance merge provides a method to ensure that only highly significant edges remain The merge routine can process layered images Full Python script and function syntaxes are given in section A 4 4 29 Mar 10 NASoftware Ltd 25 CHAPTER 4 IMAGE PROCESSING WITH InfoPACK Figure 4 6 a Figure 4 6 b Figure 4 6 a SAR image of rain forest b Segmentation using texseg 4 1 4 The texseg routine Texture segmentation involves partitioning of an image into regions each of which contains a single texture distinct from its neighbours It works in an analogous fashion to segment the difference being that instead of segmenting upon the measure of regional mean intensity texseg segments upon the Blacknell Tough measure TI log I I 1log 1 where J denotes the intensity image values This measure accounts for variation of texture within the image Since the measure is unreliable with small sample sizes and because the user should be looking to identify reasonably large regions with homogeneous texture texseg starts with a default segmentation of regions of approximately 200 pixels in size This value is user settable Figure 4 6 shows an example segmentation using texseg Note that the routine has clearly differentiated between areas of virgin rain forest and clearings This distinction would have been far less clear from using the segment routine Note that the output image
106. edure output_array easyseg image array looks intensity quality mtpol Parameters image nc This can be either a single or a multiple layered real or complex valued image de pending on the chosen method A polarimetric image should be given as a layered complex valued image output nc The coloured region map output image is of the same dimensionality as the input im age if the original image was real valued The coloured region map output image is single layered in the case of a complex valued original image intensity integer default 0 Assume the data is amplitude square root intensity looks float default 1 The looks measure of the image data If looks is negative use 1 the routine will compute this parameter from the data quality Chooses from a predefined set of segmentation parameters quality default a good compromise sensitive preserve detail or fast quick result The default is default mtpol integer default 0 Polarimetric data is stored as layers of complex data however this is the same file format as multi temporal polarimetric data To distinguish the two the mt pol parameter is used For instance 5 days of 3 channel polarimetric data would be stored as a 15 layer complex image and the mtpol 3 parameter would be used The default value of mt po1 0 uses all of the layers as a single polarimetric image tilesize integer default 512 A tilesize of 512 x 512 is used overlap integer
107. en tiles an overlap of 48 is used tmpdir directory default None Intermediate results are written to a system temporary directory Processing tiles with an overlap incurs an overhead in terms of disk space and file copying so rou tines that would not benefit from it for instance add py or 1og10 py have the overlap set to a default of 0 The tilesize and overlap defaults can be set within the GUI using the Configuration Stripmap Op tions menu A 3 Documentation Conventions In this chapter we document all of the high level user allable facilities provided in InfoPACK For each facility both function available from Python or from the InfoPACK console by importing module infopack and script available from a command line versions are listed Image Parameters e For the script versions the image files are given with the suffix nc These parameters must be the full path name of an image file e For the function version the data files are given with the suffix array These parameters must be the name of Numeric Arrays containing the image data 29 Mar 10 NASoftware Ltd 119 APPENDIX A InfoPACK HIGH LEVEL ROUTINE DOCUMENTATION e For the script version the specified file is written before exit For the function version the specified image is returned as the or a result of the function 120 NASoftware Ltd 29 Mar 10 A 4 HIGH LEVEL ROUTINES A 4 High Level Routines A4 1 segment Purpose segment takes an i
108. ensitive normal and fast The sensitive parameters should be employed where there is little scene structure or when it is on a large scale so that large segments are to be preferred The normal parameters are for general use representing a compromise between sensitivity and speed Finally the fast parameters are intended for speedy but not necessarily accurate determination of segments Note that when we are concerned with clas sification after a segmentation stage it is not essential that the segmentation should completely fill the homogeneous regions Provided that segments are large enough for the statistical uncertainty to be small and that the segments do not go over region boundaries the results will be indistinguish able Indeed there is no appreciable difference between the classification performance for the three modes in this example For the present comparison we adopt the normal mode The classified results shown in Figure 6 29 c is much closer to the ground truth with the average fraction of cor rectly classified categories rising to 0 941 and the confusion matrix in Table 6 1 c showing very little ambiguity between classes There is an obvious over smoothing of some of the boundaries so that details of the structure have been lost at the expense of rejecting unwanted flecks in the reconstruction For comparison we show the corresponding result for ML classification following segmentation of the span measure
109. entation offers a valuable precursor to temporal signature classification because it adapts the region size to the data and presents a common set of boundaries for the whole sequence of images The Upper Limit provides the most sensitive method for classification However it suffers from three disadvantages It cannot detect inhomogeneities within the defined regions The sensitivity is degraded by any inhomogeneities It only processes within defined regions with no extrapolation capability such as that possessed by segmentation Joint segmentation is somewhat less sensitive than the Upper Limit but is considerably better than using a 3x3 window which in turn is better than single pixels The 3x3 window method suffers from edge effects due to averaging over the boundaries Therefore temporal signature classification should adopt joint segmentation as an initial pre process NASoftware Ltd 29 Mar 10 6 5 HANDLING CORRELATED IMAGES Figure 6 13 Typical HH SAR image obtained from the DLR ESAR system 6 5 Handling Correlated Images 6 5 1 Introduction On many occasions SAR images are sampled at spacing less than their resolution Indeed Nyquist sampling of the detected intensity image would require spacing of half the resolution in each direc tion to retain all the information in the image Figure 6 13 illustrates part of a typical SAR image obtained using the DLR ESAR system In this mode of operation the rang
110. erwise to the false value See section B 2 29 for details of the routine parameters Image Arithmetic Two options are provided e Single image functions Provides access to the following arithmetic operations on images carried out on each pixel Scalar Add image2 imagel c Scalar Multiply image2 imagel c Scalar Divide image2 imagel c log10 image2 log y image log e image2 log image1 square root image2 yimagel square image2 imagel imagel Double Image Functions Provides access to the following operations carried out on each corresponding pair of pixels in two specified images of the same size image3 imagel op image2 op Full function specifications are given in Appendix B Stripmap Options Options are provided to set the default tilesize and overlap values These values have no effect with some evaluation versions of InfoPACK Full specifications are given in Appendix A 2 5 3 3 The Segmentation Menu Easyseg Easyseg provides a fast and easy to use segmentation routine recommended for exploratory or rou tine use It carries out an initial segmentation followed by a call to the merge routine If you neeed closer control of the segmentation process call segment and optionally merge separately See section A 4 2 for details of the routine parameters and section 4 1 2 for a longer discussion Segment Default Settings Segment a given image f
111. et us first define these quantities Effective Equivalent Number of Looks The statistics of the speckle are usually described in terms of the Effective or Equivalent Number of Looks ENL This is intended to represent the speckle component of the observed intensity Normally of course the intensity combines both speckle and RCS fluctuations However if the underlying RCS is uniform the ENL can be defined as mean ENL variance In general estimating the looks measure yields values which are smaller than the true ENL because additional RCS fluctuations are included Only if the RCS is constant or if the observed intensity is divided by the true underlying RCS would the looks measure correspond to the ENL Resolution The physical characteristics of the radar determine how far apart two objects on the ground must be in order for them to be distinguished in the image This minimum distance is the spatial resolution of the image The resolution in the range across track direction is determined by the bandwidth of the transmitted pulse larger bandwidth equals higher resolution The resolution in the azimuth along track direction is determined by the length of the radar antenna a shorter antenna gives a higher resolution Pixel spacing The pixel spacing in an image is determined by the sampling frequency of the pro cessor An image can be oversampled so that the pixel size is smaller than the resolution When this happens the
112. f images If there is no change in RCS then the data is merely gamma distributed about its mean value with an order parameter equal to the effective number of looks 3 for these images Any type of fluctuation in RCS will result in an increase in U Thus changes can be detected by applying a suitable threshold to the normalised log image formed from the sequence The exact choice of threshold is selected to yield a reasonably low rate of false detections of change in the absence of variations in the RCS When this analysis is applied to the sequence of images we obtain the results shown in Figure 6 8 If we compare the results with the expected ones we note that with single pixels a only 70 NASoftware Ltd 29 Mar 10 6 3 CHANGE DETECTION BETWEEN INTENSITY IMAGES Figure 6 5 a Detected Changes using single pixels the strongest change is visible corresponding to the red class in Figure 6 2 After applying a 3x3 window b changes corresponding to the green class are visible The result after joint segmentation c is considerably better with all classes being detected Indeed the changes resemble the predicted ones in Figure 3 When the Upper Limit is considered in which the predefined class boundaries determine the regions the result is further improved so that the result is almost indistinguishable from Figure 6 4 The performance comparison can be quantified in terms of the probability that changes are correctly detected shown
113. f the image data If looks is negative use 1 the routine will compute this parameter from the data maxiters integer default 90 The maximum number of iterations complete scans of the image data allowed regionsize integer default 200 texseg will start forming regions of a size regionsize pix els Because texture is not defined for a single pixel regionsize should be set greater than 1 for an initial look at a new dataset the default value of 200 is recommended intensity integer default 0 Assume the data is amplitude square root intensity immutable This is an image region map whose edges will be preserved in the final segmenta tion The image should be a single layer and have the same width and height as the input image For the command line script the immut able option will be a file name and for the python procedure it is a Numeric array tilesize integer default 512 A tilesize of 512 x 512 is used overlap integer default 48 To prevent obvious joins between tiles an overlap of 48 is used tmpdir directory default None Intermediate results are written to a system temporary directory Further Comments The output image is coloured with the Blacknell Tough measure The routine colseg may be used for a different colouring 126 NASoftware Ltd 29 Mar 10 A 4 HIGH LEVEL ROUTINES A 4 4 merge Purpose Given an original image file and a region map of this image merge will unify any adjacent regions which
114. fault None Intermediate results are written to a system temporary directory Further Comments edge is only applicable to segmented images Using edge to produce an edge map of other than a segmented image will produce disapointing results Instead the routine scanedge should be used 150 NASoftware Ltd 29 Mar 10 B 2 IMAGE MANIPULATION ROUTINES B 2 6 fill Purpose Creates a real valued image of the given dimensions width height and the number of layers and colours all pixels with the same uniform value Width and height parameters must be specified Command Line Script fill py image_out width height layers value tilesize tmpdir Python Procedure array out fill width height layers value Parameters layers integer default 1 value float default 0 tilesize integer default 512 A tilesize of 512 x 512 is used tmpdir directory default None Intermediate results are written to a system temporary directory B 2 7 fill gammarand Purpose Creates a real valued image of the given dimensions width height and the number of layers filled with random samples These samples have a gamma distribution with the given mean and order parameter Seeds seed1 and seed2 may be specified to give a reproducible sequence of random numbers Width and height parameters must be specified Command Line Script fill gammarand py image_out width height layers mean order
115. form RCS within the seg ments Thus this parameter value is too high The result for the default shape parameter value in 4 3 has lost detail such as the weak dashed lines visible in places in a but has also lost the spu rious spots As the value decreases the reconstruction becomes progressively smoother with more detail averaged out It is apparent that a value around the default value is probably best for this segmentation The other two parameters relate to the optimisation process and are less obviously related to the quality of the output In fact the stopping parameter has little impact on the quality or speed of reconstruction provided that the default values are used The cooling parameter has comparatively little effect on the quality of the reconstruction but has a considerable impact on execution time For convenience three predefined parameter sets are provided with segment fast provides for a quick look at the segmentation default is recommended for normal running and sensitive provides suitable settings when the interest lies in small features 22 NASoftware Ltd 29 Mar 10 4 1 SEGMENTATION Figure 4 2 Segmentation with large shape parameter Figure 4 3 Segmentation with default shape parameter 29 Mar 10 NASoftware Ltd 23 CHAPTER 4 IMAGE PROCESSING WITH InfoPACK Figure 4 4 Segmentation with small shape parameter 4 1 2 The easyseg routine The easyseg routine is included to give an ea
116. gure 6 17 show no structure related to the underlying scene and the average statistical properties could be used to derive an Effective Number of Looks ENL that should be the same as the known value 4 0 imposed by the sensor This residual ratio image is shown in Figure 6 19 Some minor evidence of the underlying scene structure is visible indicating that the reconstruction is not perfect On the other hand the derived ENL is 4 03 which is close to the expected value The fact that the ENL is approximately correct indicates that the average properties are well represented However the presence of structure effects in the residual image indicates those regions where detail has been lost in the reconstruction This is a consequence of adopting a single shape parameter If the segmentation algorithm used in segment were extended so that this parameter was allowed to adapt to the scene content the effects of structure would probably be reduced but we have not tried this A more sensitive test as to whether the residual ratio corresponds to the predicted speckle model for a system with ENL 4 0 is to compare the PDF of the residual ratio at full resolution with theory This is illustrated in Figure 6 20 While the agreement is not perfect it is clear that the PDF has approximately the correct form The estimated mean for the residual ratio is almost exactly 1 0 as predicted The observed discrepancy is not however entirely due to the fact that
117. he byte order 18 NASoftware Ltd 29 Mar 10 3 3 GENERAL FILE FORMATS width nw height nh The image size The dimensions in the image are measured in pixels option width the number of pixels per row of the image option height number of rows in the subimage you can import only some of the rows 29 Mar 10 NASoftware Ltd 19 Chapter 4 Image Processing with InfoPACK This chapter illustrates how to use the main InfoPACK routines and some of the results that can be achieved to show what they do and how the parameters affect the results Lengthier Case Studies geared more towards specific applications are given in Chapter 6 Many of the test images used in this chapter are simulated 3 look SAR data but InfoPACK can be used on SAR data from any source and with any size of looks measure and also on optical images see section 6 for examples Sample data is included with the InfoPACK distribution in the images subdirectory of the main installation directory Also located there is a README file that provides in some detail how to process the sample data with several of the routines described in this chapter 4 1 Segmentation Speckle is the most characteristic feature of SAR imagery and the removal of speckle is an impor tant step in the automatic interpretation of SAR images Many techniques have been proposed for reducing the level of speckle These include simple averaging and applying standard filters How ever thes
118. he different processing methods Method Pcor Single pixel 0 899 3 by 3 Window 0 954 Joint Segmentation 0 992 Upper Limit 1 000 If we compare the probability of false classification 1 Pcor we see that this progresses from 10 for single pixels to 4 6 for 3x3 window 0 8 for joint segmentation and 0 for the Upper Limit The considerable advantage offered by joint segmentation before classification is obvious As a final comparison we show the confusion matrices for the different methods in Table 2 Generally speaking the classification results are better than those for the 10 scenes considered when we introduced joint segmentation and classification This is because the average RCS in each class is higher than before 29 Mar 10 NASoftware Ltd 81 CHAPTER 6 CASE STUDIES e P Figure 6 4 2 c Classification using Joint Segmentation 82 NASoftware Ltd 29 Mar 10 6 4 TEMPORAL SIGNATURES OF INTENSITY IMAGES Figure 6 4 2 d Classification using Upper Limit These detailed results bear out the observations made from the classified images In particular the averaging window seems to have the most serious degrading effect on the classification of class 0 This is because all the other classes tend to be averaged down toward this class Table 6 4 2 Confusion matrices for temporal signature classification using the different processing methods a single pixel b 3 x3 window c joint segmentation
119. he ground truth regions is an inevitable consequence of speckle noise In this instance we know that the simulated image had a uniform RCS within each region The speckle statistics have of course spoilt this The High Limit uses the fact that the average intensity in every region is the same This allows the average intensity to be segmented and classified However the performance is essentially the same as that achieved with joint segmentation revealing that the latter is able to use all the infor mation in the sequence of images The Upper Limit as one would expect provides the highest statistical accuracy There are two important caveats to note however 1 This result is based on the assumption that the RCS within the region is indeed homogeneous In real applications this will not generally be the case 2 The Upper Limit classification can only be applied where region boundaries are known It cannot extrapolate beyond the ground truth Segmentation on the other hand can use the training set to enable it to classify over as large an image as is required Thus the overall conclusion is as in Study 1 segmentation provides an essential precursor to intensity classification 29 Mar 10 NASoftware Ltd 67 CHAPTER 6 CASE STUDIES 6 3 Change Detection between Intensity Images 6 3 1 Introduction In the previous study we demonstrated the capability of segmentation to improve the classification of a set of simulated 3 look S
120. he nearest pixel A reasonable approximation can be obtained by observing that the gradient of both ACFs is reduced significantly at around a lag value of 3 indicating that the speckle part of the ACF has decayed away at about this value When the averaging and resam pling by 3x3 is applied to the original image the image shown in Figure 6 5 4 a is obtained while Figure 6 5 4 b shows the corresponding segmentation It is obvious that the segmented form of the resampled image contains large uniform regions as one would expect This is in stark contrast to the original version in Figure 6 5 2 b There are no speckle blob sized segments in Figure 6 5 4 b which should be compared with the very large number visible in 6 5 2 b Figure 6 5 4 c shows the ratio of the original intensity to the reconstructed cross section after resampling This has the expected scale of correlations in the speckle blobs similar to those in Figure 6 5 2 d and com pletely different from the ratio image obtained from the original data shown in Figure 6 5 2 c and repeated here for comparison in Figure 6 5 4 d Finally the ENL derived from the ratio image is 4 00 consistent with the defined value This consistency indicates that the segmentation derived in this way can be trusted 92 NASoftware Ltd 29 Mar 10 6 5 HANDLING CORRELATED IMAGES Figure 6 5 4 d Segmentation of the resampled original image from Figure 6 5 2 a Image resampled by 3x3 b segment
121. he parameter sd is used with the following mapping p sdxo gt 0 p sdxo 255 where y is the mean and a is the variance of the data A value of sd 0 has the special meaning of minimum 0 maximum 255 Complex data cannot be directly displayed as a bitmap and so must be converted to real data first This is done by the option cmp12real Command Line Script See tojpeg py totiff py Python Procedure array_out makebitmap array_in cmpl2real sd min max verbose Parameters cmpl2real string default abs Chooses from a set of methods for converting the complex image to a real one abs absolute value mag magnitude of complex number abs x abs 172 NASoftware Ltd 29 Mar 10 B 3 FILE HANDLING FACILITIES phase Phase of complex number real Real part imag Imaginary part sd float default 2 0 min list default None This can either be a single value or a list eg min 3 ormin 0 3 7 2 max list default None verbose integer default O If set to 1 then the estimated min and max values are printed out B 3 12 nanfix Purpose Sets NaN or infinities to a specifed value Command Line Script nanfix py image_in image out nan tilesize tmpdir Python Procedure array_out nanfix array_in nan Parameters nan float default 0 Wherever a NaN or infinity occurs in the image is replaced with this value tilesize integer default 512 A ti
122. he parameters in terms of the final classi fication performance This avoids somewhat arbitrary decisions about what constitutes a good segmentation 108 NASoftware Ltd 29 Mar 10 6 8 SEGMENTATION AND CLASSIFICATION OF POLARIMETRIC DATA 6 8 2 Segmentation performance A crucial issue with applying segmentation is the selection of operating parameters In Figure 6 27 we show the result of applying ML polarimetric segmentation with different values of the shape parameter In Figure 6 27 a a shape parameter of 0 1 is adopted It is apparent that this allows small regions with a sharply varying boundary to be reconstructed However there is some evidence for flecks of speckle dominated segments By increasing the penalty for non smooth boundaries reducing the shape parameter this type of reconstruction can be suppressed Figures 6 27 b to d show the corresponding results for shape parameters of 0 01 0 001 and 0 0001 If you compare these with the original test pattern in Figure 6 25 a it is clear that reconstruction is generally very good The shape of the fields around the water course are well represented Figure 6 27 b with a shape parameter of 0 01 appears to offer a good compromise between smooth edges to regions and reproducing small structures We need to increase the shape parameter to follow weak small structure but on the other hand we need to reduce the shape parameter to eliminate speckle regions As before the only reaso
123. hicles or pylons Because they are small and isolated they can be difficult to distinguish from the fluctuations due to speckle 4 6 1 The cfar routine The cfar algorithm locates bright areas in the image which are not just the effect of speckle and checks to see whether they are small and isolated The CFAR algorithm works by calculating the likelihood of the central pixel in a scanning window coming from the background distribution as represented by the rest of the window The classification of the central pixel as a point object depends upon the comparison of this likelihood with the threshold calculated from the probability of false alarm pfa This probability can be set by the user A lower value of probability of false alarm will mean less point objects will be detected Since the result is determined by the accuracy of the statistical estimates it is advisable to have a reasonably large data sample i e window size We suggest using a 7x7 window Figure 4 15 b shows the point objects detected in figure 4 15 a through the use of cfar The cfar algorithm can be performed upon images with more than one image layer cfar can also be accessed via the filter function use method cfar Full Python script and function syntaxes for cfar are given in section A 4 11 29 Mar 10 NASoftware Ltd 39 CHAPTER 4 IMAGE PROCESSING WITH InfoPACK 4 7 References For papers to download other references and links please visit our we
124. ibit crinkle in the edges of regions This works on a logarithmic scale so that smaller eg 0 1 0 01 0 001 values produce smoother boundaries stopping float default 0 01 Unless maxiters iterations are reached the routine will termi nate when two successive iterations show a relative change in the objective function of less than stopping cooling float default 10 Determines the rate of cooling in the simulated annealing minimisation process A larger value of cooling tends to decrease the time taken by the routine but may in crease the likelihood that it gets stuck in a local minimum looks float default 1 The looks measure of the image data If looks is negative use 1 the routine will compute this parameter from the data maxiters integer default 90 The maximum number of iterations complete scans of the image data allowed regionsize integer default 1 segment will start forming regions of size regionsi ze pixels For most purposes the default value of 1 is recommended A larger value will decrease runtimes but may lead to chunky segmentations intensity integer default 0 Assume the data is amplitude square root intensity immutable This is an image region map whose edges will be preserved in the final segmenta tion The image should be a single layer and have the same width and height as the input image For the command line script the immut able option will be a file name and for the python procedure it
125. iform spacing of the original data The number of classes is determined by the user setting the parameter num_classes in the routine make_classes This generates a means file which can then be passed to the classify routine Supervised Classification With supervised classification a priori class means or a priori class means along with class order image files are input to the routine These files can be derived via e Entering the values in a text file and using the ascii2nc py utility The values could be selected by using histogram py to study the distribution of the data and pick out peak values e By running the mask_stats orclassstats routines on a set of training masks See the images directory for an example of this In any case the means nc and orders nc files can be edited further to provide iterative supervised classification Figure 4 8 shows one of a sequence of images of the Indonesian rain forest and the resultant image from a joint segmentation process Using this segmentation the images can then be classified as shown in figure 4 9 The reason for using the sequence of images for classification is that the information is in the temporal change Joint segmentation provides better definition of boundaries The low resolution and speckle associated with just a single image would give rise to a much poorer result A further example of supervised classification is given in the images directory Full Python script and function syn
126. ile with one of three predefined settings default fast or sensitive See section A 4 1 for details of the routine parameters and section 4 1 for a longer discussion 46 NASoftware Ltd 29 Mar 10 5 3 THE MAIN InfoPACK MENUS Segment Custom Setting Segment a given image file with full control over the options The default settings are shown and can be altered See section A 4 1 for details of the routine parameters and section 4 1 for a longer discussion Texture Segment Segment a given image file using texture instead of intensity information in the image The default settings allows you to choose from one of three predefined settings default fast or sen sitive You will be prompted for the segmentation parameters or select the custom settings to view and change the default values See section A 4 3 for details of the routine parameters and section 4 1 4 for a longer discussion Merge Merge removes weak edges from a segmentation It does this by merging regions that are not sig nificantly different in intensity at the specified level of significance Merge requires the segmented image the original image and a significance level See section A 4 4 for details of the routine parameters and section 4 1 3 for a longer discussion Colseg Given a segmentation image and the original imge colour the segmentation ie floodfill each seg ment with a value computed from any of a set of meas
127. in Table 6 3 2 Table 6 3 2 Probability of correct detection of changed regions for the different processing methods Method Pcor Single Pixel 0 626 3 by 3 window 0 805 Joint Segmentation 0 971 Upper Limit 0 996 If we compare the probability of false classification 1 Pcor we see that this progresses from 37 for single pixels to 19 for 3x3 window 3 for joint segmentation and 0 4 for the Up per Limit The considerable advantage offered by joint segmentation before change detection is obvious Positional change detection Positional change detection offers the opportunity to detect both the presence of a change and its position The likelihood that a change occurs at a position K in a sequence can be shown to be given by 29 Mar 10 NASoftware Ltd 71 CHAPTER 6 CASE STUDIES Figure 6 6 b Detected Changes using a 3 by 3 filter Figure 6 7 c Detected Changes using joint segmentation 72 NASoftware Ltd 29 Mar 10 6 3 CHANGE DETECTION BETWEEN INTENSITY IMAGES Figure 6 8 d Detected Changes using Prior Knowledge K Klog I M k log 12 Mlog Io where the bars denote the average values in each segment over the first K images region 1 the second M K images region 2 and all M images region 0 respectively Once again we can apply a threshold to ensure that the detected change is unlikely to be caused by speckle noise when no change is expected The predic
128. integer default 100 The number of data bins used to sort the data sdrange float default 2 The number of bins is spread over the interval u sdrange x o u sdrange x g where y and o are the mean and standard deviation of the data respectively Any data values falling outside this range are placed into the bin with either the lowest or highest data values as appropriate intensity integer default 0 If set to 1 then the image is processed as an intensity image and the data is not squared before calculating the statistics to work out the range to plot If set to 0 then the range is calculated using statistics assuming its an amplitude image norm float default None If this parameter is specified normalize the area under the histogram to this value B 4 5 stats Purpose Calculates a variety of statistics for a given image file These include image dimensions data type minimum maximum mean standard deviation and for real data looks measure and order parameter If the intensity flag is set to O the default value thus indicating the image contains amplitude data then the statistics are calculated for the square of the data i e the intensity image The mean minimum and maximum values are then squarerooted before being returned Complex data is converted internally to real data using the cmp12real method before the statistics of the real data is displayed For multi layer images the statistics from each layer is
129. is provided e Geoliff tif e Tiff Bitmap tif e ERDAS Imagine img e Atlantis MFF hdr Other export formats including postscript and png are made available from the OpenEV viewer Use the Print option and print to file Run Study This will prompt for the case study file to be opened and then show the output of the study in the console Under Linux the case study should end with 1in while under Windows the case study should end with bat 29 Mar 10 NASoftware Ltd 43 CHAPTER 5 USING THE INFOPACK GUI New Script This opens a blank IDLE shell window From the file menu selecting New Window brings up an empty IDLE editor window ready for you to type in a Python script a text file containing a list of InfoPACK or other Python commands to be executed Python script files are generally give the suffix py For details of the editor features see section 5 7 Open Script This will prompt for the script file to be opened and then open it in a new IDLE editor window An IDLE shell window is also opened for the script output when run from the editor window For details of the editor features see section 5 7 New Console This will open a new InfoPACK shell window Exit Exit InfoPACK 5 3 2 The Tools Menu Create New Image Create an image with artifical speckle generated from a specified distribution Useful when creating test images Full function specifications are given in Appendix B L
130. is a Numeric array size integer default 2 For polarimetric data only this option sets the minimum region size mtpol integer default 0 Polarimetric data is stored as layers of complex data however this is the same file format as multi temporal polarimetric data To distinguish the two the mt pol parameter 122 NASoftware Ltd 29 Mar 10 A 4 HIGH LEVEL ROUTINES is used For instance 5 days of 3 channel polarimetric data would be stored as a 15 layer complex image and the mtpol 3 parameter would be used The default value of mt po1 0 uses all of the layers as a single polarimetric image tilesize integer default 512 A tilesize of 512 x 512 is used overlap integer default 48 To prevent obvious joins between tiles an overlap of 48 is used tmpdir directory default None Intermediate results are written to a system temporary directory Further Comments The merge routine may be employed to tidy the segmentation and ensure that all the edges are statistically significant 29 Mar 10 NASoftware Ltd 123 APPENDIX A InfoPACK HIGH LEVEL ROUTINE DOCUMENTATION A 4 2 easyseg Purpose Given an original image file easyseg will produce a segmented image Easyseg is designed as a simple interface to segmentation It combines an optimal set of parameters to both segment and merge Command Line Script easyseg py image nc output nc looks intensity quality mtpol tilesize overlap tmpdir Python Proc
131. is therefore possible to study the temporal development of identical regions of these images In this study we simulate a sequence of ten images with the same six classes as in the previous study shown in Figure 6 9 but with cross section variation from sample to sample as shown in Figure 6 10 From Figure 6 10 we see that with the exception of the maroon class all the classes show a variation in RCS over the sequence Typical variations include a gradual increase analogous to crop growth for example and a sudden drop analogous to the reaping of crops The purpose of this study is to classify regions in the sequence of images in terms of this signature In fact we do not use this signature directly but train the temporal signature of each class using the training masks shown in Figure 6 11 a The validation is then performed by testing over the regions denoted by the test masks in Figure 6 11 b 78 NASoftware Ltd 29 Mar 10 6 4 TEMPORAL SIGNATURES OF INTENSITY IMAGES Image number Figure 6 10 Plot of the mean intensity in each of the classes from Figure 6 9 during the sequence of images 6 4 2 Classification methods As before we simulate a series of images having random 3 look speckle with the underlying RCS shown in Figure 6 10 We then perform Maximum Likelihood temporal signature classification on these images after different processing techniques have been applied namely Upper Limit classification
132. lculating the texture measures can be time consuming on large im ages If you do not require this information then running stats with calctex 0 will save some time tilesize integer default 512 A tilesize of 512 x 512 is used overlap integer default 48 To prevent obvious joins between tiles an overlap of 48 is used tmpdir directory default None Intermediate results are written to a system temporary directory B 4 6 xgraphtool Purpose Provides a graphical display of the data in a text file The points for each graph should be listed on a single line A new line represents a new graph 29 Mar 10 NASoftware Ltd 183 APPENDIX B LOW LEVEL AND UTILITY ROUTINE DOCUMENTATION Command Line Script xgraphtool py text _in Python Procedure None provided Parameters axissize integer default 10 Font size for axis labels legendsize integer default 10 Font size for legend labels legend integer default 1 Display legend X list x min max Min max values for x axis y list y min max Min max values for y axis title string default None Graph title xaxis string default None X axis label yaxis string default None Y axis label grid integer default 1 Draw grid behind graph colours list default colours Red Green Blue Line colours No spaces around commas allowed style list default style solid dot long_dash short_dash dot_dash Line drawing styles No spaces around commas allowed thick i
133. lesize of 512 x 512 is used tmpdir directory default None Intermediate results are written to a system temporary directory B 3 13 nc2ascii Purpose Converts an NASoftware Ltd NetCDF file into an ASCII file Command Line Script 29 Mar 10 NASoftware Ltd 173 APPENDIX B LOW LEVEL AND UTILITY ROUTINE DOCUMENTATION ne2ascii py image_in ascii_out Python Procedure None provided B 3 14 nc2raw Purpose Converts an NASoftware Ltd NetCDF file into a raw binary data file Command Line Script nc2raw py image_in raw_out byteswap Python Procedure None provided Parameters byteswap integer default 0 The option byteswap 1 transforms big endian to little endian data B 3 15 numload Purpose Loads an image file without the GIS information numload returns the GDAL file as a Numeric array numload loads the subimage of dimensions width by height starting at the pixel location given by woffset and hoffset Command Line Script None provided Python Procedure array_out numload image_in woffset hoffset width height 174 NASoftware Ltd 29 Mar 10 B 3 FILE HANDLING FACILITIES Parameters woffset integer default 0 hoffset integer default 0 width integer default None read whole image in height integer default None read whole image in B 3 16 raw2nc Purpose Converts a binary data file into an NASoftware Ltd NetCDF file The ty
134. lso classstats 29 Mar 10 NASoftware Ltd 189 APPENDIX B LOW LEVEL AND UTILITY ROUTINE DOCUMENTATION B 5 6 mask_polstats Purpose Calculates the covariance matrix from a polarimetric image within the masked regions of image mask is a 0 1 image where each layer indicates the training data for each class Command Line Script mask_polstats py maskimage_in image_in meanimage out tilesize overlap tmpdir Python Procedure stats_class mask_polstats maskarray_in array_in cmpl2real verbose Parameters cmpl2real string default abs Chooses from a set of methods for converting the complex image to a real one abs absolute value mag magnitude of complex number abs x abs phase Phase of complex number real Real part imag Imaginary part verbose integer default O If set to 1 prints out the class statistics tilesize integer default 512 A tilesize of 512 x 512 is used overlap integer default 48 To prevent obvious joins between tiles an overlap of 48 is used tmpdir directory default None Intermediate results are written to a system temporary directory B 5 7 make classes Purpose Generates equally spaced class means from input data If the minimum or maximum parameters are not given then the minimum and maximum parameters are estimated from the mean of the data plus or minus the sd spread 190 NASoftware Ltd 29 Mar 10 B 5 CLAS
135. lt The default is default These choices set the following parameter values Parameter Fast Default Sensitive Shape 0 0001 0 0001 0 0001 Stopping 0 01 0 01 0 003 Cooling 30 10 1 Looks 1 1 1 Maxiters 30 90 1000 Regionsize 200 200 200 Intensity 0 0 0 Any of these values can be over ridden by subsequent values on the command line The values fast default sensitive are provided as generally convenient settings but because of 29 Mar 10 NASoftware Ltd 125 APPENDIX A InfoPACK HIGH LEVEL ROUTINE DOCUMENTATION the iterative nature of the segmentation process and its data dependence some initial parameter searching may be worthwhile for best results shape float default 0 0001 The objective function minimised by texseg includes a shape term designed to inhibit crinkle in the edges of regions This works on a logarithmic scale so that smaller eg 0 1 0 01 0 001 values produce smoother boundaries stopping float default 0 01 Unless maxiters iterations are reached the routine will termi nate when two successive iterations show a relative change in the objective function of less than stopping cooling float default 10 Determines the rate of cooling in the simulated annealing minimisa tion process A larger value of cooling tends to decrease the time taken by the routine but may increase the likelihood that it gets stuck in a local minimum looks float default 1 The looks measure o
136. lue to use for the background xoffset integer default 0 Add this value to the x deltas yoffset integer default 0 Add this value to the y deltas intensity integer default 0 Assume the data is amplitude square root intensity tilesize integer default 512 A tilesize of 512 x 512 is used tmpdir directory default None Intermediate results are written to a system temporary directory B 2 12 layer_sum Purpose Adds all the layers in two or more images The result is the sum of corresponding pixel values in all of the layers of the input images Command Line Script layer_sum py imagel_in image2_in imagen_in image_out tilesize tmpdir Python Procedure array out layer sum arrayl in array2_in arrayn_in Parameters tilesize integer default 512 A tilesize of 512 x 512 is used tmpdir directory default None Intermediate results are written to a system temporary directory Further Comments The width and height of the input images must be the same although they can differ in the number of layers The output image will have the same width and height and only be a single layer image 29 Mar 10 NASoftware Ltd 155 APPENDIX B LOW LEVEL AND UTILITY ROUTINE DOCUMENTATION B 2 13 1log10 Purpose The result is the logarithm to the base 10 of each individual pixel value from the input image Command Line Script log10 py image_in image_out tilesize tmpdir Python Procedure array out
137. mage Statistics Image Properties Clip Image Filter Despeckle Threshold Image Arithmetic Stripmap Options Segmentation see section 5 3 3 Segment Texture Segment Merge Colseg Defleck Edge Classification see section 5 3 4 42 Classify Generate Classes Colour Classes Class Statistics Classmap to Layers Layers to Classmap Combine Matrix Split Matrix NASoftware Ltd 29 Mar 10 5 3 THE MAIN InfoPACK MENUS Change Detect see section 5 3 5 Change Detection Change Position Detect Help see section 5 3 6 Create Licence File Enter your licence details to activate the InfoPACK routines InfoPack This documentation including documentation for all InfoLIB routines Most users will need only this manual Python For details of the base Python language Numeric Python The manual for the Numeric Python extensions used by InfoPACK Command List This pops up a window listing the python scripts that come with InfoPACK About Gives the version number for this release 5 3 1 The File Menu This provides access to the following facilities Open Image Open an image file in a format supported directly by the OpenEV Image Viewer see Section 5 4 Import Image Convert an image into the InfoPACK standard nc format see chapter 3 A wide range of standard image formats is supported Export Image Convert a nc image file to an alternative format Full conversion to the following formats
138. mage and produces a cartoon segmentation consisting of a collection of ho mogeneous regions The regions are each coloured with its mean value in the case of real or single layered complex images or with the determinant of the M by M covariance matrix For M gt 1 complex images this is defined as det det X where ij cov xi j 1 1 2 M is the covariance of x and zj The image or images need not emanate from a SAR processor images from other sources eg optical may be processed either alone or in combination with SAR images The algorithm starts with an adaptive rectangular default single pixel segmentation of the image The likelihood of the data fitting this segmentation can then be calculated The judgement as to whether to move edge pixels depends on the change in likelihood of the segmentation The process governed by a simulated annealing optimisation algorithm will eventually find the global optimal solution for the segmentation of the image The computational cost increases with the area N of the image as O N log N Command Line Script segment py image nc output nc quality shape stopping cooling looks maxiters regionsize intensity immutable size mtpol tilesize overlap tmpdir Python Procedure output_array segment image_array quality shape stopping cooling looks maxiters regionsize intensity immutable size mtpol
139. matrix from the data The maximum value of the likelihood when all possible classes are compared indicates which class the particular regions should be assigned to We now compare classification of the covariance matrix using the following methods 1 Upper Limit classifying the average in each ground truth region 2 Segmentation classifying the average in each segment 3 3x3 window classifying the average over a 3x3 window 4 Lower Limit classifying the covariance matrix in each pixel The classification results are shown in Figure 6 29 Classification of a single pixel shown in Figure 6 29 a is comparatively poor with the average probability of correctly classifying each cat egory 0 615 The associated confusion matrix shown in Table 6 1 a reveals how poorly the classes 0 magenta 1 blue and 4 yellow are assigned as can be seen in Figure 6 29 a Nevertheless the underlying structure of the original classes in Figure 6 25 a is still visible Classification over a 3x3 window shown in Figure 6 29 b yields considerable improvement with the average fraction of correctly classified pixels in each class rising to 0 798 However the confusion matrix shows that there is still ambiguity between class 2 cyan and O magenta and between class 1 blue and class 4 yellow The results of classification following ML polarised segmentation depend on the selection of control parameters Three basic default sets are defined s
140. mining whether to merge regions Two adjacent regions will be merged if their intensities are not significantly different at a confidence level of pfa crinkle Same as pfa but also smoothes the regions boundaries abs Merges regions whose intensity differs by less than the absolute value set by the cost option crinkabs As abs but also smoothes the region boundaries rel Merges regions whose intensity differs by less than the relative value set by the cost option crinkrel As rel but also smoothes the region boundaries 29 Mar 10 NASoftware Ltd 127 APPENDIX A InfoPACK HIGH LEVEL ROUTINE DOCUMENTATION user As crinkle but the power of the smoothing can be set by the crinkle pa rameter cost float default 5 The definition of cost depends upon the chosen method See method for details looks float default 1 The looks measure of the image data If looks is negative use 1 the routine will compute this number from the data crinkle float default 4 0 This is an ad hoc parameter controlling how smooth the region bound aries should be Increasing values between about 1 and 4 increase the amount of smoothing intensity integer default 0 Assume the data is amplitude square root intensity mtpol integer default 0 Polarimetric data is stored as layers of complex data however this is the same file format as multi temporal polarimetric data To distinguish the two the mt pol parameter is used For instance 5 days
141. moving window of size width by height to filter the image by computing an output pixel value from within each window according to the method indicated in the routine name Command Line Script filter method py image nc output nc width height looks pfa intensity tilesize overlap tmpdir Python Procedure Use filter with the method parameter defined appropriately Parameters See parameter descriptions for filter in section A 4 11 29 Mar 10 NASoftware Ltd 143 APPENDIX A InfoPACK HIGH LEVEL ROUTINE DOCUMENTATION A4 13 cfar Purpose Single pixel CFAR target detector The CFAR algorithm works by calculating the likelihood of the central pixel in a moving window of size width by height coming from the background distribution as represented by the rest of the window The classification of the central pixel as a target depends upon the comparison of this likelihood with the threshold calculated from the probability of false alarm pfa A lower value of probability of false alarm will mean less targets will be detected Since the result is determined by the accuracy of the statistical estimates it may be advisable to have a larger data sample i e window size than the default Command Line Script cfar py image nc output nc width height looks pfa intensity tilesize overlap tmpdir Python Procedure output_array filter image array method cfar width height
142. nable test is to compare the classification probability as will be discussed below It is instructive to compare ML segmentation results with those derived using the span measure shown in Figure 6 28 Figure 6 28 a shows the ML segmentation using the normal parameters discussed later When the segment boundaries are overlaid onto the original classes in Figure 6 28 b two things are immediately obvious Firstly the segment boundaries are closely aligned with the region boundaries This means that there is little blurring of the properties of the different classes Secondly quite large segments are reconstructed Often a complete region of the original scene is represented by a single segment Even where large homogeneous regions are broken up in segmentation the individual segments are sufficiently large that the statistical properties will be well determined On the other hand there is no appreciable evidence of the field structure in the segmented span image in Figure 6 28 c revealing that there is very little information carried by the span measure Indeed the overlay of the segmentation boundaries onto the original classes illustrated in Figure 6 28 d confirm the inadequacy of the span measure in this context The Lower Limit corresponds to attempting ML polarimetric classification on individual pix els There is no reduction in the statistical uncertainty so that speckle noise would be expected to dominate Because of this much
143. nc This can be either a single or a multiple layered real or complex valued image de pending on the chosen method A polarimetric image should be given as a layered complex valued image output nc The coloured region map output image is of the same dimensionality as the input im age if the original image was real valued The coloured region map output image is single layered in the case of a complex valued original image method The measure to be used to colour the regions default me an This string should be one of mean Mean of values within the segment This measure may be applied on either real or complex images For real images the mean value is given as the intensity mean or the square root of the intensity mean depending on whether the data given is intensity or amplitude respectively inverse Inverse 1 mean of mean option bt Blacknell Tough measure within each segment bt I log I I log where I denotes the intensity values This measure is used in the textured segmentation routine texseg This measure can only be prescribed for real images 29 Mar 10 NASoftware Ltd 129 APPENDIX A InfoPACK HIGH LEVEL ROUTINE DOCUMENTATION gorder Gamma order parameter of segments The estimated order of the gamma distribution best fitting the data in the segment This measure can only be prescribed for real images korder K order parameter of segments The estimated order of the K distribution best fitting the data
144. ngle layer image Those regions in class 1 will have value 1 regions in class 2 will have value 2 and so on means Image of a priori class means per layer The class means should be a single layered real valued image of height equal to the number of layers The number of classes is then assumed to be equal to the width of the given image For complex data the width of the image is again equal to the number of classes but to store the covariance matrix the height is the number of layers squared orders Optional image of a priori class order parameters per layer The class orders should be a single layered real valued image of height equal to the number of layers The number of classes is then equal to the width of the class orders image scaling integer default 0 Flag to specify whether there should be scaling 1 or not 0 The purpose of the scaling option is to allow the temporal sequence for a given segment to be scaled to match the reference pattern in temporal signature classification The absolute value of RCS is there fore normalised out and classification depends on the relative changes within the sequence alone 29 Mar 10 NASoftware Ltd 133 APPENDIX A InfoPACK HIGH LEVEL ROUTINE DOCUMENTATION This option is valuable when members of the same class have differing absolute RCS due to factors such as ground slope intensity integer default 0 Assume the data is amplitude square root intensity tilesize integer default 512 A
145. nteger default 1 Line drawing thickness linespoints list default linespoints square dot circle cross plus triangle triangle down Draw a marker at each data point No spaces around commas allowed markersize integer default 1 Size of marker xrow integer default None Row of data file to use for x axis points If not specified use the values 0 1 2 yrow integer default None Rows of data file to use for y axis points Default is to use all rows of input file transpose integer default 1 Transpose the data array so that points for the graph are read from each line of the file rather than along a single line 184 NASoftware Ltd 29 Mar 10 B 4 IMAGE INFORMATION ROUTINES save string default None Saves the plot to the specified filename Use the extensions bmp xpm jpg png to select file type Further Comments Zooming controls with mouse e Left mouse double click reset zoom e Right mouse click zoom out centred on click location e Left mouse drag Zoom box Command line options may be specified in a file one per line and used with the config option For instance xgraphtool py config pdf ini with pdf ini containing xrow 0 yrow 1 y min title Histogram yaxis Count xaxis Deviations grid 1 style solid thick 2 29 Mar 10 NASoftware Ltd 185 APPENDIX B LOW LEVEL AND UTILITY ROUTINE DOCUMENTATION B 5 Classification Routines B 5 1 colclass Purpose Converts a
146. oPACK routines see chapter 4 e A detailed guide to the InfoPACK GUI see chapter 5 e A set of Case Studies illustrating the use of InfoPACK see chapter 6 e Complete and detailed specification for all the routines see appendices A and B 2 NASoftware Ltd 29 Mar 10 1 1 INSTALLING AND RUNNING InfoPACK 1 1 Installing and Running InfoPACK Installing InfoPACK Hardcopy installation instructions are delivered with the software cdrom Alternatively installation instructions specific to each supported operating system are in the file INSTALL txt which can be found in the top level installation directory or alternatively in the top level distribution directory which is what is delivered in compressed format on the software cdrom Running InfoPACK The routines can be accessed in several easy to use ways From the Graphical User Interface The GUI provides a point and click environment from which any of the InfoPACK routines can be invoked Successive routines may be invoked in this way For details on how to invoke the GUI for each supported operating system see the file README txt which can be found in the top level installation directory For each routine the GUI presents the user with a dialogue box to help specifying the pa rameters accepted by the routine For each routine one or more default parameter sets are available for example the segmentation routine segment provides the choice of fast default sensitive se
147. oPACK routines on your data directly Alternatively your existing GIS software may be able to export the data into one of the formats supported by GDAL If you are still unable to process your data with InfoPACK then there are a variety of methods that may be used to convert the data 3 1 The NASoftware format InfoPACK uses a simplified version of the Unidata NetCDF http www unidata ucar edu packages netcdf file format for its default for mat NetCDF is a general purpose format that allows InfoPACK to save intermediate results without loss of accuracy However full use of this flexibility would actually complicate the storing of single and layered images so we use a subset of the functionality known as the NASoftware format If your software allows you to write NetCDF files then see Appendix D for details of how to write NetCDF files in the NA Software format 3 2 Built In File Conversion Facilities All of the InfoPACK routines will read any of the formats supported by GDAL but not all of these support writing so the default is to write images in the NA Software format NetCDF files can be converted to other standard image formats using the Export Image item on the main File menu of the InfoPACK GUI see section 5 3 1 for more details 3 2 1 File Import Conversion to NetCDF Format Conversion to NetCDF format is provided via the Import item on the main File menu This supports a wide variety of standard image formats You will
148. of the Earth This process known as geo coding can also alter the statistical properties of the image We recommend geo coding after processing the image To overcome the problems caused by oversampling the image should be resampled before being processed Section 6 5 gives a detailed discussion of the effect of correlation and how to remove it from both intensity amplitude and complex single look imagery both when the degree of oversampling is known and when it is not known a priori but must be estimated from the data 2 9 Reducing the Image Size Many of the algorithms in InfoPACK make use of a physically correct model derived from the image formation process that produces SAR images This is particularly true of the speckle reduction and edge detection functions The more the statistical properties of an image deviate from those of a SAR image the less optimally these algorithms will perform Because of this it is important to resize or average down images as intensity data and not as amplitude data We provide the utilities see appendix B to help you do this 2 10 Sources of SAR Data Shuttle Imaging Radar SIR experiments SIR A flew on the space shuttle Columbia and collected data on 12 November 1981 It is L band HH polarised National Space Science Data Center Coordinated Request User Support Office NASA Goddard Space Flight Center Code 633 Building 26 Room 113 Greenbelt MD 20771 USA Tel 1 301 286 6695 Fa
149. of 3 channel polarimetric data would be stored as a 15 layer complex image and the mtpol 3 parameter would be used The default value of mt po1 0 uses all of the layers as a single polarimetric image tilesize integer default 512 A tilesize of 512 x 512 is used overlap integer default 48 To prevent obvious joins between tiles an overlap of 48 is used tmpdir directory default None Intermediate results are written to a system temporary directory Further Comments If you are interested in the individual changes found between adjacent pairs of layers in the input image use routine change detection position 136 NASoftware Ltd 29 Mar 10 A 4 HIGH LEVEL ROUTINES A4 9 change_detection_position Purpose For each region in a given region map change detection position looks for significant change over a series of original speckled images but in more detail than does change detection The routine detects large scale changes between adjacent pairs of images large scale implies that it is not tailored to detecting differences in point targets between images The images to be scanned for changes must be provided as layers in a single multi layer image if your images are separate use layer_combine to merge them into a single 3 D image The routine uses the segmentation of each layer to carry out its analysis and this must be provided too In most cases the joint segmentation produced by segmenting the multilayer image is what
150. of 3 channel polarimetric data would be stored as a 15 layer complex image and the mtpol 3 parameter would be used The default value of mt po1 0 uses all of the layers as a single polarimetric image tilesize integer default 512 A tilesize of 512 x 512 is used overlap integer default 48 To prevent obvious joins between tiles an overlap of 48 is used tmpdir directory default None Intermediate results are written to a system temporary directory Further Comments None 128 NASoftware Ltd 29 Mar 10 A 4 HIGH LEVEL ROUTINES A4 5 colseg Purpose Given an original image file and a region map of this image colseg will produce a segmented image with the regions coloured according to one of a number of measures Command Line Script colseg py region_map nc image nc output nc method looks intensity mtpol tilesize overlap tmpdir Python Procedure output_array colseg region_map_array image_array method looks intensity mtpol region_map nc The file containing the region map image This image should be one of either e A single layered real valued image of dimensions equal to the dimensions of the first layer of the original image image nc e A real valued image of exactly the same dimensionality as image nc In the case of a multiple layered region map the algorithm takes the intersection of the given layers to be the region map on which to perform the classification image
151. of InfoPACK 29 Mar 10 NASoftware Ltd 49 CHAPTER 5 USING THE INFOPACK GUI 5 4 The OpenEV Image Viewer InfoPACK provides the OpenEV image viewer to view raw or processed images This open source viewer is well suited to handling large images of a wide variety of formats It also takes account of geo referencing information when provided and can produce overlay images for example map data overlaid upon co registered SAR data Documentation for OpenEV is available from OpenEV s Help menu 5 4 1 Classification Images In order to save disk space InfoPACK classification images are stored as byte data OpenEV however interprets byte images as greyscale bitmaps and so the classes 1 2 3 get displayed as black There are two workarounds to this problem e Use the colour classes routine colclass to convert the classes to red green blue and so on e Use the image equalisation buttons See Figure 5 1 in OpenEV to remap the class values to a more useful greyscale aluolulala Figure 5 1 OpenEV image equalisation buttons 5 5 Using the InfoPACK Shell Window The Shell Window provides access to an interactive mode This window will open as part of the main screen when you fire up the InfoPACK IDE and packages needed to provide access to InfoPACK facilities are automatically loaded You can type any python statement into this window but most users will need only those facilities provided specifically by InfoPACK and documen
152. ommand Line Script None provided Python Procedure save array in image out gis woffset hoffset format Parameters gis Optional GIS information to be saved to the image file header This variable is returned from a call to the load function woffset integer default None If woffset or hof f set are set it is assumed that emptyfile has already been called to initialise the image size hoffset integer default None If woffset orhof fset are set itis assumed that emptyfile has already been called to initialise the image size format The following formats are supported VRT Virtual Raster GTiff GeoTIFF HFA Erdas Imagine Images img NASoftware Ltd NASoftware Ltd NetCDF Images nc SAR_CEOS CEOS SAR Image CEOS CEOS Image ELAS ELAS AIG Arc Info Binary Grid AAIGrid Arc Info ASCII Grid SDTS SDTS Raster 176 NASoftware Ltd 29 Mar 10 B 3 FILE HANDLING FACILITIES DTED DTED Elevation Raster PNG Portable Network Graphics JPEG JPEG JFIF MEM In Memory Raster JDEM Japanese DEM mem GIF Graphics Interchange Format gif ESAT Envisat Image Format N1 BSB Maptech BSB Nautical Charts XPM X11 PixMap Format xpm PNM Portable Pixmap Format netpbm DOQ1 USGS DOQ Old Style DOQ2 USGS DOQ New Style ENVI ENVI hdr Labelled EHdr ESRI hdr Labelled PAux PCI aux Labelled MFF Atlantis MFF Raster MFF2 Atlantis MFF2 HKV Raster EFF Eosat Fast Format
153. on the average intensity in each ground truth region over the sequence Joint Segmentation classification on the average intensity in each segment over the sequence 3x3 window classification on the average intensity over a 3x3 window over the sequence Lower Limit classification on the intensity in each pixel over the sequence The first stage 1s to perform the pre processing stage described above Next we train the tempo ral signature by estimating the properties of each class over the regions defined by the training mask in Figure 6 11 these properties are next used to provide Maximum Likelihood classification over the different temporal signatures The underlying model for the speckled data is that it is gamma distributed with order 3 corresponding to 3 look SAR and varying mean from image to image The likelihood we wish to calculate therefore corresponds to data having order 3 and pre trained mean over the sequence The likelihood of each segment belonging to each of the temporal signature categories can be calculated The ML solution for each segment then denotes the assigned class for that segment When this analysis is applied to the sequence of images we obtain the results shown in Figure 6 4 2 Corresponding average probabilities of correct classification are summarised in Table 6 4 2 With single pixels a the classification is clearly similar to the expected result in Figure 6 9 How ever there are spots of misclassified data visible
154. ords in blue 5 5 2 Indentation The Python scripting language uses indentation levels to define program blocks rather than ex plicit begin end syntax The command line facilities provide limited automated support for indentation here is an example that creates a function to produce factorials gt gt gt def factorial num result 1 while num gt 0 result result num num num 1 return result gt gt gt If you type in the above example you will find that when you hit Enter after the def clause the next line is automatically indented for you as you enter a new block When within a block you will be automatically placed at the same level as the previous statement above this happens for the num num 1 statement Each press of the Backspace key will back you out of a level of indentation above it needs to be used before the return statement After a pass return break continue or raise statement you will automatically be backed out of a level of indentation Hitting Enter when on the new line after return result will take you out of the function definition to a command prompt The IDLE IDE provided for scripts file new script and file Open script menus provides fully automated support for indentation 5 5 3 Command History A feature that is worth knowing because of the amount of typing it can save you is the command history mechanism Placing the cursor at the end of any line and pressing En
155. ore we shall compare results after the same four types of processing namely Upper Limit change detection on the average intensity in each ground truth region over the se quence 68 NASoftware Ltd 29 Mar 10 6 3 CHANGE DETECTION BETWEEN INTENSITY IMAGES Image number Figure 6 3 Plot of the mean intensity in each of the classes from Figure 6 2 during the sequence of images 29 Mar 10 NASoftware Ltd 69 CHAPTER 6 CASE STUDIES Figure 6 4 Predicted regions of change white and no change black in the image sequence Joint Segmentation change detection on the average intensity in each segment over the sequence 3x3 window change detection on the average intensity over a 3x3 window over the sequence Lower Limit change detection on the intensity in each pixel over the sequence Change detection The output of this method is expected to be a map of those regions in the sequence where changes have been detected compared with those regions where no changes were found Only two hypothe ses are possible for the entire sequence From figures 6 2 and 6 3 we note that changes are expected in the red blue and green classes only Thus the predicted changes are illustrated in Figure 6 4 The ML test for homogeneity with a gamma distributed PDF is given by the normalised log measure U log 1 log I 6 1 where the bars indicate averaging over all the pixels in a given segment for the complete se quence o
156. orm and operating system e The commands which produced the error and a description of the behaviour e If possible the images involved so that we can try to reproduce your results To report bugs via e mail send a message to support nasoftware co uk Other Addresses You can also contact us by telephone fax or mail NASoftware Ltd 1 Prospect Road Prenton Merseyside CH42 8LE United Kingdom Phone 44 151 637 1422 Fax 44 151 550 7830 29 Mar 10 NASoftware Ltd 5 CHAPTER 1 INTRODUCTION 1 4 Acknowledgements We are grateful to e the National Environmental Research Council for permission to make use of some of their very nice DLR ESAR high resolution SAR data for some of the Case Studies in this docu mentation e Dr Rod Cook and N A Software Ltd for extensive discussions during the design and imple mentation of InfoPACK e The many users whose comments and encouragement have helped develop Release 2 6 NASoftware Ltd 29 Mar 10 Chapter 2 Synthetic Aperture Radar Images A radar system illuminates an area with microwaves and records the strength and travel time of the returned signals This allows the distance or range of the reflecting objects to be determined As in an optical instrument the resolution of such a system is affected by the size of the aperture a larger aperture gives a finer resolution However scaling up to radar wavelengths has problems For example to achieve an azimuth resolu
157. orrelation properties a consequence of the segmentation tending to reconstruct the individual speckle blobs This is further borne out by the relative variance of Figure 6 5 2 c which is 14 7 compared with the ENL of 4 0 6 5 3 Speckle correlation properties The SAR image in Figure 6 13 was of a detected amplitude image In some cases the single look complex SLC image is available In this case we can perform a detailed study of the instrument correlation properties as distinct from any cross section fluctuations Consider the autocorrelation function ACF of the received complex field at lag k which is given by lt BoE gt A Eo E gt k sek y lt EoF gt lt ERE gt lt I gt provided the field E is a stationary ergodic random process Thus constructing the ACF of the complex field yields a correlation function with a value of 1 0 at zero lag and decaying to zero as the lag increases provided the field has zero mean In fact gg k corresponds to the ACF of the instrument function so that gE k h 0 h k If the SLC image is detected and the ACF of the resulting intensity constructed this will be given by lt olk gt _ lt Tolk gt 1 lt I gt lt P gt gr k m This can be split up into four separate terms Firstly there is the flat background of lt 1 gt Secondly there is the interference term corresponding to random scatterers within the instrument width This would be the
158. ot be any problems with automatic type conversion in C if the header file is not included For those routines that return an int the return value will be one of INFOSAR_OK 0 INFOSAR_NO_MEMORY 1 INFOSAR_INVALID_DIMS 1 2 INFOSAR_OUT_OF_RANGE 3 INFOSAR_INVALID_ARGS 4 INFOSAR_INVALID_LICENCE 5 where INFOSAR_OK The routine has successfully completed INFOSAR_NO MEMORY There was an error mallocing enough memory for the routine to com plete INFOSAR_INVALID_DIMS The width or height is too small to process the data INFOSAR_OUT_OF_RANGE You have asked for an invalid combination of options For instance running a median filter on segmented data INFOSAR_INVALID_ARGS One of the arguments passed to the routine has an invalid value INFOSAR_INVALID_LICENCE There was an error either in finding or reading the licence file C 3 1 Progress Function Many of the InfoPACK functions accept a function argument int progress int flag const char format This progress routine is used to return error or information message back to the calling applica tion It is also occasionally called with flag set to INFOSAR_PUR_BLIP to indicate progress of the InfoPACK routine Other values for flag are INFOSAR_PUR_MEMERROR 3 INFOSAR_PUR_ERROR 2 INFOSAR_PUR_WARNING 1 INFOSAR_PUR_BLIP 0 INFOSAR_PUR_M
159. pe of data must be specified Command Line Script raw2nc py raw_in image_out type byteswap header layers verbose Python Procedure None provided Parameters type string must be specified Type is one of D Complex Complex64 F Complex0 Complex16 Complex32 Complex8 a Float Float64 Float0 Float16 Float32 Float8 1 Int b Int0 Int8 s Intl6 37 Int32 byteswap integer default 1 When set to 1 the routine transforms big endian to little endian data Depending on how your images are constructed it is possible that you will need to reverse the order of the bytes To see whether this is the case it is suggested that you look at the statistics of the converted image to check that the values it reports look sensible if not you probably need to swap the byte order A visual inspection may also prove sufficient 29 Mar 10 NASoftware Ltd 175 APPENDIX B LOW LEVEL AND UTILITY ROUTINE DOCUMENTATION header integer default 0 When set to 0 the routine skips the header bytes layers integer default 1 To set the number of layers Band sequential BSQ format verbose integer default 0 When set to 0 the routine turns off information messages B 3 17 save Purpose Saves the Numeric array and class containing the GIS information in an NASoftware Ltd NetCDF file Options are available to save the array at given offsets and in other GDAL compatible output formats C
160. peckle oo ns nda Ge ee Reha cres ee ee oa ee dees 139 AAAI Sea oone wy ee ho BALE a oe HP ee wre Ee ie a 141 A412 filter_bt filter_det filter_enl filter_gorder filter_korder Pr leer meen ELISA a a baw es bk G RE ES 143 A A ce en a EE O a e ee ee E aT 144 B Low Level and Utility Routine Documentation 146 B 1 Documentation Conventions e go e e a soe ee we ew we a 146 BIL Prao cis a rata ee ae e ae Go a e We a 146 B 2 Image Manipulation Routines oo 2 20 2 002 ir ee ee ees 148 B21 gals os ok wok Ge ek A ew A Re RD a 148 B22 BSEWSED lu bd Lagat beet eee eh LAOS oh tee ee wes 148 B23 A II eh aed na ee ee Ne we Ee ok Boel 149 B24 DICTE oho kaw Sad bee baa haw Oa ew a ed ea ia 149 B293 crepe bo Ge Bebe Ba de Ga be he booed de aoe Bs 150 Beo Sa ss a Rae A ROR a a ee ae ee A ee 151 B27 Sagara e sone we he a A we ee ewe ee o ai i d 151 B26 ELULIMSS SA esos So Baie de walk Ge ee ed 152 B29 aa aman as A BO a es ow a ay ew a a AR ee 153 B210 ELL pol rand id me ae ee a 153 Bill Geeter seg tocne na aima 2 OR Ee ROR Oe Ree e Se doe BM 154 Beds Tayar Sie oe oboe es RR he oe a eS 155 Bilal do id hee ea Od bad a Ok ESS Bk 156 Bll LE pe hak ae bee a RAR A Beka ee ew a Gs 156 B21 median yd a des a a ee a ee ae eh Le di 157 B216 POD e eee Gig a A Ee Se i ee ee A e i a 157 BLI BO ae ae eS oe eet a Oe ee ele Oe ee Oa ed e 158 BZ POLLO tb be eA Ow Ree RAE SAO ee Ae 158 B219 pelyscan s escra diapa ed dk wt bee wed AA 159 B220 DOSIS ora O heh Se
161. pect and is in stark contrast to the original version in Figure 6 6 b There are no speckle blob sized segments in Figure 6 6 1 b which should be compared with the very large number visible in 6 6 b Figure 6 6 1 c shows the ratio of the original intensity to the reconstructed cross section after resampling This has the expected scale of correlations in the speckle blobs similar to those in Figure 6 6 d and completely different from the ratio image obtained from the original data shown in Figure 6 6 c and repeated here for comparison in Figure 6 6 1 d Finally the ENL derived from the ratio image is 4 03 consistent with the defined value This indicates that the despeckling derived in this way can be trusted Figure 6 6 1 a Figure 6 6 1 b 96 NASoftware Ltd 29 Mar 10 6 6 CORRELATED IMAGES AND DESPECKLING Figure 6 6 1 c Figure 6 6 1 d Figure 6 6 1 Segmentation of the resampled original image from Figure 6 6 a Image resampled by 3x3 b segmentation of a c ratio of a to b d corresponding ratio for original data from Figure 6 6 c 6 6 2 Conclusions This study has demonstrated the following aspects of applying despeckling to correlated data e If despeckling is applied directly to SAR intensity images that have been sampled at spacing less than the speckle correlation width which is determined by the resolution of the system the speckle blobs are faithfully reconstructed rather than averaged out
162. plex number real Real part imag Imaginary part sd float default 2 0 min list default None This can either be a single value or a list eg min 0 3 7 2 Note no spaces allowed around the comma or in python min 0 3 7 2 max list default None As min this can be a single value or a list of values verbose integer default 0 If set to 1 then the estimated min and max values are printed out tilesize integer default 512 A tilesize of 512 x 512 is used tmpdir directory default None Intermediate results are written to a system temporary directory 29 Mar 10 NASoftware Ltd 179 APPENDIX B LOW LEVEL AND UTILITY ROUTINE DOCUMENTATION B 4 Image Information Routines B 4 1 countseg Purpose Returns the number of segments in an image Command Line Script countseg py image_in Python Procedure segments countseg array_in B 4 2 gdalinfo Purpose A utility to give information about a GDAL Geospatial Data Abstraction Library supported dataset Information provided includes geocoding informarion image dimensions and data type Command Line Script gdalinfo image in Python Procedure None provided B 4 3 histo Purpose Writes the histogram of the input image to an ASCII file Command Line Script histo py image_in text_out nbins layer min max sd norm cmp1l2real 180 NASoftware Ltd 29 Mar 10 B 4 IMAGE INFORMATION ROUTINES Python Procedure
163. proves the result The accuracy could be further improved by increasing the size of the window However there is then a problem of averaging out small regions What is required is a method that adapts the region size to the data this is segmentation Segmentation followed by classification yields a considerable improvement since it provides a filter whose window size and shape is adapted to the data The fact that the segments are not matched to the ground truth regions is a consequence of speckle noise In this instance we know that the simulation was for uniform properties within each region The speckle statis tics have spoilt this Larger regions could be grown by changing the curvature parameter However this would tend to reject small regions Only the full ML polarised segmentation method offers a significant improvement over the 3x3 window It reduces the false classifications from about 20 to about 6 The Upper Limit as one would expect provides the highest statistical accuracy There are three important caveats to note however 1 The result is based on the assumption that the properties within the region are indeed homogeneous In real applications this will not generally be the case 2 The Upper Limit is incapable of showing if there are inhomogeneities within the defined reference region The consequence of any such variation is a degradation in classifica tion performance 3 The classification can only be applied wher
164. rameters set to the default values The looks measure parameter Figure 4 14 shows the result of increasing the looks measure used by the routine despeckle Fifty iterations were used so compare it with Figure 4 13 When the looks measure parameter is increased the algorithm interprets the input data as being less affected by noise This means that despeckle will be more likely to interpret fluctuations caused by speckle as genuine jumps in the cross section The output is less smooth as a result We recommend that normally you let the routine estimate the looks measure for itself 36 NASoftware Ltd 29 Mar 10 4 5 SPECKLE REDUCTION Figure c Figure d Figure 4 13 a Unprocessed SAR image b despeckled with looks measure 1 and 10 iterations c 50 iterations d 250 iterations 29 Mar 10 NASoftware Ltd 37 CHAPTER 4 IMAGE PROCESSING WITH InfoPACK Y A gt Figure 4 14 a Figure 4 14 b Figure 4 14 Despeckling the image of figure 4 13 a with 50 iterations and a looks measure 2 b looks measure 4 38 NASoftware Ltd 29 Mar 10 4 6 POINT OBJECT DETECTION Figure 4 15 a Figure 4 15 b Figure 4 15 a Original image containing potential point objects b Image processed by cfar to detect point objects 4 6 Point Object Detection Point objects are bright areas of an image only a few pixels in size They correspond to highly reflective scatterers on the ground often metal objects such as ve
165. re 6 6 a A detail from the original image shown in Figure 6 13 b The image despeckled using despeckle c The residual speckle d The expected residual speckle If we take the ratio of the original intensity to the despeckled version the result should have the properties of random speckle with the expected correlation properties and relative variance In Figure 6 6 c we show this residual speckle derived from Figures 6 6 a and b Figure 6 6 d 29 Mar 10 NASoftware Ltd 95 CHAPTER 6 CASE STUDIES comprises a simulated speckle image having approximately the correct correlation properties It is clear that 6 6 c fails to show the expected speckle correlation properties a consequence of the despeckling algorithm tending to reconstruct the individual speckle blobs This is further borne out by the relative variance of Figure 6 6 c which is 37 9 compared with the ENL of 4 0 6 6 1 Resampling to remove speckle correlations Since the original sampling was at approximately one third of the instrument resolution we would expect following the previous study that speckle correlations could be removed by averaging and shrinking the intensity data by a factor of 3 in each direction When the averaging and resampling by 3x3 is applied to the original image the image shown in Figure 6 6 1 a is obtained while Figure 6 6 1 b shows the corresponding despeckled reconstruction It is obvious that this contains large uniform regions as one would ex
166. ree of speckle noise which still requires further image processing to identify regions of similar properties i e classification Segmentation can lead directly into this stage Furthermore there is no concept of an edge in this approach whereas most information data bases such as maps require regions of similar content to be identified and their edges delineated The final drawback of this approach is that it actually misrepresents the speckle model If we 29 Mar 10 NASoftware Ltd 101 CHAPTER 6 CASE STUDIES Figure 6 22 Despeckled version of Figure 6 17 construct the residual ratio assuming that the despeckled output corresponds to the underlying RCS we obtain the result shown in Figure 6 23 While the uniform regions of the original image Figure 6 17 yield a realistic speckle appear ance all structure in the image affects the residual ratio strongly This is a consequence of the 3 x 3 window over which correlations in the data are identifed and exploited in the reconstruction pro cess When this is compared with the corresponding result following segmentation in Figure 6 19 it is clear that the process introduces considerable distortion into the reconstruction This is made apparent when the ENL is estimated yielding a value of 4 26 compared with 4 03 for segmentation This indicates the extent to which this method fails to satisfy the speckle model The effect is made still more obvious if the PDF of the residual ratio in
167. research has depended on taking 3x3 averages of the covariance matrix before classification This reduces statistical uncertainty at the expense of blurring region boundaries Thus the impact of such a process is scene dependent The purpose of the comparison is to reveal 1 the importance of using the full complex polarimetric data 2 the considerable advantage of segmentation methods over the 3x3 window and the Lower Limit 6 8 3 Comparison of classification methods Optimum classification of polarised images is based on the covariance matrix which contains all the information in the scattering process The ML classifier can be defined in terms of the likelihood In R Trace R gt R 29 Mar 10 NASoftware Ltd 109 CHAPTER 6 CASE STUDIES Figure 6 27 c Figure 6 27 d Figure 6 27 Segmentation of a single polarised image using different shape parameters a 0 1 b 0 01 c 0 001 d 0 0001 110 NASoftware Ltd 29 Mar 10 6 8 SEGMENTATION AND CLASSIFICATION OF POLARIMETRIC DATA Figure 6 28 c Figure 6 28 d Figure 6 28 a ML segmentation b Overlay of the segmentation boundaries on the original classes c Segmentation of the span measure d overlay of the span segment boundaries on the original classes 29 Mar 10 NASoftware Ltd 111 CHAPTER 6 CASE STUDIES where R is the reference covariance matrix for class C and Rc denotes its determinant and Re denotes the estimated covariance
168. rform the classification image nc This can be either a single or a multiple layered real or complex valued image de pending on the chosen method A polarimetric image should be given as a layered complex valued image output nc The deflecked output image is of the same dimensionality as the input image if the original image was real valued The deflecked output image is single layered in the case of a complex valued original image size integer default 3 Regions with this number of pixels or less will be removed from the seg mentation mtpol integer default 0 Polarimetric data is stored as layers of complex data however this is the same file format as multi temporal polarimetric data To distinguish the two the mtpol parameter is used For instance 5 days of 3 channel polarimetric data would be stored as a 15 layer complex image and the mtpol 3 parameter would be used The default value of mt po1 0 uses all of the layers as a single polarimetric image tilesize integer default 512 A tilesize of 512 x 512 is used overlap integer default 48 To prevent obvious joins between tiles an overlap of 48 is used tmpdir directory default None Intermediate results are written to a system temporary directory 29 Mar 10 NASoftware Ltd 131 APPENDIX A InfoPACK HIGH LEVEL ROUTINE DOCUMENTATION Further Comments This routine is run automatically after segment with polarimetric data if the size parameter is set 132 NASoftware L
169. rld War 11 These names have entered radar jargon and are commonly used They are listed in the table in Figure 2 1 For wavelengths in the region 30 75cm the term P band is often used Because of atmospheric absorption effects only certain sharply defined windows in the spectrum can be used for imaging purposes Band Frequency Range Wavelength Range P 0 4 1 GHz 30 75cm L 1 2 GHz 15 30cm S 2 4 GHz 7 5 15cm C 4 8 GHz 3 7 7 5cm X 8 12 GHz 2 5 3 7cm Figure 2 1 The IEE names for radar frequency bands 2 4 Polarisation When the radar transmits microwaves they are polarised into a vertical or horizontal plane The antenna also detects the scattered signal in a particular polarisation not necessarily the same as that transmitted This gives four possible images of the same scene They are labelled HH HV VH and VV for Horizontal Vertical lt transmit gt lt receive gt However if the SAR is correctly calibrated the backscatter reciprocity principal implies that the HV and VH images are the same Instruments such as the JPL AirSAR can image a scene in two orthogonal polarisations simul taneously This is known as quadrature polarisation It enables the complete scattering matrix to be computed Other instruments such as the Envisat ASAR system can produce data for both VV and HH polarisations though not the complete scattering matrix Fully polarimetric data will also be available from future planned satellit
170. roughout the sequence or not This is based on temporal texture and uses the normalised log texture for the sequence in each element The second performs a maximum likelihood search for the position of a step change in each element over the sequence It compares the likelihood of a step at all points with there being no change Thus this offers additional information compared with the first method in that it both detects the presence of a change and determines its position in the sequence To demonstrate these approaches we simulate a sequence of ten images with six different classes shown in Figure 6 2 whose cross section varies from sample to sample as shown in Figure 6 3 From Figure 6 3 we see that there is a change of 5 in the RCS of the red class at image number 4 one of 1 in the blue class at image number 5 and one of 2 in the green class at image number 6 The purpose of this study is to detect these changes 6 3 2 Comparison of classification methods As indicated above there are two different Maximum Likelihood tests that can be performed to test for changes The first is to test the hypothesis that there has been a change in each region or not at any point in the sequence We shall call this change detection The second method addresses the more sophisticated question of where any change is detected Thus it provides information both on the existence and position of such a change We shall name this positional change detection As bef
171. rval the speckle in the resultant image is reduced The reduction is not equal to the number of pixels averaged of course because the speckle is correlated However this averaging slightly mitigates against the loss of information resulting from degraded resolution In practice we find that taking samples at a spacing corresponding to about 20 correlation along each axis achieves the desired effect In this example therefore the resampling spacing should be 2 1 and 2 9 pixels respectively Thus resampling corresponds to an average and shrink operation on the intensity over 2 and 3 pixels in range and azimuth respectively Let us now investigate how this applies to the SLC imagery Figure 6 5 3 c Figure 6 5 3 d 90 NASoftware Ltd 29 Mar 10 6 5 HANDLING CORRELATED IMAGES Figure 6 5 3 f Figure 6 5 3 Comparison of original and resampled segmentation a original image b segmentation of original image c resampled image d segmentation of resampled image e ratio image for original data f ratio image after resampling In Figure 6 5 3 a we show a detail from the intensity derived from the SLC image The seg mented version using default parameters illustrated in Figure 6 5 3 b again shows the effects of speckle correlations Many of the regions in this image are clearly resulting from original speckle blobs Both these images have been rescaled for a square display Next we apply the resampling derived above
172. ry part of complex data conj Conjugate of complex data abs Amplitude modulus of complex data power Power modulus squared of complex data phase Phase of complex data format The following formats are supported VRT Virtual Raster GTiff GeoTIFF HFA Erdas Imagine Images img NASoftware Ltd NASoftware Ltd NetCDF Images nc SAR_CEOS CEOS SAR Image CEOS CEOS Image ELAS ELAS AIG Arc Info Binary Grid AAIGrid Arc Info ASCII Grid SDTS SDTS Raster DTED DTED Elevation Raster PNG Portable Network Graphics JPEG JPEG JFIF MEM In Memory Raster JDEM Japanese DEM mem GIF Graphics Interchange Format gif ESAT Envisat Image Format N1 BSB Maptech BSB Nautical Charts XPM X11 PixMap Format xpm PNM Portable Pixmap Format netpbm DOQ1 USGS DOQ Old Style DOQ2 USGS DOQ New Style NVI ENVI hdr Labelled Hdr ESRI hdr Labelled Aux PCI aux Labelled FF Atlantis MFF Raster FF2 Atlantis MFF2 HKV Raster FF Eosat Fast Format uJiBAS Fuji BAS Scanner Image FIT FIT image USGSDEM USGS Optional ASCII DEM GXF GeoSoft Grid Exchange Format THE Es Ve EE The default output format is NASoftware Ltd tilesize integer default 512 A tilesize of 512 x 512 is used 29 Mar 10 NASoftware Ltd 167 APPENDIX B LOW LEVEL AND UTILITY ROUTINE DOCUMENTATION tmpdir directory default None Intermediate results are written to a system temporary directory Further Comments The command con
173. s Sample data is included with the InfoPACK distribution in the images subdirectory of the main installation directory Also located there is a README file that explains in some detail how to process the sample data with several different routines More details on running InfoPACK and commentary on the main routines can be found in chap ter 4 Appendices A and B contain exact syntaxes for the Python Scripts and Functions including descriptions of all parameters and their default values 29 Mar 10 NASoftware Ltd 3 1 2 The Licence The Software The Software means the InfoPACK toolset developed by NASoftware Ltd including all supporting software and documentation supplied with it Your Rights as Purchaser The purchase of this manual and accompanying software from NASoftware Ltd or from an autho rised dealer confers a licence to use the software on one machine single user licence or simultane ously by a number of users multi user licence You may make as many backup copies as you need for security and use the software on any suitable machine Apart from this right to make backup copies all copyright and rights of reproduction are reserved No part of this or other documentation may be reproduced or used in any form or by any means including photocopying without the prior written consent of NASoftware Ltd Please keep the original media as proof of purchase and quote the product name and version number in any correspondence
174. s are also coloured according to the Blacknell Tough measure Joint segmentation is performed when there is more than one image layer Full Python script and function syntaxes are given in section A 4 3 4 1 5 The colseg routine colseg is a routine that will process a segmented image to colour regions according to one of a number of measures This is useful both visually and for further processing purposes since it allows variation in the scene to be highlighted according to different properties Figure 4 7 shows the result of colouring the segmentation of figure 4 6 by two different methods The colseg algorithm can be performed upon images with more than one image layer Full Python script and function syntaxes are given in section A 4 5 26 NASoftware Ltd 29 Mar 10 4 1 SEGMENTATION Figure 4 7 a Figure 4 7 b 8 8 Figure 4 7 Result of running colseg on the image of Figure 4 6 b with a colouring by mean intensity Note that the difference between clearing and virgin rain forest is much less pronounced than when the regions are coloured using the Blacknell Tough measure as in Figure 4 6 b b colouring by regional Gamma order parameter 4 1 6 The defleck routine It is often the case after a segmentation or a segmentation and merge that you end up with unwanted single pixel flecks across the image By running the Defleck routine you can get rid of these flecks without losing other information 29 Mar 10 NASoftware Ltd
175. s between layers of a multiple layer image for example a multi temporal sequence The routines accessed look for large scale changes between layers of the specified multi layer im age you need to supply also the segmented image Two options are provided Change Detect produces a single layer output image denoting areas in which changes were found between any of the images See section A 4 8 for details of the routine parameters and section 4 3 1 for a longer discussion Change Position Detect detects changes between adjacent layers The output is an im age with one less layer than the input image each output layer contains the changes found between one adjacent pair of layers in the input image This option is useful for example when monitoring the growth of crops over the seasons See section A 4 9 for details of the routine parameters and section 4 3 2 for a longer discussion 5 3 6 The Help Menu Create Licence Fil Use this to input your licence details Without these you will not be able to run any of the InfoPACK routines InfoPACK HTML Gives you access to the InfoPACK HTML documentation InfoPACK PDF Gives you access to the InfoPACK PDF documentation Python Gives you access to the Python documentation Numeric Python Gives you access to the Numeric Python documentation Command List This displays a window listing the python scripts available with InfoPACK About Gives the version number
176. s represent the span measure shown in Figure d which has often been adopted in the analysis of polarised images It too will be shown to result in poor classification performance With the exception of the three individual images or the span measure shown above other measures for polarised segmentation are all based on the covariance matrix of the image This is constructed as shown above using data from the three complex simulated images Ey y Eyy Evy The covariance matrix carries all the information about the polarised image Therefore it is 29 Mar 10 NASoftware Ltd 107 CHAPTER 6 CASE STUDIES important to process the full complex data for each element This is implemented in the Maximum Likelihood classifier and the polarised segmentation algorithm which will be described later This ML polarimetric classifier takes averages of the polarisation covariance matrix over regions which ideally correspond to areas of uniform polarisation properties Various algorithms are used to derive this region map We either treat the single pixel covariance matrix directly apply a 3x3 window to the covariance matrix perform polarised segmentation on the data calculate the Upper Limit by averaging the covariance matrix over the original ground truth mask Two different forms of polarised segmentation will be compared e ML polarimetric segmentation e segmentation of the span measure The first method provides the Maximum Likelihood
177. segmentation solution The latter is included to demonstrate how much poorer is segmentation which only uses incoherent information namely the intensity in each polarisation The ML polarimetric segmentation algorithm requires the operating parameters to be tuned for best performance As we found previously segmentation is controlled by three basic parameters 1 The shape parameter attaches a penalty to excessive curvature in the reconstruction Large values of the parameter allow many changes of direction of the edge Small values bias the reconstruction towards smoother edges Since speckle yields very crinkled edges to regions of similar intensity it is important to adopt a large curvature penalty i e a smaller shape parameter 2 The annealing method has two further parameters attached The first is the stopping param eter to assess when the reconstruction is complete For noisier data one would expect this parameter to be increased Smoother data such as that encountered on averaging speckle should allow a reduced value of the stopping parameter 3 The second parameter in the annealing method is the cooling parameter which controls the rate at which the annealing converges on the final solution If this is made too large the solution may freeze out away from the expected result It is important to derive values for these parameters that yield the best segmentation Since we have started with simulated test data we can tune t
178. shows an example sequence of four multi temporal images The changes detected by change_det ection at the default level of significance are shown in figure 4 1 1 a Full Python script and function syntaxes are given in section A 4 8 43 2 The change_detection_position routine This routine detects changes between adjacent pairs of layers in the provided image the ordering of the layers in the multiple layer image file is therefore crucial The output is a multiple layer image with number of layers equal to the number of adjacent pairs in the sequence of layers in the original image i e number of layers minus 1 Each of these layers indicates where significant changes were found between the image pair This routine is useful for example when monitoring the growth of crops over the seasons Figure 4 11 shows an example application of this routine with respect to the sequence of images of figure 4 10 If you are interested solely in whether there are changes over the whole sequence of layers in the input image use routine change_detection Full Python script and function syntaxes are given in section A 4 9 29 Mar 10 NASoftware Ltd 31 CHAPTER 4 IMAGE PROCESSING WITH InfoPACK de RE AS Figure 4 10 Multi temporal sequence of SAR images 32 NASoftware Ltd 29 Mar 10 4 3 AREA CHANGE DETECTION Figure c Figure d Figure 4 11 Detecting changes in the sequence of images of Figure a Result of change_detection
179. ss section that underlies the speckled image When the observed intensity is divided by the reconstruction the result resembles pure speckle so that the looks measure is equivalent to the ENL It is important to get a good model and the functions are tuned to work with their own estimates The estimates are usually slightly different from the real value They are close when the image consists mainly of fields and get poorer when estimated from a town scene 2 7 Multi dimensional Images A multi dimensional image is a composite of several images of the same scene The component images are registered so that equivalent pixels correspond to the same location on the ground A three dimensional image is often displayed as a colour composite in which the three individ ual images correspond to the red green and blue components Variations in colour reflect differences between the images caused by changes on the ground such as growth of vegetation or change in water content Multi dimensional images can be generated in various ways Complex Data The NetCDF image format treats complex data by storing the real and imaginary parts in separate layers hence formally a complex dataset is stored in a multidimensional image We take this for granted below but for example complex data is usually stored for the separate polarisations in a multi polarimetric image Multi temporal images The images are taken at different times possibly months apart
180. ssessed by segmentation e Joint segmentation is somewhat less sensitive than the Upper Limit but is considerably better than using a 3x3 window which in turn is better than single pixels e Positional change detection is more sensitive than change detection alone and offers addi tional important information namely the position in the sequence of any change detected 76 NASoftware Ltd 29 Mar 10 6 3 CHANGE DETECTION BETWEEN INTENSITY IMAGES Therefore we reccommend that change detection should always adopt the simultaneous de tection of change and its position 29 Mar 10 NASoftware Ltd TT CHAPTER 6 CASE STUDIES Figure 6 9 Test pattern of 6 defined classes used in this study for simulating image intensity 6 4 Temporal Signatures of Intensity Images 6 4 1 Introduction In the previous case study we demonstrated the capability of segmentation to improve the classi fication of multiple simulated 3 look SAR images Here we extend this study to the matching of regions to a known temporal signature a practical example would be the delineation of a set of dif ferent crops whose growth pattern is known This growth pattern signature will often in practice be derived from a previous data or as in this model study from a defined training mask so that the process is one of supervised classification Joint segmentation is well matched to this process since it constructs a set of segments that are common to each image It
181. st Default Sensitive Cooling 0 75 0 75 1 Looks 1 1 1 Maxiters 25 40 50 Intensity 0 0 0 Any of these values can be over ridden by subsequent values on the command line The values fast default sensitive are provided as generally convenient settings but because of the iterative nature of the smoothing process and its data dependence some initial parameter searching may be worthwhile for best results cooling float default 0 75 Determines the rate of cooling in the simulated annealing minimisa tion process A larger value of cooling tends to decrease the time taken by the routine but may increase the likelihood that it gets stuck in a local minimum looks float default 1 The looks measure of the image data If looks is negative use 1 the routine will compute this parameter from the data maxiters integer default 40 The maximum number of iterations complete scans of the image data allowed intensity integer default 0 Assume the data is amplitude square root intensity 29 Mar 10 NASoftware Ltd 139 APPENDIX A InfoPACK HIGH LEVEL ROUTINE DOCUMENTATION tilesize integer default 512 A tilesize of 512 x 512 is used overlap integer default 48 To prevent obvious joins between tiles an overlap of 48 is used tmpdir directory default None Intermediate results are written to a system temporary directory Further Comments None 140 NASoftware Ltd 29 Mar 10 A 4 HIGH LEVEL R
182. stical estimates it may be advisable to have a larger data sample i e window size than the default A lower value of probability of false alarm pf a will mean less edges will be detected Command Line Script scanedge py image nc output nc width height looks pfa intensity tilesize overlap tmpdir Python Procedure output_array filter image_array method scan width height looks pfa intensity Parameters image nc This should be either a single or a multiple layered real image output nc This is a 0 1 image with the edges marked with 1 width integer default 3 The width of the filter window height integer default 3 The height of the filter window looks float default 1 The looks measure of the image data If looks is negative use 1 the routine will compute this parameter from the data pfa float default 5 Specifies the Probability of False Alarm 107P4 if pfa gt 1 to be used in determining whether changes are significant The larger pfa is the less chance there will be of reporting insignificant changes 162 NASoftware Ltd 29 Mar 10 B 2 IMAGE MANIPULATION ROUTINES intensity integer default 0 Assume the data is amplitude square root intensity tilesize integer default 512 A tilesize of 512 x 512 is used overlap integer default 48 To prevent obvious joins between tiles an overlap of 48 is used tmpdir directory default None Intermediate results
183. structure in the residual image and also the derived ENL The results motivate our overall advice despeckling on its own provides smoothed images which retain detail and are easier to interpret by eye than the original data but for most purposes segmentation provides a better starting point for a processing chain 6 7 1 Segmentation Initially we segment the scene using the default segment parameter set except that the shape parameter is set at 0 02 The segmented result is illustrated in Figure 6 18 Most of the structure is clearly visible in the reconstruction However some of the fine structure to the right of the top of the runway has not been reproduced In addition there are a large number of small regions visible within the scene These may be partly caused by sloping field areas in the scene which cannot be presented properly by the uniform region model imposed by segmentation other than by introducing small segments Alternatively they might be the result of small statistical fluctuations caused by speckle The most revealing test for the quality of the reconstruction is to take the ratio of the original intensity image to its segmentation If the segmentation corresponded precisely to the underlying RCS then this ratio would have the properties of pure speckle The residual ratio would therefore 98 NASoftware Ltd 29 Mar 10 6 7 AN EXAMPLE SEGMENTING AND DESPECKLING A LARGE IMAGE Figure 6 18 Segmentation of the scene in fi
184. study by applying the methods discussed in the previous section to the original intensity image The first step is to construct the intensity ACF of the data which is shown in Figure 6 16 The situation is complicated by the fact that the cross section component of the ACF at lags greater than 4 pixels has an appreciable slope From the theoretical expression for the intensity ACF we see that the component of the ACF proportional to the instrument function has the form near zero 29 Mar 10 NASoftware Ltd 91 CHAPTER 6 CASE STUDIES D 5 o P a o h o 2 a Intensity autocorrelation function o 15 a ha a e o iy 0 151 i i i i 0 Pixels Figure 6 16 Plots of the intensity ACF of the data in Figure 1 Azimuth ACF full line range ACF dashed line ql 2 go ls 1 where v is the order parameter of the cross section lag of gp k E e E The original n a L xX The origin of Figure 6 16 then indicates that v 4 and that the instrument function component of the intensity ACF has a magnitude of 0 31 Thus the correct lag values for resampling are at an ACF value of about 0 26 i e lags of 2 9 azimuth and 3 3 range The nearest integer resampling number is 3 along each axis It should be noted that it is difficult to separate the instrumental and cross section components of the ACF in this fashion Fortunately we only need to approximate the resampling values to t
185. sult is the division of the corresponding pixel values between the two input images Command Line Script multiply py imagel_in image2_in image out tilesize tmpdir Python Procedure array_out Numeric multiply arrayl_in array2 in Parameters tilesize integer default 512 A tilesize of 512 x 512 is used 29 Mar 10 NASoftware Ltd 157 APPENDIX B LOW LEVEL AND UTILITY ROUTINE DOCUMENTATION tmpdir directory default None Intermediate results are written to a system temporary directory B 2 17 pca Purpose Perform Principal Component Analysis on an image Command Line Script pca py image_in image_out white posdef tilesize tmpdir Python Procedure array_out pca array_in white posdef Parameters white integer Default 0 If white is non zero then also perform whitening Divide the output by the square root of the eigen values posdef integer Default 0 If posdef is non zero then assume the covariance matrix is a square Hermitian positive definite matrix tilesize integer default 512 A tilesize of 512 x 512 is used tmpdir directory default None Intermediate results are written to a system temporary directory B 2 18 pol2covar Purpose Converts polarimetric data from HH HV VV etc to covariance format HHHH HHHV HHVV Command Line Script pol2covar py image_in image_out mtpol tilesize tmpdir Python Procedure array_out pol2covar array_in mtpol
186. sy to use quick and smooth segmentation It uses a call to segment followed by a call to merge the parameters used have been preset to provide a fast but effective overall result We recommend that you use easyseg first if you find you need closer control of the process call segment and optionally follow this with a call to merge 24 NASoftware Ltd 29 Mar 10 4 1 SEGMENTATION Figure 4 5 Merging the result from Figure 4 2 4 1 3 The merge routine For each pair of neighbouring regions merge calculates the probability that the two regions are taken from the same underlying cross section If this exceeds the user specified probability threshold then the two regions are merged This is repeated until there are no more regions suitable for merging From the original image and a region map of this image merge will join any adjacent regions where the intensity does not significantly differ The final number of regions can be controlled by the method used and associated cost e The probability threshold pfa smaller values of cost give fewer regions e The crinkle value crinkle same as pfa but also smooths the region boundaries e The absolute intensity value abs merges regions whose intensity differs by less than the entered cost value Figure 4 5 shows the result of merging the segmented image of figure 4 2 The less significant edges have been removed from the original cartoon image Whilst the edges in the results produc
187. t gt gt fron infopack inport x gt gt gt gt gt image gis load stanwick nc gt gt gt gis Float32 image has 1 layers width 538 height 768 Driver InfoSAR InfoSAR NetCDF Images nc Pixel Size 0 022517 x 0 025379 UL 961 896000 719 946000 LR 974 010254 700 454851 Description stanwick nc layeri_name Stanwick X band SAR HH polarised 3m pixels amplitude siecle Float32 gt gt gt gt map numload stanwick_mer nc gt gt gt gt gt cseg colseg map image method bt intensity 1 gt gt gt Figure 5 3 The IDLE Shell Window with open script file NASoftware Ltd 29 Mar 10 Chapter 6 Case Studies In this chapter we illustrate the capabilities of InfoPACK via some simple Case Studies 6 1 Segmentation and Classification of Intensity Images 6 1 1 Introduction In many remote sensing applications it is important to be able to classify SAR images in terms of their intensity The difficulty is the effect of speckle noise which as we shall see makes assigning the intensity within a single pixel to a given class very inaccurate The situation can be somewhat improved in general by averaging over a window of say 3x3 pixels This averages out the edges between regions which would be undesirable in a complicated scene or if small objects are of interest Provided that we can be certain that we have known ground truth boundaries that actually apply to the dat
188. t raw RawDefinition 512 512 1 ChanDefinition 1 8U 0 1 512 Swapped UpLeftX 440720 000 UpLeftY 3751320 000 LoRightX 471440 000 LoRightY 3720600 000 All PCI data types SU 16U 16S and 32R are supported Currently georeferencing projec tions and other metadata is ignored The implementation of GDAL allows for read write and creation To open a PCI labelled file select the raw data file itself The aux file which must have a common base name will be checked for automatically For more details of the PCI aux format see http www remotesensing org gdal frmt_various html 3 3 2 Using a virtual file A second approach to describing your file format would be to create a virtual file For example a vrt file with two bands that happen to both come from band one of another file would be 29 Mar 10 NASoftware Ltd 17 CHAPTER 3 IMAGE FILE FORMATS lt VRIDataset rasterXSize 512 rasterYSize 512 gt lt VRTRasterBand dataType Byte band 1 gt lt SimpleSource gt lt SourceFilename gt utm tif lt SourceFilename gt lt SourceBand gt 1 lt SourceBand gt lt SrcRect xOff 0 yOff 0 xSize 512 ySize 512 gt lt DstRect xOff 0 yOff 0 xSize 512 ySize 512 gt lt SimpleSource gt lt VRTRasterBand gt lt VRTRasterBand dataType Byte band 2 gt lt SimpleSource gt lt SourceFilename gt utm tif lt SourceFilename gt lt SourceB
189. taxes are given in section A 4 7 28 NASoftware Ltd 29 Mar 10 4 2 CLASSIFICATION Figure 4 8 a Figure 4 8 b Figure 4 9 Result of using classify for the original SAR image of Figure 4 8 a 29 Mar 10 NASoftware Ltd 29 CHAPTER 4 IMAGE PROCESSING WITH InfoPACK 4 2 2 Themask_stats routine The mask_stats routine calculates the mean and order parameters necessary for supervised clas sification from the masked regions of an image Each layer of the mask image should indicate the training data for each class The resultant file is a single layered real valued image with height equal to the number of layers and width equal to the number of classes There is an example of a temporal sequence of SAR images a sequence of associated mask images and detailed instructions on how to process the images using mask_stats and classify in the images directory Full Python script and function syntaxes are given in section B 5 5 30 NASoftware Ltd 29 Mar 10 4 3 AREA CHANGE DETECTION 4 3 Area Change Detection These routines detect large scale changes between two or more images large scale implies that they are not tailored to detecting differences in point targets between images The images to be scanned for changes must be provided as layers in a single multi layer image if your images are separate use the tools Layer Operations option in the GUI to merge them into a single multi layered image The routines use the segment
190. td 29 Mar 10 A 4 HIGH LEVEL ROUTINES A4 7 classify Purpose Given an original image and a region map of this image classify will assign each region to one of a number of classes The output is a single layer image with each region labelled with its class number 1 2 3 etc The supervised classification scheme is based upon the assumption that the data is gamma distributed around the a priori class means If necessary order parameters for each class can be specified otherwise it is assumed that each class has the same order parameter Command Line Script classify py region_map nc image nc output nc means orders scaling intensity tilesize overlap tmpdir Python Procedure output_array classify region_map_array image_array means orders scaling intensity Parameters region_map nc The file containing the region map image This image should be one of either e A single layered real valued image of dimensions equal to the dimensions of the first layer of the original image image nc e A real valued image of exactly the same dimensionality as image nc In the case of a multiple layered region map the algorithm takes the intersection of the given layers to be the region map on which to perform the classification image nc A file containing the original image This can be either a single or a multiple layered image real or complex output nc The output file The output is a real valued si
191. ted here Full InfoPACK Python function specifications are given in Appendices A and B As with the Python direct interactive mode using Python from a command window you type a Python statement into the shell window at the gt gt gt prompt and then hit the Enter key to send it to the Python interpreter InfoPACK scripts can be executed directly from within the shell For example to segment the image myimage nc to the output file mysegment nc with default parameter settings you would type gt gt gt run segment py myimage nc mysegment nc InfoPACK functions can similarly be called For example to load myimage nc into memory segment it then save the result as mysegment nc you would type 50 NASoftware Ltd 29 Mar 10 5 5 USING THE InfoPACK SHELL WINDOW gt gt gt array_in numload myimage nc gt gt gt array_seg segment array_in gt gt gt save array_seg mysegment nc If you ever find yourself in a situation where you seem to be hung and cannot get a new prompt the interpreter is likely to be in a state where it is waiting for you to enter something specific in its parse state Hitting Ctrl z will send a keyboard interrupt and should get you back to a prompt 5 5 1 Syntax Highlighting Your code is coloured as you type it in based upon the Python syntax The colour scheme may depend upon your setup but typically comments are displayed in green strings in purple and Python keyw
192. ted pattern of changes is expected to be given by the red green and blue classes in Figure 6 2 The actual detections are shown in Figure 6 3 2 The single pixel results a show that only the red change with an RCS step of 5 is reasonably detected After a 3x3 window b the green change with step of 2 is reasonably well defined and the blue change with a set of 1 is partially detected Following segmentation c all the changes are well detected Finally the Upper Limit shows almost no difference from the expected changed classes in Figure 6 2 Once again the most useful comparison is in terms of the detection probabilities as summarised in Table 6 3 2 The probability of correctly detecting the change at the correct position in the se quence is listed for the four different processing methods The average result over all 8 possible positions is compared with the specific results for the three positions 4 5 and 6 where change is predicted 29 Mar 10 NASoftware Ltd 73 CHAPTER 6 CASE STUDIES Figure 6 3 2 a Detected Changes using Single Pixels Figure 6 3 2 b Detected Changes using 3 by 3 filter 74 NASoftware Ltd 29 Mar 10 6 3 CHANGE DETECTION BETWEEN INTENSITY IMAGES Figure 6 3 2 c Detected Changes using Joint Segmentation Figure 6 3 2 d Detected Changes using Prior Knowledge The colours indicate detected changes at the different positions within the image sequence Brown position 1 magenta 2
193. ter will bring a copy of all the line Alternatively you can toggle through all previous commands sent to the interpreter by using Ctrl or Alt p and Ctrl or Alt n You can edit the line as desired before hitting the Enter key to have it sent to the interpreter 29 Mar 10 NASoftware Ltd 51 CHAPTER 5 USING THE INFOPACK GUI 5 5 4 Function Call Prompting As soon as you type in the opening bracket for a function or method call a small box will pop up below the current line giving you a tip with regards to the arguments that are expected For example if the shell window contains gt gt gt segment you will get a small highlighted window containing a tip on completing the statement segment image quality shape stopping cooling looks maxiters regionsize intensity together with summary information on the purpose of the routine and its default parameter settings The window stays up until you enter the closing bracket These tips appear for built in functions any functions or methods including class constructors from Python s library modules as well as for any function or methods defined by yourself They do not however occur for list tuple or dictionary methods For the functions or methods that you define the tip will give the names of the parameters you define The first Summary line of the function or methods document string will also be displayed This can be used to give information on
194. the central pixel as a target depends upon the comparison of this likelihood with the threshold calculated from the probability of false alarm pfa A lower value of probability of false alarm will mean less targets will be detected Since the result is determined by the accuracy of the statistical estimates it may be advisable to have a larger data sample i e window size than the default The parameters pfa and looks can be used with this method scan Scanning edge detector This edge detector works by calculating the significance of every possible edge position in every window Since the result is determined by the accuracy of the Statistical estimates it may be advisable to have a larger data sample i e window size than the default The parameters pfa and looks can be used with this method A lower value of probability of false alarm pfa will mean less edges will be detected mmse The Minimum mean square error MMSE despeckling filter This measure can only be prescribed for real images and uses the looks parameter lee The J S Lee despeckling filter This measure can only be prescribed for real images and uses the Looks parameter width integer default 3 The width of the filter window height integer default 3 The height of the filter window looks float default 1 The looks measure of the image data If looks is negative use 1 the routine will compute this number from the data This estimate can be used with some of the
195. the slope of the intensity ACF changes significantly This marks the point where the instrumental component has largely decayed away Both methods have been applied to the resampling and segmentation of data from both SLC imagery and multi look intensity imagery In both cases the residual speckle component is consistent with the expected ENL and the correlation properties similar to those expected These techniques offer the means to segment intensity data with any speckle correlation prop erties in an optimum fashion optimising both the retention of information by speckle reduc tion through averaging the intensity and correlation removal by resampling NASoftware Ltd 29 Mar 10 6 6 CORRELATED IMAGES AND DESPECKLING 6 6 Correlated Images and Despeckling Correlations in the data cause the same problems when despeckling as when segmenting Despeck ling uses a neighbourhood of 3x3 pixels to form the reconstruction which in the example studied in the previous section 6 5 2 is matched to the correlation between samples in this case The result of despeckling image 6 13 would therefore be expected to resemble the original image with the individual speckle blobs well reconstructed In Figure 6 6 a we show a detail from Figure 6 13 together with its despeckled version in Figure 6 6 b The problem cause by the correlations is apparent the speckle has been well recon structed Figure 6 6 a Figure 6 6 c Figure 6 6 d Figu
196. the types of the arguments and return value 5 5 5 Class Member Prompting 669 Once an object has been defined typing the dot command after the object name will bring up a list of possible members or methods available For example if the shell window contains gt gt gt image gis load myfile nc gt gt gt gis you will get a small scrolled window containing a list of the gis class members 5 6 Developing InfoPACK Scripts The interactive mode of calling InfoPACK modules is convenient when exploring a new dataset For more advanced use InfoPACK also provides the ability to define chains of operations scripts InfoPACK allows two levels of scripts At the simplest level InfoPACK commands are stored in a text file which can be executed by the Run Study command from the GUI For instance segment py imagel seg intensity 1 shape 3e 3 echo Display segmented image openev seg cho Test segmentation residual ratio divide py imagel seg ratio stats py ratio whole 1 intensity 1 52 NASoftware Ltd 29 Mar 10 5 7 THE IDLE INTEGRATED DEVELOPMENT ENVIRONMENT ort sys mort infopack dof pene pT Writes an InfoSAR netcdf file as raw data Typical usage would be s image nc raster image Other options are byteswap 1 Transforms big endian to little endian data sys argv 0 sys exit rc Jer Ct image swapbyte 0 if suapbyte image image byteswapped A open file wb f write im
197. tion image nc This image must be a multiple layered i e more than 1 layer real or complex valued image output nc The output file The output is real valued layer image with dimensionality equal to 29 Mar 10 NASoftware Ltd 137 APPENDIX A InfoPACK HIGH LEVEL ROUTINE DOCUMENTATION that of the original image minus one of its layers Thus each output layer corresponds to changes between each adjacent pair of layers in the original image Pixels are either coloured with value 1 indicating that they belong to regions for which there has been significant change or with value 0 that they don t pfa float default 6 Specifies the Probability of False Alarm 107P if pfa gt 1 to be used in determining whether changes are significant The larger pfa is the less chance there will be of reporting insignificant changes looks float default 1 The looks measure of the image data If looks is negative use 1 the routine will compute this parameter from the data intensity integer default 0 Assume the data is amplitude square root intensity mtpol integer default 0 Polarimetric data is stored as layers of complex data however this is the same file format as multi temporal polarimetric data To distinguish the two the mt pol parameter is used For instance 5 days of 3 channel polarimetric data would be stored as a 15 layer complex image and the mtpol 3 parameter would be used The default value of mt po1 0 uses all of the
198. tion of 100m with a 5cm radar from a range of 800km requires an aperture some 400m long If the radar is attached to a moving platform either a satellite or an aircraft then it is possible to combine reflected signals from along the flight path to synthesise a very long antenna The aperture or area used to receive signals is created artificially during the signal processing This is a Synthetic Aperture Radar SAR 2 1 Interpretation of SAR Images Each pixel in a SAR image represents the back scattered radiation from an area in the imaged scene A large pixel value bright represents a strong received signal The strength of the received signal depends on many things These include e Radar wavelength and the relative size of the scatterers Smooth flat surfaces usually reflect very little radiation back to the antenna so roads and still water will appear dark Objects much smaller than the wavelength of the radar are invisible to the detector so for a low frequency L band radar fields may also appear smooth At shorter wavelengths fields or woodland often give the image a textured appearance Polarisation of the pulses Radar pulses can be polarised into a vertical or horizontal plane For a linear feature on the ground the reflected signal is much stronger when the polarisation agrees with the orientation of the scatterer For example a pylon will appear much brighter in a VV image than in an HH image The same is true for other
199. tion of the averaged images Assigned class 0 1 2 3 4 5 True class Prob 0 0 9574 0 0257 0 0108 0 0061 0 0 0 0 1 0 0041 0 9948 0 0010 0 0 0 0 0 0 2 0 0003 0 0061 0 9934 0 0 0 0 0 0003 3 0 0 0 0013 0 0372 0 9615 0 0 0 0013 4 0 0 0 0018 0 0 0 0 0 9982 0 0 5 0 0 0 0 0 0 0 0 0 0 1 0 Table 6 2 2 e Confusion Matrix for Classi fication using the Ground Truth segments Assigned class 0 1 2 3 4 5 True class Prob 0 1 0 0 0 0 0 0 0 0 0 0 0 1 0 0 1 0 0 0 0 0 0 0 0 0 2 0 0 0 0003 0 9997 0 0 0 0 0 0 3 0 0 0 0 0 0 1 0 0 0 0 0 4 0 0 0 0 0 0 0 0 0 9992 0 0008 5 0 0 0 0 0 0 0 0 0 0 1 0 6 2 3 Conclusions On comparing the results for classification over 10 images with the equivalent results for a single image we note that the classification performance is improved considerably for all except the 3x3 window result The 3x3 window method is shown to be a poor approach because the window introduces cor relations at the boundaries of regions leading to misclassification Joint segmentation followed by 66 NASoftware Ltd 29 Mar 10 6 2 SEGMENTATION AND CLASSIFICATION OF MULTIPLE INTENSITY IMAGES classification yields a considerable improvement since it provides a filter whose window size and shape is adapted to the data The fact that the segments are not matched exactly to t
200. to the the InfoPACK entry of the Programs folder of your Start menu and click on it 5 2 2 Unix Linux On Unix systems just type InfoPACK at a command prompt 5 3 The Main InfoPACK Menus The easiest way to access the InfoPACK library is via the menus in the main InfoPACK GUI win dow Clicking on menus will generally bring up a dialog box from which files are selected pa rameters are chosen and the routine is run All menu items call the appropriate InfoPACK script All InfoPACK scripts work file to file to enable the processing of arbitrarily large images within memory As such it is always necessary to provide an output file name in the appropriate dialog box All menu commands are executed from within the Python shell When a command is processed you will see in the shell window the script you have called and the parameters you are using For ease of use all recently accessed image file names are stored which can be accessed from a recent files button in dialog boxes and the variable files in the shell window The main toolbar on the GUI screen includes the following menus File see section 5 3 1 with menu items 29 Mar 10 NASoftware Ltd 41 CHAPTER 5 USING THE INFOPACK GUI Tools Open Image Import Image Export Image Run Study New Script Open Script New Console Exit see section 5 3 2 with menu items Create New Image Layer Operations Extract Sub Image Resize Image Histogram Image I
201. to within the specified probability of false alarm are not significantly different in intensity Alternatively the minimum absolute value between regions may be specified Command Line Script merge py regionmap nc image nc output nc method cost looks intensity crinkle mtpol tilesize overlap tmpdir Python Procedure output_array merge region map array image_array method cost looks intensity crinkle mtpol Parameters region_map nc The file containing the region map image This image should be one of either e A single layered real valued image of dimensions equal to the dimensions of the first layer of the original image image nc e A real valued image of exactly the same dimensionality as image nc In the case of a multiple layered region map the algorithm takes the union of all the edges in the region maps image nc This can be either a single or a multiple layered real or complex valued image A polarimetric image should be given as a layered complex valued image output nc The merged region map output image is of the same dimensionality as the input image if the original image was real valued The merged region map output image is single layered in the case of a complex valued original image method default pfa This selects the algorithm for comparing neighbouring regions pfa Specifies the Probability of False Alarm 10 if pfa gt 1 to be used in deter
202. ttings or users can specify their own choice of parameters For the routine specified the GUI effectively calls the corresponding Python script the basic script commands are echoed on the console so that you can see what is going on In this way user scripts specifying complex processing chains can be built up experimentally and on the fly Operations are file to file and memory is cleared after the operation is complete Operation of the GUI is the subject of chapter 5 From a Python script Single routines may instead be called from the system command line prompt Users can thus develop complete Image Understanding processing chains of both InfoPACK and other routines by composing appropriate shell or batch scripts For details on how to invoke command line scripts under each supported operating system see the file README t xt which can be found in the top level installation directory From a Python shell or program All InfoPACK routines can be invoked as Python function calls from within a Python shell or Python program The routine procedures are automatically made available from the module infopack and may be called directly or from within the users own script When called directly from the console online reminders are provided for the required parameters Python provides an integrated development environment for researching and developing com plete Image Understanding processing chains including calls to InfoPACK routine
203. turns a list containing a Numeric array the actual image data and a class with the GIS informa tion Any format recognised by the GDAL library is valid The following code segments the Numeric array a normalise the original image by the inten sity value in each segment and saves the resulting file while preserving the GIS information from infopack import a gis load file img b segment a c a b save c result nc gis gis Here a gis is a list with a containing the image and gis containing the GIS information All the InfoPACK py scripts and the GUI buttons preserve GIS information in this way However if you do not care about keeping or using the GIS information then you can throw it away from infopack import a dummy load file img b segment a c a b save c result nc If you miss out the dummy then a will be a list and you will get all sorts of interesting bugs when you try to use a as an image unless you access the numeric array and gis information via list elements a 0 and a 1 respectively For convenience the routine numload only returns the numeric array and gisload only loads the GIS data A 1 4 Online Help When typing function names in a Python or InfoPACK shell the Python interpreter will display the function syntax along with a full list of possible parameters When working with Python Scripts help concerning a particular routine can be quickly foun
204. ue the work ERS Helpdesk ESRIN Via G Galilei 00044 Frascati Rome ITALY Tel 39 6 941 80666 Fax 39 6 941 80272 URL http www esa int European Envisat ASAR Envisat was launched in 2002 the C band ASAR instrument provides 25m resolution and H and V polarisations with better coverage range of incidence angles and modes of operation than ERS Radarsat ERS Helpdesk ESRIN Via G Galilei 00044 Frascati Rome ITALY Tel 39 6 941 80666 Fax 39 6 941 80272 URL http www esa int envisat instruments html Radarsat is a Canadian satellite launched in August 1995 Itis C band HH polarisation with a 20 60 variable incidence angle 14 Radarsat International Inc 275 Slater Street Suite 1203 Ottawa Ontario K1P 5H9 CANADA Tel 1 613 238 5424 Fax 41 613 238 5425 URL http radarsat space gc ca NASoftware Ltd 29 Mar 10 Chapter 3 Image File Formats There are a wide variety of file formats currently used for storing image data and it would be im practicable for InfoPACK to support them all Instead InfoPACK has chosen to support the Geospa tial Data Abstraction Library GDAL found at http www remotesensing org gdal which is a translator library for raster geospatial GIS data formats GDAL currently supports roughly 40 format drivers some of which are listed in Table 3 1 If your image is in one of the formats supported by GDAL then you should be able to run any of the Inf
205. ure and average local looks measure e Order of equivalent gamma distribution and average local order Full function specification is given in Appendix B Image Properties Provides information on the selected image including e Size e Coordinates of origin image corners and image centre e Pixel Datatype for each layer Band Clip Image Allows you to clip a specified image with specified minimum and maximum values default values of O and 100 respectively Any pixels in the image with values less than the specified minimum will be replaced by the minimum any with values greater than the maximum will be replaced by the maximum Full function specification is given in Appendix B Filter This provides access to a range of low level filters found in traditional image processing packages Filters perform a variety of operations such as averaging smoothing or edge detection using a fixed window size See section A 4 11 for details of the routine parameters None of the high level routines in InfoLIB use these filters they are provided primarily as a convenience Despeckle Despeckle an image See section A 4 10 for details of the routine parameters and section 4 5 1 for a longer discussion 29 Mar 10 NASoftware Ltd 45 CHAPTER 5 USING THE INFOPACK GUI Threshold Produce a 0 1 mask image labelling pixels above below a given threshold Image pixels that lie above the threshold are set to the true value oth
206. ures See section A 4 5 for details of the routine parameters and section 4 1 5 for a longer discussion Defleck Use Defleck to remove small spots that sometimes occur after using Segment or Merge See section A 4 6 for details of the routine parameters and section 4 1 6 for a longer discussion Edge Produce an edge map showing the edges of a segmented image See section B 2 5 for details of the routine parameters and section 4 4 2 for a longer discussion 5 3 4 The Classify Menu Classification Classify an image into a given number of classes Classify requires a segmented image the original image and a file with the class means and an optional file with the order parameters for supervised classification See section A 4 7 for details of the routine parameters and section 4 2 for a longer discussion 29 Mar 10 NASoftware Ltd 47 CHAPTER 5 USING THE INFOPACK GUI Generate Classes There are four routines under the heading Generate Classes These are e Make Classes If no apriori classes are available this routine generates a set of equally spaced class means from the input image e Mask Statistics This routine uses training masks to calculate the mean and order Statistics files needed for classification e Mask Polstats Classifying polarimetric data needs to generate the class covariances which this routines does with the help of some region masks e Class Statistics This routine generates a mean and orders
207. utine B 3 6 gissave Purpose Saves updates the class containing the GIS information stored in an NASoftware Ltd NetCDF file Command Line Script None provided Python Procedure gissave image_out gis class 29 Mar 10 NASoftware Ltd 169 APPENDIX B LOW LEVEL AND UTILITY ROUTINE DOCUMENTATION B 3 7 insert Purpose Inserts a sub image into a larger image at the given offsets Command Line Script insert py image_in insert_in image_out woffset hoffset Python Procedure See save B 3 17 Parameters woffset integer default None hoffset integer default None B 3 8 layer_combine Purpose Combines image layers held in a set of seperate images into a single file All input images must be of the same width and height Command Line Script layer_combine py imagel_in image2_in image_out tilesize tmpdir Python Procedure array_out layer_combine arrayl_in array2_in arrayn in Parameters tilesize integer default 512 A tilesize of 512 x 512 is used tmpdir directory default None Intermediate results are written to a system temporary directory 170 NASoftware Ltd 29 Mar 10 B 3 FILE HANDLING FACILITIES B 3 9 layer_split Purpose Splits a multi layered image into a numbered set of images corresponding to the individual layers gt Results are written into the files image_out_000 nc image_out_001 nc Command Line Script layer
208. values of neighbouring pixels are correlated To obtain independent samples the image must be averaged down until the pixel spacing is at least as large as the resolution If we now consider averaging the intensity in neighbouring correlated pixels the ENL will not be reduced by the expected amount because the speckle is correlated ERS PRI data is described as having an ENL of 3 0 The pixel size is about 12 5m x 12 5m and the resolution about 25m x 25m Thus averaging 2x2 pixels does not result in an ENL of 12 but a value nearer 6 10 NASoftware Ltd 29 Mar 10 2 6 THE NUMBER OF LOOKS AND InfoPACK 2 6 The Number of Looks and InfoPACK Many of the InfoPACK routines for example the utility stats py or from the GUI select Tools Image Statistics estimate the number of looks from data using the definition of ENL above Since this estimate normally contains both RCS and speckle fluctuations this estimate only corresponds to the ENL when there are no RCS variations We describe the result as the looks measure to distinguish it from the true ENL Occasionally InfoPACK routines accept the number of looks as an input In this case the defined instrument value eg 3 for full resolution ERS PRI should be used Even if you do know the number of looks in your data you may still get better results by allowing the InfoPACK functions to estimate it from the image The InfoPACK functions use the looks parameter to build a model of the surface cro
209. vert py image nc output tif format GTiff may not produce a displayable TIFF image The GeoTiff format allows for floating point data to be stored in the file which many image display programs cannot interpret B 3 3 extract Purpose Returns a sub image of given dimensions at the given offsets The width and height must be speci fied Command Line Script extract py image_in image_out width height woffset hoffset Python Procedure array_out extract array_in width height woffset hoffset Parameters woffset integer default 0 hoffset integer default 0 B 3 4 gdal_translate Purpose A utility to convert files between various GDAL Geospatial Data Abstraction Library formats The default output format is GeoTIFF There is limited support for resizing images selecting sub windows and converting between data types Typing gdal_translate provides a full help message 168 NASoftware Ltd 29 Mar 10 B 3 FILE HANDLING FACILITIES Command Line Script gdal translate of format ot type image_in image_out Python Procedure None provided B 3 5 gisload Purpose Returns the class containing the GIS information stored in an NASoftware Ltd NetCDF file Command Line Script None provided Python Procedure gis_class gisload image in Further Comments InfoPACK provides little support for manipulating the GIS data instead 1t is usually just passed to the save ro
210. will most conveniently be used The routine detects changes between adjacent pairs of layers in the provided image the ordering of the layers in the multiple layer image file is therefore crucial The output is a multiple layer image with number of layers equal to the number of adjacent pairs in the sequence of layers in the original image i e number of layers minus 1 Each of these layers indicates where significant changes were found between the image pair If you are interested solely in whether there are changes over the whole sequence of layers in the input image use the routine change_detection This routine is useful for example when monitoring the growth of crops over the seasons Command Line Script change detection position py regionmap nc image nc output nc pfa looks intensity mtpol tilesize overlap tmpdir Python Procedure output_array change_detection_position region_map_array image_array pfa mlthresh looks intensity mtpol Parameters region_map nc The file containing the region map image This image should be one of either e A single layered real valued image of dimensions equal to the dimensions of the first layer of the original image image nc e A real valued image of exactly the same dimensionality as image nc In the case of a multiple layered region map the algorithm takes the intersection of the given layers to be the region map on which to perform the change detec
211. x 1 301 286 1771 URL http nssdce gsfc nasa gov earth earthfaq html 12 NASoftware Ltd 29 Mar 10 2 10 SOURCES OF SAR DATA SIR B flew on the space shuttle Challenger and collected data on 5 October 1984 It is L band HH polarised SIR C has flown on several space shuttle missions since 1993 It has L band and C band SAR with quadrature polarisation and also X band HH polarised Radar Data Center Jet Propulsion Laboratories California Institute of Technology 4800 Oak Grove Drive Pasadena CA 91109 USA Fax 1 818 393 2640 URL http southport jpl nasa gov AirSAR An airborne SAR which collects multi frequency data at C L and P bands Radar Data Center Jet Propulsion Laboratories California Institute of Technology 4800 Oak Grove Drive Pasadena CA 91109 USA Fax 1 818 393 2640 URL Ahttp southport jpl nasa gov Japanese Earth Resources Satellite JERS 1 Fuyo 1 was launched on 11 February 1992 It is L band HH polarisation with 38 5 incidence angle National Space Development Agency of Japan NASDA Earth Observation Program Office Tokyo 105 JAPAN Tel 81 3 5470 4254 URL http hdsn eoc nasda go jp 29 Mar 10 NASoftware Ltd 13 CHAPTER 2 SYNTHETIC APERTURE RADAR IMAGES European ERS Satellites ERS 1 was launched by the European Space Agency ESA in July 1991 It is C band VV polarisation with 23 incidence angle ERS 2 was launched in 1995 to contin
212. y to call a specific speckle reduction routine first or indeed at all How ever when required despeckle will provide a despeckled version of a given image without un necessarily losing detail 4 5 1 The despeckle routine despeckle approximates a single or multiple image with a locally constant i e locally flat model of the surface cross section It fits this model using a variant of the simulated annealing optimisation method The noise model used in the procedure is determined by the looks measure for the image A higher value of looks measure implies that the data are less influenced by noise and are therefore more reliable estimates of the underlying surface The InfoPACK routines all assume the input image has uncorrelated speckle If your image is oversampled so that it includes correlations then use the shrink utility to remove them see Section 6 5 for a discussion of how to do this optimally The despeckle algorithm can be performed upon images with more than one image layer Full Python script and function syntaxes for despeck le are given in section A 4 10 The effects of the parameters of the speckle reduction functions are described below The number of iterations despeckle is iterative The longer it runs the smoother the resulting surface Figure 4 13 c shows the effect of smoothing a section of a DLR ESAR image for 10 50 the default value and 250 iterations using despeckle with the looks measure set to 1 and other pa

Download Pdf Manuals

Related Search

Related Contents

Solar Permit Application Packet - Pennsylvania Municipal Code  realproducer 11 user`s guide - Service & Support  HRV6690. CORTADORA DE FIBRA  Oven - Maytag  Logger and Link Quality Monitor (LQM) Tools  GEN/ONE Genlocking Encoder for Amiga Computers User`s Manual  Mellanox OFED for Linux Release Notes  

Copyright © All rights reserved. Failed to retrieve file