Home

IMAGE CALIBRATION AND ANALYSIS TOOLBOX USER GUIDE

1. Image 2 Convert your image to the cone catch values of your species ignore this step if you just want to show objective camera values Plugins gt Multispectral Imaging gt Image Analysis gt Convert to Cone Catch and select your camera visual system model when prompted 3 At this stage the image will be a grey scale 32 bit multi channel stack of cone catch values or camera pixel values Run Plugins gt Multispectral Imaging gt Tools gt Make Presentation Image 4 Select which cone catch channels or image channels you want to convert into which colour hich i h fi Gok A E cror eet edhe for viewing see Image 31 for a bluetit example anie Leia erh gec elel Hk aia cer Ideally select red green and blue once each Or or yellow and blue for dichromats g i if you have a dichromatic animal select yellow and blue only Leave all other unwanted channels uv Blue sw Green th ee as Ignore lw Ignore 5 Transform Applying a square root transform will dbl Ignore reduce the image contrast and make the darker regions more visible Optionally leave this as linear Transform to make non linear by selecting none Transform Square Root 6 Convert to RGB Colour This will output a normal Remember this image is for presentation RGB image that you can save as a JPG or TIF and only not for measurements i aoe use in other software packages or for printing Convert to RGB colour p FOR g 7 When presenting
2. Pattern and luminance distribu Cified in the image tion difference calculator options Region comparison A omae flowers_f 128 000 To this region The region ID to com pare to This can be the same in which case it will only be able to compare between photographs Region comparison Within photo would compare the first region ID to the second region ID comparing flower f to flower g in this example Select Between photos to compare flower f to any instances of flower g across all 43 Image Analysis photos this will compare within photo if applicable as well Click OK it will generate the comparisons and create a table of results You can then save this table as a csv file or copy and paste the data into a spreadsheet Colour differences introduction There are various methods available for comparing colours such as hue calcula tions principal components analysis or opponent Channel responses A Common method for comparing colours in non human visual systems in terms of whether two objects are likely to be disciminable or not is Vorobyev and Osorio s 1998 re ceptor noise model which calculates just noticeable difference JND values This predicts whether two similar colours are likely to be discriminable based on the signal to noise ratios of each channel to di tri and tetrachromatic visual systems JND values less than one are indiscriminable and values above 3 ar
3. Red Green Blue Eo Including square transforms e g Red Red Red can greatly increase the number of potential interactions probably best used when there are only 3 channels include Square Transforms Diagnostics plots can show whether the model conforms to assumptions of homogeneity of variance and normality of error structure Diagnostics plots Location of Rscript exe Windows only Path cAProgram Files R R 2 13 2 bin x64 Rscl OK Cancel mage 19 Options dialogue for generat ing anew cone catch mapping model from long to short wave sensitivity e g R G B or vR vG vB UB UR Where the lower case v and u refer to visible and UV pass filters and RGB refers to the camera s red green and blue chan nels The format for all sensitivity data follow the structure on the left Image 18 saved as a CSV file The order of the camera channel sensitivities is import ant and must match the order used when generating the mspec file see Error Reference source not found This example shows the order for a stand ard visible amp UV multisoectral image starting from the longest wavelength and going to the shortest See the sup plied examples in imagej plugins Cone Mapping All sensitivity data photore ceptors and camera _ photosensors must be normalised so that the sum for each row equals 1 e g create these values by dividing each cell by the sum of all the cells across all wavelengths for
4. this is ignored during RAW import Use a fixed aperture between all photos in the study if possible to have a uniform depth of field use changes in shutter soeed to control the exposure e When using a zoom lens only use the maximum or minimum focal length not an intermediate Ensure the lighting is suitable and the angle of the grey standard s relative to the light source follows a clear justifiable rule see Image 13 and that the colour and brightness of the light falling on the standard is as close as possible to that of the sample For pattern analysis or scale measurements place a scale bar level with the sample and photograph from a consistent angle e g overhead Place a label in each photograph if practical and make a note of the photo number linked to each sample e If you are placing samples against an artificial background use a dark grey or black material particularly with potentially transparent or shiny samples When photographing samples in situ in the field their natural background is normally most suitable unless you are comparing the sample 15 Taking Photos directly to its surroundings in which case it may be desirable to photograph the sample separately against a dark background so that it is measured independently of its surrounds obviously only if it can be moved without risk of harm to yourself or the sample Use a stable tripod amp remote shutter release if possible
5. Alignment Auto Align gt Offset 16 v Scaling loops 1 0f 5 Scale step size 0 005 Standard reflectance s Add your stand l custom aignmentzone ard reflectance values separated by Outpt commas if there s more than one Any IV Save configuration file image ouput Aligned Normalised 22ta gt number can be used in any order This ex Image Name testmags S S ample Image 14 shows the values for a I Rename RAW files photo that contains a 5 and 95 stand ard Leaving this box empty will mean el cee that no standards are used and the result Image 14 Multispectral image set ing image will be linear but not normal tings options ised Linear images such as this can be used for comparing lighting between photographs Customise standard levels Grey standards should ideally be grey meaning they have uniform reflectance across the whole spectrum being photographed e g the reflectance from 300nm to 700nm is uniform for UV photography However sometimes standards can get damaged or maybe two standards were used and they have very slightly different colours Use this option to correct for known differ ences in standard reflectance in each of the camera s channels You can meas ure these values by photographing your imperfect standard against a reference clean grey standard and normalise the image using the clean reference stand 25 Image Processing ard to work out the corrected reflectance values for each chan
6. STOTIS TG Sor a E E ud ae E E alae O A A E e E 46 Reporting your MethodS e esseeseeeerreererererrrerrrcerrerereeererreceerceerrerereererereererereeeereeeees 46 Creating colour or false colour images for Presentation cccccessssssseeeeeeeeeeees 47 REFERENCES xccaveduasaatenvacies senuaes a a a a a a Dia sanbianeiadda aeee 50 OVERVIEW The Image Calibration and Analysis Toolbox can be used to transform a normal di gital camera into a powerful objective imaging tool Uncalibrated digital photographs are non linear Stevens et al 2007 meaning that their pixel values do not scale uniformly with the amount of light measured by the sensor radiance Therefore uncalibrated photographs cannot be used to make objective measurements of an object s reflectance colour or pattern within or between photographs This toolbox extracts images linearly from RAW photo graphs Chakrabarti et al 2009 then controls for lighting conditions using grey standards Various colour and pattern analysis tools are included and images can be converted to animal vision images As long as none of the photos are over exposed the processing will not result in data loss when measuring reflectance values above 100 relative to the standard which is common with shiny objects or when the standard is not as well lit as other parts of the image This is because all images are opened and processed as 32 bit floating point straight from the RAW fil
7. The position and angle of the grey standards is critically important for objective measurements The light falling on the standard should match as closely as possible the light falling on the object being measured with re spect to angle relative to the light source intensity and colour see page 10 above If only one standard is being used then a value around 20 to 50 reflect ance is appropriate for most natural objects as a rule the ideal standard should have a reflectance near the upper range of your sample For example when measuring very dark objects a lower reflectance standard can be used e g 2 to 10 Multiple standards e g white and dark grey allow the software to overcome some optical glare light bleeding onto the sensor or photographing through reflective or hazy materials such as un derwater and through the surface of water When photographing from the air into water the grey standards must be positioned next to the target e g both standards underwater as close to the Image 4 When photographing roughly 2D objects the grey standard s should always be in the same plane as that object and ideally the same distance from the light source sample as possible and the back ground sky or ceiling being reflected by the surface of the water must be uniform e g this won t give correct values if there are any visible ripples or if you can see blue sky and white clouds reflected in the water surface 11 T
8. and green Select regions of interest and a scale bar Regions of interest ROIs are used to specify regions of the image that you want to measure Once you have created a multispectral image you can specify these re gions using any of the selection tools rectangle circle ellipse polygon freehand tools etc After highlighting a region press a key on your keyboard to assign that ROI to a letter When you are done press 0 to save the ROIs with this multispectral image For example Image 17 shows an image of two flowers Using the polygon tool one petal can be selected and then the letter F on the keyboard will call that ROI f1 highlight the next petal press F and it will be labelled f2 etc For the second flower the petals are marked with G instead This will make it easy in the analysis later on to compare the pooled F regions compared to the pooled G regions flower F compared to flower G but will also still allow easy between petal comparisons e g if you wanted to look at within flower differences 29 Image Processing 1 Image ah al x File Edit Image Process Analyze Plugins Window Help BOla o lt 4 Alale D sedele flowers 16 7 A 2 5 visible G Normalised 4948x3280 pixels 32 bit 310MB Measure Deselect Properties Flatten F I Show AIl W Labels Image 17 Selecting ROIs and a
9. of reflectance the amount of light reflected by an object relative to a reflectance standard colour and pattern all you need is a digital camera that can produce RAW images a good quality lens e g with no vignetting darkening at the image edges and a grey or white standard or ideally two standards dark grey and white Most pattern analysis requires a scale bar Lighting also needs to be considered carefully see page 8 onwards and Im age 3 The images produced by this set up will be objective in terms of measuring reflect ance levels being repeatable with the same equipment and robust against changes in lighting conditions but are strictly device specific Different camera models have different spectral sensitivities so measurements made with a different set up will produce slightly different results Equipment check list for objective measurements Camera body Use any camera that can take RAW photos This includes digital SLRs and mirrorless cameras Some crossover cameras with integrated lenses might also be suitable Lens Ideally use a prime i e not a zoom lens A high quality lens is desirable as this will minimise vignetting images getting darker towards the corners radial distortion e g if you photograph a chequer board the lines in the photo should all be perfectly straight and chromatic distortion Diffuse grey standard s These can be purchased from most photography suppliers and will s
10. our eyes or to a camera under sunlight versus fluorescent lighting The light source you use must cover the entire range of wavelengths you are pho tographing it is not acceptable to use two or more light sources with different emission spectra to cover the whole range you are interested in unless these light sources can be blended together effectively which is not trivial For example us ing one light source that emits human visible light should not be combined with a second light to add the UV component This is because the lights will interact with the 3D surface angles of the target making some regions appear to have colours that they do not The Iwasaki eyeColour MT70D E27 6500K arc lamp available from CP lighting www cp lighting co uk can be converted into a good UV visible band light source by removing its UV IR protective filter The filter is just visible as an oily rain bow effect when the bulb is held up to the light By using a hand held drill with a steel wire circular brush this filter can be removed without damaging the glass to increase the UVA emissions of this bulb Sensible precautions and protective equip ment goggles and gloves should be worn when doing this The bulb s emissions in the UVA range are increased by this process so eye and skin protection must be worn when working near this light source for long periods and it will fade nearby colours faster Being an arc lamp this bulb needs to be operated by a suit
11. that receptor Natural spectra should be normalised so that the highest value equals 1 e g divide 32 Image Processing each cell by the maximum value across all wavelengths for that spectrum luminant Select your illuminant as above D65 is recommended as the CIE stand ard illuminant You can add your own illuminant spectra see above Receptors Specify the visual system to wavelength 300 301 302 303 304 map to following the format in Image 20 uv 0 000027 0 0000335 0 0000432 0 0000582 0 0000818 This example shows bluetit sensitivities The 5 5 o e order of the channels is not important for N 0 photoreceptor classes but the order here will be other order in which channels are created when converting to cone catch images Training spectra Select the database of natural spectra to use when making the mapping function The wavelength band e g 400 700nm or 300 700nm should be the same for camera sensitivities illuminant soectra receptor sensitivities and train ing spectra There will be a warning if the wavelengths do not match Use stepwise model selection Model selection can be used to remove terms from the model that do not usefully improve the fit This will sow down the processing of generating the model but results in a smaller faster model for image processing In practice it seems to make little difference You can specify how the models are simplified e g forwards backwards and whet
12. you will be asked which mspec file to add these ROIs to Reloading multispectral images Once you ve generated a multispectral image you can easily re load it by going Plugins gt Multispectral Imaging gt Load Multispectral Image Select the mspec file you generated and it will automatically load the normalised aligned 32 bit image with all the ROIs you selected You can then use Plugins gt Multispectral Imaging gt Tools gt Convert to Cone Catch to manually transform individual images from camera vision to animal vision if you have the appropriate mapping function Generating a new cone mapping model We have included a number of models for converting to animal cone catch val ues based on the camera setups that we have calibrated and under D65 lighting conditions However tools are included for making your own model if you know your camera s spectral sensitivity functions Mapping from camera to animal vision is performed by simulating the animal s pre dicted photoreceptor responses to a set of thousands of natural spectra and then the camera s responses to the same spectra all under a given illuminant A poly nomial model is then generated that can predict animal photoreceptor cone catch values from camera photoreceptor values Hong et al 2001 Lovell et al 2005 Stevens et al 2007 Westland et al 2004 ImageJ gets R to perform the modelling after preparing the data so you must have R installed for this to func tion
13. zip file and copy them to your imagej plu gins folder DCRAW Each camera manufacturer has their own type of RAW file and DCRAW is an in credibly useful piece of software that can open almost all types of RAW file Dave Coffin www cybercom net dcoffin dcraw Our toolbox utilises DCRAW to open RAW files in a linear fashion Chakrabarti et al 2009 using the DCRAW Plugin for ImageJ Jarek Sacha ij plugins sourceforge net plugins dcraw Incidentally RAWTherapee also uses DCRAW to open RAW files meaning it is suitable for pre screening photographs to check whether they are well exposed see below To check DCRAW is working on your installation of ImageJ go Plugins gt Input Out put gt DCRAW Reader Select a RAW file e g NEF for Nikon CR2 for Canon and it will show an options dialogue Select the options shown in Image 8 and it should load the RAW file as a linear 16 bits per channel image Note that ImageJ treats 16 bit and 32 bit RGB images as a stack displayed in colour This is slightly concep tually different to a normal 24 bit RGB image 8 bits per channel and any meas urements you make of this image will only be from the selected channel even though you can see the channels 18 Software Use temporary directory for processing white balance None Do not automatically brighten the image Output colorspace raw Document mode no color no interpolation Document mode without scaling totally raw Read as Interpol
14. 0 to 380nm The Baader IR UV cut filter can then be used for visible soectrum images 400 to 700nm We use custom built filter sliders to switch easily between visible and UV pass filters These sliders are made on a CNC milling machine and G Code plans can be made available on request UV grade grey standard s Normal photography grey standards are not suitable as few have a flat reflectance down to 300nm Spectralon standards from Labsphere are suitable In an emergency or adverse conditions a finely sanded piece of natural white PTFE plastic wrapped in a stretched flat layer of plumber s PTFE tape could be used as a white standard Although this will not be quite as diffuse as a Spectralon standard it should have flat reflectance down to 300nm The layer of PTFE tape can also be replaced regularly as it gets dirty Never touch the surface of any PTFE Spectralon grey standards While they are highly hydrophobic and repel water extremely well they absorb oils incredibly easily such as the natural oils in your skin Also take great care not to get sun block sun screen chemicals near the surfaces UV lighting Only use one type of light source that covers the whole spectrum never attempt to use a light that only emits UV and a separate source for visible light See below for suggestions page 8 Image 3 Only use a metal coated reflective umbrella or sheets of natural white PTFE to diffuse UV light sources avoid using white umbr
15. 00 14478 766000000 15118 433000000 8134 429000000 8389 055000000 6380 996000000 6371 547000000 7088 177000000 6320 748000000 10398 112000000 gt l Image 30 Calculate pairwise colour differences using JNDs Visual system Weber fractions Select your visual system s Weber fractions These are often just calculated from relative cone abundance values you can add your own Weber fractions by making a new txt file in plugins Multisoectral Ima ging weberFractions Examples are included in the toolbox for common model visual systems with global Weber fractions of 0 05 and 0 02 Compare this region First region to use in the comparison To this region Second region to use in the comparison can be the same as the first region for between photo comparions Region comparison Specify whether the comparison should be within or between regions and photos 45 Image Analysis Click OK and the comparisons will be calculated Save the data as a csv file or copy and paste into a spreadsheet 46 Image Analysis PRESENTING DATA Statistics Once you have generated colour pattern or luminance metrics the data can be analysed in any suitable statistics package When using analyses that assume nor mality of error structure such as ANOVAs GLMs GLMMs etc these variables will of ten benefit from a log transformation Luminance distribution difference values are bounded at 2 i e the difference between the norm
16. As with colour these Scale pixels differences can be calculated auto Image 28 Pattern spectra for an egg matically between regions of interest against its background Note there are two peaks in spatial energy for the egg imply ing it has both small spots and large spots pattern energy spectra above using The peak frequency could therefore either the Pattern and Luminance Distribu be around 10 pixels or 100 pixels tion Difference Calculator Lumin ance distribution differences are calculated similarly summing the differences in the number of pixels in each luminance bin across the entire histogram This meth od for comparing luminance is recommended when the regions or objects being measured do not have a normal distribution of luminance levels For example many animal patterns have discrete high and low luminance values such as zebra stripes or egg maculation in which case a mean value would not be ap propriate Comparing the luminance histograms overcomes these distribution problems 0 01 0 008 0 006 0 004 within or between images from the 42 Image Analysis Calculating pairwise pattern and luminance differences Descriptive pattern data such as the peak energy and peak spatial frequencies of pattern spectra can be useful for comparing treatments but for pairwise compar isons when there is more than one peak such as the egg in Image 28 it makes more sense to measure pattern difference as the
17. IMAGE CALIBRATION AND ANALYSIS TOOLBOX USER GUIDE Jolyon Troscianko amp Martin Stevens Contact jt jolyon co uk Contents OVERVIEW oreinen e oea i a E A O A Ea A A O E 3 EQUIP EIT us a Aee e vas EE EE E ER 5 Objective camera specific MECSUFEMENNS cccccccccsssesssssccceccceecessesssseeeeeceeeees 5 Equipment check list for objective MECSUFEMENTS ccccceccceccesccccseeeesssecccesenses 5 Converting to cone catch IMAgEesS ssesssesssssssesssssssesssesssesssesssssssesssesssessssssssseeee 6 Equipment check list for UV photography and CONE MAPPING ccececeeeeeeeees 6 WARING d OE EEE E E E EEE eaineaanenn 8 Lighting emission SOECHA sesssesssesssssssesssesssesssssssesssesssesssessssssssssetessssseereessssseseees 8 Lighting direction amp GUTUSSM SSS 65 cvsieicieisa ssqovsvseGevesvs docsdesiansdvssdustonwivossevatevesesaisvece 10 CEST CIICICIOIS cca tah deaventaved iatadelunuas hey valet E E aase tetas ee audits weneens tule a 11 AVS SS TUNIS oen naa e a a E eae eaten pata ta E EEE ete Pte ayer E E 12 Scale eTe a E E yO eae me EEE EE E E OT 14 Taking photos CSC CNS esa Cet ia orl cata th dole Rica nea ee eee ada el ae 15 DOI A eerste conti aar a a Sore a raataa ea O AON S 17 SOWAS CRECklIS tn a a a A th aibs 17 nstallat spea a an ei a EE Miser A E E a a aa 17 DORAMA A A E EE A Lipa teamed 18 Compiling DCRAW for MACOS eesessessseesseesseeesererererererererererererrrrrrrrerrrrerssesee
18. Settings These options specify what type of multispectral stack to create and how the channels should be arranged The defaults are Visible for working with nor mal RGB photos or Visible amp UV for standard UV photography this will arrange the channels as vR vG vB UB UR See page 34 for details on creating a custom ised filter combination 24 Image Processing Grey Standards Choose Separate photos if you did not take photos of the grey standards in the same photo as the sample This is the sequential method and the photos must be taken with exactly the same camera settings and lighting condi tions shutterspeed aperture ISO etc Estimate back point This setting is useful Settings Visible amp UV x to minimise optical veiling glare where in Grey Standards ternal reflections within the lens reduce a Grey standards in Same photo x photo s contrast raising the black point F7 Estimate black point useful with one standard from zero If you only used one standard Standardreflectance s 5 95 check the Estimate black point box I Customise standard levels and it will attempt to calculate the dark tandans Move a a level this is most important in bright field Image Location _ conditions Do not tick this box if you I Images sequential alphabetically in directory have used two or more standards It is probably not required when photographs are taken in dark rooms Alignment amp Scaling
19. You only need to create a mapping function for each camera visual system illu minant soectrum database combination once Run Plugins gt Cone Mapping gt Gen erate Cone Mapping Model to create a new model and specify your settings Er ror Reference source not found 31 Image Processing wavelength 300 301 302 303 304 305 306 Camera Select your camera configuration this vR o o o o o o o wil be specific to your camera lens and filter Me 5 5 gt 5 5 5 combination You can add your own camera uB o o 0 0 0 o o sensitivity curves visual system receptor sensitivit uR o 0o 0 O O 0 O jes iluminant soectra or soectrum database to Image 18 Camera sensitivity ex ample data the relevant folder in imgej plugins Cone Map ping The format of these files should match the supplied csv files and the wavelength range and intervals must all match e g 300 700nm or 400 700nm at Inm increments The Camera sensitivity csv file should contain a row for each of the channels required The order of these channels must match the output of the multispectral image creation We have ordered channels Select Configuration Camera 400D Canon 18 55mm 400 700 J luminant D65 300 700 Receptors Bluetit 300 700 Training spectra Natural Spectra 300 700 Model Simplification Settings Use stepwise model selection Direction forward backward Criterion AIC Set the maximum number of interaction terms e g 3
20. able bal last also available from CP lighting Photography stands that have an E27 socket 9 Taking Photos can also readily be used with this bulb Consult an electrician to wire this system to gether if you are not confident in doing it yourself Lighting direction amp diffuseness The direction and diffuseness of the light source can interact with the three dimen sional shape of the target and with the shinyness or glossiness of its surfaces A point source of light such as the sun or a small light bulb is highly directional so the reflectance measured from a point relative to the grey standard will be highly dependent on its angle relative to the light source as well as the surface reflect ance Therefore under highly directional lighting the reflectance measured from a surface by your camera will only be accurate if that surface has perfect Lamber tian reflectance is very diffuse and the angle of the surface is exactly equal to that of the grey standard Clearly these conditions are almost never true so care must be taken with directional point light sources Direct sunlight is arguably a dif ferent case because measurements made by a camera in the field will be qualit atively similar to the reflectance measured by an eye and the lighting conditons are ecologically relevant However as long as uniform lighting angles relative to the target are being used point sources are suitable for comparing reflecta
21. ackground used in this photograph Image 12 below Is an over exposed ex ample of the same image with clustering on the right hand side of the histogram Image 11 Histograms typical of a good exposure There are no pixels clustered on the very right hand side of the histo gram but the red peak is close Image 12 Histograms of an over exposed image with lots of pixels clustered on the very right hand side of TOCESSINg the histogram Creating Multispectral Images Once you ve selected the photos with the best exposure you are ready to gener ate a multispectral image For human visible soectrum photos i e normal RGB photos this will be a single photo for UV photography this will generally be two photos one taken through the visible pass filter and a second through the UV pass filter Other filter combinations are supported File Image Process Analyze J Plugins Window Help BO clo lt 4 4S A macros Paintbrush Tool Shortcuts Utilities New Compile and Run Install Ctrl Shift M 3D Analyze Cone Models Examples Filters Graphics Input Output Measure Multispectral Imaging Generate Multispectral Image Load Multispectral Image Data Analysis Image Analysis Polynomial Slice Transform 32Bit Tools Image 13 Creating a new multispectral image In ImageJ go Plugins gt Multispectral Imaging gt Generate Multispectral Image Image 13 this will open a settings dialogue box
22. ad the toolbox files from www sensoryecology com or www jolyon co uk unzip the con tents and copy them to your ImageJ plugins directory On windows this directory is normally C Program Files ImageJ plugins see Image 7 On MacOS this is Applications ImageJ plugins On Ubuntu this is home user name imagej plugins if you have installed ImageJ through the software centre note this is a hidden folder on Ubuntu 17 Software Z a l J TB0818700D C gt Program Files Image gt plugins gt search plugins 2 Organize v T Open New folder az v Fr Favorites Name Date modified Type 430 15 10 2014 08 55 e folder pv Libraries ad Analyze 15 10 2014 08 55 ile folder Ji Cone Models 15 10 2014 08 56 ile folder a Homegroup d deraw 15 10 2014 09 59 File folder do Examples 15 10 2014 08 55 File folder 1 Computer Ji Filters 15 10 2014 08 55 File folder amp 11208187000 C d Graphics 15 10 2014 08 55 File folder a Storage D J Input Output 15 10 2014 08 55 File folder d jars 15 10 2014 08 55 File folder ia Network L Multispectral Imaging 15 10 2014 10 13 File folder gt Scripts 15 10 2014 08 55 File folder d Stacks 15 10 2014 08 55 File folder J Tools 15 10 2014 08 55 File folder ij dcraw_jar 15 10 2014 09 59 Executable Jar File L README txt 19 04 2014 19 57 Text Document 4 items selected Date modified 15 10 2014 09 59 Image 7 Extract the toolbox files from their
23. aking Photos When measuring 2D objects the sur face of the standard should be in the same plane as the target e g Image 4 For complex objects the standard should normally be angled relative to the light source e g flat on the ground outside Image 5 Whatever rule is used must be kept constant throughout the experiment and between treatments Image 5 When photographing more com plex 3D objects or when the angles you Remember that light spills off nearby Photograph from cannot be predicted then the grey standard s should be elec eye pacregrepnig me angled relative to the light source In the grey standards near brightly coloured field the grey standard should normally be objects that you have introduced to level with the ground so that it collects the scene This includes objects colour light from the whole sky not angled dir ful in UV many surfaces that look Cty towards the sun white to us absorb UV Camera settings Focus and exposure are the main considerations when taking photographs A tri pod is recommended if camera shake could affect the photograph and tripods are essential for UV photography as each photograph needs to be taken twice once with a visible pass filter and once with a UV pass filter with as little move ment as possible between photographs so that they can be aligned Exposure is affected by aperture ISO setting and shutter soeed also known as in tegration time The aperture af
24. al alignment Select an area with lots of in focus detail and or your sample The grey standards are also useful to use as custom alignment zones This also speeds things up because otherwise the entire image is used for alignment Using scaling is not recommended if you are selecting a relatively small custom alignment zone Save configuration file Always leave this ticked unless you are just testing This soe cifies whether the mspec file should be saved The mspec file is what s used to re load a multispectral image from RAW files and is saved alongside the RAW files Image output Specify what to output at the end for inspection For UV photos the aligned normalised 32 bit option will let you inspect the alignment at the end psuedo uv will show you the G B and UV channels as RGB looks good and lets you see UV colour ignoring red The colour visible or psuedo uv outputs can make life easier when selecting areas of interest Any output other than aligned normalised 32 bit images are only for inspecting and selecting regions of interest Any modifications to these images e g increasing the brightness to see dark ob jects better will not be saved and will not affect the multispectral image if it is re loaded later Image name Give each multispectral image a name that describes its sample and treatment This name will become the label used in batch processing You ll be asked before overwriting in case you forge
25. alised histograms have a max imum of 2 if there s no overlap between the sample s luminance values aft all Where these numbers appear to be bounded a solution is to divide all the values by 2 so the maximum is 1 and then apply a logit transform logit p log p 1 p This should make the luminance difference data normally distributed even if they are near the upper and lower bounds Reporting your methods When publishing findings based on image analysis it is important to report the equipment and methods used in sufficient detail Here is a list of the things that you should include so that your study could be replicated Camera model reporting whether it was converted to full soectrum e Lens model if it was a zoom lens what focal length was used Filters models and approximate transmission characteristics e g Baader Venus U filter transmitting from 320 to 380nm Light source or lighting regime e g direct sunlight not within 1 hour of sunrise or sunset or Iwasaki eyeColor arc lamp with its UV filter removed Grey standard reflectance values and model or material If a direct point light source was used what angle was the grey standard relative to the illuminant e g the grey standard was always angled towards the illuminant or the grey standard was level with the plane of the sample see page 10 47 Presenting data How were regions of interest selected I
26. are saved as the batch processing progresses so the process can be restarted picking up from where it stopped Granularity pattern analysis introduction We provide tools for performing a pattern analysis based on Fast Fourier bandpass filtering often called a granularity analysis This form of analysis is increasingly widely used to measure animal markings Godfrey et al 1987 Stoddard and Stevens 2010 and is loosely based on our understanding of low level neuro 40 Image Analysis physiological image processing in numerous vertebrates and invertebrates Each image is filtered at multiple spatial frequency scales and the energy at each scale is measured as the standard deviation of the filtered pixel values The pattern energy of non rectangular regions of interest are measured by extracting the se lection so that all surrounding image information is removed i e the selection area is measured against a black background This image is duplicated and the selec tion area is filled with the mean measured pixel value to remove all pattern in formation inside the selection area Identical bandpass filtering is performed on both images and the difference between the images is calculated before measur ing its energy for a shape independent measure of pattern To perform pattern analysis using the batch measurement tool one must first select which channel to use for pattern analysis Pattern processing in humans is thought to rely o
27. area between the curves in this case the grey area between the egg and its background The same principle can be used for comparing multi modal luminance histogram data The toolbox can calculate this for you Open your pattern spectrum or lumin ance histogram data e g from the flowers_f 2 000 200 207 flowers f 2 828 341 498 csv file saved by batch analysis so flowers_f 4 000 468 605 4 n flowers f 5 657 611 995 that they are in the results window flowers T 8 000 pues close any results windows open flowers_f 11 914 6 flowers_f 16 000 already and go File gt open or just drag fl f 22 627 m i ge Ptal patem _eneroy z the file into ImageJ then run Plugins gt E e this region f z flowers fi 45 280 a A ra Multispectral Imaging gt Data Analysis gt flowers_f 64 000 tothis region g gA Pattern and Luminance Disribution Dif oaar gabe ference Calculator The dialogue box flowers_f 90 510 flowers_g 2 828 will ask for a few options Image 29 flowers_g 4 000 a5 6 flowers_g 5 657 509 947 flowers_g 8 000 493 764 flowers_g 11 314 514 272 Data Specify the column name that flowers_g 16 000 568 862 t th tt A flowers g 22 627 645 984 CONTAINS e pattern energy or IUMIN flowers_g 32 000 717 886 ance values flowers_qg 46 255 740 067 flowers_g 64 000 767 129 flowers_g 90 510 711 099 flowers_g 128 000 491 354 Compare this region Select one of the regions IDs from the ROI coding spe Image 29
28. ation quality High speed low quality bilinear Half size Do not rotate or scale pixels preserve orientation and aspect ratio Show metadata in Result window OK Cancel Help Image 8 DCRAW Import options The settings specified here will open a linear image from the selected RAW file If the toolbox is not working correctly try opening a file with DCRAW fo see if this is the problem Compiling DCRAW for MacOS DCRAW for Windows and Linux seems to work reliably across different operating systems however MacOS sometimes requires recompilation of DCRAW If the Tool box does not work with the supplied files and throws out an error message when opening a RAW image you will need to compile your own DCRAW binary file this is the bit of software that converts the RAW image into a usable image for ImageJ 1 Create a new folder on your desktop e g called myDCRAW 2 Get the original source code for DCRAW from this website http www cybercom net dcoffin dcraw dcraw c in your browser go File gt Save Page As and save the code as dcraw c in your myDCRAW folder 3 Open System Preferences Choose the Keyboard option and then the Shortcuts tab Under Services check the New Terminal at Folder option Close System preferences 4 Right click your myDCRAW folder and choose the New Terminal at Folder option A terminal will load up you can check it s in the right location by typing Is and h
29. d the Step Multiplier Multiply box in the previous dialogue For birds this will be the double cones dbl as shown in Im nae ia age 25 For dichromatic mammals just select uminance Options Luminance Bands 0 off the lw channel as this is most likely to en Lowest Luminance code luminance HighestLuminance 65535 Transform Luminance Linear Se MV Output Pattern Spectrum 2 128 Step Size 1 41421356 32 0 Combining ROIs Rescale Image This is a second scaling op Prefix separate with a comma f g tion available if you want to measure the pat aig aesia Statin wiih Een aa wa tern at a different scale to that selected in bined and d ingl HF A s E r Carina va an Pe haa the previous dialogue or want to scale alll im vaio megos ages uniformly rather than based on a scale bar In general this should always be set to 1 Visual display of the energy maps saved alongside the original image Useful for checking but remember off these are auto scaled so adjsut the brightness settings on different images before comparing them directly MV Output Energy Maps x cacai Start size Specify the smallest scale to start Image 25 Pattern amp luminance bandpass filtering at It is not possible to measurement dialogue measure wavelengths lower than 2px in size so this is a sensible start value Set this to zero to switch off pattern analysis End size Pattern analysis will stop at thi
30. e 400 700nm i e human visible range then the equipment listed above for ob jective images is suitable with the exception that the spectral sensitivities of the camera must be known Equipment check list for UV photography and cone catch mappin Camera converted to full spectrum sensitivity This process involves removal of the filter covering the sensor Often this filter is replaced with a quartz sheet that transmits light from 300 to 700nm ACS www advancedcameraservices co uk and other companies provide this service commercially or you can try to do it yourself e g www jolyon co Uk 2014 07 full soectrum nx1000 Doing it yourself could damage the camera and has the added difficulty of ensuring the sensor position can be adjusted to restore focusing UV lens Most normal lenses do not transmit well in UV CoastalOptics and Nikon make excellent but expensive UV dedicated lenses Cheaper alternatives can be purchased second hand online Well known examples are the Nikkor EL 80mm older model with a metal body enlarging lens and the Novoflex Noflexar 35mm The Nikon AF S Micro Nikkor 105mm f 2 8 does have UV transmission although it is not achromatic meaning it needs re 6 Equipment focusing between visible and UV photos which complicates the alignment although the toolbox can readily deal with this UV pass and visible pass filters We recommend the Baader Venus uU filter which transmits efficiently from 32
31. e discrimin able under good lighting conditions Siddiqi et al 2004 The toolbox implementa tion of the model only applies to photopic vision daylight lighting conditions and uses the standard log version of the model Luminance JNDs follow Siddiqi et al 2004 Some authors argue that JNDs are not a suitable metric to use for compar ing highly dissimilar colours e g see Endler amp Mielke 2005 although the perform ance of different colour space models is an area that requires further investigation JNDs are currently therefore intended for comparisons between similar colours Calculating pairwise colour and luminance discrimination values Just noticeable differences JNDs tell you whether a given animal is likely to be able to distinguish two colours or levels of luminance under ideal viewing condi tions see above To calculate colour or luminance JND differences open the res ults CSV file containing your cone catch results so it s in the Results window of ImageJ and go Plugins gt Multispectral Imaging gt Data Analysis gt Calculate colour JNDs see Image 30 This calculation uses the common log transformed model and should only be used for photopic daytime vision with plenty of light Siddiqi et al 2004 Vorobyev and Osorio 1998 This tool only compares the mean colour values for each region you have measured If there are lots of very different col ours in those regions this method is not suitable If
32. e minimum not an outlier it is lt 2 6 standard deviations from the mean Using a scaling value equal to or lower than the minium will mean no images are scaled up which creates false data The smallest scale bar is in photo flowers Image 24 Output from the Batch scale bar calculation tool The minimum scale re ported here is 102 so using this scaling value would be sensible or round it down to 100 do not round up Start processing after file number While processing very large datasets it can be annoying if it stops half way through e g due to a power cut or crash The batch processing records measurements as it proceeds so there is no need to start pro cessing from the start again just put the image number to start at in this box to re start from where it got to You can look in the output files in the same directory to see where it got to previously Click OK to proceed The first multispectral image in the selected directory will be loaded so that the script knows what receptor channels are available 37 Image Analysis Next a dialogue will open asking what meas urements should be made Luminance Channel abi Image Scaling all images must be scaled uniformly Recalelmage 1 Luminance channel The luminance channel Pattern Analysis Options is used for pattern measurements but this will ee re P vary between different taxa For human vision so E this would be Lum added if you ticke
33. eing clipped or sat urated As pixels become brighter they reach this level and cannot go any higher which is a big problem for objective photography as it will result in false data being produced and any pixels that have reached saturation point which is 65535 for a 16 bit image should not be measured Most digital cameras provide on screen his tograms and these are very useful for judging whether a photo is over exposed as you take the photos though they cannot be relied on entirely because they use non linear levels and are often conservative Ideally separate red green and blue histograms should be used but some cameras only provide a grey scale his togram Single grey level histograms can be less reliable for UV photography If the grey values are calculated as the average of RGB this will under estimate the pixel levels because the green channel is not sensitive to UV while the red is most sensit ive meaning the red could be saturated Nevertheless once you have become accustomed to your camera s histogram performance it is a very useful tool for judging exposure The recommended camera mode depends on conditions In fixed lab conditions where lighting intensity will not fluctuate substantially the best solution is to use M full manual aperture and shutter soeed control and spend some time working out the best settings then use those settings for the whole experiment Note that any changes in the position or inte
34. el values by 655 35 to give percentage re flectance values relative to your grey standard s By default the channels are shown individually as greyscale images TO make an image with three channels colourful you can go Image gt Colour gt Make Composite and select colour This will not affect image measurements You can manually convert on opened mspec image to animal cone catch quanta with Plugins gt Multispectral Imaging gt Image Analysis gt Convert to cone catch and measure the values in these images Running Plugins gt Measure gt Meas ure All Slices once after loading ImageJ will measure the pixels in each image channel separately and now when you press M on the keyboard it will measure all the channels However we would recommend performing a batch image pro cessing job once you have finished preparing all your mspec images This will en sure each image is measured with exactly the same settings and will arrange the data into easily managed spreadsheets with less room for human error 35 Image Analysis Batch image analysis os Ee o File Edit Image Process Analyze 00S Window Help Boaclo lt 4 a Shortcuts Utilities New Compile and Run Install Ctrl Shift M 3D Analyze Cone Models Examples Filters Graphics Input Output Measure Multispectral Imaging Generate Multispectral Image Load Multispectral Image Data Analysis Batch Multispectral Image Analysis Image Analysis Batch Scale Bar Calculati
35. ellas or diffusers as they are very unlikely to be white in UV see page 10 Natural white PTFE sheets of about 0 25 to Imm thickness are good for diffusing through back lighting 1mm or thicker sheets are good as reflective white surfaces Tripod It is essential that there is as little movement as possible between the visible and UV photographs so that they can be perfectly aligned Image 2 A UV camera setup with our custom built filter slider that makes it easy to take the same photo in visible or UV bands 7 Equipment TAKING PHOTOS Lighting emission spectra Photography is entirely dependant on light and the way it bounces off or through objects so is an important consideration before starting data collection In an ideal world all photographs that you want to compare should be taken under uni form lighting conditions however in practice this can be impossible to achieve The emission spectrum of a light source affects its colour which can generally be controlled for by using a grey standard see below However some emission spec tra are so spikey that the spikes can interact with reflectance spectra in ways that cannot be controlled for with a grey standard e g causing metamerism see Image 3 Fluorescent light sources energy saving bulbs tube lights etc are the worst and should be avoided if at all possible Flickering can also be problematic If your photographs have odd looking horizo
36. ergy summed across all scales or amplitude a measure of pattern contrast proportion energy the maximum en ergy divided by the summed energy a measure of pattern diversity or how much one pattern size dominates mean energy and energy standard deviation In ad dition the energy spectra the raw measurement values at each scale can be output and used for pattern difference calculations see below Pattern maps can also be created these are false colour images of the filtering at each scale 41 Image Analysis produced for subjectively visualising the process Animal markings and natural scenes often have pattern energy spectra with more than one peak frequency meaning the pattern descriptive statistics above can arbitrarily jump between peaks with similar energy levels in different samples Im age 28 An alternative approach we recommend when directly comparing two samples rather than deriving intrinsic measurements to each sample is to calcu late the absolute difference between two spectra A and B across the spatial scales measured s Pattern gi y A B sS max This is equivalent to the area between ooz the two curves in Image 28 Any two patterns with very similar amounts of energy across the spatial scales meas ured will produce low pattern differd ence values irrespective of the shape of their pattern spectra These differ ences will rise as the spectra differ at or BA PEA Ha any frequency
37. ers_g 15972622 6181 615 12166 767 129 64 000 0 103 7444 557 572 658 130 558 17623 817 3929 Pattern Results AE z Luminance Results File Edit Font File Edit Font Label pattern_size pattern_energy Label luminance coverage flowers_f 2 255 256935270 flowers_f 0 03125 0 000000000 flowers_f 2 8284 341498358704 flowers_f 0 0625 0 000000000 flawers_f 4 468 605393670 jflowers_f 0 09375 0 000000000 flowers_f 5 6569 611 224656336 i 0 000085580 flowers _f 8 707843371585 a 0 003808301 flowers_f 11 3137 683 547508129 E 0 112922550 flowers_f 16 556 297741652 0 241591784 flaowers_f 22 6274 430 268170435 a i 0 198117244 flowers_f 32 343 487309027 a t 0 120196834 Wflowers_f 45 2548 280 102772654 a E 0 106290116 231 191716733 0 090415062 189 186366916 0 061360719 125 161886419 0 041677364 Image 27 Results from batch image analysis These results are also saved in spreadsheet files in the image directory The results of batch image processing are shown in Image 27 The main colour measurements mean and standard deviation for each ROI in each colour chan nel and pattern descriptive data are saved to the results window Note that lumin ance standard deviation is a measure of contrast Depending on whether you op ted to output pattern spectrum data and create luminance histograms these data will be shown in their own windows All data are saved in spreadsheets in the im age directory and the data
38. es as they are required so no very large in termediate TIFF images ever need to be saved Using two or more standards or one standard with black point estimates over comes the problem of the unknown black point of the camera sensor making the method robust in the field with various factors reducing contrast such as sensor noise or lens flare Photographing through reflective surfaces or through slightly opaque media is also made possible by using two or more grey standards This allows photography through water from the air as long as the reflected sky background is uniform underwater or though uniformly foggy misty atmospheric conditions Multispectral cameras with almost any number of bands are supported by the code and can be used for greater colour measurement confidence This toolbox is freely available on the condition that users cite the corresponding paper published with this toolbox Image Calibration and Analysis Toolbox a free 3 Overview software suite for measuring reflectance colour and pattern objectively and to animal vision and the papers relating to any visual systems and natural spectrum libraries used Image 1 Examples of flowers photographed in human visible colours left and false colour hon eybee vision right EQUIPMENT The ideal equipment you need will depend on the hypotheses you want to test Objective camera specific measurements For making objective measurements
39. essing has been performed and maintains the meta data of each image We would also recommend uploading the associated mspec files and ROI files for each image as these are very small files that enable quick data analysis See White et al 2015 for further information on repeatability in colour measure ments Creating colour or false colour images for presentation Linear images are essential when measuring reflectance values but they normally look dark and dull on your monitor This is because the dynamic range of cameras is far higher than that of your viewing media e g monitor or printed on paper meaning your camera can detect a far higher difference between the lightest 48 Presenting data and darkest parts of a scene than your monitor or printer can recreate To ac count for this photos are non linearly transformed normally with a Gamma curve to try to squeeze the image s dynamic range into a smaller range When presenting examples of your photographs it is acceptable to transform the images so that they are non linear for display on monitors and in print though it is good practice to make it clear what transform has been applied Presenting a tetrachromatic image is obviously impossible for human viewing so instead one can present a false colour or pseudo colour image where a subset of the channels are arranged into an RGB stack Image 32 1 Open your mspec image Plugins gt Multispectral Imaging gt Load Multispectral
40. f there is some ambiguity in a region s outline how was this accounted for In this case consider having the regions selected by someone who is blind to the hypothesis and treatment being selected If image scaling was used what was the scaling value and what was the average size of the sample being measured e g images were scaled to 20 px mm the average sample was 52 3 5 mm in length If you converted your images to cone catch values report where the camera spectral sensitivity functions were sourced or how they were measured what database of natural soectra were used for the modelling and where the animal s receptor spectral sensitivities came from The R values for the model fit to natural soectra could also be reported these numbers can be retrieved by opening the java file for your model in imagej plugins Cone Models If you are measuring JNDs report the Weber fraction used for the most abundant receptor channel this will normally be 0 05 or 0 02 and cone proportions If you are measuring pattern descriptive statistics or pairwise pattern differences what spatial frequencies were measured If you compared luminance distribution differences how many bins were used Will your image data be made available in a public archive RAW files should be made available rather than any other image format to ensure compatibility with these tools The use of RAW files also ensures no non linear processing or destructive proc
41. fects the amount of light allowed through the lens with an iris that changes size like the pupil in your eye A large aperture also known as a wide aperture e g f 2 lets through lots of light but that narrows the depth of field making out of focus parts of the photograph even more blurred Close up or macro photography creates even narrower depths of field so a smaller aperture e g f 16 might be necessary to get more of the scene in focus Aperture can interact with various lens imperfections such as radial chromatic dis tortion so as a rule try to keep the aperture constant across all your photos and rely on shutter soeed to change the exposure Smaller apertures are recommen 12 Taking Photos ded to get more of the scene in focus Most lenses have an optimal aperture of around f 8 for the sharpest images assuming the whole image is in focus smaller apertures can create blurring through diffraction but will make out of focus ob jects sharper ISO should also be kept constant throughout your entire study be cause this affects the signal to noise ratio of the images higher ISO produces more noise It is important that the aperture is not changed between visible and UV photos The exposure can be adjusted by altering the shutter soeed and or the ISO as long as the same ISO setting is used for all visible photos and another ISO is used for all UV photos Over exposure will result in the brightest pixels in the image b
42. gnment was selec ted and specifies how far out of alignment the pho tos can be 16 pixels is a sensible default warning messages will be shown in the auto scaling log Im age 14 if this offset doesn t seem to be sufficient Scaling Loops If you have had to re focus between visible and UV shots this has the effect of zooming the image slightly which needs to be undone for a true alignment This is most important when using lenses that are not UV achromatic like the Nikon Nikkor 105mm but not the CoastalOpt 60mm 5 6 scaling loops are normally sufficient Turn off scaling by setting this value to 1 27 Image Processing File Edit Font 5 of 6 0 99 Fit 1377 2849 Offsets x 12 y 17 1 Fit 1264 6942 Offsets x 2 y 1 0 9975 Fit 1316 5899 Offsets x 3 y 5 1 0025 Fit 1183 0113 Offsets x 3 y 3 1 0012 Fit 1214 0729 Offsets x 1 y 2 1 0038 Fit 1214 7417 Offsets x Sy 5 1 0019 Fit 1192 1647 Offsets x 2 y 3 1 0031 Fit 1192 2006 Offsets x 4 y 4 1 0022 Fit 1183 8785 Offsets x 2 y 3 1 0028 Fit 1182 0958 Offsets x 3 y 4 Image 16 Auto scaling output Scale step size A default of 0 005 is sensible this is a 0 5 scale difference In searching for the optimal scale this step size is used to start with and is then halved with each scaling loop Custom alignment zone This lets you select the area to use for alignment for both auto or manu
43. h a prefix that specifies the filter tyoe used v for visible u for UV and a suffix that describes the camera s channel R for red G for green B for blue Most UV cameras are sensitive to UV in their red and blue channels and not their green channels So the green channel is thrown out leaving us with vR vG vB UB UR Here the channels are ar ranged from longest wavelength to shortest wavelength see Image 9 Because UV multispectral images are combina tions of two photographs any slight camera shake or change in focus can cause misalign ment between the photos Even an offset of a few pixels can result in false colours being cre ated Think of the purple fringing seen at the corners of many photos where there is very high contrast this bright purple is an unwanted arte fact caused by chromatic radial distortion ef fectively a misalignment between the red green and blue channels 21 Image Processing Checking Photographs RAWTherapee Before converting and calibrating photos it is often necessary to select the photos with the best exposures e g if exposure bracketing was used or to check that they are in focus etc A good bit of software for doing this is RAWTherapee open source and available across all operating systems After installing and opening RAWTherapee ensure the processing profile is set to Neutral We would also recommend setting this as the default profile in Prefer ences gt Image P
44. he background is significant in display and crypsis Biol J Linn Soc 32 427 433 Hong G Luo M R and Rhodes P A 2001 A study of digital camera colorimetric characterisation based on polynomial modeling Color Res Appl 26 76 84 Lovell P G Tolhurst D J Parraga C A Baddeley R Leonards U Troscianko J and Troscianko T 2005 Stability of the color opponent signals under changes of illuminant in natural scenes JOSA A 22 2060 2071 Osorio D and Vorobyev M 2005 Photoreceptor sectral sensitivities in terrestrial animals adaptations for luminance and colour vision Proc R Soc B Biol Sci 272 1745 1752 R Core Team 2013 R A Language and Environment for Statistical Computing Vienna Austria R Foundation for Statistical Computing Schneider C A Rasband W S and Eliceiri K W 2012 NIH Image to ImageJ 25 years of image analysis Nat Methods 9 671 675 Siddiqi A Cronin T W Loew E R Vorobyev M and Summers K 2004 Interspecific and intraspecific views of color signals in the strawberry poison frog Dendrobates pumilio J Exp Biol 207 2471 2485 Spottiswoode C N and Stevens M 2010 Visual modeling shows that avian host parents use multiple visual cues in rejecting parasitic eggs Proc Natl Acad Sci 107 8672 8676 Stevens M Parraga C A Cuthill I C Partridge J C and Troscianko T S 2007 Using digital photography to study animal coloration B
45. her to use AIC or BIC Number of interaction terms The polynomial can specify higher level interactions between channels though in practice two or three are sufficient Include square transforms Including square transforms creates a more complex model in theory it can provide for a better fit but in practice it rarely improves the model and can create more noise Diagnostic plots These show how well the model fits assumptions of parametric models and are output to imagej plugins Cone Models Plots Rscript Path If you are running in Windows you need to add the location of your Rscript exe file that s bundled with R and performs script processing If the model 33 Image Processing 0 ojojo Image 20 Sample format for sensitivity data creation process stops doing anything when it says Waiting for R to process data there is most likely a problem Communicating with rscript exe check this file path is correct After creating a model it is saved in imagej plugins Cone Models along with its source code and the quality of the model s fit R values are shown for each photoreceptor The R values should ideally be gt 0 99 Considerably lower values imply the model has a poor fit and might fail to produce reliable results Generating a custom camera and filter arrangement Photo oR i 5 igit abies You can specify your own channel import A 3 options e g if you are using different filter Image 21 Filter arra
46. his is the optimal exposure a9 1 0 1 T2 U The camera thinks this is one stop EV over exposed Qe en fen Wa 2 8 a A This might produce a good exposure for bright 2 1l 0 1 2 targets photographed against a darker background Exposure bracketing With this setting three photos Been ee ae will be taken with 1 EV i e one photo under 2 A g eee exposed one over exposed and one normal Exposure bracketing as above but with a larger EV Mee fea faa faa l range This is recommended in highly unpredictable 2 l 9 1 2 situations or for very high contrast photos y t p Exposure bracketing with 1 EV but this time Qe nnn fan fa 2 amp they are all shifted down one stop Use this if the 2 2 0 1 2 camera tends to over expose Image 6 Exposure values and bracketing examples In aperture priority A mode you can manually shift the EV values to meet these specifications In manual M model you can change the shutter speed to shift these values Remember that over exposure of your samples can lead to complete data loss Automatic exposure bracketing is a useful feature on almost all cameras capable of taking RAW photos whereby the camera automatically adjusts the exposure range across three or more photos to intentionally under and over expose by a set number of exposure values Using exposure bracketing is a good idea if there s little cost to taking a couple of extra photos to ensure you get the perfect expo
47. hotographs extracted as linear 16 bit images Visible pass filter UV pass filter ai The relevant channels are extracted according to a camera configuraion file 2 The grey standard location s are selected manually Photograph channels are aligned automatically or manually The grey standards are measured following alignment gt A mspec file is created alongside the RAW photos BB visPhoto RAW saving the settings required E uvPhoto RAW k E mylmage mspec to recreate the normalised aligned multispectral stack 5 The multispectral image stack is converted to 32 bit floating point and normalised ready for objective measurement or conversion to cone catch quanta 6 Regions of interest and a scale bar can be selected These ROIs are saved with the mspec image for automatic measurement Regions with the same prefix can optionally be measured as one 7 The Batch processing tool can be used to measure multiple images using any selected visual system Image 9 Overview of the image processing work flow A multispectral image is a stack of images cap tured at different wavelengths A normal RGB photo is a multisoectral image with three wavelength bands captured red green and blue This software can handle a large number of additional image channels with different wavelength bands represented The most com mon extension is into the UV so in this case we give each channel a name wit
48. iol J Linn Soc 90 211 237 Stoddard M C and Stevens M 2010 Pattern mimicry of host eggs by the common cuckoo as seen through a bird s eye Proc R Soc B Biol Sci 277 1387 1393 Troscianko J 2014 A simple tool for calculating egg shape volume and surface area from digital images Ibis 156 874 878 Vorobyev M and Osorio D 1998 Receptor noise as a determinant of colour thresholds Proc R Soc Lond B Biol Sci 265 351 358 51 References Westland S Ripamonti C and Cheung V 2004 Characterisation of Cameras Comput Colour Sci Using MATLAB Second Ed 143 157 White T E Dalrymple R L Noble D W A O Hanlon J C Zurek D B and Umbers K D L 2015 Reproducible research in the study of biological coloration Anim Behav 106 51 57 52 References
49. itting enter it should list dcraw c in response 5 Paste the following command into the terminal and hit enter 19 Software lvm gcc o dcraw dcraw c Im DNO_JPEG DNO_LCMS DNO_JASPER 6 As you hit Enter the computer will download Xcode app compile and install the source code Ignore the warning messages on the terminal When it is done check the myDCRAW folder and there should be a file grey with green exe label named dcraw 7 Copy this deraw file not dcraw c across to your imagej plugins dcraw folder overwriting the existing file 8 Check whether it works in ImageJ by going Plugins gt Input Output gt DCRAW Reader and try to load a RAW image with the settings shown in Image 8 9 If it doesn t work it s possibly because the executable dcraw doesn t have permission to run To give it permission enter this into the terminal you may need to specify a different path if you ve placed imagej elsewhere chmod 777 Applications ImageJ plugins dcraw dcraw Memory Some multispectral images can take up a lot of memory If you get an error mes sage complaining that ImageJ is out of memory you can increase the memory dedicated to ImageJ in Edit gt Options gt Memory amp Threads Depending on how much memory is available on your computer you can increase the number e g 3000MB should be plenty for most multispectral images 20 Software IMAGE PROCESSING What is a multisoectral image 1 RAW p
50. n the luminance channel which is the combination of LW and MW sensit ivities and double cones in birds Osorio and Vorobyev 2005 If images are not being converted to cone catch quanta then the green channel is recommended Spottiswoode and Stevens 2010 or a combination of red and green which is available as an automated option Next the desired measurement scales must be selected in pixels along with the desired incrementation scale e g linear in creases of two pixels would provide measurements at 2 4 6 8 pixels and so on A multiplier can be used instead for example multiplying by two to yield measure ments at 2 4 8 16 pixels and beyond A size no larger than the scaled image di mensions should be used to cover the entire available range The number of scale increments should be judged based on processing speed although increasing the number of scales measured will yield progressively less additional information We find that increasing the scale from 2 by a multiple of V2 up to the largest size avail able in the scaled images produces good results in many situations Summary statistics Chiao et al 2009 Stoddard and Stevens 2010 of the pattern analysis are saved for each region of interest or pooled regions These include the maximum frequency the spatial frequency with the highest energy i e corres ponding to the dominant marking size the maximum energy the energy at the maximum frequency summed energy the en
51. nce values within a given study The effect of the surface angle i e the shape of the target and its diffuseness can be minimised by using a diffuse light source Shiny objects and complex 3D objects will therefore benefit most from diffuse lighting conditions Artificial light sources can be made more diffuse with standard photography umbrellas or diffusers in the human visible range However particular care must be taken when diffusing UV light sources as the diffuser must be UV reflective White plastic umbrellas and dif fusers should therefore not be used for UV sources Metal coated umbrellas or sheets of natural white PTFE Polytetrafluoroethylene easily bought online from plastic stockists are suitable for diffusing UV light sources In the field lighting angle and diffuseness can be difficult to control for If possible attempt to photograph only under sunny or overcast conditions not both If this is not practical or you are interested in the variation in natural lighting conditions then take note of the lighting for each photograph and take multiple photographs of each sample under all ecologically relevant lighting conditions Then lighting conditions can be entered into the statistical model during analysis 10 Taking Photos Grey standards At least one grey standard of known reflectance is required in each photograph or in a separate photograph taken under exactly the same conditions and with identical camera settings
52. nd minus symbols to change the scale to get a match The easiest way to use this is to align the top left of the image then change the scale value until the bottom right is aligned too see Image 15 26 Image Processing 2 2 1408x1804 pixels 16 bit 9 7MB 804 pixels 16 bit 9 7MB 2 2 1408x1804 pixels 16 bit 9 7MB iS The top left is aligned Iy Click and drag to align The bottom right is not aligned so needs scaling Accept J g Accept c G Click amp drag to align the layers so Change the scaling of the layers there is no blue or yellow fringe using the plus and minus buttons Align the top left of the image first until the whole image is aligned Repeatedly align the top left and scale until the bottom right is also aligned Image 15 Example of the manual alignment process Accent ute When the alignment looks perfect no blue or yellow fringing anywhere click Accept Auto alignment attempts to automatically find the best alignment between photos but can take a minute or two depending on the photo resolution and pro cessor power Automatic alignment works best on photos or regions of photos that have lots of detail shared between the visible and UV or other filter photos and does not work so well if your images have large plain areas Image 16 shows the output of auto scaling with the best scale and x y offsets Offset This is only used if Auto ali
53. nel For example if you measure a damaged 40 standard it might only have 35 reflectance in the camera s UR channel relative to the reference standard 37 in the uB channel 39 in the vB channel etc Measure each channel and input these values when prompted This setting will rarely be required and is included primarily for emer gency use to recover values if a standard is damaged Standards move between photos Tick if the standard was moved between visible and UV photos Otherwise the UV standard locations are measured after align ment to make sure exactly the same areas are being measured between different filters This setting will rarely be required Images sequential in directory If you have arranged your visible and UV photos in the same directory so that the UV photo always follows the visible photo alphabet ically this box can be ticked and it will select the UV photo automatically This also applies for more complex filter combinations and can save a bit of time searching through folders for the correct photos Alignment amp Scaling These options are ignored for visible only photography they are used with UV photography or any other number of filters to align the photo graphs Alignment There is almost always some misalignment between visible and UV pho tos which needs correcting The manual alignment option will let you manually align the photographs by eye by dragging the image with your mouse Click on the plus a
54. ngement ex Combinations or want to drop the UV blue ample for Visible amp UV photos channel for example The files specifying this are in imagej plugins Multisoectral Imaging cameras Add your own ixt file with the following format to specify your own camera filter setup Image 21 Each row specifies a filter type numbers in the RGB columns specify the position that channel should occupy in the stack zero ig nores the channel and the alignment channels specify which two channels to use when aligning the photos from different filters the first row is the reference im age so both values are zero Align is the global image number from 1 to 5 in this case align2 is the channel number relative to this filter i e red 1 green 2 blue 3 To remove the uB channel in this example you would change uv B from 4 to O and uv R from 5 to 4 In this case alignment is performed between channel 3 vis ible B and uv channel 1 uv R so this can be left as it is 34 Image Processing IMAGE ANALYSIS After generating a multispectral image and opening it as a 32 bit normalised im age you can measure the pixel values manually and these will be objective re flectance values though the colours will be camera specific By default the im ages are in the 16 bit range from zero to 65535 although being floating point the numbers can go higher if there are parts of the photo with higher reflectance than a white standard You can divide pix
55. nsity of the light source will likely affect these set tings so check histograms regularly Some cameras do a very good job of calcu lating the optimal exposure in which case A mode aperture priority can be used In this mode you set the aperture to use which should remain fixed across the whole study and the camera automatically decides on the best shutter speed You can then alter the exposure value EV or symbol on most cam eras telling the camera to under or over expose the photo by a set number of exposure values compared to what the camera thinks is best For example if you are photographing something against a dark background the camera will gener 13 Taking Photos ally try to over expose so selecting EV 1 will tell the camera to under expose slightly DSLR cameras tend to be good at working out the optimal exposure in vis ible wavelengths but are poor at working out UV exposures so Manual control M is recommended for UV photography Mirrorless cameras such as the Samsung NX1000 tend to do a better job in UV as their exposures are based on the light hit ting the sensor rather than dedicated light meters so can often be used in aper ture priority mode In either case though it is offen possible to use the exposure val ues as references that will vary between camera models and will require some ex perimentation to get used to See Image 6 for examples ji Qe ean an en Oa 2 The camera thinks t
56. ntal banding this is caused by a flick ering light source and a much longer exposure shutter soeed should be used to eliminate this effect if no other source is available Also be aware that many artifi cial light sources change their soectra as they warm up over a few minutes could change with voltage fluctuations in the mains or battery supply and can change as the bulb ages So always try to get a grey standard into each photo graph if possible rather than relying on the sequential method taking a photo graph of the standard before or after the target has been photographed 8 Taking Photos Sunlight eyeColor Arc Lamp Fluorescent bulb a pale 700 300 40 700 300 400 600 700 Pana Pa nm Image 3 Emission spectra of different light sources normalised so that the max imum spectral radiance 1 The sun is the most ecologically relevant light source and provides a good broad emission spectrum but this also varies sub stantially with time of day latitude and atmospheric conditions Arc lamps have relatively spikey emission spectra but this example Iwasaki eyeColor with its UV filter removed see below does a relatively good job of recreating sunlight Fluorescent tubes are very poor light sources for accurate colour ren dering because of their very sharp spikes with some wavelengths almost entirely absent and others over represented These spikes can interact with reflectance spectra to make colours look quite different to
57. oecify their own reflectance value The X rite colorChecker passport is convenient as it has a range of grey levels Lighting Sunlight camera flash arc lamps incandescent bulbs or white phosphor based LED lights are suitable Avoid fluorescent tube lights see page 8 Use the same light source for all photographs if possible Try to use a diffuse light source for shiny or 3 dimensionally complex objects see page 10 5 Equipment Converting to cone catch images If your hypotheses depend on the appearance of the object to a specific visual system model species or absolute measures of colour are required then cone catch images are recommended This conversion produces images based on the spectral sensitivities of a given visu al system The images are device independent different cameras or soectromet ers should all produce the same results The equipment required will depend on the sensitivities of the visual system in ques tion Many species are sensitive to ultraviolet UV light in which case a UV camera set up would be required This entails a camera converted to full soectrum sensitiv ity two filters passing visible and UV light respectively and grey standards that have a flat reflectance across the 300 700nm range such as sintered PTFE stand ards If photographs are being taken in the lab then a UV light source is also re quired see below If your model visual system is only sensitive to wavelengths in th
58. on Polynomial Slice Transform 32Bit Convert to Cone Catch Tools Pattern Colour amp Luminance Measurements Image 22 Batch image analysis To measure the colour pattern and luminance data in a series of multispectral im ages use Plugins gt Multispectral Imaging gt Image Analysis gt Batch Multispectral Im age Analysis Image 22 Make sure you select a folder that contains all your mspec files alongside their associated RAW files Once you have selected the directory a dialogue box will ask you about visual system and scaling options Image 23 Image Scaling Model Select the model for converting from For pattern analysis all images must have the A i same scale scale bars must already be added camera colours to your desired visual system If to the image ROIs our camera visual system model is not available Scale px mm fio y Y ir ae and you know the spectral sensitivities of your Set to zero to turn off scaling If you re not sure what scale to use run the Batch Scale Bar camera you can generate a mapping function Calculation tool on this folder first i see page 31 If you do not know your camera s Start processing at file number E E spectral sensitivities select none to measure the _oK Cancel Image 23 Batch processing scal 36 Image Analysis ing and visual system options normalised camera values the images will still be objective Add human luminance channel Tick this box if
59. rocessing tab gt under Default processing profile for RAW photos profile select Neutral Using any other profile might switch on automatic exposure compensation which makes it more difficult to work out whether a photo is over exposed Image 10 RAWTherapee can be used for screening RAW photos Ensure the pro cessing profile is set to Neutral with no exposure compensation applied The RGB histograms in RAWTherapee can be used to determine which photos 22 Image Processing have the optimal exposure These histograms show the number of pixels in the photo across a range of intensity levels from zero on the left to maximum satura tion on the right The ideal exposure for the whole image will have a histogram with peaks spread evenly from left to right without any pixels quite reaching the right hand side implying they are saturated If the sample you are measuring is not as bright as other parts of the image then you can check whether it appears over exposed by running your mouse over the region you will be measuring As you do this the RGB levels under the cursor will be shown below the histogram If these values reach the right hand side of the histogram they are saturated and this re gion should not be measured Image 11 right shows a histogram of a well exposed photograph The peaks on the right of the histogram correspond to the pixels of the white standard while the large number on the left represent the dark grey b
60. rrreses 19 Memoer e a cee vad N ae ead es L ETE 20 IMAGE PROCESSING acrin ieee o EN E E R e E 21 Whot is a multispectral IMAGE 2 4 caring iisessoti eid idasdudeawiaahe diate aie en ess 21 Checking Photographs RAWINEFO DEES c 6 casseseceoseveeseseseteysovsinesuceoedeverteesesens 22 Creating Multispectral IMAJES seeeseeessesrssersserseersserssersserssersserseerseeeceeeeeseereeeeeese 24 Select regions of interest and a scale DOM ssscessessssessesssccnscenssscccceseeeeesssecs 28 Reloading multispectral iMAJES seessesssessessssessseseresereseresereseresresseeeseeeseessreesresseees 30 Generating A new CONE mapping MOAEY L cccesccesscesscesscesscecscesscesscesscesseessecs 30 Generating a custom camera and filter araNngeMent esseesseeeerereresserrerrseees 33 IMAGE ANALYSIS aose e E A EE iara EEEE A ETE T tele 34 B tech image CINGIYSIS a E EROE E E AE E A A OR EAR 35 Batch image analysis FOSUITS ccc tices cssuclvan viens ten ie scteuevan la ar sel ween gus annie Cla ede 39 Granularity pattern analysis INTTODUCTION cssccsseessecsssensccnsccsscenscesscenseenssesessce 39 Calculating pairwise pattern and luminance CIffErENCES cceeeeecseeececeeeeeeees 42 Colour differences iIN rodUCtIOnN ssssessessssereesssseeessseresssneerssssereessssssssssssseeereeerees 43 Calculating pairwise colour and luminance discrimination Values 43 PRESENTING DATA ao an anea lds EAA A EE E Ea E ENE a S 46
61. s size As a rule this size should be equal to or smaller than your smallest sample s longest dimension E g of your smallest sample is 50mm long and your px mm is 20 this sample will be 50x20 1000 pixels long So the end size should be equal to or smaller than 1000px Step Size Specify what scaling number to use for increasing the scale from the start size to the end size Step multiplier Soecify whether the scale should increase exponentially or linearly The example in Image 25 will measure 13 levels between 2 and 128 pixels increas 38 Image Analysis ing as a multiple of 1 41 i e 2 2 8 4 5 7 8 etc Exponential steps generally make more sense than linear given the Gaussian filtering involved with the Fourier band pass Output pattern spectrum Tick this box if you want to measure pairwise pattern dif ferences between samples It saves the pattern data across all spatial scales oth erwise only the descriptive statistics are measured Luminance bands This saves a basic luminance histogram used for pairwise com parisons of multi modal luminance distributions For example if you are measuring samples that have discrete patches of luminance such as zebra stripes that are either black or white it does not always make sense to use the mean grey in this case to compare zebra to backgrounds as this grey level is not actually found on the zebra You can set how many levels to save around 20 to 100 might be sens ible
62. scale bar If you are planning to compare patterns between multispectral images or want to scale your images to a uniform number of pixels per unit length then you need to add a scale bar to your image To do this select the line tool and draw a line along the scale bar in your image and press S A window will pop up asking you how long the length you selected was Type in a number only no text The units of the number used here must be the same for the whole study E g the scale numbers across the whole study should all be in mm cm or m but not a combination For measuring eggs the toolbox includes an egg shape and size calculation tool that also makes it easy to select an egg shaped ROI Troscianko 2014 Select the multipoint tool right click on the point tool and select multipoint if it is not already selected then place points on the tip and base of the egg and three more points down each side of the egg 8 points in total though you can add more points Then press E and it will highlight the egg shape If you are happy with the fit of the egg shape click accept and the script will add this ROI in addition to calcu lating various egg metrics such as length width volume surface area and pointed ness shape Remember to press 0 when you re done selecting ROIs and the scale bar If you 30 Image Processing leave the log window open it will automatically know where the relevant mspec file is otherwise
63. sure you can look at the photos on the computer later to judge which is the best exposure or when the contrast of the samples being photographed is highly variable Note that some cameras automatically take all three or more photos with one press of the shutter button but most for models you need press the shutter 14 Taking Photos once for each photo Scale bar Pattern analysis requires all images be scaled to a uniform number of pixels per unit length a number that will vary with every study tools are provided to calculate this see page 37 Photographing different samples from slightly different distances can be controlled for with a scale bar nevertheless we recommend taking photo graphs at uniform distances from the subject whenever possible Particular care should be taken to photograph different treatments at the same distances for un biased pattern analysis Position the scale bar level with the sample and photo graph from overhead if possible If photographing from an angle rather than dir ectly overhead a horizontal circular disk could be used as a scale bar the disk s maximum width will always equal its diameter whatever angle it s viewed from or place a straight scale bar ruler side on to the camera Even if you initial hypothesis does not concern pattern or size it is always good practice to include a scale bar Taking photos check list Photograph in RAW format not JPG The camera white balance is not important
64. t to change the name at this stage Rename RAW files Choose whether to rename your RAW photos with the image name selected above Renaming these files can be useful to show you which RAW files have been processed and are associated with which mspec file Only use this option for renaming RAW files if you rename the files yourself the mspec file will not be able to find them Always keep the mspec files in the same folder as its associated RAW photos Click OK when you are done The settings you specify here will be saved as de 28 Image Processing faults for next time you make another image The script will guide you through the process of selecting the RAW image s high lighting the grey standards you can use a rectangle tool circle tool or polygon tool and optionally ALWAYS CHECK THE ALIGNMENT FOR UV PHOTOGRAPHY Once the multispectral image has finished it will show you the result If you selected normalised 32 bit as the output format then the script will automatically flip between the various chan nels to and you can see how well aligned they are you can also manually flip between the channels using your scroll wheel or keyboard left amp right arrows If there is movement between these images then the alignment has not worked and needs to be repeated e g using manual alignment If you selected pseudo UV as the output then misalignment will look like the blue channel being out of align ment with the red
65. the sample you are measuring has lots of discrete colours then these regions should be measured in different ROIs 44 Image Analysis or an image quantisation method should be used File Edit Image Process Analyze Plugins Window Help E o ESRA t t SASDA Pe Stk 4 e 7 4 File Edit 3943 234000000 4102 931000000 3922 849000000 8090 200000000 7600 507000000 5943 819000000 7859 438000000 8317 790000000 8567 860000000 5188 365000000 5704 825000000 4447 847000000 4583 491000000 5116 056000000 4279 495000000 7471 403000000 Font Results 4656 4450000 4589 1540000 4109 8820000 4573 1940000 5333 6760000 3107 0150000G 5403 103000000 6528 116000000 5884 473000000 4051 046000000 4467 796000000 4148 739000000 4098 471000000 3736 628000000 3887 733000000 4414 288000000 Calculate colour JND differences Visual system Weber fractions peafowl 0 05 Compare this region egg to this region Region comparison 8583 900000000 8864 813000000 9162 111000000 5641 697000000 6016 698000000 4589 012000000 4653 147000000 5314 165000000 4578 848000000 7811 201000000 406000000 within Photo OK Cancel 5833 693000000 6702 693000000 6152 486000000 4625 117000000 4935 323000000 4678 507000000 4557 957000000 4243 682000000 4478 428000000 5058 750000000 187000000 12296 3510000
66. this image report which channels OK Cancel are being displayed and that the linear values Image 31 Select chan were square root transformed to optimise it for nels to convert to a col displaying See Image 32 for an example our image suitable for hu man viewing This ex ample will create a false 49 Presenting data colour UV image from a bluetit cone catch im age 1 uy Ba y 2 Example of Gallotia galloti male and female lizards On the left is a normal human CIE XYZ image on the right is a false colour bluetit im age showing MW SW and UV channels Both images are square root trans formed from linear cone catch images for display purposes The lizard s blue cheek on the right hand image demonstrates that this patch has very high UV reflectance relative to MW and SW Image 50 Presenting data REFERENCES Chakrabarti A Scharstein D and Zickler T 2009 An Empirical Camera Model for Internet Color Vision In BMVC Citeseer p 4 Chiao C C Chubb C Buresch K Siemann L and Hanlon R T 2009 The scaling effects of substrate texture on camouflage patterning in cuttlefish Vision Res 49 1647 1656 Endler J A and Mielke P W 2005 Comparing entire colour patterns as birds see them Biol J Linn Soc 86 405 431 Godfrey D Lythgoe J N and Rumball D A 1987 Zebra stripes and tiger stripes the spatial frequency distribution of the pattern compared to that of t
67. though this number is somewhat arbitrary Larger numbers of pixels could sup port a higher number of bins If your histograms are always skewed to dark pixels you could apply the log or square transform here Combine ROls with prefix Here you can specify any ROI groupings E g entering f g as in Image 25 will group together the petals of each respective flower from the example in Image 17 treating all objects labelled F as one object and all G objects as one You can leave this box empty out to measure every ROI indi vidually e g this makes it easy to do inter flower measures and _ intra flower meas ures Output energy maps Outputting the en ergy maps saves tiff images of the pattern analysis across all spatial scales e g Im age 26 This is mostly just for checking everything is working correctly e g just run it on a sub sample of files to check it is grouping objects together correctly These are quite large files that you can delete once you have had a look at them they are not required for any measurements Image 26 Examples of pattern energy map outputs 39 Image Analysis Click OK and it will measure all mspec images in the selected folder with the set tings you have chosen Batch image analysis results File Edit Font Results Labei __ lummean maxFreq propPower sumPower flowers_f 16854 142 4324620 23370 707 843 8 000 0 136 5223 672 401 621 191 531 23460606 4464 flow
68. tripods are essential for UV photography Consider using exposure bracketing to get the best exposure and regularly check photo histograms on the camera to make sure they are well exposed The bars of the histogram should be as evenly spread from left to right as possible without quite touching the right hand side Always check the photos look good that everything important is in focus and fills the frame as much as possible 16 Taking Photos SOFTWARE Software check list ImageJ version 1 49 or more recent with Java available from imagej nih gov ij Schneider et al 2012 32 bit or 64 bit versions are supported Ubuntu available from the software centre Note that on Linux ImageJ is more stable with Java version 6 though it does generally work with version 7 Download the Toolbox files from www sensoryecology com or www jolyon co uk Unzip the contents of the zip files and place them in your imagej plugins directory see below for further details e Optional R from www r project org R Core Team 2013 required for creating new cone catch mapping models for converting from camera vision to animal vision Once the models have been made for a given camera and visual systems R is not required e Optional RAWTherapee from rawtherapee com This is useful for screening RAW photographs before processing Installation Install ImageJ following the instructions for your operating system Downlo
69. you re working with human vision This runs a script that adds human luminance as LWS MWS 2 the average of red and green thought to code for the human luminance channel Image Scaling If you are not doing pattern analysis f and have not selected scale bars in your images set this to zero and no image scaling will be per formed For pattern analysis all your images need to have the same pixels per unit length Every series of photos will have its own ideal scale that depends on the size of the samples you re working with Gen erally you want fo set this so that it reduces the size of all your images slightly enlarging the images would create false data The Batch Scale Bar Cal culation tool will tell you which scale bar is the smallest and whether this minimum value looks too small compared to your other images just to make sure you are not using an anomalously small one Try using this minimum value or round down to an integer and if the processing is going too slowly choose a smaller number halve the number and it should go about four times faster To use this tool go Plugins gt Multispectral Imaging gt Image Analys is gt Batch Scale Bar Calculation All scaling is per formed as bilinear interpolation to minimise the cre ation of artefacts File Edit Font Scale Bar Statistics Path D Scale bar count 1 Minimum px mm 102 0031 Maximum px mm 102 0031 Mean px mm 102 0031 SD px mm NaN Th

IMAGE CALIBRATION AND ANALYSIS TOOLBOX USER GUIDE

Contents

Download Pdf Manuals

Related Search

Related Contents