Image Processing Laboratory - Department of Electrical and
Contents
1. Gaussian the coefficients are all positive This produces a smoothing effect or low pass filtering of the image This can be useful for removing noise from an image 5 Why would someone want to enhance the edges that are in an image By enhancing the image edges specific particles can be easier picked out 6 Explain the benefits of using a wavelet analysis instead of a traditional FFT 2006 Cleveland Medical Devices Inc Cleveland OH 10 Property of Cleveland Medical Devices Copying and distribution prohibited CleveLabs Laboratory Course System Version 5 0 Clevel_abs amp Laboratory Course System CleveLabs Laboratory Course System Teacher Edition Image Processing Laboratory A wavelet analysis increases the time resolution in a spectral analysis A traditional FFT must be completed over a long time period to increase the spectral resolution This limits the ability to detect at what particular point in time a frequency occurred 7 What is the effect of the number of coefficients that are used to reconstruct an image using wavelet analysis A minimum number of coefficients must be used to accurately represent an image However the total number of coefficients may be much less that 100 of the number used in the original image 8 Describe a novel method that you could create to automatically quantify particles in an image This answer will depend on the creativity of the students 2006 Cleveland Medical Devices2. Inc Cleveland OH 11 Property of Cleveland Medical Devices Copying and distribution prohibited CleveLabs Laboratory Course System Version 5 0 ClevelLabs E amp 8 Laboratory Course System CleveLabs Laboratory Course System Teacher Edition Image Processing Laboratory References 1 National Instruments Corporation NI Vision IMAQ Vision Concepts Manual January 2005 2 National Instruments Corporation LabVIEW and LabWindows CVI Signal Processing Toolset User Manual December 2002 3 Webster John G Medical Instrumentation Application and Design Third Edition John Wiley and Sons New York 1998 2006 Cleveland Medical Devices Inc Cleveland OH 12 Property of Cleveland Medical Devices Copying and distribution prohibited CleveLabs Laboratory Course System Version 5 0
3. this laboratory session we will explore four kernels including gradient Laplacian smoothing and Gaussian Each convolution kernel that we will examine in this session is a linear filter These filters can act as either high or low pass filters A high pass filter such as gradient and Laplacian kernels includes coefficients both positive and negative This creates a weighted differentiation that improves contrast in the picture The effect of the filter is that areas of high transition are amplified and brought out of the image On the other hand in some of the kernel transfer functions such as smoothing and Gaussian the coefficients are all positive This produces a smoothing effect or low pass filtering of the image This can be useful for removing noise from an image ES 4 s Figure 2 The edges of the original image left can be enhanced by spatial filtering techniques right 2006 Cleveland Medical Devices Inc Cleveland OH 3 Property of Cleveland Medical Devices Copying and distribution prohibited CleveLabs Laboratory Course System Version 5 0 Clevelat Laboratory Course System CleveLabs Laboratory Course System Teacher Edition Image Processing Laboratory Specifically a gradient filter is a high pass filter that sharpens the differences in light intensity along a specific direction This can be useful for enhancing edges in an image along a particular direction A Laplacian filter sharpens the differ
4. 00 and the y resolution to 300 This will increase the size of the image _Image Translation Angle dearees sp 2006 Cleveland Medical Devices Inc Cleveland OH 6 Property of Cleveland Medical Devices Copying and distribution prohibited CleveLabs Laboratory Course System Version 5 0 Clevel_abs Laboratory Course System CleveLabs Laboratory Course System Teacher Edition Image Processing Laboratory 5 Since the processed image now has more pixels than the original image interpolation techniques must be applied to interpolate between pixels that exist and those that do not Examine the different types of interpolation techniques to see the effects on the processed image 6 Finally adjust the symmetry of the image to see the effects on the processed image 7 Now click on the tab labeled Gray Scales In this section you will examine how gray scale techniques can be utilized to improve the contrast of an image We will once again use the image of the spine for this example 8 The range of 0 255 represents the gray scale color of a pixel Adjust the scale to range from a minimum of 0 to a maximum of 50 Then click on process image and notice the effects on the processed image and the histogram Capture a screen shot of this Gray Scale Histogram SS 9 Adjust the scale to range from a minimum of O to a maximum
5. Algorithm Threshold Values Minimum Value 4 38 al Maximum Value ll 32 4 i QO Particles Found Circle Algorithm Max Radius Ci J I I i 1 I I I I i 1 I 0 50 100 Figure 3 Automated techniques for counting particles in an image can be used to save time 2006 Cleveland Medical Devices Inc Cleveland OH 4 Property of Cleveland Medical Devices Copying and distribution prohibited CleveLabs Laboratory Course System Version 5 0 Clevel_abs E amp Laboratory Course System CleveLabs Laboratory Course System Teacher Edition Image Processing Laboratory Wavelets Wavelet theory has been around since the early 1900 s It was designed as a way to improve the resolution that a standard fast Fourier transform FFT can provide In general an FFT can provide good frequency resolution over a long time period However this comes at the expense of not knowing when during that time interval that particular frequencies occurred On the other hand short time FFT s can provide a better indication of when a particular frequency occurred but at the cost of a reduced frequency resolution In very general terms wavelets provide a function that can be stretched and shifted in time They can be applied in time steps over an entire signal A wavelet function can improve the ability to detect when particular frequencies occur in time They are applied in a wide variety of applications While it is typically thought
6. Clevel_abs E Laboratory Course System CleveLabs Laboratory Course System Teacher Edition Image Processing Laboratory 2006 Cleveland Medical Devices Inc Cleveland OH Property of Cleveland Medical Devices Copying and distribution prohibited CleveLabs Laboratory Course System Version 5 0 ClevelLabs E amp Laboratory Course System CleveLabs Laboratory Course System Teacher Edition Image Processing Laboratory Introduction Image processing has become an increasingly important field in biomedical engineering today By expanding our ability to view the body we can better study and treat disease Not only do novel imaging techniques provide researchers and clinicians with the ability to diagnose and monitor disease progression but imaging can also be used to guide surgical procedures New developments of biomedical engineering imaging techniques have increased the quality and applications of imaging This remains a highly active area of research with many departments now focused on designing new imaging equipment and improving the techniques to process and i gure i ima ge proce ine analyze collected data In addition to medical applications techniques may be applied to imaging is extremely useful in other areas including machine automated fingerprint detection vision for automatically sorting parts law enforcement for identifying biometrics such as facial features or fingerprints Fig 1 and character recogni
7. Edge Enhancements Depending on the type of feature you are trying to extract from an image particular types of filtering may improve the features For example if the image is noisy it may be useful to filter out some of the artifact Similarly if you are looking to extract particular features providing a sharper contrast between edges may enhance the quality of feature extraction These types of image processing techniques can be completed using convolution kernels A convolution kernel is a two dimensional filter that can be applied to a gray scale image Convolution kernels can be used to remove noise from an image smooth an image or enhance the edges There are particular types of convolution kernels each with their own properties that produce different results when applied to an image The different types of convolution kernels are distinguished by their particular coefficients When an image is filtered the new resulting pixel is a function of itself and the weighted combination of surrounding pixels prior in the original image Variables in the convolution kernel algorithm include the number of surrounding pixels that are used in the filtering algorithm as well as the different weights coefficients that are applied to those neighboring pixels There are several different types of kernels that may be applied to images Kernels may be standard functions or user defined coefficients designed for particular applications For the purposes of
8. alysis filter is swapped with the synthesis filter Original Image 2006 Cleveland Medical Devices Inc Cleveland OH 5 Property of Cleveland Medical Devices Copying and distribution prohibited CleveLabs Laboratory Course System Version 5 0 Clevel_abs E amp Laboratory Course System CleveLabs Laboratory Course System Teacher Edition Image Processing Laboratory Experimental Methods Experimental Setup Make sure your BioRadio receiver is connected to your computer before starting this laboratory session If the receiver is not connected most of the functionality in the laboratory session will be disabled Procedure and Data Collection 1 Run the CleveLabs Course software Log in and select the Image Processing laboratory session under the Engineering Basics subheading and click on the Begin Lab button 2 Click on the tab labeled Manipulation This section will examine some basic concepts in manipulating images To the left is the original image of a spine To the right is the processed image based on the parameters that you select 3 First adjust the rotation angle of the image to examine the effects on the processed image Then adjust the offsets to examine their effect on the processed image 4 Next reset the angle and offsets to 0 Now we will adjust the resolution and interpolation methods of the processed image Set the x resolution to 2
9. at particular section This means that the section must be enlarged or re sampled Since the number of pixels in the enlarged image is greater than the number of pixels in the original image the area between the true pixels must be interpolated There are several types of interpolation functions that could be used to resample image data Some of these functions include a zero order a cubic spline a quadratic and a bi linear Each of these functions has a unique impact on the resized image A zero order function rounds to the nearest integral pixel location A bi linear function interpolates in both the horizontal and vertical direction to compute the pixel location A quadratic function uses the quadratic formula to compute the location of the pixel Finally cubic splines are also used to compute the location Gray Scales Images may be in black and white gray scale or color For the purposes of this laboratory session we will focus on gray scale images One tool for analyzing images is a histogram This can be a useful technique when analyzing gray scale images A histogram can provide information on the number of pixels that fall within specific ranges of intensity Therefore the histogram yields if there is a particular region of the image that has the same intensity For example a particular type of noise may show up in one area of the histogram Lookup tables can be used to highlight details in certain areas of an image at the cost of r
10. e the effects on the gray scale 13 Now click on the tab labeled Edge Enhancement Click on the Reload Original Image button to load the original image on the screen This will be the default image that appears 14 Next change the kernel family to Laplacian and click on the Process button Notice what features of the image change Click on the Process button again and notice how the image changes again 2006 Cleveland Medical Devices Inc Cleveland OH Property of Cleveland Medical Devices Copying and distribution prohibited CleveLabs Laboratory Course System Version 5 0 Clevel_abs Laboratory Course System CleveLabs Laboratory Course System Teacher Edition LS 16 we 18 19 20 2i 22 23 Image Processing Laborator Click on the Reload Image button to load the original image on the screen Repeat step 14 for the other kernel families Now click on the tab labeled Quantification This section will provide insight on how automated image processing algorithms are implemented to quantify specific features of an image Two algorithms will be examined including a particle algorithm and a circle algorithm Under the particle algorithm set the minimum and maximum gray scale levels that define a particle area Then click on the Process button The number of particles will automatically be displayed based on the algorithm results Now define para
11. educing the quality of other areas These tables are used to improve the contrast and brightness of an image by modifying the dynamic intensity of regions with poor contrast When a lookup table transformation is applied in a particular transfer function it changes the value of each pixel in the gray scale image A range of minimum and maximum values can be specified over which this lookup table will operate There are several transfer functions available in this laboratory session including linear log exponential square square root power x and power I x The linear transfer function increases the intensity dynamic by evenly distributing a given gray level interval over the full gray scale The log power 1 x and square root functions increase brightness and contrast in dark regions but decrease the contrast in bright regions The 2006 Cleveland Medical Devices Inc Cleveland OH 2 Property of Cleveland Medical Devices Copying and distribution prohibited CleveLabs Laboratory Course System Version 5 0 Clevel_abs E amp Laboratory Course System CleveLabs Laboratory Course System Teacher Edition Image Processing Laboratory exponential power x and square functions decrease the brightness and contrast in dark regions but increase the contrast in bright regions Finally equalization modifies the gray scale values in the image and redistributes them evenly in the gray scale range This acts to maximize the contrast in the image
12. ences in light intensity that surround a pixel This is useful for sharpening the contour of an object In contrast a smoothing filter attenuates variations in light intensity This smoothes the shape of an object and can remove details The low pass kernel functions basically tend to act as an averaging filter Finally the Gaussian filters also smooth however the effect of blurring is not as prominent as in the smoothing filter Image Quantification Several methods exist to automatically quantify features of images One method is particle analysis In general the aim of particle analysis is to use an algorithm that detects regions where there are many pixels of the same intensity Specific particles in an image can be measured based on their location area and shape One example of where this technique could be useful is a researcher that wants an automated technique for counting the number of cells in an image One method is a particle analysis A particle is defined as group of contiguous nonzero pixels The spatial characteristics of the group must satisfy specific criteria For example you may define the minimum and maximum intensity that defines a group of pixels that qualifies as a particle Once the particle is defined and located other statistics and measures can be applied to the particle for a more complete analysis For example you may want to further separate the particles based on a maximum and minimum size criteria Particle
13. endent upon the bit depth of the image For example if it 1s an 8 bit image the pixel number may range from 0 255 while if it is a 16 bit image the number could range from 0 65 535 The spatial resolution of an image refers to the number of pixels that were used to 2006 Cleveland Medical Devices Inc Cleveland OH 1 Property of Cleveland Medical Devices Copying and distribution prohibited CleveLabs Laboratory Course System Version 5 0 Clevel_abs amp Laboratory Course System CleveLabs Laboratory Course System Teacher Edition Image Processing Laboratory create the image An image is typically made up of a number of rows and columns of pixels The total number of pixels in that matrix is the image spatial resolution The more pixels the better the spatial resolution of the same size image Image Manipulation One of the simplest processing techniques that can be completed with images is translation and rotation This is fairly simple in the two dimensional example however it gets more complex in three dimensions In two dimensions all of the pixels of an image can be translated or shifted along the x or y axis Additionally the image may be rotated about its center point The ability to translate and rotate an image becomes important when a researcher is attempting to focus in on a particular aspect or feature of the image Once a particular portion of the image has been selected a researcher may want to zoom in on th
14. meters for the circle algorithm Set the minimum and maximum number that a radius should be to define a circle Then click on Process The number of circle particles will be automatically calculated Now click on the Wavelet tab Select the cell image file Adjust the number of coefficients used to reconstruct the image Find the threshold at which the image no longer is accurately represented Adjust the wavelet type to examine what effect different wavelet functions have on the reconstruction Repeat steps 21 and 22 for all images 2006 Cleveland Medical Devices Inc Cleveland OH 9 Property of Cleveland Medical Devices Copying and distribution prohibited CleveLabs Laboratory Course System Version 5 0 Clevel_abs E amp Laboratory Course System CleveLabs Laboratory Course System Teacher Edition Image Processing Laboratory Discussion Questions 1 Why is imaging important in biomedical engineering What are some practical applications of imaging Imaging is an important area in biomedical engineering It can provide insight into many functions of the body that cannot be seen with the naked eye Imaging can be used to diagnose tumors and abnormal heart conditions Additionally it can be used to guide decisions during surgery 2 Discuss how image translation rotation and re sampling can be important in biomedical imaging What impact does the interpolation technique have on the processed image Usi
15. ng these functions can help a researcher zoom in on a particular part of an image for analysis Different types of interpolation techniques can affect the degree of smoothing that occurs between original pixels 3 What are the different gray scale operators that can be used to process an image Describe the impact of each on the processed image There are several transfer functions available in this laboratory session including linear log exponential square square root power x and power 1 x The linear transfer function increases the intensity dynamic by evenly distributing a given gray level interval over the full gray scale The log power 1 x and square root functions increase brightness and contrast in dark regions but decrease the contrast in bright regions The exponential power x and square functions decrease the brightness and contrast in dark regions but increase the contrast in bright regions 4 What is the difference between a high pass and low pass spatial filter What are particular applications for each These filters can either act as high or low pass filters A high pass filter such as gradient and Laplacian kernels includes coefficients both positive and negative This creates a weighted differentiation that improves contrast in the picture The effect of the filter is that areas of high transition are amplified and brought out of the image On the other hand in some of the kernel transfer functions such as smoothing and
16. of 255 Then click on process image and notice the effects on the processed image and the histogram Capture a screen shot of this essin __ Image Processing Laborator Background Manipulation Gray Scales Edge Enhancement Quantification Wavelets Orignial Image Processed Image Gray Scale Range Minimum Value 2 0 Maximum Value 255 0 87 174 255 X Value 3mo Gray Scale Histogram i eea ri S i g 150 y A i My a Mh Way 2006 Cleveland Medical Devices Inc Cleveland OH 7 Property of Cleveland Medical Devices Copying and distribution prohibited CleveLabs Laboratory Course System Version 5 0 Cleve_abs E amp E Laboratory Course System CleveLabs Laboratory Course System Teacher Edition Image Processing Laboratory 10 Adjust the scale to range from a minimum of 100 to a maximum of 255 Then click on process image and notice the effects on the processed image and the histogram Capture a screen shot of this Image Processing Laboratory 11 Now click on the Equalize button and notice the difference in the processed image Capture a screen shot of this Background Manipulation Gray Scales Edge Enhancement Quantification Wavelets Minimum Value f 0 Maximum Value 1 00 Orignial Image rocessed Image si Gray Scale Histogram 12 Try adjusting the different operators to examin
17. of as time frequency this is spatial frequency and many of the same techniques apply Wavelet transforms allow an infinite number of basis functions to be used in an analysis There are many existing wavelet waveforms that can be used Additionally an individual can design a unique wavelet function A common application of wavelets in image processing is to de noise an image This can be completed using a 2D wavelet transform Users can select wavelet functions that best match the spatial characteristics of the image The wavelet transform of the signal is taken and the coefficients below a certain threshold are set to zero inverting the transform then reconstructing the original signal The image can be reconstructed with a significantly less percentage of the original data When reconstructing an image using wavelets there are several parameters to select The first is to select the wavelet basis function The next is to select the extension type Zero padding adds zeros at the beginning and end of the original data Symmetric folds the original signals at the beginning and end Periodic treats the original data as a periodic signal and spline symmetrically folds the original signal at the beginning and end and then smoothes with spline You also must specify the percent of coefficients to be used in the reconstruction The number of levels specifies the number of levels of the wavelet packet decomposition The swap control specifies whether the an
18. tion in personal computing devices to convert written text to digital form a od a ap aei eimi i m Background There are two main areas of research in imaging The first involves the development of hardware and sensors to acquire the image Radiography was the first tool used to view the inside of the human body X rays are electromagnetic radiation An x ray system includes a high voltage generator an x ray tube a collimator the patient an intensifying screen and a film Another imaging technique is computed tomography CT While a conventional x ray is limited because it is generated by projection from the source through the object the CT scan forms a three dimensional image by scanning in many directions The information from the scans is digitally reconstructed to form the image Magnetic resonance imaging and nuclear medicine are two other advanced imaging techniques used to gain insight into the human body The second aspect involves processing the image to improve quality and feature extraction The purposes of this laboratory session will be to introduce and apply algorithms for processing acquired images The software for this laboratory session was developed using National Instruments Vision Development Module Pixels Pixels are the discrete points that make up an image Each pixel has a particular light intensity that 1s represented by a particular numerical value The range of numbers that a pixel can represent is dep
Download Pdf Manuals
Related Search