The Pocket Handbook of Image Processing Algorithms in C Harley R. Myler Arthur R. Weeks Department of Electrical & Computer Engineering University of Central Florida Orlando, Florida a Prentice Hall P T Ft Englewood Cliffs, New Jersey 07632Latrary of Congress Catatoging-in-Publicwlion Dawa Myler, Harley R., ‘The pocket hand book of image processing algorithons in C/Hailey R. Afyler, Arthur R. Weeks. 93-779] cr Edivorial prod Cover desiga: Buyer: Mary © sth Acquisitions Englewood Cliffs, New Jersey 07632 GIF (Graphie Interchange Forman is copynghied/trade taarked by Conmpuservs, Inc... TIFF (Tagged interchange File Formal} was developed by Microsait Car- avon and Aldus Corporation, PCK is a wademark of Z-Soft Corporation, Risrosoft and MS-DOS are trademarks of Microsof Comporalion, MacPaunt and Macintosh are uademarks of Apple Computer. Alll rights reserved. No part of this book may be repreduced, an any form or by any means without permussion in writing from the publisher. Printed in Ux United States af America log k76 ISBN O-13-b4e240-3 i) 8043642240 Prentice: Hall Iniematignal ALK Lina dead” Prentice-Hall of Australie Ply. Limited, Spiny Prenuce-Hall Canada Inc, Foronia Prenvce- Hall Hispancamencana, $.A., Meneo Prentice-Hall of India Private Limited, Mew Dethe Prentice-Hall of Japan, Inc., Tokyo Simon & Schuster Asia Pte. Lid., Sagapore Editora Prentice-Hall do Brasil, Lida. Rio de JanesroTe Nancy, Krifka and Logan Te Dian, Anne, Frank, Anne and RichardTable of Contents Preface How to Use This Book Topical Layout Finding Topics Algorithm Programming Class Groupings Index Adaptive DW-MTM Fier » Adaptive Filters . Adaptive MMSE Filter Alpha-Trimmed Mean Filter Area Arithmetic Mean Filter Brightness Correction €.LE. Chromaticity Diagram Centroid Chain Code Circularity Circularly Symmetric Filter Closing (Binary) Filter Closing (Graylevel) Filter Clustering + Coding and Compression * Color Image Processing Color Saturation Correction Color Tint Correction Compactness Contta-Harmonic Mean Filter Contrast Correction Dilation (Binary) Filter sre 13 16 RRES 29 32 35 37 42 4) 4g 49 SI SSERVi Table of Contents Dilation (Graylevel) Filter Discrete Convolution Discrete Cotrelation Discrete Cosine Transform Discrete Fourier Transform Dithering Erosion (Binary) Filter Erosion (Graylevel} Filter Rip Fourier Transform Properties Gamma Noise Gaussian Filters Gaussian Noise Geometric Mean Filter Gradient Masks Graphic Incerchange Format (GIF) + Graphics Algorithms Graylevel Graylevel Histogram HSI Color Model Hadamard Transform Harmonic Mean Filter Hartley Transform High Pass Spatial Fitters Histogram Equalization Histogram Specification + Histogram Techniques Hit-Miss (Binary) Filter Homomorphic Filter 61 63 65 67 70 75 76 B 80 81 82 84 85 87 89 90 93 94 95 oF 99 102 105 199 110 112 115 116 119Table of Contents vii Hough Transform 122 Huffman Coding 125 + fmage Fundamentals 129 Inverse Filter 1x) Joint Photographic Experts Group (IPEG) 133 Laplacian Filter 133 Least Mean Squares Filter 136 Line Detector 137 Low Pass Spatial Filters 140 MacPaint File Format 141 Mask 144 Maximum Axis 145 Maximum Filter 147 Median Filter 149 + Mensuration 151 Midpoint Filter 152 Minimum Axis 154 Minimum Filter 155 Moments 157 Morphing 160 + Morphological Filters 162 Mohi-Graylevel Thresholding 163 Negative Exponential Noise 165 * Noise 167 + Nonlinear Filters 168 Nonlinear Transformations 169 Opening (Binary) Filter 171 Opening (Graylevel) Fitter 173 Optimum Thresholding 175Vili Table of Contents Outline (Binary) Filter PC Paintbrush (PCX) Perimeter Pixel Point Detector Pseudocvior Pseudocolor Display Quantization RGB Color Model Range Filter Rayleigh Noise Robert's Filter Rotate Ron Length Encoding (RLE) Salt and Pepper Noise Sampling Sealing + Segmentation Skeleton (Binary) Fikter Slant Transform Sobel Filter + Spatial Filters Spatial Frequency + Spatial Frequency Filters Spatial Masks + Storage Formats Tagged Interchange File Format (TIF) Thickening (Binary) Filter Thinning (Binary) Filter 178 180 183 184 185 ar, 1X) 193 192 194 196 198 199 202 205 208 212 218 220 221 222 223 224 225 229 234Table of Contents Thresholding Top-Hat Filter + Transforms True-Color Display Uniform Noise Walsh Transform Warping Weighted Mean Pilzer Weighted Median Filter Wiener Filter Wiener Filter (Parametric) YIQ Color Model Yp Mean Filter Zooming Appendix A Image.h File Appendix B Example Program Appendix C TIF Tags List Bibliography Glossary Subject index ix 239 al 245 252 254 261 BRR 270 271 279 281 282 297Preface xi This book is the culmination of an idea that was not ours uniquely, but one that has probably occurred to anyone who has spent long hours at the terminal writing Image Processing programs and routines. The programmer is often found surrounded by a library of texts, loaded with formulas and mathematically described algorithms. It is then necessary to engage in the arduous task of writing computer code from the mathematical description. In addition, if one dees not have an adequate background in signal processing, the difficulty of this task is further compounded. Programmers often spend time flipping from book to book, index to index, as they search fer a useful mask or filter technique. In this book we have tried to bring within a single, desktop friendly cover--in a simple, easy to use format—the most popular of the vast ensemble of image processing algorithms. We chose che C programming language as our algorithm description formar because of ir's popularity, portability and widespread use in image processing systems. Also, the C language is compact and easy to learn, and a plethora of texts and tutorials exist in the literature for boch the novice and seasoned programmer. Appendix B consains the complete code for a simple image processing program that may be used as a template for implementing any of the code in this handbook. It is our hope that all programmers (in the end, are we not all programmers?) of Image Processing systems will find this book usefui--both as a reference guide and for the implicit tutorial nature of our descriptions. Students should find our simple and direct discussion effective as a supplement to their classroom text or tutorial guides. Seasoned users will find the format convenient to keep close 10 the terminal for quick reference and easy access 19 definitions and algorithms, Harley R. Myler Arthor R. WeeksHow to Use This Book 1 TOPICAL LAYOUT This book is intended to serve as a teady reference and imaging system user and developer companion. Algorithms and definitions covered in this book are arranged alphabetically and typically occupy one or two pages. Each algorithm follows a similar format that is illustrated by the miniaturized sample page graphic io the left, below. Pizel CLASS: image Furdanverints Dascpiprtont: A puntos Pie Blemeees. 8 8 044 vast panne ln nine, ‘The omc er af tel ‘eden ie Supa dency das erp i, Sebpasnal Sea bt Lined nay S9 Cw mamory of the me Wieser 4 MUAY Uiikgh Tek rem Below share 4 casings MIE tee ra eae BAY Da warril por EXAMPLE: Each topic is identified by name in bold print in che header region of each page with the page number. The elass of algorithms to which the topic belongs is identified next. There are fifteen classes, which are: « Adaptive Filters + Coding and Compression + Color + Graphics + Histogrem Operations » Image Fundamentals + Mensuration + Motphological Filters + Noise + Nonlinear Filters + Segmentation « Spatial Filters + Spatial Frequency Filters + Storage Formats + Transforms PROG RAI EXAMPLE: fe Bhi x 1? Pantie -/ ‘shar TReRWA SIAL LUT fe Wid eV bd oboed Mi aage «t anae sasamnant i (7411 TA! Fe M6 5 The Eyoantag zou pixel Zips nterntoed br Fleet dibeestien| of Chea Read reo F4] | SLT AME ALEO Gund aarn, Same ihe Each class is described on a separate page and appears in alphabetic order with the topics. Classes are distinguished from topics by a bullet (+) in front of the class name in the page header. A description of the algorithm or the definition of the topic follows the class designation, This is followed by an example of the algorithm or description discussed. Examples tmay be graphic illustrations or inages. The algorithm in C is then given, if appropriate to the topic, followed by suggested follow-on topics under the SEE ALSO heading.2 How to Use This Book FINDING FOPICS Well-known topics, such as threshelding, may be looked up tapidly by thumbing through she headers at the top of each page. If the class of an algorithm is known, you may go directly to the class description, which will list all the topics that fall under thar class. The Table of Contents in the frone of che book lists all of the topics and class descriptions as they appear in alphabetical order with their page numbers. The Class Groupings Index, following this discussion, lists the classes alphabetically and their topics along with page numbers, The Subject Index, at the end of the book, lists subject areas and page mumbers. ALGORITHM PROGRAMMING The C programming language algonthms given in this book have been written in as simple a fashion as possible, with as litde dependence on sophisticated programming technique as possible. In the words of Einstein, “make everything as simple as possible, but not simpler.” It is a telatively easy maiier for the experienced programmer to make the C routines presented more efficient and elegant, however, it is not a simple matter for an inexperienced programmer to read and understand, i.c., simplify, a program that has been written in a complex and obtuse way with the goal of computational efficiency. Throughout the book an image data structure has been used that allows flexibility while maintaining clariry. Image tepresentation within a computer is best expressed as a rectangular array; however, when dealing with more than one pixel data type across a set of algorithms, a C data structure is the best solution for maintaining consistency throughout the algorithm ensemble. The data structure used here contains the rectangular size of the image, the cype of data each pixel in the image represents, and a pointer to the image data self,How to Use This Book 3 Another consideration in the use of a data structure. are the differences that exist between hardware platforms. Images, by nature, are large data objects. The structure that we use assumes that the user has allocated a data space large enough to accommodate the image being processed within any restrictive boundaries that may exist on the hardware being used. The example graphic below shows our daca structure and its relationship to an image. Struct Image{ Columns int Rows; int Cols; unsigned char *Data; unsigned char Type; Ve The *Dara pointer is a contiguous sequence of bytes whose type is described by the Type variable. The following four types of image data pixels are used: =4 Unsigned Character (BASIC) Type =2 Integer (UINT) Type =4 Float (REAL) Type =8 Complex (CMPLX) For example, a rectangular image with Rows = 64 and Columns = 256 and a Type = 0 will require that 16,384 (256 x 64) bytes of storage be allocated for the “Data pointer. If Type = 2 and if an integer on the host machine requires 4 bytes, then four times the previons amount will have to be. allocated. One must also apply special data type modifiers when programming MS-DOS™ systems. The far and Auge keywords must precede large data space pointers. Proper use of these modifiers is discussed further in Appendix B. A complex image is a special case. Here, each datum represents a real and imaginary component using a float data type. If the host computer requires 4 bytes to represent a float value, then each datum in the *Data buffer will consist of 8 bytes: the first 4 bytes for the real part of the datum, and4 How to Use This Book the second 4 bytes for the imaginary part, Consider the diagram below as representing an image, where each pixel is represented by a square and the total number of pixels is unspecified. The pixels have been lettered to facilitate the discussion, The first row of the image consists of the set of pixels {A, B, ..., C}, the second row is (D, E, .... F}, and the last row is {G, H, .... 1]. We can easily generalize this to sets of columns, as well. Using our image data structure and assuming pixels of Type 0, the arrangement of data for this image in memory would be as follows, where each pixel requires one byte of storage: Data AE) SPE) FS) Data will point to a series of bytes that represent the pixel values in column-row order. If Type = 2 and an integer in our machine requires two bytes of storage (often this is the size of an int} then the intemal representation will be: Data [A]4feTa)-+-[e[c}o]p jeje)... So pixel 'A' requires 2 bytes of storage, and so on. If the image is complex (Type = 8} and a float in the machine requires 4 bytes, the memory map will be:Hew to Use This Book 5 Data [ALAA ay ala] 4] Aye] 2 [8] 8 je) 8 7B |B]: + real imaginary real imaginary A A B B A complicity in programming arises in the accessing of the data using cartesian coordinates. The data structure allows for easy Tepresentation of a large number of pixel data formats and conforms readily 10 the idea of memory as a contiguous sequence of bytes; however, access of the data by coordinate is tricky. We must convert from the cartesian coordinate pair into a single value so that we may index into the Linear image data array. Consider the diagram below. We can visualize the pixels of the small 4 x 3 image co the left , with pixel values 4 through EL, in terms of their row-column coordinates. The programming description could be specified as the array, Img. Pixel G would then be Img[1][2]. Alternatively, we can specify an image structure as described earlier, with the ¢ statement struct Image Img. To access pixel G using our structure, we have to index into the data vector one row and three columns. The data in either case is stored as a sequence of bytes. The Data pointer of the image structure will be initialized to point to the start of his sequence, as shown in the graphic to the right, after the space has been allocated. Using the siracture format, the value of pixel G would be: *(Img.Data + (R * Img.Cols} + C} Where R is the row number (indexed from zero) and C is the column, For this example, *(img.Data + 1*4 + 2) will yield6 How to Use This Book the value, G. There are other ways to formulate access schemes into linear arrays thar have owo-dimensional data mapped to them, and some of these are illustrated in the various program sections of the book. A C program header file listing for the data structure is given in Appendix A, and a complete image processing program listing is described in Appendix B. The routines presented in this book assume differemt image pixel data formats and the user must observe some caution when incorporating the algorithms ia theic own programs. The most popular pixel size is, of course, the unsigned character. For this reason, it was chosen as the base pointer data type in our image structure. Image transforms such as the Fourier and filters that operate in the spatial frequency domain use compfex pixels. Other processes use float pixels, We have indicated the type of pixe] data required by the algorithm in the description, but in the interest of compactness and simplicity have put no error detection mechanisms in the code. It is assumed that the user has allocated the correct data space for the algorithm called. To simplify this concept of variable data widths for image pixels, we have provided an example of a mottiple data format image /O routine at the end of Appendix B. Finally, a Glossary has been included with a large range of image processing terms defined as a convenience to the teader,Class Groupings Index 7 This index is for reference by class membership. The classes and page numbers are: + Adapiive Filters + Coding and Compression + Color Image Processing ¢ Graphics Algorithms + Histogram Techniques « Image Fundamentals + Mensuration + Morphotogical Filters + Noise +¢ Nonlinear Filters + Segmentation + Spatial Filters + Spatial Frequency Filters » Storage Formats + Transforms Adapuve DW-MTM Filter 13 Adaptive MMSE Filter 1? Chain Codes 32 Huffman Coding 125 Run Length Encoding 2008 Class Groupings Index + Color Image Processing Class C.LE. Chromaticity Diagrara 29 Color Saturation Correction 49 Color Tint Correction Sl HSI Coter Mode? a? Pseudocolor 187 Pseudocolor Display 190 RGB Color Model . 192 True-Color Display 245 YIQ Color Model 264 » Graphics Algorithms Class Dithering 74 Flip 30 Morphing 160 Rotate 199 Warping 252 Zooming 268 * Histogram Techniques Class Brightness Correction 27 Contrast Correction 57 Graylevel Histogram 95 Histogram Equalization 110 Histogram Specification 11z Nonlinear Transformations 169Class Groupings Index 9 + Image Fundamentals Class Discrete Convolution 63 Discrete Correlation 65 Graylevel 94 Mask 144 Pixel 184 Quantization. 191 Sampling 204 Scaling 205 Spatial Frequency 221 + Mensuration Class Area 23 Centroid 30 Circularity 35 Clustering 44 Compaciness 33 Maximum Axis 145 Minimum Axis 154 Moments 157 Perimeter 183 Closing (Binary) Filter 40 Closing (Graylevel) Filter 42 Dilation (Binary) Filter 59 Dilation (Graylevel) Fiiter 61 Erosion (Binary) Filter 76 Erosion (Graylevel) Filter wr Hit -Miss (Bioary} Filter 11610 Class Groupings Index + Morphological Filters Class (continued) Opening (Binary) Filter V1 Opening (Graylevei} Fitter 173 Outline (Binary) Filter 178 Skeleton (Binary) Filter 208 Thickening (Binary) Fitter 229 Thinning (Binary) Filter 234 Top-Hat (Graylevei) Filter 241 + Naise Class Gamma Noise $2 Gaussian Noise 35 Negative Exponential Noise 163 Rayleigh Noise 196 Salt and Pepper Noise 202 Uniform Noise 246 + Nontinear Fitters Class Alpha-Trimmed Mean Filter 20 Contra-Harmonic Filter 54 Seomewic Mean Filter 87 Hamneonic Mean Filter 102 Maximum Filter 147 Median Filter 149 Midpoint Filter 152 Minimum Filter 155 Range Filter 194 Weighted Median Filter 257 Yp Mean Filter 265Class Groupings Index 11 « Segmentation Class Line Detector 137 Multi-Graylevel Thresholding 163 Optimum Thresholding 175 Poim Detector 185 Thresholding 239 + Spatial Filters Class Arithmetic Mean Filter 25 Gaussian Filters 84 Gradient Mask 89 High Pass Spatial Filters 109 Laplacian Filter 135 Low Pass Spatial Filters 140 Robert's Filter 198 Sobel Filter 218 Spatial Masks 233 Weighted Mean Filter 254 + Spatial Frequency Filters Class Circularly Symmetric Filter 37 Homomorphic Filter 119 Inverse Filter 130 Least Mean Squares Filter 136 Wiener Filter 260 Wiener Filter (Parametric) 26112 Class Groupings Index + Storage Formats Class Graphics Interchange Format (GIF) 90 Joint Photographic Experts Group (JPEG) 133 MacPaint File Format (MAC) 141 PC Paintbrush (PCX} 180 Tagged interchange File Format (TIF) 225 Discrete Cosine Transform 6? Discrete Fourier Transform 70 Fourier Transform Properties 81 Hadamard Transform 9 Hartley Transform 105 Hough Transform 122 Stance Transform 212 Waish Transform 248Adaptive DW-MTM Filter 13 CLASS: Adaptive Filters DESCRIPTION: The adaptive double window modified trimmed mean (DW- MTM) filter overcomes the difficullies of using the MMSE filter in the presence of impulsive noise by using the median estimator (9 estimate the local mean. A new loval mean is then computed using oaly pixels within a small graylevel Tange about the median. This effectively removes outers in ihe calculation of the mean estimate, bence improving the averall performance of the mean filter in the presence of outliers such as salt and pepper noise. The adaptive DW-MTM filter algeritam is described as follows. Given a pixel located at x. y within the image, a median fiker (MED{g(x, y)}) is contputed within an n x n local region surrounding the location x, y. The median value computed from this filter is used 30 estimate the mean value of the n Rows x IMAGE->Cols pixel image stored in the structure IMAGE. The program also assumes that the standard deviation of the noise and the threshoid parameter K is passed to the program upon execuuon. The program first computes the median in a 3 x 3 local window. This median value is then used to determine which points are excluded in the 5 x 5 mean filter calculation. Upon completion of the program, the fitered image is stored in the structure IMAGE1. DWHTM_filter{struct Image *IMAGE, struct image *IMAGE!. fleat NSTB, float K} { int x, ¥, xl, yl, med(9]; int median; int gray. i. j, temp; dong int tokai, sum, R: R=IMAGE-=Cols; for(Y= 2; YCols-2; M++} {Adaptive DW-MTM Filter 15 ; vie=l; yl+s} for (xls-1l; xl<=1; x1l++) { med{tetal]= * CIMAGE->Data+ S4+x1+(long! (Ysy1)*R)? totalstotal+]; } fordijel; je-8; 44+) { temp = medi]; i=j-l; while(ie=0 && med[i] »temp) { med(it+lj= medlil; ded-1, } med [i+i)= temp; médian=med{4]; sum=0; total=0; for{yl=-2; yl<«=2; yl++) for(tul=-2; xl<=2; x1++} { gray=: * (IMAGE->Data +KXexl+ (long) (Y¥tyl) "Ble if tgray>=(median-K*NSTD} } if (gray<={median+K*NSTD}} t sum=aum+gray + total = total +1; t } *{IMAGEL--Datard+ (long) T*R) =funsigned char) (ffloat)sum/(floatitotall; t I |“ SSBE ALSO: Adaptive MMSE Filter16 * Adaptive Filters Adaptive Filters DESCRIPTION: Depending un the type of noise that is present within an image, the type of filter that is chosen to remove the noise can radically affect the important details thal must remain udaffected by che filtering operation, For example, Gaussian type aoise is better filtered using a mean filver while Sait and Pepper type noise is better filtered using a median filter. The disadvansage of using a mean filter over a median filter is that the mean filter removes high spatial frequencies which blurs the edges within an image. An ideal filter to use on an image is a filler that adaptively changes its [iltcring characieristics depending on the image content within a local window, For example, if the image content within a local window contains only edges, then a median filler would be used 1o preserve the edge details. If, wilhin a local window, a uniform background is detected, then the filler would change ils charactlerislics 10 perform a mean filter within this local window. Filless that dynamically change their fillering characteristics as they are scanned through the image are known as adaptive filters. CLASS MEMBERSHIP: DW-MTM Filter MMSE Filter SEE ALSO: Noise. Nonlinear and Spatial FiltersAdaptive MMSE Filter i7 CLASS: Adaptive Filters DESCRIPTION; The Minimum Mean-Square Error filter (MMSE) makes use of the knowledge of the local variance to determine if a mean filter is to be applied te the local region of an image. This filer works best for additive type short tail noise. The ouiput of ihe MMSE finer for a pixel located al the coordinates x, y in an image g(x, y) is 2 on 1 ¥F= UT) 80 7+ 9 K. A, 1s? where Pe is the vanance of the neise, and 3 is ihe local variance About ahe pixel located at x, y. The parameter K is the output from a jocal mean fier which is usually a 3 x 3 or 5 X S$ mean filer. In background tegions the local variance is approximately equal to the noise vafiance reducing the adaptive MMSE to approximaiely a local mean filtor. Near edges, the local variance will be much larger than the noise variance and the adaptive MMSE filter reduces to the original unfiltered image. EXAMPLE: tal {oy (a) The original addidve Gaussian noise corrupted image {variance = 200) and (8) the 5 x 5 MMSE filtered image.18 Adaptive MMSE Filter ALGORITHM: The program assumes thal the original image is a 256 graylevel x IMAGE->Rows x IMAGE->Cols pixel image stored in the structure IMAGE. The program also assumes ahat the variance of the noise is Known and is passed te the program upon execulion. The program first computes the local variance over a N x N window from the fecal first and secood moments, Next, the program computes the MMSE filter oulpmt using the N x N mean computed in the first step, Upon completion of the program, the fitered image is slored in the structure IMAGE1. MMSE _filber(struct Image *IMAGE, Struct Image *IMAGE], float NVAR} { ant x, ¥, «li, yl, N, gi long int rotal, sum, suml, R; float FSECOND, FVAR, FMEAN; R=IMAGE->Cols; N=5; for{¥= Nr2; YRows-N/2: Y++} for (ReN/2;X sum l= total =0; for(yl=-N/2; yle«=N/2; yl+4) { for l[Ml=-N/2; xl<=N/2; x1l+4} { sumesum + * (IMAGE--Data + Kexl+(long}) (¥+yi)*R}; sunlssuml + tlong)*(iMAGE-=sData + HX4x1+¢long) (Y¥+y1)*R)* {long} * ({IMAGE->Data + Mexl+ (long) (Y+y1)*R) total = total +1: I PSECOND=(float}suml/ (float) total; PHEAN = (float) sum: {floatitotal; FVARSFSECOND - FMEAN*FMEAN; if(FVAR ==0.0)Adaptive MMSE Filter 19 g=f{int) (PMEAN+.5); else g=tint) ((1-NVAR/FVAR}* *1 TMAGE-sbata+k+(loang)¥*R) + NVAR/FVAR* FHEAN+ .5) 7 ifig>255) ge255; lftged) g=0; * {IMAGEL->Data+X+ (longl ¥*R) =9; } } SEE ALSO: Adaptive DW-MTM Filter20 Alpha-Trimmed Mean Filter CLASS: Nonlinear Filters DESCRIPTION: The a(pha-trimmed mean filter is based on urder statistics and varies between a median and a mean iilter. It is used when an image contains both short and kong tailed types of noise. For example, many images vontain both Gaussian and salt and pepper type noise. Ty define the alpha-trimmed. mean filter, all pixels sucrounding the pixel at the coordinate x. ¥ in the image A which are specified by an input mask A(i} are order from minimum to maximuni graylevel value Ay SAp SAqSAgS- SAN] SAN, The alpha-trimmed mean filter is hen given as N-P AlphaMean (A) =—— © Aj, “Arise where N - 2P is the total number of pixels included in the average calculation. EXAMPLE: {a} (5) {a} A Gaussian and salt and pepper noise corrupted image and (&) the alpha-trimmed mean filtered image using a 5 x 5 square mask and P = 3,Alpha-Trimmed Mean Filter 21 ALGORITHM: The program assumes that the onginal image is a 256 graylevel x IMAGE->Rows x IMAGE->Cols pixel image stored in the structure (MAGE. The program computes the alpha-irimmed mean filter over a sel of pixels contained within a square N x N region of the image centered at the pixel X,Y. The size of the filtering operation is determined by the variable N and should be set to an odd number and be less than 12. The parameter P passed to the program determines how many ot the endpoint pixels are climinated from the urdered data. For example, if P = 1, the pixels with the minimum and maximum graylevel values are not included in the avérage calculation. For the special cases of P=(and P=(N * Nj/2 - 1/2 the alpha-trimmed mean filter becomes the mean and median filters respectively. Upon completion of the program, ihe alpha-wimmed mean filtered image is stored in the structure IMAGE}, AlphaMéanistruct Image *IMAGE, int P, struct Image *IMAGE1) { int #. ¥, I, J, SUM, 2; int N, AR[121], A; Nel; for {¥sN.'2; YRows-N-Z; Y++! for(XeN/2; KeIMAGE->Cols-Nys2: K+4i { 2-0; fortd=-N/2; d<=sNv2; J++! for (I=-N/2; T<=N/2; I++) { AR[2] =* {IMAGE-+Data+# +I+tlong: (¥4+7) *IMAGE- =Cols}; Geri for(J=1; J<«=NtN-1;74++) 4 A BRIT: T= whi lets 0 && AR{I] +A) R(T+1]=AR [1]; =I-1; ~yeo es22 Alpha-Trimmed Mean Filter AR(I+1]=A; } SUM=0;2Z=0: for(JsP; TxeN*N-1-P; J++} } { SUM = 50M + ARIJ]; Lee; I * (IMAGE]-»DataeX+(longiy *IMAGE->Cols} = ijunsigned char} { (float) sums (float}2Z 4.5); SEE ALSO: Geometric, Yp. Harmonic and Arithmetic Mean Filters, Median. and other Nonlinear Filters.Area 23 CLASS: Mensuration DESCRIPTION: Area algorithms measure the number of pixels contained within an object in an image. The boundary of the object tmust be defined, or the region wherein the object lies as well as the pixel value to be counted as part of the object must be known. In the image below, a region is delineated and a pair of objects lie within the region. The area algorithm given Tetums an area calculation for an object given a specific gtaylevel. When the physical size of a pixel is known, the actual area in size units can be calculated. The example below shows the original image on the left. To the right is a thresholded, binarized, and eroded image that has a region drawn around a white object that represents the “end" key. The white pixels in this region, when counted, represent the area of the object. EXAMPLE: ALGORITHM: The algorithm accepts an irnage structure pointer, In, that is assumed te be an 8-bit unsigned character picture. The coordinates of the region containing the object, {xl y1,x2,y2), and the object graylevel, GbjVal, are aiso passed. The area is computed by counting the number of pixels in the area at the graylevel value. This vatue is returned. Objects that contain spaces or multiple objects within che region Containing pixels at the object graylevel value are also24 Area counted, Other methods of area calcuiation can avoid these problems, but require the contour pixels of the object. If the contout is known, the area can be calculated tine by tine by counting candidate pixels between contour coordinates. The macro pix is used for simplified access 10 the pixels of the image by coordinates, édefine pix(iIm,x,.yi 4 *(Im->Bata + (x}*Im->Cols + (¥}} /* Compute and return area for objects */ int area(struct Image *In,int «l,int y1, int “2,int ¥2,unsigned char Obj¥al) { ieng i.j,rows: int area value = 0; fortimal; is\-x2; ++4) for(jsyl; jRows x IMAGE->Cols pixel image stored in the structure IMAGE. The program then performs aN x N arithmetic mean filter on the umage. The size of the fitering operalion is determined by the variable N and should be set tw an odd number. Upun completion af the program the filtered image is stored in the suucture IMAGE]. Meantstruct Image *IMAGE, struct Image *IMAGEL) { int 4, ¥; int I, d; int M, SUM; Ns?; for{Y=sN/2; YDatatk +I+tlong) (y+) *IMAGE->Cols) ; } } *{IMAGE]->Data+X+!long?Y *IMAGE->Cols) = SUM/N/N: } SEE ALSO: Median, Minimum, Maximum, and othe: Nonlinear FiltersBrightness Correction 27 CLASS: Histogram Operation DESCRIPTION: Brightness correction is used to enhance the visual appearance of an image. Brightness modification of an image is defined as $j =g) + brightness . where sj is the ith graylevel value of the brightness enhanced image, and gj is the ith graylevei value of the original image. EXAMPLE: {a} () (a) The original image and (#) the brightness corrected image. ALGORITHM: The program assumes that the original image is a 256 graylevel x IMAGE->Rows « IMAGE->Cols pixel image stored in the structure IMAGE. The program adds the value passed tc the program in the brightness parameter to each pixel in the image. Upon completion of the program, the brighiness corrected image is slored in the structure IMAGE]. Brightness (struct Image *IMAGE, struct Image “IMAGE1, int brightness) {2 Brightness Correction int xX, ¥, I: for[y=0; YCols; X++) 1 = *(IMAGE->Data+k+ (long) ¥* IMAGE->Cols} + brightness; if{I> 255) 1=255; ATi1<9) Ist; * (IMAGE1- »Data+K+!long)¥* IMAGE->Cols} = 1; } SEE ALSCr Graylevel Histogram, Histogram Specification, Contrast Correction, and Nonlinear Graylevel Transforma- lionsC.LE. Chromaticity Diagram 29 CLASS: Color Image Processing DESCRIPTION: C.LE. Chromaticity Diagram Wavelength in GREEN+ Nanometérs + WHITE n=» ZMmMDG RED Axis 1 =Red + Green + Blue or Blue = 1- Red - Green SEE ALSO. RGB and YIQ Color Models, Hue and Saturation Correction, and Pseudocolor30 Centroid CLASS: Mensuration DESCRIPTION; Cenreid refers to center and is a measure of the geographic center of an object. The centroid is expressed in coordinates and is calculated using the following expression: N N 1 1 Xe = % Ye =ahY i=l iz] where Xc and Y¢ are the centroid coordinates, X and ¥ ate the coordinates of the ith object pixel, and A is the object area. The centroid may also be calculated as the first central moment. The example graphic illustrates the computation of the centroid of an object using the formula given above. EXAMPLE: Xc = 39/13 = 3 O123456709 Ye = 32/13 =4 - eatogrezia ALGORITHM: The algorithm accepts an 8-bit unsigned character image, In and the coordinates of the region containing the object (<1,y1,x2,y2) and the object graylevel, ObjVal. The object area is computed by calling the area function (See Area). Row and column coordinates are then accumulated for the object. These accumulations are divided by the computed area for the object and che resulting centroid coordinates are Teturmed as the x and y members of the coord data suructure. Integer arithmetic assures that a coordinate value will be returned. If area_value computes to zero, negative coordinates are returned.Centroid /* coordinates structure */ struct coord { fleat x,¥r be /* Compute and return centroid */ void *centroid(struct Image *In,int x1, aint yl,int “2,int y2, unsigned char ObjVal, struct coord *coords} long i,4; int area_value, Xcent=0, Ycent=0; area_value = area(In, xl, yl, x2, y2,O0bjVal}: if (area_value == 0) { coords->x = -l; coords->y = -1; return; t for (i=xl; i¢=_2; ++i) for{j=yl; j<=y2: ++)){ if (pix (In, i, 4)==ObjVal}h{ Keent += ji Yoent += i; } coords->x = Xcent/area_value + 1; coords->y = Yeent/area value + 1; return; SEE ALSO: Area, Maximum Axis, Minimum Axis, Moments 3132 Chain Code CLASS: Coding and Compression DESCRIPTION: Chain Coding is used to define object boundaries for the purpose of measurement and identification. A chain code follows the boundary of an object and indicates the change in direction according to a 4- or 8-direction pattern as shown here: 1 3 The code is dependent on where on the boundary the code sampling starts, but may be normalized using a oumber of techniques. EXAMPLE: “ ™~ Contour Mappings ~ Contour \7 \7 A contour is shown with the directional mappings given to the right. The $-direction chain code for this contour would be: FTITTTLOOT I.Chain Code 33 ALGORITHM: The routine is passed an image structure pointer of unsigned character data containing the open ended contour, In, the x and ¥ coordinates of the start ef the comtour to be coded, and a threshold value, thresh, used to determine membership in the contour from a pixel neighborhood. The routine is also supplied a character suring pointer, code, which must be initialized to che maximurn expected length of the code. The routine returns the length of the code in the int poiater length. Caution coust be exercised in that the code string will comain zeroes that indicate the right horizontal direction and Not siring termination. The routine uses a macro, idx, to simplify coordinate access into the image array. #* compute 8-direction chain code given start point of contour */ ¥define idx{m,n)/ *{in->Datat (mi +( in} *In->Cols)) chain_8(struct Image *In, int x, int y, char *code,int *Length, unsigned char thresh) *length = 1; foris:){ idx(x,y} = 0; if (idx (x+1, ¥} >thresh)} [{ ik => */ *l(code++) = O7 tH; +e{tlength}y continue: ) iftidx(xti, y-1) >thresh)} [ im fowy *{ooder+) = 1; -oys teas ++ (*length}; continue;34 Chain Code LE (idxix,y-l}othreshy { Ae yp re x(code+4+) = 2; --y; ++ *length): continue: i Lf (idx(x-1,y-l)sthresh} { ee yee *(code++} = 3; TTY 2TcHe ++ (*length) : continues } if (idx (#-1, y) >thresh} [ fk <- #f s{eodert+) = 4; oxi ++ (*length}: continue; } Lf (idx (x-1, ytl)>thresh) { fe f o*f *(codet++}) = 5; tty pees +4 (length) ; continue: } if(idxtx, y+1) >thresh) { ft | wf *(code++} = 6; tty? ++ (*length) : continue; } dE (ide (xt+1, y4+1}>thresh) | fe NF *{codet+) = 7; ttysttxy ++(*length); continue; I return; t SEE ALSO: MensurationCircularity 35 CLASS: Mensuration DESCRIPTION: Circularity is the measure of an abject’s closeness to circular shape. It can be expressed as the variance of the boundary pixel's distance to the centroid, If the object is perfecdy circular, this variance will be zero, The limit on circularity is one-half the major axis. ALGORITHM: The circularity algorithm requires both she boundary pixels of the object in question as well as the centroid. Fhe boundary may be determined by submacting the single pixel erosion (See Erosion) of the object from the object. The algorithm accepts an $-bit unsigned character image, In and the coordinates of the region containing the object (x1, yl, x2, y2) boundary and the object grayievel, ObjVal. The object centroid is computed by calling the centroid function (See Centroid): then a euclidian distance measure is applied to each boundary pixel with the centroid. These values are accumulated and the variance calculated. f* coordinates structure */ struct coord { float x,¥: he /* Compute and return circularity */ double circularity (struct Image *In, long x1, leng yi, long x2,long y2, unsigned char GbjVal) { struct coord Ocenter; Jong i,j, 1Tows,points=07 double mean=0.6, temp, stdev=0.0, variance; cantroid{In, ¥1,yl,#2,y2, Ob jVal, cOcenter}; /* compute mean */ for{i=xl; lData; N= filt->Rows; for¢iwO,1sN-1; L */ ‘* upper left quadrant */ *(E+CA*N+ GP) © (E+ (G*N+]} +1) = Huy: /* lower right quadrant */ *CF+ (Liam) = *(F+(2*N4+m}41) = Huy; /* upper right quadrant */ e(f+ti%N+m)) = *¢f£4+(itN+m) +1) = Buv: /* lower left quadrant */ S(ft4L N+ j)) = *(L+(L*N+5}4+1) = Buy: } SEE ALSO: Fourier Transform40 Closing (Binary) Filter CLASS: Morphological Filters DESCRIPTION: Morphological closing of @ binary object is defined as the dilation of that object followed by ihe erasion of the dilated abject. The closing filter operation will reduce small inward bumps and small holes within the object, Closet, B= (A BBOB EXAMPLES: Cd * OBJECT A OBJECT A * * {a} (b} {a) The original binary image of object A and (6) the closed image of object A with a circular structuring function &. Procasing | Procauang (a) The original image and (#} the closed image.Closing (Binary) Filter 41 ALGORITHM: The program assumes that the original binary image is an IMAGE->Rows x IMAGE->Cols piael image with a background graylevel value of 0 and an object graylevel value of 255 (object) and is stored in the structure IMAGE. The N x N structuring function is stored in array MASK{]f]. Upon completion of the program, the closed image is stored in the structure FILTER. The binary erosion and dilation functions used by the afgorithnt can be found under binary erosion and dilation respectively. fdefine MN 5 Closeistruct Image *IMAGE, int MASK[} [MN], struct Image "FILTER} t aint X, ¥; DilationtIMAGE. MASK, FILTER! ; for(¥=0; YCols; K++! ( *IMAGE->Data+ X + tlongi¥ *IMAGE-> Cols: = *{PILTER->Darat+ X + : long; ¥ * IMAGE->Cols); } ErosiontIMAGE, MASK, FILTER); FILTER-»Rows =[MAGE->Rows; PILTER-»Cols=IMAGE- Cols; } SEE ALSO: Binary Dilation, Erosion, and Opening Filters42 Closing (Graylevel) Filter CLASS: Morphological Filters DESCRIPTION: Morphological closing of an image is defined as the gtaylevel dilation of the image follawed by the graylevel erosion of the dilated image. The closing filter operation will reduce small negative oriented graylevel regions and hegative graylevel noise regions generated from salt and pepper noise. Close(d, B= (ASRS EXAMPLES: (@) {b) fa) The original image and (&) the closed image using an all zero 7 x 7 mask.Closing (Graylevel) Filter 43 ALGORITHM: The program assumes that the original image is a 256 graylevel x IMAGE->Rows x IMAGE->Cols pixel image stored in the structure IMAGE. The N x N structuring fanciion is stored in the array MASK[][]. Upon completion of the program, the closed image is stored in the structure FILTER. The graylevel erosion and dilation functions used by the algorithm can be found under graylevel erosion and dilation respectively. #define N 5 cCloseGray Istruct Image "IMAGE, int MASK[] (NJ, struct Image *FILTER) { ion xX. YF DilationGray (IMAGE, MASK, FILTER) ; fort¥=0; YRows; Y++] [ for{X=0; XDatat+x+ (long) ¥* IMAGE-=>Cols)= *(FILTER-> Data+X+{ long) ¥* IMAGE=>Cols); a } ErosionGray (IMAGE, MASK, FILTER); FILTER ~>RoOws = IMAGE->Rows 7 FILTER ~-=Cols=IMAGE->Cols; } SEE ALSO: Graylevel Dilation, Erosion, Top-Hat, and Opening Filters44 Clustering CLASS: Mensuration DESCRIPTION: Clustering is the process of counting and labeling of objects within an image. Clusters consist of pixel groupings that are related to oné another by a predetermined measure. This measure can be defined as a distance between clusters or a similarity measure such as pixel value, or it may be a complex set of identifiers and constraints that define membership in a cluster. Fhe cluster may ultimately be mapped into a binary image as a unique object. One of the simplest approaches is to specify a size in pixels tha: the clusters should be. If we assume a binary image, the clustering algorithm processes each pixel and when one is found that is nonzero, it becomes part of the first cluster and is marked. The next nonzero pixel found is tested to see if the distance between it and the previous pixel is less than or equal to the desired cluster pixel size. If it is, itis marked as a member of the first ctuster and the search coatinues. If it is not, it becomes the first member of the second cluster and is marked accordingly. This process continues until all pixels have been evaluated. The exampie shows a graphic of a set of similar-sized objects that have been labelled by grayscale value, EXAMPLE:Clustering 45 ALGORITHM: The following variables, with their descriptions, are used in the cluster analysis fuactioa: i,jek Loop variables. Tr Cluster count variable. In Unsigned character image data structure pointer to Binary image, In=>Rows Rows in Image In->Cols Columns in Image ¢lustx[] Temporary storage for x— coordinate of last pixel in cluster []. eclusty[] See clustx, new clust Plag for new cluster. edist Euclidean distance. ¢lust_size Size of clusters to be found, At the completion of the function, the Im data array contains pixel values that reflect their membership in the clusters found that meet the criteria of being clust_size apart. Clusters of pixels that are larger than clust_size size will be. partitioned into pieces as they are found in the image. The image is scanned from the upper left corner origin ia column-row order. cluster({struct Image *In, int clust_size} { int mn, 4, 4, k, new _clust; float edist, clustx[256),clusty[256]> f* initialize cluster count variable n */ n=; #* double—loop through each pixel of binary image */ for{ie0; i < In->Rows; ++i) for(jed; 4 < In->Cols; ++i) if(n==0) {46 Clustering /* only do for lst cluster */ if(pix(In,i,j) !=0){ nm=1; /* Ist cluster found */ /* store K coord, */ elust™[1] = i; **® store y coord. */ olusty({1) = 4: ?#* mazk pixel as cluster 1 */ pintIn,i, 4} = 17 } } /* test for membership in all known clusters */ else if{pin(In,i,j} != 0) ¢* marker for new cluster */ new_clust = 0; /* compute Euclidean distance */ for(k = 1; k <- ni ++kif edist = sqrt ((i-clueex[ki}* (i-clustx[k]}+ {j-clusty[k]}* (j-clusty(k]}): /* test against cluster size desired */ iffedist <= ¢lust_size} [ /* set pixel te cluster number */ pix(In-i,j) = funsigned char)k?: Aew_¢lust = 1; kant: } } /* add a new cluster */ if{new_elust == 0 && n <= 255) { nen +t 1; clustu(n] = i: clusty[n] = j: pix({In,i,j) = (unsigned charin:+ Coding and Compression 47 Coding and Compression Class DESCRIPTION: Coding is a broad term that can refer to data encryption, Tepresentation, or compression schemes. Although images are often encrypted for security purposes, these methods are beyond the scope of this book. Only one representation coding is covered in this book, chain codes. Chain codes play an important role in image understanding and computer vision studies and allow the user to describe contours and edges in a compact and useful way. Compression is an important aspect of image processing simply because the data sets are so large. Clearly, an RGB, 512 x 512 pixel image at 24 bits per pixel requires 786,432 bytes of storage. Two such pictures exceed the capacity of most magnetic floppy diskettes, If the image data is fairly correlated, the compression ratio possible can exceed 50 10 1. Of course, the basic drawback to compression is the time it takes to compress and the time it takes to decompress. However, special hardware is available that can maximize the speed of the compress/decompress cycle to real-time image display rates. Although image Storage Formats constitute a separate class in themselves, formats often incorporate compression schemes as a part of their standard. CLASS MEMBERSHIP: Chain Codes Huffman Coding Run Length Encoding (RLE) SEE ALSO: Storage Formats48 * Color Image Processing Color Image Processing DESCRIPTION: Anew and very powerful field of image processing is color image processing. Color image processing can be separated inte two general areas: (1) the enhancement of color images, and {2} the pseudocoloring of graylevel images. The en- hancemeat of a color image involves either the enhancement of the three calor (Red, Bluc, and Green) components of an image or the enhancement of thé tint or saturation of an image. In contrast, pseudocoloring makes use of color to represent the graylevel yalues of an image, In this way, certain attributes of an image can be easily visualized. The human eye is much more sensitive to color variations than to graylevel variations. For example, arteries within an x-ray image can be hilighted in red tw improve the accuracy in locating Blocked arieries. CLASS MEMBERSHIP: C.LE, Chromaticity Diagram Color Saturation Correction Color Tint Correction HSI Color Model Pseudoculor Pseuducolor Display RGB Cojur Model True-color Display YIQ Color ModelColor Saturation Correction 49 CLASS: Color Image Processing DESCRIPTION: This algorithm adjusts the color saturation of an RGB color image in a similar manner as the color control on a color television. The algorithm is based upon the R ~ ¥, G ~ Y, and B — Y color model used in many of today's color twlevisions. The cotor components in terms of R, G, and B are R—¥ =6.70R - 0.596 - 0.11B B— ¥ =-0.30R - 0,59G + 0,89B G—Y =-0.30R + 0.41G - 0.2 1B, where the (luminance component Y is defined as Y¥ = 0.30R + 0.596 + 0.11B ALGORITHM: The program assumes thai the original image is a 24-bit color image of spatial size of RED->Rows =x RED->Cois pixels. The program assumes that the image is a RGB image aml is stored in three structures: RED, GREEN, and BLUE. Passed co the program in che variable saturation is an integer humber that is greater chan zero lhat represents in percent the amount wf color tp be added back tu the illuminance or black and white tmage. The first thing the program does is compute the R -'¥,G - Y. and B - ¥ components. Next, it scales these values using ihe variable saturation and adds them back io the iuminance component to generate the new RGB values. Upon completion of the program, the saturated corrected image is returned in the three structures RED, GREEN, and BLUE. Saturation struct Image *RED, struct Image *ELUE, struct Image *GREEN, int saturation) f int X, R¥, BY, G¥, RY1, BYl, G¥1; int ¥, Rk. B. G Bl. Rl. Gl; long inr K;50 Color Saturation Correction for,¥=0: YCols; Rl= *(RED->Dara+Ki; Gl= *(GREEN->Data+Kl; Bl= *(BLUE->Data+Kl i RY1l= {fant} {7o* (int}R1- $9* (int }Gl-11*(intiBi)/100); BYl=((int} (-30* (int) Rl- 59* (int }G1+89* {int)B1)/100}; GYl=(int) ({-30% Lint} Ri+ 41*(inct}Gl-ll*(int}Bl)/ido); Yeidine) (30*(int}R1l+ 59* (int) Gi+11* (int)B1)/100); BY={BYl*saturation) /100; RY=(R¥1*saturation} /100; GY=(GY1*saturat ion) /100; Re R¥ 4¥:G=e GY + ¥iB = BY + ¥: iftR <@) if(B <0} ifte <0) Z G£(R2255} Re255; LE(B 2255) Be2So; 1ftG 2255) Ge256; * (GREEN->Data+K) = lunsigned char}; * (BLUE->Data+K} = (unsigned char) Bs; * (RED- »Data+k) = (unsigned char! RB: } } SEE ALSO: RGB, HSI and ¥1Q Color Models, €.1.E. Color Chart, Pseudocotor, and Color Tint CorrectionColor Tint Correction 51 CLASS: Color Image Processing DESCRIPTION: This algorithm adjusts the lint of an RGB color image in a similar manner as the tint control on a color television. The algorithm is based upon the R- ¥, G— Y, and B — ¥ color mode! used in many of today’s color televisions. The color components in terms of R, G, and B are R-Y¥ =0,70R —- 0.596 - GIB B-Y =~0.30R ~0.59G + 0.89B G-¥ =~0.30R + 0.41G - 0.2 1B, where the illuminance component Y is defined as Y =0.30R + 0.59G +0.11B ALGORITHM: The program assumes thal the original image is a 24-bit color image of spatial size of RED->Rows * RED->Cols pixels. The program assumes that the image is a RGB image and is stored in three equal sized structures: RED, GREEN, and BLUE. Passed to the program in the variable tint is an integer number in degrees between -180 and +180 that sepresenis the angle in which ihe tint of the image is to be fotated. Positive angles give the color image more of a green tint while negative angles give the color image more of ared unt. The first thing the program does is compute the R- ¥ and B - Y components. Neat. it rotates these two yeclors by the angle given in the variable tint. Finally, the program then recompnies the new RGB values placing them m the three color image arrays. Upon completion of the program, the tint corrected image is reiucned in the three Structures RED, GREEN, and BLUE. Bue{struct Image *RED, struct Image *BLUE, struct image *GREEW, int = tinti { int X, RY, BY, GY, RYi, BY1; int ¥. R, B, G. 5, ¢, Ri, GI, Bl; long int K;52 Color Tint Correction float theta; thetas(3.14159*(floatirint) $160.0; Ce286*cos ( (double) thetal+ 5=256*sin( (double) theta); foriY=0; YeRED=->Rows; Y44} { fortX=sQ; EKeRED->Cols; K++} { K=X+llong) Y*RED->Cols; Rls *{RED->Data+K}: Gi= * (GREEN->Data+k); Bl= *{BLUE->Data+Kl; RYl= (ints (70* (inti RI- 59* {int )G1l-11* (int) B1)/100); BYl=({int) (-304 (int}R1- 59* (int) G1+89* (ine }E1}/100); Ye(iint) ¢ 30* (int}R1+ 59* (int )G1l+11* (int}B1)/100); BYsiC*BYl - S*RYL) (2567 RY¥s{S*BY1l + C*RYL) /256; G¥=t{ine) (-51l*Ry - 19"BY) /1001; R= RY +¥; if(R <0) if(B <0} ifig <0) G=0; if(R2255) Re255; if(B 2255) Bs255; if(G +255) Ge255; * {GREEN- sDabatkK) = vunsigned char)G; *(BLUE-=Data+Ki= lunsigned char}B; *(RED->Data+K)= (unsigned char!R; } GY + %;B = BY + ¥; 3 SEE ALSO: RGB, HSI and YIQ Color Models, C.LE. Colar Chart. Pseudocolor, and Color Saturation CorrectionCompactness 53 CLASS: Mensuration DESCRIPTION: Compaciness is an object distribution measure similar te circularity, but is defined as the perimeter squared divided by the area. The compactness value will be minimal for a circular object, but it is not insensitive to noise in the perimeter measure. Objects with noisy boundaries will have a large perimeter measure with respect to similar sized objects with smooth boundaries, hence different compactness values in spite of circular shape. No algorithm is given as compaciness can be readily calculated from the area of the object and the area of the object boundary (perimeter). Examples are shown, however, of compactness measurements on various objects. EXAMPLE: Each of the objects shown has the same area, 32 pixels. Compactness for each object is as follows: A: 7.03 B:8 C:15.485 D: 15.15 Objects A and EB are similarly shaped and have very close compactness values. Objects C and D, however, are very different in shape in spite of identical compactness values. SEE ALSO: Area, Circularity, Perimeter54 Contra-Harmonic Mean Filter CLASS: Nonlinear Filters DESCRIPTION: The contra-harmonic mean filler is member of a set of nonlinéar mean filters which are better at removing Gaussian type noise and preserving edge features than the: arithmetic mean filler. The contra-harmonic mean filter is very good at Temoving positive culliers for negative values of P and negative outliers for positive values of P. If all the pixets included in the calculation of the contra-harmonic mean are zero, the output of the conwa-harmonic fiker will also be zero. The definition of the contra-harmonic mean filter is EA +i, y+ pPel G.j 3 mask, P = 2),Contra-Harmonic Mean Filter 55 ALGORITHM: The program assumes that the original image is a 256 gtaylevel x IMAGE->Rows x TMAGE->Cols pixel image stored in the structure IMAGE. The program computes the contra-harmonic filter over a set of pixels contained within a square N x N region of ihe image cenlered at the pixel X, Y. The program expects the order of the filter io be passed to it upon execution. The size of the filtering operation is determined by ihe variable N and should be sei to an add number and should be less than 12, Upon completion of the program, the filtered image is slored in the siructure. IMAGE1. ContraharmonicMéanistruct Image *IMAGE. int P, struct Image *IMAGEL1: { imt x, Y, I. J, a int N; imt AR(121), aA; float SUM; float SUM1; N=5; for{Y=N/2; ye IMAGE-2Rows-N/Z: Yre! for(X=N/2; X<=IMAGE->Cols-N/2; ¥++) f a=0; for(J=-N/2; JeaN/2; J++: for tIa-N/2; IesNs2y; I++} i AR[Z] =* |IMAGE->Data+x 4Teilong! (Y+d) *IMAGE->Calsi; Zee; } Bed; SUM=0.0; SUM150,0; for{J=0; Je=sN*N-1;7++} { if(AR[J]==0 && FeO! Bel; else { SUM=SUM+powl idouble)AR(J], idouble} (P+lil;56 Contra-Harmonic Mean Filter SUML=3UM1+powl (double! AR[J], (double) PF; } } if (Z==1} *{IMAGE]-=Data+xX +(long)¥ *IMAGE->Cols) =0; else C if (SQM1==0.0) A=0.07 else Asdint) (SUM/SUM1); ifi{R >255) As 255: * (IMAGE]-+Data+X+ {longi ¥ *IMAGE->Cols} = tunsigned char)A; SEE ALSO: Geometric, Yp, Harmonic and Arithmetic mean Filters, Median, and other Nonlinear Filters