Matlab Programs for Image Processing

1.The utility functions

The functions provided are :

- to load image called filename having the dimension xdim and ydim, and which is stored as datatype precision
rawread(filename, xdim, ydim, precision)

rawwrite(filename, matrix, precision)

- to save matrix in file called

filename with datatype precision . For internal storage purposes Matlab treats an image as a matrix. For example to load an image into a matrix try
image=rawread(ctscan.raw, 256, 256);

You can choose to treat rawread, rawwrite as blac boxes. The source code is given at the end of this document.

Rawread.m ;--------------------------------------%Reads a raw intesity image file, written with a given precision (by default uint! ). % img " R#$R%#&(fname,ncols,nrows,precision); % function raw"rawread(fname, cols, rows, prec) if nargin""', precision" uint! ; elseif nargin""(, precision"prec; else, error( $rong number of arguments) *yntax+ ,rawread(fname,cols,rows), ); return; end fid " fopen(fname, r ); raw " fread(fid,-cols,rows.,precision) ; fclose(fid);

;---------------------------;rawwrite
function count"rawwrite(fname, array, prec) if nargin""/, precision" uint! ; elseif nargin""', precision"prec;

else, error( $rong number of arguments) *yntax+ ,rawwrite(fname,array), ); return; end fid " fopen(fname, w ); count " fwrite(fid,array,precision) ; fclose(fid);

;-----------------------------------------------------------1) Exercise 1 Rea an is!lay an image Rea " is!.m ;---------------------------------------------filename# $ctscan.raw%; image#rawrea &filename' ()*' ()*); Imagesc&image);

50

100

150

200

250 50 100 150 200 250

Result+ ;-------------------------------The image a!!ear coloure an image+ Ty!e in the comman win ow+ axis(square); colormap(gray); The result now+ istorte . In or er to is!lay a gray un istorte

50

100

150

200

250 50 100 150 200 250

Exercise () Plot a line !rofile through an image line"!rofile.m

%Reads a raw image file and display a line profile through its middle % it uses the rawread function filename" ctscan.raw ; % the filename to be read image"rawread(filename, /01, /01); figure(2); 3magesc(image); % display axis( s4uare ); colormap( gray ); line"2/!; figure(/); plot( image(line,2+/01)); image(line,2+/01)

50

100

150

200

250 50 100 150 200 250

250

200

150

100

50

50

100

150

200

250

300

;------------------------------------------------------------exercise ,. Transform a gray image into a binary image To ma e any graylevel image into a a binary image !",# image$ at say a threshold of half the maximum brightness value try :
maximum"max(max(img))5/.6; binimg"(img7maximum); The max operator returns a vector

if applied to a matrix and a number if applied to a vector - so to get the maximum graylevel value in a matrix it has to be applied twice. The same is true for the median, std and min functions. - o"binary.m
filename" ctscan.raw ; % the name of the file to be read image"rawread(filename, /01, /01); figure(2); 3magesc(image); % display original image axis( s4uare ); colormap( gray ); % binari8e maximum"max(max(image)); binimg"(image7maximum5/.6); figure(/); 3magesc(binimg); % display the binary image axis( s4uare ); colormap( gray ); % save filenameout" ctscanbin.raw ;

rawwrite(filenameout, binimg); % it writes not as bit but with the same precision as the original image

50

100

150

200

250 50 100 150 200 250

Find the way to store the image as bits% in order to show the smaller si&e of the file'' ( -------------------.on/olutions + Matlab has facilities for both #) and *) convolutions as implemented in the functions conv and conv/. For example to get a simple edge detection mas convolved with img try :
result"conv/(img,--2 -2 -2;-2 ! -2;-2 -2 -2., same );

(------------------------------------------------------------------0.1ourier Transform 1ourier Transforms+ There are also facilities for #) and *) forward and inverse Fourier Transforms - try the commands fft, fft/, ifft and ifft/. The command fftshift centers the Fourier spectrum origin to the center of the image. +ote that the results of these are complex and to display them one needs to decide on a reasonable representation such as for example the modulus. ,s an example to display the centered Fourier transform for the image in img type :
fouimg"fft/(img); fouimg"fftshift(fouimg); imagesc(abs(fouimg));

Result with shifting

50

50

100

100

150

150

200

200

250 50 100 150 200 250

250 50 100 150 200 250

fouimg2"/9 log26 (2:abs(fouimg))

Result without shifting

50

100

150

200

250 50 100 150 200 250

The phase can be displayed by using imagesc(angle(fouimg))

50

100

150

200

250 50 100 150 200 250

The phase

and the real and imaginary parts using the real and imag operators.

50

50

100

100

150

150

200

200

250 50 100 150 200 250

250 50 100 150 200 250

-eal part

imaginary part

.rogram fft"image.m
filename" ctscan.raw ; % the filename to be read img"rawread(filename, /01, /01); figure(2); 3magesc(img); % display axis( s4uare ); colormap( gray ); fouimg"fft/(img); fouimg"fftshift(fouimg);

fouimg2"/9 log26 (2: abs(fouimg));% compresion of the dinamic range figure(/); 3magesc(abs(fouimg2)); axis( s4uare ); colormap( gray ); figure('); 3magesc((fouimg2)); axis( s4uare ); colormap( gray );

% display the modulus without compression range

% display the modulus with compression range

figure((); imagesc(angle(fouimg)) ; axis( s4uare ); colormap( gray );

% display the phase angle

/t is worth remembering that multiplication in the Fourier domain in the case of an e0uation of the form 1!u,v$2F!u,v$3!u,v$ would actually be implemented by elementwise multiplication of F and 3 using the .9 operator and not normal matrix multiplication. ( (------------------------------------------------------------------------------------------------Exercise ) The low pass filter: do4lowpass.m
filename" ctscan.raw ; % the filename to be read image"rawread(filename, /01, /01); 3magesc(image); % display axis( s4uare ); colormap( gray ); s8e"si8e(image); rows " s8e(2); cols " s8e(/); %cutoff is the cutoff fre4uency of the filter 6 - 6.0. % n is the order of the filter, the higher n is the sharper % the transition is. (n must be an integer 7" 2). cutoff"6.0; n"2.6; % ; and < matrices with ranges normalised to :5- 6.0 x " (ones(rows,2) 9 -2+cols. - (fix(cols5/):2))5cols; y " (-2+rows. 9 ones(2,cols) - (fix(rows5/):2))5rows;

radius " s4rt(x.=/ : y.=/); radius relative to centre.

% # matrix with every pixel "

filt"2 .5 (2.6 : (radius .5 cutoff).=(/9n)) ; f " fftshift(filt); % >he filter figure(/); surf(filt); im " normalise(image); % Rescale values 6-2 (and cast % to ?double if needed). @@>3m " fft/(im); % >aAe @@> of image him " real(ifft/(@@>3m.9f)); % #pply the filter, invert figure('); imagesc(him); axis( s4uare ); colormap( gray );

cutoff2".5

cutoff2".# the filter

50

100

150

200

250 50 100 150 200 250

50

100

150

200

250 50 100 150 200 250

Exercise * 3igh pass filter( do4highpass.m /t is obtained as:

filt" 2.6 - filtlow; where fillow is the lowpass filter described above

50

100

150

200

250 50 100 150 200 250

The result of the filtering Exercise 2 3igh4oost filtering

a high-boost Butterworth filter. f " (2-25boost)9highpassfilter : 25boost;

1 0.9 0.8 0.7 0.6 0.5 0.4 0.3 0.2 0.1 0 0 50 100 150 200 250 300

difference between the highpass ( red) and highboost filter for boost"26;

Cighboost leaves a bit of low frec4uency information

50

50

100

100

150

150

200

200

250 50 100 150 200 250

250 50 100 150 200 250

6oost2*

6oost2#"

Exercise 5 3omomor!hic filtering+ o"homor!hic.m The homomorphic filtering%

im " normalise(im); cast % to ?double @@>log3m " fft/(log(im:.62)); offset % Rescale values 6-2 (and if needed). % >aAe @@> of log (with % to avoid log of 6).

him " exp(real(ifft/(@@>log3m.9h))); % #pply the high boost filter, invert % fft, and invert the log.

50

100

150

200

250 50 100 150 200 250

6ray le/el transformations Exercise 7+ 8inear .ontrast Enhancement: do4lcontrast.m Typically, high contrast images are visually more appealing.
0.45 0.4 0.35 0.3 0.25 0.2 0.15 0.1 0.05 0 0 50 100 150 200 250 300

Transfer functions for alfa"6.20

Red-7 alfa Blue-7 /9 alfa Dreen-7'9alfa

CT Scan 50 100 150 200 250 50 100 150 200 250 2* alfa 50 100 150 200 250 50 100 150 200 250 50 100 150 200 250 50 100 150 200 250

alfa

50 100 150 200 250 3* alfa

50 100 150 200 250

Exercise 11+ 6ama .orrection : do4adgamma.m The drawbac of the linear contrast enhancemnt is that it leads to saturation. This may be avoided by employing a non-linear contrast ad7ustment scheme such as gamma correction
newim " newim.=(25g);

8here g 9 # increases contrast and g:# reduces contrast

1 0.9 0.8 0.7

1 0.9 0.8 0.7 0.6

0.6 0.5 0.5 0.4 0.4 0.3 0.3 0.2 0.1 0 50 100 150 200 250 300 0.2 0.1 0 0 50 100 150 200 250 300

g25

12".;

50

50

100

100

150

150

200

200

250 50 100 150 200 250

250 50 100 150 200 250

g25

12".;

%Ealues of g in the range 6-2 enhance contrast of bright % regions, values 7 2 enhance contrast in darA % regions.

Exercise 1(+ 9 :ust .ontrast : do4ad7constrast.m <ne drawbac of the gamma correction is that the gray values are mapped in an asymmetric fashion with respect to the middle level gray.This may be alleviated by employing a sigmoidal non-linearity .
newim " 2.5(2 : exp(gain9(cutoff-newim))); % #pply *igmoid function

1 0.9 0.8 0.7

1 0.9 0.8 0.7

0.6 0.6 0.5 0.5 0.4 0.4 0.3 0.2 0.1 0 0 50 100 150 200 250 300 0.3 0.2 0.1 0 50 100 150 200 250 300

cutoff"6.0;

cutoff"6./;

50

50

100

100

150

150

200

200

250 50 100 150 200 250

250 50 100 150 200 250

cutoff"6.0;

cutoff"6./;

(-----------------------------------------------------.rograms ta en from http:==robotics.eecs.ber eley.edu=>mayi=imgproc=index.html

/. Gaussian_m.m

6aussian 1ilter ;emos

These demos show the basic effects of !*)$ 1aussian filter: smoothing the image and wiping off the noise. 1enerally spea ing, for a noise-affected image, smoothing it by 1aussian function is the first thing to do before any other further processings, such as edge detection. The effectiveness of the 1aussian function is different for different choosing the standard deviation sigma of the 1aussian filter. You can see this from the following demos.

<moothing =onnoisy Image

lena.gif filtere with sigma # , filtere with sigma # 1

=oise .ancelling
noisy lena filtere with sigma # , filtere with sigma #1

!The noise is generated by M,T?,6 function ".5@randn!;#*$$

/t uses the function d2gauss.m

lena 50 100 150 200 250 50 100 150 200 250 smooth 50 100 150 200 250 50 100 150 200 250 50 100 150 200 250 50 100 150 200 250 50 100 150 200 250 50 100 150 200 250 noise cancel noisy lena

The parametes of the function could be changed:

Farameters of the Daussian filter+ n2"26;sigma2"';n/"26;sigma/"';theta"6; % >he amplitude of the noise+ noise"6.2;

,. .anny E ge etector

.anny E ge ;etector ;emos

There are some results of applying Aanny edge detector to real image The thresholding parameter alfa is fix as ".#. The si&e of the filters is fixed as #"@#". These images are all gray images though they might show a little strange in your browser. To see clearer, 7ust clic these images and you will find the difference especially from the Bresult imagesB, that is, the titled Bafter thinningB ones. The safe way to see the correct

display of these images is to grab these images and show them by BxvB or BM,T?,6B. 8hile, you are encouraged to use the given M,T?,6 codes and get these images in M,T?,6 by yourself. Try to change the parameters to get more sense about how these parameters affect the edge detection.

The results of choosing the stan ar the e ge etectors as ,.

lena.gif /ertical e ges

e/iation sigma of
hori>ontal e ges

norm of the gra ient

after threshol ing

after thinning

The results of choosing the stan ar the e ge etectors as 1.

lena.gif /ertical e ges

e/iation sigma of
hori>ontal e ges

norm of the gra ient

after threshol ing

after thinning

/t uses the function d*dgauss.m /t wor s

Original Image 50 100 150 200 250 50 100 Iy 50 100 150 200 250 50 100 150 200 After Thresholding 50 100 150 200 250 50 100 150 200 250 250 50 100 150 200 250 50 150 200 250 50 100 150 200 250 50

Ix

100 150 200 Norm of Gradient

250

100 150 200 After Thinning

250

50 100 150 200 250 50 100 150 200 250

d*dgauss can be used as )erivative of a 1aussian ?in s of materials for image processing and Matlab #.http:==amath.colorado.edu=courses=CD*"=*"""Epr=?abs=8or sheets=Matlab4tutorial=mat labimpr.html /ntroduction to image processing in Matlab # by Kristian Sandberg, Department of Applied Mathematics, University of Colorado at Boulder

exercise 1, Fdge )etectors G morphology in Matlab% .rewit operators Aode: do4prewitt.m

FrwGim " edge(image, prewitt );

50

100

150

200

250 50 100 150 200 250

Fxercise #C Eobel operators Aode: do4sobel.m

sobGim " edge(image, sobel );

50

100

150

200

250 50 100 150 200 250

)ifference between the sobel and prewittH Aode: do4sobel4prewitt.m

50

100

150

200

250 50 100 150 200 250

Fxercise #I. -oberts <perators Aode: do4-oberts

50

100

150

200

250 50 100 150 200 250

3istogram based segmentation http:==arl.stanford.edu=pro7ects=helicopters=matlab.html

Fxercise #D. )o histogram Aode: do4histogram.m

imhist(im2);

50

100

150

200

250 50 100 150 200 250

5000

4000

3000

2000

1000

0 0 50 100 150 200 250

Fxercise #J. Fixed thresholding Aode: do4thresholding.m 68 2 /M*68!/,?FKF?$ converts the intensity image / to blac and white. 6ased on the histogram Thresh 2J". The output binary image 68 has values of " !blac $ for all pixels in the input image with luminance less than ?FKF? and # !white$ for all other pixels.

50

100

150

200

250 50 100 150 200 250

Thresh2#*"( L only the bones are ept.

50

100

150

200

250 50 100 150 200 250

Fxercise #M Morphology ;ilation+ code: do4dilation.m

&ifferent Aernels H'"ones('); H0"ones(0); HI"ones(I); HJ"ones(J);

50

100

150

200

250 50 100 150 200 250

thresh2#*"

50

100

150

200

250 50 100 150 200 250

dilated with N5

50

100

150

200

250 50 100 150 200 250

dilated with N; etc..

Fxercise *" Frosion

50

100

150

200

250 50 100 150 200 250

Froded with N5

50

100

150

200

250 50 100 150 200 250

eroded with N; Fxercise *": do4opening.m Morphology could be also applied with bwmorph function that will use a N5 ernel(

50

100

150

200

250 50 100 150 200 250

op"bwmorph(B$, open );

Fxercise *# Alosing Aode: do4closing.m

50

100

150

200

250 50 100 150 200 250

closing Fxercise *5 Thining )o4thining.m

50

100

150

200

250 50 100 150 200 250

Fx Fxercise *5 Thic ening )o4thic ening.m

50

100

150

200

250 50 100 150 200 250