image enhancement using PSO algorithm

particle.m

clc;
 clear all;
 close all;
 %Reading image
 Img   = imread('w.jpg');
 I     = imresize(Img,[256 256]); 
 I     = rgb2gray(I);
 [m,n] = size(I);
 %PSO implementation
 P       = 5;        %swarm size
 iter    = 6;        %number of iterations
 c1      = 2;
 c2      = 2;        %usually c1=c2=2
 r1      = rand;
 r2      = rand;
 fit_val = [];       %matrix for storing fitness values
 P_best  = [];       %matrix for storing pbest values
 pbest   = 0;
 gbest   = [];
 %particle initialization
 for i = 1:P
 %updating particle position
       a(i) = (1.5).*rand(1,1);
       b(i) = (0.5).*rand(1,1);
       c(i) = rand(1,1);
       k(i) = 0.5+1.*rand(1,1);
       fprintf('particle  %d:: a=%f ; b=%f ; c=%f ;k=%f \n',i,a(i),b(i),c(i),k(i));
 %updating particle velocity
       v1(i)  =  0.15*a(i);
       v2(i)  =  0.5*b(i);
       v3(i)  =  0.15*c(i);
       v4(i)  =  0.5*k(i);
       f_v    =  []; 
       P_best =  [];
       pbest  =  [];
  end
  %iterations
   for it=1:iter
       IG=I;
       fprintf('\n');
       fprintf('--> ITERATION %d <--\n',it)   
       for i=1:P
            fprintf('\n.....PARTICLE (%d)....',i);
            figure
            imshow(IG);
            pause(1);
           IG              = uint8(trans_fcn(IG,a(i),b(i),c(i),k(i)));
           [fitness,out1]  = f_fcn(m,n,IG);
           f_v(it,i)       = fitness;
           pbest(i)        = max(f_v(:,i));
           P_best(it,i)    = pbest(i); 
           fprintf('FITNESS VALUE of (%d) iteration :\t\t\n',it); 
           disp(f_v()); 
           fprintf('p_best_values: \n'); 
           disp(P_best(it,:));
           gbest(it) = max(f_v(:)); %calculating gbest
          
           v1(i) = (1/it).*v1(i) + c1.*r1.*(gbest(it)-a(i)) + c2.*r2.*(pbest(i)-a(i));
           a(i)  = a(i) + v1(i);
%            a(i) = ((a(i)-0.1139724)/(7.7605725-0.113924))*1.5;      % 0 to 1.5
           fprintf('a = %d \n',a(i));
           
           v2(i)= (1/it).*v2(i) + c1.*r1.*(gbest(it)-b(i)) + c2.*r2.*(pbest(i)-b(i));
           b(i) = b(i) + v2(i);
%            b(i) = ((b(i)-(0.31926))/(0.9866753-(0.31926)))*0.5;    % 0 to 0.5                  
           fprintf('b = %d \n',b(i));

           v3(i) = (1/it).*v3(i) + c1.*r1.*(gbest(it)-c(i)) + c2.*r2.*(pbest(i)-c(i));
           c(i)  = c(i) + v3(i);
%            c(i)  = (c(i)- 0.1117058)/(2.78389-0.7249451);          % 0 to 1                    
           fprintf('c = %d \n',c(i));

           v4(i) = (1/it).*v4(i) + c1.*r1.*(gbest(it)-k(i)) + c2.*r2.*(pbest(i)-k(i));
           k(i)  = k(i) + v4(i);
%            k(i) = 0.5+(((k(i)-(0.193213))/(0.9783839-(0.193213)))* 1);          
           fprintf('k = %d',k(i));                                % 0.5 to 1.5
           fprintf('\n');
           
%    if (out1.n_edgels<500||out1.E<100000||out1.H<0.2123)
%                break;
%        end
       end
       fprintf('gbest of iteration %d :-> %f', it, gbest(it));
       fprintf('\n');
       fprintf('.......Iteration %d stops......', it)
end
   figure
       x= [1:iter];
       y= gbest ;
       plot(x,y)
%     figure
%     bar(gbest)
%     imshow(IG);
---------------------------------------------------------------------------------
trans_fcn.m
function IG = trans_fcn(IG,a,b,c,k)
 % function to transform image to enhanced version
 D   =  mean2(double(IG)); %global mean
 s   =  stdfilt(IG);       %local std dev
 SE  =  ones(3,3);
 SE  =  SE/sum(SE(:));
 lm  =  conv2(double(IG),double(SE),'same');
 lm  =  uint8(lm); %cast back to uint8, this will round values
 x   =  (s+b);
 w   =  (k*D);
 K   =  (w./x);
 IG  =  (K.*(double(IG)-(c*double(lm)))+(double(lm).^a));
------------------------------------------------------------------------------------
function [f,out] = f_fcn(m,n,IG)
n_edgels = 0;
T        = [m,n];  
E        = 0;

Is = imgradient(IG,'sobel'); % it will return binary image Is where the function finds edges it will return 1 else 0 

%Is = histeq(IG);
n_edgels = nnz(Is);    % number of edge pixels
T = uint8(Is).*IG;
E= sum(T(:));          % sum of edge intensities
H=entropy(IG);         % entropy of enhanced image
% fitness value  
f = log(log(double(E))).*(n_edgels/(m*n)).*exp(H);  % we can get the density of the sparse matrix by nnz(s)/prod(size(s));
out.E        = E;
out.n_edgels = n_edgels;
out.H        = H;
out.f        = f;
fprintf('................................\n');
fprintf('Sum of pixel intensities %d\n',E);
fprintf('Number of edge pixels %d\n',n_edgels);
fprintf('Entropy of image %d\n', H);
fprintf('----- fitness value is  %f ------ ',f);
fprintf('\n');
end

------------------------------------------------------------------------------------------

This is the code to enhance the digital images using particle swarm optimization algortithm, but the resultant images are not accurate. also there is some problem in the code. Any suggestions for some improvement?

[Walter Roberson]

Is there an error message?

[KALYAN ACHARJYA]

but the resultant images are not accurate

Elaborate? Be specific please. Also requested you to do format of the code by clicking on Insert code button.

Answer 1

[SHILPI GUPTA]

prticle.m


clc;
 clear all;
 close all;
 %Reading image
 Img   = imread('w.jpg');
 I     = imresize(Img,[256 256]); 
 I     = rgb2gray(I);
 [m,n] = size(I);
 %PSO implementation
 P       = 5;        %swarm size
 iter    = 6;        %number of iterations
 c1      = 2;
 c2      = 2;        %usually c1=c2=2
 r1      = rand;
 r2      = rand;
 fit_val = [];       %matrix for storing fitness values
 P_best  = [];       %matrix for storing pbest values
 pbest   = 0;
 gbest   = [];
 %particle initialization
 for i = 1:P
 %updating particle position
       a(i) = (1.5).*rand(1,1);
       b(i) = (0.5).*rand(1,1);
       c(i) = rand(1,1);
       k(i) = 0.5+1.*rand(1,1);
       fprintf('particle  %d:: a=%f ; b=%f ; c=%f ;k=%f \n',i,a(i),b(i),c(i),k(i));
 %updating particle velocity
       v1(i)  =  0.15*a(i);
       v2(i)  =  0.5*b(i);
       v3(i)  =  0.15*c(i);
       v4(i)  =  0.5*k(i);
       f_v    =  []; 
       P_best =  [];
       pbest  =  [];
  end
  %iterations
   for it=1:iter
       IG=I;
       fprintf('\n');
       fprintf('--> ITERATION %d <--\n',it)   
       for i=1:P
            fprintf('\n.....PARTICLE (%d)....',i);
            figure
            imshow(IG);
            pause(1);
           IG              = uint8(trans_fcn(IG,a(i),b(i),c(i),k(i)));
           [fitness,out1]  = f_fcn(m,n,IG);
           f_v(it,i)       = fitness;
           pbest(i)        = max(f_v(:,i));
           P_best(it,i)    = pbest(i); 
           fprintf('FITNESS VALUE of (%d) iteration :\t\t\n',it); 
           disp(f_v()); 
           fprintf('p_best_values: \n'); 
           disp(P_best(it,:));
           gbest(it) = max(f_v(:)); %calculating gbest
          
           v1(i) = (1/it).*v1(i) + c1.*r1.*(gbest(it)-a(i)) + c2.*r2.*(pbest(i)-a(i));
           a(i)  = a(i) + v1(i);
           fprintf('a = %d \n',a(i));
           
           v2(i)= (1/it).*v2(i) + c1.*r1.*(gbest(it)-b(i)) + c2.*r2.*(pbest(i)-b(i));
           b(i) = b(i) + v2(i);
           fprintf('b = %d \n',b(i));

           v3(i) = (1/it).*v3(i) + c1.*r1.*(gbest(it)-c(i)) + c2.*r2.*(pbest(i)-c(i));
           c(i)  = c(i) + v3(i);
           fprintf('c = %d \n',c(i));

           v4(i) = (1/it).*v4(i) + c1.*r1.*(gbest(it)-k(i)) + c2.*r2.*(pbest(i)-k(i));
           k(i)  = k(i) + v4(i);
           fprintf('k = %d',k(i));                              
           fprintf('\n');
     end
       fprintf('gbest of iteration %d :-> %f', it, gbest(it));
       fprintf('\n');
       fprintf('.......Iteration %d stops......', it)
end
   figure
       x= [1:iter];
       y= gbest ;
       plot(x,y)
.....................................................................
       
       f_fcn.m
       
function [f,out] = f_fcn(m,n,IG)
n_edgels = 0;
T        = [m,n];  
E        = 0;

Is = imgradient(IG,'sobel'); % it will return binary image Is where the function finds edges it will return 1 else 0 

%Is = histeq(IG);
n_edgels = nnz(Is);    % number of edge pixels
T = uint8(Is).*IG;
E= sum(T(:));          % sum of edge intensities
H=entropy(IG);         % entropy of enhanced image
% fitness value  
f = log(log(double(E))).*(n_edgels/(m*n)).*exp(H);  % we can get the density of the sparse matrix by nnz(s)/prod(size(s));
out.E        = E;
out.n_edgels = n_edgels;
out.H        = H;
out.f        = f;
fprintf('................................\n');
fprintf('Sum of pixel intensities %d\n',E);
fprintf('Number of edge pixels %d\n',n_edgels);
fprintf('Entropy of image %d\n', H);
fprintf('----- fitness value is  %f ------ ',f);
fprintf('\n');
end
.................................................................
    
 trans_fcn.m

 function IG = trans_fcn(IG,a,b,c,k)
 % function to transform image to enhanced version
 D   =  mean2(double(IG)); %global mean
 s   =  stdfilt(IG);       %local std dev
 SE  =  ones(3,3);
 SE  =  SE/sum(SE(:));
 lm  =  conv2(double(IG),double(SE),'same');
 lm  =  uint8(lm); %cast back to uint8, this will round values
 x   =  (s+b);
 w   =  (k*D);
 K   =  (w./x);
 IG  =  (K.*(double(IG)-(c*double(lm)))+(double(lm).^a));
 
 

 
       
       
       

[SHILPI GUPTA]

particle.m is the main program.

The required images are enhanced ones with sharpened edges. But the images that are coming dont show connected and sharpened edges. Moreover the enhancement of the image is not done. Do suggest me proper wayout for this . Thanks.

[Image Analyst]

Not down here - up there in your original question. Oh never mind, I'll do that. You just concentrate on attaching 'w.jpg' to your original post at the top, NOT down here.

[SHILPI GUPTA]

Did that.

I'm getting NAN values for pbest and a,b,c,k after 1 or 2 iterations. Any suggestions for the improvement?

Answer 2

[Manisha Pandey]

Hi Shilpi

Did you get the NAN value problem resolved? Help me please if any update regarding the same.

[mashal zafar]

is this code working

[Salem Mahdi]

I done kwon

[Alexandra Kate Mariano]

Is there a new code that is working?