image processing with parallel approach

s=load('matlab.mat','imagest1','imagest2','Train_Number');

imagest1=s.imagest1;
imagest2=s.imagest2;
Train_Number=s.Train_Number;

eyeimage=imread('C:\Users\PRIYAM\Desktop\project\project pictures and links\CASIA Iris Image Database (version 1.0)\002\1\009.bmp');

%imshow(eyeimage)
radpixels=50;
angpixels=1000;
[circleiris, circlepupil, imagewithnoise,] = segmentiris(eyeimage);
[polar_array, noise_array] = normaliseiris(imagewithnoise, circleiris(2),circleiris(1), circleiris(3),circlepupil(2), circlepupil(1), circlepupil(3),...
 eyeimage,radpixels,angpixels);
subplot(2,2,2)
imshow(polar_array)
%imshow(polar_array)
[icodeV,icodeH]=  irisfeature(polar_array);

for i=1:Train_Number
  HD=0;  
    for j=1:500

          HD=HD+((xor(icodeV(1,j),imagest2(i,j))+xor(icodeH(1,j),imagest1(i,j)))/2);

      end
      hd(i)=HD;

end

loc=0;
for i=1:Train_Number
    if (hd(i)<149.5)
     loc=i; 

      fprintf('image matched with image number %d \n with hamming distance %d',loc,hd(i));

      end

end
if (loc==0)
    disp('image not matched with any image');
end;toc

How to parallelize it? I have used parfor, but the exexcution time remains same. Please help me to make the code snippet parallel. What should be the alternative parfor code for it?

Answer 1

[Walter Roberson]

Vectorize, not parfor!

For example,

loc = find(hd < 149.5, 1, 'first');
if ~isempty(loc)
  fprintf('image matched etc');
else
  fprintf('image not matched');
end

[Edric Ellis]

I would say: vectorize first, and then you'll probably be in a better position to use PARFOR if you wish. Plus, of course, vectorization might get you intrinsic multithreading (depending on your operation), which definitely beats PARFOR on a single machine.