Vectorization of For loop

Hi every one!

I am trying to vectorize the following for loops as this part of the code is the main bottleneck and taking hours together to get the output

can someone pls help me

Thanks in advance

[rowA colA]=size(blockA);
[rowB colB]=size(blockB);

blockA=double(blockA);
blockB=double(blockB);

m=1;
flag=0;
blockC=zeros([500 4]);

for i=1:rowA
    for j=1:rowB

   fi = fft2(blockA(i,1:bs*bs));
   fr = fft2(blockB(j,1:bs*bs));
  %    perform phase correlation (amplitude is normalized)
   fc = fi .* conj(fr);
   fcn = fc ./ abs(fc); 
   c1 = real(ifft2(fcn));

   % find the peak in the correlation plane
   [v index] = max(c1(:));

   if v>cut_off
       blockC(m,1)=blockB(j,bs*bs+1);
       blockC(m,2)=blockB(j,bs*bs+2);
       blockC(m,3)=v;
       blockC(m,4)=2;
       flag=1;    
       m=m+1;
    end

    end
      if flag==1
       blockC(m,1)=blockA(i,bs*bs+1);
       blockC(m,2)=blockA(i,bs*bs+2);
       blockC(m,3)=v;
       blockC(m,4)=1;
       m=m+1;
      flag=0;
      end

end

[Sean de Wolski]

fyi:

c1 = abs(ifft2(fcn));

[Sean de Wolski]

So are you trying to find the displacement field between two images by block processing them and then moving the pieces according to their vector? Please explain the goal.

(A large focus of my MS thesis is in phase correlation displacement fields this piques my interest)

Answer 1

[Walter Roberson]

Notice that

fi = fft2(blockA(i,1:bs*bs));

does not change as j changes, so there is no need for it to be within the range of the "j" for loop. You can calculate it right after the "i" for loop starts. This will save you quite a number of fft2's.

[Sean de Wolski]

bs doesn't change, calculate bs*bs once.