Is there an efficient and fast way to process pixel neighborhoods without loops?

Hi all, I am looking for an efficient and fast way to process local window without loops. I want to apply a nonlinear operation on the local neighborhood with an exponential function. The output of this operation is not a scalar rather than being a weight for ever pixel in the neighborhoods. Later, I use these weights to calculate the final value of the central pixel. Here is a code snippet of what I did:

for i=1:r 
  for j=1:c
      x=i:i+wsize-1;  y=j:j+wsize-1    % wsize=3 (the window size)
      if(max(x)>r || (max(y)>c))
          break
      end

        Win=im1(x,y);  Win=Win(:);

        C=im2(x,y);    C=C(:);      % current window of size wsize
        avgC=mean(C);

        Gij=exp(-(C-(avgC(i,j)))./C); 

        Wij=(G_ij)./sum(G_ij);

        im3(i,j)=sum(Win.*Wij);        
    end
end

Note: im1 and im2 are different while the desired output will be in im3.

Answer 1

[Image Analyst]

There is a function called nlfilter() that can apply your custom function to a moving neighborhood. I attach a demo.

[Ahmed Elazab]

Dear Image Analyst, Thanks so much for reply. I tried to use nlfilter function before but it is much slower than using the code above.

[Image Analyst]

Just get rid of the loops and vectorize it. Use conv2() or imfilter().

Answer 2

[Tony Richardson]

I've written up a brief description (see the url below) of a fairly straight-forward method that I've found to be much faster that nlfilter. It does not use loops (vectorized) and still easily allows for non-linear processing of neighborhoods (for example - median processing). (nlfilter allows non-linear processing, but conv2 and imfilter do not.)

Fast Pixel Neighborhood Processing (pdf)