How to pick a sliding window of pixels arround a pixel ?

Hi, I used the following loop to select at each iteration a window of 5x5 pixels around a pixel of a given image 'im' in this case:

for ii=3:size(im,1)-2
    for jj=3:size(im,2)-2          
       win=im(ii-2:ii+2,jj-2:jj+2);
    end
end

How do i change it to enlarge the window to 32x32 ?

Answer 1

[Image Analyst]

Change the 2 to 15 and 16. Note that usually window sizes are an odd number so that the filter does not create a half pixel shift in the output image. Because your size is not odd, that's why you have to subtract 15 but add 16, or vice versa, instead of the same number for each direction.

And, why don't you use nlfilter(), like in my attached demo, which handles all of that for you?

[Ynne]

Big Thanks. I am using the function in this link https://www.mathworks.com/matlabcentral/fileexchange/46093-color-auto-correlogram to compute the autocorrelogram for each pixel of a given image according to a sliding window 32x32. here is the code

 for ii=16:size(im,1)-15
    for jj=16:size(im,2)-15     
       temp_im=im(ii-15:ii+15,jj-15:jj+15,1:3);         
       correlogram=color_auto_correlogram(temp_im, distance_vector);
    end
end
Thus i want to have for each pixel given his neigborhood a feature vector. The problem according to the time consumed by this code without getting any result. Is there any error ?

[Image Analyst]

You're just overwriting a single scalar variable, correlogram, every time instead of creating an image. I think you want correlogram(jj, ii) or something (not sure what color_auto_correlogram returns).

[Ynne]

I'am trying to insert the result into a matrix as following, but i have the same the problem

 feat=zeros(M,N,64);
 for ii=16:size(im,1)-15
    for jj=16:size(im,2)-15     
       temp_im=im(ii-15:ii+15,jj-15:jj+15,1:3);         
       correlogram=color_auto_correlogram(temp_im, distance_vector);
       for k=1:64
            feat(ii,jj,k)=correlogram(k);
        end
    end
end

[Image Analyst]

correlagram may not have 64 elements.

After you get it, set a breakpoint there and execute this line

>> whos correlogram

What does it say?

[Ynne]

Name              Size            Bytes  Class     Attributes

correlogram      64x1               512  double