Storing values in an array

I want to calculate the distance to each pixel from the center pixel and sort the highest 4 values. I used the following code for that. But it gives the error "Subscript indices must either be real positive integers or logicals". How to solve it?

 totalVal = 0; 

 for h = 1 : H
    for w = 1:W

         %value = pdist2([cen h],[cen w],'euclidean');
         value(h,w) = sqrt((cen-h)^2 + (cen-w)^2); 
          end
end

 f3Row = reshape(value, 1, []); 
 [sortedDist, sortedInds] = sort(f3Row,'descend');
 highestDist = (sortedDist(1:4));

for num = 1:n
    totalVal = totalVal+highestDist(num);
end

 f3 = totalVal; 
 fprintf('Feature 03 : ');
 disp(f3);

[Jan]

Please post the error message. It is much easier to fix an error than to guess, what the error is.

[NC]

Question was edited with the error message.

Answer 1

[Walter Roberson]

centy = H/2;
centx = W/2;
[Hg, Wg] = ndgrid(1:H, 1:W);
value = sqrt((Hg - centy).^2 + (Wg - centx).^2);
sortedDist = sort(value(:), 'descend');
f3 = sum(sortedDist(1:4));

However, you can predict this value without doing any of this. The value will always be 4 * sqrt(H/2^2 + W/2)^2

Perhaps you are only to consider pixels that meet some particular condition, like belonging to a non-square ROI? And perhaps the center point is not the center of the image but is instead the centroid of the non-square ROI ?

[NC]

Thank you very much. Actually I want to Fourier transform the image and then get the distances because the high frequencies are aggregated around the center. These distances are used as a feature as an input to a SVM.

[Walter Roberson]

The code you were showing deals with the euclidean distance of a pixel location from something, and does not have any relationship to pixel content or content of the fft() or fft2() of an image.

I guess you could do

fimg = fft2(YourImage);
fimgss = fimg(2:end,2:end);    %skip the zero-order components
absfimg = abs(fimgss);
meanabs = mean(absfimg(:));
[sortdiff, diffidx] = sort(abs( absfimg(:) - meanabs ), 'descend');
ssfar4 = diffidx(1:4);
f13frabs = sum(sortdiff(1:4));   %sum of the abs of the frequency differences
f13frorig = sum(fimgss(ssfar4));  %sum of the complex frequency differences

Even though I took care to skip the zero-order coefficients along the left and top boundary, the ssfar4 indices turn out to correspond to pixels at the upper left and lower right corners. This is because for visual images, the fft2 often has peaks at the edges (and at both edges, because of conjugate symmetry.)

[NC]

When I use this code I got the error

Subscript indices must either be real positive integers or logicals.
Error in completeFeatures (line 126) %completeFeatures is file name
f13frabs = sum(sortdiff(1:4))

What should be done here?

[Walter Roberson]

The only way that could have happened there is if you had created your own variable named sum . Do not name your own variable sum .

[NC]

Thank you. The problem was resolved by renaming existing sum variable.