0

As part of a larger image processing pipeline I have lots of RGB images where local regions should be smoothed with medfilt2. The regions are given by non-zero pixels of a gray mask with the same size and are less than 10% of the whole image area. Instead of smoothing the whole images by applying medfilt2 to each channel which is very time consuming I first extracted the regions with

outimage2 = uint8(zeros(h, w, 3));
[r,c] = find(outmask);
for i = 1:length(r); outimage2(r(i),c(i)) = outimage(r(i),c(i)); end;

and then apply medfilt2 to the channels of outimage2. This is 2-3 times faster than applying medfilt2 to the whole outimage because averaging over outimage2 with its many zeros seems faster than averaging over arbitrary numbers. But it could be much faster if medfilt2 could be restricted to the regions without doing anything on the rest of the image. I have tried to generalize the solution from image-processing-algorithms-on-the-roi with 

outimage21 = outimage(:,:,1); 
outimage21 = medfilt2(outimage21(outmask > 0), [medfilt_val(1) medfilt_val(2)]);

for the first channel but I get an "Index exceeds matrix dimensions"-error.

Günter Bachelier
  • 121
  • 8
  • 1
    Your first bit of code is the same as `outimage2(outmask>0)=outimage(outmask>0)` and will not let you speed up processing, your image size is not changing. You could try to extract the rectangular regions of the mask and smooth within each of them, but it's not trivial code. Look for `regionprops` to find the bounding boxes of each region in your mask, then loop over each region and use regular indexing to extract them. – Cris Luengo Dec 30 '17 at 16:45
  • Note that `[medfilt_val(1) medfilt_val(2)]` is the same as `medfilt_val(1:2)`, but slower. But your second bit of code will never work, you get a 1D image out of that indexing. – Cris Luengo Dec 30 '17 at 16:47
  • Possible duplicate of [Matlab, how to apply medfilt2 function on specific pixel locations](https://stackoverflow.com/questions/45245987/matlab-how-to-apply-medfilt2-function-on-specific-pixel-locations) – Leander Moesinger Dec 31 '17 at 14:43
  • Especially read rayryengs comment in the duplicate target – Leander Moesinger Dec 31 '17 at 14:43

0 Answers0