efficient use of extracting mean over segments

Hello, I have image (Iseg) where i have number of segments. Each segments is represented by unique number in Iseg. The segments are around (50000). I have another image(Idata) which is same size that of Iseg. For every segment in Iseq i want to extract mean value of segments from Idata.

Rightnow i am using like this

NumRegion = max(max(Iseg));
mean_magnitude = zeros(size(Iseg));
for i=1:NumRegion
    [iIndList] = find(Iseg == i);
    mean_magnitude(iIndList) = mean(Idata(iIndList));  
end

It works fine, but the problem is that it takes a long time. about 10 mins in my machine. Is there a effecient way to do it.

[Jan]

What is a "segment"?

[Image Analyst]

I'm not sure either. Is a "segment" a blob in a binary image? And he wants the mean gray level of a gray level image for each blob region? Sounds like Iseg is a "labeled" image like you'd get from bwlabel (connected components labeling) where each blob has a unique ID number assigned at every pixel in the blob.

[Sukuchha]

Image Analyst, you are right. Iseg is a grayscale image, where each blob has a unique ID number assigned at every pixel in the blob.

Answer 1

[Image Analyst]

Do you, by chance, mean this:

measurements = regionprops(Iseg, Idata, 'MeanIntensity);
allMeanIntensities= = [measurements.MeanIntensity];

[Sukuchha]

Thanks for taking time for answering.

The code you provided correctly calculates meanIntensity for each blob.But the problem now is allMeanIntensities is vector of size equal to number of blobs. But i want to have an image (size equal to Iseg and Idata) where each pixels in a blob is represented by corresponding allMeanIntensities.

[Sukuchha]

used a custum function 'find and replace' using ismember to map back allMeanIntensities to raster image. Thx

[Image Analyst]

I would have used intlut() on the labeled image.

[Sukuchha]

i had a look at intlut before but couldnot figure out how to use it. Could you please show one example(or pseudo code)?

[Image Analyst]

Something like (untested)

newImage = intlut(labeledImage, allMeanIntensities);

Answer 2

[Jan]

At first you can cleanup the loop a little bit using logical indexing:

NumRegion      = max(max(Iseg));
mean_magnitude = zeros(size(Iseg));
for i = 1:NumRegion
  iIndList = (Iseg == i);
  mean_magnitude(iIndList) = mean(Idata(iIndList));  
end

I assume accumarray is faster. I'd try to run a loop over Iseq instead also, but I cannot test it due to the absence of test data. It might be helpful if you post some, e.g. produced by some rand calls.

[Sukuchha]

The looping is much slower. The code given by image analyst calculated mean intensities over each blob in much faster way but problem now to to make a raster image out of it. measurements = regionprops(Iseg, Idata, 'MeanIntensity); allMeanIntensities= = [measurements.MeanIntensity];