Compute texture for small parts of the image

Hello,

I would like to ask about how to computing the texture of a part of the image efficiently, i.e., I applied superpixel segmentation (SLIC) on an image to get smaller homogenous regions, how to compute the texture for each superpixel efficiently (e.g., fast execution)?

any help will be appreciated,

best regards,

Answer 1

[Image Analyst]

Depends on how you define texture. One way is to use the labeled image from superpixels() to get the mean intensity of the StDev image

windowSize = 9; % Whatever...
stDevImage = stdfilt(grayImage, ones(windowSize)); % A local standard deviation.
% Get the standard deviation averaged over each superpixel region.
props = regionprops(labeledImage, stDevImage, 'MeanIntensity');
for k = 1 : length(props)
    fprintf('The mean standard deviation in superpixel #%d of %d is %f.\n', ...
        k, length(props), props(k).MeanIntensity)
end

[MatlabUser]

Could you please explain to me why to use the StDev?

I was expecting to compute gradient or transformation.

[Image Analyst]

I don't know what you mean by transformation. The gradient will tend to find edges, which are regions of high standard deviation. But stdfilt() gives you control over the window size and imgradient() does not, so you can use stdfilt() to better tune your filter to your particular image.

[MatlabUser]

I meant in frequency domain,

Got it, thank you so much for explaining,

One last question please, can I use the superpixel instead of window size to compute the texture for each superpixel's centroid?

[Image Analyst]

No, I don't believe so, unless the help says superpizel() returns a texture. It returns a labeled image which you can then use with a texture image and regionprops() to determine texture in each superpixel region.

[MatlabUser]

Got it, thank you so much for explaining,