How to calculate standard deviation from ROI of image after thresholding?

Hello, I have images and I use graythresh threshold to select object from the image. I want to get standard deviation of the pixel value from the ROI of the image. From regionprop, I can only get maxintensity, minintensity, meanintensity, pixelvalues, and weightedcentroid.

level=graythresh(I2);
BW=im2bw(I2,level);
figure, imshow(BW);

%ROI

L=bwlabel(BW);
s=regionprops(L,I2, 'MaxIntensity', 'MeanIntensity','MinIntensity','PixelValues');
max_I=s.MaxIntensity
min_I=s.MinIntensity
%average_I=s.MeanIntensity
pix_val=s.PixelValues;

M=max(max(L));
for i=1:M
    sumAv(i)=s(i).MeanIntensity;
end
average_I=mean(sumAv)

I need to find standard deviation from the ROI. Can anyone help me to solve this. I've been looking for a way, unfortunately, I'm not very good with Matlab. This is my first time using it.

Thank you in advance.

Answer 1

[DGM]

You have the pixel values, so you can just use those.

% a grayscale image
inpict = imread('cameraman.tif');

% an antialiased mask with several blobs
mask = imread('cmanpolyblobs.png');
mask = mask>128; % we don't need the antialiasing
imshow(mask)

% get regionprops
S = regionprops(mask,inpict,'pixelvalues','meanintensity','centroid');

% these are represented in unit scale, 
% that way the results are independent of the input class
nblobs = numel(S);
regionmu = im2double(cast(vertcat(S.MeanIntensity),class(inpict)));
regionsigma = zeros(nblobs,1);
for k = 1:nblobs
    regionsigma(k) = std(im2double(S(k).PixelValues));
end

% create a composite image, and plot the blob stats
% so that we have some idea that they make sense
alph = im2double(mask);
BG = im2double(inpict);
comp = alph.*BG + 0.5*(1-alph).*BG;
imshow(comp); hold on
for k = 1:nblobs
    c = S(k).Centroid;
    ht = text(c(1),c(2),sprintf('mean:%.4f\nstd:%.4f',regionmu(k),regionsigma(k)));
    ht.Color = 'k';
    ht.HorizontalAlignment = 'center';
    ht.BackgroundColor = 'r';
    
    % the centroid of this blob isn't a good place to put the annotation
    % so let's just move it over
    if k == 1
        ht.Position(1) = ht.Position(1) + 70;
    end
end

Oh wow. That's a jumbled mess on the forum. Oh well.

Either way, we can see that the dark region on the coat is very smooth, and has a low standard deviation. The camera and face have a lot of contrast, and so the std is high. The grass and sky are in between.