How to retrieve specific pixels from an image based on the label

Hello everyone,

I am conducting some works on image restoration, particularly on specular reflection removal. I have obtained some results and I would like to evaluate the performance of my method using some quantitative metrics suh PSNR, SSIM, COV, etc. However, I would like to compute these metrics only for the regions that have restored. Please, how can I retrieve only the pixels of the restored regions from the processed images? This is the code I wrote for computing the evaluation metrics for the entire image.

allPSNR = zeros(1, numImages);
allSSIM = zeros(1, numImages);

for k = 1 : numImages
	baseFileName = theFiles(k).name;
	fullFileName = fullfile(theFiles(k).folder, baseFileName);
    
    baseFileName1 = theFiles1(k).name;
	fullFileName1 = fullfile(theFiles1(k).folder, baseFileName1);

	lab = imread(fullFileName);
    img = imread(fullFileName1);
    
    allPSNR(k) = psnr(img, lab);
    allSSIM(k) = ssim(img, lab);   
end
    
meanPSNR = mean(allPSNR);
meanSSIM = mean(allSSIM);

[Image Analyst]

How are you defining the regions? Do you have some binary image or something that maps out the regions you want to consider and compare?

[Patrice Gaofei]

Thank you for your comments. Yes, I do have the binary image that will help map out the regions. That is what I call label.

[Patrice Gaofei]

Please Mr./Mrs. Image Analyst, any hep?

Answer 1

[Image Analyst]

See if it will work for one-dimensional vectors:

    allPSNR(k) = psnr(img(binaryImage), lab(binaryImage));
    allSSIM(k) = ssim(img(binaryImage), lab(binaryImage));   

[Patrice Gaofei]

Thank you very much for your help. I have got it working as follows.

    img2 = img(logical(mask));
    lab2 = lab(logical(mask));
    
    allPSNR(k) = psnr(img2, lab2);
    allSSIM(k) = ssim(img2, lab2); 

[Patrice Gaofei]

I have noticed that img2 and lab2 are both column vectors with several elements with the same values. Please, could you clarify why several elements are having the same the value? In my opinion, these values of the vector elements are pixel values. In this case, does it mean that there are different pixels with the same values?

[Image Analyst]

If you have a 2-D array, and give it a logical mask, it will pick out all the values and put them into a 1-D list. This is because if you're taking some values but discarding others, there is no way it could be a 2-D list. For example if m =

[1, 2, 3;

4, 5, 6;

7, 8, 9]

and your mask is

[ 0 1 0

1 0 1

1 1 0]

You'd need to extract out ONLY the 2, 4, 6, 7, and 8. There is no way to have that be a 2-D array if you're not wanting to extract the 1,3,5, and 9.

Now, if the values are equal, then the values are equal, and if they're not equal, they're not equal. If the mask is the same for both img and lab, which it is, then you're comparing the same corresponding pixels in each image, not different ones. They may have the same value or they may be unequal.

[Patrice Gaofei]

What I mean is that the obtained 1-D list for either img and lab contains same values at different postions (index) of the list. Is this ok? This is what I would like to understand. I thought that the pixel values at different locations of the same image cannot be equal. Sorry if my question seems silly. I am just trying to get it clear.

Thank you for your time and patience.

[Image Analyst]

Of course pixel values at different locations can be equal. What if you had a completely uniform image where every single pixel in the entire image had the same gray scale value or RGB color? Maybe I'm not understanding what you're trying to say.

[Patrice Gaofei]

Ok Sir/Madam! Thank you very much for the clarification.