what is wrong whith my script?

I need to calculate the SNR but this code wrong

code:

for i=0:(size(im,1)-1)
    for j=0:(size(im,2)-1)
        som1 = som1+im(i,j)^2;
        som2 = som2+(noisy_im(i,j)-im(i,j))^2;
    end
    SNR = 10*log10(som1/som2)
end

where is the error?

Answer 1

[Andrei Bobrov]

It's MATLAB

n = size(im);
SNR = zeros(n(1),1);
som1 = 0;
som2 = 0;
for ii=1:n(1)
    for jj=1:n(2)
        som1 = som1+im(ii,jj)^2;
        som2 = som2+(noisy_im(ii,jj)-im(ii,jj))^2;
    end 
    SNR(ii) = 10*log10(som1/som2);
end

vectorize variant:

a = sum(im.^2,2);
b = sum((noisy_im-im).^2,2);

SNR = 10*log10(cumsum(a)./cumsum(b));

[Andrei Bobrov]

corrected

[Image Analyst]

They do different things. The loop version gives the SNR of every row (but not really) while the vectorized is of the whole image. I said "not really" because you're not reinitializing som1 and som2 to 0 at the beginning of each row. So it's some sort of cumulative SNR, which is hard to interpret.

Answer 2

[Image Analyst]

Your first error is starting with 0 as the index. In MATLAB indices start with 1. The next error is that SNR gets overwritten in each row so it will end up with only a single value, not an array like you would get if you had an index for SNR. The SNR can be pulled out of both loops and be calculated after the loops exit.

You appear to want to calculate the peak signal to noise ratio. There is a function in the Image Processing Toolbox for that:

SNR = psnr(noisy_im, im);

[nahla]

I'm debutant

I need to calculate the SNR of an image then

if the SNR < threshold I denoise the image else I do any thing but the problem here is to calculate the SNR of an image I should denoise it by using fitler!

the question: there is an other solution to test the image noise before his amelioration?

[Image Analyst]

But you already have the "noise free" image - you called "im". So why do you need to denoise anything?

[nahla]

in the first I denoised my original image "noisy_im" to get the filtred image "im" then I calculate the psnr(noisy_im,im) I do all that to evaluate the quality of the original image "noisy_im". I saw that it is not logic that I denoise the image to evaluate its quality. So I want to know if i can evaluate image quality without denoising it? and then if the image has a bad quality I will denoise it. else I will keep it. i hope that is clear now