Calculation of psnr value in RGB image. I have calculated psnr using this code. But the result obtained is a complex number. What should be the error??

if true

function PSNR = psnrgb(I,W)

[m,n,p] = size(I); [h,w,q] = size(W); if m ~= h n ~= w p~=q error('Two images must have the same size.') end for k=1:p for i=1:m for j=1:n delta=sum(I(i,j,k)-W(i,j,k).^2); end end end

delta = 1/(m*n*p) * delta ; PSNR = 10 * log10( 255^2/delta );

end end

Answer 1

[DGM]

There are plenty of problems.

function PSNR = psnrgb(I,W)
    [m,n,p] = size(I); 
    [h,w,q] = size(W); 

    % obviously, this test expression is invalid as written
    if m ~= h n ~= w p ~= q 
        error('Two images must have the same size.') 
    end

    % this entire set of loops does nothing but calculate
    % the squared difference of the last pixel pair alone.
    % while the rest of the entire image is ignored, 
    % the inputs are presumed to be uint8, so it's a 50% chance 
    % that this answer for that one single pixel is still completely wrong.
    for k=1:p 
        for i=1:m 
            for j=1:n 
                delta = sum(I(i,j,k)-W(i,j,k).^2); 
            end
        end
    end

    % convert the sum to a mean, but we don't have a sum of anything
    delta = 1/(m*n*p) * delta ;

    % this presumes the input is uint8
    PSNR = 10 * log10( 255^2/delta );
end

We don't need to try fixing that. IPT already has psnr(). If you want to write your own, you can:

% the inputs
A = imread('peppers.png'); % the reference
B = imnoise(A,'gaussian',0,0.1); % the image to test

% compare the example against IPT psnr()
psnrrgb(B,A)

ans = 11.8650

psnr(B,A)

ans = 11.8650

function P = psnrrgb(inpict,ref)
    % synopsis goes here
    
    % test the inputs
    if ~isequal(size(inpict),size(ref))
        error('Images must both have the same size')
    end
    if ~strcmp(class(inpict),class(ref))
        error('Images must both have the same class')
    end
    
    % calculate the output
    err = double(inpict)-double(ref); % error
    mse = mean(err(:).^2); % mean-square error
    dr = diff(double(getrangefromclass(inpict))); % dynamic range for this numeric class
    P = 10*log10(dr^2/mse);
end