how I can measure the robustness of watermark?

I am working on one of the watermarking algorithm, I wanna measure the robustness of the watermark Image the PNSR used for original Image , I could not use it for watermark because it is double Image , the measure should done between the Watermark and Extracted Watermark, all of images are unit8 class any suggestion please?

[Image Analyst]

Why do you say that? The documentation clearly says psnr() can bu used with both double and uint8 images. Are you trying to use psnr() on the watermark recovered from the original and the watermark recovered from the attacked image? Like

p = psnr(originalWatermark, attachedWatermark);

Is your watermark recovery algorithm smart enough to handles attack like cropping, rotation, coloration/intensity change, etc.?

[Eliza]

My compassion between the original watermark and the extracted watermark, I did arrive to the attacks level yet. The algorithm I am working on, is to prove the robustness of watermark and compare it to the extracted WM by just looking to the image. I wanna measure it by calculation either the difference between them or the similarity

This is the original watermark

32x32x3 uint8

This is the extracted watermark

32x32 uint8

The PSNR function as

function y=psnr(processed,original)
% processed=im2double(processed);
 original=im2double(original);
[m n]=size(original);

%mserror
error=processed - original;
se=error.*error;
sumse=sum(sum(se));
mse=sumse/(m*n);
%mserror

ma=max(max(processed));
y=10*log10(ma*ma/mse);
display(y);

[Image Analyst]

psnr() is a built-in function since R2014a. Why are you overwriting it with your own?

ALso, if you embedded a color watermark, and you're extracting a grayscale version of it, then your extraction method is not right. You should be able to extract the very same image you embedded.

[Eliza]

I used it but if it gives me negative value what does it mean?

Answer 1

[Walter Roberson]

double() the image that is uint8 before calculating the PSNR against the double() array.

[Eliza]

the error in the dimension between the Original WM and Extracted WM

[Subhadeep Koley]

[row col]=size(original_wm);
extracted_wm=imresize(extracted_wm,[row,col]);

after executing this piece of code psnr() will work fine.

Apart from psnr you can also test structural symmetry between two images by calling

ssimval=ssim(extracted,original);

Hope this helps :)