1

Say, I have original image, noised and denoised. By PSNR value I want to find out how good does my denoising algorithm works.

For which pair of the images should I calculate it? I've tried it for original+noised, noised+denoised and original+denoised and got PSNR = 28, 29 and 31 respectively.

What value shall be appropriate?

I'm calculating PNSR as here.

ans
  • 173
  • 7

1 Answers1

1

With original image $o$ fixed, the higher, the better. If PSRN($o,d$) is bigger than PSNR($o,n$), you can claim an improvement in the decibel measure. It is difficult to interpret PSRN($d,n$), since you compare two more-or-less noisy data.

Is PSNR a good measure of subjective image quality remains an open question.

Laurent Duval
  • 31,850
  • 3
  • 33
  • 101
  • Thanks, I got it. But another question comes in mind: What if I have same image (PSNR(o,n) is the same), but denoised by another algorithm and its PSNR(o,d) is higher (than original psnr(o,d)), but it looks much worse visually. How could it be explained? I'm sorry, if such a question is inappropriate for a comment. – ans Jun 08 '19 at 15:44
  • This unfortunately happens: add a constant (a bias) to the "better" denoising image, and the PSNR will get worse, while it remains good visually. PSNR is very debated what that, hence other methods, like SSIM, have been developed, but nothing fully effective so far – Laurent Duval Jun 08 '19 at 15:57