[XviD-devel] Re: PSNR question/bug

[Rémi Guyomarch]

On Sat, Jan 25, 2003 at 03:11:33PM +0100, Marco elcabesa Belli wrote:
> looking inside image.c i found psnr fouction, after calc of sum of  (diff)^2  
> you make  psnr_y=255*255/result
> andthen SNR = 10 log10(psnr_y)
> 
> is this right??
> 
> isn't better do
> 
> psnr_y =sum (diff)^2
> s= sum (original pixel) ^2
> 
> snr= s/psnr;
> SNR =10 log10(snr);

Why an error of X in a realtively bright pixel whould be more
"important" than the same X error on a relatively dark pixel ?
(or the reverse, I don't recall if Y=255 means white or black)

Even if it was the case, think about Cb and Cr ...

-- 
Rémi