[XviD-devel] PSNR question/bug

Christoph Lampert xvid-devel@xvid.org
Sat, 25 Jan 2003 16:32:45 +0100 (CET)


On Sat, 25 Jan 2003, Marco "elcabesa" Belli wrote:

> after some calc.. i understood that 40 is a good PSNR using only 8 bit for y=)

I guess PSNR above 40 has hardly any visible erorr (in realistic
scenarios; it's surely possible to create a picture where PSNR of 40 
looks horrible). But you cannot really say PSNR of 36 for one scene is
better than 33 for another etc., that depends on too many factors. In
generally, clips with more "structure" (hair, grass) have lower PSNR,
because more high frequencies are quantized away, but these are also less
visible, so the visual quality might even be better. 
Compare PSNR only for different encodings of the same clip, and you are on
the safe side. 

gruel