[MPlayer-users] PSNR Questions

[Martin Pavon]

On Sat, Feb 15, 2003 at 05:32:04PM +0000, Ian wrote:
> [Automatic answer: RTFM (read DOCS, FAQ), also read DOCS/bugreports.html]
> -----BEGIN PGP SIGNED MESSAGE-----
> Hash: SHA1
> 
> D Richard Felker III wrote:
> 
> > Huh?? This is nonsense. You can't compare PSNR relative to two
> > different reference 'signals'. When encoding, the source is considered
> > the signal, and the difference between the encoded picture and the
> > source is the noise, so the source has infinite PSNR automatically.
> 
> OK, this is the first time I've seen it stated, a formula would be nice 
> too, but in any case see below . . .
> 
> > Yes, theoretically. Computational issues might make it come out
> > different tho.
> 
> Regardless, since in my experience better quality material (ie. the 
> originals!) has a lower PSNR than the copies, I can't take it seriously. 
>   Obviously your experience is different.
> 
> > Huh?? Do you even know what PSNR means??
> 
> Can't find the reference (it seems to be unavailable ATM), but I thought 
> it was peak signal to noise ratio.  I can't be that much of a dimwit if 
> I'm using the encoder successfully, now can I? ;)
> 

Hi!!!!

Sometimes ago, I found a formula for PSNR in the following link:

http://appliedit.arc.nasa.gov/videotech/psnr.html

A short cut & paste give us:


MSE = 1 / n * ( Sum(i)(j) |P(i)(j) - Q(i)(j)|^2 )

 Where P  and Q are  two images.  i and j  are the horizontal  and vertical
 locations  of a  pixel. P(i)(j)  is  the value  of the  pixel at  location
 (i)(j). n is the total number of pixels in the image.


and


PSNR =  10 log_10 (b^2 / MSE)

 Where b  is the peak value  for a pixel,  typically 255 for 8  bit pixels.
 PSNR is usually quoted in decibels (dB), a logarithmic scale.


I hope someone can find this usefull

Regards,

Martín

-- 
Martín Pavón
martin_199ar at yahoo.com.ar
http://mate.dm.uba.ar/~mpavon/