[MPlayer-users] Re: [-] Re: Crop before deinterlacing OK?

[Rich Felker]

On Tue, Jul 26, 2005 at 11:23:28PM +0200, Matthias Wieser wrote:
> Am Dienstag, 26. Juli 2005 21:47 schrieb Rich Felker:
> > On Tue, Jul 26, 2005 at 09:05:48PM +0200, Matthias Wieser wrote:
> > > > and then pp=l5 will also result in worse psnr during
> > > > compression,
> > >
> > > No, pp=l5 is nearer to the original, so psnr is higher.
> >
> > It's pointless to compare psnr unless you compare the final psnr AFTER
> > encoding versus the original picture..
> 
> If the image prior to the encoding is already bad, the codec won't improve 
> it.

The codec always _reduces_ the quality of the image. But it will
reduce it a lot more if it contains more high frequency content. Here
are the possible paths:

original ------------> lb blended ------------> encoded
         high loss in             low loss
         the form blur

original ------------> l5 blended ------------> encoded
         low loss in              high loss in
         the form of ringing      the form of blocking/ringing

This is why I say using l5 is potentially a bad idea. At least it
needs measurement (psnr) or better double-blind testing to decide
which is better (but both are very bad anyway).

Rich