[XviD-devel] Xvid v's H.264/AVC

[Hans K. Rosbach]

> > I assume that this is not as valid when re-encoded to lower resolution
> > and/or
> > with some kind of filtering before re-encoding..?
>
> You are right, possibly not. Although the resizing filters might
> have to somehow smooth the image as well to avoid aliasing.
> What kind of filtering did you think of?

For cartoons I tested out some special cartoon filters for virtualdub
that actally made a nice improvement in quality. Not closer to source,
but more pleasing image. It essentially found black lines and made them
stronger and found big areas of one color and smoothed out any artifacts
in them. I used several such filters, dont remember names right now.

> P.S. I might have stated it wrong in the original posting: It's not that
> MPEG-4 Qpel is bad at encoding MPEG-2 material or anything. It's just that
> MPEG-4 Hpel is extremely good at it, otherwise it would be impossible to
> reduce a 8 Mbps to less than 2 Mbps with almost no visual quality loss.
> QPel doesn't make improve it any further.
> But this is also a problem of measuring performance, because a high PSNR
> value (which many people take as "good quality") here only means that the
> MPEG-4 result is close to MPEG-2's output including all visual artefacts
> that that might have had.
> In fact, some people say that Qpel images look better than Hpel ones, even
> at same PSNR.

Yes, PSNR is a lousy method of measuring visual quality.

-=Dead2=-