[MPlayer-users] Chroma interpolation

[D Richard Felker III]

On Thu, Mar 04, 2004 at 08:26:35PM +0200, Ville Saari wrote:
> On Wed, Mar 03, 2004 at 01:46:13PM -0500, D Richard Felker III wrote:
> 
> > > It messes up the chroma if the video is progressive. And even if it
> ...
> > Both of these problems were fixed a while back in cvs. I don't
> > remember if that was before pre3 or afterwards though. Anyway, it does
> > interpolate the fields separately, so it's not quite as good quality
> > as doing them together would give, but looks fine in practice.
> 
> But it has to mess up the chroma for progressive video. Otherwise it
> wouldn't do what it is supposed to do for interlaced stuff!

Nope, that's what I thought when I originally wrote the filter. Then I
read a proper explanation of how interlaced 4:2:0 works. It's really
quite clever. :)

> My mplayer is pulled from the cvs some time in February. I compared its
> output to 1.0pre1 which happened to still linger on my hard disk and
> the chroma combing defect appears to be softer on the newer version.
> But it is still there.
> 
> It looks like if it blends the fields partially together. But if it does,
> then it must cause chroma-only-ghosting in its intended usage?

Nope. It does linear interpolation using the correct field only and
proper weights. If you get slight artifacts, this is because the
samples are sufficiently far apart that interpolation isn't very good.
But there's definitely no mixing between fields. And the interpolation
is done correctly so that if there's no very-high-frequency vertical
content in the chroma (which isn't fully true in practice), it should
look as good as doing everything progressively.

Rich