[MPlayer-dev-eng] New inverse-telecine filter

[Zoltan Hidvegi]

> > NTSC have the same vertical chroma resolution as luma, but one quarter
> > of the horizontal resolution, that's why I say that NTSC is 422, while
> > YV12 is 420.
> 
>   You said NTSC is 4:1:1, why?  Studio NTSC is always stored and
> transmitted as 4:2:2 (half the horizontal resolution for chroma, same
> vertical resolution).  The only 4:1:1 storage and transmission format in
> common use that I know of is DV.

The problem is that broadcast NTSC has a very limited bandwidth, and
2/3 of that is taken by the luma info.  Color NTSC is really just an
afterthought, with color hacked on top of black and white NTSC.
Studio NTSC is 4:2:2, but that is cut down by the limitation of the
broadcast.  That said, it is probably better to oversample in the TV
card and scale down, so capturing 4:2:2 is good.

> > Much of the content on TV is originally progressive, broken up to
> > interlaced fields, such as telecined movies or video from a 30fps
> > progressive camera.
> 
>   I would say that most is true interlaced content, or at least more
> than you imply (TLC, CNN, Comedy, any Sports broadcasts, ...).

It depends on what you watch of course.  I do not care much about CNN
or any other TV news if I can listen to NPR, and I do not watch
sports, what's left is mostly progressive.  And most of the ads are
also progressive, either film telecined or 30fps, and that's alredy
more than 25% of all TV (even though that's the 25% I usually cut
out).

>   If you are intending to convert to YV12 I would agree with you:
> capture in 4:2:2 and do all processing before converting to any 4:2:0
> format, and indicate in your output whether the YV12 frame is interlaced
> or progressive (like the progressive_frame flag in MPEG2).  Similarly,
> if you're just watching TV, keep it in 4:2:2 right to the display.
> That's what I do in tvtime.

Sure, for live TV, go with 4:2:2.

>   But if youre capturing in YV12, then it's a bug in bttv if it does not
> give interlaced chroma, there is no way around that.

But what I'm saying is that if it did give interlaced chroma, than it
would destroy the chroma of progressive frames, so encoding movies
from TV would look bad.  There is no correct way to subsample mixed
interlaced/progressive 4:2:2 into 4:2:0 without knowing what is
interlaced and what is progressive.

Zoli