[mythtv] [Draft] Filter documentation.

[D Banerjee]

From: "Kenneth Aafløy" <ke-aa at frisurf.no>
> D Banerjee wrote:
> > > [About postprocess filter in libavcodec/MythTV/mplayer]
> >
> > Actually, it's the same code. Non-operational right now.
>
> ? Just tested it on my epia m10k, and it's working fine,
> very slow on this low yeild cpu though :)
>
> Kenneth

Let me rephrase: the deblocking/deringing has almost no visible effect.

Just to push it to extremes, this is what I did - recorded 4 minutes of
video at a grand total for 12.8MB. Blocks, blocks, everywhere. I've tried
regular deblockers, x1, and the RK filter, dering, default. Sometimes I
think I see some blurring, but it could just be bias as this is a 'no-blind'
test :) Definately full of blocks. If I play back the same video using
xvid/divx codec (thanks to the mythtv directshow plugin) with full
deblocking (with or without deringing) I can't make out a single block, and
it doesn't blur regular video either :)

Imho if people can identify blocks above 2200 kbits in except for very rare
circumstances, we don't have effective postprocessing.

FYI, the RK filter is junk. I implemented some "clean room" filters
according to their paper, and it's just plain wrong. The
postproccess_template.c comments "but introduces ugly horizontal patterns".
If you see the paper, it becomes quite apparent what their mistake is, I
don't think they even bothered to code it up and test themselves. You can
find it here: http://citeseer.nj.nec.com/544456.html It's actually quite
easy to fix so it has _some_ positive effect.