any recommended readings about the theory for video compression ?

[gigah72]

i would like to know places there i can find interesting documents
like psychovisual condiserations to reduce bitsize if coding mpeg,
possible settings for any thresholds needed and why,
different algo's for motion estimation
and all kind of this stuff.
how the xvid team did it, or guesses why cinemacraft's cce gets
better results with 3-5 passes and what data is in the *.vaf file.

looking at source code's is no option, i'm not a programmer.

greets,
g.

[sh0dan]

If you search the XVID forum you'll find loads of links.

Also look in the xvid.org forums and the xvid.org mailling lists.

[gigah72]

thanx sh0dan,
but search for what ? the keywords from my post gave several pages of
result with threads containing many many answers. it will take
some days to search doom9 and after that xvid forums ?
if you know how to improve my search her or there, i would like to
know.
wouldn't it be possible to have a sticky on this topic ?
theory is always improving methods and if someone does spot an
interesting publication, this would be the place to post.

greets,
g.

[sh0dan]

If you can learn it in a matter of days, you're good off ;)

MPEG compression is a complex matter, without much precise and "ultimate" documentation - all I can say is that most things I've learned I've picked up from these places I mentioned.

There are loads of ME-discussions (and many different approaches has been discussed) both on doom9 and xvid.org - I suppose you already know of quantization?

Regarding CCE: The VAF-files are probably mostly motion vectors. The additional passes tries different motion vectors to see if they can achieve the same (or better) quality using these motion vectors. This will slightly change the bitrate curve, and they then compensate for this in the following pass, while still trying to find better motion vectors. This is my guess anyway. They might also apply different filtering, but i don't think so.