custom quantizer matrices "between" H263 and MPEG

[torsius]

i hope i'm not late or ignorant on this topic.
would it be possible or useful to implement a new kind of modulation type somewhat similar to the kinds included currently in koepi's builds?

the variation i had in mind would use a function that instead of a 'salt and pepper' approach to the modulation, would create a custom matrix that is a medium between the h263 and mpeg. the closeness of the matrix to h263 or mpeg would be a result of high/low quantizer or high/low bitrate, whichever would result in better compressibility/quality.

if this idea is stupid or just doesn't make any kind of sense, i'm sorry, but i was thinking about it today when i was trying to decide on which matrix type to use and thinking of the plus/minus of each of them :)

[NuclearFusi0n]

http://forum.doom9.org/showthread.php?s=&threadid=34493

[torsius]

using a single custom matrix isn't exactly what i'm talking about, maybe a function that switches between the "good, better, and best" spoken of in that thread would be closer to what was trying to say.

[Mango Madness]

this has been proposed before even suggested up to like 16 matrices if i remember correctly. We'd have to do much more extensive testing to the custom matrices before something like this could be implemented. Though, i doubt it be too difficult to do. I"m sure there's plenty more to come for xvid before h.264 takes over.

[ookzDVD]

How Tsunami's default Matrix ?
I think it's good enough to try.

[bilu]

Read this thread (http://forum.doom9.org/showthread.php?s=&threadid=45695) , could be what you want.

Bilu

[Suikun]

Using different quantizer matrices in one encode is not that easy.
There was some discussion about it already, and it seems that modulating between different matrices doesn't comply to the MPEG-4 specs.

[Rrrough]

that modulating between different matrices doesn't comply to the MPEG-4 specs.
You can store the custom quantization matrices in a VOL-header.
AFAIK MPEG-4 specs only describe that for I-frames, they don't say anything about P- or B-frames. So changing the matrix between every GOP seems standard-compliant, doing that within a GOP (P- or B-frame-wise) probably isn't standard-compliant.

cheers

[bilu]

Related news for Avisynth users:

http://forum.doom9.org/showthread.php?postid=266216#post266216

Bilu

[Defiler]

However, storing custom matrices like that can quickly erase any compressibility gains you earned, because you've increased the entropy of the bitstream. If I recall, this was the (valid) argument used when this issue was raised in the past.