[XviD-devel] Re: Luma vs Chroma Quant

[Michael Militzer]

Hi,

Quoting "Hans K. Rosbach" <hk at circlestorm.org>:

> Okay, this is the fourth time I'm sending this mail since jan 8.
> Hopefully it'll arrive this time..
> 
> ---
> 
> During a discussion with some of my friends, one of them
> came up with an interesting question..
> 
> Is it possible to encode chroma/luma components of the
> picture with different quant values? Thus one would for
> example be able to down-prioritize the chroma in favor of
> the luma..

No, it's not possible to use different quant values for luma and chroma.
 
> I cannot really guess what the result would be like, but I'm
> quite sure the ability to adjust this would be one of the many
> tweakers wet dreams. (Not that it necessarily makes anything
> better tho..)
> 
> Do we encode all components of each block with the same
> quant today? If so, is it MPEG4 compliant to encode them
> with different quants?

The quantizer value can only be changed per MB (so one quantizer value for
all 6 subblocks). Coding special chroma quant values into the bitstream is
thus not MPEG-4 compliant.

> If yes to both above, would it maybe be more effective to
> consider this also in two-pass where we can test its relative
> compressibility in advance?

Even though you can't use a different quant value for chroma blocks, it's
still possible to lower quality and save some bits for chroma by simply
changing the quantized chroma coefficients (similar to what trellis does).
However, the chroma planes compress really well already, so saving some
bits here has no big influence on the overall global bitrate. So it seems
pretty doubtful if the bitrate drop would justify the lower picture
quality.

bye,
Michael