[XviD-devel] questions:chroma half pixel interpolate.

[Christoph Lampert]

On Thu, 20 Jan 2005, ??? wrote:
>     when xvid to code p frame,it will first do one frame halfpel
> interpolate,why it only do luminance halfpel interpolate? and let chroma
> halfpel interpolate in motion estimate and motion compensate block by
> block. why not treat chroma half interpolate as same luminance? It is
> only to save memory? I think treat chroma half interpolate as luminace
> are faster,can xvid expert tell me why? Best regard xiongshu deng

To interpolate the whole picture into halfpel (frame-based) is only useful 
(meaning faster) for doing HalfPel Motion Estimation where many positions 
have to be checked, not for Motion Compensation where only 1 halfpel 
position per block is needed. So for Compansation alone, block-based 
halfpel calculation is faster. 

In normal Motion Estimation, we only use Luminance to find motion vectors, 
so we only interpolate Luminance frame based, and the chrominance we do 
block-based, if it's needed at all.

gruel