[XviD-devel] GME and Block ME

[Radek Czyz]

Hi again,

Saturday, April 19, 2003, 12:00:40 AM, gruel wrote:


> 32x32 blocks is too coarse (I believe), at least in low sizes (352x288 and
> below). I rather think we should do a good MEAnalysis and reduce later ME
> instead of doing a coarse MEANalysis and full ME.

MEAnalysis performs search in 32x32 blocks, but it works like inter4v
mode - four vectors are evaluated each time. It will not catch
movement of smaller objects (diamond will not follow them) but it's
good for gmc, isn't it?
 

> It's kind of an advanced "TOOSMALL_LIMIT". Sometimes reducing a qunatized
> coefficient to 1 less than the "correct" result can save many bits. But of
> course the quantization error gets larger.
> So you introduce a lagrangian to weight distorsion versus bits needed and
> optimize quantized coefficients to optimal.

So this is exactly what I thought. In that case, what is ffmpeg
actually doing? At constant quant 2 it increases filesize by 10% and
improves PSNR by almost 1dB .. ?_?

> In theory (mathematically) it should be identical to SSE (DCT is an
> orthogonal transform).

Isn't that 'sum of absolute coefficients'? As far as I remember sum of
absolute coefficients is the power of the signal, which is sse.

Radek