[XviD-devel] Early SKIP and early DC-only

[Christoph Lampert]

Hi,

I've come across this paper: 

Statistical Computation of Discrete Cosine Transform in Video Encoders
http://citeseer.nj.nec.com/sun98statistical.html

which finally gives some background about when early SKIP of a block might
be a good choice. The authors calculate statistical thresholds for 8x8-SAD
values based on the quantizer parameter Q when SKIP and only_DC is likely
to happen. 

They propose:          SAD < t1 * Q     -> all DCT to 0 (skip)
           t1 * Q  <=  SAD < t2 * Q     -> calc only DC
           t2 * Q  <=  SAD 

Where t1 and t2 are thresholds: 

t1=24.39  t2=60.48    for 99% confidelity
t2=36.58  t2=90.71    for 95%

As expected the maths behind it is rather "elementary", but maybe somebody
want to have a look himself.

gruel