[XviD-devel] some bframes ideas

[Marco Al]

From: "Christoph Lampert" <chl at math.uni-bonn.de>

> Sorry, I may be a mathematician, but this is too fast for me.
> Do you want to modify dequantization?

Yep, as I said ... during dequantization. Since the b-frames arent
references that doesnt present any problems. Anyway it is not my idea the
paper describes it better, the basic idea is that if the distribution of
coefficients is not uniform then if you use a mid-tread quantizer and
reconstruct to the center of the quantization range during dequantization
the error will be biased. This bias is predictable, and as such can be
removed.

Marco