[theora] NHW codec - progress

[Raphael Canut]

Hello,

I am slowly working on the NHW codec, I think now that I will remove the
(multistage) residual coding, because it consumes too much bits while
leaving still some artifacts.Instead, it seems better to simply have a
finer quantization (and use a quantization step of 4 instead of 8).For the
higher compression settings, I will need a good "psycho-visual"
quantization to know which wavelet coefficients remove/diminish.

I have also new ideas for the compression (entropy coding) scheme.Right
now, my compression scheme for the wavelet coefficients is equivalent to
ZPAQ -method 4 (but very faster).With my new ideas, I think I can improve
compression by 1-2%, which will put my scheme on the same level as ZPAQ
-method 5!

I haven't made much research, but I would also need better YUV420 2x2
up/down-sampling functions (mines are too basic and not really good).If you
would know good ones, do not hesitate to contact me.Maybe a Daala developer
could point me to the ones used for Daala still image?

With all this, I think I could make a 0.1.4 version, but I don't know when
it will be released.

Cheers,
Raphael
-------------- next part --------------
An HTML attachment was scrubbed...
URL: http://lists.xiph.org/pipermail/theora/attachments/20160316/e79cde69/attachment.htm