[theora] NHW Codec - some progress

[Raphael Canut]

Hello,

I have some progress for the NHW codec.

I recently talked with a researcher that has some interest in the NHW
codec.He has studied several directional wavelet transforms (notably HWD
directional transform), and he came to the conclusion that directional
transforms will bring marginal improvement, not significant in the case of
my compression codec.This also confirms what I have read from the Rududu
author who tested directional wavelets in his codec without luck.

I have potentially found a very good candidate for the YUV420 2x2
upsampling.I haven't studied the impact of this processing on the codec,
but I believe that better downsampling and upsampling algorithms for U and
V components will bring better quality to the codec.Here is the upsampling
I'm thinking of, the demo page is very impressive:
https://kdepepo.wordpress.com/2012/08/15/reverse-antialiasing-for-image-scaling/
.

I have also (yet) another idea for the quantization/compression scheme to
gain compression and create good lower quality settings.I will try to code
and test it this month.

To finish, I maybe get used to notice neatness/sharpness while evaluating
image quality, and not enough precision (and that's certainly why I prefer
the results of the NHW codec in 50-55% of the cases than x265).The SMPTE
recommendation for visual quality review is to make it with at least 30
subjects, to have a significant average result.I don't have the means to
conduct a study on 30 subjects, but maybe I can call the community.If we
have 30 members that have tested the NHW codec that give their opinion,
maybe we could have a first but solid idea of its quality.

So again, all the developers that have tested the NHW codec are very
welcome to give me their opinion, would be so much helpful!

Many thanks!
Cheers,
Raphael
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://lists.xiph.org/pipermail/theora/attachments/20170307/ecedc958/attachment.html>