[prev in list] [next in list] [prev in thread] [next in thread] 

List:       theora
Subject:    [theora] NHW Project - lower quality settings - high compression
From:       Raphael Canut <nhwcodec () gmail ! com>
Date:       2017-12-11 15:34:03
Message-ID: CAKE58qEN-uVr+m=-YwYbz1rGoSx+TCqj6DDWR1E2uvgUCFhKWg () mail ! gmail ! com
[Download RAW message or body]

[Attachment #2 (multipart/alternative)]


Hello,

I was testing quickly some other lower quality settings -l3, -l4 because
high compression is what really lacks to the NHW codec.

In fact I can remove -5Ko of residual coding on Y,U,V comps for -l3 and I
can remove -10Ko always of residual coding for -l4, and -l3, -l4 settings
will be really better than the current ones implemented.

I can also increase the wavelet coefficients threshold (of the dead zone),
currently it is 8, can increase to 9,10,11,12,... according to the
compression wanted.The problem with this operation is that I will increase
aliasing and it will become no more acceptable... I have coded a
post-processing function in the NHW decoder that is supposed to remove
aliasing but it doesn't work at all (and I still haven't removed
it...).Would you know an efficient processing, algorithm that removes
aliaising? It would help me a lot!!!

Because the next quality settings could be here: -l3,-l4,-l5,-l6 with
-5Ko,-10Ko,-15Ko,-20Ko corresponding roughly to 1.5bpp->1bpp (for image not
video) with the processings explained above.

Then for the next settings, I could start to increase quantization of the
DC parts, and have a better analysis of wavelet coefficients, this is the
most difficult part.

So again, I think it will finally not be so difficult to go down to 1bpp
but it implies to have a post-processing that removes aliasing? What do you
think of this approach?

Any help is very welcome!

Cheers,
Raphael

[Attachment #5 (text/html)]

<div dir="ltr"><div><div><div><div><div><div><div><div><div>Hello,<br><br></div>I was \
testing quickly some other lower quality settings -l3, -l4 because high compression \
is what really lacks to the NHW codec.<br><br></div>In fact I can remove -5Ko of \
residual coding on Y,U,V comps for -l3 and I can remove -10Ko always of residual \
coding for -l4, and -l3, -l4 settings will be really better than the current ones \
implemented.<br><br></div>I can also increase the wavelet coefficients threshold (of \
the dead zone), currently it is 8, can increase to 9,10,11,12,... according to the \
compression wanted.The problem with this operation is that I will increase aliasing \
and it will become no more acceptable... I have coded a post-processing function in \
the NHW decoder that is supposed to remove aliasing but it doesn&#39;t work at all \
(and I still haven&#39;t removed it...).Would you know an efficient processing, \
algorithm that removes aliaising? It would help me a lot!!!<br><br></div>Because the \
next quality settings could be here: -l3,-l4,-l5,-l6 with -5Ko,-10Ko,-15Ko,-20Ko \
corresponding roughly to 1.5bpp-&gt;1bpp (for image not video) with the processings \
explained above.<br><br></div>Then for the next settings, I could start to increase \
quantization of the DC parts, and have a better analysis of wavelet coefficients, \
this is the most difficult part.<br><br></div>So again, I think it will finally not \
be so difficult to go down to 1bpp but it implies to have a post-processing that \
removes aliasing? What do you think of this approach?<br><br></div>Any help is very \
welcome!<br><br></div>Cheers,<br></div>Raphael<br></div>


[Attachment #6 (text/plain)]

_______________________________________________
theora mailing list
theora@xiph.org
http://lists.xiph.org/mailman/listinfo/theora


[prev in list] [next in list] [prev in thread] [next in thread]