NHW Project - new -l17 "extreme" compression quality setting

--===============2274610573220278808==
Content-Type: multipart/alternative; boundary="000000000000f8f8db0581031e2f"

--000000000000f8f8db0581031e2f
Content-Type: text/plain; charset="UTF-8"

 Hello,

For those interested, I have added a new -l17 "extreme" compression quality
setting to the NHW Project.I find this setting competitive with x265
(HEVC).Actually I have tested -l17 on 15 rather good quality images, and on
these images I prefer the results of the NHW Project.

We can still save 2.5KB per .nhw compressed files.

More at: http://nhwcodec.blogspot.com/ .

I start wondering if for extreme compression I should add a post-processing
function in the decoder that would remove some aliasing for some images,
just like HEVC intra has post-processing functions like deblocking and SAO
filters.

As detecting aliasing seems quite difficult, would it be possible to use
for example machine learning, that would be trained to detect aliasing
patterns?

But first, do you find the results of the NHW Project at extreme
compression good in their current state? Or should they be post-processed
with aliasing and other?

Any feedback is very welcome!

Cheers,
Raphael

--000000000000f8f8db0581031e2f
Content-Type: text/html; charset="UTF-8"

<div dir="ltr">
Hello,<br>
<br>
For those interested, I have added a new -l17 &quot;extreme&quot; compression 
quality setting to the NHW Project.I find this setting competitive with 
x265 (HEVC).Actually I have tested -l17 on 15 rather good quality 
images, and on these images I prefer the results of the NHW Project.<br>
<br>
We can still save 2.5KB per .nhw compressed files.<br>
<br>
More at: <a href="http://nhwcodec.blogspot.com/" target="_blank">http://nhwcodec.blogspot.com/</a> .<br>
<br>
I start wondering if for extreme compression I should add a 
post-processing function in the decoder that would remove some aliasing 
for some images, just like HEVC intra has post-processing functions like
 deblocking and SAO filters.<br>
<br>
As detecting aliasing seems quite difficult, would it be possible to use
 for example machine learning, that would be trained to detect aliasing 
patterns?<br>
<br>
But first, do you find the results of the NHW Project at extreme 
compression good in their current state? Or should they be 
post-processed with aliasing and other?<br>
<br>
Any feedback is very welcome!<br>
<br>
Cheers,<br>
Raphael

</div>

--000000000000f8f8db0581031e2f--

--===============2274610573220278808==
Content-Type: text/plain; charset="utf-8"
MIME-Version: 1.0
Content-Transfer-Encoding: base64
Content-Disposition: inline

X19fX19fX19fX19fX19fX19fX19fX19fX19fX19fX19fX19fX19fX19fX19fX18KdGhlb3JhIG1h
aWxpbmcgbGlzdAp0aGVvcmFAeGlwaC5vcmcKaHR0cDovL2xpc3RzLnhpcGgub3JnL21haWxtYW4v
bGlzdGluZm8vdGhlb3JhCg==

--===============2274610573220278808==--