mjpeg2000 vs H264

pieter1976 · 16th March 2005, 15:21

I found a comparison of mjpeg2000 with H264 which is rather interesting. It seems that Mjpeg2000 works better for 1920x1080 video material.

Here is the link

pieter1976 · 16th March 2005, 15:31

I also have been working on a postprocessing filter specialy for Jpeg2000 files. It has a positive effect on the PSNR values.
Is anybody interrested in filters like this?

thegeby · 16th March 2005, 15:40

The comparison seems to be purely by quality. Neither speed, nor computing resources, seems to have been a factor. It does not seem surprising that a compression algorithm designed for individual pictures perform better, quality-wise. By photographic standards, 1920x1080 would after all be considered rather crummy....

Anyhting that pushes the quality envelope should be welcomed, however

Tommy Carrot · 16th March 2005, 16:31

I'm not sure if you noticed that it was a keyframe-only comparison, so it doesn't tell anything about the video coding performances, only that jpeg2k has higher PSNR rate than the h.264 keyframes.

drmpeg · 17th March 2005, 09:20

Here's a newer paper that compares JPEG2000 with H.264 FRExt on 720p
images. Seems the 8x8 transform helps quite a bit.

http://ftp3.itu.ch/av-arch/jvt-site/...JVT-N010d1.zip

Ron

pieter1976 · 17th March 2005, 20:10

Here is a postprocessing filter that improves the PSNR levels.
link

Does a similar opensource projects exist?

pest · 18th March 2005, 14:20

it's easy to code a i-frame only wavelet
coder which kicks jpeg2000s ass...
you can add a sbr-like algorithm to the
dequantization to further improve the texture quality
the problem is a good motion compensation
and current state of the art techniques are just
a way too complex..but reference coders beat h264 psnr-wise
just wait...

pest

edit: spelling

16th March 2005, 15:21	#1 \| Link
pieter1976 Registered User Join Date: Jan 2004 Location: Netherlands Posts: 85	mjpeg2000 vs H264 I found a comparison of mjpeg2000 with H264 which is rather interesting. It seems that Mjpeg2000 works better for 1920x1080 video material. Here is the link Last edited by pieter1976; 18th March 2005 at 19:39.

17th March 2005, 20:10	#6 \| Link
pieter1976 Registered User Join Date: Jan 2004 Location: Netherlands Posts: 85	Here is a postprocessing filter that improves the PSNR levels. link Does a similar opensource projects exist? Last edited by pieter1976; 18th March 2005 at 22:50.

18th March 2005, 14:20	#7 \| Link
pest Registered User Join Date: Feb 2005 Posts: 39	it's easy to code a i-frame only wavelet coder which kicks jpeg2000s ass... you can add a sbr-like algorithm to the dequantization to further improve the texture quality the problem is a good motion compensation and current state of the art techniques are just a way too complex..but reference coders beat h264 psnr-wise just wait... pest edit: spelling Last edited by pest; 18th March 2005 at 16:25.

16th March 2005, 15:31	#2 \| Link
pieter1976 Registered User Join Date: Jan 2004 Location: Netherlands Posts: 85	I also have been working on a postprocessing filter specialy for Jpeg2000 files. It has a positive effect on the PSNR values. Is anybody interrested in filters like this?

16th March 2005, 15:40	#3 \| Link
thegeby Registered User Join Date: Feb 2004 Posts: 107	The comparison seems to be purely by quality. Neither speed, nor computing resources, seems to have been a factor. It does not seem surprising that a compression algorithm designed for individual pictures perform better, quality-wise. By photographic standards, 1920x1080 would after all be considered rather crummy.... Anyhting that pushes the quality envelope should be welcomed, however

16th March 2005, 16:31	#4 \| Link
Tommy Carrot Registered User Join Date: Mar 2002 Posts: 864	I'm not sure if you noticed that it was a keyframe-only comparison, so it doesn't tell anything about the video coding performances, only that jpeg2k has higher PSNR rate than the h.264 keyframes.

17th March 2005, 09:20	#5 \| Link
drmpeg Registered User Join Date: Jan 2003 Location: Silicon Valley Posts: 455	Here's a newer paper that compares JPEG2000 with H.264 FRExt on 720p images. Seems the 8x8 transform helps quite a bit. http://ftp3.itu.ch/av-arch/jvt-site/...JVT-N010d1.zip Ron