WAvelet codec

[trican]

Anyone care to comment on the following recent article:

"Codec from Canada, CRC-WVC, outperforms H.264 video with wavelets "
http://www.videsignline.com/showArticle.jhtml?printableArticle=true&articleId=196602553

From the conclusions section
..."In terms of PSNR and bit rates, the wavelet-based video codec at its current stage of development performs as well as, or better than, H.264 High Profile. The wavelet-based codec is usually more efficient for video sequences with a high spatial resolution because the wavelet transform can remove the correlation between pixels within a large region. In addition, it inherently provides flexible scalability feature. For similar or even lower PSNR's, the wavelet-based video codec generally provides a better subjective picture quality than H.264. H.264 suffers from blocking artifacts to which the human visual system is very sensitive. The wavelet-based video codec described in this article represents a viable alternative to H.264. It is still at an early stage of development. A number of techniques, such as intra prediction, R-D optimization, and advanced entropy coding, may significantly increase the compression efficiency of the codec."

[Sirber]

it's BS until they beat x264 with a real encoder, not papers.

[trican]

Granted thats true. Though, speaking more generally I wouldn't be overly surprised to see wavelet transforms becoming increasingly common over (variants of) the DCT in future codecs. If even only as a way to eliminate the blocking artifacts and the assoicated computational burden of deblocking filters in hybrid codecs.

Does anyone know how (poorly?) Dirac fairs against H.264?

[akupenguin]

Wavelets can only keep the video block-free, they can't deblock a video that has blocks due to motion compensation. As such, all wavelet codecs I know of use obmc or mesh or wavelet-domain mc or some other advanced form of motion compensation. This may look better, but it doesn't reduce computational complexity compared to a deblocker.
Wavelets also do not inherently provide much more scalability for video than dct does. You can drop the high frequency wavelet subbands to get a lowres picture, or you can drop the high frequency dct coefficients for a similar effect. That's fine for still images, but the hard part of video scalability is in the motion compensation: Either you use the whole picture for motion compensation in which case any scaled-down verion will have artifacts accumulate over time, or you use only the base layer which loses some compression efficiency compared to the non-scalable version, or some similar compromise.

[PatchWorKs]

Just a question: any news about dirac ?

[celtic_druid]

http://dirac.cvs.sourceforge.net/*checkout*/dirac/compress/ChangeLog