[Ffmpeg-devel] Questions about snow

Michael Niedermayer michaelni
Thu Nov 30 19:41:49 CET 2006


Hi

On Thu, Nov 30, 2006 at 12:25:46PM -0500, Rich Felker wrote:
> On Thu, Nov 30, 2006 at 02:46:51PM +0100, Luca Abeni wrote:
> > Hi all,
> > 
> > I'll probably have to work on wavelet-based video codecs (well, I'll
> > have to work on scalable video encoding, and I've been told that wavelet
> > are inherently scalable), and I am thinking about using the snow codec
> > for performing some experiments.
> 
> Wavelets are scalable, but motion comp is not. Making a motion comp
> system that correctly commutes with scaling would be an interesting
> project but I'm not sure if it can be done..

there are various papers about this on citeseer ...
one way is to simply do MC in wavelet domain so that MC is simply done
within each subband the quality is shit though IIRC
the problem simply is that if you shift a image then some signal which
could be entirely represented within one subband will leak VERY significantly
into the other bands but MC as defined above cant deal with that ...


the other more generic way is to simply take a lower resolution video
and encode the high resolution stuff relative to either the scaled up
low resolution video or relative to the last frame with MC, then switch
this blockwise based on rate distortion, the upscaling + difference stuff
could be a normal inverse wavelet or some non linear directional scaling
or whatever you like ...
this can be done with pretty much any codec be oit wavelet, block dct, or
even some vector quantizer or whatever

[...]
-- 
Michael     GnuPG fingerprint: 9FF2128B147EF6730BADF133611EC787040B0FAB

In the past you could go to a library and read, borrow or copy any book
Today you'd get arrested for mere telling someone where the library is




More information about the ffmpeg-devel mailing list