[Ffmpeg-devel] [RFC, Request for testing, PATCH] snow wavelet fliping

[Michael Niedermayer]

Hi

On Tue, Jan 02, 2007 at 06:33:44PM +0100, Stefan de Konink wrote:
> On Tue, 2 Jan 2007, Michael Niedermayer wrote:
> 
> > the attached patch rotates the residual by 180? for every 2nd frame
> > before performing the wavelet transform (!only encoding is implemented!)
> 
> > and comments about this are even more welcome
> 
> Could you tell me, maybe 'us', why this works? And maybe what the results
> are for flipping the residual?

lets see
example signal is
  0   0   1   2  -6   2   1   0
the 5/3 wavelet transform of that is
 -1  -4  16  36 -30  36  16  -4   /8

if now we flip the signal
  0   1   2  -6   2   1   0   0
then the 5/3 wavelet transform of that is
  0   0   0 -64   0   0   0   0   /8


-> some signals simply can be represented much more compactly in the
   wavelet domain if they are shifted a little

if we now always encode images with the same shift then there is a much
higher chance that some signals will be always hard to store, if OTOH
we flip the wavelet after each frame then what cannot easily be stored
(= gets quantized to 0) in the even frames can with some luck more easily
be stored in the odd frames

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

If you really think that XML is the answer, then you definitly missunderstood
the question -- Attila Kinali
-------------- next part --------------
A non-text attachment was scrubbed...
Name: not available
Type: application/pgp-signature
Size: 189 bytes
Desc: not available
URL: <http://lists.mplayerhq.hu/pipermail/ffmpeg-devel/attachments/20070102/f61159b0/attachment.pgp>