[Ffmpeg-devel] [Fwd: [CinCVS] New anti-patented algorithm for better Y'CbCr]

Michael Niedermayer michaelni
Sun Mar 18 19:51:18 CET 2007


Hi

On Sun, Mar 18, 2007 at 01:29:20PM -0500, Rich Felker wrote:
> On Sun, Mar 18, 2007 at 05:46:57PM +0100, Stefan de Konink wrote:
> > -----BEGIN PGP SIGNED MESSAGE-----
> > Hash: SHA512
> > 
> > This message was on the Cinelerra mailinglist, maybe some devs are
> > interested in the methods used.
> > 
> > 
> > Stefan
> > 
> > - -------- Original Message --------
> > 
> > I think I have already sent you an e-mail about Twibright Hyperluma. I
> > have developed two even better algorithms that reach amazing SNR
> > concerning colour detail distortion when encoding into 4:2:0 or 4:2:2
> > Y'CbCr.
> > 
> > The are called Hyperluma 2 and Luminaplex. See
> > http://ronja.twibright.com/hyperluma.php
> 
> Just from a casual reading of the page, it sounds stupid. The author
> somehow things a 12.96 megabyte lookup table in his implementation
> "improves performance", which signals incompetence right away.
> 
> Also, it sounds like what he's trying to compensate for his crappy
> devices that use nearest-neighbor for chroma resampling. Obviously
> this is idiotic to begin with. If you want high quality output, use a
> proper resampler..

same feeling here ...


> All of this is based on a very casual reading; the page was long and
> ill-organized so that I couldn't quickly figure out what problem the
> author claims to be solving. If you have a better idea what it's about
> please fill me and others in on it.

my feeling is that its about 2 things
1. a decoder with nearest neighbor upsampler and encoder with missmatching
   downsampler are less optimal then a encoder which tries to select values
   so the error after the decoder is minimized (trivial linear algebra)
2. normal error meassure is wrong and the author rather uses gamma corrected
   RGB to calculate errors which certainly has some physical truth to it but 
   as humans instead of photon counters watch the video iam a little
   suspicious how correct this arguing is, the examples shown at least 
   dont look better to me also human brightness perception isnt linear ...

[...]

-- 
Michael     GnuPG fingerprint: 9FF2128B147EF6730BADF133611EC787040B0FAB

I am the wisest man alive, for I know one thing, and that is that I know
nothing. -- Socrates
-------------- 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/20070318/6333b26d/attachment.pgp>



More information about the ffmpeg-devel mailing list