[FFmpeg-devel] [PATCH] avfilter/vf_dctdnoiz: add YUV444P support

[Clément Bœsch]

On Mon, Aug 29, 2016 at 10:00:31PM +0200, Paul B Mahol wrote:
> On Mon, Aug 29, 2016 at 8:52 PM, Clément Bœsch <u at pkh.me> wrote:
> 
> > On Mon, Aug 29, 2016 at 07:00:54PM +0200, Paul B Mahol wrote:
> > > Hi,
> > >
> > > patch attached.
> >
> > The color decorrelation is an important part of the denoising algorithm
> > described in http://www.ipol.im/pub/art/2011/ys-dct/ (see 2.3)
> >
> > Can we really consider the YUV planes as properly decorrelated ones? It
> > seems to me that allowing YUV444 means that sometimes the denoising will
> > be different (randomly if picked).
> >
> > --
> > Clément B.
> > _______________________________________________
> > ffmpeg-devel mailing list
> > ffmpeg-devel at ffmpeg.org
> > http://ffmpeg.org/mailman/listinfo/ffmpeg-devel
> >
> 
> They also say that gray should not be (de)correlated at all.

in the context where there is a single plane, so of course you don't have
anything to decorrelate with.

> Y is basically gray.
> U/V is already correlated with Y.
> 

U and V are correlated with Y, so with the decorrelation you introduce
which is a noop for YUV (IIUC), you will loose the benefit of having the
"PSNR improvement from about 1 to 3 dB".

> Those number to me look random, they should be set via filter options.

you mean the 3-point DCT basis?

-- 
Clément B.