[MPlayer-users] Comparison of different deinterlace and denoise filters

[D Richard Felker III]

On Wed, Oct 15, 2003 at 11:21:29AM +0200, Matthias Wieser wrote:
> > > - pp=tn affects color ?!
> >
> > This filter is brain damaged by design and should never be used.
> 
> I know, but it's fast and because of it's description in "mplayer 
> -pphelp", some people will use it.

:((

> > It
> > can and will ADD blocking artifacts to your movie (since it switches
> > itself on and off on a per-block basis).
> 
> Woult it help, if the tn filter would fade out the denoising near the 
> edges of the blocks?

Yes. Want to send a patch? :)

> > > - pp=md  looks similar to pp=fd
> I found, that md produces jagged edges on interlaced parts. Does md use no 
> filter after deinterlacing?

Right, no filter. You can try pp=md,pp=lb for better results. Or use
the unsharp filter.

> > This is true for your test case. Not true at all in general. pp=fd
> > will create ghosts for the same reason pp=lb does, however its ghosts
> > come in the form of outlines rather than solid ghosts. Try a scene
> > with motion.
> 
> I think, this effect is only visible, if you know, what you are looking 
> for.

Try it on anime. The black outlines with ghost VERY BAD.

> Where can I find more information about how pp=md and pp=fd work?

pp=md replaces a pixel on an odd line by the median of itself, the
pixel above it, and the one below it.

pp=fd uses a convolution (5 point i think) to replace pixels on odd
lines based mostly on the surrounding even lines.

> > > - pp=lb is horrible for fast movements or moving text
> >
> > Agree mostly. Basically all deinterlacers are horrible...
> 
> 50Hz (60Hz) Avis rock! ;-)

;)

BTW, what card did you use to capture your samples? The chroma looked
rather wrong in many of them. If you use bttv, you MUST capture
interlaced content in yuy2 mode; the yv12 is broken and throws away
the odd field of chroma! :(

Rich