[MPlayer-dev-eng] [PATCH] Convolution deinterlacer

[D Richard Felker III]

On Sat, Mar 29, 2003 at 02:43:03PM +0100, Michael Niedermayer wrote:
> Hi
> 
> On Saturday 29 March 2003 07:59, D Richard Felker III wrote:
> > On Sat, Mar 29, 2003 at 12:47:36AM -0500, James Bostick Crowson wrote:
> [...]
> 
> >
> > > Because of the number of problems with this patch that I don't know how
> > > to fix, I understand if it's not applied, but I would appreciate if
> > > someone were to go through and fix them, since I believe the quality
> > > really is better in many cases.
> >
> > I think it should be applied, after these problems are fixed... Like I
> > said, I don't think it's useful for anime, but it might be very nice
> > for real interlaced material. The other deinterlacers rather suck...
> all linear filter type deinterlacers suck, the correct solution is to do some 
> moton estimation & compensation before mixing the fields
> 
> anyway i just reimplemeneted the -1 2 6 2 -1 type lowpass deinterlacer with 
> mmx, fixed cliping & no line shift ...

Michael, have you looking at my inverse telecine (detc) filter at all?
This isn't entirely related, but I'm curious if you have any
suggestions on how to make it work better. I keep thinking it's ok,
then finding material that it breaks on, then fixing it, then finding
that my fix made it oversensitive and that it broke other stuff, etc.

It would all be nice and simple if we were dealing with uncompressed,
freshly-telecined frames, but lots of movies are encoded horribly and
have tons of quantization noise, making it hard to tell when fields
from successive frames are identical...

Rich