[MPlayer-users] which deinterlace filter

[D Richard Felker III]

On Sun, Oct 12, 2003 at 12:53:39PM +0200, Matthias Wieser wrote:
> [Automatic answer: RTFM (read DOCS, FAQ), also read DOCS/bugreports.html]
> Am Freitag, 10. Oktober 2003 22:17 schrieb D Richard Felker III:
> 
> > > 'lb' is better, as it doesn't have the problems listed above, but it
> > > has some of it's own.  It needs both fields to match-up, exactly, so
> > > it is quite prone to producing 'ghosts' either if your interlacing
> > > isn't perfect, or if you have a telecine signal that the detelecining
> > > filter doesn't fix 100%.
> >
> > Agree.
> 
> ??
> Isn't "lb" this "linear blend" filter which only blends both fields into a 
> one new?
> I think it's one of the worst because
> - vertical resolution is decreased

Incorrect. The vertical blend filter is either invertible, or almost
invertible (depending on what it does at the top and bottom
boundaries), so either no resolution is lost, or at most one dimension
of resolution is lost.

On the other hand, pure interpolative filters (li, ci), almost-pure
interpolative ones (fd), and hybrid ones (md) lose a LOT of
resolution! Other than md, they throw away half the lines! With md the
results are much less predictable; it will only throw away local
minima and maxima, but there can be lots of those in a picture with
high frequency content or high motion combing. It can throw away
anywhere from nothing to 50% of the image.

> - moving objects become heavily blurred

Yes. Linear blend is HORRIBLE when applied to telecined video since
you're blending two images from 1/24 sec apart, which could be very
different! On the other hand, with true interlaced NTSC video, fields
are only separated by 1/60 sec in time, so blurring will be much less
significant. Personally, I find the slight blurring to be a very small
price to pay to avoid losing half the resolution in stills.

Rich