[MPlayer-users] Re: [-] Question on converting to DivX4...

[Rich Felker]

On Mon, Jul 11, 2005 at 08:12:13PM -0700, RC wrote:
> On Mon, 11 Jul 2005 18:28:29 +0200
> Matthias Wieser <mwieser at gmx.de> wrote:
> 
> > Just take a test video consisting of alternating black and white 
> > horizontal lines. Without pp=lb you will see horizontal lines, with
> > pp=lb you will see a solid grey colour. => Information is lost.
> 
> I realize information is lost by the bluring pp=lb does, but that
> doesn't equate to 50%.

ok, you can compute the amount if you like.. i've been too lazy to do
it so far, but if you like i can try...

> Hmm, how about an actual test?  Assuming a 480i source, compare "- vf
> scale=:240" with "-vf pp=lb". The latter looks noticably better.

scale=:240 loses more than half the information with any scaler except
sinc. anyway it's all a matter of what your content is. there are
images that are completely undamaged by pp=lb (for instance solid
color image :) and there are images that are totally destroyed (the
horizontal lines example). the only good way i know to measure the
destructiveness is with the size of the (algebraic) kernel.

with that aside, i can assure you that temporal blending is very
destructive from any practical standpoint. i've been working on a
deblending filter to repair a really nasty blended-telecined dvd
source, and it's incredibly difficult, even when most of the necessary
information is there. your case is much worse. see, when you blend 2
fields together, the information lost is not the details of the actual
fields (altho some of that is lost too), but rather which information
goes with which field. you've reduced 60fps to 30fps in an essentially
irreparable way..

rich