[MPlayer-users] Re: Crop before deinterlacing OK?

[Rich Felker]

On Mon, Jul 25, 2005 at 03:02:22PM +0200, Matthias Wieser wrote:
> Am Montag, 25. Juli 2005 14:43 schrieb Rich Felker:
> > On Mon, Jul 25, 2005 at 11:56:06AM +0200, Matthias Wieser wrote:
> > > Am Montag, 25. Juli 2005 00:48 schrieb Rich Felker:
> > > >
> > > > It has all the same disadvantages as lb, plus the disadvantages of
> > > > a sharpener (hurting compressibility and not improving quality).
> > >
> > > Maybe not improving psnr but I think it improves visual quality. I
> > > have done some tests with different test patterns and the
> > > deinterlaced result of pp=l5 is much more comparable to the original
> > > as the one from pp=lb.
> > >
> > > (btw.: when measuring the psnr between the original and the scaled
> > > down and encoded video, sws=2 or sws=6 often results in higher PSNR
> > > than sws=1 or sws=7)
> > >
> > > > It's
> > > > exactly the same thing as running pp=lb followed by a simple
> > > > -1,4,-1 sharpening filter.
> > >
> > > But as pp=lb by the very nature of the system introduces some
> > > vertical blurring this sharpening might improve the overall quality.
> >
> > If you weren't going to encode, yes. But a sharpening filter will just
> > hurt compressibility and lower the psnr of the actual encode.
> 
> Do you always apply an gaussian blur filter before encoding because 
> mencoder -lavcopts [...]:psnr will then show a higher psnr? ;-)

No, obviously this is bogus reasoning, but actually it will improve
quality at a fixed bitrate unless the bitrate is extremely high.

> The PSNR between the original frame and a pp=l5 processed frame is 
> significantly higher than the PSNR between the original and a pp=lb 
> processed frame. So why not encode the frames which are nearer to the 
> original?

Neither is anywhere near the original if your original is interlaced.
On the other hand, if the original is progressive, both will change
the image, and then pp=l5 will also result in worse psnr during
compression, so you take two quality hits.

> > The ghosting you see is 
> > just because pp=lb created a nasty ghost. It's visible on CRT too.
> 
> The ghosting I see is the ghosting from pp=lb and the ghosting from my 
> TFT. As the ghosting from pp=lb lowers contrast of moving edges, the 
> ghosting from my TFT increases.

To me it sounds like nonsense. I agree this happens, but I don't think
it's anywhere near enough to be the cause unless you have a really
shitty TFT. I'll believe it when someone does some proper tests...

Rich