[MPlayer-users] lanczos or bicubic spline or..?

[D Richard Felker III]

On Tue, Feb 03, 2004 at 10:56:52PM -0800, * wrote:
> --- D Richard Felker III <dalias at aerifal.cx> wrote:
> > On Sun, Feb 01, 2004 at 11:11:05PM -0800, * wrote:
> > > Thanks (again) Rich!  I think I'm a little less confused now.  I don't
> > > know if this is complete nonsense or not, but I ran a test upscaling 10
> > > times in a row to compare:
> > >
> > > mplayer movie.vob -nosound -vo png -vf crop=716:452:0:12,
> > > scale=720:384,scale=736:400,scale=768:416,scale=800:432,scale=832:448,
> > > scale=864:464,scale=896:480,scale=912:496,scale=944:512,scale=976:528
> > > -sws #
> >
> > IMO this is not a legitimate test. When encoding a movie you don't
> > scale 8 times, and you especially should never upscale. Try it with
> > just _one_ scale filter, to the size you'll actually be encoding, and
> > compare the results.
> >
> > Also, compare the results _after_ encoding to mpeg4, since that's what
> > will ultimately matter.
> >
> Okay, thanks for walking me through this.  With one scale filter there
> are the following PSNR:
> 
> mencoder movie.vob -nosound -ovc lavc -lavcopts psnr:mbd=2:mv0:trell:
> cbp:precmp=2:cmp=2:subcmp=2:vmax_b_frames=1:predia=2:dia=2:preme=2:
> vme=4:v4mv:vbitrate=875 -vf scale=976:528 -ss 360 -frames 360 -sws #

Why on earth are you upscaling your movie to encode it?? This will
only hurt the quality. Either leave the size alone and encode aspect
(but broken players won't support this), or scale it down to a
reasonable size. There's no way you'll get remotely tolerable quality
at 875 kbit/sec for 976x528 video. You'd need at least 2mbit. The only
reason the first few frames look ok is that the rate control hasn't
had time to kick in yet.

> 0 PSNR: Y:41.37, Cb:44.15, Cr:44.26, All:42.13
> 1 PSNR: Y:41.37, Cb:44.11, Cr:44.25, All:42.12
> 2 PSNR: Y:40.27, Cb:43.15, Cr:43.30, All:41.05
> 3 PSNR: Y:40.69, Cb:43.67, Cr:43.80, All:41.49
> 4 PSNR: Y:38.18, Cb:41.84, Cr:41.95, All:39.10
> 5 PSNR: Y:41.37, Cb:44.11, Cr:44.25, All:42.12
> 6 PSNR: Y:40.31, Cb:43.66, Cr:43.79, All:41.18
> 7 PSNR: Y:41.96, Cb:44.56, Cr:44.70, All:42.68
> 8 PSNR: Y:39.76, Cb:42.68, Cr:42.84, All:40.55
> 9 PSNR: Y:39.98, Cb:42.85, Cr:42.99, All:40.75
> 10 PSNR: Y:40.12, Cb:43.01, Cr:43.16, All:40.90

These come from using the psnr option to lavc?? Or...??

> And extracting 1 frame:
> 
> mplayer test.avi -nosound -vo png -frames 1
> 
> http://box.dnsalias.net/mexico/0.png
> http://box.dnsalias.net/mexico/1.png
> http://box.dnsalias.net/mexico/2.png
> http://box.dnsalias.net/mexico/3.png
> http://box.dnsalias.net/mexico/4.png
> http://box.dnsalias.net/mexico/5.png
> http://box.dnsalias.net/mexico/6.png
> http://box.dnsalias.net/mexico/7.png
> http://box.dnsalias.net/mexico/8.png
> http://box.dnsalias.net/mexico/9.png
> http://box.dnsalias.net/mexico/10.png
> 
> This seems to be pretty consistent.  Gaussian (7) gets highest PSNR and
> nearest-neighbor (4) lowest.
> 
> It is impossible for me to tell the difference between (2) and (6).  But

(2) and (6) behave identically on the luma, but treat the chroma
differently. (2) uses bicubic scaling for both while (6) uses simple
bilinear scaling for chroma. This is both faster and makes more sense
usually, since chroma samples are too far apart for bicubic
interpolation to be reasonable.

> I have to say that (9) looks a tiny bit sharper.  Don't you think?  Like
> there is more texture in the skin.

I don't notice any difference. Certainly not from a viewing distance.

> > The PSNR I'm talking about is not from the scale filters, but from the
> > mpeg4-encoder after the scaling takes place. What I said is that if
> > you give the mpeg4 encoder a sharper image, your resulting movie file
> > will have worse (lower) PSNR relative to the input it gets, than if
> > you give the mpeg4 encoder a more blurred image.
> >
> Okay.  So basically PSNR is completely useless for telling which of the
> -sws settings have a higher quality?  I don't want either (7) or (4).

There are many ways to measure "quality", which is a very subjective
term. One of them is by how much information is lost, i.e. how closely
you can recover the original image by inverting the scale operation.
Sinc will naturally win for this measure, but from a perceptual
standpoint, it looks very bad.

A second, more reasonable way to measure quality would be according to
how good it looks to you, but this is very hard to measure without
double-blind (or multi-blind) testing.

Finally, you need to take into account the quality _after_ encoding to
mpeg4. If you're going to keep increasing your bitrate until you're
happy with the quality, or use constant-qscale, then this doesn't
really matter much. But if you want the output to fit in a fixed size,
then keeping more fine detail will necessarily result in more ringing
and blocking artefacts.

If you're getting your PSNR's from lavc, they're measuring exactly one
thing: how close the encoded video matches the (already scaled) video
you're passing into it. They tell you how lossy the mpeg4 encoding is
when done after your scaling, not how lossy the scaling is.

> > IMO you should pick between 2 and 6. 6 probably makes more sense for
> > 4:2:0-sampled sources.
> >
> I am ashamed to say I do not know what is 4:2:0-sampled source?  You
> mean with hqdn3d?

Anything mpeg-type will be 4:2:0. It just refers to the way chroma
subsampling is done (half resolution in both horizontal and vertical
directions).

Rich