[MEncoder-users] broken mencoder options (was: High quality encoding of anime and broken mencoder options)

Jorge Peixoto de Morais Neto please.no.spam.here at gmail.com
Mon Mar 10 17:34:44 CET 2008


On Sun, 09 Mar 2008 23:25:20 -0700
RC <cooleyr at gmail.com> wrote:

> On Sun, 9 Mar 2008 15:32:39 -0300
> Jorge Peixoto de Morais Neto <please.no.spam.here at gmail.com> wrote:
> 
> > 3.2) If they actually increase the video quality, but diminish PSNR
> > (which is possible but unlikely), it should be noted in the
> > documentation
> 
> PSNR being contrary to perception is NOT unlikely, in fact it's very
> common.  See nr, *mask, qns, etc.  Even cmp=2 is a good example, as
> it makes live action quite blocky, yet (usually) improves PSNR.
> 
> A warning that PSNR isn't perfect would be pointless.  The only
> real question is: What gave you the idea that it is?
PSNR is certainly not perfect. What it tells you is that if
video A has higher PSNR than video B than there is a big
*chance* that video A is higher quality(1) than video B.

The documentation *does* point that qns and *mask are not good for
increasing PSNR. The reason the documentation includes this
information is (probably) because this information is not what we
expect - we expect that better quality results in higher PSNR.
If *cmp=2 also has this behavior, than it should be pointed out in the
documentation, in my oppinion.

1: For me, I want the encode to preserve information, not to just "look
good". Imagine a blurring filter that can make the image more
compressible and therefore reduce blocking for a given bitrate. I would
prefer not to use the filter, because I want to preserve information,
not just to make the image "look good". For example, if there is fine
print in the video, I want it to be readable. And for the low bitrates
I want, I cannot have the video "look good, without blocking" and
preserve fine detail at the same time, so I tolerate blocking. In fact,
in my case (one who wants to preserve high detail) I think the PSNR is
a good metric.

By the way, something I like in the Theora codec is that it
prefers to preserve fine detail than to eliminate blocking - or at
least it behaved like this in the last video I tested. Since I care
more about detail than about about blockiness, I like the Theora
approach.

-- 
Software is like sex: it is better when it is free. --Linus Torvalds



More information about the MEncoder-users mailing list