[theora] Fixed Quantizer - Fixed Quality

[Christoph Lampert]

On Wed, 26 Mar 2003, Dan Miller wrote:

> > From: Marco Al [mailto:marco at simplex.nl]
> ...
> > I think the general assumption was that you meant quantizer by Q, not
> > quality. Christoph most certainly means quantizer with Q.
> 
> Fair enough.  I guess then my point is that offering some sort of raw
> 'Quantizer' knob to an end user of a codec is a baad idea.  The user
> usually wants to go for maximum quality M (Q could be confusing),
> limited to peak datarate P, with average datarate D.  These are the
> sorts of knobs a good codec should be presenting to the world. --- >8

Yes, sorry for the confusion. I have no "quality" setting to work with,
because of the usual question: What is "quality"?

Of course, fixed quality would be optimal, but then the question is, "how
to we achieve it?"

I see four possible way to control data flow:

1) Fixed Bitrate/CBR (one or twopass). This will not yield to optimal
   quality. It will yield to fixed bitrate, just as it's intended. Twopass
   for constant bitrate is not common for AVI codecs. 

The rest are VBR methods: 

2) Fixed Quantizer (no 2-pass needed). This will not yield to optimal
quality (my example was meant to show this), but it's the easiest to test
with, because no "intelligence" (ratecontrol/quality control) is tested at
the same time. 

3) Fixed Distorsion (two-pass, or one with side information): It's easy to
control compression to achieve fixed PSNR (within a margin of error, of
course), just like CBR. But this does not yield to "fixed quality", 
because of psychovisual masking effects. 

4) Fixed Quality (twopass, or one with side information). This is what
everybody wants, but I don't know of any "real" way to achieve this. Many
very advanced two-pass methods try to, but since nobody can agree what
"quality" is, and even less people know how to measure it (no, not by
PSNR).

When benchmarking with 1) or 4), the "control" algorithms is
tested together with the compression of the codec itself. Using 3) for 
PSNR distorsion isn't very helpful, a better quality metrics would be
needed to make 3) into 4). Testing 2) in combintion with the
resulting PSNR tells at least something. 

Christoph 

<p>--- >8 ----
List archives:  http://www.xiph.org/archives/
Ogg project homepage: http://www.xiph.org/ogg/
To unsubscribe from this list, send a message to 'theora-request at xiph.org'
containing only the word 'unsubscribe' in the body.  No subject is needed.
Unsubscribe messages sent to the list will be ignored/filtered.