[vorbis-dev] video codec

[Steve Underwood]

Marco Al wrote:
> 
> From: "Steve Underwood" <steveu at coppice.org>
> 
> > I have a Bugs Bunny cartoon. Its a bit old and noisy. I put it through a
> > codec that tends to glide over the noise. Those big areas of noisy
> > colour, which should be bland and textureless, have become smoother. The
> > result might actually look better than the original. I am happy.
> >
> > Now I come to a natural history film. A close up of a cat with mottled
> > fur. Its a moving, and the picture is a mass of rapidly changing detail
> > - a lot like noise. The codec suppresses it. Now I'm not so happy.
> 
> Thats not the point, its the user's responsibility to present the image as he
> would like them compressed. Suggesting using a noise polluted source to test is
> doing things backwards... if it removes them its ok and if it doesnt remove them
> its still ok, well that was a usefull test. Test with an image sequency which
> naturally has high frequency components.

One last go, before I shut up.

What you say is really no answer. The user seldom has control. If I have
a noisy movie to compress, that is what I have to compress. Clean it up?
Well, I can filter a bit. If I filter a lot, though, I'll loose lots of
detail. In practice few noisy video streams are filtered hard, as the
loss of detail is just too much. We normally compromise - a little less
noise, with a little less detail. The bottom line is our video source is
still quite noisy, and that is what you have to compress - unless you
have deep prior knowledge about the scene, like my Bugs Bunny example.

A related case. Digital cell phones sound OK in a quiet room, but sound
pretty awful out in the street where there is background noise. The
nature of voice compression makes this difficult to avoid. However, in
the competitive processes to choice the winning vocoder 100% of all
testing is conducted in a studio environment. Not a since utterance
other than a single clean human voice has been used for the competitive
evaluation of the vocoders we use today. There wasn't even a
non-competitive broader evaluation, to get a more rounded picture of the
vocoder's performance. This reflects directly in the design groups.
Those people trying to win the competitions (at least all the ones I
have seen) take the attitude that there is no point testing with source
material that won't be used in the competition. In general they don't
even bother to try to understand the overall behaviour of the vocoders
they design. A vocoder which achieves a more rounded balance of quality
in varying environments - the kind of thing most users really want - is
possible, but that vocoder will always loose the competition.

I think this sucks. I was amazed when I was first introduced to this
insanity. Test hard with the problem cases, but keep a close eye on them
all.

Regards,
Steve

--- >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 'vorbis-dev-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.