[vorbis] Video codec

[Jelle Foks]

Tom Bishop wrote:
> 
> >
> > > the MPEG effects were not
> > > that noticable on a 27" tube, but when played on a 43" projection system
> > > suffers very noticably.
> >
> > Which the best noticable degradation: Ringing? Blocking? Blur? Skewed
> > areas? In the colors or in the luminance? I'm sorry, I haven't seen the
> > tivo in Europe yet.
> 
> I have replayTV.. I would say "Blocking" ..though I do not have that
> much understanding of terminology, there are prominent blocks that appear
> during action.. still shots look reasonable.

That does sound like blocking, or maybe completely running out of bits
and choosing to use 'skipped macroblocks' where it shouldn't but has to
because it's reaching the bitrate ceiling (which basically means that
it's skipping parts of the image, displaying the previous reference
image in the skipped locations). In frames with more motion, more bits
are needed to encode the motion vectors, so most encoders adaptively
increase the quantization in frames with motion. Often that can be done
without it being visible, because we see less and less details as the
motion increases. But of course, you can always go too far.

If it only occurs in a fraction of the images, then it raises the
question why they didn't use a VBR compressor with a higher bit-rate
ceiling. If it only happens occasionally, then it wouldn't increase the
average bit rate too much, which is the most important factor for video
on hard disk anyway. Maybe they're using an MPEG encoder chip that only
supports fixed bit-rate.

If blocking is the only problem, then the image quality at that setting
can probably be improved by using a decompressor with some (better)
deblocking algorithms. I'd say that there is room for a replayTV version
2 with improved image quality while still being fully compatable with
the current version.

Blocking shouldn't occur at 2.22mbit/s at SIF resolution, but could very
well occur at CCIR601/656 resolution at that bitrate, so I think we can
assume that replayTV is encoding at full CCIR601/656 resolution.

Of course, when Ogg Video is ready, the should be able to use that for
optimal performance ;-))

I'll describe what I mean with the terms below. Some explanations may be
incomplete, some effects are pretty rare, but I think I'm pretty much on
the spot.

Well, MPEG video is built up of quantized DCT coefficients, so the
effects of degradation that you can see is directly related to that.

Blocking occurs if the quantization of the DC coefficient is too strong,
in wich case the boundaries between the macroblocks (16x16 pixels)
become visible, and you'll see blocks in the image where the colour or
image intensity varies from the surrounding blocks.

Ringing occurs if the quantization of the highest frequency DCT
coefficients is too strong for the image content. In this case, you'll
see ripples, like little waves. Ringing is mostly seen on parts of the
image where there are sharp edges, especially sharp corners.

Blur occurs if the quantization of almost all higher frequency DCT
coefficients is too strong, or when the encoder chooses to remove
higher-frequency DCT coefficients from the compressed stream to save
bits.  Blurring also occurs if blocking is really occurring, but the
decoder uses a lowpass filter to try to encounter the effect (happens
sometimes in the lowest bit-rate H.26x-type streams).

Skewed area's happens mostly in a transmission if there are transmission
errors in combination with movement in the image. When seeing that, I'm
thinking of corrupt or otherwise nonoptimal motion vectors. Regional
skipping of macroblocks during frames with motion may also result in
such an effect.

If more image degradation occurs in the colors (chrominance) than in the
luminance, then that may be a sign that the encoder tries to reduce the
bit-rate by underestimating the importance of colors, for example by
using stronger quantization on the chrominance blocks, or by using more
subsampling for the chrominance pixels.

Cya,

Jelle.

--- >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-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.