[vorbis] Video codec

Chrissy and Raul chrissyandraul at hotmail.com
Tue Sep 12 10:02:06 PDT 2000



These mailings are getting very detailed, this is good.  Answers below:

>From: Ralph Giles <giles at snow.ashlu.bc.ca>
>
>On Mon, 11 Sep 2000, Chrissy and Raul wrote:
>
> > >Totally agreed. Let's start making list and try to agree upon it?
>
>I assume these are for purposes of comparison and guidelines for common
>source types, rather than requirements. I was hoping we also "already
>agree" to allow arb. frame size and rate. One of my pet peeves with the
>DVD spec is the way they include the letterbox mask in the frame they
>compress, just to keep things within the spec'd size and aspect ratio.

When we do arbitrary frame sizes that are not exact multiples of the 
original scanning rate (of camera CCD/CMOS sensor, of film scanner or of 
original A/D converter) there is a softening of the picture that happens due 
to the arbitrary subsampling.  Using filters to reduce say 720 to 640 pixels 
causes fine lines to disappear, create weird patters and edges to be 
softened.  Companies like Faroudja, Snell and Wilcox and now Teranex, go 
through incredible pains to scale up to large sizes and the pictures are 
still softened because you are making pixels out of nothing.  When you are 
scaling down you face similar problems.  I want to stay away from softening 
the picture by using exact SUB-multiples of the original scanning.  Even 
hardware manufacturers stick to multiples: for example the new Compaq 
portable PocketPC has a 320x240 display.  The softening of adapting a 
picture to a certain display with different native resolution can create 
maddening artifacts.

>
> > Frame Rates:
> >
> > - 15 fps (half 30 fps)
> > - 24 fps (Film native rate.  I heard that in Europe 24 fps movie DVDs 
>are
> > presented at 25fps which make them end sooner, good, more time for the
> > family, is this true?)
> > - 30 fps
>
>Likewise, we're not worrying that 30 fps isn't really 30 fps in NTSC
>video?

The difference is small enough to not be a problem.

>
>I'd also suggest 16 fps for early cinema, and 60 for action sequences
>(ask someone who plays first person shooters). Rates of 8 and 12 fps are
>sometimes used for animation.

60 fps is definitely good, the hardware requirements are huge, though.

>
> > SIF (320x240 resolution)
> > [...]
> > QSIF (160x120 resolution)
>
>We also need other frame sizes for film. Common aspect ratios are:
>
>1.33:1  (4x3)
>1.78:1  (16x9)
>1.85:1  (US standard for 'just ordinary movies')
>2.35:1  ("cinemascope")
>
>Here are some examples based on multiples of 240, by way of example:
>
>vert.	x1.33	x1.78	x1.85	x2.35
>240	320	426	444	564
>480	640	854	888	1128
>720	960	1280	1332	1692
>960	1280	1706	1776	2256
>
>Note that block-oriented codecs like mpeg want the sizes to be multiples
>of 8, 16, or 32, of which every few of these qualify. For a given aspect
>ratio one can usually find something though. For example, the "Lord of the
>Rings" trailer (quicktime) is available in 240x104, 320x136, 480x204, and
>640x272. All divisible by 8 and all but the smallest at 2.3529:1.
>
>As Monty pointed out, the proposed wavelet codec doesn't care about
>blocks. I don't know anything about the proposed non-power-of-2 transform
>though. Are things like small prime factors important like with the fft?

One issue with Wavelets is that they require access to larger areas at the 
same time than 8x8 DCTs, isn't that true?

>
>The US HDTV standard also suggests some standard frames:
>
>640x480   (4x3)
>1280x720  (16x9)
>1920x1080 (16x9)
>
>All but the largest support 24,30, and 60 fps (actually 60/1.001).
>
>IMO we should be planning to target content production and a replacement
>for DVD-video as well as web streaming. That means large formats. The
>acknowledged digital-equivalent vertical resolution of 35mm film is 4k
>lines (these days; that's about equivalent to the 70mm epics of 25+ years
>ago). Most special effects work is done at 2k lines and some folks claim
>to see the difference. That's not to say you always get that, of course:
>it depends on film stock, exposure, lens quality, development, etc. At
>some point, you're just resolving the grain. OTOH, rumour says the new
>"digital film" camera being used to shoot the next "Star Wars" are only 1k
>lines (960, I think) and they're still having to be more careful with set
>painting because they give sharper detail than film.

Viewing any image at the right distance with the right contrast and color 
can give you the impression of reality.  If you see the interlacing lines in 
your TV step back until you don't see them, that is a good place to watch.  
The eyes does wonders integrating separate lines, it is only a matter of how 
close you want to be to the screen, how much it becomes part of your 
perceived surroundings.

>
>See this recent slashdot article for further comments:
>
>http://slashdot.org/article.pl?sid=00/09/11/1710204
>
>
>Speaking of square pixels, the advantage is obvious to most software
>people, but we'll need to be ready to defend this one. One thing I didn't
>realize is that effectively *all* 35mm film work uses an anamorphic
>(cylindrical) lens to squeeze the director's chosen aspect ratio onto a
>standard-sized area of film. (The process is reversed in projection.) So
>people are used to this, and from the point of view of hardware
>implementations, it makes a lot of sense to have a fixed frame-size plus
>(even arbitrary) horizontal scaling.

Here I push for *optically* correcting anamorphically encoded digital 
images.  There is also Super35, like Titanic, that shoot everything in the 
original 4:3 aspect ratio and then cut top and bottom.

>
>
> > I am using three scenes for the video compression experiments: hockey,
> > mongolian rock concert, mongolian flag and plaza.
>
>Do you own the copyright on these clips? Would you be willing to make them
>available in uncompressed form (on the net, or mail someone a cd) so we
>can make a start on a test suite?

My company owns the mongolian rock concert and the mongolian flag and plaza. 
  We can get an easy release.  We need to shoot hockey and basketball at 
somebody's neighborhood and get releases from the neighborhood players.

RAUL LOPEZ

>
>Cheers,
>  -ralph
>
>--
>giles at ashlu.bc.ca
>
>
>
>--- >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.

_________________________________________________________________________
Get Your Private, Free E-mail from MSN Hotmail at http://www.hotmail.com.

Share information about yourself, create your own public profile at 
http://profiles.msn.com.

--- >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.



More information about the Vorbis mailing list