[theora-dev] What sort of math i required?

[Holger Waechtler]

Timothy B. Terriberry wrote:
> There are two example wavelet codecs CVS, tarkin and w3d. IIRC, w3d is 
> the more recent one, but they both have been untouched for along time 
> now. That said, there are things that may perform better than wavelets, 
> such as lapped transforms.
> 
> Also, one need not throw away block motion estimation when using 
> wavelets. The motion estimation and residual coding stages can be made 
> completely independent. Whether or not that's a good idea is another story.

Does anybody of you knows whether it has been tried to compress 
(piecewise) continous motion flow instead of motion blocks?

In the last few years Michael Black published some papers how to compute 
dense flow fields robustly and efficiently, the algorithms are well 
suited for SIMD implementations on today's CPUs and GPUs. 
(http://www.cs.brown.edu/people/black/framework.html)

The resulting flow fields should be well compressible again with 
DCT/wavelet techniques. Has this been tried, does anybody knows research 
results about this?

> There isn't much improvement you can get on the actual entropy encoding 
> side. Huffman is cheap and free. Arithmetic coding is good and slow, and 
> may eventually be free (original patent was 1979, IIRC; I don't know 
> what other patents, if any, on specific implementation techniques would 
> also need to lapse before it becomes usable), and there are fast 
> variations of arithmetic coding that do not require multiplications and 
> divisions at the loss of some efficiency. Pick whatever meets your 
> requirements.

The bad thing about Huffman codes is that it's hard to implement them 
using adaptive statistics/tables without too much computational cost. 
Static Huffman codes are used widely and perform well when the tables 
are carefully optimized.

Holger

--- >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-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.