[Mp4-tech] Bottom limit on bin/bit ratio in H.264

[Shevach Riabtsev]

Dear experts
According to JVT-C038 document, the theoretical upper bound on bin/bit
ratio is 64:1. 
I am interested in the estimation of the bottom (floor) bound of the
bin/bit ratio. 
Because of the adaptivity of CABAC probability models, the bottom limit
should be aproximately 1:1 (and the limit is achived on uniformly
distributed noise according to Information Theory).  Therefore on slices
with large number of MBs it is expected the bin/bit ratio is more 1:1
(since the adaptation period will require at most 64 MBs and this enough
to tune to the actual source entropy). 
On the other hand the CABAC adaptivity is not perfect, there are bins
coded as bypass (i.e. with the fixed model LPS=0.5), there is
end_of_slice_flag syntax element that coded with a fixed non-uniform
probability model, albeit its impact is minor.
What happens on low-resolution pictures? Indeed, while the probability
models are being tuned to the source entropy, the picture is finished. 
The above considerations cause me to think that the bottom limit of
bin/bit ration might be much less 1:1.  
Are there any JVT documents or papers on the estimation of bottom bound
of bin/bit ratio? Did someone experience streams with bin/bit ratio less
1:1?
Regards,  Shevach
Broadcom
-------------- next part --------------
An HTML attachment was scrubbed...
URL: /pipermail/mp4-tech/attachments/20070521/cac53987/attachment.html