[Theora-dev] Question about Huffman Tables in Setup Header

[VP3HoSwiYO]

Timothy B. Terriberry wrote:

   > VP3HoSwiYO wrote:
   >
   >>If we want to use this space, we find out best(or proper) performance
   >>80 frequency counts(Huffman code sets) from tens thousands of frame.
   >>Dose anyone have an idea or algorithm to find out them?
   >>
   >
   > This is a generic clustering problem. Clustering is NP-hard, but there
   > are a number of reasonable algorithms that will give you an answer, if
   > not necessarily the optimal one, such as K-means based algorithms.
   >
   >

Thanks your advice. But I'm sorry I mistook my question,
and we need optimal Huffman code sets. If it is not optimal, it is not
worth using it.
About 16 DC Huffman codes, Algorithms which you show may be effective.
But about AC Huffman codes, more complex algorithms will be needed, I think.


I think this bit space is useless, because using this is troublesome.
it will need much time to find optimal Huffman code sets.
and then, using these codes, we must re-encode tokens in video stream.
Dose anyone try to use this?
If nobody try, This will be really useless.