|
|
|
|
|
|
by minitech
723 days ago
|
|
|
> Given a message with two symbols in a 1:1 ratio, typical entropy encoders (Huffman, ANS, etc) would require 1 bit per symbol, with 0 representing one symbol and 1 representing the other. If you make use of the same information required here (exact counts) and update the probability model as you go, the result becomes the same for arithmetic coding/ANS (e.g. even with Huffman, after all symbols of one kind are seen, the remaining symbol is encoded in 0 bits). |
|
|