Talk:Canonical Huffman code

From Pin Eight
Revision as of 03:59, 26 April 2012 by Tepples (talk | contribs) (Distro + entropy code: Test of new QuestyCaptcha (disguised as a query about unnecessary complexity) succeeded.)
Jump to: navigation, search

Distro + entropy code

A Canonical Huffman code as opposed to a Red Hat range code? What about a Gentoo Foundation gamma code, given the power law distribution of word frequencies? --98.226.71.46 20:47, 25 April 2012 (MST)

Not that Canonical, and not that universal code either. I've used gamma codes before for run-lengths in one of my NES-based image codecs. But in the case of Huffword, replacing the table of counts per length with a fixed universal code might save a little space in the decoder, but the difference between the size of a generic Exp-Golomb decoder (of which gamma is a special case) and the size of the first half of a CH decoder is probably minuscule. Most of the space in a CH decoder model is in the table from symbols to indices, and a universal code won't save any on that. In fact, Huffword uses two layers of CH: one for words and one for characters in the dictionary. --Tepples 20:59, 25 April 2012 (MST)