<div dir="ltr">Hi<div><br></div><div style><br></div><div style>I've been invited to join the discussion on Daala working group</div><div style>on a very dedicated matter which contributes to the codec compression effectiveness : the entropy coder.</div>
<div style><br></div><div style>Tim's presentation (<a href="http://people.xiph.org/~tterribe/pubs/lca2012/auckland/intro_to_video1.pdf">http://people.xiph.org/~tterribe/pubs/lca2012/auckland/intro_to_video1.pdf</a>, starting slide 51) makes a detailed presentation on available techniques for this coder, mainly Huffman and Arithmetic coders.</div>
<div style><br></div><div style>One recent new technique that could be of interest to Daala is the Finite State Entropy Coder :</div><div style><a href="https://github.com/Cyan4973/FiniteStateEntropy">https://github.com/Cyan4973/FiniteStateEntropy</a><br>
</div><div style>or</div><div style><a href="http://fastcompression.blogspot.fr/2013/12/finite-state-entropy-new-breed-of.html">http://fastcompression.blogspot.fr/2013/12/finite-state-entropy-new-breed-of.html</a><br></div>
<div style><br></div><div style>To summarize its property, it gets the speed of Huffman, and the accuracy of Arithmetic coders. For coders with very high prediction capability, it makes a big difference in accessible compression ratio (closer to Shannon Limit).</div>
<div style>Possibly of importance, it does its job while using only basic operations, add, shifts and masks.</div><div style><br></div><div style>The open source version on github (BSD license) can be downloaded and includes a benchmark program to test its effectiveness on any sample dataset.</div>
<div style><br></div><div style><br></div><div style>Best regards</div><div style><br></div></div>