Game Development Reference
In-Depth Information
Chapter 6
Entropy Coding
This chapter gives an overview of entropy coding in AVS2, especially for the
transform coefficient coding. It consists of four parts. The first part provides the
related basic concepts of entropy coding. In the second part, we describe the Context-
based 2D Variable Length Coding (C2DVLC) in AVS1 which provides the clue of
the basic coding symbols in AVS2. The third part describes the entropy coding called
Context-based Binary Arithmetic Coding (CBAC) in AVS2, and the last part con-
cludes this chapter.
6.1 Concepts of Entropy Coding
Similar to the first generation of AVS video coding standard (AVS1) (Fan et al.
2004
),
the second generation of AVS video coding standard, named AVS2 (He et al.
2013
),
still adopts the motion-compensated hybrid coding framework. In such a framework,
intra- and interpredictions are first used to remove the spatial and temporal corre-
lations, generating prediction errors. Then the prediction errors are transformed by
DCT, quantized, and finally entropy coded. Entropy coding is used for data com-
pression and stream organization.
Context-based entropy coding uses context modeling to exploit statistical behav-
iors of sequentially observed symbols for higher coding efficiency. The context mod-
eling is used to select a specific context from a given context set for a symbol which
will be coded. Generally speaking, a specific context corresponds to a specific statis-
tical behavior. Since the estimated conditional entropy based on the selected context
for the conditional coding of a symbol should maximally approach the actual one,
coding conditioned on contexts can adapt to symbols' local statistical variations and
therefore can improve coding efficiency. One of the most famous context modeling
methods is Rissanen's context modeling for universal data coding (Rissanen et al.
1983
). It can theoretically approach the bound of the minimal code length based on
the concept of stochastic complexity (Weinberger et al.
1996
). Besides, there are
