Probabilistic and Q-coder algorithms for binary source adaptation

Abstract

Familiarity with arithmetic coding is assumed. Early probability estimators for binary sources employ deterministic techniques, see [1], for a survey. One way to characterize a binary source is by the value of the more probable symbol (MPS) and less probable symbol (LPS) and by skewness of the distribution. The negative log2 of p(LPS) serves as a skewness measure. Many binary adapters quantize the skewness range to a small number of indices k, and use an indexchange decision to alter the estimate.