Probability estimation of binary information sources for image coding

Probability estimation of symbol occurrence in adaptive entropy coding used for efficient image coding can be classified into two types. One is Bayesian probability estimation, in which the probability is estimated by using accumulated occurrence counts of the information source symbols. The other is state transition probability estimation in which the probability is estimated through a state transition not always caused by every occurrence of information source symbols. In this paper, we examine the characteristic of both of the probability estimation methods, and propose a new probability estimation method in which either of the two is switched adaptively. We confirmed that the proposed probability estimation provides a higher compression performance on random sequences having wide range of occurrence probabilities, and also works well on coding of various images