Possibility distribution based lossless coding and its optimization

Adaptive estimated probability distribution is used by lossless coding, i.e. entropy coding to encode the source when probability distribution is not stable. However, probability distribution prediction inaccuracy and its influence on coding performance cannot be quantitatively evaluated. In this paper, we treat the probability distribution prediction as a possibility distribution prediction process and use estimated possibility distribution to code the source. Possibility describes the likelihood of an event in one symbol while probability describe the frequency of an event in infinite symbols. We find that code rate is the cost for inaccurate possibility distribution prediction. We also increase the possibility distribution prediction accuracy from two aspects: 1. The best Exponential Smoothing Prediction (ESP) model parameter is derived. 2. A source reorganization algorithm which can significantly increase the correlation between accurate possibility distributions of different symbols is designed. Applying the proposed optimizations on lossless coding part in HEVC, great coding efficiency improvement can be achieved.