Overlapped Context Modeling Using Feature Mapping Functions in the Adaptive Arithmetic Coding Process for Lossless Encoding

Abstract

Context modeling plays a critical role in the adaptive arithmetic coding process. It classifies the causal part into several classes according to the features extracted from the causal neighboring pixels. However, when the feature value is around the border of the ranges of two adjacent contexts, its corresponding probability model cannot be estimated accurately. In this paper, we propose an advanced way for context assignment. We make the contexts overlapped in both the training phase and the coding phase. With the proposed method, more than one context wm be assigned for each input data. Then, the probability model generated by weighted combination is applied to encode the input data. Then, the frequency table corresponds to the context whose range overlaps with the input data value wm be adjusted. Experimental results on lossless image coding show that, with the proposed algorithm, a high coding efficiency can be achieved.