Towards optimised context selection in scalable wavelet based video coding

Abstract

The effectiveness of arithmetic coding in image and video compression depends on probability estimation of symbols to be encoded. Considering context in which these symbols occur can lead to improved compression. In wavelet-based coding same contexts are usually used for all wavelet subbands of same type across different scales of wavelet transform. Efficiency of different strategies for context selection has not yet been fully analysed for application in scalable wavelet-based video coding. In this paper an algorithm for context modelling based on tree structure is adapted for optimisation of contexts for symbols generated during Embedded ZeroBlock Coding (EZBC) of wavelet coefficients. With the proposed technique optimised contexts have been adaptively obtained for different wavelet subbands and EZBC quadtree levels. Comparison with predefined context models, which are used in common approach, shows that context models based on tree structure together with advanced modelling strategy significantly improve overall compression.