High order context modeling and entropy coding of multimedia data.

摘要

Arguably the most important component of a multimedia signal compression system is statistical context modeling and entropy coding of the signals. Statistical context modeling can expose the long-term memory of the source to an entropy coder so that it can achieve a code length approaching to the high-order conditional entropy.; This thesis presents some new algorithmic techniques for high-order context modeling, examines and demonstrates the efficacy of these techniques in multimedia data compression. The proposed context modeling algorithms are guided by universal source coding principle, and assume no knowledge about the source being coded. The context model is constructed by an on-line learning process which estimates the conditional probability of future samples based on the statistics of the samples being coded (the so-far observable).; Given the popularity of wavelet-based compression techniques in the past decade, this research focuses on context modeling and entropy coding of wavelet coefficients. Comprehensive and comparative studies were carried out on adaptive entropy coding of wavelet coefficients of various multimedia data contents, including image, video, 3D volume data, and audio. We have obtained some of the best compression results so far in the literature.