Redundancy Analysis for Variable Length Coding

Abstract

Typical VLC schemes use fixed Huffman tables, thus the coding efficiency is far from the real Huffman coder. By employing multiple fixed code tables, the performance should be increased and approach the ideal Huffman coder. Since the fixed table VLC method is not the "minimum redundant" one, there is a believe that the redundancy between symbols still exists and there should be methods which could further remove it, and the possible way is encoding concatenation of codewords together, i.e., applying a post-processing on the VLC code streams. Some models are given to describe how much equivalent redundancy exist within the VLC code streams by using the channel coding concepts. Starting from those models, we introduce some modifications to evaluate the inter symbol redundancy for source coding. The estimation based on the models for the redundancies of H.263 and AVS are shown, which may lead us to understand the post-processing efficiency could achieve.