High-order entropy coding for tree structured vector quantization

This paper presents an efficient noiseless encoding scheme for a balanced/unbalanced tree structured vector quantization (TSVQ) system utilizing high order statistics of TSVQ indices. The proposed encoding scheme includes a reduction method of both the number of conditioning states and the size of entropy table at each encoder state. In addition, to effectively manage a given amount of memory resource, the memory constrained encoding problem is examined. Simulation results show the proposed scheme brings remarkable bitrate reduction even under a moderate amount of memory.