Multi-subspace tree-structured vector quantizer design algorithms

To further reduce the searching complexity and memory requirement of the tree-structured vector quantization (TSVQ), new multi-subspace TSVQ design algorithms and encoding techniques are proposed in this paper. The proposed multi-subspace TSVQ design algorithms perform the vector quantization in spatial domain while using specially designed subspace distortions in transform domain as cost functions for the optimization process. The dimensionality and basis of subspace distortions are selected based on the local statistics of the partition associated with each non-terminal node of the tree. Experimental results show that extremely low subspace dimension can be used in multi-subspace TSVQ based on a fixed transform domain to obtain a similar performance as the conventional TSVQ or single subspace TSVQ. In addition, the proposed generalized multi-subspace TSVQ design algorithm is a general TSVQ design algorithm and which can also be utilized as a fast codebook design algorithm.