A novel adaptive subspace distortion measurement technique for image vector quantization

An adaptive subspace distortion measurement technique is proposed to further reduce the complexity of the tree-structured vector quantization (TSVQ). The Euclidean distortion is adaptively replaced by dimensionally reduced transform-domain subspace distortions during tree codebook generation. To achieve good approximation of the Euclidean distortion, the dimensionality and basis of subspace distortions are selected on the basis of the local statistics of the partition associated with each nonterminal node of the tree. Thus, substantial reductions of both computational complexity and memory requirement can be obtained. Experimental results show that exceptionally low subspace dimension can be used in multisubspace TSVQ based on a fixed transform-domain to obtain a performance similar to that of conventional or single-subspace TSVQ