DCT-based high speed vector quantization using classified weighted tree-structured codebook

Abstract

We propose a high-speed vector quantizer with classified weighted tree-structured codebook based on the characteristics of DCT (discrete cosine transform) coefficients. To reduce the encoding complexity and the edge degradation, we employ both the CVQ (classified vector quantization) and the modified TSVQ (tree search vector quantization) approach. In this scheme, input vectors are classified into four edge-oriented classes by simple classification algorithm employing two DCT coefficients. For each class, a weighted tree-structured codebook, whose search vectors have lower dimensions in average than the input vector, is designed by using BTSOFM (binary tree-structured self-organizing feature maps). When searching the tree, only the specified DCT coefficients of the input vector are referenced, hence resulting in a fast tree search. Simulation results show that the encoding speed of the proposed vector quantizer is twice as fast as that of the TSVQ with a little improvement of image quality at 0.625 bpp.