Wavelet-based image compression using classified interpolative vector quantization

Abstract

A new method of nonlinear interpolative vector quantization for wavelet-transform image coding is proposed. Training and quantization are carried out by classifying the coefficient vectors in the wavelet-transform domain according to their directional activity energy. A product codebook is used during encoding and decoding to allocate bit assignments among these classes adaptively. Compared to the conventional interpolative vector quantization technique, the new method gives superior compression performance and also reduces the complexity of codebook training and quantization. The PSNR improvement is about 1 dB at the compression ratio of 50.