Tree-structured vector quantization using direction and resolution information in wavelet transform domain

Abstract

A coding technique, based on WT (wavelet transform) and TSVQ (tree-structured vector quantization), is proposed in this paper. Wavelet transformed image is composed of several subimages according to resolutions and edge directions, and has a particular PDF (probability density function), the generalized Gaussian distribution. We propose an improved tree- structured VQ coder based on the properties of wavelet transform. Edge information extracted from the subimages in the wavelet transform domain has been used to reduce the distortion. A new vector formation scheme and a new tree growing algorithm has been presented in this paper to reduce the distortion rate in the reconstructed image. Finally, in order to allow the receiver a picture as quickly as possible at minimum cost, we propose a progressive transmission scheme using unbalanced tree structured codebook. It is shown that unbalanced TSVQ is well adapted to progressive transmission. Simulations results indicate that the quality of the reconstructed image is excellent in the range of 0.30 - 0.40 bit/pixel.