Information theory, wavelets, and image compression

Abstract

We examine practical, theoretical, and speculative aspects of wavelet transform-based image compression. Section I summarizes objectives and compares experimental results using a JPEG-standard cosine-based algorithm with a wavelet based algorithm developed at ISS. Section II analyzes image compression requirements using information theory to explain why wavelet transform-based image compression works well. The wavelet transform is shown to be a simple transform that effectively exploits second-order image statistics. Section III speculates about next-generation image compression and pattern recognition. It outlines a research plan to develop a probabilistic image model that incorporates higher-order image statistics by using wavelet expansions to provide a convergent series of finite dimensional marginal image probability densities. Physicists have successfully used similar cell cluster expansions to analyze lattice fields, Ising models, and Euclidean quantum fields. © 1996 John Wiley & Sons, Inc.