Adaptive techniques for lossless data compression

Abstract

Data compression techniques have many applications in medical signal and image processing. In medical imaging, lossless image compression is required. According to information theory, a fundamental problem in data compression is to estimate the probability distribution function (pdf) of the signal given the data seen so far. The estimation should be as close as possible to the true pdf. For non-stationary signals, an adaptive estimation technique must be used. In this paper we address this problem by reviewing the current practices in compressing digital image and audio data. We show that the popular prediction plus entropy coding approach is only a rough approximation to that suggested by information theory. We then discuss a Bayesian approach to improve the prediction performance. We also propose another Bayesian approach for adaptive pdf estimation.