Bridging Compression to Wavelet Thresholding as a Denoising Method
S. Grace Chang, Bin Yu, Martin Vetterli · Infoscience (Ecole Polytechnique Fédérale de Lausanne) · 1997
Some past work has suggested that lossy compression can be a good denoising tool. Building on this theme, we make the connection that quantization of transform coefficients approximates the operation of Donoho-Johnstone's wavelet thresholding, to conclude that compression (via coefficient quantization) is appropriate for filtering noise from signal. The method of quantization is scale adaptive and is facilitated by a criterion similar to Rissanen's minimum description length principle. Results show that a small number of quantization levels achieves almost the same performance of full precision thresholding, suggesting that denoising is mainly due to the zero-zone and that the full precision of the thresheld coefficients is of secondary importance. 1. INTRODUCTION The purpose of this paper is to explain why lossy compression can be appropriate for signal denoising. More specifically, we wish to show that quantization (a common step in compression) of transform coefficients achieves de...