{"title":"Signal de-noising using adaptive Bayesian wavelet shrinkage","authors":"H. Chipman, E.D. Kolacxyk, R. McCulloch","doi":"10.1109/TFSA.1996.547454","DOIUrl":null,"url":null,"abstract":"Shrinkage of the empirical wavelet coefficients is an effective way to de-noise signals possessing sparse wavelet transforms. This article outlines a Bayesian approach to wavelet shrinkage, in which the form of the shrinkage function is induced by a particular choice of prior distributions placed on the wavelet coefficients. Our priors are chosen to be mixtures of two normal distributions, one wide and the other narrow, so as to effectively model the sparseness inherent in the wavelet representations of many signals. This particular choice of prior also allows us to obtain a closed-form expression for the shrinkage function (posterior mean) and for the corresponding uncertainty (posterior variance). This uncertainty information is used in turn to generate uncertainty bands for the full signal reconstruction. An automatic, level-dependent scheme is used to adapt the shrinkage functions to each resolution level of coefficients, although subjective information may be incorporated quite easily.","PeriodicalId":415923,"journal":{"name":"Proceedings of Third International Symposium on Time-Frequency and Time-Scale Analysis (TFTS-96)","volume":"3 1","pages":"0"},"PeriodicalIF":0.0000,"publicationDate":"1996-06-18","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":"11","resultStr":null,"platform":"Semanticscholar","paperid":null,"PeriodicalName":"Proceedings of Third International Symposium on Time-Frequency and Time-Scale Analysis (TFTS-96)","FirstCategoryId":"1085","ListUrlMain":"https://doi.org/10.1109/TFSA.1996.547454","RegionNum":0,"RegionCategory":null,"ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":null,"EPubDate":"","PubModel":"","JCR":"","JCRName":"","Score":null,"Total":0}
引用次数: 11
Abstract
Shrinkage of the empirical wavelet coefficients is an effective way to de-noise signals possessing sparse wavelet transforms. This article outlines a Bayesian approach to wavelet shrinkage, in which the form of the shrinkage function is induced by a particular choice of prior distributions placed on the wavelet coefficients. Our priors are chosen to be mixtures of two normal distributions, one wide and the other narrow, so as to effectively model the sparseness inherent in the wavelet representations of many signals. This particular choice of prior also allows us to obtain a closed-form expression for the shrinkage function (posterior mean) and for the corresponding uncertainty (posterior variance). This uncertainty information is used in turn to generate uncertainty bands for the full signal reconstruction. An automatic, level-dependent scheme is used to adapt the shrinkage functions to each resolution level of coefficients, although subjective information may be incorporated quite easily.