2 - Débruitage par ondelettes et segmentation de signaux non-stationnaires : réinterprétation d'un algorithme itératif et application à la phonoentérographie

This communication deals with wavelet-based denoising techniques of non-stationary signals, in order to extract informative events. The practical application concerns physiological bowel sounds processing, with a view to medical diagnosis and monitoring. This work continues and develops a recent publication placed in the same framework [14]. The method for separating the stationary part from the non-stationary part of a signal presented by Hadjileontiadis et al. [15, 14] stems from a denoising algorithm introduced by Coifman and Wickerhauser [6, 7]. This method involves two user-tuned parameters. We propose a novel version of this algorithm, based on a fixed-point interpretation. This modification allows to eliminate one of the parameters and to determine an inferior limit for the second, depending on the probability distribution of the wavelet coefficients. This revisited version also improves significantly the computational efficiency. We present the results and compare them with other denoising algorithms, both on simulated signals and on real bowel sounds.