A Wiener filter improvement combining wavelet domains

An improvement of an interesting approach for denoising in the wavelet domain using an empirical Wiener filter (Choi, H. and Baraniuk, R., Proc. IEEE-SP Int. Symp. on Time-frequency and Time-scale Anal., 1998) is presented. It is based on a well-balanced matching of selection and attenuation performances using more than one wavelet basis. Experimental results are better than other wavelet-Wiener based techniques, both in terms of objective (signal-to-noise ratio, SNR) and subjective quality, requiring a moderate computing time.