Wavelet-based empirical Wiener filtering

Existing denoising schemes rarely use multiple-bases representations and if they do, they do not address the choice of the different bases. We present a new denoising scheme based on multiple bases processing. The multiple bases used in the denoising algorithm are generated via unitary transforms. These unitary transforms also allow the construction of new wavelet bases. In the new domains spanned by the multiple bases, we apply a simple hard thresholding technique as well as a more complex Wiener filtering scheme. Preliminary results suggest that the resulting algorithms can deliver significantly improved performance over the undecimated wavelet transform without being computationally more expensive.