Adaptive Wiener Filtering using Polyharmonic Wavelet Packets
Master Semester Project
A common approach for image denoising is Wiener filtering, which provides us with the (global) minimum mean squared error (MMSE) solution. Recently, we have showed the optimal Wiener filter solution for fractal-like images. The best function space for this type of image models is generated by the polyharmonic B-splines. The aim of this project is to design an spatially adaptive version of this algorithm, using the polyharmonic B-spline wavelets. First, these wavelets span the same optimal space as the polyharmonic B-splines (which are the scaling functions). Second, this wavelet transform can also be applied in "wavelet packet"-style to have a fine spectral decomposition. The power spectrum model will be fitted locally for each wavelet coefficient through scale, which can then be used to derive optimal local scaling parameters derived from the optimal Wiener filter solution. The number of iterations will correspond to an elegant trade-off between the spatial and the spectral domain.
- Dimitri Van De Ville, firstname.lastname@example.org, 021 693 51 42, BM 4.140
- Michael Unser, email@example.com, 021 693 51 75, BM 4.136
- Sathish Ramani