Medley filters — Simple tools for efficient signal smoothing
Maciej Jan Niedźwiecki, Szymon Cackowski · European Signal Processing Conference · 2010
Medley filters are defined as convex combinations of elementary smoothing filters (averaging, median) with different smoothing bandwidths. It is shown that when adaptive weights of such a mixture are evaluated using the recently proposed Bayesian rules, one obtains a tool which often outperforms the state-of-the-art wavelet-based smoothing algorithms. Additionally, unlike wavelet-based procedures, medley filters can easily cope with non-Gaussian (Laplacian, uniform) and temporarily inhomogeneous measurement noise.