The learning type of mean and median hybrid filters
Misaki Meguro, Akira Taguchi · 2002
In this paper, a new adaptive filter, called a learning type of mean and median hybrid (LMMH) filters, is introduced. This filter is a combination of FIR filtering and order statistics (OS) filtering for removal all kinds of distributed noise. LMMH filter is regarded as the extension of MMH filters which can't be learned. On the other hand, LMMH filters can be optimized by using a priori information on the input signal. A procedure for designing optimal LMMH filters under the mean square error criterion has been developed. Experimental results show that the performances of the optimal LMMH filter are superior to those of the Wiener filter and the OS filter, for signal corrupted by from short- to long-tailed distributed noise. The filters are applied to image restoration.