Design of multi-mask aperture filters

Aperture filters compose a recently introduced class of non-linear operators used in signal processing. Their operation involves filtering of signals that are observed within a window of finite width and height. They allow a tractable design of non-linear filters by reducing the search space. This paper presents an adaptation to the original design involving multiple masks with shapes chosen to fit commonly occurring patterns of the input signal. The information obtained using the different masks is efficiently combined to produce the multi-mask filter, which is optimal with respect to the class. This paper demonstrates that for smaller training set sizes the multi-mask filter is well trained and therefore performs better than a single aperture filter with the same data.