Design of two-dimensional digital filters by using the singular value decomposition

Abstract

A method for the design of quadrantally symmetric 2-D digital filters is described. Like the method outlined by Twogood and Mitra, the method proposed here is based on the singular value decomposition. The salient difference between the two methods is that the proposed method uses one or more parallel correction sections which reduce the approximation error. In this way, reduced passband ripple and increased minimum stopband attenuation can easily be achieved. An important merit of the method is that the required 2-D filter is decomposed into a set of 1-D digital subfilters, which are much easier to design by optimization than the original 2-D filter.