Feedback Control based Spatial Filter Window Size Selection for Image Processing

Abstract

Window-based spatial filtering is usually the first pre-processing operation performed, primarily for noise removal, in most image processing applications. In most filter designs, the window size (WS) is selected after performing a search that minimizes some criterion. Such a procedure is dependent on the choice of the initial WS as well as the search domain. In this work, an integral control based algorithm is proposed, where the WS is posed as the control variable that is designed so that the Power Spectral Density (PSD) of the unfiltered image-its output, tracks a reference PSD. The reference PSD is specified such that it consists of only the high energy components. It will be shown that the proposed algorithm is immune to the initial WS value, type of noise, and can be implemented for the design of several types of spatial filters.