Spatial filters for pattern detection and image enhancement

Abstract

A theory of spatial frequency domain filters is presented from the standpoint of leastsquare linear estimation. The general expression of this filter is obtained with few assumptions, thus allowing considerable flexibility in its application. Noise common to pictorial information usually can be described as a mutually exclusive process, whereas much of the available signal processing theory assumes additive noise. Types of pictorial noise discussed in this paper include the background of signal patterns, pattern obscurity, and image modulation due to the recording process. It is shown that the additive noise assumption is valid under certain conditions. The general filter is described in terms of spatial frequency estimates which are conveniently expanded for particular applications. A detection filter for determinative patterns is obtained and compared to the matched filter. A filter is derived to restore patterns degraded by noise and a linear distortion filter.