Edge Detection With Matched Filters

Abstract

This paper suggests that a matched filter be used for image edge detection to maximize the output signal-to-noise of an edge operator.We assume stochastic models for edge,background and noise,define the frequency transfer function of matched filters mid achieve the optimum edge detection in frequency dowain.This paper proposes two new operators in spatial domain for recognizing edges in two-dimensional images. The first, called the average difference filter(ADF),combines the outputs of two bandpass fitters at two orthogonal orientations which are motivated by two-dimensional matched filters. The second edge operator called the median difference filter(MDF)is a modification of ADF.The parameters of the filters can be chosen to approximate matched fitters if the estimated parameters of models for edge, background and noise are proper.In addition to exhibiting low computational cost, the two kinds of fitters are shown to otitperform other previous algorithms for edge detection in noisy images.