Entropy-weighted Bayesian Approach to Edge Finding for Object Perception

Abstract

Why edge feature is considered far most important for establishing a perceptual contour in human vision system is based on two dependent viewpoints (a) robust ability to define/extract edges from heterogeneous objects or textures and (b) a subsequent step to decide which edges are significant enough to be preserved for object perception, namely the perceptual edges. In this paper, we present a method not only capable of finding perceptual edges but also allowing them to be used for constructing contours with good continuity. The method mainly comprises two stages: (i) a linear mask filter and non-linear filters (median filter and morphology) are applied to obtain fine-and coarse-edge features, respectively. (ii) An algorithm based on Entropy-weighted Bayesian decision making used to determine perceptual edges is carried out. Extensive simulation results are provided to show noise resistance, and the capability of approximating human visual perception is revealed by testing results of gestalt images.