A neural architecture applied to the enhancement of noisy binary images without prior knowledge

Abstract

The authors present the formulation of an improved neural architecture, a modified adaptive resonance theory (ART), for the enhancement of binary images in the presence of noise. The two-layer ART model developed by G.A. Carpenter and S. Grossberg (1987) is further incorporated into a four-layer network. The operation and performance of ART1 in classifying binary input patterns is first investigated. Based on ART1, a noise filtering architecture is devised whereby preestablished recognition categories are used as region or contour detection exemplars in order to fill in the gaps and smooth the contours of a noisy binary image without any prior knowledge of the image itself.<>