Properties of cellular neural networks in selected image processing applications

Abstract

Summary form only given. Concerns the use of stable analog cellular neural networks (CNN) for image processing. CNN architecture can be treated as a space-invariant iterative nonlinear filter. The authors compare CNNs and other techniques in image processing. The analysis is performed for two kinds of tasks for which nonlinear filters are commonly used: noise suppression and edge detection. Two synthesized test images, 64*64 pixels each, are used in experiments. One consists of solid blocks of different shapes and the other contains thin lines and sharp corners. The images are added with zero-mean Gaussian noise and impulsive noise. The efficiency of noise removal is examined. The limiter type M filter, a type of median filter, is considered. Edge detection by various filters and operators is compared.<>