Genetic algorithms with fuzzy fitness function for object extraction using cellular networks

Abstract

Setting up a Cellular Neural Network (CNN) for a particular task needs a proper selection of circuit parameters (cloning template) which determines the dynamics of the network. The present paper provides a methodology, demonstrating the capability of Genetic Algorithms with a fuzzy fitness function, for automatic selection of cloning templates when a CNN is used in extracting object regions from noisy images. Fuzzy geometrical properties of image are used as the basis of fitness function. The proposed method relieves the CNN from using heuristics for the template selection procedure, and performs consistently well in noisy environments.