Genetic programming of a CNN multi-template tree for automatic generation of analogic algorithms

Abstract

A fruitful field of cellular neural net (CNN) research is the development of analogic algorithms utilizing combinations of single templates to perform complex image processing task; dedicated to industrial applications, vision problems in robotics, pattern analysis, etc. In this work a software implementation for the automatic generation of analogic algorithms by mean of a genetic search is presented First, we briefly present an improved automatic templates generation. Next, an algorithm for generating templates in cascade will be showed like the natural and original extension of the already known tool. Lastly, the multitemplate tree concept derived from the AI field is applied in the automatic generation of analog algorithms, and its solution based in both genetic evolutionary search and heuristic methods are exposed.