Exploring genetic programming for modeling character shape

In the framework of an evolutionary approach to machine learning, this paper proposes to use genetic programming as a tool to implement a learning module whose purpose is that of finding the set of prototypes to be used by a handwritten character recognition system. After discussing the rationale behind this choice, we describe the structural character shape representation adopted and the coding scheme for transforming such a two dimensional representation into a vector-based one, especially suitable for genetic programming. Then, the basic principles according to which the approach has been designed are presented, together with the genotype´s structure, the fitness function and the genetic operators devised to deal with the problem at hand. The results of a preliminary experiment performed on a standard database of handwritten characters are eventually reported