A new evolutionary learning model for handwritten character prototyping
Luigi Pietro Cordella, Claudio De Stefano, Antonio Della Cioppa, Angelo Marcelli · 2003
The work reported in this paper is aimed at exploiting evolutionary learning algorithms for producing the set of prototypes to be used by a handwriting recognition system. In this paper we propose a new evolutionary learning model that combines the power of search of a classical evolutionary algorithm, namely a breeder genetic algorithm, with a novel mechanism for implementing the interaction between the evolving population and the environment. The proposed model allows the system to search for the prototypes by means of a simple iterative strategy rather than through a parallel and adaptive search, as generally happens in evolutionary learning. Experimental results on handwritten digits have shown that the performance of the proposed algorithm is similar to that exhibited by more complex evolutionary learning algorithms, and better than that provided by a neural network.