A genetic learning system for on-line character recognition

In this paper we present the result of an investigation on a new method to perform online character recognition. The method is based on a genetic algorithm used as the engine of a learning system to produce prototypes of the characters, and on a string matcher to perform the classification. The learning mechanism, provided by a genetic algorithm, allows the system to have both a writer independent core and an adaptation scheme to finely tune the recognizer to the writer´s style. Preliminary experiments have shown that the method is very promising, since it produces prototypes general enough to cope with the large variability encountered when handling specimen produced by different writers. Moreover, it provides a natural and effective writer-dependent learning of new symbols