Use of an evolutive base of models in a system for reading printed texts

The recognition system of printed text that we present uses a base of models of characters to recognize the prototypes of various patterns of characters encountered in successive readings. To identify, the prototypes, we have developed a decision making method based on the use of adaptive adherences borrowed from the rule of pretopology. This approach which we have termed adaptable ε(x)-neighbors allows an adjustment of classes using a minimum quantity of information. To reorganize the representation space, we have elaborated a system in which the models are associated to weights for which a continuous correction process is conducted. This process effects a permanent evolution of the models and continuously questions their influence in recognition. Weight management that entails the creation and the removal of models places our system in a context of continuous learning