Reconnaissance de caractères manuscrits par intelligence collective

Abstract

The general problematic that directs our work of this thesis focuses on the collective interactions between autonomous entities and on emergency of a complex system. This technique relates to the manner of dividing a problem between entities relatively simple but able to self-organize in order to achieve in a collective way a common goal. It forms a part of the vaster problematic of collective Intelligence. This distributed vision caused an increasing attention of researchers for the quality of its applications and its advantages such as parallelism, speed, flexibility… etc. Swarm intelligence constitutes a sub-field of collective intelligence. The work described in this thesis gives place to applications mainly inspired of the collective behaviors of birds and bees for characters recognition and particularly for recognition of handwritten Arabic numerals and English letters. This thesis thus presents four contributions based on two types of neural networks namely MLP and RBF neural networks. Within the framework of distributed vision, we tried to focus our efforts on the application of new “swarm intelligence” methods such as PSO, the bees algorithm and artificial bees colony optimization applying them as statistical classifiers and then as optimizers of the two types of neural networks or as species in a cooperative co-evolutionnary system. We have also used classifiers combination to improve accuracy. The Interest and originality of the presented approaches for characters recognition exist in the aspect of collective intelligence or the cooperation of several MLP or RBF neural networks by offering enhanced results than other approaches employed in literature. The developed approaches were studied by showing their interest in terms of recognition rate.