A Multi-Agent Design for Pattern Recognition Systems based on Multi-classifiers Context

Pattern recognition has to face very complex tasks, often requiring collaboration between several classifiers to perform a well classification performance and to have good generalization ability. Multi-classification systems are mechanisms for taking into account the decision of a set of classifiers to ameliorate the generalization performance. Therefore such systems are dependant from the contributed classifiers that haven’t the possibility to modify their behaviors. In this paper, we propose to design Pattern Recognition system using Multi-classification context with cooperative classifiers and negotiation context. In this purpose multi-classification systems are viewed from both the static and the dynamic points of view. Our main goal is to provide a framework to perform a rigorous agent-based design process for this kind of systems both in the case of a single classifier decision, and combined multi-classifiers scenario. Our methodology is based on the classifiers’ decisions as well as the result of the combination of these decisions with combination methods. Classifiers are here considered as agents with cognitive comportment and progressives’ abilities of live. Combination method is used as additional technique for delivering decision.