SVM synthesis by hierarchical structures of learning automata application for handwritten digits recognition

In this paper, a new SVM (Support Vector Machines) synthesis method is presented. This method is based essentially on training criterion optimization of this machine by a set of hierarchical structures of learning automata. This methodology is adopted for the development of off-line isolated handwritten digits recognition system. A comparison is taken between this new approach and that of a standard approach for SVM synthesis. These two methodologies are also compared with a neural network based classification method. The obtained results show the performances of the new suggested method for SVM synthesis.