Sélection des modèles SVM pour la reconnaissance des caractères arabes manuscrits

Abstract

The work presented in this memorial fits into the framework of the recognition of handwritten Arabic characters recognition, and meets the need to test a new method of learning: the support vector machine (SVM: Support Vectors Machines) applied to the recognition the training is a fundamental characteristic based on the selection and optimization of setting kernel and kinds of classifiers (one against one, one against rest) of model support vector. The choice of these parameters has a great influence on the final classifier performance and also the calculation time. One of this meta-heuristic techniques used for optimization and selection called genetic algorithms (GA) are favorable in such a field because of their characteristics. The proposed system combines the SVM with this selection technology (optimization parameters), for recognition of handwritten Arabic characters recognition in the four forms (isolated, beginning, middle and end) the experiment conducted on a 3920 image database. The proposed contributions have been validated and supported by analyzes showing their advantages and limitations. The results are encouraging and open new research perspectives.