COMPARATIVE STUDY AN D NEW APPROACH MULTI CLASSIFIERS: APPLICATION TO THE RECOGNITION OF ARABI C NUMERALS

Generally, most of the problems of recognition are due to the classificat ion of the feature vectors, or to the mismatch between learning conditions and test, so to improve the robustness of the existing recognition systems, it is necessary to efficient extraction, to reduce these differences between the reference corpus (learning) and test, and finally achieve a reliable and valid comparison. In fact , the task at first is devoted to study and present the different classifications algorithms (classifiers) and fly on the theoretical basis of algorithms recognition, which allows us to address the principle of oper ation of systems recognition in their totality. And in a second step, we compared the different classification algorithms to identify the advantages and disadvantages of such algorithms, and assess the level and performance of the system already made, to d irectly filter and keep only the positive side of each method . In the end, an analysis and comparison of the results were made along this work, which has led us to propose a hybrid classifier, and provides us a significant increase in the rate of speech re cognition perfor ming a biometric authentication tool powerful enough as possible.