A generic multi-strategy classification approach for Arabic Handwriting Recognition AHR

Arabic Handwriting Recognition AHR is a current research subject and has tremendous potential in the future because of the great variability of the morphological and typographical features of Arabic handwriting. Nowadays, the combination of classifiers makes it possible to benefit from their possible complementarities, on the one hand, and to improve the quality of an Arabic handwriting recognition system compared to each of classifiers, on the other hand. In this context, our main contribution was the proposal of a word representation model using geometric, statistic and global features and a hybrid recognition approach based on a generic multi-strategy classification algorithm that combines different dynamic classification methods using “Adaboost” and “Bagging” techniques with a static classification method using as a rule of combination “the majority vote”. These methods are applied to a set of statistical classifiers, such as: Support Vectors Machine “SVM”, Multi Layer Perceptron “MLP”, Hierarchical Naïve Bayes “HNB”, and Radial Basis Function Network “RBF Network” in order to increase their performance. The evaluation of our proposed Arabic handwriting recognition system is carried out on the data base from Tunisian cities names IFN/ENIT and has achieved promising results.