Preprocessing Algorithms for Arabic Handwriting Recognition Systems

Preprocessing is one of basic phases of handwritten text recognition and it is crucial to reach high recognition rate. In this paper, we present several algorithms for Arabic handwritten text which are based on inherent properties of Arabic writing. These algorithms include noise removal and smoothing, diacritics detection, contour tracing/correction, baseline estimation, slope correction and detecting/correcting touching descenders. In first stage of propositions validation, each presented method is individually tested on the commonly used IFN/ENIT database. Then, the influence of the presented algorithms on the recognition rate is studied based on K-NN classifier and hybrid features. The obtained results show the efficiency of the proposed algorithms and their positive impact on features discrimination and recognition performance.