Combining Statistical and Structural Approaches for Arabic Handwriting Recognition

A novel technique to recognize full range of shapes of Arabic handwritten characters is presented. This is done by combining statistical and structural approaches for character recognition. The statistical method first recognizes the main body of a character using modified direction features and Support Vector Machines. Then in structural classification, dot-descriptors are used to recognize the exact shape of an Arabic character. On IfN/ENIT benchmarking database, we achieved 96.71% accuracy for character main-bodies and 94.52% accuracy on the complete characters. Our results compare favorably with the state of the art.