Histogram Clustering and Hybrid Classifier for Handwritten Arabic Characters Recognition

M. Syiam, T.M. Nazmy, A.E. Fahmy, H. Fathi, and K. Ali (Egypt)


Arabic OCR, histogram clustering, decision tree, MLP. 1-


Segmentation of the handwritten Arabic characters is still one of the most difficult problems to develop a reliable Arabic OCR. This paper presents a complete Arabic OCR system that uses histogram clustering method for the segmentation of the Arabic word. This method gives the ability to process different user styles, and manages the variability of pen strokes. Also, a new algorithm for separating overlapped characters was proposed to support the proposed technique for segmentation. The feature extraction process was based on a combination between the PCA network and characters geometric features. A classifier for hundred of Arabic character images was designed using a decision tree induction algorithm, and MLP network. A segmentation correctness of 96% was achieved while the recognition rate of the whole system was 91.5 %.

Important Links:

Go Back