Arabic handwritten character recognition

Abstract

The main objective of this work is to produce an automatic system that can accurately recognize handwritten Arabic characters. This system should be able to extract images of characters, read and process each character, and identify the class of the character as one of the twenty-eight Arabic alphabets. The system is also expected to recognize different styles of handwriting. This project uses a database that contains images of handwritten Arabic characters where each character has a number of samples reflecting its four possible shapes: stand-alone, beginning, middle, and end. The developed recognition system consists of three phases: preprocessing, in which data is subjected to noise removal, binarization, and translation and scale normalization operations; feature extraction, which is achieved by extracting three types of features, namely: Zernike moments of up to the 20th order, the seven classical moments by Hu, and horizontal and vertical image projections; and finally classification, which is realized by a back-propagation feed-forward neural network classifier. The decision making is made by minimum Euclidean distance matching. The three phases of the system are implemented by MATLAB 7.1-based programmes. The simulation results of the experiments were reported. The results show that the proposed system achieves good recognition rates compared to other existing systems. The system can be successfully adapted to meet the demands of a form-reader application.