On Offline Arabic Character Recognition

Machine recognition of characters has received considerable research interest in the area of pattern recognition in the past few decades. This chapter presents the design and implementation of a system that recognizes machine printed Arabic characters. The proposed system consists of four stages: preprocessing of the text, segmentation of the text into individual characters, feature extraction using the moment invariant technique and recognition of characters based on two approaches. In the preprocessing of the text we deal with two problems: isolated pixel removal and drift detection and correction. Next, the given text is segmented into individual characters using horizontal and vertical projection profiles. Moment invariants are used for feature extraction for each character. Finally, the system is trained and the characters are recognized. Recognition of characters is attempted using two approaches, a syntactic approach and a neural network approach.