This paper introduces a statistical approach for the recognition of Arabic characters. As a first step, the character is segmented into two parts. The first part is dots and Hamza, and the second part is the character body. This reduces the number of character classes from 28 to 18. The second step is the extracting of character features. This is achieved using Zernike moments that are used for mapping character image onto a set of complex orthogonal polynomials. The third step is character classification. A Backpropagation neural network based appmach is used for the classification of Arabic characters represented by invariant features (Zernike moments based features). The approach has been evaluated using printed characters, obtained from scanned documents, with different fonts and sizes from a large database collected for this purpose. The overall recognition rate was 99.3 % for this system.
Arabic diaracters, Zernike moments, Backpropagation
All articles published by Studies in Informatics and Control (SIC) are published in accordance with the terms and conditions of the Creative Commons Attribution License (CC BY-NC-SA 4.0).