Features selection techniques for off-line handwritten isolated Arabic characters

Abstract

Offline Handwritten isolated Arabic characters’ software has become a highly demand application to the machine reading of bank and post offices. In the past few years, several approaches have been used in the development of handwritten recognition applications. However, the recognition of handwritten Arabic characters is a difficult task because of the similar appearance of some different characters.In this study, the moments: contour sequence, geometric and Zernike moments are employed on handwritten characters to select the efficient features. The classification and recognition process are applied using Neural Network technique and the results are analyzed to determine the necessity of thinning and unthinning processes. The database consists of 6885 images of characters: 75% of training and 25% of testing in the network. Matlab tool is implemented to perform the classification and recognition processes. Results obtained have shown that thinning process should be excluded as it deteriorates the recognition accuracy. The experiments resulted 97.58% in Contour Sequence moments with unthinning for classification and 95.25% for recognition process. Thus, Contour Sequence moments with unthinning process exhibited the highest recognition rate as compared to Geometric moments and Zernike moments.