Abstract
Fast Fourier transform (FFT) is used successfully in computing the Fourier descriptors which are used in object and character recognition. In this article, an Arabic character recognition algorithm using modified fourier spectrum (MFS) is presented. Ten descriptors are estimated from the Fourier spectrum of the character contour by subtracting the imaginary part from the real part (and not from the amplitude of the Fourier spectrum as is usually the case). Ten MFS descriptors are extracted and used for the recognition of Arabic characters. Experimental results using 10 MFS descriptors resulted in an average recognition rate of 95.9%. The analysis of the sparse matrix indicates that the major part of the errors is due to few similar characters. The new technique, based on MFS descriptors, was compared with the Fourier descriptors calculated from the amplitude of the FFT spectrum. Experimental results have shown that the MFS-based technique is faster to compute than the FFT-based technique. However, the Fourier descriptors, initially, have a better recognition rate than MFS descriptors (96.9% vs. 95.9%). Using the holes' and dots' features to resolve the problematic characters reduces the error rate of the MFS technique more than that of the Fourier descriptor technique. This article introduced MFS-based features that are faster to compute than Fourier descriptors and have fewer errors utilizing the dots and holes features of Arabic characters. Both techniques may be used in combination or in a multi-classifier system to enhance the Arabic recognition system rate.