Using Mel Frequency Cepstral Coefficient method for online Arabic characters handwriting recognition

This paper presents an approach to extract features of online Arabic handwritten characters by combining offline features with Mel Frequency Cepstral Coefficients (MFCCs); indeed, these latter are commonly used as features in speech recognition systems. In this work, we have adapted MFCC method to online handwriting recognition area, and we investigate the classification performance of the MFCC with Hidden Markov Models (HMMs) for online Arabic handwritten character recognition, by varying some MFCC and HMM parameters such as sampling frequency, frame size, frame increment and number of HMM states. Besides, we have proposed a new solution of the problem of distributing points unevenly along the stroke curve, due to the variation in writing speed. This solution is appropriate for the online Arabic handwriting recognition systems for the reason of preserving information of the original character signal. The proposed system is evaluated using NOUN dataset and it gives an excellent recognition rate up to 96% which outperforms that reported by NOUN dataset owner in [1,2].