Online Arabic handwriting recognition using continuous Gaussian mixture HMMS

In this paper, we present a recognizer structure aimed at recognizing online Arabic handwriting written in continuous form. The basic units of recognition used are strokes, which are sub-letter parts. To recognize strokes we used hidden Markov models (HMMs) to model each stroke. Decision logic was then used to interpret the output of stroke HMMs, converting their output into recognized-words. Data collected from six writers was used to validate the functionality of the system. Experimental simulation of the proposed system resulted in promising recognition rates (>75%), which is significantly better than currently available solutions.