On-line Arabic Handwritten Personal Names Recognition System Based on HMM

In this paper a new on-line handwriting recognition system for Arabic personal names based on Hidden Markov Model (HMM) is presented. The system is trained with the ADAB-database using two different methods: manually segmented characters and non-segmented words. This work presents a recognition system dealing with a large vocabulary of 2800 Arabic personal names using a new lexicon reduction method that depends on the delayed strokes formation and the number of strokes. Besides, a new delayed strokes detection method is used to reduce the temporal variation of the on-line sequence. A dataset of on-line Arabic handwritten names has been collected to validate the system and a highly encouraging recognition rate is achieved compared to the results of commercially available recognition systems on the same dataset.