Large Vocabulary Hybrid DNN/HMM Arabic Online Handwriting Recognition System

Abstract

Online Arabic handwriting recognition is a di cult problem since it is naturally both cursive and unconstrained. The analysis of Arabic script is further com-plicated due to obligatory dots/stokes that are placed above or below most letters and usually written de-layed in order. In addition, Arabic language is rich in morphology and syntax which makes it a must for a good online handwriting system to handle large vocabulary lexicon. Previously, Hidden Markov Model (HMM) with sequence reordering have provided a successful solution for most of the di culties inherent in recognizing Arabic handwriting. Recently, Deep Neu-ral Networks (DNN) have shown to provide signi cant improvement when integrated with HMM. In this paper we introduce the e orts done to build a large vocabulary Arabic HWR system using hybrid DNN/HMM model. This system used over segmentation to provide e cient decoding. The developed system was tested using a test set of 12k words written by 100 writers with lexicon size of 125k words. The system achieved an accuracy of 71.62%, 89.61% in rst recognized word and top ve recognized words respectively which to our knowledge is the best reported result for large vocabulary Arabic HWR.