Arabic Handwriting Recognition Using Variable Duration HMM

The present paper describes a complete system for the recognition of unconstrained handwritten Arabic words us- ing over-segmentation of characters and Variable Duration Hidden Markov Model (VDHMM). First, a segmentation algorithm is used to translate the 2-D image into 1-D se- quence of sub-character symbols. This sequence of symbols is modeled by the VDHMM. The shape information of char- acter and sub-character symbols is compactly represented by forty-five features in the feature space. The feature vec- tor is modeled as an independently distributed multivariate discrete distribution. The linguistic knowledge about char- acter transition is modeled as a Markov chain where each character in the alphabet is a state and bigram probabili- ties are the state transition probabilities. In this context, the variable duration state is used to resolve the segmentation ambiguity among the consecutive characters. Using Arabic handwritten data from two different sources, detailed exper- imental results are described to demonstrate the success of the proposed scheme.