Arabic Handwritten Word Recognition based on Dynamic Bayesian Network

Distinguishing an Arabic handwritten text is a hard task because the Arabic word is morphologically complex and the writing style from one model is highly variable, like the recognition of words representing the names of Tunisian cities. Actually, this is the first work based on the Dynamic Hierarchical Bayesian Network (DHBN). Its objective is to get the best model by learning the structure and parameter of Arabic handwriting to decrease the complexity of the recognition process by allowing the partial recognition. In fact, we propose segmenting the word based on a vertical smoothed histogram projection using various width values to put down the segmentation error. After that, we extract the characteristics of each cell using the Zernike and HU moments, which are invariant to rotation, translation and scaling. Then, the sub-character is estimated at the lowest level of the Bayesian Network (BN) and the character is estimated at the highest level of the BN. The overall Arabic words are processed by a dynamic BN. Our approach is tested using the IFN/ENIT database, where the experiment results are very promising.