A System for Off-Line Arabic Handwritten Word Recognition Based on Bayesian Approach
Akram Khémiri, Afef Kacem Echi, Abdel Belaı̈d, Mourad Elloumi · 2016
In this work, a system based on a Bayesian approach, for the off-line recognition of handwritten arabic words, is proposed. Different structural features such as ascenders, descenders, loops and diacritic, are extracted from word's image, tacking into account the morphology of handwritten arabic words. For accurate features extraction, we proposed a novel method to estimate the word's baseline and evaluated it using the IFN-ENIT Tunisian city names dataset ground-truth. The extracted features are used as input to some variants of Bayesian networks, notably Naïve Bayes (NB), Tree Augmented naïve bayes Network (TAN), Horizontal and Vertical Hidden Markov Model (VH-HMM) and Dynamic Bayesian Network (DBN). Results are reported on the benchmarking IFN/ENIT which indicate the robustness and the effectiveness of the proposed approach. The best word recognition rate we obtained achieves 90.02% for the bi-stream VH-HMM.