A Hybrid NN/HMM Modeling Technique for Online Arabic Handwriting Recognition

TL;DR

This paper introduces a hybrid neural network and HMM approach for online Arabic handwriting recognition, achieving high accuracy by segmenting strokes and combining neural and probabilistic models.

Contribution

The work presents a novel hybrid NN/HMM system that effectively segments and recognizes online Arabic handwriting, outperforming existing systems.

Findings

96.4% character recognition accuracy on ADAB database

Significant improvement over state-of-the-art systems

Effective stroke segmentation using Beta-Elliptical strategy

Abstract

In this work we propose a hybrid NN/HMM model for online Arabic handwriting recognition. The proposed system is based on Hidden Markov Models (HMMs) and Multi Layer Perceptron Neural Networks (MLPNNs). The input signal is segmented to continuous strokes called segments based on the Beta-Elliptical strategy by inspecting the extremum points of the curvilinear velocity profile. A neural network trained with segment level contextual information is used to extract class character probabilities. The output of this network is decoded by HMMs to provide character level recognition. In evaluations on the ADAB database, we achieved 96.4% character recognition accuracy that is statistically significantly important in comparison with character recognition accuracies obtained from state-of-the-art online Arabic systems.8