Arabic characters recognition by using multi recognition models system

Abstract EN

In this paper, an Arabic character recognition scheme based on multi recognition models was presented.

Three of these models are artificial neural networks that were used to classify the Arabic character, these networks are : Backpropagation, Self- Organization Map, and Linear Vector Quantization.

Each network was trained and tested using the same clean and noisy Arabic character patterns.

In the proposed scheme, the unknown character is checked by the three networks and those outputs are fed to the post processing which compare the three outputs together, and if two at least of outputs are the same then the final decision represent this character, otherwise, the three output are fed to the Dynamic Time Warping to gives the final decision concerning the identity of unknown character.

Simulation results are provided and show that the proposed scheme produces higher success rates by numerical measures compared with the results of each model alone when they were applied to classified Arabic characters with different noise ratios.

The current scheme is efficient and can be extended by adding other methods as classifiers.