RECOGNITION OF ARABIC HANDWRITTEN CHARACTERS USING RESIDUAL NEURAL NETWORKS

Abstract

This study proposes the use of Residual Neural Networks (ResNets) to recognise Arabic offline isolated handwritten characters including Arabic digits. ResNets is a deep learning approach which showed effectiveness in many applications more than conventional machine learning approaches. The proposed approach consists of three main phases: pre-processing phase, training the ResNet on the training set, and testing the trained ResNet on the datasets. The evaluation of the proposed approach is performed on three available datasets: MADBase, AIA9K, and AHCD. The proposed approach achieved accuracies of 99.8%, 99.05% and 99.55% on these datasets, respectively. It also achieved a validation accuracy of 98.9% on the constructed dataset based on the three datasets. [JJCIT 2021; 7(2.000): 192-205]

Keywords

• residual network
• deep learning
• deep neural networks
• arabic handwritten
• characters recognition