Arabic Handwritten Characters Recognition Using Convolutional Neural Network

Abstract

Automatic handwritten characters' recognition is one of Artificial intelligence applications which is considered an interesting research area and important in various fields. Many studies have been conducted for the recognition of English handwritten characters and fewer works are available for the Arabic language because of the diversity in characters' shapes according to their positions in the words. Convolutional Neural Networks are efficient for handwritten characters' recognition. In this paper, a Convolutional Neural Network has been proposed for handwritten characters' recognition. The model has been trained on a dataset of 16,800 images of handwritten Arabic characters with different shapes to perform classification. The proposed model achieved high recognition accuracy of 97.2%, outperforming other state-of-art models. When applying data augmentation, the model achieved better results and accuracy of 97.7%.