Using Convolutional Neural Networks, Arabic Handwritten Character Recognition

Abstract

Recognizing handwritten Arabic numbers is a challenging research topic. Impulsive by this research topic proposed two convolutional neural networks for recognizing Arabic handwritten numerals. Two proposed models have been analyzed using different filter sizes. The Arabic Number dataset exported from Kaggle was trained. The simplest proposed model achieved high recognition accuracy of 99.92%, outperforming the other complex with a more reasonable accuracy. For the MADBase dataset