Recognition of Arabic digits using a convolutional neural network
Mouhssine El Atillah, Khalid El Fazazy, Jamal Riffi · 2022
Arabic is the most widely spoken language in the world. However, optical recognition of Arabic handwriting by deep learning networks remains inadequate. Recently, some studies have been based on this field and have produced remarkable results in the recognition of alphabets and Arabic numerals. This article focuses on Arabic numbers recognition problem. We use a convolutional neural network with minimal parameters to overcome the overfitting problem. Preceded by the morphological gradient method to detect images contours. This model applies to the Arabic manuscript numbers database, which consists of 70,000 images available in Kaggle [1]. Our model provides 99.80% classification accuracy with a minimum loss of 0.96%.