Building a Convolutional Neural Network from Scratch for Arabic Handwritten Recognition

Arabic character recognition has attracted significant attention due to its crucial role in various applications. Despite significant advancements in deep learning, Arabic handwritten text recognition remains a challenging task due to the language’s complex script, high variability in handwriting styles, and connectivity between letters. In this study, we propose a novel approach to Arabic character recognition using a Convolutional Neural Network (CNN) developed from the ground up. The architecture of our CNN is meticulously designed to capture the intricate features inherent in Arabic characters by leveraging the representational power of deep learning. The model is trained and evaluated on two well-established benchmark datasets: AHAWP and IFHCDB. Experimental results demonstrate the model’s effectiveness in addressing the complexities of Arabic character recognition, achieving competitive accuracy 93.95% and 93.51% on IFHCDB AHAWP respectively while surpassing certain existing approaches. Furthermore, our methodology integrates architectural innovations and dataset exploration, incorporating the recently introduced AHAWP dataset to advance the field of Arabic character recognition.