Abstract—Automatic offline recognition of handwritten Arabic characters poses a significant challenge in various application domains. While notable progress has been made in this area, challenges remain, particularly regarding children’s handwriting. Indeed, the latter often has less legible characters, which complicates the task of automatic recognition. In this paper, we propose two innovative deep learning-based approaches to effectively identify children’s Arabic handwriting. The developed models are based on the promising architectures of Variational Autoencoders (VAEs) and Generative Adversarial Networks (GANs). They were trained on the Hijja dataset, a rich collection of handwritten Arabic letters by Arabic-speaking schoolchildren in Saudi Arabia in 2019. The rise of generative adversarial learning has sparked particular interest in Variational Autoencoders (VAEs) and GAN networks. However, the intrinsic limitations of GANs in terms of inference have led to the adoption of hybrid models combining the strengths of VAEs and GANs. The results obtained are particularly encouraging, with our hybrid models achieving a remarkable accuracy rate of 97%, surpassing existing models in the literature.
 
Keywords—handwritten Arabic, Variational Autoencoders (VAE), Generative Adversarial Networks (GAN), Hijja database


Copyright © 2024 by the authors. This is an open access article distributed under the Creative Commons Attribution License (CC BY-NC-ND 4.0), which permits use, distribution and reproduction in any medium, provided that the article is properly cited, the use is non-commercial and no modifications or adaptations are made.