Empowering Accessibility in Handwritten Arabic Text Recognition for Visually Impaired Individuals through Optimized Generative Adversarial Network (GAN) model

Abstract

Listen

Translate article

Recognizing handwritten Arabic writing poses unique challenges for individuals with visual impairments due to the diverse range of calligraphic styles employed. This paper presents a novel approach to enhance the precision of recognizing handwritten Arabic language. The method involves utilizing synthetic images created by an optimized generative adversarial network (GAN). A novel GAN architecture is introduced to effectively address the intricacies of Arabic script, considering its diverse forms, variations, and contextual intricacy. The generator is trained based on features such as size, orientation, and style using a conditional GAN architecture. Thanks to style embedding approaches that accurately capture the intricacies of Arabic calligraphy, the generator is now capable of producing text with a significantly enhanced level of authenticity. The Fréchet inception distance and the inception score are metrics utilized to assess the diversity and quality, respectively, of the generated images. The text recognition model is utilized as an inherent evaluation to examine the ability of the GAN to recognize handwritten Arabic text for visually impaired individuals. Utilizing transfer learning techniques and pre-trained convolutional neural networks to extract features enables the GAN to comprehend the patterns of Arabic writing. Hyperparameter tuning involves evaluating different learning rate schedules, batch sizes, denoising filters, and image enhancing techniques to maximize performance. The proposed model achieves an accuracy level of 0.99 and a validation loss level of 0.01 on the specified dataset. The results demonstrate that the proposed optimized GAN architecture is proficient at generating intricate synthetic handwritten Arabic text that closely resembles real-world examples. The internal evaluation findings demonstrate a substantial enhancement in recognition accuracy, thus confirming the effectiveness of the improvements made. This highlights the practical applicability of GANs in handwritten Arabic text recognition. This revolutionary approach based on GANs considers the intricacies of handwritten Arabic text. It enhances Arabic script recognition and creates opportunities for digitizing documents, preserving culture, and advancing natural language processing.