Enhancing Arabic Alphabet Sign Language Recognition with VGG16 Deep Learning Investigation
Ahmed M. Elshaer, Yousef Ambioh, Ziad Soliman, Omar Abdelaziz Ahmed, Miral Elnakib, Mohamed Safwat, Salma M. Elsayed, Mahmoud Khalid · 2024
This paper presents ArAsl (Arabic Alphabet Sign Language) recognition, a system aimed at fostering communication between deaf and hearing individuals by converting Arabic sign language gestures into text or speech. The system utilizes visual recognition of hand gestures from image inputs, employing a novel algorithm that leverages hand geometry and distinct hand shapes for each sign. The Visual Graphics Group (VGG16) model is implemented for letter classification. Through extensive experiments conducted on real-world datasets, our algorithm demonstrates superior performance, outperforming other competitive algorithms. The system achieves an impressive accuracy rate of 96.05% in recognizing Arabic hand sign-based letters, establishing its credibility as a highly dependable solution for facilitating effective communication between the deaf and hearing communities. Further analysis of the confusion matrix and ROC curves reveals particularly strong performance with labels like “ain,” “al,” and “laam,” indicating the model's ability to accurately classify these challenging categories with exceptional frequency.