Alphabet Recognition in Arabic Sign Language: A Machine Learning Perspective

Khattab, Mahmoud M.; Zeki, Akram M.; Matter, Safaa S. Matter; Abdella, Mohamed A.; Atiia, Rada A. E.; Soliman, Amr Mohmed

doi:10.21608/qarts.2024.267418.1882

Alphabet Recognition in Arabic Sign Language: A Machine Learning Perspective

نوع المستند : المقالة الأصلية

المؤلفون

¹ Faculty of Information and Communication Technology, International Islamic University Malaysia, Kuala Lumpur, Malaysia

² Department of Computer Science, Applied College, King Khalid University, Abha, Saudi Arabia

³ Faculty of Social Sciences, Imam Muhammad Ibn Saud Islamic University, Saudi Arabia

⁴ College of Education, King Khalid University, Abha, Saudi Arabia

⁵ 4College of Education, King Khalid University, Abha, Saudi Arabia

10.21608/qarts.2024.267418.1882

المستخلص

Pattern recognition in human-computer interaction systems has gained significant attention in recent years, particularly in computer vision and machine learning applications. One prominent application is the recognition of hand gestures used in communication with deaf individuals, specifically in identifying the dashed letters within Quranic surahs. This paper proposes a new alphabet-based Arabic sign language recognition model, which employs a vision-based approach. The system comprises four stages: data acquisition, data preprocessing, feature extraction, and classification. The proposed model accommodates three types of datasets: bare hands against a dark background, bare hands against a light background, and hands wearing dark-colored gloves. The process begins with capturing an image of the alphabet gesture, followed by hand separation and background isolation. Hand features are then extracted based on the chosen method. In terms of classification, supervised learning techniques are employed to classify the 28-letter Arabic alphabet using 9,240 images. The focus is on classifying the 14 alphabetic letters representing the initial Quranic surahs in the Quranic sign language. The experimental results demonstrate that the new proposed model has achieved an impressive accuracy of 99.5% using the k nearest neighbor classifier.

الكلمات الرئيسية