Arabic Sign Language Hand Gesture Recognition Using the Support Vector Machine Algorithm

Abstract

Arabic Sign Language (ArSL) plays crucial role in facilitating communication for hearing-impaired community in Arabic-speaking countries and hand gesture recognition systems can contribute to improving accessibility and enabling communication and communication with them. Hand gesture recognition (HGR) has wide range of applications, including virtual environments, intelligent monitoring, sign language interpretation, medical systems, etc. Translating Arabic Sign Language using hand gestures and machine learning (ML) algorithms is one of the most important applications we have created. To develop a system for recognizing hand gestures in Arabic Sign Language using SVM, which is one of the widely used machine learning techniques? To develop a powerful classifier for hand gesture recognition By training the model to improve the hyper-level to effectively separate different classes of hand gestures based on the extracted features and evaluating the performance of the classifier using different evaluation metrics to determine its accuracy and generalization capabilities, we need dataset of hand gesture Samples labeled with their corresponding meanings. The dataset will include features extracted from hand gestures, such as hand shape, movement, and position. It should be noted that the accuracy of the recognition system depends on the quality of dataset, feature selection, and SVM parameters. Also, pre-processing steps such as hand segmentation and normalization may be necessary to improve performance. Present paper proposes static hand gesture recognition system for ArSL. Meanwhile, it uses multi-class support vector machine (MSVM) algorithm. The current study discovered a histogram of oriented gradients (HOG) from each sample image. In addition to performing principal component analysis (PCA) on HOG image samples with 100% accuracy. Test results on ArSL showed that this method is very effective and with high accuracy. Whereas, using the Z-score normalization method, the features and sigma belonging to one class became more closely related and separated from the other class.