Deep Learning-Based Siamese Neural Network for Masked Face Recognition

Abstract

Face recognition is one of the essential elements of security and surveillance, and is widely used in mobile devices and public surveillance systems. However, face blockage poses a significant challenge for developing practical and effective applications in this area. With the spread of the COVID-19 pandemic, wearing face masks has become essential in public places to reduce the transmission of the virus. Despite their health importance, masks have complicated the task of facial recognition, as the face is only partially exposed, hindering traditional recognition systems and increasing the difficulty of the work of security personnel. In this study, the Siamese neural network was used as an innovative approach to human face recognition under partial face-occlusion. The model's performance was tested using RMFRD and MFR2 databases, and the results showed high accuracy. On the RMFRD dataset, the ResNet50, EfficientNet, and Xception models achieved 99% accuracy, while the MobileNet model achieved 99% accuracy. On the MFR2 dataset, the ResNet50, EfficientNet, and Xception models achieved 99% accuracy, while the MobileNet model achieved 98% accuracy. This approach shows great effectiveness in dealing with the challenges posed by face obscuration, as a result enhancing the capabilities of facial recognition systems in real-life scenarios and enabling their more efficient use in security and surveillance. The Siamese network has proven to be highly effective at recognizing masked faces, making it highly relevant for applications in security, human-computer interaction, and many other areas affected by the COVID-19 pandemic.