Identifying individuals within masked groups: Challenges and solutions using deep neural networks

With the spread of any infectious virus, wearing masks has become essential to reduce transmission, as recommended by global health organizations. However, masks pose challenges to traditional face recognition systems by obscuring key facial features, this paper presents a sophisticated deep learning-based model for masked face recognition, proven effective in both individual and group scenarios. The approach achieves high accuracy of up to 99.8% by integrating an MTCNN for accurate face detection, a FaceNet for high-accuracy feature extraction, and a ResNet-18 network for identity classification, using a dataset of undergraduate students captured under various lighting conditions and angles. Initial experiments focused on single face recognition, and the framework was subsequently modified to include group images using InsightFace (RetinaFace) for multi-face detection and DeepFace (FaceNet) for identity matching, while maintaining high accuracy. This extension enables the system to operate efficiently in crowded environments, ensuring its practicality. In addition to the technical contributions, the system is designed to address key constraints related to scalability and ethical compliance during implementation. Extensive testing on diverse datasets has demonstrated the robustness of the framework and its superior performance compared to standard recognition methodologies. Overall, this study contributes to the development of face recognition technology by providing a robust and scalable solution for identifying masked faces across diverse practical applications.