Occlusion Resilient Face Recognition and Alert System based on Siamese Network
P Shreyas, Rahul Krishna A, Jay Surya C, Arathi Krishna R, R. Vidhya Lavanya · 2025
Face recognition is widely used in various applications. While current techniques perform well in controlled settings, accurately recognizing occluded faces remains challenging. This is crucial in computer vision for surveillance and public safety, where partial occlusions often compromise identification accuracy. This work aims to develop a model for detecting occluded faces and improving recognition accuracy in partial occlusion scenarios. We also plan to integrate this model with a real-time monitoring module accessible via a web-based GUI. Experiments conducted using the publicly available Labeled Faces dataset, along with a custom dataset, demonstrated 84.49% accuracy for Siamese Model. The system was tested in real-time, providing administrators with email alerts for unknown individuals through an intuitive interface.