People Identification Through Facial Recognition and Anti-Spoofing Using Deep Learning

Abstract

This research effort uses cutting-edge anti-spoofing techniques in conjunction with deep learning approaches to address the issue of spoofing assaults on facial recognition systems. A diversified dataset containing real facial photos and several spoofing attack scenarios is compiled as the project's first step. Then, data pretreatment methods are used to guarantee data consistency and the best model performance. The research makes use of MobileNet and VGG-16, two well-known deep-learning architectures, to build reliable facial recognition models. A thorough evaluation using well-established metrics including classification reports, accuracy scores, and confusion matrices is undertaken after thorough training and validation. It's significant because this research incorporates real-time anti-spoofing capabilities, which go beyond traditional facial recognition jobs. Webcam functionality is added to the deployed models to assess real-time images in comparison to reference passport-size photos. Dynamically shifting boundary box colors—blue for real faces and red for detected fake images—indicate the anti-spoofing technology. The project's conclusion contains a thorough comparison of the MobileNet and VGG-16 models that identifies and compares each model's advantages and disadvantages. Real-time demos also highlight the anti-spoofing methodology's effectiveness in practice.