Anti-Spoofing Detection Model Using Transfer Learning Techniques for Smart Door Security Systems

Abstract

This study introduces a robust anti-spoofing detection model specifically designed for smart door security systems, targeting critical vulnerabilities present in current facial recognition technologies. Utilising transfer learning-based architectures, particularly VGG16 and MobileNet, the proposed approach integrates pre-trained weights alongside advanced image augmentation techniques to improve the model's capability to identify various spoofing attacks, including print, replay, and 3D mask attacks. The VGG16-based model achieved an impressive accuracy of 98.75%, while the MobileNet-based model recorded an accuracy of 97.82%, showcasing exceptional performance in differentiating between genuine and spoofed images. Evaluations using metrics such as precision, recall, and F1- score further confirmed the robustness and efficiency of the models. With its real-time applicability and computational efficiency, this system is well-suited for deployment in smart homes and IoT-enabled security frameworks. By addressing limitations related to dataset generalisation, robustness, and scalability, this research significantly enhances the reliability and security of biometric-based authentication systems, offering a scalable framework for future smart security applications.