Comprehensive Vulnerability Evaluation of Face Recognition Systems to Template Inversion Attacks via 3D Face...

Area of Science:

• Computer Vision
• Biometrics
• Cybersecurity

Background:

• State-of-the-art face recognition systems are susceptible to sophisticated attacks.
• Template inversion attacks pose a significant threat to biometric data security.
• 3D face reconstruction offers a novel approach to exploit facial templates.

Purpose of the Study:

• To comprehensively evaluate the vulnerability of face recognition systems to template inversion attacks using 3D face reconstruction.
• To propose and validate a new method (GaFaR) for 3D face reconstruction from facial templates.
• To assess the effectiveness of proposed attack methods in both whitebox and blackbox scenarios.

Main Methods:

• Developed GaFaR: a novel method for 3D face reconstruction from facial templates using a pretrained geometry-aware face generation network.
• Employed a semi-supervised approach with real and synthetic face images to train a mapping from facial templates to the latent space of a face generator network.
• Utilized generative adversarial network (GAN)-based framework for real images and direct mapping for synthetic images.
• Optimized camera parameters of the GNeRF model to enhance attack success rates.
• Conducted whitebox and blackbox attacks against face recognition systems.

Main Results:

• Demonstrated the effectiveness of the proposed GaFaR method in reconstructing 3D faces from templates.
• Achieved high attack success rates in both whitebox and blackbox scenarios.
• Evaluated the transferability of the attack across different face recognition systems on MOBIO and LFW datasets.
• Successfully performed practical presentation attacks using digital screen replay and printed photographs.

Conclusions:

• Face recognition systems are vulnerable to template inversion attacks facilitated by 3D face reconstruction.
• The proposed GaFaR method offers a potent tool for assessing and enhancing the security of face recognition systems.
• Findings highlight the need for developing more robust defenses against advanced biometric attacks.