Exploring 2D and 3D facial recognition systems and its security
Author:
Zakaria Frane ’28Co-Authors:
Faculty Mentor(s):
SingChun Lee, Computer ScienceFunding Source:
The emerging scholar programAbstract
Two-dimensional (2D) face recognition is widely deployed for authentication and security, yet it remains vulnerable to spoofing attacks using printed photos, replay videos, and deepfakes. Its reliability also degrades when real-world conditions drift from the enrollment image, especially under changing lighting, facial
expressions, and occlusions such as masks or glasses, creating ongoing concerns about both security and robustness. Three-dimensional (3D) facial recognition is often presented as a stronger alternative because it can exploit depth and facial geometry, but its practical resilience and attack surface under adversarial conditions still need clearer, reproducible evidence.
This study evaluated and compared the accuracy, robustness, and security of 3D facial recognition systems against 2D baselines, benchmarking open-source pipelines across public datasets and controlled laboratory experiments. Testing systematically varies illumination (direction and intensity), expression changes, and occlusions, measuring performance in verification and identification tasks. To assess security, the systems were exposed to spoofing attempts using printed media and 3D-printed facial models, recording attack success rates and characteristic failure patterns.
We revealed where 3D methods provide meaningful gains (for example, reduced sensitivity to harsh lighting and certain occlusions) and where weaknesses persisted (for example, vulnerability to high-fidelity 3D replicas or sensor-specific artifacts). By linking failure modes to specific conditions, the work aims to propose targeted upgrades, such as depth-consistency checks, temporal liveness cues, and multi-modal fusion, to harden 3D recognition. Overall, the research clarified trade-offs between 2D and 3D facial recognition and supports the development of more robust and secure 3D authentication in realistic environments.