UiAs: User-Independent 3D Facial Anti-Spoofing via Multi-modal Wireless Signals

Cross-modal subtraction of mmWave and acoustic responses enables liveness detection that generalizes to unseen users without enrollment.

Zhiwei chen; Lebin Lyu; Yimo Zhang; Dingyu Zhong; Yijie Li; Yichao Chen; Dian Ding; Jiguo Yu; Xiaosong Zhang; Yongzhao Zhang· 2026· DOI 10.48550/arXiv.2608.29084

The core problem

Face authentication is widely deployed in security-sensitive applications, yet increasingly realistic 3D spoofing attacks pose growing threats. High-fidelity 3D masks can reproduce facial appearance and geometry, but they cannot replicate the intrinsic physical responses of living tissue. These responses can be actively probed by wireless signals, offering a promising avenue for liveness detection. However, the resulting liveness cues captured by wireless signals are entangled with user-dependent facial geometry, which limits cross-user generalization. UiAs addresses this challenge by presenting a multimodal, user-independent 3D facial anti-spoofing system that leverages electromagnetic (mmWave) and mechanical (acoustic) waves. The two modalities share similar user-dependent geometric variations, allowing UiAs to suppress them through cross-modal subtraction while preserving modality-specific liveness cues. Their complementary physical responses further improve live/spoof discrimination. In practical deployments, multiple materials (e.g., skin, hair, eyeglasses, or face coverings) may also bias liveness representations, while spoofing materials are diverse and open-ended. UiAs tack

Innovation

UiAs was evaluated against real 3D spoofing attacks, including high-fidelity 3D masks. The system achieved an accuracy of 93.25% for unseen users without any user-specific physical-signal enrollment. This result demonstrates strong cross-user generalization, a key challenge in anti-spoofing systems. The evaluation included scenarios with various materials such as skin, hair, eyeglasses, and face coverings, as well as diverse spoofing materials. The skin-anchored contrastive learning effectively mitigated biases from these materials, maintaining robust performance. Comparative analysis with single-modality approaches showed that the multimodal fusion and cross-modal subtraction significantly improved discrimination between live and spoofed faces. The system's ability to operate without enrollment makes it suitable for deployment in security-sensitive applications where user cooperation for enrollment may not be feasible.

Face authentication is widely deployed in security-sensitive applications, yet increasingly realistic 3D spoofing attacks pose growing threats. High-fidelity 3D masks can reproduce facial appearance and geometry, but they cannot replicate the intrinsic physical responses of living tissue. These responses can be actively probed by wireless signals, offering a promising avenue for liveness detection. However, the resulting liveness cues captured by wireless signals are entangled with user-dependent facial geometry, which limits cross-user generalization. UiAs addresses this challenge by presenting a multimodal, user-independent 3D facial anti-spoofing system that leverages electromagnetic (mmWave) and mechanical (acoustic) waves. The two modalities share similar user-dependent geometric variations, allowing UiAs to suppress them through cross-modal subtraction while preserving modality-specific liveness cues. Their complementary physical responses further improve live/spoof discrimination. In practical deployments, multiple materials (e.g., skin, hair, eyeglasses, or face coverings) may also bias liveness representations, while spoofing materials are diverse and open-ended. UiAs tackles both issues through skin-anchored contrastive learning. Evaluated with real 3D spoofing attacks, UiAs achieves 93.25% accuracy for unseen users without user-specific physical-signal enrollment.

UiAs employs a dual-modality sensing architecture that actively probes facial tissue using millimeter-wave (mmWave) and acoustic signals. The core insight is that both modalities capture a mixture of user-dependent geometric features and modality-specific liveness cues. By subtracting the shared geometric component across modalities, UiAs isolates the liveness-specific responses that distinguish living tissue from spoofing materials.

Why it matters

The key innovation of UiAs lies in its exploitation of the shared geometric component across mmWave and acoustic modalities to cancel out user-dependent variations. This cross-modal subtraction is effective because both modalities interact with facial geometry in similar ways, while their liveness cues arise from distinct physical mechanisms: mmWave senses electromagnetic properties like water content and dielectric constant, whereas acoustic waves sense mechanical properties like stiffness and damping. These complementary cues enhance discrimination. The skin-anchored contrastive learning addresses the open-ended nature of spoofing materials by using skin as a consistent anchor. This approach allows the model to generalize to unseen spoofing materials without requiring exhaustive training data for every possible material. However, the system's performance may degrade in scenarios with extreme occlusion or when the face is covered by materials that significantly alter both modalities' responses. Future work could explore additional modalities or adaptive learning to further improve robustness. Overall, UiAs represents a significant step toward user-independent, enrollment-free 3D facial anti-spoofing, with potential applications in mobile authentication, access control, and border security.

Who should read this

CS practitioners and researchers

Opening member content…