Liveness Detection

Production face liveness with temporal consistency, behind billion-scale identity checks.

Frame-based liveness detectors at the time of this work (Xu et al., 2021) suffered from prediction inconsistency across time: a single frame might flip between live and spoof, producing flicker that broke the downstream face recognition system. The fix combined a temporal smoothing model with an adapter that turns short-clip predictions into a stable per-session decision.

This paper received the Best Paper Award at the 5th ChaLearn Face Anti-Spoofing Workshop @ ICCV 2021, and the methods were folded into Amazon Rekognition’s online liveness detection. The system has since handled billions of annual identity checks against presentation attacks, and its constraints (legitimate users must not be flagged while attackers shift from physical spoofs to GenAI deepfakes) are exactly what motivated the continuous lifelong-learning research direction that I work on now.

Several patents grew out of this program: liveness from motion, face, and context cues (Xu et al., 2025), gesture/facial-expression/concurrency validation (Zhang et al., 2025), and synthetic-image evaluation of biometric authorization (Xu et al., 2024).

The lifelong-update problem this program raises also motivated newer research on source-free adaptation (Li et al., 2025): an optimal-transport-guided adaptor lets a deployed anti-spoofing model adapt at the client side at test time, with only a small sample of target data and without touching the base model parameters or training set.

References

2025

  1. Patent
    Liveness Detection Based on Motion, Face, and Context Cues
    Xiang Xu, Mingze Xu, Zheng Zhang, and 4 more authors
    2025
    US Patent 12,300,035 B1
  2. Patent
    Liveness Detection Based on Gesture Validation, Facial Expression Analysis, and Concurrency Validation
    Zheng Zhang, Xiang Xu, Hao Chen, and 2 more authors
    2025
    US Patent 12,333,863
  3. Optimal Transport-Guided Source-Free Adaptation for Face Anti-Spoofing
    Zhuowei Li, Tianchen Zhao, Xiang Xu, and 7 more authors
    In IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR), 2025

2024

  1. Patent
    Evaluating Biometric Authorization Systems with Synthesized Images
    Xiang Xu, Hao Zhou, Jonathan Wu, and 1 more author
    2024
    US Patent 12,039,027

2021

  1. On Improving Temporal Consistency for Online Face Liveness Detection System
    Xiang Xu, Yuanjun Xiong, and Wei Xia
    In IEEE/CVF International Conference on Computer Vision Workshops (ICCVW), 2021
    Best Paper Award