Liveness Detection for Emotion AI
Anti-spoofing mechanisms to distinguish genuine human expressions from synthetic inputs
Liveness detection is a critical component in ensuring the authenticity of emotional input signals. Our research develops advanced anti-spoofing mechanisms that can reliably distinguish between genuine human emotional expressions and synthetic or manipulated inputs, such as deepfake videos, printed photos, or 3D masks.
Research Focus: Developing robust face anti-spoofing systems that maintain performance across different devices, lighting conditions, and presentation attacks while ensuring real-time efficiency for practical deployment.
Key Contributions
- DiffFAS Framework: First introduction of generative diffusion models to face anti-spoofing tasks through spatio-temporal progressive denoising mechanisms
- DADM Method: Dual alignment of domain and modality for enhanced face anti-spoofing performance
- Consistency Regularization: Novel regularization techniques for deep face anti-spoofing systems
- Cross-domain Robustness: Developing models that maintain performance across different devices, lighting conditions, and presentation attacks
- Real-time Detection: Efficient algorithms suitable for real-world deployment in emotion AI systems
DiffFAS: Face Anti-Spoofing via Generative Diffusion Models
European Conference on Computer Vision (ECCV), 2024
source code
DADM: Dual Alignment of Domain and Modality for Face Anti-spoofing
International Conference on Computer Vision (ICCV), 2025
source code
Consistency Regularization for Deep Face Anti-Spoofing
IEEE Transactions on Information Forensics and Security, Vol. 18, pp. 1127-1140, 2023
source code