DEFAEK: Domain Effective Fast Adaptive Network for Face Anti-Spoofing

Area of Science:

• Computer Science
• Artificial Intelligence
• Machine Learning

Background:

• Current deep learning face anti-spoofing (FAS) models are resource-intensive, requiring large datasets and powerful hardware.
• These models struggle with generalization to unseen spoof types and environmental variations, limiting their real-world applicability.

Purpose of the Study:

• To develop a fast-learning, lightweight, and robust face anti-spoofing approach.
• To enable effective adaptation to new spoof types and environmental domains without extensive retraining.

Main Methods:

• Proposed Domain Effective Fast Adaptive nEt-worK (DEFAEK), an optimization-based meta-learning framework.
• Treated environmental differences as domains, simulating domain shifts during training.
• Incorporated metric learning with careful sample selection for enhanced meta-learning efficiency.

Main Results:

• DEFAEK demonstrated strong generalization capabilities, learning environment-independent features.
• The lightweight model achieved robust performance on unseen spoof classes.
• Achieved competitive results compared to state-of-the-art methods in terms of parameters, FLOPS, and accuracy.

Conclusions:

• DEFAEK offers an efficient and effective solution for face anti-spoofing on resource-constrained devices.
• The meta-learning approach enables rapid adaptation and robust generalization against diverse presentation attacks.