Hybrid close-up model for active face liveness.

Area of Science:

• Computer Science
• Biometrics
• Artificial Intelligence

Background:

• Face Anti-Spoofing (FAS) is crucial for secure facial authentication.
• Active liveness detection analyzes user behavior and signals for authenticity.
• Current active liveness detection faces challenges due to limited public datasets and reproducibility issues.

Purpose of the Study:

• Introduce a new dataset, UFPR-Close-Up, for active liveness detection research.
• Propose a novel active spoof detector combining distortion features and spatial embeddings.
• Enhance the scientific community's ability to develop and evaluate face anti-spoofing solutions.

Main Methods:

• Collected the UFPR-Close-Up dataset featuring genuine and spoof samples during close-up user interactions.
• Generated spoof samples using diverse face images and presentation attack instruments.
• Developed a new active spoof detector integrating distortion features with spatial embeddings.

Main Results:

• Achieved an Average Classification Error Rate (ACER) of 4.33% on the full-data protocol.
• The proposed detector consistently outperformed existing active liveness models across four evaluation protocols.
• The UFPR-Close-Up dataset facilitates reproducible research in face anti-spoofing.

Conclusions:

• The UFPR-Close-Up dataset addresses the need for public, shareable data in active liveness detection.
• The novel detector demonstrates superior performance in distinguishing genuine faces from spoof attempts.
• This work contributes to more robust and reliable facial authentication systems.