Real masks and spoof faces: On the masked face presentation attack detection

Area of Science:

• Computer Science
• Biometrics
• Security Engineering

Background:

• Face masks are crucial for COVID-19 transmission reduction, complicating face recognition (FR) systems.
• Face presentation attack detection (PAD) is vital for FR system security.
• The impact of masked attacks on PAD performance remains under-explored.

Purpose of the Study:

• To investigate the effect of masked attacks on PAD performance.
• To evaluate the vulnerability of FR systems to masked attacks.
• To introduce novel masked attack methodologies reflecting real-world scenarios.

Main Methods:

• Development of novel masked attacks using real face masks on presentations and subjects.
• Evaluation of seven state-of-the-art PAD algorithms against masked attacks.
• Assessment of FR system vulnerability under masked attack conditions.

Main Results:

• Masked attacks pose a significant threat to the operation and security of FR systems.
• Existing PAD algorithms show reduced performance when faced with masked attacks.
• Real-world masked attacks present a serious security concern for biometric systems.

Conclusions:

• The security of face recognition systems is compromised by masked attacks.
• Further research is needed to develop robust PAD methods against masked threats.
• Adapting FR and PAD systems for mask-wearing scenarios is essential for future security.