Evaluating and Enhancing Face Anti-Spoofing Algorithms for Light Makeup: A General Detection Approach.

Area of Science:

• Computer Science
• Biometrics
• Artificial Intelligence

Background:

• Face anti-spoofing systems are crucial for security but are often challenged by subtle facial modifications like light makeup.
• Existing research and datasets largely overlook the impact of light makeup on face anti-spoofing performance.
• Inaccurate spoofing detection due to light makeup can cause significant user inconvenience.

Purpose of the Study:

• To address the lack of data and algorithms for face anti-spoofing with light makeup.
• To evaluate the performance of current anti-spoofing algorithms on lightly made-up faces.
• To propose a novel, robust face anti-spoofing algorithm specifically for light makeup scenarios.

Main Methods:

• Creation of a new face anti-spoofing database incorporating light makeup faces.
• Assessment of existing face anti-spoofing algorithms using the novel database.
• Development of a specialized algorithm featuring makeup augmentation, batch channel normalization, Exponential Moving Average (EMA) backbone updates, asymmetric virtual triplet loss, and nearest neighbor supervised contrastive learning.

Main Results:

• Established face anti-spoofing algorithms showed reduced performance on light makeup faces.
• The proposed specialized algorithm demonstrated superior detection capabilities for light makeup faces.
• The new database and evaluation provide a benchmark for future research in this area.

Conclusions:

• Light makeup presents a significant challenge for current face anti-spoofing technologies.
• The developed algorithm effectively enhances face anti-spoofing accuracy in the presence of light makeup.
• Further research is needed to address the nuances of facial variations in biometric security systems.