Face Privacy Protection Method for Autonomous Sensors Based on Hierarchical Format-Preserving Encryption

Area of Science:

• Computer Science
• Cybersecurity
• Automotive Engineering

Background:

• Connected automated vehicles (CAVs) utilize advanced sensors that collect sensitive facial biometric data, creating significant privacy risks.
• Current automotive data privacy solutions often prioritize security over data usability, hindering the effectiveness of downstream applications.
• The need for robust privacy-preserving methods that maintain data utility for environmental perception in CAVs is critical.

Purpose of the Study:

• To introduce a novel hierarchical format-preserving encryption (H-FPE) method for facial data privacy in autonomous sensors.
• To develop a privacy-preserving framework for face detection that balances encryption strength with functional usability.
• To evaluate the effectiveness of the H-FPE method in protecting facial data while preserving essential detection capabilities.

Main Methods:

• The study proposes a region-specific encryption strategy within a YOLOv11-based face detection framework.
• An SM4-based Feistel structure with pseudo-random functions is employed to ensure RGB value constraints and image format integrity.
• Encryption strength is dynamically adjusted based on the importance of different facial regions.

Main Results:

• The H-FPE method achieved over 98% entropy efficiency and a 9.4% improvement in entropy increase compared to traditional methods.
• Pedestrian detection performance was maintained, with F1-scores exceeding 97% and only a 0.5% difference compared to the thumbnail-preserving encryption (TPE) method.
• The proposed method demonstrated substantially stronger privacy protection while preserving critical detection capabilities.

Conclusions:

• The hierarchical format-preserving encryption (H-FPE) method effectively balances privacy protection and functional usability for facial data in autonomous sensors.
• This approach offers a robust solution for protecting sensitive facial data in CAVs without compromising essential detection tasks.
• The region-specific encryption strategy ensures optimal privacy tailored to data sensitivity and application requirements.