Privacy-preserving method for face recognition based on homomorphic encryption

Area of Science:

• Computer Science
• Biometrics
• Cryptography

Background:

• Facial recognition technology is widespread, but plaintext data storage and processing pose significant privacy risks.
• Existing methods lack robust privacy protection for sensitive facial data.

Purpose of the Study:

• To develop a privacy-preserving facial recognition framework using approximate homomorphic encryption (HE_FaceNet).
• To enhance the computational efficiency of HE_FaceNet through a clustering-based optimization scheme.

Main Methods:

• Facial feature templates extracted using a pre-trained model and then encrypted.
• Encrypted template matching via Euclidean distance, with recognition after decryption.
• Clustering algorithms applied to group similar encrypted facial features for accelerated searching.

Main Results:

• HE_FaceNet effectively protects facial data privacy while maintaining high recognition accuracy.
• The clustering optimization significantly improves computational efficiency.
• The optimized framework demonstrates high accuracy and efficiency across various facial datasets.

Conclusions:

• HE_FaceNet provides a viable solution for privacy-preserving facial recognition.
• Clustering-based optimization is crucial for the practical application of HE-based facial recognition systems.