Scholarlink

Facial recognition technology is pervasive but has serious privacy issues because it stores data centrally. This study introduces a privacy-friendly solution combining FaceNet-based feature extraction with federated learning (FL) and fully homomorphic encryption (FHE). FL allows for decentralized training of models so that sensitive facial information is kept on local devices instead of being transmitted. To further secure, FHE in the form of the CKKS scheme enables computation on encrypted embeddings without decryption, safeguarding data during the process. Experimental results show excellent recognition performance with 94.29% accuracy, 97.86% precision, 94.29% recall, and 92.76% F1-score. The framework presents a secure, scalable solution for privacy-sensitive applications like healthcare and surveillance.

错误：搜索内容不能为空，请输入英文关键词
错误：关键词超出字数限制，请精简
高级检索

Privacy-Preserving Face Recognition System with Federated Learning and Fully Homomorphic Encryption

  • 会议论文

  • 发表日期：

    2026年

查看全文
  • N. Aravindhraj,
  • N. Shanthi,
  • M. Muthuraja,
  • S. M. Jaishruthie,
  • C. M. Mohanraj,
  • T. Muruganantham

摘要

Facial recognition technology is pervasive but has serious privacy issues because it stores data centrally. This study introduces a privacy-friendly solution combining FaceNet-based feature extraction with federated learning (FL) and fully homomorphic encryption (FHE). FL allows for decentralized training of models so that sensitive facial information is kept on local devices instead of being transmitted. To further secure, FHE in the form of the CKKS scheme enables computation on encrypted embeddings without decryption, safeguarding data during the process. Experimental results show excellent recognition performance with 94.29% accuracy, 97.86% precision, 94.29% recall, and 92.76% F1-score. The framework presents a secure, scalable solution for privacy-sensitive applications like healthcare and surveillance.