基于FHE的高性能隐私保护人脸认证系统
网络出版日期: 2026-05-29
基金资助
国家重点研发计划(2024YFB4505400)
版权
High performance privacy preserving face authentication system based on fully homomorphic encryption
Online published: 2026-05-29
Copyright
本文研究了基于全同态加密(fully homomorphic encryption,FHE)的隐私保护人脸认证系统,目标是在不泄露人脸特征的前提下,实现云端匹配并降低计算与通信开销。基于CKKS(Cheon-Kim-Kim-Song)近似同态计算方案,构建了“端侧加密、云端同态计算、可信机构解密判决”的三层架构,实现了密文特征的生成、传输、存储与匹配。针对128维人脸特征匹配需求,设计了多槽打包与批量余弦相似度评估流程,在密文域并行完成内积、模长等关键计算,且仅返回加密匹配分数。针对同态运算的性能瓶颈,采用巴雷特模约简算法并进行汇编级优化,结合OpenMP多线程技术与指令级并行技术,提升系统计算吞吐量。Intel平台的实验结果表明,优化后的巴雷特模约简算法较普通实现加速约1.96倍。通过配置32个加速线程及特定加密参数(15个模数Q、1个模数P,模数QP的总比特数为881),在批量大小为128的配置下,特征密文匹配时延由68.1 ms降至11.4 ms。表明该系统在隐私保护约束下具备高并发实时认证的性能。
赵野 , 赵文鹏 , 王文超 , 彭惠康 , 柳笛 , 张海春 . 基于FHE的高性能隐私保护人脸认证系统[J]. 网络空间安全科学学报, 2026 , 4(2) : 50 -60 . DOI: 10.20172/j.issn.2097-3136.260212
A privacy-preserving face authentication system based on fully homomorphic encryption (FHE) was investigated, with the objective of enabling cloud-side matching without disclosing facial features and reducing computation and communication overhead. Based on the CKKS (Cheon-Kim-Kim-Song) approximate homomorphic computation scheme, a three-tier architecture of client-side encryption, cloud-side homomorphic computation, and trusted authority decryption and decision was constructed to support the generation, transmission, storage, and matching of ciphertext features. For 128-dimensional facial feature matching, a multi-slot packing and batched cosine similarity evaluation workflow was designed to complete inner product and norm computations in the ciphertext domain in parallel and return only encrypted matching scores. To address the bottlenecks of homomorphic computation, Barrett modular reduction algorithm was adopted and optimized at the assembly level, and OpenMP multithreading technology and instruction-level parallelism were combined to improve the system's computational throughput. Experimental results on an Intel platform indicate that the optimized Barrett modular reduction algorithm achieves a 1.96-fold speedup over the conventional implementation. With 32 acceleration threads and specific encryption parameters (Q=15, P=1, total modulus bit-width 881), the feature ciphertext matching latency for a batch size of 128 was reduced from 68.1 ms to 11.4 ms. The results demonstrate that the system establishes a solid performance foundation for high-concurrency and real-time authentication while strictly complying with privacy-preserving constraints.
表 1 系统设置Table 1 System configuration |
| 配置项目 | 详细参数 |
| 处理器 | Intel Core i9-13900K |
| 计算架构 | x86-64 |
| 核心/线程数 | 24核 (8个性能核 + 16个能效核) / 32线程 |
| 性能核频率 | 4.3~5.8 GHz |
| 能效核频率 | 2.2~3.0GHz |
| 内存 | 64GB DDR5 4800MHz |
表 2 不同模型下的识别准确率Table 2 Recognition accuracy under different models |
| 模型 | 识别准确率 |
| FaceNet | 98.6% |
| VGG-Face | 97.4% |
表 3 模约减算法性能Table 3 Performance of modular reduction algorithms |
| 模约简算法 | 耗时/ms |
| 试商除法 | |
| 普通巴雷特 | |
| 优化巴雷特 | 1846 |
表 4 同态算法的性能结果Table 4 Performance results of homomorphic algorithms |
| 算法 | 单线程耗时/μs | 多线程耗时/μs |
| CKKS加密 | ||
| CKKS密文加密文 | ||
| CKKS密文加明文 | 214 | |
| CKKS密文乘密文 | ||
| CKKS密文乘明文 | ||
| CKKS密文旋转 | ||
| CKKS密文共轭 | ||
| 特征密文匹配 |
| 1 |
Cheon J H, Han K, Kim A, et al. A full RNS variant of approximate homomorphic encryption[C]//Selected Areas in Cryptography – SAC 2018. Cham: Springer International Publishing, 2019: 347-368.
|
| 2 |
Deng Z Y, Chiang H H, Kang L W, et al. A lightweight deep learning model for real-time face recognition[J]. IET Image Processing, 2023, 17 (13): 3869- 3883.
|
| 3 |
Thaher T, Mafarja M, Saffarini M, et al. A comprehensive review of face detection techniques for occluded faces: methods, datasets, and open challenges[J]. Computer Modeling in Engineering & Sciences, 2025, 143 (3): 2615- 2673.
|
| 4 |
Yu Z T, Qin Y X, Li X B, et al. Deep learning for face anti-spoofing: a survey[J]. IEEE Transactions on Pattern Analysis and Machine Intelligence, 2023, 45 (5): 5609- 5631.
|
| 5 |
Gonzalez R C. Digital image processing[M]. 3rd ed. New Delhi: Pearson Education India, 2009: 1-845.
|
| 6 |
Guo G D, Zhang N. A survey on deep learning based face recognition[J]. Computer Vision and Image Understanding, 2019, 189, 102805.
|
| 7 |
Deng J K, Guo J, Xue N N, et al. ArcFace: additive angular margin loss for deep face recognition[C]//Proceedings of the 2019 IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR) . Piscataway: IEEE Press, 2019: 4685-4694.
|
| 8 |
Schroff F, Kalenichenko D, Philbin J. FaceNet: a unified embedding for face recognition and clustering[C]//Proceedings of the 2015 IEEE Conference on Computer Vision and Pattern Recognition (CVPR) . Piscataway: IEEE Press, 2015: 815-823.
|
| 9 |
Parkhi O M, Vedaldi A, Zisserman A. Deep face recognition[C]//Proceedings of the Proceedings ofthe British Machine Vision Conference 2015. British Machine Vision Association, 2015: 41.1-41.12.
|
| 10 |
Cao Q, Shen L, Xie W D, et al. VGGFace2: a dataset for recognising faces across pose and age[C]//Proceedings of the 2018 13th IEEE International Conference on Automatic Face & Gesture Recognition (FG 2018) . Piscataway: IEEE Press, 2018: 67-74.
|
| 11 |
Cheon J H, Kim A, Kim M, et al. Homomorphic encryption for arithmetic of approximate numbers[M]//Advances in Cryptology – ASIACRYPT 2017. ChamSpringer International Publishing2017: 409-437.
|
| 12 |
Brakerski Z, Gentry C, Vaikuntanathan V. (leveled) fully homomorphic encryption without bootstrapping[J]. ACM Transactions on Computation Theory, 2014, 6 (3): 1- 36.
|
| 13 |
Fan J, Vercauteren F. Somewhat practical fully homomorphic encryption[J]. Cryptology ePrint Archive, 2012 (144): 1- 36.
|
| 14 |
Lyubashevsky V, Peikert C, Regev O. On ideal lattices and learning with errors over rings[C]//Advances in Cryptology – EUROCRYPT 2010. Berlin, HeidelbergSpringer, 2010: 1-23.
|
| 15 |
Boemer F, Lao Y X, Cammarota R, et al. nGraph-HE: a graph compiler for deep learning on homomorphically encrypted data[C]//Proceedings of the 16th ACM International Conference on Computing Frontiers. New York: ACM, 2019: 3-13.
|
| 16 |
Cheon J H, Han K, Kim A, et al. Bootstrapping for approximate homomorphic encryption[C]//Advances in Cryptology – EUROCRYPT 2018. ChamSpringer International Publishing, 2018: 360-384.
|
| 17 |
Han K, Ki D. Better bootstrapping for approximate homomorphic encryption[C]//Topics in Cryptology – CT-RSA 2020. ChamSpringer International Publishing, 2020: 364-390.
|
| 18 |
Lee Y, Lee J W, Kim Y S, et al. High-precision bootstrapping for approximate homomorphic encryption by error variance minimization[C]//Advances in Cryptology – EUROCRYPT 2022. ChamSpringer International Publishing, 2022: 551-580.
|
| 19 |
Alexandru A, Kim A, Polyakov Y. General functional bootstrapping using CKKS[C]//Advances in Cryptology – CRYPTO 2025. ChamSpringer Nature Switzerland, 2025: 304-337.
|
| 20 |
Chillotti I, Gama N, Georgieva M, et al. TFHE: fast fully homomorphic encryption over the torus[J]. Journal of Cryptology, 2020, 33 (1): 34- 91.
|
| 21 |
Chillotti I, Joye M, Paillier P. Programmable bootstrapping enables efficient homomorphic inference of deep neural networks[M]//Cyber Security Cryptography and Machine Learning. ChamSpringer International Publishing, 2021: 1-19.
|
| 22 |
Podschwadt R, Takabi D, Hu P Z, et al. A survey of deep learning architectures for privacy-preserving machine learning with fully homomorphic encryption[J]. IEEE Access, 2022, 10, 117477- 117500.
|
| 23 |
Lu W J, Huang Z C, Hong C, et al. PEGASUS: bridging polynomial and non-polynomial evaluations in homomorphic encryption[C]//Proceedings of the 2021 IEEE Symposium on Security and Privacy (SP) . Piscataway: IEEE Press, 2021: 1057-1073.
|
| 24 |
Wang Z W, He H Q, Zhao L T, et al. Chameleon: an efficient FHE scheme switching acceleration on GPUs[J]. IEEE Transactions on Parallel and Distributed Systems, 2025, 36 (11): 2264- 2280.
|
| 25 |
Clet P E, Stan O, Zuber M. BFV, CKKS, TFHE: which one is the best for a secure neural network evaluation in the cloud? [C]//Applied Cryptography and Network Security Workshops. Cham: Springer, 2021: 279-300.
|
| 26 |
Bossuat J P, Mouchet C, Troncoso-Pastoriza J, et al. Efficient bootstrapping for approximate homomorphic encryption with non-sparse keys[C]//Advances in Cryptology – EUROCRYPT 2021. ChamSpringer International Publishing, 2021: 587-617.
|
| 27 |
Wang J D, Yu L X, Zhuang W, et al. Research on vector extension of instruction set architecture[C]//Proceedings of the 2024 3rd International Conference on Cloud Computing, Big Data Application and Software Engineering (CBASE). Piscataway: IEEE Press, 2024: 378-385.
|
| 28 |
Bosselaers A, Govaerts R, Vandewalle J. Comparison of three modular reduction functions[C]//Advances in Cryptology — CRYPTO’ 93. Berlin, HeidelbergSpringer, 2007: 175-186.
|
| 29 |
Dagum L, Menon R. OpenMP: an industry standard API for shared-memory programming[J]. IEEE Computational Science and Engineering, 1998, 5 (1): 46- 55.
|
/
| 〈 |
|
〉 |