FHE-相容的分布式人脸识别方案设计与分析

Design and analysis of FHE-compatible distributed face recognition scheme

密码协议与人脸识别的结合需解决两者的相容性、因协议约束导致的低识别率和低效率问题。融合C/S和管道-过滤器结构提出分布式人脸识别系统(DFRS)体系结构,引入并改进整数环上的全同态加密(FHE),以此为基础设计远程管道协议,提出FHE-相容的DFRS方案(FHE*DFRS)。采用与FHE相容的欧式距离分类,采用Gabor小波和主成分分析进行特征提取,提高识别率,将欧式距离计算转化为点积计算,降低轮次复杂度。实验验证与分析结果表明,该方案能够保护隐私,具有较高的识别率和效率。

Combining cryptographic protocol with face recognition leads to some problems such as the compatibility of the two classes of algorithms,low recognition rate and low efficiency caused by protocol constraints.The distributed face recognition system architecture was proposed by combining C/S distributed model with pipe-filter architecture,fully homomorphic encryption algorithm over integer domain was introduced and improved to design remote pipe protocol and FHE-compatible distributed face recognition scheme（FHE＊DFRS）.Classifier algorithm of Euclidean distance metric was utilized to solve the compatibility problem.Furthermore,Gabor filter and principal component analysis method were combined to extract features to increase the recognition rate,and the Euclidean distance computation was converted into dot product computation to reduce rounds complexity.The experiments and analysis demonstrate that the scheme can protect privacy with high efficiency and high recognition rate.