摘要
基于深度学习的大规模人脸检索系统通常面临两个主要困难:海量高维人脸特征导致查询速度慢;深度特征维数高,导致传统的树状索引算法失效。针对这些困难,设计并实现一种大规模人脸检索系统。使用ArcFace作为人脸特征提取器,通过构建正轴体LSH(Locality-Sensitive Hashin)和VLH(Variable Length Hashing)索引,加速大规模K近邻特征查询。通过有针对性地设计存储结构与访问机制,降低系统部署成本,提升系统可扩展性。
Large-scale face retrieval systems based on deep learning method usually face two main difficulties:massive high-dimensional face features lead to slow query speed;high-dimensional depth features lead to the invalidation of traditional tree-based indexing algorithm.Aiming at these difficulties,we introduce a large-scale face retrieval system.The system used ArcFace as the face feature extractor,accelerated large-scale K-nearest neighbor feature query by constructing LSH hashing and VLH index.The storage structure and access mechanism was designed pertinently to reduce the cost of system deployment and improve system scalability.
作者
余若晟
徐超
张帆
Yu Ruosheng;Xu Chao;Zhang Fan(College of Computer Science and Technology,Fudan University,Shanghai 200433,China)
出处
《计算机应用与软件》
北大核心
2021年第3期119-123,共5页
Computer Applications and Software
基金
国家自然科学基金项目(61672165)
上海市科技人才计划项目(17XD1425000)。
关键词
人脸检索
高维空间索引
监控视频
系统设计
Face retrieval
High dimension indexing
Video surveillance
System design
