计算机科学与探索 ›› 2021, Vol. 15 ›› Issue (7): 1195-1206.DOI: 10.3778/j.issn.1673-9418.2012010

• 综述·探索 • 上一篇    下一篇

人脸识别系统的活体检测综述

马玉琨,徐姚文,赵欣,徐涛,王泽瑞   

  1. 1. 河南科技学院 人工智能学院,河南 新乡 453003
    2. 北京工业大学 信息学部,北京 100124
    3. 河南科技学院 信息工程学院,河南 新乡 453003
  • 出版日期:2021-07-01 发布日期:2021-07-09

Review of Presentation Attack Detection in Face Recognition System

MA Yukun, XU Yaowen, ZHAO Xin, XU Tao, WANG Zerui   

  1. 1. School of Artificial Intelligence, Henan Institute of Science and Technology, Xinxiang, Henan 453003, China
    2. Faculty of Information Technology, Beijing University of Technology, Beijing 100124, China
    3. School of Information Engineering, Henan Institute of Science and Technology, Xinxiang, Henan 453003, China
  • Online:2021-07-01 Published:2021-07-09

摘要:

人脸识别系统的快速发展对人脸活体检测技术提出了新要求,包括检测实时性、面对复杂环境的泛化性、对多种攻击类型的鲁棒性以及用户体验的友好性等。主要阐述了人脸活体检测的必要性,对方法进行了分类、整理和总结,根据所提特征的不同,将活体检测分为基于手工特征的方法和基于深度学习的方法,并将近期针对算法泛化性的研究进展归纳为基于辅助监督信号方法、基于域适应域泛化的方法、基于特征解耦的方法、基于噪声建模的方法、基于异常检测的方法,对每类方法的代表性算法进行了分析介绍,详细总结了每类方法的基本思想和优缺点。从各方面系统地概括了人脸活体检测问题,包括不同类型的呈现攻击、先进的人脸活体检测方法、常用公共数据库、标准化评价指标、测试方法等的介绍。此外,还讨论了该领域的难点与挑战,总结了未来的研究方向和发展趋势。

关键词: 人脸识别, 活体检测, 深度学习, 纹理特征

Abstract:

As the rapid development of face recognition technologies, face presentation attack detection (face liveness detection) techniques are facing new requirements, including instantaneity of?detection, generalization in complicated environments, robustness against various attack types, friendliness of user experience and etc. In?this?paper, the necessity of face presentation attack detection is explained, and the methods are classified, compared and summarized. In general, they can be divided as methods based on manual features and methods based on deep learning. Furthermore, the recent researches on algorithm generalization are summarized as the following categories: methods based on auxiliary supervision, methods based on domain adaptive or domain generalization, methods based on disentangled representation, methods based on noise modeling, and methods based on anomaly detection. Representative algorithm of each type of method are analyzed, and the basic ideas of each method are summarized in detail, as well as the merit and demerit. Issues surrounding face presentation attack detections are systematically summarized from different perspectives, including different types of presentation attacks, advanced detection methods, popular public databases, the standardized evaluation indicators, and testing method. In addition, the difficulties and challenges in this field are also discussed, and the future research direction and development trend are summarized.

Key words: face recognition, presentation attack detection, deep learning, texture?features