人脸欺诈检测最新进展及典型方法

A Survey on the Latest Development and Typical Methods of Face Anti-Spoofing

以人脸为生物特征的身份认证由于其简单、快捷和无接触,深受用户喜爱。然而,不法分子针对人脸的易于伪造和认证场景的多样性,恶意攻击人脸识别系统,给国家安全和人民生活带来极大危害,有必要深入研究人脸的反欺诈技术。该文收集了经典和代表性论文68篇,其中包括近2年来国际主流期刊和会议的相关论文50余篇,从7大类来梳理近期人脸欺诈检测的技术脉络,把握领域的最新动向。为了使读者洞悉各类的技术精髓,对典型方法进行了重点介绍,并对各类进行了技术小结。在方法介绍结束之后,对2019年以来新推出的6个人脸欺诈检测数据库进行了简要介绍,便于读者对本领域的最新算法开发和验证平台有所了解。结语部分对人脸欺诈检测7类方法之间的关联、在理论和应用上的总体现状、存在的主要问题以及发展的方向作了简单的总结和展望。

Identity authentication based on face recognition is widely applied because of its simplicity, convenience and non-contact nature. However, face recognition systems in real-world scenarios are subject to malicious attacks, which severely threatens national security and people’s daily life. It is therefore urgent to explore counter measures of face spoofing. This survey collects 68 classical and typical research articles, among which over 50 ones were published in the major international academic journals and conferences in the past two years. We organize those articles according to their technical motivation into 7 classes. More information is given to some typical works so that readers may gain some insight into the latest trends in this area. We further summarize each class. After reviewing the methods, we make a brief introduction to 6 face spoofing detection databases released since 2019, which contributes to a better understanding of the development and validation platforms of face spoofing detection. In the last section, we conclude the relations among the 7 classes, describe their general status in theoretical research and industrial applications, exhibit the major challenges and point out the research directions.