虚假数字人脸内容生成与检测技术  

Digitally Forged Face Content Creation and Detection

作　　者：蔺琛皓 沈超[1,2] 邓静怡[1,2] 胡鹏斌 王骞 马仕清 李琦 管晓宏[1,2] LIN Chen-Hao;SHEN Chao;DENG Jing-Yi;HU Peng-Bin;WANG Qian;MA Shi-Qing;LI Qi;GUAN Xiao-Hong(School of Cyber Science and Engineering,Faculty of Electronic and Information Engineering,Xi’an Jiaotong University,Xi’an 710049;Key Laboratory for Intelligent Networks and Network Security(Xi’an Jiaotong University),Xi’an 710049;School of Cyber Science and Engineering,Wuhan University,Wuhan 430072;Department of Computer Science,Rutgers University,New Jersey 08854 USA;Institute for Network Sciences and Cyberspace,Tsinghua University,Beijing 100084)

机构地区：[1]西安交通大学电子与信息学部网络空间安全学院,西安710049 [2]智能网络与网络安全教育部重点实验室(西安交通大学),西安710049 [3]武汉大学网络安全学院,武汉430072 [4]罗格斯大学计算机学院,美国新泽西州08854 [5]清华大学网络科学与网络空间研究院,北京100084

出　　处：《计算机学报》2023年第3期469-498,共30页Chinese Journal of Computers

基　　金：科技创新2030-“新一代人工智能”重大项目(No.2020AAA0107702);国家自然科学基金重大项目(62006181,U20A20177,61822309,61703301,U21B2018);陕西重点研发计划项目(2021ZDLGY01-02)资助.

摘　　要：近年来,以深度学习算法为代表的人工智能技术在安防视频监控、个人隐私保护、自动驾驶等领域广泛应用.尤其在人脸识别等领域,深度学习方法显示出超越人类感知及辨别的能力,为人类的日常生活带来了诸多便利.然而,利用人工智能生成、对抗、伪造等技术产生的虚假数字人脸给个人隐私安全、社会安全乃至国家安全等方面带来了诸多风险和挑战.本文通过回顾虚假数字人脸内容生成与检测的相关研究工作,揭示其对国民、国家安全造成的潜在威胁.具体来说,本文首先介绍虚假数字人脸内容的攻击对象及攻击类型,从两种攻击对象—人工智能系统及人类感知系统,两大攻击类型—人脸对抗样本及人脸深度篡改,归纳、分析相应的生成、攻击及检测、防御技术.最后,本文讨论和展望虚假数字人脸内容生成与检测技术未来的研究方向和发展趋势.In recent years,artificial intelligence technologies,represented by deep learning algorithms,have been widely applied in many fields including smart video surveillance,privacy protection,autonomous driving and so on.Particularly in the field of face recognition,deep learning based methods have shown the ability of surpassing human perception and brought great conveniences to our daily lives.However,digitally forged face generated by adversarial and deepfake techniques poses huge risks and challenges for individual privacy,social security and even national security.This paper reviews previous work on the creation and detection of digitally forged face content,revealing potential risks to individual privacy,social security and national security.Specifically,we firstly introduce the attack targets and attack types of digitally forged face content.Secondly,we summarize and analyze digitally forged face content creation,attack,detection and defense technologies in terms of two attack targets:artificial intelligence system and human perception system,and in terms of two attack types:adversarial face example and deep face manipulation.Finally,the directions of future research on digitally forged face content creation and detection are discussed.

关 键 词：虚假数字内容 人脸对抗样本 人脸深度篡改 人工智能安全 隐私保护 

分 类 号：TP309[自动化与计算机技术—计算机系统结构]