一种利用背景光流特征的虚假人脸检测方法  

Faceanti-spoofing method using the optical flow features of back ground

作　　者：孔月萍[1] 刘楚 朱旭东[1] KONG Yueping;LIU Chu;ZHU Xudong(Schoolof Information and Control Engineering,Xi’an University of Architecture and Technology,Xi’an 710055,China)

机构地区：[1]西安建筑科技大学信息与控制工程学院,陕西西安710055

出　　处：《西安电子科技大学学报》2021年第5期86-91,99,共7页Journal of Xidian University

基　　金：国家重点研发计划项目(2019YFD1100901);陕西省自然科学基金(2019JM-183)。

摘　　要：人脸识别系统会受到照片视频的虚假人脸攻击,但手持照片、视频攻击存在抖动现象,以光流表征运动信息,提出一种利用背景光流特征的假人脸检测方法。该方法聚焦真实人脸背景与虚假人脸背景之间的运动差异性,分析人脸外区域的光流角度分布,背景对比区的运动状况,通过评估对比区的运动一致性虚假人脸。新方法在公开数据集ReplayAttack和CASIA-FASD中的测试准确率分别为97.87%和90.95%,可有效甄别含背景区手持照片视频的虚假人脸攻击。The face recognition system is vulnerable to spoofing attacks by presenting photosor videos of a valid user,but there is a hand shaking existing when the illegal user holds the photosor videos.Focused on this phenomenon to represent motion information by optical flow features,an ovelface anti-spoofing method which exploits the background of optical flow features is proposed.The method took advantage of the background motion difference between the registered user and the invalid user to analyze the optical flow angle distribution of outside the face in multi comparing regions.And the similarity function is built to measure the correlation degree between the discograms of optical flow angle distribution in comparing regions.According to the measurements of the background comparing regions,their motion consistency have been evaluated,then the real face or the attacked face will be detected.Experiments have been done on the publicface anti-spoofing database of the Replay Attack and CASIA-FASD.The accuracy of the new method is 97.87%and 90.95%,respectively,which shows that the new method can effectively detect the attacker hold photosor videos with the background region.

关 键 词：虚假人脸检测 背景运动 光流计算 直方图 

分 类 号：TP391[自动化与计算机技术—计算机应用技术]