摘要
利用合法用户的脸部视频进行回放假冒攻击,是目前人脸认证系统的重要安全威胁。针对此问题,本文提出了一种仅用单个普通摄像头来抵抗人脸视频假冒攻击的方法。不同于以往从人脸区域中获取假冒线索进行活体检测的方法,本文通过人脸输入图像与场景参考图像之间的背景对比,从人脸周围背景区域中寻求视频假冒攻击线索。首先,本文在尺度空间里构建人脸周围区域图像的背景特征点集合;然后,利用背景特征点集合建立识别场地背景和人脸背景的Gabor背景描述子,并用融合相位补差的相似度来进行背景比对。实验表明该方法能有效地识别视频回放假冒攻击。
Video-replay spoofing is a serious threat for face authentication/verification system. A real-time approach to aetect video-replay attack using a generic webcamera is presented. Our approach extracts video-spoofing clues in the background around a face region, which is different from the previous in-vivo detection methods that use clues inside a face region. Firstly, a set of background feature points are extracted in scale space. Secondly, the Gabor descriptors of feature points for the background are calculated. To compare two background images, we employ a phase compensated similarity of the descriptors to qualitatively measure attacks. Experimental results show that it can detect video-replay attacks efficiently.
出处
《电路与系统学报》
CSCD
北大核心
2010年第2期39-46,共8页
Journal of Circuits and Systems
基金
国家自然科学基金(60525202)
国家863计划(2008AA01Z149)
长江学者和创新团队发展计划资助(IRT0652)
