基于人脸边缘图像的人脸防伪

Face Anti-Spoofing Based on Face Edge Maps

由于自动化人脸识别系统在越来越多的应用中得到广泛的部署,使用照片、视频或3D面具进行自动化脸部识别系统欺骗攻击变得越来越受到关注。为了保护自动化人脸识别系统免受欺骗攻击,提出许多使用图像特征以及分类器的方法。与现有的从原始人脸图像中提取特征的方法不同,将原始人脸图像转换为边缘图,并学习人脸边缘图上的特征用来防范人脸欺骗攻击。与原始人脸图像相比,人脸边缘图放大真实人脸和虚假人脸之间的差异,同时抑制人脸图像中的噪声。所提出的方法已经在两个公共数据库上进行评估,并与最先进的方法进行比较。结果表明,该方法可以更有效地区分真人脸和假脸,特别是在交叉数据库场景中。

Because of automatic face recognition system has been widely deployed in more and more applications, the method of using photo, video or 3D mask to break automatic face recognition system has gained more and more attention. In order to protect automatic face recognition sys- tem from deception attack, many methods of using image features and classifier have been proposed. Different from the existing methods of extracting features from original face images, transforms the original face images into edge images, and the features on the faces are learned to prevent face spoofing attacks. Compared with the original face image, the face edge map magnifies the difference between the real face and the false face, and inhibits the noise in the face image. The proposed approach has been evaluated on two public databases and com- pared with the most advanced methods. The results show that the method can be more effectively divided into true face and false face, espe- cially in the cross database scene.