Performance Evaluation of Face Anti-Spoofing Method Using Deep Metric Learning from a Few Frames of Face Video
Performance Evaluation of Face Anti-Spoofing Method Using Deep Metric Learning from a Few Frames of Face Video
复制标题
DOI：
--
复制
发表时间：
2020-12
期刊：
影响因子：
--
通讯作者：
Koichi Ito;Asateru Kimura;T. Aoki
中科院分区：
文献类型：
--
作者：
Koichi Ito;Asateru Kimura;T. Aoki
面部识别和深度学习技术的最新进展使我们能够从远处的相机拍摄的图像中识别个人。另一方面，存在一个问题，即恶意的人可以通过提供注册用户的面部照片或视频来冒充注册用户。基于视频输入的欺骗检测，可以提取比图像更多的特征，目前研究还不多。本文提出了一种从少量帧数的视频图像中检测欺骗的方法。该方法使用3D卷积神经网络（3D CNN）从视频图像中提取特征。我们还使用深度度量学习来提高检测的准确性。我们通过使用大规模欺骗攻击数据集的性能评估实验证明了所提出方法的有效性。
Recent advances in face recognition and deep learn-ing technologies are enabling us to identify individuals from images captured by a camera from a distance. On the other hand, there is a problem that a malicious person can impersonate the registered user by presenting a photo or video of the registered user’s face. Spoofing detection using video input, from which more features can be extracted than images, has not been studied very much. In this paper, we propose a method for detecting spoofing from video images of a small number of frames. The proposed method uses features extracted from video images using 3D Convolutional Neural Network (3D CNN). We also use deep metric learning to improve the accuracy of detection. We demonstrate the effectiveness of the proposed method through performance evaluation experiments using a large-scale spoofing attack dataset.