基于3D卷积神经网络的活体人脸检测

发布时间：2018-06-13 01:43

  本文选题：D卷积神经网络 + 活体人脸检测　； 参考：《信号处理》2017年11期

【摘要】：非法入侵者通过伪装人脸骗取系统认证,给人脸认证系统带来了严重的威胁。因此,活体人脸检测成了人脸认证系统走向实用必须解决的一个重要课题。现有活体人脸检测方法多为基于照片的人脸攻击方面的研究成果,对于基于视频的人脸攻击,效果并不理想。3D卷积神经网络(Convolutional Neural Network,CNN)具有深度学习的特点,能自动学到图像的分布式特征表示;与2D卷积相比,它能学到连续视频帧的动作信息。本文结合3D卷积神经网络的特性,提出利用3D卷积实现视频人脸伪装检测。通过提取3D卷积神经网络最后全连接层学到的时间空间特征,训练SVM(Support Vector Machine)分类器,实现真实人脸和伪装人脸的分类。实验采用两个人脸伪装公开数据库Replay Attack和CASIA,实现多尺度内部数据库测试和交叉数据库测试。实验结果相对于纹理特征及2D卷积方法有较大提高,可应用于视频人脸攻击的活体人脸检测。
[Abstract]:Illegal intruders deceptive system authentication by camouflage face, bring serious threat to face authentication system. Therefore, face detection in vivo has become an important issue that must be solved in the face authentication system. Most of the existing face detection methods are based on photos. For video-based face attacks, the effect is not ideal. 3D convolutional neural network (CNN) has the characteristics of deep learning. It can automatically learn the distributed feature representation of images, and it can learn the action information of continuous video frames compared with 2D convolution. Based on the characteristics of 3D convolution neural network, a video face camouflage detection based on 3D convolution is proposed in this paper. By extracting the temporal and spatial features of 3D convolutional neural network and training SVM support Vector Machine, the classification of real face and camouflage face is realized. Two human face camouflaged open databases, replay Attack and CASIA, are used to realize multi-scale internal database testing and cross-database testing. Compared with the texture features and 2D convolution, the experimental results can be applied to live face detection of video face attacks.
【作者单位】： 五邑大学信息工程学院;
【基金】：国家自然科学基金项目(61372193,61070167) 广东高校优秀青年教师培训计划资助项目(SYQ2014001) 广东省特色创新项目(2015KTSCX143,2015KTSCX145) 广东省青年创新项目(2016KQNCX171)
【分类号】：TP183;TP391.41
，

本文编号：2012074