计算机工程 ›› 2019, Vol. 45 ›› Issue (5): 256-260.doi: 10.19678/j.issn.1000-3428.0050721

• 多媒体技术及应用 • 上一篇    下一篇

基于微调策略的多线索融合人脸活体检测

胡斐1a,文畅1a,谢凯1b,贺建飚2   

  1. 1.长江大学 a.计算机科学学院; b.电子信息学院,湖北 荆州 434023; 2.中南大学 信息科学与工程学院,长沙 410083
  • 收稿日期:2018-03-12 出版日期:2019-05-15 发布日期:2019-05-15
  • 作者简介:胡斐(1996—),男,硕士研究生,主研方向为信号处理、图像与视频处理;文畅(通信作者),讲师;谢凯,教授;贺建飚,副教授。
  • 基金项目:

    国家自然科学基金(61272147)。

Multi-cue fusion face liveness detection based on fine-tuning strategy

HU Fei1a,WEN Chang1a,XIE Kai1b,HE Jianbiao2   

  1. 1a.School of Computer Science; 1b.School of Electronic Information,Yangtze University,Jingzhou,Hubei 434023,China; 2.College of Information Science and Engineering,Central South University,Changsha 410083,China
  • Received:2018-03-12 Online:2019-05-15 Published:2019-05-15

摘要:

为解决身份认证过程中可能会出现的打印攻击、视频重播攻击等安全问题,提出一种多线索融合人脸活体检测方法。利用金字塔LK光流追踪视频帧并将其进行剪切波变换,以获取图像质量特征,通过卷积神经网络对数据集进行网络微调,得到真假活体。在Print-attack数据库和CISIA数据库上进行实验,结果表明,与LFDNet方法相比,该方法具有较高的人脸活体检测准确率,可用于抵制欺骗攻击。

关键词: 卷积神经网络, 微调策略, 剪切波变换, 金字塔LK光流, 人脸活体检测

Abstract:

To solve the security problems such as print attacks and video replay attacks that may occur during the authentication process,a multi-cue fusion method for detecting human faces is proposed.The pyramid Lucas-Kanade(LK) optical flow is used to track the video frame,and it is subjected to shear wave transform to obtain image quality features.The Convolutional Neural Network(CNN) is used to fine tune the data set to obtain true and false liveness bodies.Experiments on the Print-attack database and the CISIA database show that compared with the LFDNet method,this method has higher face liveness detection accuracy and can be applied to spoofing attacks resistance.

Key words: Convolutional Neural Network(CNN), fine-tuning strategy, shear wave transform, pyramid Lucas-Kanade(LK) optical flow, face liveness detection

中图分类号: