ISSN 1000-1239 CN 11-1777/TP

计算机研究与发展 ›› 2018, Vol. 55 ›› Issue (3): 643-650.doi: 10.7544/issn1000-1239.2018.20160417

• 图形图像 • 上一篇    下一篇

基于LBP和多层DCT的人脸活体检测算法

田野1,项世军1,2   

  1. 1(暨南大学信息科学技术学院 广州 510632); 2(信息安全国家重点实验室(中国科学院信息工程研究所) 北京 100093) (shijun_xiang@qq.com)
  • 出版日期: 2018-03-01
  • 基金资助: 
    国家自然科学基金面上项目(61772234,61272414);信息安全国家重点实验室开放基金项目(2016-MS-07)

LBP and Multilayer DCT Based Anti-Spoofing Countermeasure in Face Liveness Detection

Tian Ye1, Xiang Shijun1,2   

  1. 1(School of Information Science and Technology, Jinan University, Guangzhou 510632); 2(State Key Laboratory of Information Security (Institute of Information Engineering, Chinese Academy of Sciences), Beijing 100093)
  • Online: 2018-03-01

摘要: 随着安全性成为制约人脸识别系统应用的最大瓶颈,提高人脸识别系统的抗欺骗攻击能力已成为亟待解决的问题.针对基于视频的人脸欺骗攻击,基于局部二值模式(local binary patterns, LBP)和多层离散余弦变换(discrete cosine transform, DCT)提出了一种新的人脸活体检测算法.其基本思想是首先从目标视频中每隔一定帧数提取1张人脸图像;其次对提取出的每张人脸图像进行LBP操作得到低级特征描述子(LBP算子);然后在LBP特征上进行多层DCT变换得到高级特征描述子(LBP-MDCT算子);最后将得到的高级特征描述子送入支持向量机(support vector machine, SVM)中判断该视频是非法用户实施的人脸欺骗攻击还是合法用户的进入请求.通过在Replay-Attack和CASIA-FASD数据库上与现有的人脸活体检测算法做比较,验证了该算法能够取得优异的检测效果且十分简单、高效.

关键词: 人脸活体检测, 局部二值模式, 多层离散余弦变换, Replay-Attack数据库, CASIA-FASD数据库

Abstract: As security problem has become the tightest bottleneck in the application of face recognition systems, rendering a face recognition system robust against spoof attacks is of great significance to be dealt with. In this paper, aimed at video-based facial spoof attacks, an innovative face antispoofing algorithm based on local binary patterns (LBP) and multilayer discrete cosine transform (DCT) is proposed. First, we extract face images from a target video at a fixed time interval. Second, the low-level descriptors, i.e., the LBP features are generated for each extracted face image. After that, we perform multilayer DCT on the low-level descriptors to obtain the high-level descriptors (LBP-MDCT features). To be more exact, in each layer, the DCT operation is implemented along the ordinate axis of the obtained low-level descriptors, namely the time axis of the entire target video. In the last stage, the high-level descriptors are fed into a support vector machine (SVM) classifier to determine whether the target video is a spoof attack or a valid access. In contrast to existing approaches, the outstanding experimental results attained by the proposed approach on two widely-used datasets (Replay-Attack dataset and CASIA-FASD dataset) demonstrat its performance superiority as well as its low complexity and high efficiency.

Key words: face antispoofing, local binary patterns (LBP), multilayer discrete cosine transform (DCT), Replay-Attack database, CASIA-FASD database

中图分类号: