基于哈尔小波分解和等价局部2值模式的人脸防欺骗算法  

Anti-spoofing Algorithm for Face Using Haar Wavelet Decomposition and Uniform Local Binary Pattern

作　　者：李冰[1] 苏天宇 由磊[1] 王宝亮[1] Li Bing;Su Tianyu;You Lei;Wang Baoliang(School of Electronic Information Engineering, Tianjin University, Tianjin 300072, Chin)

机构地区：[1]天津大学电子信息工程学院,天津300072

出　　处：《南开大学学报（自然科学版）》2018年第3期37-43,共7页Acta Scientiarum Naturalium Universitatis Nankaiensis

基　　金：国家自然科学基金(61202380);天津市自然科学基金(12JCQNJC00300)

摘　　要：针对人脸识别技术在身份认证中极易受到假冒用户欺骗的问题,提出1种基于哈尔(Haar)小波分解和等价局部2值模式(Uniform Local Binary Pattern,ULBP)的算法,致力于检测假冒的照片人脸.算法对人脸图像进行4级Haar小波分解,计算1-4级的高频子图系数矩阵的均值、方差作为特征,并提取ULBP特征谱的统计直方图,形成最终的75维特征向量,训练支持向量机(SVM)以判定人脸是否来自活体.在公开的NUAA、REPLAY-ATTACK数据库上实验,最高准确率分别为99.96%和96.26%,最高受试者工作特征曲线下方面积(Area Under Curve,AUC)为1.实验结果表明算法准确度高,计算复杂度小,对不同介质的假人脸图像都能有效检测,能够用来保障人脸识别系统的可靠性.Facing the problem face recognition technology was vulnerable to attacks by illegal users in identity authentication, an approach based on Haar wavelet decomposition and uniform local binary pattern was proposed to devote to defend from the counterfeit facial photographs. The algorithm decomposed face images by four-level Haar wavelet decomposition, calculated the means, variances of high frequency subbands coefficients matrixes which were obtained from one-level to four-level Haar wavelet decomposition,extracted ULBP histogram, and they acted as final 75 dimensional feature vectors to train SVM classifier to determine whether the given face is genuine or not. It was conducted on NUAA and REPLAY-ATTACK database, the highest accuracy were 99.96% and 96.26% respectively and the highest area under curve was 1. Experimental results show that the algorithm achieves high accuracy and low computational complexity, detects face images from different media effectively and it can be used to ensure the reliability in face recognition systems.

关 键 词：人脸识别 防欺骗 哈尔小波分解 等价局部2值模式 

分 类 号：TP391.4[自动化与计算机技术—计算机应用技术]