基于RGB-D生物特征识别的安全认证  

The Security Authentication of Biometrics Recognition Based on RGB-D

作　　者：肖源源 谢晓尧[2] XIAO Yuanyuan;XIE Xiaoyao(School of Computer,Guizhou University,Guiyang 550025,China;Key Laboratory of Information and Computing Science of Guizhou Province,Guizhou Normal University,Guiyang 550001,China)

机构地区：[1]贵州大学计算机科学与技术学院,贵州贵阳550025 [2]贵州师范大学贵州省信息与计算科学重点实验室,贵州贵阳550001

出　　处：《郑州大学学报（理学版）》2018年第4期26-30,共5页Journal of Zhengzhou University:Natural Science Edition

基　　金：贵州省自然科学基金项目(黔科合LH字[2014]7641);贵州大学青年基金项目(贵大自青基合字(2012)023);贵州省基础研究重大项目(黔科合J字[2014]2001);贵州省科技合作计划重点项目(黔科合LH字[2015]7763);住房和城乡建设部科学技术计划项目(2016-K3-009);全国统计科学研究项目(2016LY81);贵州省信息安全等级保护测评项目

摘　　要：针对传统口令的安全性问题,以生物特征技术作为更安全的替代方案,将基于RGB-D的生物特征技术用于身份的安全认证.采用熵和显著性对RGB图和深度图做图像预处理,用卷积神经网络代替了方向梯度直方图进行特征提取,并与几何特征属性相融合.该研究改进了RISE算法,并采用EURECOM数据集做实验,为复杂条件下的人脸识别的生物特征认证提出了一种有效的方法.To overcome the security problem of traditional password,biometric technology was used as a safer alternative.In view of application of biological features in practice,the technology of biological features based on RGB-D was used for identity authentication. Entropy and saliency were used to preprocess RGB images and depth maps. Instead of histogram of oriented gradient（HOG）, convolutional neural networks（CNN） were used to extract features,which were fused with geometric feature attributes. Using EURECOM datasets to do experiments, the study improved the RISE algorithm. It was an effective method proposed for biometrics face recognition under complex conditions.

关 键 词：熵生物特征 显著性 卷积神经网络 

分 类 号：TP391.9[自动化与计算机技术—计算机应用技术]