光学学报, 2021, 41 (6): 0610002, 网络出版: 2021-04-07
具有活体检测功能的手背静脉身份识别方法研究 下载: 1037次
Recognition Method of Dorsal Hand Vein with Liveness Detection Function
图像处理 模式识别 近红外成像 手背静脉 活体检测 主成分分析 马氏距离 image processing pattern recognition near infrared imaging dorsal hand vein liveness detection principal component analysis Mahalanobis distance
摘要
针对身份识别容易被仿冒和造假的问题,提出了一种利用近红外相机捕获手背静脉同时具有活体检测功能的身份识别方法,手背静脉图像提供静脉特征作为身份识别的依据,与此同时获取的脉搏波的周期性特征作为活体检测的标志。利用自行搭建的手背静脉和脉搏波捕获实验装置,研究了70个个体的手背静脉图像以及活体和假体的静脉图像特征,并提出了提高身份识别准确率的算法。采用主成分分析对活体静脉特征向量进行降维,降低分类算法的复杂度,结合马氏距离去除异常样本,以提高识别精度,再采用参数优化的随机森林算法和支持向量机算法实现了手背静脉的精准识别。结果表明:基于手背静脉特征结合随机森林算法和支持向量机算法可以对不同个体进行身份识别,识别准确率分别为99.28%和99.86%,识别时间分别为0.368 s和0.110 s。
Abstract
In order to solve the problem that identification is easy to be counterfeited and faked, we proposed an identification method that used a near infrared camera to capture the dorsal hand veins and had a liveness detection function. The vein features in the images of dorsal hand veins provided a basis for the identification and the periodic features of the pulse waves acquired at the same time were taken as the sign of liveness detection. Specifically, a self-developed experimental setup of capturing dorsal hand veins and pulse waves was adopted to study the characteristics of the dorsal hand vein images from 70 individuals and the vein images from the living and false bodies, and an algorithm of improving the identification accuracy was proposed. Furthermore, principal component analysis was applied to reduce the dimension of the vein feature vector in the living body and simplify the classification algorithm, and Mahalanobis distance was combined to remove abnormal samples so as to improve the recognition accuracy. Then, the parameter-optimized random forests (RF) algorithm and support vector machine (SVM) algorithm were employed to achieve accurate identification of dorsal hand veins. The results show that the identification of different individuals can be performed by combining the features of dorsal hand veins with the RF and SVM algorithms. The recognition accuracy is 99.28% and 99.86%, and the recognition time is 0.368 s and 0.110 s, respectively.
陈秀莲, 黄梅珍, 富雨超. 具有活体检测功能的手背静脉身份识别方法研究[J]. 光学学报, 2021, 41(6): 0610002. Xiulian Chen, Meizhen Huang, Yuchao Fu. Recognition Method of Dorsal Hand Vein with Liveness Detection Function[J]. Acta Optica Sinica, 2021, 41(6): 0610002.