基于特征子集与特征区分度的生物认证方法  

Biometric Authentication Method Based on Feature Subset and Feature Discrimination

作　　者：王娜 李劲松 姚明海[1] WANG Na;LI Jin-song;YAO Ming-hai(School of Information Science and Technology,Bohai University,Jinzhou 121013,China)

机构地区：[1]渤海大学信息科学与技术学院,辽宁锦州121013

出　　处：《计算机技术与发展》2020年第12期51-55,共5页Computer Technology and Development

基　　金：辽宁省自然科学基金项目(2019-ZD-0503);辽宁省教育科学技术项目(LQ2017004)。

摘　　要：生物认证是信息安全领域研究的热点问题,已经成为社会安全各个领域用于身份识别的重要技术手段。随着数字图像获取技术和采集设备的快速发展,生物认证图像数据在采集过程中往往会出现高维度、高冗余现象。为了解决生物认证数据在计算过程中出现的维度高、冗余信息多、计算复杂度高的问题,在生物数据处理过程中构建了基于特征子集与特征区分度的特征选择方法。该方法首先利用改进的随机子空间方法和费舍尔得分法分别对特征排序;然后,将两种方法选择的特征结果进行加权融合得到全新的特征排序;最后,利用顺序前向搜索策略进行特征选择。为验证方法的有效性,将该方法与传统方法分别在五个经典的生物认证数据库上进行了比较。实验结果证明该方法获得了非常高的识别准确度。Biometric authentication is a hotspot issue in the information security field,which has become an important technical means for identity recognition in various fields of social security.With the rapid development of digital image acquisition technology and shooting equipment,high dimension and high redundancy often appear in the process of biological image data acquisition.In this paper,the biometric methods based on feature subset and feature discrimination is proposed to solve the problem of the biometric authentication data high dimensionality and redundancy.Firstly,modified random subspace method and Fisher score method are employed to pre-rank the feature.Then,the new feature ranking is obtained by fusing the feature selection results.Finally,sequential forward search method is utilized to select the most significant feature subset.In order to verify its effectiveness,the proposed method is compared with the traditional method on five classic biometric authentication databases.The experiment shows that the proposed method has high recognition accuracy.

关 键 词：特征选择 随机子空间 费舍尔得分 生物认证 特征融合 

分 类 号：TP391[自动化与计算机技术—计算机应用技术]