基于相关性和支持度分析的人脸识别方法  

Face recognition method based on correlation and support analysis

作　　者：刘鸿雁 姚明海[2] LIU Hongyan;YAO Minghai(Graduate College of Bohai University,Jinzhou 121013,China;College of Information Science and Technology,Bohai University,Jinzhou 121013,China)

机构地区：[1]渤海大学研究生学院,辽宁锦州121013 [2]渤海大学信息科学与技术学院,辽宁锦州121013

出　　处：《渤海大学学报（自然科学版）》2020年第3期271-276,共6页Journal of Bohai University:Natural Science Edition

基　　金：辽宁省自然科学基金项目(No:2019-ZD-0503);辽宁省教育厅科研项目(No:WJ2020004,No:LJ2020003);辽宁省普通高等教育本科教学改革研究项目.

摘　　要：为了提高人脸识别算法识别准确性和算法的计算效果,提出一种基于相关性和支持度分析的特征选择方法.采用互信息理论来分析特征数据与类别及特征数据间的相关性;通过费舍尔得分法判断特征对类别的支持度;利用皮尔逊相关系数法实现特征数据冗余度的分析.最后,从有监督和无监督角度评价特征数据,选择出最优特征集合.通过在AR、ORL、CMU PIE、UMIST、Yaleb五个人脸数据库上进行对比实验证明本文提出的方法在识别效果上有显著的提升.This paper proposes a feature selection method based on correlation and support analysis to improve the recognition accuracy and computational efficiency of face recognition algorithms.Mutual Informa⁃tion theory is used to analyze the correlation between features,categories and features.The support degree of the feature to the category is judged by Fisher score method.The feature redundancy is analyzed by Pearson correla⁃tion coefficient method.Finally,the optimal feature subset is selected from the perspective of supervised and unsupervised feature evaluation.The experimental results on five face databases of AR,ORL,CMU PIE,UMIST and Yaleb show that the proposed method has a significant improvement in recognition effect.

关 键 词：人脸识别 特征选择 相关性分析 支持度分析 冗余度分析 

分 类 号：TP399[自动化与计算机技术—计算机应用技术]