基于一致判别相关分析的低分辨率人脸识别算法  

Low Resolution Face Recognition Algorithm Based on Consistent Discriminant Cor⁃relation Analysis

作　　者：张恩豪 陈晓红[1] ZHANG Enhao;CHEN Xiaohong(College of Science,Nanjing University of Aeronautics and Astronautics,Nanjing,211106,China)

机构地区：[1]南京航空航天大学理学院,南京211106

出　　处：《数据采集与处理》2020年第6期1163-1173,共11页Journal of Data Acquisition and Processing

基　　金：国家自然科学基金(61703206,61661136001)资助项目;中央高校基本科研业务费专项资金(NG2019004)资助项目。

摘　　要：相比于高分辨率(High resolution,HR)人脸图像,低分辨率(Low resolution,LR)人脸图像的识别效果较差。针对此问题,已有研究者提出基于典型相关分析和核典型相关分析的LR人脸识别算法,但其并未考虑样本的类信息和视图间的一致性。本文同时利用数据的类信息和视图间的一致性信息,提出一致判别相关分析(Consistent discriminant correlation analysis,CDCA),进而得到基于CDCA的LR人脸识别算法。该算法先利用主成分分析从HR和LR人脸图像中提取主成分特征,然后利用CDCA学习HR和LR人脸的特征投影矩阵,进而实现LR人脸识别。实验结果表明,相比现有的LR人脸识别算法,该算法具有较好的识别效果和鲁棒性。Compared with high-resolution(HR)face image,the low-resolution(LR)face image recognition effect is poorer.Researchers have put forward several LR face recognition algorithms based on the canonical correlation analysis(CCA)and kernel canonical correlation analysis(KCCA)to solve this problem,which ignored the supervised information and the consistency information between different views.In this paper,we put forward a novel dimensionality reduction algorithm—consistent discriminant correlation analysis(CDCA)by virtue of the class information and consistency information of different views.Furthermore,we design a LR face recognition algorithm based on CDCA.Concretely,we extract the principal component features from HR and LR face images respectively,use CDCA to learn the characteristic projection matrix of HR and LR face,and realize LR face recognition with the help of projection matrix.The experimental results show the superiority of the proposed method on recognition effect and robustness compared with the existing LR face recognition algorithms.

关 键 词：低分辨率人脸识别 典型相关分析 主成分分析 径向基函数 

分 类 号：TP39[自动化与计算机技术—计算机应用技术] TP301[自动化与计算机技术—计算机科学与技术]