松弛耦合非负矩阵分解的低分辨率人脸识别算法  

Low Resolution Face Recognition Algorithm Based on Relaxed Coupled Nonnegative Matrix Factorization

作　　者：王超 赵阳 裴继红[1] Wang Chao;Zhao Yang;Pei Jihong(Key Laboratory of ATR National Defense Science and Technology,Shenzhen University,Shenzhen,Guangdong 518060,China)

机构地区：[1]深圳大学ATR国防科技重点实验室,广东深圳518060

出　　处：《信号处理》2020年第7期1127-1135,共9页Journal of Signal Processing

基　　金：国家自然科学基金项目(61871269,61331021);广东省基础与应用基础研究基金(2019A1515011861);深圳市科技计划项目(JCYJ20190808151615540)。

摘　　要：针对实际监控场景中经常遇到的人脸图像分辨率较低的问题,本文提出了一种利用松弛耦合非负矩阵分解的低分辨率人脸识别算法(RCNMF)。首先,对高低分辨率人脸图像进行非负矩阵分解(NMF),同时使高低分辨率人脸图像的组合系数保持松弛耦合,从而得到含有原图像特征信息的基矩阵。然后,通过低分辨率图像的基矩阵提取训练和测试样本的特征。最后进行识别。实验结果验证了与其他几种基于耦合映射的低分辨率人脸识别方法相比,RCNMF算法的识别性能更好。同时通过实验验证了RCNMF算法的收敛性。In order to solve the problem of low resolution of face image in the actual monitoring scene,this paper proposes a algorithm for low resolution face image recognition which utlize the relaxation coupled nonnegative matrix factorization(RCNMF).Firstly,the nonnegative matrix factorization(NMF)is performed for both high-resolution and low-resolution face images.Meanwhile,the combination coefficients of high-resolution and low-resolution face images are kept loose coupling,so as to obtain the basis matrices with feature information of original face images.Secondly,the features of training samples and test samples are extracted by the basis matrix of low resolution images.Finally,the identification process is carried out.Our experiments verify that the proposed RCNMF algorithm is more effective to solve low resolution face recognition problem than the other state-of-the-art methods based on coupled mapping.At the same time,the convergence of the proposed RCNMF algorithm is verified by experiments.

关 键 词：低分辨率人脸识别 非负矩阵分解 基矩阵 

分 类 号：TP391[自动化与计算机技术—计算机应用技术]