基于低秩矩阵分解和协作表示的人脸图像分类  

Face image classification based on low rank matrix decomposition and collaborative representation

作　　者：阴亚芳[1] 李倩[1] 廖延娜[2] 

机构地区：[1]西安邮电大学电子工程学院,陕西西安710121 [2]西安邮电大学理学院,陕西西安710121

出　　处：《西安邮电大学学报》2015年第1期70-74,共5页Journal of Xi’an University of Posts and Telecommunications

基　　金：陕西省教育厅科学研究计划资助项目(12JK0559)

摘　　要：为了降低干扰因素对人脸识别准确率的影响,提出一种基于低秩矩阵分解和协作表示的人脸图像分类算法。针对噪声阴影影响,使用鲁棒主分量分析(RPCA)对人脸数据进行低秩矩阵分解,去除干扰,得到较为干净的人脸图像。通过协作表示分类方法对经RPCA处理后的低秩分量图像进行分类,通过归一化的最小重构误差来判定测试样本的具体类标。在3个人脸数据集上对算法性能进行仿真测试,结果表明,该算法在10%到50%的标记率下较其他算法均能提升分类识别率。In order to address the problem of low recognition accuracy in the case with strong interference, a new algorithm of face image classification based on low rank matrix decomposition and collaborative representation is proposed in this paper. Firstly, the robust principle component analysis （RPCA） method for low rank matrix decomposition of face data set is used to obtain the clean face images. Secondly, images are classified after RPCA by collaborative representation and the labels of test sample are determined with the normalized minimum reconstruction error. Performance comparisons are done between the new algorithm and classical classification algorithms. Experimental results show that this method can improve the classification accuracy efficiently when the label rate is ranged from 10 % to 50%.

关 键 词：低秩矩阵分解 协作表示 分类 监督学习 人脸识别 

分 类 号：TP391.4[自动化与计算机技术—计算机应用技术]