有监督低秩子空间恢复的正则鲁棒稀疏表示人脸识别算法  

Regularized Robust Sparse Representation Face Recognition Algorithm Based on Supervised Low-Rank Subspace Recovery

作　　者：胡正平[1] 白帆[1] 王蒙[1] 孙哲[1] 

机构地区：[1]燕山大学信息科学与工程学院,河北秦皇岛066004

出　　处：《信号处理》2016年第11期1299-1307,共9页Journal of Signal Processing

基　　金：国家自然科学基金(61071199);河北省自然科学基金(F2016203422)

摘　　要：针对训练样本和测试样本均存在光照及遮挡时,破坏图像低秩结构问题,本文提出基于监督低秩子空间恢复的正则鲁棒稀疏表示人脸识别算法。首先,将所有训练样本构造成矩阵D,对矩阵D进行监督低秩矩阵分解,分解为低秩类相关结构A,低秩类内差异结构B和稀疏误差结构E;然后用主成分分析方法找到类相关结构A低秩子空间的变换矩阵;再通过变换矩阵将训练样本和测试样本投影到低秩子空间;最后,在低秩子空间中,通过正则鲁棒稀疏编码进行加权分类识别。在AR和Extended Yale B公开人脸数据库上的实验结果验证本文算法的有效性及鲁棒性。Due to training samples and query sample always filled with lighting and occlusion, then the image low-rank structure was destroyed. In view of this problem, a regularized robust sparse face representation algorithm which based on supervised low-rank subspace recovery is proposed. Firstly, the matrix D that is structured by all training samples is de- composed into low-rank class-specific structure A, low-rank non-class-specific structure B and sparse error structure E by the supervised low-rank decomposition. PCA is applied on the low-rank class-specific structure A to obtain the transform matrix; and then to project training samples and query samples onto low-rank subspace by using the transform matrix. Fi- nally, utilizing weighted classification based on regularized robust sparse coding to classify the query image. Expe.rimental results on AR and Extended Yale B face databases verify the effectiveness and robustness of our method.

关 键 词：人脸识别 低秩分解 主成分分析 正则鲁棒编码 

分 类 号：TP391.41[自动化与计算机技术—计算机应用技术]