基于局部特征与核低秩表示的人脸识别算法

针对人脸识别中的遮挡、伪装、光照及表情变化等问题,提出一种基于局部特征与核低秩表示的人脸识别算法。首先,对训练和测试的样本图片进行LBP特征的提取;然后将其通过映射函数投影到高维特征空间中进行后续操作,投影到高维空间中的特征矩阵通过降维处理后采用低秩表示的方法来提取样本之间的共同特征;最后根据低秩表示的结果进行分类识别。实验证明算法在对遮挡、伪装以及光照变化等噪声的影响鲁棒性更强,同时较当前的一些人脸识别算法的识别率也有了显著的提高。...

Full description

Saved in:
Bibliographic Details
Published in电子技术应用 Vol. 42; no. 9; pp. 126 - 128
Main Author 首照宇 杨晓帆 李萌芽
Format Journal Article
LanguageChinese
Published 桂林电子科技大学认知无线电与信息处理教育部重点实验室,广西桂林,541004 2016
Subjects
LBP
低秩表示
映射函数
核函数
特征提取
特征提取
feature extraction
low-rank representation
映射函数
kernel method
核函数
低秩表示
LBP
mapping function
Online AccessGet full text
ISSN0258-7998
DOI10.16157/j.issn.0258-7998.2016.09.033

Cover

More Information
Summary:针对人脸识别中的遮挡、伪装、光照及表情变化等问题,提出一种基于局部特征与核低秩表示的人脸识别算法。首先,对训练和测试的样本图片进行LBP特征的提取;然后将其通过映射函数投影到高维特征空间中进行后续操作,投影到高维空间中的特征矩阵通过降维处理后采用低秩表示的方法来提取样本之间的共同特征;最后根据低秩表示的结果进行分类识别。实验证明算法在对遮挡、伪装以及光照变化等噪声的影响鲁棒性更强,同时较当前的一些人脸识别算法的识别率也有了显著的提高。
Bibliography:For the problem of face recognition, such as disguise, occlusion, illumination and expression changes, a face recognition based on local feature and kernel low rank representation is proposed. First, it extracts the LBP feature of training and testing sam-ple image. Then it is projected onto the high dimensional feature space by the mapping function. The feature matrix which is pro-jected into the high dimension space is used to extract the common features of the samples by the method of low rank representa-tion after dimension reduction. Finally, it carries on the classified recognition based on the method of residual approximation. The ex-perimental results show that the proposed algorithm is robust to occlusion, camouflage and illumination changes. At the same time,compared with some of current face recognition algorithms, the recognition rate of this proposed algorithm has been significantly im-proved.
Shou Zhaoyu, Yang Xiaofan, Li Mengya （Key Laboratory of Cognitive Radio and Information Processing, Guihn
ISSN:0258-7998
DOI:10.16157/j.issn.0258-7998.2016.09.033