摘要
用现有的人脸识别方法处理人脸姿态和光照的变化仍有一定的难度,本文提出一种基于图像重构和l_0范数稀疏表示的人脸识别算法:首先,采用深度学习网络提取人脸特征;然后,根据提取的特征重构人脸图像;最后,用l_0范数快速稀疏分类的识别算法在重构图像上进行识别.基于FERET人脸数据库的实验结果表明,本算法可在姿态变化比较大的情况下保持较高的人脸识别率以及较快的识别速度.
It is still difficult to use the existing face recognition methods to deal with the changes in face attitude and illumination. This paper proposes a face recognition algorithm based on image reconstruction and l0 norm sparse representation classification. First, face features are extracted by the depth learning network; then, the face image is reconstructed based on the feature extracted; and finally, the recognition algorithm based on the fast sparse classification algorithm is used to identify the reconstructed images. Experimental results based on FERET face database demonstrate that the proposed algorithm can greatly improve recognition rate and speed under various conditions such as large attitude change.
作者
曾军英
赵晓晓
林作永
谌瑶
冯武林
ZENG Jun-ying;ZHAO Xiao-xiao;LIN Zuo-yong;SHEN Yao;FENG Wu-lin(School of Information Engineering, Wuyi University, Jiangmen 529020, Chin)
出处
《五邑大学学报(自然科学版)》
CAS
2018年第2期18-22,共5页
Journal of Wuyi University(Natural Science Edition)
基金
广东高等学校优秀青年培养计划项目(SYQ2014001)
广东省特色创新类项目(2015KTSCX143)
广东省青年创新人才类项目(2015KQNCX165
2015KQNCX172)
关键词
人脸识别
深度学习
特征提取
图像重构
l0范数
稀疏表示
face recognition
deep learning
feature extraction
image reconstruction
l0 norm
sparserepresentation classification