您的位置:首页  > 论文页面

基于改进的稀疏表示的人脸识别

发表时间:2017-02-28  浏览量:1236  下载量:249
全部作者: 吴佩,邹承明,赵广辉
作者单位: 武汉理工大学计算机科学与技术学院
摘 要: 稀疏表示(sparse representation,SR)作为一种热门分类方法已经成功地应用到人脸识别中,并且能实现较高的识别率,但是SR 中用l^1 范数求解最优SR 的计算过程复杂、耗时长。为解决这一问题,提出一种基于Split Bregman 与近邻样本的SR 算法进行人脸识别。首先,采用主成分分析(principal component analysis,PCA)对图像进行降维,提取局部特征;其次,通过奇异值分解(singular value decomposition,SVD)求解最小二乘解的方法解决l^2 范数问题,求解出训练样本图像的近邻样本;然后,基于该近邻样本用Split Bregman迭代算法求解l^1 范数最小化问题,得出对测试样本的SR;最后,通过分类得出人脸识别结果。本文算法分别在ORL、CMU PIE 和AR 人脸库中进行实验,结果表明,算法对表情、光照和有遮挡的干扰具有较强的鲁棒性,同时相较于传统的稀疏表示的分类算法(sparse representation based classification,SRC)与K 近邻稀疏表示的分类算法(K nearest neighbor sparse representation based classification,KNN-SRC),本文算法的识别速度更快。
关 键 词: 人工智能;主成分分析;稀疏表示;奇异值分解;近邻样本;Split Bregman
Title: Face recognition using improved sparse representation
Author: WU Pei, ZOU Chengming, ZHAO Guanghui
Organization: School of Computer Science and Technology, Wuhan University of Technology
Abstract: As an effective classification method, sparse representation (SR) has been successfully used in the field of face recognition. However, the solution of l^1 norm in SR is complicated and costs a lot of time. To solve this problem, this paper proposes an improved SR method based on neighbor samples and Split Bregman algorithm for face recognition. First, principal component analysis (PCA) is used to reduce the dimensionality of all images. Then, a method of solving least squares through singular value decomposition (SVD) is used to solve l^2 norm. According to the least squares, the neighbors of the original samples can be solved. Next, based on the neighbor samples, the Split Bregman iterative algorithm is used to solve l^1 norm. Thus, the final SR can be obtained. Finally, the result of face recognition can be obtained through classification. A series of experiments have been done through the proposed algorithm in different face recognition datasets, ORL, CMU PIE and AR respectively. The results show that the proposed algorithm is robust. Besides, the effectiveness of the proposed method is faster than that of the traditional sparse representation based classification (SRC) and K nearest neighbor sparse representation based classification (KNN-SRC) algorithm.
Key words: artificial intelligence; principal component analysis; sparse representation; singular value decomposition; neighbor samples; Split Bregman
发表期数: 2017年2月第4期
引用格式: 吴佩,邹承明,赵广辉. 基于改进的稀疏表示的人脸识别[J]. 中国科技论文在线精品论文,2017,10(4):412-420.
 
1 评论数 0
暂无评论
友情链接