«上一篇/Previous Article|本期目录/Table of Contents|下一篇/Next Article»

《计算机技术与发展》[ISSN:1006-6977/CN:61-1281/TN]

卷:

28

期数:

2018年01期

页码:

69-73

栏目:

智能、算法、系统工程

出版日期:

2018-01-10

文章信息/Info

Title:

Face Recognition Algorithm Based on HOG Feature and DSPP Dimension Reduction

文章编号:

1673-629X(2018)01-0069-05

作者:

田亚娜1 ; 童 莹2 ; 曹雪虹2

1. 南京邮电大学 通信与信息工程学院,江苏 南京 210003;
2. 南京工程学院 通信工程学院,江苏 南京 211167

Author(s):

TIAN Ya-na 1 ; TONG Ying 2 ; CAO Xue-hong 2

1.School of Communication and Information Engineering,Nanjing University of Posts and
Telecommunications,Nanjing 210003,China;
2.School of Communication Engineering,Nanjing Institute of Technology,Nanjing 211167,China)

关键词:

稀疏保留投影; 有监督; 降维; 梯度方向直方图; 人脸识别

Keywords:

sparsity preserving projections; supervised; dimension reduction; HOG; face recognition

分类号:

TP273

DOI:

10.3969/ j. issn.1673-629X.2018.01.015

文献标志码:

A

摘要:

稀疏保留投影(SPP)是一种以保持数据的稀疏表示结构为目的的降维方法,该方法仅考虑了数据的稀疏重构关系而没有充分利用样本的类别信息。 为提高分类识别性能,提出一种有监督的判别稀疏保留投影方法(DSPP)。 首先在构建样本的稀疏重构关系时,通过样本系数和类内样本平均系数的差来重新表示类内紧凑度,同时考虑不同类样本的类别信息和相同类样本的类内紧凑度信息,优化得到具有较强鉴别能力的稀疏表示系数;再通过最小化重构误差准则来得到最优投影,从而提取有效的人脸信息;最后用稀疏表示分类方法进行人脸分类识别。 HOG 算子可以很好地表征人脸图像的局部特征同时有很好的鲁棒性,文中在 HOG 算子提取图像特征的基础上,用 DSPP 方法对图像特征降维后再进行人脸识别分类。 实验结果表明,结合 HOG 特征和 DSPP 算法的人脸识别在 Extended Yale B 和 LFW 库上的平均识别率分别达到 98. 33％ 和 77. 93％,相比其他方法有较好的识别结果。

Abstract:

Sparsity preserving projections (SPP) is a kind of good dimensionality reduction algorithm for maintaining the sparse representation structure of data. It only considers the sparse reconstruction relations of the data without using the class information of the samples. To mprove the performance of classification and recognition,we proposed a supervised discriminant sparse preserving projection (DSPP). Firstly,when constructing the sparse reconstruction relations of the samples,the class information of different samples and the intraclass dispersion
information of the same class of samples are taken into account at the same time,which the intraclass dispersion is represented using the ifference between the average coefficient of the sample coefficients and the intraclass sample coefficient. And then the sparse representation tructure with strong discriminating ability are obtained. Then,the optimal projection is obtained by minimizing the reconstruction error to extract the effective face information. Finally,the sparse representation classification is used to face recognition. HOG operator can well represent the local features of face images and has good robustness. Based on the feature extraction of face image by HOG operator,we use DSPP  to  reduce the dimension of image features and then classify them. Experiment shows that the average recognition rate of face recognition ased on HOG feature and DSPP algorithm achieves 98. 33％ and 77. 93％ on Extended Yale B and LFW face databases respectively,and he recognition rates are obviously higher than that of some other related methods.

常用功能

更新日期/Last Update: 2018-03-12