J4 ›› 2014, Vol. 41 ›› Issue (2): 144-150.doi: 10.3969/j.issn.1001-2400.2014.02.024

• 研究论文 • 上一篇    下一篇

一种用于人脸识别的非迭代GLRAM算法

赵扬扬;周水生;武亚静   

  1. (西安电子科技大学 数学与统计学院,陕西 西安  710071)
  • 收稿日期:2013-01-08 出版日期:2014-04-20 发布日期:2014-05-30
  • 通讯作者: 赵扬扬
  • 作者简介:赵扬扬(1988-),男,西安电子科技大学硕士研究生,E-mail: 314409630@qq.com.
  • 基金资助:

    国家自然科学基金资助项目(61179040, 61072144)

Non-iterative GLRAM algorithm for face recognition

ZHAO Yangyang;ZHOU Shuisheng;WU Yajing   

  1. (School of Mathematics and Statistics, Xidian Univ., Xi'an  710071, China)
  • Received:2013-01-08 Online:2014-04-20 Published:2014-05-30
  • Contact: ZHAO Yangyang

摘要:

利用二维主成分分析算法通过协方差矩阵获得右投影变换矩阵,进一步对其投影特征矩阵降维获得左投影变换矩阵,提出了一种矩阵广义低秩逼近的新的非迭代算法.ORL和AR人脸数据库的实验研究表明,新的非迭代算法在图像重建和图像识别都取得了和矩阵广义低秩逼近的迭代算法相近的效果,同时节省了大量的训练时间,而较二维主成分分析,新算法以较大的压缩率取得了更好的图像重建效果和识别率.

关键词: 人脸识别, 数据降维, 矩阵的广义低秩逼近, 二维主成分分析(2DPCA)

Abstract:

In this paper, we get the right projection transform matrix by the covariance matrix of the 2DPCA algorithm, and gain the left projection transform matrix by dimensional reduction of the feature matrix of the 2DPCA. Then we propose a new non-iterative algorithm for generalized low rank approximation of matrices (NGLRAM).Experiments on ORL and AR face database show that the new NGLRAM saves a lot of training time to get the similar performance with GLRAM in image reconstruction and image recognition. Compared with the 2DPCA, the NGLRAM can lead to better results in image reconstruction and image recognition at a larger compression rate.

Key words: face recognition, data dimension reduction, generalized low rank approximations of matrices(GLRAM), two-dimensional principal component analysis(2DPCA)