计算机工程与应用 ›› 2011, Vol. 47 ›› Issue (35): 180-183.

• 图形、图像、模式识别 • 上一篇    下一篇

一种自适应邻域选择半监督判别分析算法

刘云东1,李 鸿1,白万荣2,刘 罡3   

  1. 1.宿州学院 信息工程学院,安徽 宿州 234000
    2.兰州理工大学 计算机与通信学院,兰州 730050
    3.中国人民解放军 68611部队 装备处
  • 收稿日期:1900-01-01 修回日期:1900-01-01 出版日期:2011-12-11 发布日期:2011-12-11

Semi-supervised discriminant analysis algorithm based on adaptive neighborhood selection

LIU Yundong1,LI Hong1,BAI Wanrong2,LIU Gang3   

  1. 1.School of Information Engineering,Suzhou University,Suzhou,Anhui 234000,China
    2.College of Computer and Communication,Lanzhou University of Technology,Lanzhou 730050,China
    3.Department of Equipment,No.68611 Troops,The Chinese People’s Liberation Army
  • Received:1900-01-01 Revised:1900-01-01 Online:2011-12-11 Published:2011-12-11

摘要: 为克服边界Fisher判别分析(MFA)只利用少量有标记样本和构建邻域不能充分反映流形学习对邻域要求的缺点,提出一种基于局部线性结构的自适应邻域选择半监督判别分析的算法。采用自适应算法扩大或者缩小近邻系数[k]来构建邻域以保持局部线性结构。MFA通过少量有类别标签样本进行降维的同时UDP对大量无标签样本进行学习,以半监督的方法对高维人脸数据进行维数约减。最后,在ORL和YALE人脸数据库通过实验结果验证了该算法的有效性。

关键词: 边界Fisher判别分析, 无监督鉴别投影, 半监督, 局部线性结构, 邻域选择

Abstract: A new semi-supervised discriminant analysis algorithm adaptive neighborhood selection algorithm based on local linearity is proposed for the disadvantage of Marginal Fisher Analysis(MFA),which can only make use of a few labeled samples and construct a reasonable neighborhood for each point.An adaptive algorithm to expand or narrow neighbor coefficient [k] is adopted to keep the local linear structure.The MFA can make use of small amount of labeled samples and the UDP can study a large numbers of unlabeled samples,so the method can use semi-supervised dimensionality reduction algorithm for high dimensional data of face.Finally,the effectiveness of the proposed methods is validated through the experimental results on ORL and YALE face databases.

Key words: Marginal Fisher Analysis(MFA), Unsupervised Discriminant Projection(UDP), semi-supervised, local linear structure, neighborhood selection