Computer Engineering and Applications ›› 2013, Vol. 49 ›› Issue (4): 224-226.

Previous Articles     Next Articles

Margin correlation analysis with its application to expression recognition

HUANG Yong   

  1. Department of Information Engineering, Liuzhou Railway Vocational Technical College, Liuzhou, Guangxi 545007, China
  • Online:2013-02-15 Published:2013-02-18

边际关联分析及其在表情识别中的应用

黄  勇   

  1. 柳州铁道职业技术学院 信息工程系,广西 柳州 545007

Abstract: The facial expression recognition method based on Margin canonical Correlation Analysis(MCA) is proposed in this paper. Unlike CCA、MML, which all maximize the total or ensemble correlation over all training samples. Motivated by the correlation analysis and the idea of margin learning, MCA devotes to maximizing the individual correlations between given instances and its associated labels and is established by solving a relaxed quadratic programming with box-constraints. Experimental result on JAFFE and CED-WYU show that its effectiveness.

Key words: Canonical Correlation Analysis(CCA), Maximum Margin Learning(MML), Margin Correlation Analysis(MCA), expression recognition

摘要: 提出了一种基于边际关联分析MCA的人脸表情识别方法。传统的CCA、MML等处理的是所有训练样本的全局关联系数。受关联分析和边际学习启发,MCA专注于样本与对应类标间的个体关联,而非整体或全局关联。基于JAFFE和CED-WYU两个表情数据库的识别结果证实了MCA特征提取方法的有效性。

关键词: 典型关联分析, 边际学习最大化, 边际关联分析, 表情识别