白化散度差矩阵的独立元分析应用于表情识别  

Independent component analysis based on whiten scatter difference matrix for expression recognition

作　　者：李春芝[1] 陈晓华[1] 

机构地区：[1]湖州师范学院信息与工程学院,浙江湖州313000

出　　处：《计算机应用研究》2011年第11期4361-4363,4367,共4页Application Research of Computers

基　　金：国家自然科学基金资助项目(60872057);浙江省自然科学基金资助项目(R1090244;Y1080212;Y1100095);浙江省科技计划资助项目(2010C33G2200008);湖州市自然科学资金资助项目(2010YZ04);湖州科技攻关计划资助项目(2010GG22)

摘　　要：提出基于白化散度差矩阵的独立元分析算法,增加不同类表情之间的类间距离,减弱人脸个体差异性信息对表情识别的干扰,避免传统的二维主元分析方法(2DPCA)以总体散布矩阵作为产生矩阵,有效地简化了白化实现过程,提高了白化性能,削弱了光照、姿态等噪声对表情识别的影响。该算法首先采用散度差矩阵求白化矩阵,由快速固定点算法(FASTICA)求解样本独立元,最终由最近邻准则实现表情识别。实验结果表明,提出的算法要优于传统的2DPCA及ICA算法,为表情识别提供了一条有效途径。As the traditional ICA does not consider the impornance of the independent components for classification and recognition.This paper proposed a method,which obtained scatter difference matrix by calculation of expression face matrix and neutral face matrix,abolished the total scatter matrix as a generation matrix which had been employed by 2DPCA.As a result,increased difference of between-class scatter,and weakened the noisy from the variety of face.Firstly this method whitened scatter difference matrix.Secondly,calculated independent components by FASTICA.Finally,used a nearest neighbor rule for expression recognition.Experiment result shows that correct recognition rate by the method is higher than that by 2DPCA and traditional ICA,and is valid for expression recognition.

关 键 词：散度差矩阵 白化散度差矩阵 独立元分析 最近邻准则 表情识别 

分 类 号：TP391[自动化与计算机技术—计算机应用技术]