摘要
提出了一种基于分类性能的二维主分量特征选择方法。即将二维主分量分析中图像总体散布矩阵的特征向量在二维线性鉴别分析的目标函数上进行投影,并选择分类性更好的特征向量进行投影。另外,为了保持原有的二维主分量分析主特征的优点,对最后的投影特征向量进行组合,也就是最后的投影特征向量选取对图像重建和图像分类分别起着重要作用的特征进行组合。在XM2VTS标准人脸库上的试验结果表明,所提出的方法融合了两种具有互补性的图像并行特征,在识别性能上优于传统的二维主分量分析方法。
A new two-dimensional principal component analysis method based on feature selection is proposed. Firstly,the feature vectors of image total scatter matrix that have good classification performance are selected to project to the object function of two-dimensional linear discriminant analysis. In addition,to make the best use of the advantages of two-dimensional principal component analysis, the two kinds of vectors got from two-dimensional principal component analysis are combined into new feature vectors. The method is applied to XM2VTS face database,and the experimental result shows that the proposed method is more available than the traditional two-dimensional principal component analysis.
出处
《计算机应用与软件》
CSCD
北大核心
2008年第2期71-73,共3页
Computer Applications and Software
基金
江苏省教育厅自然科学基金项目(05KJD510248)
关键词
二维主分量分析
特征选择
特征融合
人脸识别
Two-dimensional principal component analysis
Feature selection
Feature fusion
Face recognition
