作　　者： ; ; ; ;

机构地区： 河南大学计算机与信息工程学院

出　　处： 《仪器仪表学报》 2009年第9期1880-1885,共6页

摘　　要： 在分析2DLDA方法的基础上,给出类间离散度和类内离散度矩阵另一种形式的定义。基于这种类间离散度和类内离散度矩阵定义的二维线性判别分析方法即为扩展2DLDA方法。通过对2DLDA方法和扩展2DLDA方法提取的人脸图像特征分析可知,2DLDA提取的主要是人脸图像水平方向上的判别信息,扩展2DLDA提取的主要是人脸图像垂直方向上的判别信息。因此,称2DLDA为水平方向2DLDA,扩展2DLDA为垂直方向2DLDA。水平和垂直方向2DLDA将同一原始人脸图像映射到两个不同的特征空间,并得到互补的两类人脸图像特征。最后,设计一种特征融合方法,对这两类人脸图像特征进行融合,并将其用于人脸识别。在ORL和Yale人脸数据库上的实验结果证明,本文提出的人脸识别方法具有较高的平均识别率,鲁棒性更好。 Based on the analysis of 2DLDA, this paper presents an alternative definition for the image between-class scatter matrix and image within-class scatter matrix. The two-dimensional linear discriminant analysis based on these matrixes is called extended 2DLDA. Based on 2DLDA and extended 2DLDA, two classes of fea- tures of a face image can be obtained. By analyzing the two classes of features, the following conclusions can be drawn： the feature extracted by 2DLDA is mainly the horizontal discriminant characteristic, whereas the feature extracted by extended 2DLDA is mainly the vertical discriminant characteristic. So, 2DLDA can be called horizontal 2DLDA and extended 2DLDA can be called vertical 2DLDA. These two 2DLDAs transform an original face image into different spaces respectively, and two classes of features can be obtained, which complement each other. By fusing the two classes of features, this paper proposes a new face recognition method. Experimental resuits on the ORL and Yale face image databases show that the proposed face recognition method has higher classification accuracy and is more robust.

关 键 词： 双向 特征融合 人脸识别

领　　域： [自动化与计算机技术] [自动化与计算机技术]