Abstract: In this paper, a novel local preserving based discriminant analysis algorithm, trace ratio based local discriminant analysis algorithm using adaptive neighborhood graph, is proposed. To implement the algorithm, it adaptively constructs within-class and between neighborhood graph according to samples distribution, preserves local structure of the data manifold and utilizes its discriminant information to define local within-class scatter matrix and local between-class scatter matrix, ultimately gains optimal subspace by iteratively maximizing the trace ratio of local within-class scatter matrix and local between-class scatter matrix. The experiments on ORL and Yale face database demonstrate the effectiveness of the proposed algorithm.

Key words: subspace learning, trace ratio criterion, adaptive neighborhood graph, local discriminant analysis, face recognition

摘要： 提出了一种新的局部保持鉴别分析算法：基于迹比准则与自适应近邻图嵌入的局部保持鉴别分析算法。根据样本分布特性自适应构建类内和类间近邻图，保持数据的局部结构并且利用数据的鉴别信息，定义局部类内离差矩阵以及局部类间离差矩阵，采用迹比Fisher判别函数作为目标函数，通过迭代的方法最大化局部类间离差矩阵与类内离差矩阵的迹比值，解得最优子空间。在ORL和Yale人脸数据库上的实验表明该方法是有效的。

关键词: 子空间学习, 迹比准则, 自适应近邻图, 局部鉴别分析, 人脸识别