一种用于人脸识别的图优化线性判别分析算法

摘要

A dimensionality reduction algorithm with linear discriminant analysis based on graph-optimisation for face recognition is proposed in the paper.The algorithm first defines the associated adjacency matrix of same class nature and the mutex adjacency matrix of different class nature ; Then takes these two adjacency matrixes as the function factors to set up respectively the weight matrix between two different samples; Finally,the data dimensionality reduction is achieved by the related projection of these two metric weight matrixes.Experiments on Yale,YaleB and UMIST face datasets verify the effectiveness of the algorithm.%提出一个面向人脸识别的基于图优化的线性判别分析降维算法.该算法首先定义同类性的关联邻接矩阵和异类性的互斥邻接矩阵;然后以两个邻接矩阵作为作用因子分别建立两种不同样本之间的权值矩阵;最后通过这两个度量权值矩阵的相关投影完成数据的降维.在Yale、YaleB和UMIST人脸库的实验验证了该算法的有效性.