局部保持分类投影的人脸识别算法

Method of Face Recognition Based on Local Preserving Discriminating Projection

局部保持投影(LPP)是一种有效的特征提取算法,在模式识别领域得到了广泛的应用.然而,LPP本质上是属于无监督的特征提取算法,没有利用训练样本的类别信息,同时LPP在计算映射矩阵时出现了邻域重叠问题.为了进一步提高LPP的性能,提出了一种局部保持分类投影(LPDP)算法.该算法首先利用样本数据的类别信息计算类间散度矩阵和类内散度矩阵,然后再把类间散度矩阵与类内散度矩阵之差引入到LPP的目标函数中,从而求得具有良好分类特性的投影向量.在ORL和Yale人脸数据库上的实验结果证明,提出的LPDP方法用于人脸识别方法具有较高的识别率,鲁棒性更好,表明所提出的方法是有效的.

Locality preserving projection is an effective method which can extract the feature and reduce dimensionality,and has been used in some field of pattern recognition.For the unsupervised nature of LPP and the overlapping problem,this paper presents a supervised algorithm of Local Preserving Discriminating Projection.Local Preserving Discriminating Projection firstly calculates the intra-class scatter and the inter-class scatter using the label of the sample data.Introducing them to objective function,we get classified projection vector.Experimental results on ORL and Yale databases suggest that the proposed Local Preserving Discriminating Projection provides a better way to solve these problems and demonstrates the effectiveness of the proposed method.