判别近邻保持嵌入人脸识别  

Discriminant neighborhood preserving embedding algorithm for face recognition

作　　者：田玉敏[1] 云艳娥[1] 马天骏[1] 

机构地区：[1]西安电子科技大学计算机学院,陕西西安710071

出　　处：《西安电子科技大学学报》2011年第3期24-28,98,共6页Journal of Xidian University

基　　金：陕西省自然科学基础研究计划资助项目(2010JM8011)

摘　　要：针对普通近邻保持嵌入算法侧重保持样本的局部结构,而没有考虑样本的类判别信息以及小样本问题,提出了一种新的人脸识别算法———判别近邻保持嵌入算法.在近邻保持嵌入算法的基础上,将最大散度差准则引入到其目标函数中.在嵌入低维空间后,类内样本保持它们固有的近邻几何结构关系,而类间样本彼此分离,能够充分提取具有判别力的特征.在AT&T人脸数据库上进行的对比实验表明,与主成分分析、线性判别分析以及近邻保持嵌入算法相比,判别近邻保持嵌入算法的最高识别率分别提高了15.35%、6.47%和6.94%;在Yale人脸数据库上进行的对比实验表明,判别近邻保持嵌入算法的最高识别率分别提高了20.27%、5.63%和2.27%.Neighborhood preserving embedding （NPE） emphasizes the face sample manifold local structure, without taking into account the sample class discriminant information and the small sample problem. The algorithm is based on NPE and the maximum scatter difference criterion （MSDC） is introduced to its objective function. After being embedded into a low dimensional subspace, the samples of the same class maintain their intrinsic neighbor relations while the samples of the different classes are far from each other. And then the most discriminative feature is extracted. Experiments in the AT ＆ T face database show that the highest recognition rate of the algorithm has increased by 15.35%, 6. 47%, and 6. 94%, respectively, comparied with the PCA, LDA and NPE algorithm. Comparative experiments in the Yale face database show that the highest recognition rate of the algorithm has increased by 20. 27%, 5.63%, and 2. 27%, respectively.

关 键 词：人脸识别 近邻保持嵌入 最大散度差准则 

分 类 号：TP391.4[自动化与计算机技术—计算机应用技术]