最大边缘准则图形嵌入在人脸识别中的应用

摘要

Locally Linear Embedding(LLE) is an effective nonlinear dimensionality reduction method,which can maintain the same topology relation ship with the original space.However,it could not be widely applied in dimensionality reduction processing,visualization and classification.For these problems,a new dimensionality reduction and data classification method-embedding method based on maximum margin criterion(EM/MMC) is proposed in this paper.This algorithm gathers together the nearest samples as well as maximizes class intervals.The experiment of two widely used face databases (ORL and Yale) show that the proposed approach is effective and feasible.%局部线性嵌入是一种有效地非线性维数约减方法,它能保持降维后的数据与原空间有相同的拓扑关系.但是这种方法在降维处理、可视化以及数据分类方面应用不是很广泛,针对上述问题,提出了一种新的、有效的降维以及数据分类方法——基于最大边缘准则图形嵌入方法.该方法首先构建最近邻关系图聚合数据点之间的最近邻样本,同时最大化类间间隔,保证不同类之间数据可分性大,从而更好地实现数据分类.最后,该方法的有效性分别在ORL及Yale两大人脸库上得到了验证.