二维判别监督LPP算法在人脸识别中的应用

Application of Two-dimensional Discrimination and Supervision LPP in Face Identification

一维常规LPP方法应用于人脸识别数据时,由于通常存在矩阵奇异性问题,相应特征方程不可直接求解;目前已提出了二维局部保持投影算法(2DLPP)可以解决上述问题。但需要指出的是,2DLPP是一个非监督的学习方法,其只考虑了数据的距离关系,而忽视了合理处理不同类别样本间关系的问题。本文将介绍的2DLPP改进方法——二维判别监督局部保持投影(2DDSLPP)的方法能利用监督学习的手段改进2DLPP,提高其分类性能。通过实验证明,可以大大提高识别率。

Because of the matrices singularity,the corresponding characteristic equations can not be directly solved when the one dimensional regular LPP method is applied to face recognition data,the two dimensional locality preserving projection（2DLPP）can directly solve the above problems.But it should be pointed out that,2DLPP is an unsupervised learning method,it only considers the distance relationship of the data,and ignores to reasonably deal with the relationship among different categories.This article will introduce the improved method of 2DLPP——two-dimensional discrimination and supervision locality preserving projection（2DDSLPP）,it can use supervised learning method to improve 2DLPP,improve its classification performance.The experiment proves that the recognition rate is improved greatly.