基于相关性度量的伪主成分分析

采用数据样本间的相关性作为相似性度量,并引入样本的类信息,提出一种新的降维方法,即伪主成分分析(Pseudo-PCA),该方法尽可能地保持原样本的变化信息,同时又使得降维后的同类数据样本尽可能保持相似。此外,将这种思想方法成功推广到近年来提出的2DPCA,MatPCA和(2D)2PCA。在ORL,Yale和AR等人脸数据集上的实验表明,该类方法的识别率高于相应的基于欧氏距离的PCA,2DPCA,M atPCA和(2D)2PCA等方法。

Correlation Metric Based Pseudo-Principal Component Analysis

A new dimensionality reduction method,called the pseudo-PCA,is proposed,in which the correlation between the samples is taken as the similarity metric.Meanwhile the class information of the samples is incorporated.Pseudo-PCA can preserve the variation information of the samples and enable the data within the same class to be similar to each other.Moreover,the idea of pseudo-PCA is generalized to the recently proposed 2DPCA,MatPCA and(2D)~2PCA.Experimental results on ORL,Yale and AR face datasets show that pseudo-PCA,-2DPCA,-MatPCA and-(2D)~2PCA based on the correlation metric outperform PCA,2DPCA,MatPCA and(2D)~2PCA based on the Euclidian distance.