一种基于局部排序PCA的线性鉴别算法

主分量分析(Principal Component Analysis,PCA)是模式识别领域中一种重要的特征抽取方法,该方法通过K-L展开式来抽取样本的主要特征。基于此,提出一种拓展的PCA人脸识别方法,即分块排序PCA人脸识别方法(MSPCA)。分块排序PCA方法先对图像矩阵进行分块,对所有分块得到的子图像矩阵利用PCA方法求出矩阵的所有特征值所对应的特征向量并加以标识；然后找出这些所有的特征值中k个最大的特征值所对应的特征向量,用这些特征向量分别去抽取所属的子图像的特征；最后,在MSPCA的基础上,将抽取子图像所得到的特征矩阵合并,把这个合并后的特征矩阵作为新的样本进行PCA+LDA。与PCA和PCA+LDA方法相比,分块排序PCA由于使用子图像矩阵,可以避免使用奇异值分解理论,从而更加简便。在ORL人脸库上的实验结果表明,所提出的方法在识别性能上明显优于经典的PCA和PCA+LDA方法。

PCA Face Recognition Algorithm Based on Local Feature

Principal component analysis is an important feature extraction method of pattern recognition,and the main features of the method are extracted from the sample by KL expansion.Therefore,we proposed a method of face recognition PCA expansion,called modular sorting PCA face recognition method(MSPCA).MSPCA method first divides the image matrix into blocks.The feature vectors,corresponded to all the feature value,are obtained from the sub-image matrix of all sub blocks by using PCA method,and the feature vectors are identified.Then the method finds out the feature vectors,corresponded to k maximum feature value of all eigenvalues.These feature vectors are taken to extract the characteristic of sub-image.Finally,based on the MSPCA,the feature matrix extracted from sub-images is merged,and then the combined feature matrix is as a new sample to implement PCA+LDA.Compared with the PCA method and PCA+LDA method,because of the use of sub-image matrix,MSPCA avoids using singular value decomposition theory,which makes it easier.Experimental results on ORL face database show that the proposed method outperforms the classical identification PCA and PCA+LDA methods.