一种改进的模块PCA方法及其在人脸识别中的应用  

Improved Modular PCA Method and Its Application in Face Recognition

作　　者：李晓东[1] 费树岷[1] 张涛[1] 

机构地区：[1]东南大学自动化学院复杂工程系统测量与控制教育部重点实验室,江苏南京210096

出　　处：《测控技术》2008年第11期19-21,24,共4页Measurement & Control Technology

摘　　要：提出了一种改进的模块PCA方法,即基于类内平均脸的分块PCA算法。该算法对每一类训练样本中每个训练样本的每一子块求类内平均脸,并用类内平均脸对训练样本类内的相应子块进行规范化处理,然后由所有规范化后的子块构成总体散布矩阵,从而得到最优投影矩阵;由训练集的全体子块的平均值对训练样本的子块和测试样本的子块进行规范化后投影到最优投影矩阵,得到识别特征;最后用最近距离分类器分类。在ORL人脸库上的试验结果表明,提出的方法在识别性能上明显优于普通模块PCA方法。Based on improved modular PCA, that is modular PCA based on the within class average face, a human face recognition technique based on PCA is presented. Firstly, the with-in class mean of each sub-image of all training samples in each class is calculated, and using it to normalize corresponding sub-image of with-in class. After that, the best projecting matrix from general matrix that is made up of all normalized sub-images can be obtained accordingly. Secondly, when all sub-images of training samples and testing samples are projected to the best projecting matrix that has been got above, the recognition features is produced; Finally, the nearest distance classification is used to distinguish each face. The experiment results on ORL face database indicate that improved modular PCA is obviously superior to that of general modular PCA.

关 键 词：主成分分析 类内平均脸 分块PCA 特征矩阵 人脸识别 

分 类 号：TP391[自动化与计算机技术—计算机应用技术]