一种改进的模块2DPCA人脸识别新方法  

New Face Recognition Method Based on Improved Modular 2DPCA

作　　者：李晓东[1] 费树岷[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,南京210096

出　　处：《系统仿真学报》2009年第15期4672-4675,共4页Journal of System Simulation

基　　金：国家自然科学基金(60574006)

摘　　要：提出了一种改进的模块2DPCA方法,即基于类内平均脸的分块2DPCA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人脸库上的实验结果表明,提出的方法在识别性能上明显优于2DPCA方法和普通模块2DPCA方法。An improved modular 2DPCA method, that is modular 2DPCA method based on With-in Class Average Face, was proposed. Firstly, the with-in class mean of each sub-image of all training samples in each class was calculated, and it was used 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 could be obtained accordingly. Secondly, when all sub-images of training samples and testing samples were projected to the best projecting matrix that was gotten above, the recognition features was produced, Finally, the nearest distance classification was used to distinguish each face. The experiment results on ORL face database indicate that the improved modular 2DPCA is obviously superior to that of 2DPCA and general modular 2DPCA.

关 键 词：二维主成分分析 类内平均脸 模块化二维主成分分析 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]