一种基于分块差图像的2DPCA人脸识别方法  

A Human Face Recognition Method of 2DPCA Based on Modular Residual Image

作　　者：施志刚[1] 

机构地区：[1]南通航运职业技术学院管理信息系,江苏南通226010

出　　处：《内蒙古师范大学学报（自然科学汉文版）》2015年第3期380-384,共5页Journal of Inner Mongolia Normal University(Natural Science Edition)

摘　　要：用2DPCA方法对分块差图像进行特征提取,以进一步提高人脸识别的效果.首先利用2DPCA算法对原始样本进行重构,建立差图像样本空间,以减轻光照差异对人脸识别带来的干扰;然后基于新的图像样本实施分块,考虑到不同位置分块图像的局部特征差异,为获取最佳的鉴别特征,采用2DPCA方法对所有对应位置的分块差图像训练样本各自提取特征;最后根据各训练样本和测试样本在特征空间的最小距离,即每个训练样本和测试样本所有对应分块图像在各自特征空间投影距离之和的最小值,完成模式分类.仿真实验结果表明,该方法使识别效果得到了有效地改善.In order to improve the human face recognition rate,a human face recognition method that uses two-dimensional principal component analysis（2DPCA）to extract features on the residual image is proposed in this paper.In this method,firstly,the 2DPCA algorithm is used to reconstruct the original samples for constructing the residual image sample space,in order to reduce the interference due to illumination variation in human face recognition;then the new image samples are blocked,taking into account the local feature differences of the block images in different positions,and the 2DPCA algorithm is used to extract feature for all the training samples of the block residual image in the corresponding position independently,in order to obtain the optimal discriminated features;finally,according to the minimum distance between each training sample and testing sample in the feature space,which is the minimum sum of distances between each training sample and testing sample block images in the corresponding position projecting in the respective feature space,so that pattern classification would be finished.The results of simulation experiments verify that the proposed method improve the recognition effectiveness.

关 键 词：差图像 2DPCA 重构 分块图像 对应位置 最佳鉴别特征 

分 类 号：TP391.41[自动化与计算机技术—计算机应用技术]