基于Bagging的LDA集成算法及其在人脸识别中的应用  

Face image recognition based on Bagging LDA ensemble method

作　　者：周中侠[1,2] 王正群[1] 徐春林 李峰[1] 薛巍[1] 

机构地区：[1]扬州大学信息工程学院,江苏扬州225127 [2]国网沾化县供电公司,山东滨州256800 [3]北方激光科技集团有限公司激光应用技术部,江苏扬州225009

出　　处：《扬州大学学报（自然科学版）》2015年第1期41-44,共4页Journal of Yangzhou University：Natural Science Edition

基　　金：国家自然科学基金资助项目(61402395);江苏省高校自然科学基金资助项目(10KJB510027)

摘　　要：针对线性判别分析(linear discriminant analysis,LDA)在处理人脸图像时对局部变化识别不稳定性的问题,提出一种基于Bagging的LDA人脸识别方法 Bagging LDA.首先将人脸图像集划分成若干子图像集,然后将Bagging应用于每个子图像集并融合初始图像集构建多个LDA分类器,选择其中一个分类性能最好的分类器作为集成基分类器,最后使用权重投票的方法将各个基分类器的分类结果进行组合.仿真结果表明,Bagging LDA算法识别性能好,精度较高,在图像遮挡、表情变化等条件下仍具有较强的稳定性.When LDA deals with the high dimensional face image, recognition instability of the local change is a challenging problem. In order to solve it, anew method named Bagging LDA is proposed based on LDA and ensemble learning method in this paper. More specifically, a face image is divided into several sub-image set, and then a series of training set of LDA based on Bagging resample sub-image set and initial sub-image set are constructed. In the stage of classifiers ensemble, a component classifier with best performance is selected from a set classifiers. Weighting combination of all component classifiers gives ensemble classifier output. Experiments on ORL and YALE face database show that the proposed Bagging LDA method is effect in recognition performance.

关 键 词：人脸识别 集成学习 线性判别分析 BAGGING算法 

分 类 号：TP391.4[自动化与计算机技术—计算机应用技术]