基于RSM的判别分析集成方法  

Discriminant analysis method ensemble based on RSM

作　　者：周中侠[1] 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]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：《计算机工程与设计》2013年第12期4247-4251,共5页Computer Engineering and Design

基　　金：国家自然科学基金项目(61175111);江苏省高校自然科学基金项目(10KJB510027)

摘　　要：针对非监督鉴别投影(UDP)算法分类能力弱和性能不稳定问题,提出了一种基于RSM(random subspace method)的判别分析方法集成(DAC_EL)。该方法对训练样本的特征向量空间进行随机采样,得到多个UDP投影变换,因此获得多个既有差异又互补的投影子空间。由测试样本测试集成各基分类器分类精度,分类精度作为基分类器集成权重。基分类器分类结果线性权重集成给出集成分类器的输出结果。在ORL和YALE人脸图像库上实验结果表明,DAC_EL方法性能明显优于UDP方法和基于多数投票法的UDP集成方法,能够提高人脸识别的分类能力和稳定性。A discrimination analysis classification based on ensemble learning method （DAC_EL） is .proposed to improve the recognition performance of a single unsupervised discriminant projection （LIDP）. At the training stage, based on random sampling feature of training set, lots of UDP projecting transformations are gotten from RSM （Random Subspace Method）. Lastly, get many differences and complementary projection subspace. By test samples, the accuracy of individual classifiers in the ensemble are gotten, which are used as their weights in linear weight ensemble. The experimental results on ORL and YALE illustrate that the performance of DAC_EL method is superior to single UDP classifier and majority votes UDP ensemble, and significant- ly improve the classification ability and stability of the face recognition.

关 键 词：随机子空间方法 非监督鉴别投影 集成学习 子空间 人脸识别 

分 类 号：TP391[自动化与计算机技术—计算机应用技术]