计算机应用研究2017，Vol.34Issue(5)：1560-1564,5.DOI:10.3969/j.issn.1001-3695.2017.05.063
基于最大边界准则的稀疏局部嵌入特征提取方法
Sparse local embedding feature extraction method based on maximum margin criterion
摘要
Abstract
The local linear embedding(LLE) was unable to take advantage of the discrimination information of the samples and the maximum margin criterion (MMC) had a weak performance on the nonlinear data.Therefore this paper proposed a feature extraction method called sparse local embedding based on maximum margin criterion (SLE/MMC).With the preservation of local nearest neighbor premise,firstly,the similar samples were gathering together as much as possible in the intrinsic graph.Secondly,the samples of different classes were as far as possible from each other in the penalty graph.Finally,it used the elastic net regression to obtain an optimal sparse projection matrix.In order to avoid the "small sample size" problem,it constructed the objective function by MMC.The experiment results on ORL,Yale and UMIST show that,compared with other methods (PCA,LLE and MMC),SLE/MMC has a higher recognition rate,indicating that this method is more efficient in feature extraction.关键词
特征提取/局部线性嵌入/最大边界准则/弹性网回归Key words
feature extraction/local linear embedding/maximum margin criterion/elastic net regression分类
信息技术与安全科学引用本文复制引用
刘毛溪,万鸣华,孙成立,王巧丽..基于最大边界准则的稀疏局部嵌入特征提取方法[J].计算机应用研究,2017,34(5):1560-1564,5.基金项目
国家自然科学基金资助项目(61462064,61272077,61203243,61262019,61362031) （61462064,61272077,61203243,61262019,61362031）
高维信息智能感知与系统教育部重点实验室(南京理工大学)基金资助项目(30920140122006) （南京理工大学）
中国博士后基金资助项目(2014T70453,2013M530223) （2014T70453,2013M530223）
