Uncorrelated Maximum Locality Preserving Projections

摘要：

Dimensionality reduction algorithms,which aim toselect a small set of efficient and discriminant features.In this paper,a new manifold learning algorithm,called Uncorrelated Maximum Locality PreservingProjections(UMLPP),to identify the underlying mani-fold structure of a data set.UMLPP considers both thebetween-class scatter and the within-class scatter inthe processing of manifold learning.Equivalently,thegoal of UMLPP is to preserve the intrinsic graphcharacterizes the interelass compactness and connectseach data point with its neighboring points of the sameclass. Different from Principal componentanalysis(PCA)that aims to find a linear mapping whichpreserves total variance by maximizing the trace offeature variance,While locality preserving projections(LPP)that is in favor of preserving the local structureof the data set.We choose proper dimension ofsubspace that detects the intrinsic manifold structurefor classification tasks.Extensive experiments on facerecognition demonstrate that the new featureextractors are effective,stable and efficient.

作者: Lin Kezheng Lin Sheng

作者单位: Harbin University of Science and Technology,Harbin 150080,China

会议类型: 国际会议

会议名称: 2008 3rd International Conference on Intelligent System and Knowledge Engineering(第三届智能系统与知识工程国际会议)(ISKE 2008)

会议地点: 厦门

会议语种:英文

页码: 1310-1313

在线出版日期: 2008-11-17（万方平台首次上网日期，不代表论文的发表时间）