点击次数：

论文类型：期刊论文

发表时间：2008-12-01

发表刊物：Journal of Information and Computational Science

收录刊物：EI、Scopus

卷号：5

期号：6

页面范围：2537-2544

ISSN号：15487741

摘要：Locality preserving projection (LPP) aims at finding an embedded subspace that preserves the local structure of data. Though LPP can provide intrinsic compact representation for image data, it has limitations on image recognition. In this paper, an improved algorithm called kernel scatter-difference based discriminant locality preserving projection (KSDLPP) is proposed. KSDLPP uses kernel trick method to map the input data into an implicit feature space where a scatter-difference discriminant rule based LPP is employed to seek a low-dimensional manifold subspace. Not only does KSDLPP describe complex nonlinear structure of the images, but it also avoids the singularity problem of high-dimensional data matrix and offers better classification capability. Experiment results on public face and palmprint databases also demonstrate the effective recognition performance of the KSDLPP algorithm. © 2008 Binary Information Press December 2008.