Learning a local maximal marginal embedding: Subspace
Cairong Zhao,Chuancai Liu
2010-01-01
Abstract:Many problems in information processing involve some form, ofdimensionality reduction. In graph embedding algorithm, how to construct a,neighborhood, graph that can correctly reflect the relationship between samplesis extremely vital. This paper reconstructs the interclass neighborhood, graphand, develops a, new approach for dimensionality reduction of high dimensionaldata, called, local maximal marginal (interclass) embedding (LMME), to manifoldlearning and, pattern recognition. LMME can be seen as a, linear approach of a,multimanifolds-based learning framework which integrates the information ofneighbor and, class relations. LMME characterizes the local maximal marginalscatter as well as the local intraclass compactness, seeking to find aprojection that maximizes the local maximal margin and, minimizes the localintraclass scatter. This characteristic makes LMME more powerful than the mostup-to-data, method, Marginal Fisher Analysis (MFA), and, maintains all theadvantages of MFA. The proposed, algorithm, is applied, to face recognition and,is examined, using the USPS handwriting digital database and, AR face imagedatabase. The experimental results show LMME outperforms PCA, LDA and, MFA,owing to the locally discriminating nature. This demonstrates that LMME is aneffective method for face recognition. ICIC International © 2010.