M-Isomap: Orthogonal Constrained Marginal Isomap for Nonlinear Dimensionality Reduction

被引:78
作者
Zhang, Zhao [1 ]
Chow, Tommy W. S. [1 ]
Zhao, Mingbo [1 ]
机构
[1] City Univ Hong Kong, Dept Elect Engn, Kowloon, Hong Kong, Peoples R China
关键词
Isomap; manifold learning; nonlinear dimensionality reduction (DR); pairwise constraints (PCs); visualization; EIGENMAPS;
D O I
10.1109/TSMCB.2012.2202901
中图分类号
TP [自动化技术、计算机技术];
学科分类号
0812 ;
摘要
Isomap is a well-known nonlinear dimensionality reduction (DR) method, aiming at preserving geodesic distances of all similarity pairs for delivering highly nonlinear manifolds. Isomap is efficient in visualizing synthetic data sets, but it usually delivers unsatisfactory results in benchmark cases. This paper incorporates the pairwise constraints into Isomap and proposes a marginal Isomap (M-Isomap) for manifold learning. The pairwise Cannot-Link and Must-Link constraints are used to specify the types of neighborhoods. M-Isomap computes the shortest path distances over constrained neighborhood graphs and guides the nonlinear DR through separating the interclass neighbors. As a result, large margins between both inter-and intraclass clusters are delivered and enhanced compactness of intracluster points is achieved at the same time. The validity of M-Isomap is examined by extensive simulations over synthetic, University of California, Irvine, and benchmark real Olivetti Research Library, YALE, and CMU Pose, Illumination, and Expression databases. The data visualization and clustering power of M-Isomap are compared with those of six related DR methods. The visualization results show that M-Isomap is able to deliver more separate clusters. Clustering evaluations also demonstrate that M-Isomap delivers comparable or even better results than some state-of-the-art DR algorithms.
引用
收藏
页码:180 / 191
页数:12
相关论文
共 24 条
[1]  
[Anonymous], 1994, Introduction to parallel computing: design and analysis of algorithms
[2]  
[Anonymous], 1994, Multidimensional Scaling
[3]  
[Anonymous], ADV NEURAL INFORM PR
[4]  
[Anonymous], 1980, Multivariate Analysis
[5]   Laplacian eigenmaps for dimensionality reduction and data representation [J].
Belkin, M ;
Niyogi, P .
NEURAL COMPUTATION, 2003, 15 (06) :1373-1396
[6]   Document clustering using locality preserving indexing [J].
Cai, D ;
He, XF ;
Han, JW .
IEEE TRANSACTIONS ON KNOWLEDGE AND DATA ENGINEERING, 2005, 17 (12) :1624-1637
[7]  
de Ridder D, 2003, LECT NOTES COMPUT SC, V2714, P333
[8]   Hessian eigenmaps: Locally linear embedding techniques for high-dimensional data [J].
Donoho, DL ;
Grimes, C .
PROCEEDINGS OF THE NATIONAL ACADEMY OF SCIENCES OF THE UNITED STATES OF AMERICA, 2003, 100 (10) :5591-5596
[9]   Supervised nonlinear dimensionality reduction for visualization and classification [J].
Geng, X ;
Zhan, DC ;
Zhou, ZH .
IEEE TRANSACTIONS ON SYSTEMS MAN AND CYBERNETICS PART B-CYBERNETICS, 2005, 35 (06) :1098-1107
[10]   A generalized Foley-Sammon transform based on generalized fisher discriminant criterion and its application to face recognition [J].
Guo, YF ;
Li, SJ ;
Yang, JY ;
Shu, TT ;
Wu, LD .
PATTERN RECOGNITION LETTERS, 2003, 24 (1-3) :147-158