EURASIP Journal on Advances in Signal Processing 
Volume 2008 (2008), Article ID 862015, 12 pages
doi:10.1155/2008/862015
Research Article

A Metric Multidimensional Scaling-Based Nonlinear Manifold Learning Approach for Unsupervised Data Reduction

M. Brucher,1,2 Ch. Heinrich,1 F. Heitz,1 and J.-P. Armspach2

1Laboratoire des Sciences de l'Image, de l'Informatique et de la Télédétection, LSIIT, UMR 7005, CNRS-Université Louis Pasteur, Strasbourg 1, Boulevard S. Brant, BP 10413, 67412 Illkirch Cedex, France
2Laboratoire d'Imagerie et de Neurosciences Cognitives, LINC, UMR 7191, CNRS-Université Louis Pasteur, Strasbourg 1, LINC-IPB, 4, rue Kirschleger, 67085 Strasbourg Cedex, France

Received 30 September 2007; Revised 21 January 2008; Accepted 7 March 2008

Recommended by Olivier Lezoray

Abstract

Manifold learning may be seen as a procedure aiming at capturing the degrees of freedom and structure characterizing a set of high-dimensional data, such as images or patterns. The usual goals are data understanding, visualization, classification, and the computation of means. In a linear framework, this problem is typically addressed by principal component analysis (PCA). We propose here a nonlinear extension to PCA. Firstly, the reduced variables are determined in the metric multidimensional scaling framework. Secondly, regression of the original variables with respect to the reduced variables is achieved considering a piecewise linear model. Both steps parameterize the (noisy) manifold holding the original data. Finally, we address the projection of data onto the manifold. The problem is cast in a Bayesian framework. Application of the proposed approach to standard data sets such as the COIL-20 database is presented.