Video analysis via nonlinear dimensionality reduction

Abstract

In this work we present an application of nonlinear dimensionality reduction techniques for video analysis. We review several methods for dimensionality reduction and then concentrate on the study of Diffusion Maps. First we show how diffusion maps can be applied to video analysis. For that end we study how to select the values of the parameters involved. This is crucial as a bad parameter selection produces misleading results. Using color histograms as features we present several results on how to use diffusion maps for video analysis. © Springer-Verlag Berlin Heidelberg 2007.