Image characterization from statistical reduction of local patterns

Abstract

This paper tackles the image characterization problem from a statistical analysis of local patterns in one or several images. The induced image characteristics are not defined a priori, but depends on the content of the images to process. These characteristics are also simple image descriptors and thus considering an histogram of these elementary descriptors enables to apply "bags of words" techniques. Relevance of the approach is assessed when dealing with the image recognition problem in a robot application framework. © 2009 Springer-Verlag Berlin Heidelberg.