An indexing model of remote sensing images

Abstract

This paper is concerned with the problem of indexing remote sensing images. This kind of images has a semantics mainly related to the image spectral properties. For these reasons the spectral properties can be considered as effective image descriptors. The model proposed in this paper assumes that the image descriptors are spectral regions and their spectral signatures. By the application of a clustering algorithm each image is segmented into a set of spectral regions to be associated to basic (pre-defined) ground cover classes. To take into account the uncertainty that often affects the cluster labelling process the indexing model generates for each region and each reference class a possibility degree indicating the possibility that the region corresponds to that class. The uncertainty of this association is explicitly modelled, and allows the definition of a more flexible image representation with respect to a crisp approach. © Springer-Verlag 2004.