Object oriented hierarchical classification of high resolution remote sensing images

The appearance of the satellite images in very high resolution is a real opportunity for the geographical identification of objects in urban zones. These images provide a huge amount of data about land cover surface and allow the perception of objects on the ground which was not observable in lower resolutions e.g. Ikonos images. Nevertheless, their heterogeneousness perturbs the methods of classic classification, also called pixel based methods. In this paper we propose an object oriented approach for extracting urban objects. Our approach is divided into two steps: the first is a hierarchical segmentation based on region-merging according spatial (texture) and spectral (NDVI, IB) criteria. The second is a regions classification using the non supervised approach.