A morphological process of high resolution remote sensing imagery for significant landscape unit segmentation

We propose an auto-adaptive segmentation approach (non-linear) based on a description of satellite images using an adjacency graph. The represented space is modeled using the Voronoi diagrams and their dual graph (Delaunay triangulation). The morphological transformations of opening and closing are applied to obtain suitable and significant regions conserving the principal features of the most important landscape units. The results of this approach have been tested using a SPOT5 image of Toulouse, France.