Scene understanding for settlements from metric resolution imagery

Summary for only given, as follows. Satellite and airborne remote sensing has reached a new level of sophistication, but available interpretation methodologies cannot cope with the huge amounts of acquired data. At the same time, the new generations of metric resolution sensors, e.g: Ikonos, and the huge potential of synthetic aperture radar SAR have opened up the perspective on novel applications related to the understanding of high complexity settlement scenes. At meter resolution, mainly for man-made scenes, the complexity of the scene structures and of imaging phenomenology can be very high. This is reflected in the complexity of the observed images. New methods are needed for their interpretation, both for 2- and 3- dimensional analysis. The present article proposes methods of 2-dimensional information extraction from SAR and optical metric resolution observations. The presented methods are based on several assumptions: 1) the most compact encoding of the data is by the probabilistic model that describes it best, 2) a large number of sources of information coexist within the same observation data set, 3) the understanding of a scene requires complementary or multisensor observations. Thus, the concept developed aims at a description of the scene using data and information fusion. The primitive features, signal parameters, are extracted using model based methods such to obtain a quasi complete description of image content, i.e. radiometric/polarimetric attributes, structural information, geometric features, analysis at multiple scales. A hierarchic Bayesian modeling and learning paradigm is used for data and information fusion and for interactive exploration of the scene identity. The article presents the image formation phenomenology relevant for metric resolution observations, a concept for quasi-complete image content characterization, methods for supervised learning for scene classification by information. fusion, and examples using X-SAR, E-SAR (X band, and L band polarimetric) and Ikonos images