AUTOMATIC EXTRACTION OF LANDCOVER THEMES ON DIGITAL ORTHOPHOTOS IN MOUNTAINOUS AREA FOR MAPPING AT 1/25K

摘要

IGN-F is undertaking the production of a new digital 1/25k national base map. In this context, we developed an automatic process for extracting lancover information out of digital orthophotos. This process must be robust, self-adaptive for variation of image quality, date of acquisition, geographical location..., and must be usable also in an updating context. For this, we used a general strategy of information fusion from different sources: DTM, general geographic a priori knowledge, old or imprecise information. The uncertainty is managed throughout the full process until the final cartographic representation. More precisely, the images are first corrected for atmospheric effects, shadows, sun exposition... Probability functions for the different themes, knowing image radiometry, altitude, slope, orientation, are built. All these informations are merged using a simple form of Dempster Shafer rule on pixel basis. The pixel information is then extended to regions obtained by an image segmentation.