Automatic Image-To-Map-Registration of Remote Sensing Data

With this paper, a fully automatic system for the registration of satellite image data with vector GIS data is presented. The implemented approach is based on polygonal objects which are extracted from the image (region growing segmentation) and from the GIS data set (merging of polygons with the same object class). Based on absolute affine invariant features, corresponding pairs of objects are determined by a backtracking depth-first tree search algorithm. Investigations concerning the robustness and discrimination power of the used absolute affine invariants are presented. Within the tree search algorithm, unary and binary constraints are applied in order to find a consistent subset of polygon pairs from all matching hypothesis. Furthermore, a coarse global affine transformation is determined in this step. Based on this transformation, all hypothesis being consistent with the global transformation are determined in a verification step. An individual affine transformation for each object pair is computed by using the formalism of standard positions., The individual transformations are refined for each verified sensor-model polygon pair in a refinement step. Finally, homologuous points (GCPs) with prefixed standard deviation of the residuals are extracted from the polygon boundaries. The whole process is exemplarily applied to a 25km x 20km section of the German ATKIS Vector data base and corresponding original SPOT PAN data. In this work, only forest polygons were used.