Satellite Image Registration through Structural Matching

摘要

A method is proposed for registration of satellite images based on the matching of their high-level symbolic description. The method consists of the following steps. First the images are segmented and the objects extracted are represented using a set of shape and grey-level descriptors. For each image, the spatial relations between the objects is computed and a description in the form of a directed graph is obtained. Finally, the best match between the structures is computed using relaxation techniques. The advantages of the method are that it can be made insensitive to a large class of geometric and radiometric transformations. Through a proper choice of attribute descriptors and spatial relations, a matching can be found even in the presence of geometric distortions like scaling, rotation, and others. Radiometric distortions can also be handled in a similar way so it can be used to match images of very different natures such as images obtained with different sensors or even to match satellite images and maps. The method is illustrated with several examples using satellite images and images and maps.