Abstract: First, a coarse registration model was built. Second, after image edge extraction, Hough transform was employed to extract matched line. Virtual corner dataset was constructed on the basis of the matched lines, description of the dataset was performed, the initial homonymous points are extracted according to the constraint geometry and mutual information. Finally, RANSAC algorithm was used to obtain the final results of the matching, a great amount of high precision control point pairs are used to construct TIN for tiny facet primitive rectification. Experiment results demonstrate that the presented method can extract more matching points than that SIFT matching method can. The Root of Mean Square Error of the matching points obtained by this method is lower than 1 pixel.

Key words: remote sensing technology, registration, multi-source, high resolution satellite images