This paper presents a proposed approach of automatic image to image registration based on matching edge segments. Here, edge segments are extracted from the two images and their signatures are generated using chain coding. Segment signatures are utilized in a matching procedure to find common parts of edge segments. Cross correlation matching is then employed to refine the positions of midpoints of those common parts in the two images. The matching results can then used to compute the transformation parameters necessary for the transformation of the image to be matched.
The test images consist of two satellite images of different platforms and with different resolutions. One image is a QuickBird image with 0.6 m ground sample distance and the other image is a SPOT-5 image with a 2.5 m ground sample distance. The images cover the same area which is a somewhat urban area with insignificant height variation. The results reached by applying the proposed approach on the test images have proved the validity and efficiency of matching edge segments in automatic image-to-image registration.
|
