A Comparison of Registration Techniques on Quickbird Satellite Images

In remote sensing, because of the wide diversity of image characteristics (size, spatial and radiometric resolution, terrain relief, observation poses, etc.), image registration methods that may work well on certain satellite images will not produce acceptable results for others, requiring more powerful techniques. A variety of registration techniques for images with non-rigid geometric differences has been proposed, including piecewise (linear or cubic) functions, weighted mean functions, radial basis functions, Bspline functions, etc. This paper compares three of them: polynomial, piecewise linear and radial basis –thin plate spline– functions; and evaluates their accuracy according to two well-known metrics: root mean square error (RMSE) and circular error with 90% confidence (CE90). The comparison focuses on pan-sharpened QuickBird images (0.6 meters/pixel) acquired on different dates, from different observation attitudes, and sensing different land covers: urban area, high relief terrain and a combination of both. The experimental results show that local methods (as the radial basis functions) perform much better than global ones based on polynomial functions since they can exploit the geometrical information captured by a (desirable) large number of control points. * Corresponding author.