Abstract :
As an important task in image processing, image registration has applications in medical analysis, remote sensing and computer vision etc.. However, since images taken by different sensors, automatic image registration, especially automatic feature matching, is difficult. Depending on the fact that the distortion between those images is near affine transformation, affine model is selected as the coarse model, and its parameters are determined by improved feature consensus method together with convex hull of points set. Control points pairs are formed by searching the counterpart points within a threshold in the coarse registration result. Parameters of mapping polynomials are determined by control point pairs to refine the registration. Experimental results indicated that this approach is an efficient way to align images using feature points automatically.
