Affine Matching of Planar Sets

To recognize an object in an image, we must determine the best-fit transformation which maps an object model into the image data. In this paper, we propose a new alignment approach to recovering those parameters, based on centroid alignment of corresponding feature groups built in the model and data. To derive such groups of features, we exploit a clustering technique that minimizes intraclass scatter in coordinates that have been normalized up to rotations using invariant properties of planar patches. The present method uses only a single pair of 2D model and data pictures even though the object is 3D. Experimental results both through computer simulations and tests on natural pictures show that the proposed method can tolerate considerable perturbations of features including even partial occlusions of the surface. c © 1998 Academic Press