Affine and Projective Normalization of Planar Curves and Regions

Abstract:
Recent research has shown that invariant indexing can speed up the recognition process in computer vision. Extraction of invariant features can be done by choosing first a canonical reference frame, and then features in this reference frame. This paper gives methods for extracting invariants for planar curves under affine and projective transformations. The invariants can be used semilocally to recognize occluded objects. For affine transformations, there are methods giving a unique reference frame, with continuity in the Hausdorff metric. This is not possible in the projective case. Continuity can, however, be kept by sacrificing uniqueness