Parameter space warping: shape-based correspondence between morphologically different objects.

Abstract

This paper presents a novel and comprehensive method for the automated determination of correspondences between two morphologically different two-dimensional (2-D) or three-dimensional (3-D) objects. Correspondences are determined by warping parametric representations of the objects to be matched. The warp is guided by the minimization of a similarity criterion function that measures features related to structural correspondence, including Euclidian point-to-point distance and differences in normals and curvature. The method uses a continuous harmonic parameterization for both the object and the warp, which provides: 1) a high degree of computational efficiency; 2) robust extraction of differential features, not subject to discretization errors or noise amplification in differentiation; 3) direct formulation of constraints to avoid overlaps in the resulting correspondence set; and 4) a scale-space paradigm of object shape and warp. The new method does not search for individual landmarks, but operates with a complete, integrated representation of the object geometry. The method was tested on 2-D and 3-D objects with substantial shape differences. Results demonstrated substantial improvements of 2%-33% in correspondence accuracy and 15%-59% in correspondence quality compared with direct registration methods.