On a study of invariant features in nonrigid transformations

Abstract

Invariant features are of prime importance in computer vision. These features are any properties of the surface that are not changed during a transformation, and which can be used for object recognition or motion analysis. Two categories of invariancies can be identified: invariance with respect to the rigid motion, and invariance with respect to the nonrigid motion. The authors concentrate on the differential geometric properties that can be used to detect the nonrigid motion of a given surface. Analysis of invariant features of a surface that can be used in recognition of nonrigid motion has been addressed. A novel method for estimating point correspondences between surfaces undergoing a small deformation is presented.<>