A non-parametric invariant approach to image registration and its application to image rectification

摘要

This paper offers a non-parametric approach to image matching and its application to geometric correction. Geometric invariants are shape or surface descriptors which remain invariant under geometrical transformations such as projection or viewpoint change. Non-parametric invariants are used as a distance measure in the matching process. A new invariant is proposed than can handle slight non-linear effects and partial occlusions, which are so frequent in remote sensing imagery.