Affine Invariants of Planar Sets

Abstract:
Recent research has indicated that invariants can be useful in computer vision for identification and pose determination of objects. The idea is to find functions that are invariant under a set of transformations acting on a configuration space. This paper describes some new viewpoints on the construction and use of such invariants. The key idea is that any kind of features like derivatives, distinguished points or integrative features can be used to construct invariants. In a given viewing situation one should choose those features that are most stable. As examples, affine invariants for planar smooth curves, planar regions, and planar point configurations are given. The properties of the invariants are illustrated with experiments