Invariant pattern recognition: A review

摘要

In this document we review and compare some of the classical and modern techniques for solving the problem of invariant pattern recognition. Such techniques include integral transforms, construction of algebraic moments and the use of structured neural networks. In all cases we assume that the nature of the invariance group is known a priori. Many of the methods described apply to specific geometrical transformation groups; however some of the techniques are highly general and applicable to large classes of invariance groups. We also review some results regarding the existence and structure of invariants under certain kinds of groups.