Image registration in machine vision is a very popular and useful research topic in both theoretical and practical problems. Geometric pattern matching is one of the most promising research directions in image registration. Major difficulties for pattern matching are performance, scale- and rotation-invariance, and robustness. It is hard to achieve these three objectives at the same time. Circular template sampling intrinsically bears rotation-invariance with lower computational complexity. In the proposed method, under circular sampling, the templates were sampled uniformly on 24-way to retrieve the consecutive feature vectors from the edge points. The scaling and rotation problem of the target pattern with respect to the template pattern is simplified as a problem of proportion of the length and an angle difference of the feature vectors. In the proposed, the scale- and rotation-invariance is elegantly achieved with high performance by representing a geometric pattern as a set of feature vectors.
