Faster affine invariant point pattern matching using graphical models

This paper proposed a fast junction tree algorithm for performing affine invariant point pattern matching using probabilistic Graphical models. First we utilize the convex hull of point-sets and least square method to obtain three non-collinear reference points, which are then used to compute affine-invariant features. Then we propose a dynamic programming algorithm on junction tree for point pattern matching problem which is efficient and guaranteed to find a maximum probability configuration. Experiments were carried out on both synthetic data and real-world data. The results show our algorithm is significantly faster and have higher accuracy compare to existing algorithm.