Pattern matching by sequential subdivision of transformation space

Pattern matching is a well-known pattern recognition technique. This paper proposes a novel pattern matching algorithm that searches transformation space by sequential subdivision. The algorithm subdivides the transformation space in depth-first manner by conducting Boolean operations on the constraint sets that are defined by pairs of template points and target points. For constrained polynomial transformations that have no more than two parameters on each coordinate, a constraint set can be represented as a 2D polygon or a Cartesian product of 2D polygons. Then, the Boolean operations can be computed through generic polygon clipping algorithms. Preliminary experiments on randomly generated point patterns show that the algorithm is effective and efficient under practical conditions.