Ant colony system with extremal dynamics for point matching and pose estimation

For a point-based image registration method, point matching is a hard and a computationally intensive task, especially when issues of noisy and outlying data have to be considered. In this paper we cast the problem as a combinatorial optimization task and describe a global optimization method to achieve robust point matching and pose estimation for image registration purposes. The basic idea is to use an ant colony system (ACS) as a population based search strategy to evolve promising starting solutions, i.e affine transformations. An appropriate local search inspired by extremal optimization is developed and embedded within the search strategy to refine the solutions found. Experimental results are very promising and show the ability of the method to cope with outliers and achieve robust pose estimation.