Affine invariant matching of broken boundaries based on particle swarm optimization

Affine invariant matching of broken image contours with model shapes is an important but difficult research topic in computer vision. One of the effective approaches to date encapsulates the process as an optimization problem which determines, with the use of a Simple Genetic Algorithm (SGA), the best matching score between pairs of object boundaries. Despite the moderate success of methods developed in this direction, the overall success rate is generally low and inconsistent amongst test trials. This unfavorable outcome could be due to the lack of adequate exploitation in an enormous and erratic search space, which is rather common in the context of shape matching. In this paper, a novel scheme based on Particle Swarm Optimization (PSO) is presented to overcome these problems. Experimental results reveal that the proposed method has outperformed SGA and Real Coded Genetic Algorithm (RCGA) in terms of speed, stability and success rate. In addition, the evolutionary behavior of PSO also permits the use of repeated trials to further enhance the success rate towards perfection with relatively fewer iterations.