Particle swarm optimization for point pattern matching

The technique for point pattern matching (PPM) is essential to many image analysis and computer vision tasks. Given two point patterns, the PPM technique finds an optimal transformation for one point pattern such that a distance measure from the transformed point pattern to the other is minimized. This paper presents a new PPM algorithm based on particle swarm optimization (PSO). The set of transformation parameters is encoded as a real-valued vector called particle. A swarm of particles are initiated at random and fly through the transformation space for targeting the optimal transformation. The proposed algorithm is validated through both synthetic datasets and real fingerprint images. The experimental results manifest that the PSO-based method is robust against practical scenarios such as positional perturbations, contaminations, and drop-outs from the point sets. The PSO algorithm is also shown to be superior to a genetic algorithm and a simulated annealing algorithm on both effectiveness and efficiency.