Simultaneously Multiple-object Pattern Matching based on Multi-swarms Particle Swarm Optimization

Simultaneously multiple objects matching under 2D affine transformations (six parameters or degree of freedom) is key, yet not resolved problem in computer vision. This paper proposes an optimized and efficient pattern matching method by maximizing the Normalized Cross Correlation (NCC) coefficient measure based on Multi-swarms Particle Swarm Optimization (MPSO). PSO is a kind of stochastic optimized algorithm which is widely used in pattern recognition and computer vision. As a modification version of PSO, Multi-swarms PSO is designed to find multiple local optimums that approximate to the global optimum which correspond to multiple objects under 2D affine transforms. Experimental results using images demonstrate good performance in multi-Object pattern matching which involves huge computation complexity.