Second-order graph model ant and colony optimization based image matching

Image matching is a fundamental problem in the computer vision field. This paper focuses on image matching based on the graph structure model. The methods of the graph model establishment in the second-order or high-order constraint are studied. In order to overcome the defects of traditional optimal algorithms which fall easily into the local optimal solution, this paper adopts the ant colony optimization algorithm to optimize the match score function and proposes an high-order graph matching algorithm based on ant colony optimization. It first applies the tensor matching algorithm to initialize the pheromone matrix to provide a good start point, adopts the affinity tensor to provide the priori knowledge for computing the heuristic factor, then calculates the transition probability using the pheromone and heuristic factor, and finally updates the pheromone in two ways by the solutions which have been searched. The two updating rules of pheromone are local and global. Experimental results show that this algorithm can get a higher matching accuracy and has a stronger robustness against deformation noises and outliers compared with others.