A Study Of High-order Graph Matching Method Based On Ant Colony Optimization

Image registration is the foundation of many image technologies and applied in many fields. As a key step in the field of image analysis,image registration is always the hot issue. As the important problems in image registration, high order graph matching attracts many scholars research.This paper first introduces the development of image registration and analyzes the related theory of image registration, and summarized the current main methods of registration. Secondly, we analysised the main steps of the registration method based on image features. High-order graph matching has attracted many researching. High-order graph matching method means that when we establishing correspondence we should consider the constraint conditions of high order. This paper introduces the relevant theories of higher order graph matching problem in detail.In order to overcome the shortcoming that traditional matching methods is easy to fall into laocal optimum, this paper proposed a new method to solve the problem of high order graph matching. The first step of this method is establishing a tensor which represents the affinity relation of two feature sets, and constructing the objective function with the tensor. Then we compute the heuristic accroding to tensor,and compute transition probability according heuristic and pheromone.Finally,we update the pheromone according the solution and get the matching resualt. Because we apply high-order matching method, the method is better than second-order matching method with higher accuracy. At the same time, the experiments of this method shows that the proposed algorithm perform better than other advanced matching methods in anti-noise field.The main innovation points of our methods are:(1)we apply the ant colony algorithm into the high-order graph matching problem,which make the solution global optimal and promote the accuracy.(2)we compute the heuristic with tensor which can accelerate the searching speed.(3)in each iteration,we save the best path which lead the searching direction towards to the best one.In the last chapter of this paper, we summarize our work and analyze the next work we can improve.