Research Of Image Feature Matching Algorithm Based On Graph Theory

Image matching refers to the spatial alignment of images got under different imaging conditions to determine the geometric mapping relationship between images,and then make the images to match.Image matching technology is one of the key technologies in computer vision,which is the basic problem in image processing and analysis.Image matching is becoming more and more widely used in the fields of object recognition,texture discovery and analysis,image information fusion,image retrieval and so on,which is of great research significance.Because feature-based image matching algorithm has the good stability and robustness on image scaling change and affine transformation,it has been widely concerned by scholars both at home and abroad.As a tool for describing data,graph model can effectively represent the structural features of images,and keep the interrelationship between regions.The use of graph model to achieve image feature point matching is favored by academia.The image feature point matching method based on graph theory is a hot and difficult problem in recent years because of its good adaptability and high matching precision.In this paper,we focus on the research of image feature matching method based on graph theory.The main study contents can be concluded as following:(1)This paper analyzes the theoretical significance and practical value of image matching,and summarizes the research situation of image matching both at home and abroad.We focus on the overview of image feature matching theory.Firstly,we introduce the basic concept and matrix representation of the graph.Then we introduce two key techniques in image feature matching:feature extraction and feature description.At last,the classic SIFT image feature matching algorithm is analyzed in detail.The related theory of graph and the study of SIFT algorithm have laid an important theoretical basis for the development of image matching algorithm in this paper.(2)On the image feature point matching and the thought of hierarchical clustering,this paper proposed an image matching algorithm based on top-down split clustering.The main idea of this algorithm is to use mutual k neighbor graph model to represent the corresponding relationship between images.In the mutual k neighbor graph representation model,the vertices represent the corresponding relationship between feature points.The edges between the vertices represent the geometric compatibility of corresponding relationships.The definition of the cluster density function can be measured whether or not belong to the same cluster.Under normal circumstances,the larger the value of cluster density,the more likely the correct cluster.The algorithm can not only obtain the corresponding relationship between images,but also indicate which corresponding relationship belongs to the same object.The corresponding relationship within the same cluster has higher geometric compatibility,and the compatibility of corresponding relationship between different clusters is lower,so different objects will present different clusters.On the basis of mutual k neighbor graph representation model,the cluster of the figure is obtained by the cluster detection method,and the idea of split clustering is used.Finally,the goal of image matching can be achieved by restoring the corresponding relationship within the cluster.Experimental results on real images show that the image matching algorithm of top-down split clustering is superior to ACC algorithm in matching performance,and improves the recall and precision of image matching.The effect picture and the quantitative analysis of experiment show that the algorithm has good matching results.(3)In order to further improve the accuracy of the image feature matching algorithm,this paper proposed a feature description and feature matching algorithm based on local neighborhood graph.By constructing a local neighborhood graph for each feature point,the structure information of the image is deeply mined.The algorithm firstly detects the initial feature points by FAST and SURT algorithm,and then constructs local neighborhood graphs for all the feature points.Each local graph is composed of the feature point and its neighboring feature points,thus forming a novel feature description method.On the basis of this novel feature descriptor,a similarity measure function and an energy function are given.In view of this,a feature matching algorithm based on local neighborhood graph model is proposed.In order to verify the effectiveness of the algorithm,two experiments were carried out:Gaussian noise simulation experiment and real image matching experiment.The purpose of the Gaussian noise simulation experiment is to analyze the effect of outliers and deformation noise on the performance of the algorithm,while the experiment on the real image database is to verify the accuracy of the algorithm in image feature matching.The instance diagrams of experiment and quantitative analysis results show that compared with SM algorithm,the feature matching algorithm based on local neighborhood graph has some advantages.