Research On Algorithms Of Point Pattern Matching Based On Attributed Graph Theory

Image matching technology is a research focus in pattern recognition and computer vision, which is the technology promise of image fusion, image mosaic, et al. The research in this aspect has obtained abundant achievement in recent several decades. However, due to the impacts of human factor, device and external scene etc., the images to be matched maybe exist some problems, such as geometric transformation, illumination and shield, which made the research task in this area still very arduous. Feature-based image matching technology has good reliability and robustness, which has become the mainstream of image matching method in recent years. Moreover, as the important features of image, point pattern has an important significance and value. The study of point pattern matching has become a hot topic of current research.As a tool for describing the data, graph can retain the relationship between the area and structure and show the structural features effectively. Currently, using graph model to deal with the problem of point pattern matching is the favor of many researchers.On the basis of the traditional spectral graph theory applied to the point pattern matching, this thesis draws on the idea of local descriptors of the image using the statistics of the gradient or brightness to describe the local area of the image. Starting from the point of structure-property relationship, a point pattern matching algorithm based on spectral graph theroy is proposed. A structural spectral descriptors is obtained by utilizing the statistics of graph spectra and spectral gap to show the feature of the points. Under rotation, translation, the proposed descriptors can be invariant with good discriminant ability. Meanwhile, an objective function is defined by combining geometric consistency represented by neighborhood relationship, and then the matching problem is formulated as an optimization problem with one-to-one correspondence constraints. Finally, the solution of the defined objective function is given by using probabilistic relaxation.In order to verify the effectiveness of the algorithm, the experiments are carried out from two aspects of the synthetic data and real-world images. The experiments of the simulated data are to quantify noise of the position and outliers on the performance of the algorithm. Then detailed experiments on real images are also given to verify the correctness and effectiveness of the algorithm in image matching applications. The results of the experiments applied to both synthetic data and real-world images validate that the algorithm has better robustness to outliers and noise.