Research On The Application Of Graph Spectral Theory In Image Registration

Vision is the most useful perception of humam.Image is an important way for human beings to cognize the object in the world and conveys information. With the rapid development of information, research on image processing is pervasive. The research field of image processing is very wide, among which image registration is an important branch.Image registration means to find an optimal geometry transformation that can transform the images of the same scene but in different coordinate systems, for the reason that they are sampled from different sensors, taken at different times or form different viewpoints, into the same coordinate system. It applied in the aspect of medical image analysis,remote sensing image of processing, computing vision and so on. In the last few decades, thousands of paper were published on the topic of image registration, but there are still many problems needed to be solved and much works needed to be done.The current image registration techniques can be divided into area-information based registration and feature-based registration. The former is usually used in specific images, which is more precision and cost more time than the latter. It's the most important that to find the better similarity to catch matching in two or more images for the latter.Graph, which is an important and effective way to describe the feature information of structure, is a useful tool of describing dimensionality date. Graph spectral theory is one of branch of graph, which is made of matrix theory and combination theory to research matrix spectral such as proximity matrix, weighting proximity matrix, Laplace matrix and weighting Laplace matrix.Based on the theory of graph spectrum as the main theoretical tool, it is researched on the algorithms of image registration. The main research works and achievements are outlined as follow: First of all, it is studied that the concepts, class and steps of image registration. Then it is analyzed the choosing of geometric model transform, evaluation criterion and algorithms of feature-based registration. It is also studied the graph spectral theory which be applied in the aspect of matching, segmentation and classification.Secondly, an algorithm of image registration based on graph spectral theory. Firstly, a graph matrices was constructed according to the feature points of two related images respectively. Then, with SVD a matching matrix denoting the matching degree among feature points was constructed by using the results of the decomposition. Finally, the strategy used the homogeneous matrix, coordinate transfer and the image interpolation to register. Experimental results indicate its feasibility and higher registration precision.Finally, an algorithm of image registration based on graph spectral theory and the texture image analysis. Firstly, the SIFT is introduced into the field of image registration for the feature points of two related images. Secondly, with the algorithm of image registration based on graph spectral theory, a graph matrices was constructed according to windows texture analysis.Then,with SVD a matching matrix denoting the matching degree among feature points was constructed by using the results of the decomposition. Finally, the strategy used the homogeneous matrix, coordinate transfer and the image interpolation to register. Experimental results indicate its higher registration precision.