Remote Sensing Image Registration Method Based On Multi-feature And Local Neighborhood Information

Image registration is the process of matching and overlaying images which are obtained at different times and different viewpoints,from different sensors,or under different conditions.Image registration is a critical technology in image processing.It has been widely applied in many areas,such as computer vision,remote sensing data,medical images field.With the development of the science and technology,the way of people to obtain image becomes diverse.But the images obtained by different sensors exists obvious differences and limitations.So the registration of remote sensing images becomes difficult.According to the different characteristics of images,people have proposed kinds of registration methods.We can divide the registration methods into three categories: the gray level information based methods,the transform domain based methods and the feature based methods.Nowadays,with the deeply research of image registration,there is a growing requirement for the accuracy of image registration.After the research of the existence of image registration techniques and the analysis of the experiments,this paper proposed the following two methods to improve the matching accuracy of remote sensing image registration.A remote sensing image registration algorithm based on anisotropic scale space and multi-feature is proposed.The method is used to solve the problem that the remote sensing images are affected by noise.It uses the anisotropic scale space instead of the traditional Gaussian scale space,not only can remove the noise but also can preserve the edge detail information effectively.And the corner features can be abstracted by using the Harris corner detector based on the second-order moment matrix.The texture features can be abstracted by using the Hessian matrix.The two kinds of features enrich the characteristic information and increase the number of feature points as well as.Finally,the image gray information is divided into two categories according to a clustering method,and the point matching is carried out in different classes to get the best transformation matrix,so the two images can be matched with the optimal transform parameters.The experiments show that the method can increase the number of feature points and improve the matching accuracy of remote sensing images.A new point matching algorithm based on local neighborhood information is proposed.The method is used to solve the problem that the traditional matching method has many wrong correspondences.It combines the feature-based and area-based method.That is,combines the feature information and the local neighborhood gray-level information around the feature points.By extending the neighborhood of the feature points,a similarity measure is used to measure the similarity of the neighborhood image blocks.And the feature points corresponding to the irrelevant neighborhood image blocks are deleted.The remaining features are the correct matches.Then increase the number of correct matches on similar image blocks.The proposed point matching algorithm can not only remove the imprecise features,but also can increase the number of correct matches,which generally improves the accuracy and efficiency of the registration results.