Research On Remote Sensing Image Matching Method Based On Point And Line Features

Image matching is the core problem of remote sensing image registration.Its purpose is to find the corresponding matching primitives with the same name,and realize the one-to-one mapping of two or more images in terms of features,textures,and gray levels,so that the images can be spatially mapped.When there are too many similar regions in the image,the degree of similarity between feature descriptors is relatively high.Traditional matching methods such as SIFT can easily cause mismatches to be too concentrated,and it is difficult to obtain highprecision points with the same name.At the same time,for remote sensing images acquired by multiple sensors,in areas where texture information is relatively lacking,point features are difficult to extract and unstable,and straight line feature extraction is relatively easy,but the uncertainty of straight-line feature attributes directly affects the accuracy of line matching.Based on this,this paper mainly carries out research from the following two aspects :(1)How to improve the initial matching accuracy of remote sensing images with small feature differences in similar regions.For images with many similar regions,based on the problem that feature point matching on the entire spatial scale is greatly affected by noise and poor initial matching accuracy,in this paper,an effective optimization plan has been proposed in the feature point matching stage,and introduces regional constraints to limit the feature point matching area,So that the matching area changes from the entire image to the corresponding block image.Experiments show that the method in this paper not only improves the initial matching accuracy,but also makes the algorithm more stable to viewpoint,angle,illumination,blur,etc.;(2)How to improve the stability and accuracy of straight line matching.Aiming at the difficulty of obtaining accurate and stable lines with the same name based on the similarity of line descriptors,a line matching algorithm has been proposed in this paper,which calculates the homography matrix through the local points of the same name obtained by GMS,and use matrix constraints and geometric relationships between lines to quickly obtain lines with the same name.Experimental results show that the proposed method can greatly improve the accuracy of line feature matching.Finally,the algorithm of remote sensing image Mosaic is realized by using the feature point-and-line matching method in this paper,and a Web-based image Mosaic system is developed to realize the fast Mosaic of multiple remote sensing images.