| Remote sensing image registration is a key technology in remote sensing matching.It refers to the process of calculating the corresponding geometric relationship between two or more remote sensing image groups with a certain overlapping ratio.Generally,the quality of remote sensing image acquisition is affected by factors such as the satellite’s in-orbit motion state and shooting weather.How to effectively using the existing image resources and accurately establishing the matching relationship between the large-scale correlated images is of vital importance to the accuracy of surveying and mapping and the construction accuracy of 3D models of ground objects.Different from the traditional image matching method,a single remote sensing image has the characteristics of wide map,multiple repeated textures,large amount of data,and large relative tilt angle,and it faces many difficulties in the actual matching process.Scholars have tried a variety of algorithms for these problems.This article has made two improvements and researches on the basis of the predecessors.One is to address the problem of multiple repeated textures in remote sensing images,and an improved method for removing feature points from mismatching is proposed;the other is to carry out several methods on the basic frequency domain matching algorithm.To improve the accuracy of remote sensing image matching.The main content of the article is as follows:I.The article proposes a texture mismatch removal algorithm based on regional constraints.The feature descriptors of weak texture or similar texture regions in remote sensing images are similar,so it is difficult to remove the mismatching of feature points in the same region.Aiming at this problem,this method introduces geometric region constraints,constrains the position of feature points at similar textures,and removes mismatches between weak textures and similar texture regions.In order to improve the matching efficiency,the article proposes a strategy to improve the search efficiency.This strategy can quickly search for similar wrong feature matching points,and improve the efficiency of large-scale remote sensing image mismatch removal.II.The article proposes a method for solving the information of histogram of directional gradient based on frequency domain.This method combines the image single-point neighborhood direction gradient histogram information and the frequency-domain cross-power spectrum initial matching information of the single-point area map to improve the regional matching accuracy of remote sensing images.Experimental verification shows that this method can improve the accuracy of sub-pixel single-point image sparse matching.III.The article proposes a semi-dense image matching algorithm based on frequency domain correlation.Aerial triangulation uses the sparse matching results of remote sensing images as data input.If the number of matching points is too sparse,it will be difficult to support the accurate calculation of aerial triangulation.To solve this problem,this article uses sparse matching information as the initial discrete point set,uses Delaunay triangle to generate a grid,maps the sparse matching point set into a dense matching network,and finally obtains a semi-dense image matching result.This method is more flexible than traditional methods,and provides a new idea for remote sensing matching from sparse and fast to dense.In summary,according to the characteristics of remote sensing images,this article adds reasonable constraints to the image matching process by introducing regional constraints,directional gradients,and frequency domain related information,and designs regionalized frequency domain matching algorithms and feature point mismatch elimination.algorithm.Experimental results prove that the algorithm proposed in this article can overcome the impact of weak texture and similar texture on matching accuracy,obtain sub-pixel image sparse matching results,and can flexibly obtain semi-dense matching results through sparse matching point pairs. |
PDF Full Text Request