Research On Special Texture Aerial Image Matching Method

As one of the key steps of air triangulation,Image measurement is of great significance to the automation of photogrammetry.Image matching technology has been a research hotspot from the initial manual measurement method to the digital image-related technology and now to the feature-based image matching in computer vision.With the rapid development of information technology,image matching technology has also been constantly developed,making the accuracy and efficiency of aerial triangulation greatly improved,and aerial triangulation is more and more widely used.The purpose of image matching is to find the same feature points on two or more images with overlapping regions,so that the number,distribution and precision of matching points can meet the needs of photogrammetry.However,aerial photography usually contains large areas of desert,woodland,and arable land as images contents,and textures appear as poor textures and repetitive textures,these images are called special texture images.Due to the low signal-to-noise ratio of the poor texture image content,the gray-based image matching is susceptible to noise interference,and the matching success rate is low.When using the feature-based image matching method,it is difficult to obtain an effective matching point;the repeated texture image characteristics will lead to descriptors are poorly distinguishable,ambiguous,and multiple matches are formed,gray-based and feature-based image matching methods are often difficult to work with.Aiming at the difficulty of matching the special texture aerial image,this paper proposes a special texture image matching scheme by studying the existing matching algorithm.The main research contents are as follows:(1)Analysis of the characteristics of commonly used image matching algorithms.The applicability and limitations of common matching methods in special texture aerial image matching are studied.Including gray-based image matching algorithm(correlation coefficient method,least squares matching method)and feature-based image matching algorithm(ORB algorithm,SIFT algorithm,MROGH algorithm)(2)Combine the color invariants and the descriptors of MROGH.Taking full advantage of the color information of the image and the spatial relationship of the sampling points of the feature region,the color invariant is introduced into the MROGH descriptor to improve the uniqueness and robustness of the feature point to describe the characteristics of the special texture aerial image.(3)Special texture aerial image matching strategy.Feature matching based on the relationship of the epipolar geometry is used to reduce the matching time by searching for matching points in the same-named region.The matching is improved by the mismatching method based on the graph model and the least squares matching method based on features to improve matching rate and matching accuracy.(4)Comparative experimental study of typical methods.This paper compares the method with SIFT algorithm and MROGH algorithm,and compares the number of correct matching points,distribution,positioning accuracy and matching time.Experiments show that the proposed method can obtain more correct matching points and the matching precision can reach the sub-pixel level when processing special texture aerial image matching.