Research On Registration Algorithm And Improvement Of Remote Sensing Images

Image registration is a hot research direction in current image processing,and it has been widely used in fields such as remote sensing,medicine,and computer vision.As a front-end technology for image stitching and fusion,its accuracy directly affects the subsequent processing accuracy in related fields.In the field of remote sensing processing,due to the problems caused by the multi-source,multi-temporal,multi-view,and ultra-high resolution of remote sensing images,higher performance is required for the registration method.This paper mainly studies the registration algorithm of remote sensing images as follows:(1)The systematic research of this paper analyzes the current development status of image registration,introduces the main process of image pre-processing,image enhancement and geometric correction,and analyzes its basic process with the feature-based matching method as an example.The related classification of registration algorithm is introduced.(2)The four classic algorithms in the three major classifications of image registration methods and the algorithms based on mutual information,FFT,Harris and SIFT are studied in depth.Through experiments,we can verify their adaptability under various conditions,and summarize and analyze their advantages and disadvantages.Through experiments,it is found that the classic algorithm will lead to the problem of insufficient number of feature points when the matching target of the remote sensing image is small,which will lead to the problem that the number of matching points is insufficient or not,which leads to the problem of unmatchable pairs.This also points the way for the next step.(3)In view of the problems caused by the above situation,it is obtained through experiments that a four-direction Sobel operator is introduced during remote image registration.Before processing,the Sobel operator is used to sharpen the reference image and the image to be registered to highlight Its important edge details,so as to increase the number of feature points,and then use SURF method for image registration,and finally use RANSAC method to filter out the mismatched points.Experiments prove that this algorithm can extract more feature points with more uniform distribution when comparing the traditional algorithm and other algorithms combined with edge extraction to deal with the above situation,thereby generating more correct registration points.Compared with the SURF algorithm,the number of feature points is increased by an average of about 5 times,and the number of matching logarithms is increased by an average of about 3 times,which effectively avoids registration failures caused by too few feature points or uneven distribution,and then improves Matching accuracy can also meet real-time requirements.Provided a guarantee for the subsequent image processing work.In this paper,the improvement of the classic SURF algorithm effectively improves the problem of insufficient matching points caused by image size differences,improves the performance of the algorithm,and provides a guarantee for subsequent image processing.