Image Registration Method Research For High Resolution Remote Sensing Images Based On Edge Point Features

Remote sensing image registration is the basis and prerequisite for remote sensing applications.The accuracy of registration has a direct impact on the effect of follow-up applications.The automatic processing and improved registration efficiency of registration has a great significance to the analysis of mass remote sensing data,is also the focus of attention.SIFT(Scale Invariant Feature Transform)and SURF(Speed up Robust Features)are typical commonly used methods in the remote sensing image registration.SIFT has scale invariance,high robustness in noise interference and affine transformation,but it is computationally intensive and time-consuming.Moreover,the Euclidean distance is used to produce more false matching point pairs.The SURF algorithm optimizes the registration time and remains the invariant of scale and affine transformation,but the registration accuracy needs to be improved and the matching rate is low.With the improving resolution of remote sensing images,the size and amount of remote sensing images are constantly increasing.Meanwhile,with the development of remote sensing applications,the performance of image registration is also requested by higher and higher performance.The feature registration method overcomes the limitations of grayscale registration method,which is a research focus of remote sensing image registration.Therefore,for high resolution remote sensing images with rich texture information,an automatic image registration method is proposed in this thesis based on edge point features.The method includes five steps:(1)Both the reference image and the image to be registered are performed by one-level Haar wavelet transform to get the low frequency approximate images to match,and then,the registration of the original images is completed according to the matching result of the approximate images.It can effectively reduce the amount of calculation and improve registration speed.(2)Edges in the optical image are extracted by the Canny operator,and edges in the SAR image are extracted by the ROA(ratio of averages)operator.Then the edge line features are transformed into point features.The use of edge point features can get accurate positioning,but also can obtain stable features.(3)In the feature matching session,the main and auxiliary directions of the point features are considered,so that each point feature has multiple directions.It can enhance the robustness of image registration.Then the initial matching points are determined by the ratio— the minimum angle to the minor angle— which is less than a threshold.(4)In the matching point pairs filtering session,in order to improve the registration accuracy,random sample consensus is improved by adding the constraint condition and the good matching point pairs are selected to fit the model parameters.(5)In the affine session,the block thought is used to uniformly choose matching point pairs so that it can be evenly distributed in the image to avoid the local optimal problem in the registration and further improve the image registration accuracy.In order to verify the efficiency of the method,experiments are carried out in several different situations: the same sensor optical image registration,the same sensor SAR image registration,image registration between different bands,image registration with different resolution,and image registration of different satellite sensors.The proposed method is compared with the typical SIFT algorithm and SURF algorithm and registration results are evaluated by using four quantitative evaluation indexes i.e.matching rate,matching efficiency,root mean square error and time consuming.The experimental results show that the method has high registration accuracy,better robustness and effectiveness in processing efficiency.