Research On Algorithms Of SAR And Optical Image Registration

Image registration can be described as a process of aligning two images,namely,the reference and sensed images,which can be obtained by different sensors at different times,or from different viewpoints.This process is a preparation step for pattern recognition,image fusion,change detection,image mosaic,and so on.Synthetic aperture radar(SAR)and optical images are the most typical types among remote sensing images.Synthetic aperture radar is a kind of high-resolution remote sensing radar under all-day and all-weather situations,but it does not contain spectral information.Optical image is rich in spectral information,but it is easily affected by the atmospheric attenuation and weather conditions.Thus the study of the registration between these two images is of great significance.This dissertation embarks from the significance of registration between SAR and optical images,has summarized the current development of image registration at home and abroad.We analyze the existing problems of registration between SAR and optical images,and proposes an improved method.The main work is summarized as the following three aspects:(1)This dissertation focuses on the scale invariant feature transform(SIFT)algorithm,which is based on the current development of registration between SAR and optical images.We analyze the advantages and disadvantages of existing algorithms,including the three steps of image registration:feature detection,description and matching.(2)Registration between SAR and optical images is time-consuming and has poor accuracy when based on the scale-invariant feature transform(SIFT)algorithm.In this letter we propose a novel method to solve this problem.First,we smooth SAR image by using bilateral filter(BF).BF is also good at preserving edges in the image as opposed to Gaussian smoothing,which is used in the original SIFT.Then,keypoints are detected in the Difference-of-Gaussian(DOG)scale space and SIFT descriptors are generated.Next,we adopt the fast library for approximate nearest neighbors(FLANN)algorithm which can search matching points fast in high-dimensional space.Last,progressive sample consensus(PROSAC)algorithm is utilized to exclude false matches.(3)In order to further improve the registration accuracy of SAR and optical images,a progressive algorithm,which is based on the combination of SIFT and Canny edge detection,is proposed in this dissertation.Firstly,we detect candidate keypoints by SIFT algorithm,then find out the edge points of image using Canny edge detection algorithm.It would remove some candidate keypoints by analysis whether the coordinates are equal with the candidate and edge points.Next,SIFT descriptors are generated from those correct keypoints.Last,we adopt the FLANN algorithm for matching.Experimental results and analysis show that our approach is significantly more accurate and much faster than the original SIFT.