Font Size: a A A

Research On The Registration Technology Of Optical And SAR Images Based On Point Features

Posted on:2022-11-30Degree:MasterType:Thesis
Country:ChinaCandidate:Y C MiaoFull Text:PDF
GTID:2518306764999529Subject:Computer Software and Application of Computer
Abstract/Summary:PDF Full Text Request
The optical image has the characteristics of high image resolution,good imaging effect,close to human vision,and easy to interpret image;Synthetic aperture radar(SAR)has the advantages of all-weather,all-day,and strong penetration;SAR images can achieve a good ability of information complementation and have important application value in national defense or production and life,and the premise of using the two information together is image registration technology.However,the registration of optical and SAR images is restricted by many conditions,such as the differences in grayscale and geometric characteristics between the two images,and the speckle noise of SAR images.For this problem,to improve the registration accuracy and robustness of the two images,this paper makes corresponding improvements on the registration algorithm based on point features,and studies an effective optical and SAR image registration algorithm:(1)To solve the problem of poor real-time registration between optical and SAR images,an improved registration method of optical and SAR heterogeneous remote sensing images based on optical-SAR scale invariant feature transformation is proposed.First,a nonlinear diffusion filter(NDF)is introduced to construct the scale space of the two images.For the optical image and the SAR image,the multi-scale Sobel operator and the multi-scale ratio of exponentially weighted averages(ROEWA)operator are used to calculate the uniform gradient information of the two images.Secondly,the image blocking strategy is adopted to block the scale space after removing the first layer of scale space.Harris feature points are extracted according to the relevant gradient information to obtain stable and uniform point features.Next,using the results of the gradient calculation,a descriptor is constructed with the help of the gradient location and orientation histogram(GLOH),and the constructed descriptor is normalized to overcome the nonlinear radiation between images.Finally,the Euclidean distance(ED)is used for feature matching,and the Fast Sample Consensus(FSC)algorithm is used to eliminate false matches.The algorithm is robust to visible light and SAR image registration,which improves the matching accuracy and accuracy of the algorithm.(2)For feature-based registration,geometric constraints are often used to eliminate mismatches to estimate that image transformation requires a predefined transformation model,the problem of transforming feature matching into spatial clustering with outliers is studied.The method includes four main parts: initial matching feature point set generation,weighted distance definition and calculation,adaptive parameter estimation,and mismatch point elimination.First,the feature-based registration algorithm is used to get the initial matching feature point set;second,the Euclidean distance is calculated for the points in the initial point set to determine the weighted distance and nearest neighbor of each point;then,the adaptive parameter estimation is used to calculate search for the time complexity required for the maximum and minimum distances,cluster the points and iterate until the time complexity is within the proposed range;finally,use density-based spatial clustering of applications with noise(DBSCAN)determines the core points and abnormal points of the clustering sample,removes the abnormal points,and registers the cluster points with similar motion patterns to complete the matching.The false matching elimination method increases the number of correct matching point pairs and improves the accuracy of the false matching elimination.In summary,the main research work of this paper is to study the heterologous image registration algorithm suitable for optical and SAR images and provide research ideas in feature extraction,feature description,and feature matching.The work has a certain reference value.
Keywords/Search Tags:Optical and SAR image registration, Nonlinear diffusion filter, Block strategy, Spatial clustering
PDF Full Text Request
Related items