Research On Methods Of Enhancement And Registration Of SAR Image

Synthetic Aperture Radar(SAR)has been widely used and developed rapidly in military reconnaissance,disaster monitoring and geological survey since the birth of it.SAR can reflects the target structure information,work all-day and all-weather condition compared with the optical sensors.However,SAR images are hard to understand according to the coherent imaging mechanism,and not consistent with the experiential knowledge in optical image.Therefore SAR image pseudo-color enhancement and image registration between SAR image and optical image,have been researched in the thesis.The main works of this thesis are listed as follows:(1)The pseudo-color enhancement and registration of SAR image are firstly studied.The difference between the traditional correlation algorithm and the algorithm used in the thesis has been analyzed,and the contrast experiment was done.Based on the analysis of the algorithms of Polarization matrix decomposition,Freeman decomposition was adopted to realize pseudo-color enhancement of SAR image.Not only the needs of stability and real-time,but also the character of the large-scene data,SIFT was chosen in the thesis for image registration.(2)When the image is enhanced based on the Freeman decomposition,by analyzing the power of double-bounce and surface scatterings,some pixels of the image are found negative.By analyzing the formula of Freeman decomposition and the background of the algorithm,four possible cases are found.By employing the de-orientation,separating Helix component,modifying the volume model and employing span constrain.The conditions of negative power are eliminated and the image is enhanced preferably according to the experiments.(3)Due to the difference of image-forming between SAR and optical images,the mismatching of feature points occurs during the image registration process of SIFT algorithm.Aiming at the problem,the spatial consistent matching is employed to filtrate pairs of the feature points.After the operation,the matrix of transform is obtained by using the random sample consensus(RANSAC)algorithm and the result of registration is achieved.Due to the rich character of corners and line intersections in the image,the Harris-SCM algorithm and Line Intersections-SCM are adopted to realize the image registration.Due to the failure of exhaustive algorithm/NCC algorithm /SVD algorithm for corners and line intersections match,the descriptor algorithm used in SIFT was employed to match the points.