Research On High Spatial Resolution Optical Image And SAR Registration Algorithm

With the rapid development of satellite remote sensing and sensor technology,the comprehensive utilization of multi-source,multi-scale and multi-temporal remote sensing image data has become a current development trend and a hot research issue.Among them,the high precision and rapid registration of multi-mode remote sensing images is one of the core technologies.Optical sensor and Synthetic Aperture Radar(SAR)are two important imaging methods in satellite earth observation system.SAR can break through the limitation of imaging of optical sensor and make up for the discontinuity of earth observation caused by optical sensor due to weather and other special circumstances.Therefore,SAR and optical images can be comprehensively utilized to achieve effective information complementarity between them through image fusion,which has a wide application prospect in special feature information extraction and target recognition.However,the imaging principles of different sensors are also different,and there are great geometric and radiation differences between them.Therefore,how to effectively solve the precise registration of optical image and SAR has become a hot research topic at present,and it is also one of the key technologies in the project I participated in.This paper focuses on the accurate automatic registration algorithm of high spatial resolution optical image and SAR.In a lot of reading at home and abroad under the premise of frontier literature,study the characteristics of SAR image,to find and analysis suitable for optical image and SAR registration algorithm and unsolved problems in the algorithm,using optical remote sensing images,such as GF1,GF2 and Google Earth,and SAR images,such as GF3 and Sentinel 1A,are used to carry out registration experiments for verification.On this basis,an improved algorithm is proposed to achieve the high-precision matching between SAR and optical images.The main research content of this paper focuses on the following two aspects:(1)A registration algorithm based on coupled gradient mutual information and Partial Volume(PV)interpolation is proposed.For optical image and SAR with on time accuracy,low efficiency,large amount of calculation and easy to appear false match and a series of problems,this paper puts forward the algorithm,the main process is: Firstly,the principle of Mutual Information(MI)method is analyzed,and image gradient is introduced into gray MI,improved Gradient Mutual Information method is proposed,which is used as the similarity measure of SAR and optical image registration.Secondly,the improved gradient mutual information is calculated by the improved PV interpolation method,and the cosine function is used as the interpolation kernel function to expand the neighborhood range of the interpolation.Finally,the two improved algorithms are coupled,and Powell optimization algorithm is used to perform local optimization search,so as to realize the registration problem of SAR and optical image.(2)An automatic registration algorithm for step-by-step refinement of special texture images based on coupling marker control watershed and mathematical morphology is proposed.In view of the existing algorithm cannot find the stable matching feature points lead to can't registration,the paper will be water as a special texture,puts forward the algorithm,the concrete process as follows: first of all,according to the characteristics of SAR images,the marker-controlled watershed algorithm was improved so that it could accurately extract the water region in the image,and the binary image containing only water region and non-water region by binary and morphological processing,so as to complete the accurate extraction of the water region in SAR and optical images.Secondly,the centroid of the water region is extracted for coarse registration between images to improve the search efficiency of the subsequent algorithm.Finally,Particle Swarm Optimization(PSO)algorithm is used to obtain the optimal transformation parameters when the similarity measure is maximized,and then the spatial transformation of SAR image is carried out by using this parameters,and the precise registration of SAR and optical image is realized.This algorithm can transform image registration into binary image registration and combine image segmentation and registration.Paper a lot of experimental comparison,the results show that: in this paper,two algorithms are effective and algorithm one inhibits the emergence of local extremum,reduce the probability of false match the image,at the same time of computation is greatly reduced the registration precision is improved,the realization of optical image between the SAR and automatic registration,apply to most of the optical image and SAR registration problems.Algorithm two aims at image registration with water body with higher accuracy and smoother edge,and effectively solves the problem of finding matching points of the same name between SAR and optical high-resolution remote sensing images with large difference in gray scale and resolution.