Research On Super Resolution Reconstruction Algorithm Of Satellite Video Based On Global Motion Estimation

As a new product in the field of remote sensing,satellite video has application prospect widely in many fields.Although the resolution of the satellite video can satisfy the requirements of some practical applications,but with the more extensive and in-depth satellite to earth observation applications,people needs satellite video that have a higher resolution.Super resolution reconstruction technique is a process that uses a set of low resolution image sequences containing non redundant complementary information to generate a high resolution image.The technology can make up for the deficiencies of improving the spatial resolution by improving the hardware,and save more cost.The thesis analyzes the Characteristics of satellite video,research on the key steps of the super resolution reconstruction technology.and mainly focuses on following aspects:1.Applicability analysis of global motion estimation.In this paper,we first study the key problem of motion estimation in super resolution reconstruction.Compared to ordinary video,satellite video has large data volume,The static region contains a small part of the moving object,Aiming at these characteristics,this paper chooses the global motion estimation method to solve the video frame displacement.Because of the moving target,the accuracy of global motion estimation is affected by the satellite video image.Therefore,this paper discusses the applicability of the global motion estimation method to satellite video super resolution reconstruction.The Vandewalle method and the Pyramid LK(Lucas-Kanade)optical flow method are used to estimate the global motion estimation and Super resolution reconstruction algorithm for satellite video are studied based on global motion estimation.2.In the reconstruction process,the phenomena of "black mesh" and "jagged edges"are caused due to the low resolution pixel to the high resolution grid map is not accurate.And the reconstruction of the algorithm is an integer multiple,But the best multiple is a decimal number.To solve these problems,this paper proposes F-MSA(SA Fractional-median)algorithm based on SA(Shift and Add).The F-MSA uses low and high resolution images to search the appropriate low resolution pixels more robustly and accurately.The method can provide high resolution images with arbitrary multiple.Experiments results show that the F-MSA algorithm has good reconstruction effect and low computational complexity,saving time.3.In order to ensure the accuracy of the reconstruction,the static region detail information is increased in the use of the global motion estimation for satellite video image super-resolution reconstruction.Kriging interpolation algorithm is first analyzed to find the correlation between the pixels in a certain neighborhood of high resolution pixels,and then set up the Kriging linear equations to calculate the optimal unbiased estimates.Experiments show that this method can provide more details in the static area,and the edge of the object can be more clear when the correctness of the moving target(vehicle)is reconstructed.The method also can satisfy the need of the super resolution reconstruction with arbitrary numbers...