Research On The Registration Technology Of Optical And SAR Remote Sensing Images

Optical image has the advantages of high resolution,intuitive imaging and easy interpretation and so on.Synthetic Aperture Radar(SAR)has strong penetrating power and can work all-day and multi-weather conditions.Combining the information of two images can give full play to their respective advantages and obtain the feature of the target scene to the maximum extent,which has a wide range of applications in both military and civil fields.The premise of these applications is stable and reliable highprecision image registration technology,so it is of great significance to study the image registration technology.This dissertation focuses on the registration technology of optical and SAR images,aiming at improving the accuracy efficiency and robustness of the algorithm.For the speckle noise of SAR images and the difference of radiation and geometric characteristics between optical and SAR images,we studied three registration algorithm of optical and SAR images,the specific research contents are as follows:(1)Aiming at the problem that SIFT algorithm often fails in the processing of optical and SAR image registration,an optical and SAR image registration method based on nonlinear diffusion filter and improved SIFT is proposed.The scale space of the image is constructed by nonlinear diffusion filter.The edge and detail features of the image can be better preserved while the speckle noise is suppressed.Multi-scale Sobel operator and multi-scale ROEWA operator are used to calculate the consistency gradient information of optical and SAR images,repectively.Based on the consistency gradient information,two kinds of Harris scale spaces are constructed to extract feature points with high repeatability and robustness to noise.Finally,descriptors are constructed based on the gradient location and orientation histogram.The position,scale and orientation of the feature are used to constrain the feature,and the feature matching is completed.Experimental results show that the proposed algorithm is superior to the improved SIFT algorithm in terms of successful matching rate and matching accuracy,and retains the rotation and scale invariance of SIFT algorithm.(2)In order to solve the problem that the line of sight pointing error exists in the process of image acquisition by airborne optoelectronic platform,which leads to a large nonlinear radiation difference between optical and SAR images,a fast optical and SAR images registration framework suitable for airborne optoelectronic platform is designed.Geographic information is used for coarse registration to determine the overlapping area between the two images,and the rotation difference between the two images is roughly eliminated.The scale difference between the two images is eliminated by resampling according to the resolution information of the sensor.Then the overlapping images are used for fine registration,and a certain number of corner points and grid points are extracted from the reference image by block-Harris.The gradient information of the image is used to construct the 3D dense feature expression,and then the 3D structure information descriptor is constructed for each feature point.In order to increase the matching speed,the descriptor is transferred to the frequency domain,and the sum of square differences is used as the similarity measure criterion,and the similarity matching is implemented by template matching.Experimental results show that the method has good real-time performance and high matching accuracy,which is suitable for engineering application.(3)Aiming at the problem that the speckle noise of SAR image and the nonlinear radiation difference between optical and SAR images lead to the poor robustness of image registration algorithm,a robust optical and SAR remote sensing image registration algorithm(ROS-PC)is proposed,which includes a novel feature detector(UMPC-Harris)and a novel local feature descriptor(HOSMI).The PC multi-moment maps are constructed by using PC moment information in UMPC-Harris,and stable edge points and corner points are selected as feature points by voting strategy on the multi-moment maps.HOMSI uses multi-scale MIM and PC oriention instead of image gradient amplitude and oriention feature to construct local feature descriptor,so as to overcome the problem that gradient information is sensitive to gray difference between images.Experimental results show that the proposed ROS-PC algorithm can effectively suppress the radiation difference between optical and SAR images,and improve the robustness of the matching algorithm.The methods proposed in this paper provides a new research idea in the aspects of feature extraction,feature description and feature matching,and provides a certain theoretical basis and design reference for the realization of high precision and strong robust optical and SAR image registration algorithm.