Research On Key Technology Of Visible And SAR Remote Sensing Image Registration

The acquisition methods of remote sensing images mainly include photoelectric imaging and radar imaging.The visible and SAR images obtained by the two patterns have various characteristics respectively.For instance,visible images have the advantages of visual,easy understanding and rich in information,however,they are easily influenced by illumination,clouds,and seasons.Meanwhile,SAR images have the advantages of all-day,multi-weather,and not being affected by illumination,but also have the disadvantages of not intuitive,difficult to understand,and have strong scattering characteristics.Therefore,the visible and SAR image fusion can combine their advantages and improve the efficiency of target reconnaissance and enrich the useful information.As a key technology of image fusion,visible and SAR image registration is crucially important.This dissertation focuses on the nuts and bolts of visible and SAR image registration technique,and it aims at improving the accuracy and robustness of visible and SAR image registration.In this dissertation,the geometric and radiometric differences between visible and SAR images are discussed,while many visible and SAR image registration algorithms are studied in detail.The main research contents of this dissertation are summarized as follows:1.Considering that the classical SIFT algorithm is not ideal for visible and SAR image registration,an improved algorithm based on SIFT is proposed.Firstly,in view of the different noise types between visible and SAR images,Sobel and ROEWA operators are used not only to extract consistent gradients but also to construct the scale space for visible and SAR images.Then,the significant key points are extracted by combining the Harris scale space and the threshold constraint in the second order matrix.Meanwhile,the histogram of main orientation difference and the gradient orientation information are used to construct a robust descriptor,which can improve the efficiency of descriptor construction and the consistency of descriptors between the correspondences.Experimental results show that the proposed algorithm can effectively register visible and SAR images,and the registration accuracy of the proposed algorithm is improved compared with OS-SIFT.2.Affected by the radiometric characteristics of ground objects,a large number of interference points are easy to be extracted from the regions where the ground objects with strong scattering phenomenon,such as metals or buildings,are located.This seriously affects the accuracy and robustness of visible and SAR image registration algorithms.To solve the problem,this dissertation proposed an algorithm for visible and SAR image registration based on the complexity analysis and binary descriptor.Firstly,in the edge images obtained from the maximum moment of the phase congruency,multi-orientation sliding window and morphological operation are used to automatically distinguish the dense regions with strong scattering in visible and SAR images,and these regions are masked to avoid the extractions of the interference points.Then,a novel binary local self-similarity descriptor is proposed,which constructs the self-similarity of multi-orientation among the bins in the support regions,and converts the self-similarity into binary modality.The descriptor can solve the problems of high computational cost of the traditional LSS descriptor and sensitivity to nonlinear differences between visible and SAR images.Experiments on several groups of visible and SAR images with significant nonlinear radiometric differences show that the proposed algorithm is insensitive to strong scattering phenomenon,and can improve the registration accuracy and robustness effectively.3.Compared with the algorithms based on spatial domain,the algorithms based on the phase congruency are more robust to nonlinear radiometric differences between visible and SAR images.Therefore,the algorithms based on the phase congruency are more suitable for visible and SAR image registration.However,most of the existing algorithms based on the phase congruency can only effective for visible and SAR images with small scale and rotation differences.So,this dissertation designs a registration framework based on the phase congruency for visible and SAR image registration,and a multi-scale feature detector(PC-Harris)and a descriptor with rotation invariance(PCLG)are proposed.The framework can effectively register visible and SAR images with large scale and rotation differences.In order to overcome the influence of nonlinear radiometric characteristics in SAR images,the multi-scale space is obtained by convolving the maximum moment of the phase congruency with the log-Gabor filter,and Harris operator is combined to extract repeatable correspondences from visible and SAR images.In the descriptor construction process,the multi-scale and multi-orientation characteristics of the log-Gabor filter are used to represent the support regions where the key points are located,and the histograms are constructed based on the orientation information and the amplitude responses of each pixel in the support regions.Under the polar coordinate system,the descriptors with scale and rotation characteristics are generated based on the unique descriptor partition form.Experiments on several visible and SAR images indicate that the proposed framework can deal with the large scale and rotation differences between images.In sum,this dissertation mainly focused on the key technologies of key point extraction and descriptor construction in the visible and SAR image registration algorithm.The algorithms designed in this dissertation provides reference significance for the study of multi-modal remote sensing image registration technology.