| With the booming development of remote sensing technology in China,various remote sensing image data with different sources have been proposed recently and the complementary information of multi-source remote sensing images has also become a hot spot in recent years.Multi-source collaborative methods can be conducive to obtain more spatial information of the image.In addition,the complementarity between the sensors can effectively overcome the limitations of a single sensor in a complex environment.Multi-source image registration is a pre-processing process in the field of remote sensing image analysis which is a process of realizing temporal and spatial registration of data acquired by different sensors at different viewing angles and at different times.The registration of optical image and synthetic aperture radar(SAR)image is the most crucial and difficult issue of multi-source image registration.This thesis focuses on the registration of optical image and SAR image.The optical image has high resolution,but it is extremely susceptible to the impact of harsh environments which will cause the loss of spectral details.SAR image have strong penetrating power to vegetation,clouds and snow,but they will be disturbed by coherent speckle noise.In addition,the imaging mechanism of optical image and SAR image is different,and the gray information is vastly different,which may lead to the failure of the performance of the two image registrations.For the above problems,we make a study on optical image and SAR image registration algorithms.The main contents can be summarized as following:(1)In this thesis,the essential difference between optical image and SAR image from the imaging theory,and the registration of optical image and SAR image in domestic and international references have been analyzed first.We also point out deficiency of the traditional SIFT registration algorithm and the limitations of the state-of-the-art.Based on the above analysis,the follow-up research is to propose a method with higher registration accuracy for the defects of traditional algorithm.(2)In order to solve the two vital problems,one is that the gray level of optical image and SAR image is quite different,the other is that the SAR image is easy to be affected by speckle noise,and its feature extraction is difficult,which will lead to the failure of traditional registration algorithm,this thesis proposes an improved SIFT-Like image registration algorithm.Different gradient operators are used to calculate the gradient of optical image and SAR image respectively,and the corresponding scale space is constructed to limit the main direction angle and then the feature descriptor is constructed.The simulation results show that the algorithm can solve the problem of low matching rate and poor robustness of feature points in the process of feature extraction of optical image and SAR image.(3)To address the issues of mismatches and low registration accuracy in optical image and SAR image registration process,this thesis proposes a method of optical image and SAR image registration based on the geometric constraints.Through the spatial geometry characteristics between similar feature points,the feature descriptors are locally optimized.The experimental results show that the proposed algorithm has better matching accuracy than other algorithms.To conclude,several sets of optical image and SAR image are used as experimental data to compare the validity of the proposed algorithm with the traditional SIFT algorithm and other existing algorithms.The experimental results show that the improved algorithm in this thesis has a significant improvement in the correct matching rate and registration accuracy compared with other algorithms.Meanwhile,the research results of this thesis can provide some valuable results for the following research of image fusion and target recognition. |