Study Of SAR And Optical Images Registration Based On SIFT And Mutual Information

With the development of sensor technology,the types of remote sensing images are becoming more and more diverse.Since different types of images contain different information,how to use the complementary information in multi-sensor images for practical observation in earth observation has become a hot topic in recent years.As an important image analysis of remote sensing preprocessing step,the accuracy of multi-sensor image registration has a high impact on remote sensing image processing tasks such as image fusion and change detection.Therefore,multi-sensor image registration problems have also attracted the attention of scholars.The registration of SAR images and optical images is the crucial and difficult issue in the study of multi-sensor image registration.In this thesis,we focus on the SAR images and optical images and study the image registration method based on point features and the image registration method based on mutual information.In order to achieve accurate registration of SAR and optical images,we fully exploit the useful information of SAR images and optical images,and simultaneously consider the effects of multiplicative speckle noise in SAR images.The main contents and the acquired research results in this dissertation can be summarized as the following two parts:1.Due to the different imaging mechanisms of SAR images and optical images,the intensity and texture between the two images may be different.Traditional image registration method based on point features cannot achieve registration of SAR and optical images due to the unreliability of feature descriptors.In the chapter 3 of this thesis,we propose a registration method for SAR and optical images based on ROEWA-SIFT.This method uses improved ratio of exponentially weighted average(ROEWA)to calculate edge information and direction information of an image and determine the orientation of key points based on the information as well as the feature descriptors.To improve the uniqueness of descriptors,a set of 256-dimensional feature descriptors is constructed.Besides,bi-directional matching is used to determine the matching relationship between the feature points.And the distance relationship between matching points is utilized to further remove false matching points.At last,according to the matching feature points,calculating the transformation matrix and the final registration result is obtained.The experimental results show that the proposed method can be used for the registration of SAR and optical images.2.Considering the differences in intensity between SAR and optical images and the effects of multiplicative speckle noise in SAR images,the method based on the conventional mutual information is instability due to neglecting the structural information.In the chapter 4,a novel similarity measure termed as structural conditional mutual information(SCMI)is proposed.Structural information which is edge strength information and orientation information can be obtained from phase congruency at first.Meanwhile,the non-local means is utilized to reduce the influence of noise while calculating the phase congruency information.Conditional mutual information is calculated by using a 3-D joint histogram which is combined with the intensity and structural information of image.A weight coefficient function is calculated by using phase congruency edge strength and orientation information,and multiply this function with the conditional mutual information to construct the structural conditional mutual information.Experimental results illustrate that the proposed method achieves higher accuracy and robustness in SAR and optical images registration.