Research On SAR And Optical Images Registration Based On Structural Feature And Mutual Information

With the development of remote sensing image technology,data obtained from different sensors provides a large amount of available data which can be used for many remote sensing image applications.The data provided by multi-modal images have good complementarity,so they have played a huge role in the fields of remote sensing,aircraft and monitoring.Multi-modal image registration,as a preprocessing procedure for image classification,change detection,and target recognition,has a non-negligible impact.Among them,SAR and optical images registration has been an important and difficult point in multimodal image registration in recent years.In this paper,further research is performed on difficult problems such as noise interference,large intensity difference,and geometric distortion in the registration of SAR and optical images.We combined mutual information with structural feature information of the image and proposed two new registration algorithms which improves the accuracy and robustness of registration.The research contents and innovations of this paper are as follows:1.Due to the different imaging mechanisms of SAR and optical images,there is a large intensity difference between them.In addition,the problem of multiplicative speckle noise inherent in SAR images cannot be ignored,and traditional mutual information only consider the intensity correlation between reference image and the registered image,ignoreing the structural information of them.When it is applied to the registration of SAR and optical images,it cannot solve the existing problems.Therefore,in Chapter 3,based on the combination of intensity balanced and Gaussian-Gamma-shaped bi-windows,a new similarity measure named Gaussian-Gamma balanced mutual information is proposed.This method introduces intensity balanced to preprocess the image and adjust intensity difference between the image.At the same time,it can reduce the noise interference to a certain extent when the structural features of the image are subsequently extracted.Then,the feature information points are extracted to reconstruct the Gaussian-Gamma-shaped bi-windows.Combining with mutual information,its edge strength information and angle information are obtained,therefore a new similarity measure is obtained.This method effectively combines the structural feature information of the image with the intensity information,and makes up the shortcomings of mutual information.Experiments confirm that the method of this chapter used on SAR and optical images registration can obtain higher accuracy and robustness.2.Aiming at the problem that mutual information is easy to fall into local extremes when applied to SAR and optical images registration,and considering the general problems of SAR and optical images registration,chapter 4 proposes the coarse and fine two-phase registration framework combined with wavelet decomposition.It is divided into a coarse registration phase and a fine registration phase to achieve the search of optimal transformation parameters gradually.In the coarse registration stage,the low-frequency image obtained from wavelet decomposition is used to construct a similarity measure using phase consistency and Hausdorff distance to obtain the coarse transformation parameters,and then use parameters as the initial search parameters in the fine registration stage.In the fine registration stage,the original image is divided into blocks and Harris algorithm are used to obtain feature points,then the mutual information of the blocks are combined in a certain way as the similarity measure.In order to further reduce the probability of local extreme,the particle swarm optimization is introduced as search algorithm.It is verified by multiple sets of experiments that the method achieves high accuracy and robustness on SAR and optical images registration.