Study On Registration Of Sar Image And Optical Image

In recent years, with the development of space technology and information technology, remote sensing is widely applied in land using, environmental monitoring, urban planning, and weather report est. For the aim of the common use of optical and synthetic aperture radar (SAR) remote sensing, to overcome the disadvantages of two types of remote sensing alone and highlight their advantages together, and to reflect more surface features at same time, we need to registry SAR images and optical images. However, due to difference between SAR image and optical image is large, the registration between them is a difficult staff. This thesis is based on National Natural Science Foundation Project "The study of integrated fusion machine learning theory of SAR and optical remote sensing and parallel processing method" and Basic Operating Expenses of Central Level Nonprofit Research Institutions" SAR and optical remote sensing integrated processing technology based on integrated study", focuses on SAR image and optical image registration algorithm study, and forms a registration model to realize higher precision registration between two types of image. This thesis focused on the following two aspects:(1) Feature-based multi-scale image initial registration model: There are four steps in this model:establishing the pyramids of images, edge feature extraction, feature matching and image transform and resample. The Gauss operator is used as the image scale segmentation nucleus to establish the pyramids of SAR and optical images, which is used to simplify the after processing and improve the precision of registration. Many existing edge detection algorithms are analyzed, multi-scale space edge detection algorithms are proposed in this thesis. Hausdorff distance is used in hybrid multi-scale matching, and appropriate spatial transformation model is established. The moving image would converse and resample at last to get the final result. And the result would be analyzed.(2) The further refined registration based on similarity measures: Due to the position error in feature extraction and feature matching, the total accuracy is not high. We improve the accuracy of the results by local registration based on similarity measures. Summarizing and analyzing the similarity measures, we propose the appropriate similarity measure for SAR image and optical image. The proof experiment is designed in our thesis to prove that. And it is used in further registration method.