Automatic Registration Of Optical And SAR Image Based On Multi-features And Multi-measures

With the development of remote sensing, there are many measures to acquire remote sensing data. However, how to effectively integrate multi-sensor, multi-resolution, multi-spectral and multi-temporal remote sensing data, has become an urgent problem to be solved at this stage, and most of all, multi-sensor image registration is the core problem. SAR data has the advantages of all-time and all-weather, it can compensate the disadvantage of optical sensor, which is easily influenced by bad weather. Thus, their fusion is conducive and meaningful. However, they have different imaging mechanism, it makes the optical image and SAR image have different radiometric and geometric properties. And making the Optical-o-SAR image registration became more and more difficult.Nowadays, most of registration methods need control points acquired by manual, or only meet some certain images, can’t meet the requirements of emergency processing. Focused on it, the mainly contribution of this paper is to propose an automatic optical-to-SAR image registration method using multi-features and multi-measures:(1) Image segmentation method using multi-scale level setThe remote sensing image segmentation result largely affects the results of information extraction and image registration. Compare to optical image, SAR image has very strong speckle noises, it makes SAR image segmentation become a hot-topics and difficult problem. In recent years, because level set method has the ability to deal with topological changes and can naturally integrate boundary information and regional information together, it become one of the mainstreams of SAR image segmentation method. And classical level set method did not consider the speckle characteristics of SAR images, moreover, in the classical approach; the level set function is initialized by a SDF, meaning that pixels far from the zero level set (decided by spatial distance) are hard to move across the boundary. This results in that objects far from the zero level set being hard to detect. Based on it, a multi-scale level set approach for SAR image segmentation is proposed in this paper. Because the multi-scale analysis of SAR images preserves their highest resolution features while additionally making use of sets of images at lower resolutions to improve specific functions, the proposed method is useful for removing the influence of speckle whilst at the same time preserving important structural information. The Gamma distribution is one of the most commonly model employed to represent the statistical characteristics of speckle noise in the SAR image and it is introduced to define the energy functional. Moreover, based on the multi-scale level set framework, an improved multilayer approach is introduced for multi-region segmentation. To obtain a fast and more accurate result, a novel OTSU segmentation result is used to represent the initial segmentation curve. The experiments with synthetic and real SAR images demonstrate the effectiveness of the new method.(2) Integration of segmentation and registration method for Optical-to-SAR image registrationMost segmentation-based registration method separated the segmentation and registration into two independent steps. However, almost all of the segment-based registration methods rely on the segmentation algorithm for segmenting the primitives to be matched; and matching results largely dependent on the results of segmentation, poor segmentation result would generate unsatisfactory matching result. Thus, to perform just once segmentation before matching is usually leading to failed matching, repeatedly segmentation and matching is needed. Based on the above, to avoid failed registration caused by poor image segmentation, a simultaneous segmentation and feature-based registration method was proposed. Moreover, we proposed a two-stage registration strategy, first, an iterative level set and SIFT (ILS-SIFT) method was proposed for image registration. Here, a uniform level set segmentation model for optical and SAR images was presented to segment conjugate features. And, SIFT is employed to determine whether the registration is successful. Second, to solve the problem of incomplete segmentation result, we also presents a global constraint-based triangulation optimization registration method, which first filtered through a number of objects using shape similarity, and then use the triangular growing algorithm to match the remaining objects.(3) An iterative registration method using multi-scale line featuresImage registration based on line features is an classical and effective measure, however, the classical line-based registration method has some problems, firstly, because the largely radiometric difference between optical and SAR images, the conjugate feature extraction is difficult, second, even if the conjugate lines can be extracted, position error is still exist. Based on it, this paper proposes a method of fully automatic optical-SAR image registration based on linear features and the integration of Voronoi decomposition and Graph Spectral Theory (VSPM). Compared with traditional optical and SAR images automatic registration methods, the proposed method has three advantages:1) The proposed VSPM can search for all possible matching point pairs from the global perspective, and take the local constraints into account at the same time. Thus, the proposed method can improve the conjugate feature matching accuracy, and solve the matching error caused by many-to-many in line features and positioning errors.2) A multi-level strategy is adopted in the method to reduce the interference of details/speckles and remain main linear structures in both optical and SAR images.3) A new method that combines feature extraction and feature matching into an iterative procedure is proposed, which can overcome the problem of serious dependence on feature extraction result, low reliability and low accuracy existing in traditional methods.(4) A fast coarse registration method based on visual saliency featureVisual attention mechanism studies show that the human visual system will first focus on some of the visual saliency objectives, namely image local features, these local features can first be used as a mark for optical and SAR images registration, such as rivers, lakes and iconic buildings. Based on it, this chapter presents a fast coarse registration method based on visual saliency feature. The main idea is to use notable regional characteristics and structural features to match. For significant regional characteristics, this paper combines visual saliency model and the level set segmentation method for objects extraction, and an improved shape curve method was used for registration. For significant structural features, we use special structure and shape for coarse registration. Meanwhile, in order to make classical Itti model more suitable for water in SAR imagery, an improved TW-Itti model is employed to detect those suspected regions, which take into account texture features generated by the Gray Level Co-occurrence Matrix (GLCM) algorithm. Experiment performance that our method can effectively get interested areas; improve the image registration speed and accuracy.Finally, the author developed an experimental system of optical and SAR image registration, it is based on the SmartSARImage platform, which was developed by the State Key Laboratory of Information and Engineering in Surveying, Mapping and Remote Sensing. And in order to verify the validity and practicability of the registration method in this paper, we focused on three practical cases, which are SAR image geometric correction using satellite image and UAV image, flood detection of Poyang Lake in2009.