Method Of Automatic Registration And Mosaic For SAR Images

With development and progress of the SAR remote sensing technology, the amount of image data increases dramatically because of the fast and diversity means to obtain SAR image and the low cost to acquire SAR image. For traditional processing of SAR orthoimage products, due to depending on manual interaction, the work is too much and the efficiency is too low, which have greatly limited SAR image application. It has the profound practical significance to automatically produce orthoimage products and many key technologies involved in the procedure need be solved urgently. Among them, there are three core problems. The first one is SAR image automatic registration. For the inherent characteristics of SAR system, there are many speckles and local deformations in SAR images. Therefore, traditional image registration methods have poor ability to obtain a sufficient number of tie points with uniformed spatial distribution and high matching precision, which may affect the accuracy of the subsequent image rectification and registration. How to get the adequate tie points in image and realize SAR image automatic registration, is one of the main problems of this research. The second one is that the brightness is not balanced for a multiple band, multiple polarization aerial or satellite SAR images, and color is inconsistent between polarization synthesis pseudo-color SAR images. It can be only achieved an orthoimage with a low quality and poor visual effect by penetrating image rectification and mosaic directly. How to solve the SAR image brightness and color compensation problem is one of key targets of this paper. The third one is the method of SAR image mosaic, which is based on the seam-line network. Comparing with image serial embedded method for image sequence mosaic, although the image mosaic methods based on the seam-line network are more efficient, but for the images with big or small overlaps, it is hard to ensure satisfactory effect for existing algorithms, how to mosaic SAR images with any of overlap automatically is the third key problem we need study in this paper.The paper aims to solve several key issues for image automatic registration, image color consistency processing and large range image mosaic in the process of making SAR orthoimage. SAR-SIFT registration method is improved for SAR images, which could increase feature spacial distribution and coverage. A new method of global color compensation with local color adjustment is proposed to solve the color consistency among SAR images. The method based on traditional Voronoi diagram, Voronoi diagram with considering overlap and fast scanning line filling with spatial index is proposed to mosaic large range SAR images. A SAR orthoimage producing software is developed to realize SAR image automatic registration and mosaic. The main works of the research are as follows:1) SAR image automatic registration methods; 2) SAR image brightness compensation methods; 3) Methods of the color compensation between full polarization SAR images; 4) Methods of the seam-line network construction and mosaic for SAR images.The main contribution of the thesis and innovation lies in:In SAR image registration, according to the characteristics of SAR image, a SAR image automatic registration algorithm is designed, which is robust to speckles, rotations, and affines. It is called as the uniform SIFT-Like algorithm for SAR image registration, that is USAR-SIFT. On the one hand, the algorithm replaces difference gradient of image local window by ratio gradient to overcome the inherent strong multiplicative noise in SAR image, and uses the method of the scale space feature description to weaken local image deformation caused by the slant distance imaging. On the other hand, it can solve the distribution density and coverage problems of feature points due to using the feature point scale space proportional assignment strategy and local sub-window feature points optimization methods by aid of Voronoi diagram;In SAR image brightness compensation, from the global perspective of image, the method based-on Mask theory is improved for SAR image brightness compensation and a new algorithm is proposed based-on range direction differential brightness compensation techniques in this paper, which is named range differential brightness compensation method. The methods have solved the problem of image brightness unbalance in SAR imaging.In color compensation for full polarization synthesis pseudo-color SAR image, the new algorithm is proposed in this paper, from the perspective of global part of the surveyed area, the hue, brightness, contrast and other statistical information of all images are taken into account to build an adaptive macro surveyed area color compensation model, which integrated Wallis linear transformation and histogram mapping method into one algorithm. The model need not be specified a reference image, which improves its flexibility and versatility. The image fusion method of the multivariate linear regression model for transition and adjustment of the image color and contrast between two overlapping images is employed to realize the automatic processing of the image color balance.In the process of SAR image mosaic, the Douglas-Cooke boundary simplifying method is improved in the paper. Combining with the advantages of the seam-line network constructed by the method based on general Voronoi diagram and the method based on overlapping area Voronoi diagram, a new adaptive image mosaic algorithm is proposed. A new algorithm could integrate two classical image mosaic algorithms smartly and seamlessly to generate seam-line network quickly and bring their advantages together to overcome the deficiency of them respectively. A scan-line fast filling algorithm with spatial index is used to mosaic images. The new algorithm has solved the problems of hole" and fracture in the result mosaic image.