| The development of space technology has experienced from aviation toaerospace in the twentieth century, and gradually evolved into the integrated patternof land, sea, space and air, Due to needs of different areas a variety of optical remotesensing satellite have been rapidly developed. It has not only a strong demand forthe development of high resolution satellite, but also a gradually increasedhigh-resolution remote sensing satellite images index requirement. At the same time,along with the rapid development of digital image processing technology, imageregistration has become one of the basic work in the remote sensing imageprocessing technology, designed to provide basic conditions for the follow-up ofRemote Sensing Applications. Image registration techniques are widely recognizedin a variety of research areas and play an important role in high-resolution remotesensing applications.The background of this research is that a high-resolution remote sensingsatellite images adjacent two intervals up to tens of seconds, due to the impact ofsatellite attitude and orbit changes, the two inter-frames corresponding to the imageadjacent pixel will produce more translation, and may even rotate, and the overlapbetween two images are only20%. At present, the satellite there are many sources ofinterference of spaceborne camera optical axis pointing accuracy, the registrationaccuracy of only tens to hundreds of pixel adjacent to the original image interframestar, it is difficult to meet a high-resolution satellite like meta-level imaging needs.In this paper, the precision of the satellite image registration needs and inter-frame tobe the characteristics of image registration, do the following aspects of the work andthe effectiveness of the proposed method through simulation, the specific work areas follows:1) The state of art of image registration techniques comprises domestic andinternational research and existing problems of them have been analyzed. Theresearch background has been described, and based on that, the research ideas,content and arrangement of the chapters are given.2) The theories of remote sensing image registration are introduced. Theymainly include the definition, elements of the major methods and the general steps of image registration and development of space remote sensing.3) The process of image acquisition is introduced, and of which characteristicsof images due to a variety of factors are described. Through experiments andanalysis, comparison of a variety of image enhancement has been completed, anddepending on the research background, the most suitable method for this research ofimage enhancements has been selected.4) The pre-processing method suitable for this paper has been analyzed throughnumerical simulation. With the problem of a little amounts of cloud covered in theimage, the conclusion that only of the feature point extraction, or of the edgeextraction will affect the registration accuracy had made firstly, and then edgeextraction combined with feature point extraction method has been proposed with afirst use of the former. It has been proved that the proposed method is moreappropriate for image registration in the paper. Based on the comparison with thecommonly used edge detection operators, corner detection operator and its improvedoperator, the proposed feature extraction algorithm is consisted of an improvedCanny operator and the Harris operator, and the accuracy of the extraction algorithmhas been analyzed through the simulation.5) With the characteristics of translation and rotation existed in the images, theZernike moment-based image registration method is proposed, whose accuracy hasbeen analyzed through numerical simulation and the image registration parametersare given. The effectiveness of this method has been verified through simulation,and the accuracy of the image registration is up to the pixel level.6) Then the augmented Hu’s invariant moments method has been researched,and based on the stability of the virtual triangle, the augmented Hu’s invariantmoments of every three of those extracted feature points are computed and theimages are matched by correlation coefficients. Finally, numerical simulation resultswith several groups of image registration parameters showed that the proposedmethod can improve effectively the registration accuracy whose index can be ofpixel level. |
