Research On Algorithms Of SAR Image And Optical Image Fusion

The theory of multisensor data fusion (MDF) was unnoticeable at beginning of 1970s. Explanation and evaluation of amount of information are demanded, along with advancement of technology and diversification of channels to obtain information. In some special situations data generated from single sensor is unsatisfactory and people realize the importance of multisensor data fusion more and more. MDF acts as an important position in the local war last decade. From then on it is notified exoterically by people and has become a hot research area. MDF is a system which performs multilayer and multilevel processing using lots of data from multisensor and gives information better than those from any single sensor. The result of data fusion is a single image which is more suitable for human and machine analysis and perception. It can alleviate commander's workload and improve the confidence and credibility of the system in great measure.In this paper author enunciates the architecture and implementation of MDF, then three algorithms for extraction of special targets feature in SAR images are presented: First one is for extraction of fractal feature based on fractal dimension brown-moving field (fDBMF). Second is for extraction of circular targets based on bi-parameter constant false alarm rate (BP-CFAR) and image morphology. Last one is for extraction of linear target based on directional and local neighborhood property. After extracting targets feature we can select controlling points in the region of interest (ROI) from complex background in SAR image conveniently, then use these points to match SAR and optical images. In the next step, author analyses the algorithms of image registration including polynomial registration and resample. The result shows that the registration precision error is less than 1 pixel.For interesting targets detection application, based on above research author presents and develops three key algorithms to perform data fusion about SAR and optical images:The first algorithm is choosing the bigger local model coefficient between SAR image and optical image based on mltiscale decomposition to perform invert multiscale transform. At beginning, the multiresolution analysis and achievement are discussed. Then according to the different frequency property and statistical property of interesting targets, author selects bigger local model maximum between SAR subimage and optical subimage via multiscale wavelet transform representation. In the end, by using the selected local model maximum, author finishes data fusion through invert wavelet transform. At the same time author compares fusion result with original SAR and optical images to evaluate this algorithm. The second algorithm is how to get mass function in the Dempster-Shafer evidence reasoning. We know that D-S evidence theory is very useful in many application, but there is not a successful method could get mass function simply and easily in the literature. In this part author first discusses the basic theory and correlative character of evidence theory, as well as orthogonal fusion. Secondly author put emphasis on analysis of the process of mass function achieving from SAR image and optical image information. Then author presents a method which can generate mass function based on interesting targets feature. Finally, fusion of different information from SAR and optical images according to the orthogonal fusion in D-S evidence theory is taken. The results show that the fusion image could detect targets of interesting more precise than anyone source image. And the performance was presented in the end.The third algorithm is image classification via rough set reasoning. Rough set theory is a new theory and it is developed in recent years. At beginning the basic processing methods of rough set theory in indefinite reasoning is introduced generally: including the relationship of object indiscernibility, the representation in rough set for uncertain knowledge, rough set decision and the reduction of decision table. Because the rou...