The Research Of Image Registration Method Based On Clonal Selection Algorithm

With the development of the science and technology, the ability of human to obtain image improved quickly. Sensors of different physical characteristics generate more different remote sensing images, and the images generated by different sensors exists obvious differences and limitations. Only according to the analysis of one kind image data is difficult to meet the actual demands. So we need to combine the remote sensing image acquired from different approaches by the image fusion technology so as to have a comprehensive and accurate understanding and awareness. One of the key technologies of image fusion is image registration. According to the different characteristics of images, people have proposed kinds of registration methods. Such as the feature-based image registration algorithms, gray-based and methods based on transformation domain. In recent years, the most common method is the maximum mutual information(MI), because it is easily to realize and has no need of pretreatment such as extraction feature segmentation or any other features. So it is easy to be performed and applied to remote sensing images of different patterns. But because of the large amount calculation, this method is very easy to reach local optimum and it is very sensitive to the initial values. Thus, some improved optimization algorithms come into being.This paper first introduces the method of image registration based on mutual information, using the local search operator Powell algorithm to find the optimal parameters of registration measure function. However, the Powell operator belongs to local search operators, so it is sensitive to the initial values and has a bad performance on image registration. In order to overcome the problems of Powell and MI(mutual information) which are sensitive to the initial values, this paper proposes a Coarse-to-Fine image registration method. This method use the Fourier-Mellin transform to get the initial values of the two images, and achieve the purpose of improving the accuracy of registration.Secondly, the image registration problems can be transformed into the problems of optimizing the similarity measure function. The artificial immune algorithm was considered as one of the most potential intelligent searching algorithms. So this paper proposed to apply the Clonal Selection Algorithm(CSA) of artificial immune systems to the image registration problems. Compared with some mature evolutionary algorithm, Clonal Selection Algorithm(CSA) has a very obvious advantage because of its varieties of immune system mechanism. And it can effectively avoid the premature convergence and guarantee the global optimal solution at the same time. Mounts of experiments show that the methods above are applicability and robustness in remote sensing image registration.