Multi-resolution Remote Sensing Image Registration And Fusion Method

Currently, satellite remote sensing systems will provide large amounts of images representing multiple-time or simultaneous observations of the same features by different sensors. To make full use of these measurements, developing the technology for fusing information from different images is justified. Image registration and fusion algorithm is one of the crucial parts in such technology.Multiresolution-based image registration which is a widely used method typically employs discrete wavelet transform to represent the original images at several spatial resolutions, followed by progressive alignment of the image representations by going from the coarsest one to the finest. Multiresolution-based method is faster than those that register images at full resolution. However, recent research shows that image registration methods based on discrete wavelet transform are sensitive to image translation. In this paper, we study the mechanism how image translation affects these image registration algorithms. An analytic-wavelet-based algorithm is proposed to solve this problem. Both theoretic analysis and practical data demonstrate that our method is robust against image translation. Besides, we also explore the search strategy of multiresolution-based image registration. It is known that there are often an excessive number of local maxima when mutual information is used as the target function. In this paper, we introduce a new multiresolution-based registration method by using dual target functions, namely utilizing mutual information of both high-frequency subband and low-frequency subband as target functions. Analysis shows that the dual target function strategy can effectively prevent optimization algorithm from ending up with local maximum. Then, a new hybrid optimization method based on artificial immune system and Powell search is also presented to find the global maximum more quickly. The new optimization can find the global maximum more efficiently. Experiment results using remote sensing images confirm the effectiveness of our method.Moreover, traditional injection-model-based multispectral and panchromatic images fusion method is investigated in this paper. Since injection model parameters are typically affected by numerical instability and may give unsteady detail injection, a novel image fusion method is introduced. Spatial detail of panchromatic image is extracted by the undecimated wavelet, the parameters of injection model are set through immune optimization algorithm. The new fusion method is free of manually defined parameters and overcomes the numerical instability problem. Both visual and object comparisons of the final image fusion result shows the improvement of the proposed method.