| With the development of remote sensing technology,the detection bands of air-borne or spaceborne imaging spectrometers make a great progress.From a single band to multiple bands,or even hundreds or thousands of bands,which leads to multi-spectral or hyper-spectral images with rich bands.In addition,due to the difference of remote sensing platforms and sensor imaging principles,various remote sensing images with different information are obtained.In order to fully exploit the common comple-mentary information,multi-source remote sensing images fusion has become one of the popular topics in the research of remote sensing.A conventional problem for multi-source remote sensing images fusion is fusing the panchromatic(PAN)image and multi-spectral(MS)image,which aims at sharpening the spatial resolution of the MS image with the details of PAN.In this paper,we pay attention to enrich the spectral information of the PAN image and improve the spatial resolution of the MS image in the fusion of PAN and MS images.Based on the proposed spectral mapping fusion model,two new image fusion models are proposed,which promote the fusion results from the spatial and transform domain,respectively.The main researches are as follows:1.Different from the general fusion model which only uses the spatial struc-ture information of the PAN image,a fusion pre-processing method that expands the spectral information of the PAN image is proposed.Specifically,based on image super-resolution and hybrid color mapping,the correspondence relationship between two pairs of high and low resolution PAN images and MS images are established.Then by cou-pling the relationship with imaging model,a spectral mapping based fusion model is proposed,which produces a new PAN(SPAN)image with certain spectral information.a fast Gauss-Seidel type inertial proximal alternating linearized minimization algorithm is used to solve the fusion model,and the convergence of the algorithm is analyzed.2.Considering the SPAN image obtained form spectral mapping fusion has less spectral information,a spatial detail injection technology is employed to enrich the spectral as well as spatial information,and a new fusion model and algorithm is pro-posed.Compared with other state-of-art fusion methods,experimental results show that the proposed method has advantages on both spectral information fidelity and spatial structure preservation.3.Considering the SPAN image obtained form spectral mapping fusion has less spectral information,a refining fusion method in shearlet transform domain is proposed.The method fully uses the merits of shearlet transform,which decomposes image with more directions and scales.There are two main contributes in the proposed method.One is a two-step fusion promotion method,which makes full use of the comprehensive information of PAN,MS and SPAN images.And another is the data-depended fusion strategy,which utilizes the features of different source images to design a fusion rules based on the fourth-order correlation coefficient and local variance for low-frequency and high-frequency coefficients.Comparing with other fusion methods,the fusion image has improvement in visual effects as well as evaluation indexes,which further shows that the proposed method can achieve a better fused images with higher spatial and spectral resolutions. |
