A New Pan-sharpening Algorithm Based On Variational Method

Due to the limitation of satellite airborne storage technology,image bandwidth transmission technology,and economic factors,it is difficult for a remote sensing image to provide accurate spectral information while providing high spatial resolution.As remote sensing images are more and more widely used in various aspects of social life,national economy and national security,the demand for high quality remote sensing images is increasing.In remote sensing image processing,pan-sharpening is used to obtain a multi-spectral image with high spatial resolution by combining low resolution multi-spectral images with a corresponding high resolution panchromatic images.Based on the variational method and the previous research results,new spectral information preservation terms and spatial information injection terms based on new hypotheses are constructed in this paper.Two new pan-sharpening algorithms are proposed.The main work of this paper is divided into the following parts.First of all,in order to maintain the spectral information of original multi-spectral image,and avoid reducing the accuracy of the fusion result because of using the up-sampled multi-spectral image directly,the spectral constraint term is constructed based on the assumption that the fusion result in the image domain and the gradient domain should be similar to the original multi-spectral image after down sampling.In order to maintain the correlation information between the bands in the original multi-spectral image,by considering the degradation model in the spectral correlation constraint,the proportional relationship between the bands of fusion result become more accurately.Secondly,in order to inject spatial information into the original multi-spectral image,an adaptive weight injection mechanism is proposed to inject different spatial information into each band of the multi-spectral image by considering the difference in spectral coverage between the multi-spectral image sensor and the panchromatic image sensor,avoiding spectral distortion for injecting too much spatial information.Finally,in view of the problem of substitute component estimation in IHS-like methods,a variational method is proposed to estimate the substitute component.First,a spatial information constraint term using gradient is proposed to improve the spatial resolution of the fusion result based on the assumption that the smaller the difference in high frequency information between the substitute component and the panchromatic image,the higher the spatial resolution of the fusion result.Secondly,a spectralinformation constraint term based on semiblind deconvolution is proposed to maintain the spectral information of the original multi-spectral image based on the assumption that spectral distortion of the fusion result can be alleviated when low frequency information of the substitute component approximate intensity component as closely as possible.In addition,in order to verify the effectiveness of proposed methods in this paper,the proposed methods are compared with six state-of-the-art methods using Pleiades,Geoeye and Quick Bird data as test data in subjective and objective aspects.On the whole,the results show that the methods proposed in this paper perform better than others whether in subjective or objective evaluation.