Double non-local adaptive structural priors for pan-sharpening

In this letter, we propose a novel pan-sharpening model with double non-local adaptive structural priors under the plug-and-play (PnP) framework, which aims to generate a high-resolution multispectral (HRMS) image by preserving the spectral information from a low-resolution multispectral (LRMS) image and the spatial content from a panchromatic (Pan) image. Specifically, for modelling the first non-local adaptive structural prior, we particularly exploit the non-local self-similarity (NSS) prior on the difference image of HRMS and Pan which is hence imposed as a non-local PnP prior term to enhance the spatial structures. Moreover, for modelling the second non-local adaptive structural prior, we rely on the convolution neural network (CNN)-based pan-sharpening method and particularly exploit another NSS prior on the difference image of HRMS and CNN pan-sharpened MS which is hence imposed as another non-local PnP prior term to further preserve spectral information and spatial structures simultaneously. Then, we construct the proposed model with double non-local PnP prior terms and the common degradation-based spectral fidelity term between HRMS and LRMS images. Furthermore, the alternating direction method of multipliers (ADMM) is adopted to solve the proposed model. Finally, extensive experiments exhibit the excellent performance of the proposed method.