A Model-Based Method for Pan-Sharpening of Multi-Spectral Images using Sparse Representation

Pan-sharpening (PS) is fusion of the low-resolution multi-spectral (LRM) image with the corresponding high resolution panchromatic (HRP) one, which aims to reach the high-resolution multi-spectral (HRM) image. Due to the importance of designing well-adapted dictionaries for the pansharpening problem as the fundamental challenge in the pan-sharpening methods utilizing sparse representation, we present a novel strategy to tackle this issue. Our method takes the image formation model into account in a patch-based strategy and exploits local spectral-spatial information. To make a fair tradeoff between spectral and spatial information of each patch, some local parameters are considered and estimated adaptively. In addition, the energy ratio between LRM/LRP patches is considered to locally reconstruct the same ratio for the corresponding HRM/HRP patches. This model-based pan-sharpening strategy led to a closed-form solution, which results in precise HRM dictionary atoms. To obtain the unknown HRM image, after designing the HRM dictionary from input Pan and multi-spectral images, sparse representation over LRM and HRM dictionaries will be conducted in the sparse coding stage. The proposed method has been applied to two different datasets collected by WorldView-3 and GeoEye-1, and then compared with some popular and state-of-the-art methods, qualitatively and quantitatively.