Patch Based Pansharpening Using Weighted Nuclear Norm Minimization

Abstract

This paper proposed a multispectral (MS) and panchromatic (PAN) image fusion method based on low-rank assumption captured by weighted nuclear norm minimization (WNNM). In this method, low-rank matrix factorization is considered to model the relationship between low spatial resolution (LR) and high spatial resolution (HR) MS images. In the formulation, MS and PAN images are partitioned into patches and then clustered to further guarantee the low-rank property. Besides, WNNM is used to capture the prior about singular values, in which larger singular values are shrunked with smaller weights. By WNNM, the spatial details in MS images can be well enhanced. Finally, the fusion model is established by combining the low-rank matrix factorization with the fidelity term about PAN image. The experimental results on degraded and real datasets demonstrate the effectiveness of the proposed method.