Fusing hyperspectral and multispectral images using smooth graph signal modelling

Abstract

ABSTRACT Hyperspectral image (HSI) and multispectral image (MSI) fusion aiming to improve HSI spatial resolution has attracted increasing research interests in the recent decade. Fusing hyperspectral and multispectral images is formulated as a linear inverse problem (LIP), which solution is considered to live in a lower dimensional subspace spanned by the high-resolution HSI vectors. LIP is generally ill-posed and does not have a unique solution. Therefore, some prior information or regularization terms are required to convert it to a well-posed LIP. In this paper, the smooth graph signal modelling is suggested to incorporate the spatio-spectral joint structures of the HSIs. As spatially near pixels have been expected to have similar spectral responses, the image graph model is considered to be smooth if strongly connected nodes have similar values. We suggest the consistency of subspace projection fractions corresponding to the nearby nodes by exploiting the graph Laplacian matrix and investigate an optimization algorithm based on the alternating direction method of multipliers (ADMM). Experiments on the real well-known datasets demonstrate the superiority of the proposed method compared with the current state-of-the-art HSI and MSI fusion approaches.