Regularizing Subspace Representation for Fusing Hyperspectral and Multispectral Images

Fusing a low spatial resolution hyperspectral image (LR-HSI) with a high spatial but low spectral resolution multispectral image (HR-MSI) has been regarded as an effective approach to obtain high resolution HSI (HR-HSI). While matrix factorization-based approaches obtained promising performance for HSI-MSI fusion, the mixed noise introduced into the LR-HSIs inevitably influence the accuracy of the estimated endmembers and representation coefficients. Therefore, to effectively fuse the LR-HSI and HR-MSI information, and meanwhile avoid the impact of mixed noise, in this article, we introduce a denoising regularized subspace representation-based HSI-MSI fusion paradigm. To do so, by assuming that the spectral subspaces of the desired HR-HSI and the acquired LR-HSI are similar, we reformulate the fusion problem as the estimation of the low-dimensional subspace and representation coefficients. More precisely, first, a robust subspace is estimated by the residual statistics on the median filtered images, which can detect pixels contaminated by mixed noise. Then, the well-known block-matching and 3-D filtering regularizer is incorporated into the alternating direction method of multipliers algorithm, which serves as a convex surrogate for estimating the coefficients, thereby promoting the self-similarity of images. Experimental results, carried out on simulated and real datasets, demonstrate the effectiveness of the proposed approach in terms of preserving both spatial details and texture. Furthermore, significantly improved image quality is observed when compared to those of other methods that do not consider noise effects.