Multispectral and Panchromatic Image Fusion With Guaranteed Detail Injection-Enhanced Low-Rank Model

Abstract

Based on the detail injection framework, this letter proposes a novel guaranteed detail injection-enhanced low-rank (DILR) model for multispectral and panchromatic (Pan) image fusion (i.e., pansharpening), which fuses a low-resolution multispectral image (LrMSI) and a Pan image to acquire a high-resolution multispectral image (HrMSI). Specifically, despite using the detail injection fidelity constraint that injects the spatial details of Pan into the LrMSI for spectral and spatial fidelity, we novelly reinterpret the detail injection fidelity constraint as the band-wise low-rank prior, theoretically and practically from the perspective of structural correlation modeling, and propose a novel nuclear norm-based DILR prior term for enhancing the structural correlations among LrMSI, Pan, and HrMSI. Moreover, we derive an efficient algorithm to optimize the proposed DILR model and further give the detailed convergence proof of the proposed DILR algorithm. Finally, we provide the reduced-resolution and full-resolution experimental results to verify the superiority of DILR.