Iteratively Regularizing Hyperspectral and Multispectral Image Fusion With Framelets

Abstract

Hyperspectral (HS) and multispectral (MS) image fusion mainly focuses on transferring spatial details from high spatial resolution (HR) MS images (MSIs) to low spatial resolution (LR) HS images (HSIs). Recent investigations introduce prior regularizations, such as sparsity, low-rankness, or total variation, to enhance fusion quality by denoising latent factor images. This article proposes a new HS and MS image fusion approach using two kinds of regularization. First, we design a flexible plug-and-play framelet for fusion purposes, which denoises factor images by leveraging high-pass and low-pass filters for simultaneously promoting sparsity and spatial smoothness properties. Second, we iteratively regularize the fusion task by enhancing the quality of the LR-HSI and the HR-MSI, updating the input image pairs by injecting components from the resulting HR-HSI. The proposed model is solved by the alternating direction method of multipliers. Experimental results on simulated and real datasets indicate the superiority of the proposed approach compared to some state-of-the-art methods.