Pansharpening Based on Low-Rank Fuzzy Fusion and Detail Supplement

Abstract

Pansharpening is a technique used to reconstruct a high-resolution (HR) multispectral (MS) image by combining an HR panchromatic (PAN) image with a low-resolution MS image. In recent years, the detail-injection model has demonstrated excellent performance in pansharpening, thus receiving wide attention. Obtaining appropriate details is vital for the detail-injection model. Therefore, this article presents a detail optimization approach to obtain more precise high-frequency (HF) details for pansharpening. The proposed method comprises two steps. In the first step, we design a low-rank fuzzy fusion model to fuse the HF details of the PAN and MS images. In this model, the high frequencies of the PAN and upsampled MS images are decomposed into low-rank and sparse components, and the corresponding fusion rules are designed according to their characteristics. Because some details of the PAN image are replaced with those of the MS image, using them directly as injection details may result in redundant information or spatial distortion. To solve this problem and further optimize the details, in the second step, we construct an adaptive detail supplement model. Based on the similarity and correlation between the fused HF and the original HF of the PAN image, the fused details are supplemented to obtain the final injection details. Experimental results on the IKONOS, Pleiades, QuickBird, and WorldView-2 datasets demonstrate that the proposed algorithm is better than the state-of-the-art methods in maintaining spectral information and improving spatial details.