Bridging the Domain Gap: A Simple Domain Matching Method for Reference-based Image Super-Resolution in Remote Sensing

TL;DR

This work addresses the domain gap problem in reference-based image super-resolution for remote sensing by introducing a Domain Matching (DM) module that can be plugged into existing RefSR models without extra training. The method combines Gray Matching to reduce cross-domain disparities in matching, with a texture-transfer pipeline that uses Whitening and Coloring Transform (WCT) and Phase Replacement (PR) to adapt textures while preserving structural information. The authors validate their approach on the RRSSRD dataset, showing consistent PSNR and SSIM gains and improved matching quality (SSIM-s) across cross-sensor scenarios. The proposed DM module enhances real-world RefSR performance for electro-optical imagery, offering a simple, effective solution to domain gaps between LR and Ref images collected by different satellites. This has practical implications for improving SR-driven downstream tasks in remote sensing where cross-sensor heterogeneity is common.

Abstract

Recently, reference-based image super-resolution (RefSR) has shown excellent performance in image super-resolution (SR) tasks. The main idea of RefSR is to utilize additional information from the reference (Ref) image to recover the high-frequency components in low-resolution (LR) images. By transferring relevant textures through feature matching, RefSR models outperform existing single image super-resolution (SISR) models. However, their performance significantly declines when a domain gap between Ref and LR images exists, which often occurs in real-world scenarios, such as satellite imaging. In this letter, we introduce a Domain Matching (DM) module that can be seamlessly integrated with existing RefSR models to enhance their performance in a plug-and-play manner. To the best of our knowledge, we are the first to explore Domain Matching-based RefSR in remote sensing image processing. Our analysis reveals that their domain gaps often occur in different satellites, and our model effectively addresses these challenges, whereas existing models struggle. Our experiments demonstrate that the proposed DM module improves SR performance both qualitatively and quantitatively for remote sensing super-resolution tasks.

Figures (3)

• Figure 1: Overview We applied a grayscale transform to the input images before feeding it into the matching encoder during the correspondence matching process. In the texture transfer process, $F^t_{Ref}$ is the domain adapted by $F^t_{LR}$ to enhance the information of transferring textures. In this experiment, the domain adaptation module indicates WCT with phase replacement technique. $EO_A$ and $EO_B$ refer to electro-optical images captured by different satellites.
• Figure 2: Visualization of matching results of DATSR with and without grayscale transformation in the corresponding matching process. The matching result indicates the outcome after identifying the most relevant reference feature to the input, implying that a better matching is achieved as it becomes closer to the input. Our approach demonstrates a much cleaner and more effective retrieval of detailed information compared to the existing baseline.
• Figure 3: Qualitative comparison of baseline RefSR models and proposed methods. Our proposed method shows better performance compared to previous SOTA models and recovers more details and shows good visual quality.