Reference Based Super Resolution

Reference-based super-resolution (RefSR) aims to enhance low-resolution images by leveraging information from a high-resolution reference image, overcoming limitations of single-image super-resolution. Current research focuses on improving correspondence matching between the low- and high-resolution images, often employing transformer networks, attention mechanisms, and diffusion models to effectively transfer texture and detail. These advancements are improving image quality in various applications, including remote sensing, medical imaging (e.g., OCT angiography), and even enhancing the visual fidelity of rendered images in computer vision and robotics. The development of large-scale multi-reference datasets and efficient algorithms is driving further progress in this field.