New Techniques for Preserving Global Structure and Denoising with Low Information Loss in Single-Image Super-Resolution

This work identifies and addresses two important technical challenges in single-image super-resolution: (1) how to upsample an image without magnifying noise and (2) how to preserve large scale structure when upsampling. We summarize the techniques we developed for our second place entry in Track 1 (Bicubic Downsampling), seventh place entry in Track 2 (Realistic Adverse Conditions), and seventh place entry in Track 3 (Realistic difficult) in the 2018 NTIRE Super-Resolution Challenge. Furthermore, we present new neural network architectures that specifically address the two challenges listed above: denoising and preservation of large-scale structure.

1. Introduction

Super-resolution (SR) is a classic problem in image processing where the goal is to generate a high resolution image from one or more low resolution images. Applications of super-resolution are wide-ranging. For instance, SR is important for allowing modern high-definition displays to function properly when showing video recorded at lower resolutions. SR also has many applications in medical imaging, such as reducing noise in images stemming from uncontrollable patient motions (11). This work focuses on single image super-resolution, which is useful for photographic enhancement, license plate recognition, satellite imaging, and other remote sensing applications such as recognition of a military target (16).