Burst Super-Resolution with Diffusion Models for Improving Perceptual Quality

TL;DR

This paper introduces a burst super-resolution method using diffusion models that enhances perceptual quality by focusing on detail refinement, addressing alignment issues, and optimizing the reverse diffusion process for image restoration.

Contribution

It proposes a novel burst SR approach with diffusion models that skips global structure reconstruction and emphasizes detail refinement, improving perceptual quality.

Findings

Improved perceptual quality metrics on burst SR tasks.

Effective handling of misaligned burst images.

Enhanced detail and boundary sharpness in reconstructed images.

Abstract

While burst LR images are useful for improving the SR image quality compared with a single LR image, prior SR networks accepting the burst LR images are trained in a deterministic manner, which is known to produce a blurry SR image. In addition, it is difficult to perfectly align the burst LR images, making the SR image more blurry. Since such blurry images are perceptually degraded, we aim to reconstruct the sharp high-fidelity boundaries. Such high-fidelity images can be reconstructed by diffusion models. However, prior SR methods using the diffusion model are not properly optimized for the burst SR task. Specifically, the reverse process starting from a random sample is not optimized for image enhancement and restoration methods, including burst SR. In our proposed method, on the other hand, burst LR features are used to reconstruct the initial burst SR image that is fed into an intermediate step in the diffusion model. This reverse process from the intermediate step 1) skips diffusion steps for reconstructing the global structure of the image and 2) focuses on steps for refining detailed textures. Our experimental results demonstrate that our method can improve the scores of the perceptual quality metrics. Code: https://github.com/placerkyo/BSRD