July 10, 2025

Joint Restoration Method Based on Super-resolution Network and Diffusion Model

XMXin MiaoRuijin Hospital CLChenghan LiCalifornia Institute of Technology MXMingfei XinHarbin Engineering University

Key Points

The proposed method significantly improves image clarity and physical authenticity, enhancing underwater imaging.
Experiments show superior performance in PSNR and SSIM metrics, particularly in complex occlusion scenarios.
The approach integrates a super-resolution network with a diffusion model for effective image restoration.
This advancement may transform image processing techniques in challenging underwater environments.

Abstract

To address the challenges of low image resolution obtained by underwater robots and poor mask inpainting effect, a joint restoration method based on super-resolution network and diffusion model is proposed. Through the degradation model and deep learning, the mapping between high and low resolution images is achieved, and combined with the reverse optimization ability of the diffusion model, the image clarity and physical authenticity are significantly improved. Experiments show that the proposed method outperforms existing technologies in terms of PSNR, SSIM, UIQM and LPIPS, especially in the restoration effect under complex occlusion scenarios. This research effectively enhances the visual perception ability of underwater robots and provides new ideas for image processing in complex water areas.

Demander à l'IA

Bookmark

View Full Paper