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

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.