Fig. 4: Overview of the diffusion model architecture and guided restoration pipeline.
Schematic illustration of the denoising diffusion process and the proposed guided restoration framework. (Top) Forward diffusion process, where a clean manuscript image x0 is gradually corrupted with Gaussian noise to obtain xT. (Bottom) Reverse diffusion process, where the model iteratively reconstructs x0 from xT under the guidance of an observed degraded image y. In this framework, y represents the observed measurement or degraded input (e.g., blurred, stained, or incomplete image). By conditioning the reverse process on y, the model reconstructs a clean and faithful restoration of the original manuscript.
