Transformer-based Diffusion Model for Single Image Deblurring

A Transformer-based diffusion model for single image deblurring is proposed in this paper, which combines diffusion model and transformer block. Specifically, the blurred image is first used as a condition to guide the diffusion model to reverse restore and generate deblurred clean background image. Secondly, a U-shape Spatially Gate Transformer Net for Noise Estimation(USGTN-NE) is proposed, which utilizes blurred image and noisy states to estimate the distribution of noise. Introducing Spatially Adaptive Feature Modulation(SAFM) into the network to capture multi-scale and global information, combined with gate mechanisms to enhance the network's noise prediction capability. The experimental results show that the PSNR index and SSIM index of our proposal are higher than those of other competitors. The subjective quality has also been improved, and the restored images are visually closer to the real situation.