Image rain removal algorithm based on conditional diffusion model
Yuxin Li, Lijun Liu, Baoying Ma · 2023
Rainfall image restoration is one of the research hotspots in the field of computer vision. For raindrops attached to the camera lens or glass, it will hinder the visualization of the complete image and also lead to image degradation. Most of the existing methods for converting raindrop-degraded images into clean images use generative adversarial neural networks (GANs) to solve this problem. At present, the denoising probability diffusion model has been widely used in various fields. Therefore, this paper proposes a raindrop image restoration algorithm based on conditional diffusion model, which improves the U-Net network used for noise prediction in diffusion model. The model is verified and evaluated on the raindrop removal dataset, and has strong generalization ability for raindrop image restoration in the real world.