Raindrop Removal using Image Inpainting

Image Restoration under severe weather circumstances has drawn a lot of interest for many computer vision applications. In order to deliver accurate and high quality surveillance in the context of smart cities, image de-raining is a crucial subject that has been explored extensively in recent years. In order to handle the challenge of removing raindrops, two different strategies are adopted in this research study: The Diffusion model and the Generative adversarial network model. By considering the recent improvements in image deraining methods, this research study proposes a novel technique that makes use of conditional generative adversarial network with adversarial loss, which provides a factor to loss functions and regulates the output for achieving the improved results. In addition, diffusion modelling, a novel patch-based method is used to perform image restoration. Diffusion probabilistic frameworks are used for normalising noise over affected regions. This research study compares and evaluates how well these two techniques perform in eliminating the raindrops from images. This study demonstrates that the diffusion model outperforms the GAN technique in terms of qualitative assessments and visual appearance by conducting a comparative analysis on actual and synthetic data.