Title:An Effective Two-Branch Model-Based Deep Network for Single Image Deraining

Abstract:Removing rain effects from an image is of importance for various applications such as autonomous driving, drone piloting, and photo editing. Conventional methods rely on some heuristics to handcraft various priors to remove or separate the rain effects from an image. Recent deep learning models are proposed to learn end-to-end methods to complete this task. However, they often fail to obtain satisfactory results in many realistic scenarios, especially when the observed images suffer from heavy rain. Heavy rain brings not only rain streaks but also haze-like effect caused by the accumulation of tiny raindrops. Different from the existing deep learning deraining methods that mainly focus on handling the rain streaks, we design a deep neural network by incorporating a physical raining image model. Specifically, in the proposed model, two branches are designed to handle both the rain streaks and haze-like effects. An additional submodule is jointly trained to finally refine the results, which give the model flexibility to control the strength of removing the mist. Extensive experiments on several datasets show that our method outperforms the state-of-the-art in both objective assessments and visual quality.