Rain Removal for Single Image based on DenseNet

Under rainy conditions, computer vision systems will be disturbed by rain, resulting in degraded image quality. So, it is important to remove rain streaks from images and preserve image information meanwhile. For removing rain from a single image, the existing methods, whether based on prior knowledge or neural network, are either excessive or incomplete. In order to effectively remove rain streaks from a single image, the guided filtering is proposed for image pre-processing, the DenseNet is used as the basic network, and the dense blocks with different structures are used to build rain removal network. The experimental results show that, the proposed method can effectively extract the rain streaks residual mapping, remove the rain streaks in the rain image, and better preserve and restore the original background of the image.