Single image deraining via nonlocal squeeze-and-excitation enhancing network

Abstract

Raindrop blur or rain streaks can severely degrade the visual quality of the images, which causes many practical vision systems to fail to work, such as autonomous driving and video surveillance. Hence, it is important to address the problem of single image de-raining. In this paper, we propose a novel deep network for single image de-raining. The proposed network consists of three stages, including encoder stage, Dense Non-Local Residual Block (DNLRB) stage, and decoder stage. As spatial contextual information has been analyzed to be meaningful for image de-raining (Huang et al. ??), we adopt squeeze-and-excitation enhancing on feature maps in each convolution layer for capturing spatial contextual information. In addition, to better leverage spatial contextual information for extracting rain components, the non-local mean operation has been embed in DNLRB. Both quantitative and qualitative experimental results demonstrate the proposed method performs favorably against the state-of-the-art de-raining methods. The source codes will be available at https://supercong94.wixsite.com/supercong94 .