Image deraining aims to restore the clean scenes of rainy images, which facilitates a number of outdoor vision systems, such as autonomous driving, unmanned aerial vehicles and surveillance systems. This paper proposes a high-resolution detail-recovering image deraining network (HDRD-Net) to effectively remove rain streaks and recover lost details, as well as improving the quality of derained images. HDRD-Net consists of three sub-networks. First, we combine the residual network and Squeeze-and-Excitation block for rain streak removal. Second, we integrate the Structure Detail Context Aggregation block into the detail-recovering network to extract detail features form rainy images. Third, a dual super-resolution reconstruction network is utilized to enhance the quality of derained images. In addition, we extend the Rain100 dataset by incorporating low-resolution rainy images to construct a new Rain100++ dataset for high-resolution image deraining. Experimental results on several datasets show that HDRD-Net outperforms state-of-the-art methods in terms of rain removal, detail preservation and visual quality.
Bibliographical noteFunding Information:
The work described in this paper was fully supported by a grant from the Research Grants Council of the Hong Kong Special Administrative Region, China (UGC/FDS16/E01/19), Hong Kong Metropolitan University Research Grant (No. RD/2021/09), the One-off Special Fund from Central and Faculty Fund in Support of Research from 2019/20 to 2021/22 (MIT02/19-20), the Research Cluster Fund (RG 78/2019-2020R), the Interdisciplinary Research Scheme of the Dean’s Research Fund 2019-20 (FLASS/DRF/IDS-2) of The Education University of Hong Kong, and the Lam Woo Research Fund (LWI20011) of Lingnan University, Hong Kong.
© 2022, The Author(s), under exclusive licence to Springer Science+Business Media, LLC, part of Springer Nature.
- Image deraining
- Outdoor vision systems