Single-image rain removal using deep residual network
- PDF / 2,686,677 Bytes
- 8 Pages / 595.276 x 790.866 pts Page_size
- 71 Downloads / 233 Views
ORIGINAL PAPER
Single-image rain removal using deep residual network Genji YUAN1,2 · Jinjiang LI1,2
· Zhen HUA1
Received: 19 March 2020 / Revised: 9 September 2020 / Accepted: 13 October 2020 © Springer-Verlag London Ltd., part of Springer Nature 2020
Abstract Rain not only reduces visibility, but also prevents many computer vision algorithms from functioning properly. Removing rain from a single image is an important issue in the field of computer vision. Rain usually produces a phenomenon similar to fog, which is more pronounced when the rain is dense. We constructed a rain model, including rain streaks and fog generated by rainfall. Based on this model, we constructed a multi-task learning network to learn the rain streaks and fog in the image as well as the final clean background. To take advantage of contextual information, we use an expanded convolution network to handle different levels of rain and fog. We learn the residual images of fog and rain based on the deep residual network to avoid the influence of inaccurate parameter estimation on the results. Experiments show that the method in this paper can effectively remove rain and fog in the image. Keywords Rain removal · Rain model · Expanded convolution network · Residual network
1 Introduction Most computer tasks assume that the quality of the image is high enough. However, various degradations often occur in real-world scenes (especially in outdoor scenes). Removal of rain streaks from images is complex. Existing imageformation models [1] are inadequate to cover some of the important factors in real rainfall images, such as the veils due to rain accumulation and the shape and direction of different rain streaks. Moreover, due to the uneven distribution of rain density in the image, most rain removal methods tend to remove texture detail from non-rainy areas, ignoring the overlap between the rain pattern and the background texture pattern, resulting in areas that are too smooth. This paper proposes a novel rain-image formation model, which can better describe the rain conditions in real scenes. Then, this paper designs an effective deep learning frame-
B
Jinjiang LI [email protected] Genji YUAN [email protected] Zhen HUA [email protected]
1
School of Computer Science and Technology, Shandong Technology and Business University, Yantai 264005, China
2
Co-innovation Center of Shandong Colleges and Universities: Future Intelligent Computing, Yantai 264005, China
work based on this model. To better remove rain streaks, we explored possible rain streaks models and deep learning frameworks to jointly remove rain and fog from a single image. In the rain-image formation model in this paper, we use the alpha mapping of rain streaks, where [0,1] represents the visible rain streaks in the pixel, and the rest are 0. We also simulated the fog-like appearance and the various shapes and directions of overlapping rain streaks. Based on the model in this paper, we construct a deep network for removing rain and fog in the image. We use dark channel pr
Data Loading...
