A neural network approach to remove rain using reconstruction and feature losses
- PDF / 2,224,162 Bytes
- 10 Pages / 595.276 x 790.866 pts Page_size
- 60 Downloads / 184 Views
(0123456789().,-volV)(0123456789(). ,- volV)
GREEN AND HUMAN INFORMATION TECHNOLOGY 2019
A neural network approach to remove rain using reconstruction and feature losses Kamran Javed1,2 • Ghulam Hussain3 • Furqan Shaukat1 • Seong Oun Hwang4 Received: 9 May 2019 / Accepted: 5 October 2019 Springer-Verlag London Ltd., part of Springer Nature 2019
Abstract Rain streaks can eclipse some information of an image taken during rainfall which can degrade the performance of a vision system. While existing rain removing methods can recover the semantic structure, they lack natural texture recovery. The aim of this work is to recover the hidden structure and texture under the rain streaks with fine details. We propose a novel generative adversarial network with two discriminators to remove rain called rain removal generative adversarial network, where a combination of reconstruction, feature and adversarial losses is used for low level, structural and natural recovery, respectively. We have found that exploiting low-level l1 loss with high-level structural similarity loss as a reconstruction loss is quite effective in attaining visually plausible and consistent texture. Qualitative and quantitative evaluations on our synthetically created dataset and a benchmark dataset show substantial performance gain than state-of-the-art rain removing methods. Keywords Rain removal Generative adversarial network Structural similarity loss UNET Pix2Pix
1 Introduction Rain streaks introduce a lot of undesirable noises in the images taken in rainy conditions. It does not only deteriorate semantic structure, but it also adds unnecessary texture to the images, which harms the performance of surveillance systems. Removing rain streaks can enhance the performance of & Seong Oun Hwang [email protected] Kamran Javed [email protected] Ghulam Hussain [email protected] Furqan Shaukat [email protected] 1
Faculty of Electronics and Electrical Engineering, University of Engineering and Technology, Taxila 47050, Pakistan
2
College of Information and Communication Engineering, Sungkyunkwan University, Suwon 16419, Korea
3
Quaid-e-Awam University of Engineering, Science and Technology, Larkana, Pakistan
4
Department of Software and Communications Engineering, Hongik University, Sejong, Korea
outdoor vision systems in rainy conditions. However, it is difficult to recover visually plausible and consistent texture under the rain streaks. To solve this problem, many techniques have been proposed in recent times but most of them lack recovering natural looking texture. Previous works in this area can be classified into two broad categories: image-based methods and video-based methods. Rain streaks in sequence of frames can be easily identified and thus removed using the redundant temporal information present in the frames of video [1, 9]. Garg et al. exploited correlation to design an algorithm for rain streak detection [9]. To eliminate streaks, the algorithm used an average pixel value computed
Data Loading...
