Removal of visual disruption caused by rain using cycle-consistent generative adversarial networks

Tang, L. M. , Lim, L. H. and Siebert, P. (2019) Removal of visual disruption caused by rain using cycle-consistent generative adversarial networks. In: European Conference of Computer Vision (ECCV), Workshop on Autonomous Navigation in Unconstrained Environments. Series: Lecture notes in computer science. Springer: Cham, pp. 551-566. ISBN 9783030110208 (doi: 10.1007/978-3-030-11021-5_34)

Abstract

This paper addresses the problem of removing rain disruption from images without blurring scene content, thereby retaining the visual quality of the image. This is particularly important in maintaining the performance of outdoor vision systems, which deteriorates with increasing rain disruption or degradation on the visual quality of the image. In this paper, the Cycle-Consistent Generative Adversarial Network (CycleGAN) is proposed as a more promising rain removal algorithm, as compared to the state-of-the-art Image De-raining Conditional Generative Adversarial Network (ID-CGAN). One of the main advantages of the CycleGAN is its ability to learn the underlying relationship between the rain and rain-free domain without the need of paired domain examples, which is essential for rain removal as it is not possible to obtain the rain-free image under dynamic outdoor conditions. Based on the physical properties and the various types of rain phenomena [10], five broad categories of real rain distortions are proposed, which can be applied to the majority of outdoor rain conditions. For a fair comparison, both the ID-CGAN and CycleGAN were trained on the same set of 700 synthesized rain-and-ground-truth image-pairs. Subsequently, both networks were tested on real rain images, which fall broadly under these five categories. A comparison of the performance between the CycleGAN and the ID-CGAN demonstrated that the CycleGAN is superior in removing real rain distortions.

• Item Type: Book Sections
• Additional Information: Paper first presented at First International Workshop On Autonomous Navigation in Unconstrained Environments, Munich, Germany, 08 Sep 2018.
• Status: Published
• Glasgow Author(s) Enlighten ID: Tang, Dr Lai Meng and Siebert, Dr Paul and Lim, Dr Li Hong Idris
• Authors: Tang, L. M., Lim, L. H., and Siebert, P.
• Subjects: Q Science > Q Science (General)
• College/School: College of Science and Engineering > School of Computing Science; College of Science and Engineering > School of Engineering > Electronics and Nanoscale Engineering
• Publisher: Springer
• ISBN: 9783030110208
• Published Online: 23 January 2019