WSUIE : Weakly Supervised Underwater Image Enhancement for Improved Visual Perception

Research output: Journal Publications and Reviews (RGC: 21, 22, 62)21_Publication in refereed journalpeer-review

View graph of relations

Author(s)

  • Lin Hong
  • Xin Wang
  • Zhenlong Xiao
  • Gan Zhang
  • Jun Liu

Related Research Unit(s)

Detail(s)

Original languageEnglish
Pages (from-to)8237-8244
Journal / PublicationIEEE Robotics and Automation Letters
Volume6
Issue number4
Online published18 Aug 2021
Publication statusPublished - Oct 2021

Abstract

Underwater images suffer from degradation and blur inevitably due to the scattering and absorption of light as it propagates through the water, which hinders the development of underwater visual perception. Existing deep underwater image enhancement methods mainly rely on strong supervision in the form of large-scale datasets of aligned raw/enhanced underwater image pairs for model training, however, aligned image pairs are unavailable in most underwater cases. In this work, we address this problem by proposing a novel weakly supervised underwater image enhancement (WSUIE) method. Specifically, a novel generative adversarial network (GAN)-based architecture is firstly designed to enhance underwater images by an unpaired image-to-image transformation from domain X (raw underwater images) to domain Y (arbitrary high-quality images), which alleviates the demand for aligned underwater image pairs. Then, a new objective function is formulated by exploring inherent depth information of underwater images to increase the depth sensitivity of our method. In addition, a dataset with unaligned image pairs (named UUIE dataset) is provided for the underwater image enhancement model training, many qualitative and quantitative evaluations of the WSUIE method are carried out on this dataset, the results show that the WSUIE can provide improved visual perception performance while enhancing underwater images.

Research Area(s)

  • Generative adversarial networks, generative adversarial networks (GAN), Image color analysis, Image enhancement, Task analysis, Training, Underwater image enhancement, underwater visual perception, Visual perception, Visualization, weakly supervised learning

Citation Format(s)

WSUIE : Weakly Supervised Underwater Image Enhancement for Improved Visual Perception. / Hong, Lin; Wang, Xin; Xiao, Zhenlong; Zhang, Gan; Liu, Jun.

In: IEEE Robotics and Automation Letters, Vol. 6, No. 4, 10.2021, p. 8237-8244.

Research output: Journal Publications and Reviews (RGC: 21, 22, 62)21_Publication in refereed journalpeer-review