COC-UFGAN: Underwater image enhancement based on color opponent compensation and dual-subnet underwater fusion generative adversarial network

Zhenkai Liu,Xinxiao Fu,Chi Lin,Haiyong Xu
DOI: https://doi.org/10.1016/j.jvcir.2024.104101
IF: 2.887
2024-02-29
Journal of Visual Communication and Image Representation
Abstract:Due to the complex underwater environment and the attenuation of light, the underwater image is produced with various distortions such as color loss, low contrast, noise, blur, and haze-like, which bring significant challenges to underwater image applications. To alleviate these problems, a novel color opponent compensation and a dual-subnet underwater fusion generative adversarial network (COC-UFGAN) are proposed by simulating the information processing mechanism in the human visual system (HVS). Specifically, considering the color opponent mechanism of the primary visual cortex and the characteristics of underwater imaging, an opponent domain is constructed consisting of four color opponent channels: green–red, yellow-blue, cyan-red and black-white. Then, a novel local-to-global color opponent compensation method is designed to restore color loss in underwater imaging. Furthermore, considering the complexity of underwater imaging, a dual-subnet underwater fusion generative adversarial network is designed. Finally, experimental results on synthetic and natural underwater images demonstrate the superiority of the proposed COC-UFGAN over the state-of-the-art methods qualitatively and quantitatively.
computer science, information systems, software engineering
What problem does this paper attempt to address?