Fusion-based graph neural networks for synergistic underwater image enhancement

Chengpei Xu, Wenhao Zhou, Zhixiong Huang, Yuanfang Zhang, Yan Zhang, Weimin Wang, Feng Xia

Abstract

Highlights•Challenges with Conventional Approaches: While CNN-based UIE methods effectively extract local features, their constrained receptive fields limit the handling of comprehensive contextual information.•Graph Neural Networks for Enhanced Range: GNNs address these limitations by effectively processing relational information across extended spatial contexts, capturing long-range features in underwater images.•Innovative Fusion-based Approach: Our FUGN integrates the strengths of CNNs and GNNs, using image segmentation into blocks optimized for GNN processing with Sobel and Gaussian Blur operators to enhance texture and gradient-based similarities.•Proven Effectiveness of FUGN: Extensive experimental validations of FUGN demonstrate significant improvements in the quality and fidelity of enhanced underwater images.
BibTeX
@article{xu2025fusion,
  author = {Chengpei Xu and Wenhao Zhou and Zhixiong Huang and Yuanfang Zhang and Yan Zhang and Weimin Wang and Feng Xia},
  title = {{Fusion-based graph neural networks for synergistic underwater image enhancement}},
  journal = {Information Fusion},
  year = {2025},
  url = {https://openreview.net/forum?id=VldaoktuKm},
}

Loading related content...

Figure