WDFN: Wavelet-Based Decomposition-Fusion Network for Low-Light Underwater Image Enhancement
Liqun Zhou, Hao Tan, Tingting Liu, Yanyan Zhu, Jishen Liang, Yang Tao · IEEE Transactions on Geoscience and Remote Sensing · 2025
Existing underwater image enhancement methods are primarily designed for well-illuminated environments. It is a challenge to improve the quality of low-light underwater images effectively. Therefore, a wavelet-based decomposition-fusion network (WDFN) is proposed to address hybrid degradation of images in low-light underwater environments. Firstly, the bright compensation process is guided by the reverse luminance map. Then, an adaptive grouped structure enhancement module is designed based on the characteristics of water absorption. A detail enhancement module is designed by using the directional characteristics of high-frequency sub-images. The features of the structure enhancement module and the detail enhancement module are fused by using interactive weighting. Finally, a color correction module is designed based on white balance theory. Given the scarcity of low-light underwater datasets, the LU100 dataset is built to validate the generalization ability of the model. Extensive qualitative and quantitative experimental results demonstrate that the proposed WDFN effectively enhances brightness, preserves structural integrity, improves detail, and balances color. The dataset and source code are available at https://github.com/lqjw81/WDFN.