Adaptive Dual-domain Learning for Underwater Image Enhancement

TL;DR

This work tackles the challenge of underwater image enhancement by addressing spatial-region and spectral-band degradation simultaneously. It introduces SS-UIE, a dual-domain adaptive network that pairs a spatially global MCSS path with a spectrally global SWSA path in parallel, enabling degradation-level aware UIE with linear complexity. A Frequency-Wise Loss guides the model to recover high-frequency textures by applying frequency-domain supervision and dynamic weighting. Across UIEB, LSUI, and non-reference U45 datasets, SS-UIE achieves state-of-the-art results with lower computational cost and real-time inference, demonstrating practical potential for underwater imaging tasks.

Abstract

Recently, learning-based Underwater Image Enhancement (UIE) methods have demonstrated promising performance. However, existing learning-based methods still face two challenges. 1) They rarely consider the inconsistent degradation levels in different spatial regions and spectral bands simultaneously. 2) They treat all regions equally, ignoring that the regions with high-frequency details are more difficult to reconstruct. To address these challenges, we propose a novel UIE method based on spatial-spectral dual-domain adaptive learning, termed SS-UIE. Specifically, we first introduce a spatial-wise Multi-scale Cycle Selective Scan (MCSS) module and a Spectral-Wise Self-Attention (SWSA) module, both with linear complexity, and combine them in parallel to form a basic Spatial-Spectral block (SS-block). Benefiting from the global receptive field of MCSS and SWSA, SS-block can effectively model the degradation levels of different spatial regions and spectral bands, thereby enabling degradation level-based dual-domain adaptive UIE. By stacking multiple SS-blocks, we build our SS-UIE network. Additionally, a Frequency-Wise Loss (FWL) is introduced to narrow the frequency-wise discrepancy and reinforce the model's attention on the regions with high-frequency details. Extensive experiments validate that the SS-UIE technique outperforms state-of-the-art UIE methods while requiring cheaper computational and memory costs.

Figures (8)

• Figure 1: PSNR-Parameters-FLOPs comparisons with existing UIE methods. The vert axis is PSNR (dB), the horizontal axis is FLOPs (computational cost), and the circle radius is the parameter number (memory cost) of each UIE model. Our SS-UIE outperforms state-of-the-art (SOTA) methods while requiring fewer FLOPs and Params.
• Figure 2: The overall structure of SS-UIE. We combine the spatial-wise Multi-scale Cycle Selective Scan (MCSS) module with the Spectral-Wise Self-Attention (FWSA) module in parallel to form the Spatial-Spectral block (SS-block). MCSS and SWSA can obtain the spatial-wise and spectral-wise global receptive field with linear complexity, respectively. The parallel design facilitates complementary interactions between spatial and spectral features, enabling the SS-block to capture the degradation levels in different spatial regions and spectral bands, thereby achieving degradation level-based dual-domain adaptive UIE.
• Figure 3: The framework of MCSS module. Given the input data $\mathbf{X}$, MCSS first unfolds input patches into sequences along multiple distinct traversal paths (i.e., multi-scale cycle scan), processes each patch sequence using a separate S6 block in parallel, and subsequently reshapes and merges the resultant sequences to form the output map (i.e., feature-merge).
• Figure 4: Visual comparison of enhancement results sampled from the test set of LSUI and UIEB dataset. The highest PSNR scores are marked in yellow. It can be seen that the enhancement results of our method are the closest to the ground truth.
• Figure 5: Visual comparison of the non-reference evaluation sampled from the U45 dataset. Compared with existing methods, our method exhibits fewer color casts and artifacts, and recovers high-frequency local details better.