AquaDiff: Diffusion-Based Underwater Image Enhancement for Addressing Color Distortion

TL;DR

Underwater images suffer severe color distortion from wavelength-dependent absorption and scattering, hindering vision-based tasks. AquaDiff introduces a diffusion-based underwater image enhancement framework that uses chromatic prior-guided channel compensation and cross-attention conditioning to align color with preserved structure, aided by an enhanced U-Net backbone and a cross-domain consistency loss. The method achieves superior color fidelity and competitive perceptual quality on diverse benchmarks, validated through extensive quantitative and qualitative experiments and ablation studies. This work demonstrates the practicality and robustness of diffusion-based UIE with underwater priors for real-world underwater perception systems.

Abstract

Underwater images are severely degraded by wavelength-dependent light absorption and scattering, resulting in color distortion, low contrast, and loss of fine details that hinder vision-based underwater applications. To address these challenges, we propose AquaDiff, a diffusion-based underwater image enhancement framework designed to correct chromatic distortions while preserving structural and perceptual fidelity. AquaDiff integrates a chromatic prior-guided color compensation strategy with a conditional diffusion process, where cross-attention dynamically fuses degraded inputs and noisy latent states at each denoising step. An enhanced denoising backbone with residual dense blocks and multi-resolution attention captures both global color context and local details. Furthermore, a novel cross-domain consistency loss jointly enforces pixel-level accuracy, perceptual similarity, structural integrity, and frequency-domain fidelity. Extensive experiments on multiple challenging underwater benchmarks demonstrate that AquaDiff provides good results as compared to the state-of-the-art traditional, CNN-, GAN-, and diffusion-based methods, achieving superior color correction and competitive overall image quality across diverse underwater conditions.

Figures (6)

• Figure 1: Overview of the proposed AquaDiff framework.The forward diffusion process progressively corrupts the clean reference image $x_0$ into a noisy latent representation $x_t$. During the reverse diffusion process, a denoiser receives the noisy image $x_t$, the chromatic prior--guided conditioning image $y$, and the diffusion timestep $t$, and predicts the noise component $\hat{\varepsilon}_t$.Cross-attention conditioning, denoted as $\otimes\,\mathrm{Cross\hbox{-}Att}(x_t, y)$, is employed to effectively fuse structural information from $x_t$ with chromatic prior guidance from $y$. The predicted noise is then used in the diffusion sampling step to obtain the refined image $x_{t-1}$. This iterative denoising process is repeated across timesteps to progressively recover the enhanced underwater image.
• Figure 2: Quantitative comparison of underwater image enhancement methods across the U45, S16, and C60 datasets. The radar plot illustrates performance trends, while the bar charts present absolute UIQM and UCIQE scores. AquaDiff consistently achieves high values, highlighting its capability in restoring color fidelity, contrast, and overall visual quality.
• Figure 3: Qualitative comparison of our proposed method AquaDiff with traditional methods namely UDCP drews2013transmission, UIBLA peng2017underwater and deep learning-based methods namely UWCNN li2020underwater, Water-Net li2019underwater, Ucolor li2021underwater, MLFcGAN liu2019mlfcgan, FUnIEGAN islam2020fast, Shallow-uwnet naik2021shallow, DiffWater guan2023diffwater for underwater image enhancment on the U-90 dataset.
• Figure 4: Qualitative comparison of our proposed method AquaDiff with traditional methods namely UDCP drews2013transmission, UIBLA peng2017underwater and deep learning-based methods namely UWCNN li2020underwater, Water-Net li2019underwater, Ucolor li2021underwater, MLFcGAN liu2019mlfcgan, FUnIEGAN islam2020fast, Shallow-uwnet naik2021shallow, DiffWater guan2023diffwater for underwater image enhancement on the U-45 dataset.
• Figure 5: Qualitative comparison of our proposed method AquaDiff with traditional methods namely UDCP drews2013transmission, UIBLA peng2017underwater and deep learning-based methods namely UWCNN li2020underwater, Water-Net li2019underwater, Ucolor li2021underwater, MLFcGAN liu2019mlfcgan, FUnIEGAN islam2020fast, Shallow-uwnet naik2021shallow, DiffWater guan2023diffwater for underwater image enhancement on the C60 dataset.