[論文レビュー] Development of Domain-Invariant Visual Enhancement and Restoration (DIVER) Approach for Underwater Images

Rajini Makam, Sharanya Patil+3 more|arXiv (Cornell University)|Jan 30, 2026

Image Enhancement Techniques被引用数 0

ひとこと要約

DIVER は、実証補正と物理ガイド付きモデルを組み合わせて、さまざまな水質と照明条件下で水中画像を向上・復元する無監督のドメイン不変アーキテクチャです。基準手法を上回り、知覚的一貫性と放射測定的一貫性を改善します。

ABSTRACT

Underwater images suffer severe degradation due to wavelength-dependent attenuation, scattering, and illumination non-uniformity that vary across water types and depths. We propose an unsupervised Domain-Invariant Visual Enhancement and Restoration (DIVER) framework that integrates empirical correction with physics-guided modeling for robust underwater image enhancement. DIVER first applies either IlluminateNet for adaptive luminance enhancement or a Spectral Equalization Filter for spectral normalization. An Adaptive Optical Correction Module then refines hue and contrast using channel-adaptive filtering, while Hydro-OpticNet employs physics-constrained learning to compensate for backscatter and wavelength-dependent attenuation. The parameters of IlluminateNet and Hydro-OpticNet are optimized via unsupervised learning using a composite loss function. DIVER is evaluated on eight diverse datasets covering shallow, deep, and highly turbid environments, including both naturally low-light and artificially illuminated scenes, using reference and non-reference metrics. While state-of-the-art methods such as WaterNet, UDNet, and Phaseformer perform reasonably in shallow water, their performance degrades in deep, unevenly illuminated, or artificially lit conditions. In contrast, DIVER consistently achieves best or near-best performance across all datasets, demonstrating strong domain-invariant capability. DIVER yields at least a 9% improvement over SOTA methods in UCIQE. On the low-light SeaThru dataset, where color-palette references enable direct evaluation of color restoration, DIVER achieves at least a 4.9% reduction in GPMAE compared to existing methods. Beyond visual quality, DIVER also improves robotic perception by enhancing ORB-based keypoint repeatability and matching performance, confirming its robustness across diverse underwater environments.

研究の動機と目的

• Motivate robust underwater image enhancement across diverse water types, depths, and turbidity without ground-truth data.
• Develop an unsupervised framework that generalizes across shallow, deep, and turbid environments.
• Integrate empirical corrections with physics-guided modeling for radiometrically consistent restoration.

提案手法

• Use illumination assessment to route inputs to IlluminateNet for low-light enhancement or to the Spectral Equalization Filter for spectral normalization.
• Apply Adaptive Optical Correction Module to improve hue, contrast, and suppress chromatic speckle artifacts.
• Employ Hydro-OpticNet (VeilNet and AttenNet) to perform physics-guided backscatter removal and wavelength-dependent attenuation compensation.
• Train Learnable parameters via unsupervised losses including L_G, L_L, L_B, L_C, and L_S for the respective modules.
• Model backscatter U_B with depth-dependent functions and estimate depth with a transformer-based monocular depth model (DepthAnythingV2).
• Represent attenuation through an inverse mapping alpha_A(z) to achieve radiometric restoration.

Figure 1: Illustration of the underwater image formation process. As natural light penetrates the water column, shorter wavelengths (blue/green) dominate due to wavelength-dependent absorption, while scattering from suspended particles introduces haze and color distortion. The resulting captured ima

実験結果

リサーチクエスチョン

• RQ1 Can an unsupervised, domain-invariant architecture restore underwater images across shallow, deep, and turbid waters without ground-truth data?
• RQ2 How does combining empirical corrections with physics-guided modules affect color fidelity, contrast, and radiometric consistency under diverse conditions?
• RQ3 Do depth-aware backscatter removal and wavelength-dependent attenuation compensation improve perceptual quality and downstream perception tasks?
• RQ4 What are the comparative gains over state-of-the-art methods on standard underwater datasets using both reference and no-reference metrics?
• RQ5 Is the approach robust to illumination variations, turbidity, and artificial lighting across datasets?

主な発見

• DIVER は、参照指標（PSNR, SSIM）とノンリファレンス指標（UCIQE, UIQM, BRISQUE, GPMAE）で、8つの多様な水中データセットにおいて最高またはほぼ最高の性能を達成します。
• UCIQE は従来手法を少なくとも22%上回り、UDNetを34%上回します。
• SeaThru の低照度データセットでは、GPMAE が SOTA 手法より少なくとも4.9%改善します。
• DIVER は ORB デスクリプタ利用時にキーポイント再現性とマッチ精度を向上させ、ロボティック知覚の改善を示します。
• DIVER はドメイン不変の性能を維持し、WaterNet、UDNet、Phaseformer、UDCP などの方法を、困難な照明と濁度条件下で上回ります。

Figure 2: The proposed DIVER architecture for domain-invariant underwater image enhancement. An illumination assessment first determines whether the input image is low-lit. Low-light inputs are routed to the CNN-based IlluminateNet and optimized using the losses $\mathcal{L}_{G}$ and $\mathcal{L}_{L