VRS-UIE: Value-Driven Reordering Scanning for Underwater Image Enhancement

TL;DR

This work tackles underwater image enhancement by rethinking state-space scanning: a pixelwise value map guides the scanning order to prioritize informative regions, enabling purer state propagation for color and structure reconstruction. The core contributions are MVGL, which combines implicit self-similarity with an explicit DINOv2 prior to produce a value map; the Mamba–Conv Mixer (MCM), a dual-path block that couples value-driven global sequencing with adaptive local filtering; and the Cross-Feature Bridge (CFB) for robust multi-level fusion. Together, they yield a new state-of-the-art on FR and NR underwater benchmarks, with a lightweight variant VRS-UIE-S suitable for real-time deployment. The approach demonstrates substantial gains in color fidelity and structural preservation while mitigating water bias, offering a practical, explainable improvement for underwater vision systems.

Abstract

State Space Models (SSMs) have emerged as a promising backbone for vision tasks due to their linear complexity and global receptive field. However, in the context of Underwater Image Enhancement (UIE), the standard sequential scanning mechanism is fundamentally challenged by the unique statistical distribution characteristics of underwater scenes. The predominance of large-portion, homogeneous but useless oceanic backgrounds can dilute the feature representation responses of sparse yet valuable targets, thereby impeding effective state propagation and compromising the model's ability to preserve both local semantics and global structure. To address this limitation, we propose a novel Value-Driven Reordering Scanning framework for UIE, termed VRS-UIE. Its core innovation is a Multi-Granularity Value Guidance Learning (MVGL) module that generates a pixel-aligned value map to dynamically reorder the SSM's scanning sequence. This prioritizes informative regions to facilitate the long-range state propagation of salient features. Building upon the MVGL, we design a Mamba-Conv Mixer (MCM) block that synergistically integrates priority-driven global sequencing with dynamically adjusted local convolutions, thereby effectively modeling both large-portion oceanic backgrounds and high-value semantic targets. A Cross-Feature Bridge (CFB) further refines multi-level feature fusion. Extensive experiments demonstrate that our VRS-UIE framework sets a new state-of-the-art, delivering superior enhancement performance (surpassing WMamba by 0.89 dB on average) by effectively suppressing water bias and preserving structural and color fidelity. Furthermore, by incorporating efficient convolutional operators and resolution rescaling, we construct a light-weight yet effective scheme, VRS-UIE-S, suitable for real-time UIE applications.

Figures (9)

• Figure 1: (a) Distribution characteristics of underwater images: informative regions are sparse but valuable (green boxes), while the oceanic backgrounds is large-portion, homogeneous and useless (red boxes). (b) When an SSM uses a (1) fixed scan , valuable pixels are dispersed along the sequence and their feature responses are barely activated and diluted by useless oceanic backgrounds during long-range propagation. (2) MVGL (ours) applies a value-priority schedule: a learnable value map reorders tokens so high-value pixels are prioritized and positioned together, enabling purer and more efficient state propagation.
• Figure 2: An overview of the proposed Value-Driven Reordering Scanning framework (VRS-UIE). (a) The overall pipeline, which is based on a U-Net architecture with MCM blocks at each stage. (b) The Multi-Granularity Value Guidance Learning (MVGL) module. (c) The Mixer module within the MCM block, featuring a dual-path design with a reordering Mamba path and a dynamic convolution, which are designed to explore contextual information from regions based on semantic relevance or structural significance.
• Figure 3: Illustration of the proposed Cross-level Feature Bridging (CFB) module.
• Figure 4: Full reference comparison on different scenes. For each scene we report PSNR/SSIM (higher is better) with respect to the reference shown in the last column. The curve plot displays an image histogram, with a bar graph in the upper-right corner showing the average pixel value for each channel.
• Figure 5: Restoration Under Extreme Degradation. For each scene we report PSNR/SSIM (higher is better) with respect to the reference shown in the last column.