A Multi-Scale Underwater Laser Image Restoration Method with Polarization Feature Constraints

Abstract

Underwater laser imaging is widely used for deep-sea exploration, autonomous underwater navigation, and inspection of marine infrastructure, where high-precision observation under optically challenging conditions is required. These imaging systems are inherently limited by absorption attenuation, volume scattering, and backscattered noise, leading to reduced visibility, low contrast, and loss of structural details. In this study, we propose a Polarization-Constrained Multi-Scale Defogging and Restoration (PCMS-DR) algorithm designed for turbid and heterogeneous aquatic environments, specifically targeting submerged engineered structures, pipelines, and other objects of interest. The method integrates polarization feature constraints with multi-scale decomposition, enabling robust separation of backscattered light and target-reflected signals while preserving high-frequency structural information. To quantitatively evaluate the proposed framework, two complementary validation strategies are adopted. First, polarization-resolved Monte Carlo photon propagation simulations are conducted to generate physically consistent synthetic underwater laser images for controlled analysis under different scattering conditions. Second, real-world validation is performed using 12 self-acquired coastal underwater laser imaging scenes collected under representative aquatic environments. The simulation and experimental datasets are analyzed separately to ensure that the quantitative evaluation accurately reflects both physical restoration capability and practical applicability. Quantitative results from the Monte Carlo simulation experiments demonstrate that the proposed PCMS-DR method achieves a peak PSNR of 26.52 dB and an SSIM of 0.836 under representative low-turbidity conditions, outperforming polarization-only and multi-scale-only baselines. In addition, evaluation on the self-acquired real underwater laser imaging dataset containing 12 coastal scenes indicates an average backscatter suppression ratio of 22.3% and a local contrast enhancement ratio of 1.62. These results confirm that the proposed method improves image visibility and structural preservation across both controlled simulations and practical imaging scenarios. Furthermore, evaluation on 12 real coastal underwater laser imaging scenes demonstrates an average backscatter suppression ratio of 22.3% and a local contrast enhancement ratio of 1.62, indicating improved visibility, contrast, and structural fidelity. The principal novelty of the proposed framework lies in the unified integration of polarization-constrained backscatter modeling, multi-scale transmission estimation, and physically guided detail restoration within a single optimization framework. The experimental results demonstrate that the proposed PCMS-DR framework provides physically interpretable and effective restoration performance under the tested turbidity range for underwater laser imaging applications.