@Article{cmc.2026.085999,
AUTHOR = {Jinshuo Ma, Yang Li, Can Guo, Wen Gao, Ruiming Zhang},
TITLE = {Spectral-Semantic Decoupled Rectification Network for Non-Uniform Underwater Image Restoration},
JOURNAL = {Computers, Materials \& Continua},
VOLUME = {89},
YEAR = {2026},
NUMBER = {1},
PAGES = {0--0},
URL = {http://www.techscience.com/cmc/v89n1/68460},
ISSN = {1546-2226},
ABSTRACT = {Underwater image restoration is severely hindered by a tightly coupled degradation process: wavelength-dependent spectral distortion combined with non-uniform, multi-scale spatial scattering. Standard Convolutional Neural Networks (CNNs) and rigid physical priors frequently fail in these dynamic environments, limited by restricted receptive fields, overlooked inter-channel spectral correlations, and severe over-enhancement in photon-starved regions. To break this bottleneck, we propose the Phased Feature Rectification Network (PFR-Net), a decoupled architecture that transforms the ill-posed restoration task into a sequential global spectral calibration and deep semantic refinement paradigm. In the first phase, an efficient Multi-Layer Perceptron (MLP)-based Color Mapping (MLP-CM) module acts as a front-end calibrator to rectify global color casts with minimal parameter overhead. Following this, semantic structures are systematically reconstructed through Dilated Multi-Scale Residual Blocks (DMSRB), which are designed to prevent the truncation of delicate structural gradients and dynamically stabilize content-adaptive feature propagation. To bridge the gap between spatial processing and spectral consistency, we introduce Cross-Covariance Global Attention (XCA), capturing long-range inter-channel dependencies with a linear spatial complexity of <math id="mml-ieqn-1"><mrow><mrow><mi>&#x1D4AA;</mi></mrow></mrow><mo stretchy="false">(</mo><mi>N</mi><msup><mi>C</mi><mn>2</mn></msup><mo stretchy="false">)</mo></math>. Finally, we formulate an illumination-aware Adaptive Dark Channel Prior (DCP) loss that dynamically modulates penalty constraints based on global scene radiance, robustly suppressing backscatter while shielding dark regions from visual artifacts. Extensive zero-shot cross-dataset evaluations validate the effectiveness of our paradigm; PFR-Net achieves competitive performance, including a PSNR of 27.625 and an SSIM of 0.885 on the rigorous LSUI400 benchmark. By synergizing physically grounded modulations with decoupled feature learning, PFR-Net provides a robust and high-fidelity solution for underwater visual perception.},
DOI = {10.32604/cmc.2026.085999}
}