A Multi-Color-Space Residual Attention Network for No-Reference Underwater Image Quality Assessment

A B S T R A C T Underwater image quality assessment (UIQA) plays an important role in evaluating the performance of underwater image enhancement algorithms and supporting practical marine applications. However, due to complex factors commonly existing in underwater imaging processes, such as light absorption, scattering, and non-uniform degradation, existing no-reference image quality assessment methods often fail to produce evaluation results consistent with human subjective perception in underwater scenarios.To address this issue, this paper proposes a no-reference UIQA network termed UWMCQR-Net. Inspired by human visual perception mechanisms, the proposed method introduces a multi-color-space joint modeling strategy, which integrates RGB, luminance, and chrominance information within a unified residual learning framework to characterize complex underwater quality degradations, including luminance attenuation, structural blur, and color distortion. Meanwhile, an attention-enhanced residual module is incorporated to effectively strengthen the modeling capability of perceptually important degradation regions and discriminative quality-related features.Quantitative comparisons on benchmark datasets, including LUIQD, SAUD, and UWIQA, show that UWMCQR-Net outperforms widely used IQA and UIQA methods in terms of correlation with subjective quality scores. Furthermore, the results on unseen underwater scenarios indicate the proposed method’s generalization ability.Overall, UWMCQR-Net is able to achieve stable, accurate, and perception-consistent underwater image quality prediction while maintaining a controllable model complexity, providing an effective solution for practical UIQA tasks.