PIGUIQA: A Physical Imaging Guided Perceptual Framework for Underwater Image Quality Assessment

TL;DR

The paper tackles underwater image quality assessment by integrating a physics-based imaging model with deep perceptual networks. It explicitly models direct transmission attenuation and backward scattering, and uses patchwise distortion metrics alongside a neighborhood-attention based local module and a ResNet50-based global aggregator to predict quality scores. Across SAUD2.0, UID2021, and UWIQA, PIGUIQA achieves state-of-the-art correlations with MOS and demonstrates strong cross-dataset generalization, attributed to the physics-informed design. The work provides a practical, open-source framework for reliable UIQA that can guide underwater image enhancement workflows.

Abstract

In this paper, we propose a Physical Imaging Guided perceptual framework for Underwater Image Quality Assessment (UIQA), termed PIGUIQA. First, we formulate UIQA as a comprehensive problem that considers the combined effects of direct transmission attenuation and backward scattering on image perception. By leveraging underwater radiative transfer theory, we systematically integrate physics-based imaging estimations to establish quantitative metrics for these distortions. Second, recognizing spatial variations in image content significance and human perceptual sensitivity to distortions, we design a module built upon a neighborhood attention mechanism for local perception of images. This module effectively captures subtle features in images, thereby enhancing the adaptive perception of distortions on the basis of local information. Third, by employing a global perceptual aggregator that further integrates holistic image scene with underwater distortion information, the proposed model accurately predicts image quality scores. Extensive experiments across multiple benchmarks demonstrate that PIGUIQA achieves state-of-the-art performance while maintaining robust cross-dataset generalizability. The implementation is publicly available at https://github.com/WeizhiXian/PIGUIQA

Figures (8)

• Figure 1: Illustration of the underwater optical imaging model with resulting major distortions. The scattering effects of light in water, caused by visible suspended particles and dissolved organic matter, result in varying degrees of deviation in the light transmission path. This alteration in the energy distribution of the light beam significantly degrades the quality of underwater images. The distorted images, which are sampled from various datasets, typically exhibit multiple types of distortions.
• Figure 2: Overall structure of the proposed UIQA method. The symbols "$\otimes$" and "Ⓒ" denote the Hadamard product and concatenation operations, respectively.
• Figure 3: Illustration of underwater imaging estimation via Eq. (\ref{['eq:syreanet']}). (a): Input distorted underwater images $I_{dis}$. (b): Estimated transmission attenuation map $\hat{T}$. (c): Estimated background clutter map $\hat{B}$. (d): Restored underwater images $\hat{I}_{per}$. In the transmission attenuation map, darker areas signify more severe scattering and greater distortion, whereas in the background clutter map, brighter areas indicate a greater degree of background noise.
• Figure 4: The structure of the local perceptual module. It utilizes the NA mechanism, a technique that combines local inductive biases with translational invariance, to effectively aggregate pixel features and their surrounding neighborhood. This module enables a detailed analysis of each pixel in relation to its immediate environment.
• Figure 5: Scatter plots of the quality scores predicted by the proposed model against the MOS. The red curves are the 5PL fitting functions.