基于成像双曲线修正模型的水下图像增强

Underwater Image Enhancement Based on Hyperbolic Imaging Correction Model

摘要:

水下光学图像是获取水下信息最直接高效的途径，但其常因水体光吸收和散射出现严重的色彩失真与对比度下降。当前主流基于深度学习的水下图像增强方法需大量配对数据训练，且存在网络复杂、计算成本高等局限。针对上述问题，提出一种基于零参考学习的水下图像双曲线修正增强模型。该模型将增强问题转化为估计特定参数图任务：第 1 阶段（色彩增强）基于雾成像原理构建水下色彩曲线修正模型，设计轻量级网络估计其动态调整参数，以消除色散与色偏，并调整像素动态范围；第 2 阶段（亮度优化）受人眼视觉系统启发，引入非线性亮度映射与水下亮度曲线修正模型，解决前段处理导致的动态范围压缩与亮度受限问题，校正全局亮度并修正模型偏差。核心贡献在于零参考学习特性，训练完全无需任何水下配对数据集。基准数据集实验表明，提出的方法在主客观评价上均具竞争力，验证了所提双曲线修正成像框架在水下视觉任务中具有显著潜力和应用前景。

Abstract:

Underwater optical images are the most direct and efficient way to obtain underwater information， but they often suffer from severe color distortion and contrast degradation due to light absorption and scattering. Prevailing deep learning methods require extensive paired data and have limitations such as complex network architectures and high computational costs. In this paper，a hyperbolic imaging correction model for underwater image enhancement based on zero-reference learning framework is proposed. It transforms image enhancement problem into a specific parametric graph estimation task. In the color enhancement stage，a physically grounded color correction model derived from haze imaging principle is developed. A lightweight network is designed to estimate dynamic adjustment parameters and to eliminate dispersion and color cast，while optimizing pixel-level dynamic range. In the luminance optimization stage，a non-linear luminance mapping function coupled with an adaptive brightness correction model is deployed，which is inspired by the human visual system. Thus，residual dynamic range compression and illumination constraints from prior processing are rectified，global luminance is calibrated and model bias is corrected. The core contribution of this model lies in the zero-reference learning feature，and the framework operates without paired underwater data. Experiments on benchmark datasets show that the proposed method is competitive in both subjective and quantitative metrics，validating the methodological efficacy and practical viability of this physics-inspired correction framework for underwater vision.