基于暗通道先验与自适应物理模型补偿的水下图像增强方法

Underwater Image Enhancement Method Based on Dark Channel Prior and Adaptive Physical Model Compensation

摘要: 水下图像在海洋探测、资源勘查与工程运维中应用价值显著，但受水体光吸收与散射作用影响，易出现色偏、对比度低、局部照度不足等质量退化问题。本文提出一种融合暗通道先验与自适应物理模型补偿的水下图像增强方法，先依据水下光波长衰减特性与暗通道先验掩码补偿RGB通道偏差，再通过自适应亮度区间的非线性拉伸算法，实现全局和局部对比度增强，最后以最亮通道为补偿源，结合自适应暗通道掩码实现低照度区域增强。与深度学习方法相比，本算法物理建模明确、可解释性强，在UIEB数据集验证结果表明，SSIM、UIQM、UCIQE指标分别为0.8986、2.8984、0.6283，在Orin系列开发板上处理速率满足工业标准，形成了一套适用于工程应用的水下图像增强处理工具，可有效支撑潜航器下游感知任务性能提升。

Abstract: Underwater images present significant application value in marine exploration, resource exploration and engineering operation and maintenance. However, due to the absorption and scattering of light by water bodies, they are prone to quality degradation such as color cast, low contrast and insufficient local illumination. This paper proposes an underwater image enhancement method that integrates dark channel prior and adaptive physical model compensation. Firstly, the RGB channel deviation is compensated based on the attenuation characteristics of underwater light wavelengths and the dark channel prior mask. Then, a nonlinear stretching algorithm with adaptive brightness intervals is adopted to achieve global and local contrast enhancement. Finally, with the brightest channel as the compensation source, the enhancement of low-illumination regions is realized combined with the adaptive dark channel mask. Compared with deep learning-based methods, the proposed algorithm has explicit physical modeling and strong interpretability. The verification results on the UIEB dataset show that the SSIM, UIQM and UCIQE metrics are 0.8986, 2.8984 and 0.6283 respectively. The processing speed on Orin series development boards meets industrial standards, and a complete set of underwater image enhancement processing tools suitable for engineering applications is developed, which can effectively improve the performance of downstream perception tasks for underwater vehicles.