有效地实现单幅水下降质图像复原对水下资源探索及环境监控领域的清晰图像获取具有极其重要的意义。为解决常用暗通道先验方法来复原图像时, 背景光的估计易受白色物体干扰, 且无法有效估计前景中白色物体透射率, 复原质量不高的问题。本文提出了自适应背景光估计与非局部先验的水下图像复原算法。首先根据背景光具有高亮度及平坦性的特点, 利用阈值分割算法获得背景光的候选区域, 再通过图像的色调信息从候选信息中选取最佳的背景光点。随后, 利用各颜色通道光的波长与散射系数的相关性, 提出了适用于水下图像的非局部先验, 并利用该先验估计各通道的透射率。最后针对复原结果中, 因水下介质, 微生物, 水流影响而产生的加性噪声, 设计去噪的最小优化问题, 并利用引导滤波求解该问题, 以去除复原结果中的加性噪声。实验表明: 该算法在确保运行效率的基础上, 准确地估计透射率, 较常用算法的复原精度提高了约18%。证明了该算法能有效用于单幅水下图像复原的工程实践中。

It is significant to realize effective single underwater image restoration for acquiring clear image in underwater exploration and underwater environment monitoring field. Most existing algorithms use dark channel priors to restore images, which lead to inaccurate estimates of the background light and transmission map. Hence, a novel method with adaptive background light estimation and nonlocal prior was proposed. Firstly, the candidate water light regions could be obtained by a threshold segmentation algorithm owing to the fact that water light regions have the properties of flat and high brightness. Then, the water light value could be decided from the candidate regions by the dominant tone of the input image. Secondly, the nonlocal prior was built to estimate the transmission map by taking into account the wavelength dependence of the attenuation. Finally, in order to remove the additive noise from the medium and microorganisms, a minimal optimization problem with the solution strategy of guided filter was proposed for obtaining the de-noising result. The experimental results verify that the proposed algorithm not only ensures operation efficiency, but can also estimate the correct transmission map. In general, the restoration precision has improved by 18% compared with the existing algorithm. It can be used in the engineering practice of restoring a single underwater image.