Abstract

As an important underlying visual processing task, underwater image enhancement techniques have received a lot of attention from researchers due to their importance in marine engineering and lake ecosystem optimization. However, various underwater image enhancement algorithms have been proposed to be evaluated mainly with marine water body datasets, and it is not clear whether these algorithms can be performed on datasets collected from inland lakes in the freshwaters. To bridge this gap, for the first time, we construct an underwater image dataset for highland and freshwater-circumstances (HFUI) using 1000 real images to complement the underwater image datasets. In addition, we propose an unsupervised underwater image enhancement algorithm (HUFI-Net) specifically for this dataset to correct the sharpness and color of the images. This algorithm, as well as current advanced underwater image enhancement algorithms, was investigated qualitatively and quantitatively using this dataset to evaluate the effectiveness and limitations of various algorithms and to provide novel ideas for future underwater image enhancement research. We also further validate the generalization and effectiveness of the algorithm on the underwater image datasets of the UIEB and the RUIE.

中文翻译：

---

一种新颖的高原和淡水环境数据集：推进水下图像增强

摘要

作为一项重要的底层视觉处理任务，水下图像增强技术因其在海洋工程和湖泊生态系统优化中的重要性而受到了研究人员的广泛关注。然而，已经提出的各种水下图像增强算法主要针对海洋水体数据集进行评估，并且尚不清楚这些算法是否可以在从淡水内陆湖泊收集的数据集上执行。为了弥补这一差距，我们首次使用 1000 张真实图像构建了高原和淡水环境水下图像数据集（HFUI）来补充水下图像数据集。此外，我们专门针对该数据集提出了一种无监督水下图像增强算法（HUFI-Net）来校正图像的清晰度和颜色。利用该数据集对该算法以及当前先进的水下图像增强算法进行了定性和定量研究，以评估各种算法的有效性和局限性，为未来水下图像增强研究提供新思路。我们还在UIEB和RUIE的水下图像数据集上进一步验证了该算法的泛化性和有效性。