A Comprehensive Synthetic Underwater Image Dataset

The dataset contains 100 photographic images that are treated as ground-truth images. On each ground truth image, the effects of underwater environment are applied and 150 synthetic underwater images are generated. Hence, the data set contains 100 ground-truth photographic images and 100*150=15000 synthetic underwater images. Four effects of the underwater environment, i.e., color cast, blurring, contrast reduction, and low light, are considered. These effects are applied individually and in combinations. Four effects result in a total of 15 combinations, and the effect of each combination is varied by considering 10 levels. This results in a total of 150 images for a single ground-truth image. In addition to this, 21 focus metrics are evaluated on all these 1,50,100 images. The metrics calculated are Absolute central moment (ACMO), Brenner's focus measure (BREN), Image curvature (CURV), Gray-level variance (GLVA), Gray-level local variance (GLLV), Gray-level variance normalized (GLVN), Squared gradient (GRAS), Helmli's measure (HELM), Histogram entropy (HISE), Histogram range (HISR), Energy of Laplacian (LAPE), Diagonal Laplacian (LAPD), Modified Laplacian (LAPM), Variance of Laplacian (LAPV), Tenengrad variance (TENV), Vollat's correlation (VOLA), Wavelet ratio (WAVR), Wavelet sum (WAVS), and Wavelet variance (WAVV). The literature categorizes these metrics as gradient-based and non-gradient-based.

本数据集包含100张作为基准真值图像（ground-truth images）的摄影图像。针对每张基准真值图像，通过施加水下环境干扰效果，可生成150张合成水下图像。因此本数据集共包含100张基准真值摄影图像与100×150=15000张合成水下图像。本次实验共考虑四类水下环境干扰因素，分别为色偏（color cast）、模糊（blurring）、对比度降低（contrast reduction）以及低光照（low light）。这些干扰效果既会被单独施加，也会以组合形式施加。四类干扰因素共计可组合出15种干扰方案，且每种组合的干扰强度会通过10个等级进行调节，由此单张基准真值图像可生成150张合成水下图像。除此之外，研究人员会对全部15000张图像计算21项对焦评价指标，具体包括：绝对中心矩（Absolute central moment, ACMO）、布伦纳对焦测度（Brenner's focus measure, BREN）、图像曲率（Image curvature, CURV）、灰度方差（Gray-level variance, GLVA）、灰度局部方差（Gray-level local variance, GLLV）、归一化灰度方差（Gray-level variance normalized, GLVN）、平方梯度（Squared gradient, GRAS）、赫尔姆利测度（Helmli's measure, HELM）、直方图熵（Histogram entropy, HISE）、直方图极差（Histogram range, HISR）、拉普拉斯能量（Energy of Laplacian, LAPE）、对角拉普拉斯（Diagonal Laplacian, LAPD）、改进型拉普拉斯（Modified Laplacian, LAPM）、拉普拉斯方差（Variance of Laplacian, LAPV）、坦南梯度方差（Tenengrad variance, TENV）、沃拉特相关性（Vollat's correlation, VOLA）、小波比值（Wavelet ratio, WAVR）、小波和（Wavelet sum, WAVS）以及小波方差（Wavelet variance, WAVV）。现有学术文献将上述指标划分为基于梯度的对焦指标与非基于梯度的对焦指标两类。

搜集汇总

背景与挑战

背景概述

该数据集是一个综合性的合成水下图像数据集，包含100张真实照片和15000张合成图像，模拟了水下环境的多种视觉效果，并提供了详细的聚焦度量和统计指标评估。