NH-HAZE

图像去雾是近年来被广泛研究的病态问题。由于缺乏参考数据集，去雾方法的客观性能评估是主要障碍之一。虽然合成数据集已显示出重要的局限性，但最近引入的少数现实数据集假设整个场景的雾度均匀。由于在许多实际情况下，雾度不是均匀分布的，我们引入了 NH-HAZE，这是一个非均匀真实数据集，具有成对的真实雾度和相应的无雾度图像。这是第一个非同质图像去雾数据集，包含 55 个户外场景。使用模拟雾霾场景真实情况的专业雾度发生器在场景中引入非均匀雾度。此外，这项工作对使用 NH-HAZE 数据集评估的几种最先进的单图像去雾方法进行了客观评估。

Image dehazing is an ill-posed problem that has been extensively studied in recent years. One of the primary obstacles to the objective performance evaluation of dehazing methods is the lack of reference datasets. Although synthetic datasets have demonstrated significant limitations, the few recently introduced real-world datasets assume uniform haze distribution across the entire scene. Since haze is often non-uniformly distributed in many practical scenarios, we introduce NH-HAZE, a non-uniform real-world dataset containing paired real hazy and corresponding haze-free images. This is the first non-homogeneous image dehazing dataset, consisting of 55 outdoor scenes. Non-uniform haze is introduced into the scenes using a professional haze generator that simulates real-world haze conditions. Additionally, this work conducts an objective evaluation of several state-of-the-art single-image dehazing methods using the NH-HAZE dataset.

搜集汇总

背景与挑战

背景概述

NH-HAZE是一个非均匀真实图像去雾数据集，包含55个户外场景的成对有雾和无雾图像，专为评估单图像去雾方法设计，由苏黎世联邦理工学院等机构于2020年发布。