Dai Mingrui, Shi Weifeng, Li Guohua
Institute of Computing Technology, China Academy of Railway Sciences Corporation Limited, Beijing, China.
PLoS One. 2025 Jun 2;20(6):e0323536. doi: 10.1371/journal.pone.0323536. eCollection 2025.
Although the application of image-based fog density estimation brings excellent convenience and low-cost methods, the accuracy of such methods still needs to be improved, and further research is encouraged on accuracy evaluation methods. To improve the accuracy and computational efficiency of fog density estimation in images, we first construct three image features based on the image dark channel information, the image saturation information, and the proportion of gray noise points, respectively. Then, we use a feature fusion method to estimate fog density in the images. In addition, two indicators have been constructed to evaluate the accuracy of various fog density estimation methods. These two indicators are the sequential error indicator and the proportional error indicator, which are calculated using fog image sequences with known density values. These two new indicators enable the evaluation of any fog density estimation method in terms of the ability to maintain order and ratio values. The experimental results show that the proposed method can effectively estimate the fog densities of images and display the best performance among the eight latest image-based methods for estimating fog density; the three features used in the proposed method significantly impact the effectiveness of image-based fog density estimation. The proposed method has been illustrated for fog density analysis of indoor and outdoor surveillance videos. The source code is available at https://github.com/Dai-MR/ImageFogDensityEsitmation.
尽管基于图像的雾密度估计方法带来了极大的便利和低成本的方式，但此类方法的准确性仍有待提高，鼓励对准确性评估方法进行进一步研究。为提高图像中雾密度估计的准确性和计算效率，我们首先分别基于图像暗通道信息、图像饱和度信息和灰度噪声点比例构建了三个图像特征。然后，我们使用一种特征融合方法来估计图像中的雾密度。此外，还构建了两个指标来评估各种雾密度估计方法的准确性。这两个指标是顺序误差指标和比例误差指标，它们是使用具有已知密度值的雾图像序列计算得出的。这两个新指标能够从保持顺序和比值的能力方面评估任何雾密度估计方法。实验结果表明，所提出的方法能够有效地估计图像的雾密度，并且在所比较的八种最新的基于图像的雾密度估计方法中表现最佳；所提出的方法中使用的三个特征对基于图像的雾密度估计的有效性有显著影响。所提出的方法已用于室内和室外监控视频的雾密度分析。源代码可在https://github.com/Dai-MR/ImageFogDensityEsitmation获取。
