基于多尺度引导滤波的实时视频去雾算法  

Real Time Video Defogging Algorithm Based on Multi⁃Scale Guided Filtering

作　　者：宁贝 杨明[1] NING Bei;YANG Ming(School of Mathematics,North University of China,Taiyuan 030051,China)

机构地区：[1]中北大学数学学院,山西太原030051

出　　处：《中北大学学报（自然科学版）》2024年第4期439-447,共9页Journal of North University of China(Natural Science Edition)

摘　　要：为解决当前去雾算法中存在的透射率估计不准确、天空区域颜色恢复较差、运行速度较慢等问题,本文提出了一种基于多尺度引导滤波的实时视频去雾算法。首先,采用自动白平衡算法对有雾图像进行颜色校正,颜色校正后的图像与暗原色置信因子作为引导滤波,随后使用金字塔采样技术获取缩小后的图像,接着使用四叉树算法来估算透射率和大气光强度,并且不断地迭代上采样和引导图像滤波,从而有效防止了信息的损失,最后达到最佳的传输效果。此外,将单图像去雾算法扩展到了实时视频去雾,通过使传输值在时间上一致来减少去雾视频中的闪烁伪影。实验结果表明,该算法的运行速率较快,去雾效果明显,在视频去雾中闪烁伪影较少。相比于屏蔽泊松方程去雾算法,本文算法的平均梯度提高了67.08%,相比于暗通道先验逐帧视频去雾算法,本算法的时间加快了81.32%,符合实时视频运行稳定、处理快速的要求。In order to address the issues of inaccurate transmission estimation,poor color restoration in sky regions,and slow operation speed in current defogging algorithms,this paper proposed a real-time video defogging algorithm based on multi-scale guided filtering.First,the automatic white balance algorithm was used to correct the color of the foggy image.The confidence factor of the color corrected image and the dark primary color were used as the guiding filter.Then,the pyramid sampling technology was used to obtain the reduced image.Then,the quadtree method was used to estimate the transmittance and atmospheric light intensity,and the iterative up sampling and guiding image filtering were continuously used to effectively prevent the loss of information and finally achieve the best transmission effect.In addition,we extended the single image defogging algorithm to real-time video defogging,reducing flickering artifacts in defogging videos by ensuring consistent transmission values over time.The experimental results show that the algorithm has a fast running speed,significant defogging effect,and less flickering artifacts in video defogging. Compared with the shielded Poisson equation defogging algorithm, this algorithmhas faster time 81. 32% and higher average gradient 67. 08%, which meets the requirements of fastand stable real-time video processing.

关 键 词：暗通道先验 导向滤波 四叉树 金字塔分层 自动白平衡 

分 类 号：TP391.41[自动化与计算机技术—计算机应用技术]