基于天空检测和超像素分割的图像去雾方法  

Single image dehazing based on sky detection and superpixel segmentation

作　　者：高仁强 陈亮雄[1] 孙秀峰[1] 王欢欢 高真 GAO Renqiang;CHEN Liangxiong;SUN Xiufeng;WANG Huanhuan;GAO Zhen(National Engineering Laboratory of Estuary Hydropower Technology,Guangdong Research Institute of Water Resources and Hydropower,Guangzhou 510635,China;Guangzhou Institute of Geography,Guangdong Academy of Sciences,Guangzhou 510070,China)

机构地区：[1]广东省水利水电科学研究院河口水利技术国家地方联合工程实验室,广州510635 [2]广东省科学院广州地理研究所,广州510070

出　　处：《南京信息工程大学学报》2024年第5期630-642,共13页Journal of Nanjing University of Information Science & Technology

基　　金：广东省重点领域研发计划(2020B0101130018);广东省水利科技创新项目(2022-02,2024-08);广东省科学院专项资金(2024GDASZH-2024010101)。

摘　　要：针对经典图像去雾算法在边缘区域易产生光晕效应、天空等明亮区域还原失真、色调偏移等问题,提出一种基于天空检测和超像素分割的改进暗通道图像去雾新方法(Dark Channel Prior based on Sky Detection and Super Pixel,SSPDCP).首先对雾图采用HSV变换提取亮度分量进行自适应阈值分割;然后应用图像连通分析技术识别天空域;接着利用天空域估计大气光值,针对天空和非天空区域分别建立各自的透射率计算模型,并基于构建的超像素级透射率融合模型获得融合透射率图,以促进边界区域的平滑过渡,采用多尺度引导滤波精化透射率图;最后应用大气散射模型完成图像复原并进行亮度增强处理,实现无雾图像的自然恢复.该方法识别的天空区域较为连续完整,以超像素代替方形窗口可以有效克服局部块效应的影响,大气光值和透射率图估计更为客观准确.从主观定性和客观定量评价方面来看,该方法复原的图像具有整体误差小、信噪比优良、结构相似度高等优势.本文所提出的图像去雾新方法能有效抑制边缘区域的光晕效应,且复原的天空区域明亮自然,图像去雾质量相比现有方法有进一步提升.To address issues perplexing classic image dehazing methods,including halo effect in edge regions,color distortion in bright areas like sky,and hue shifts,we propose a novel image dehazing approach based on improved dark channel prior(SSPDCP:Dark Channel Prior based on Sky Detection and Super Pixel).This approach first ap-plies HSV color transformation to hazy images to extract the brightness component for adaptive-threshold segmenta-tion.Then it utilizes image connectivity analysis to identify the sky regions,from which the atmospheric light value is estimated,and separate transmittance maps of sky and non-sky areas are computed with a luminance model and a superpixel segmentation-based dark channel prior model,respectively.Subsequently,a superpixel-based fusion model is proposed to obtain a comprehensive transmittance map,ensuring smooth transition in boundary areas,which is fur-ther refined by multi-scale guided filtering.Finally,the dehazed image is naturally restored via the atmospheric scat-tering model and brightness enhancement processing.Experimental results show that the proposed approach identifies sky regions more continuously and completely,moreover,by employing superpixels instead of square windows,it ef-fectively mitigates halo effects in acquiring transmittance maps.The estimation of atmospheric light values and trans-mittance maps is more objective and accurate.Both subjective qualitative and objective quantitative evaluations reveal advantages such as low overall error,excellent signal-to-noise ratio,and high structural similarity in dehazed images.Compared to the state-of-the-art methods,the proposed approach restores skies more naturally,weakens halo effect in edge regions,and achieves qualitative and quantitative improvements in dehazing performance.

关 键 词：图像去雾 超像素分割 暗通道先验 天空域识别 大气散射模型 

分 类 号：TP391[自动化与计算机技术—计算机应用技术]