DOI:10.16136/j.joel.2023.02.0231基于非局部先验的雾线优化图像去雾算法聂新蕾1,张长胜1*,钱俊兵2(1.昆明理工大学信息工程与自动化学院,云南昆明650500;2.昆明理工大学民航与航空学院,云南昆明650500)摘要:针对非局部先验去雾算法中雾线端点像素位置精确度不足的问题,提出了雾线优化的非局部先验图像去雾算法。首先分析雾线理论,结合暗通道理论确定最大聚类雾线真实端点,以其为已知条件补偿小聚类雾线端点与大气光之间的距离,根据类内不同像素与雾线对应夹角预估单个像素雾线端点进而求得像素级优化后的透射率,最后根据图像局部灰度值差异融合暗通道先验(darkchannelprior,DCP)和非局部先验透射率得最终透射率图。将本文算法与其余3种去雾算法在多幅户外雾图下通过主观及客观两方面分析比较,实验结果表明该算法能取得更好的去雾效果,尤其在天空区域图像复原效果较为突出。关键词:图像去雾;暗通道先验(DCP);非局部先验;雾线优化;雾线端点中图分类号:TP391文献标识码:A文章编号:1005-0086(2023)02-0140-07Haze-lineoptimizationimagedehazingalgorithmbasedonnon-lo-calpriorNIEXinlei1,ZHANGChangsheng1*,QIANJunbing2(1.CollegeofInformationEngineeringandAutomation,KunmingUniversityofScienceandTechnology,Kun-ming,Yunnan650500,China;2.CollegeofCivilAviationandAeronautics,KunmingUniversityofScienceandTechnology,Kunming,Yunnan650500,China)Abstract:Thepixelpositionofthehaze-lineendpointisnotaccurateenoughinthenon-localpriordehaz-ingalgorithm.Toresolvethisproblem,animagedehazingalgorithmbasedonnon-localpriorwithanop-timizedhaze-linewasproposedinthisstudy.Weanalyzedhaze-linetheoryandcombinedthedarkchan-neltheorytofindtherealhaze-lineendpointofthelargestcluster.Then,wetookitastheknowncondi-tionstocompensatethemaximumdistancebetweenotherhaze-lineendpointsofsmallclusterandtheat-mosphericlight.Accordingtothedifferentpixelsintheclassandthecorrespondingangles,thehaze-lineendpointofindividualpixelwasestimated,andthen,thetransmissionofeverypixelafteroptimizationwasrefined.Finally,localgreyvaluedifferencefusiondarkchannelprior(DCP)andnon-localpriortransmissionwasusedtoproduceourtransmissionmap.Wecomparedouralgorithmwiththreeexistingalgorithmsbyapplyingthemtomultip...
