预览加载中，请您耐心等待几秒...

基于暗原色先验的快速单幅图像去雾算法.docx

预览
1/2
2/2

在线预览结束，喜欢就下载吧，查找使用更方便

下载提示 文本预览

如果您无法下载资料，请参考说明：

1、部分资料下载需要金币，请确保您的账户上有足够的金币

2、已购买过的文档，再次下载不重复扣费

3、资料包下载后请先用软件解压，在使用对应软件打开

基于暗原色先验的快速单幅图像去雾算法 摘要 图像去雾一直是计算机视觉领域中一个重要的研究方向，涉及到许多实际应用。本文提出了一种基于暗原色先验的快速单幅图像去雾算法。该方法通过对图像中较暗区域进行预测，去除雾霾影响，提高图像视觉质量。实验结果表明，该方法在去雾效果和速度上均有显著提升。 关键词：图像去雾，暗原色，预测，影响 Introduction Inrecentyears,imagedehazinghasbeenachallengingtaskincomputervision.Inmanypracticalapplications,suchasremotesensing,surveillance,andoutdoorphotography,imageclarityishamperedbyfog,haze,orsmog.Removingthesefactors,andenhancingthevisibilityofoutdoorsceneshasbecomeanimportantresearcharea.Inthispaper,wepresentanovelfastsingle-imagedehazingalgorithmbasedondarkchannelprior. Background Thetraditionalapproachtosingle-imagedehazingisbasedontheassumptionthatthetransmissionofthehazyimagedecreasesexponentiallywiththedepth.However,inreal-worldscenes,thehazeisnotdistributedevenly,andthistheoryisnotapplicableinmanysituations.Therefore,manyresearchershaveproposedalternativemethodstoovercomethisproblem.Onesuchmethodistousethedarkchannelprior,whichproposedbyKaimingHe. Methodology Theproposedmethodconsistsofthreesteps:predictingthedarkchannel,estimatingthetransmissionmap,andgeneratingthedehazedimage.Inthefirststep,thedarkchannelispredictedbyfindingtheminimumpixelvalueinalocalwindow.Then,thetransmissionmapisestimatedbysolvingalinearequationusingthedarkchannelandtheoriginalimage.Finally,thedehazedimageisgeneratedbydividingtheoriginalimagebytheestimatedtransmissionmap. Toimprovethespeedofthealgorithm,weusetheguidedfilterinthesecondstep.Theguidedfilterisafastandefficientimageprocessingtechniquethatpreservestheedgesandtexturesintheimage.Byusingtheguidedfilter,wecanestimatethetransmissionmapmoreaccuratelyandefficiently. ExperimentalResults Theproposedmethodwascomparedwithseveralstate-of-the-artmethodsusingbenchmarkdatasets.TheexperimentswereconductedonaworkstationwithanInteli7processorand16GBofRAM.Theresultsshowthatourmethodoutperformsothermethodsinvisualqualityandspeed. Conclusion Inthispaper,wepresentedanovelfastsingle-imagedehazingalgorithmbasedondarkchannelprior.Theproposedmethodusestheguidedfiltertoestimatethetrans