A Global Sampling Method for Alpha Matting

A Global Sampling Method for Alpha Matting

ID:41634342

大小:1.91 MB

页数:8页

时间:2019-08-29

A Global Sampling Method for Alpha Matting_第1页
A Global Sampling Method for Alpha Matting_第2页
A Global Sampling Method for Alpha Matting_第3页
A Global Sampling Method for Alpha Matting_第4页
A Global Sampling Method for Alpha Matting_第5页
资源描述:

《A Global Sampling Method for Alpha Matting》由会员上传分享,免费在线阅读,更多相关内容在学术论文-天天文库

1、AGlobalSamplingMethodforAlphaMattingKaimingHe1,ChristophRhemann2,CarstenRother3,XiaoouTang1,JianSun412TheChineseUniversityofHongKong,ViennaUniversityofTechnology,34MicrosoftResearchCambridge,MicrosoftResearchAsiaAbstractExistingmattingmethodscanbecategorizedaspropagation-based

2、orsampling-based.Propagation-basedAlphamattingreferstotheproblemofsoftlyextractingmethodstreattheproblemasinterpolatingtheunknownal-theforegroundfromanimage.Givenatrimap(specify-phavaluesfromtheknownregions.Theinterpolationcaningknownforeground/backgroundandunknownpixels),abed

3、onebysolvinganaffinitymatrix[17,8,11,16,9],bystraightforwardwaytocomputethealphavalueistosam-optimizingMarkovRandomFields[18],orbycomputingplesomeknownforegroundandbackgroundcolorsforeachgeodesicdistance[2].Thesemethodsmainlyrelyontheunknownpixel.Existingsampling-basedmattingme

4、thodsimage’scontinuitytoestimatethealphamatte,anddonotoftencollectsamplesneartheunknownpixelsonly.Theyexplicitlyaccountfortheforegroundandbackgroundcol-failifgoodsamplescannotbefoundnearby.ors.Theyhaveshownsuccessinmanycases,butmayfailInthispaper,weproposeaglobalsamplingmethod

5、thatwhentheforegroundhaslongandthinstructuresorholes.usesallsamplesavailableintheimage.Ourglobalsam-Theirperformancecanbeimprovedwhencombinedwithplesetavoidsmissinggoodsamples.Asimplebutef-sampling-basedmethods(e.g.,[19,12,7]).fectivecostfunctionisdefinedtotackletheambiguityinS

6、ampling-basedmethodsfirstestimatetheforegroundthesampleselectionprocess.Tohandlethecomputationalandbackgroundcolorsandthencomputethealphamatte.complexityintroducedbythelargenumberofsamples,weEarliermethodslikeRuzonandTomasi’swork[15]andposethesamplingtaskasacorrespondenceproble

7、m.TheBayesianMatting[6],fitaparametricmodeltothecol-correspondencesearchisefficientlyachievedbygeneraliz-ordistributions.Buttheyarelessvalidwhentheimageingarandomizedalgorithmpreviouslydesignedforpatchdoesnotsatisfythemodel.Recentsampling-basedmeth-matching[3].Avarietyofexperime

8、ntsshowthatourglobalods[19,12,7]aremostlynon-parametric:theyp

当前文档最多预览五页,下载文档查看全文

此文档下载收益归作者所有

当前文档最多预览五页,下载文档查看全文
温馨提示:
1. 部分包含数学公式或PPT动画的文件,查看预览时可能会显示错乱或异常,文件下载后无此问题,请放心下载。
2. 本文档由用户上传,版权归属用户,天天文库负责整理代发布。如果您对本文档版权有争议请及时联系客服。
3. 下载前请仔细阅读文档内容,确认文档内容符合您的需求后进行下载,若出现内容与标题不符可向本站投诉处理。
4. 下载文档时可能由于网络波动等原因无法下载或下载错误,付费完成后未能成功下载的用户请联系客服处理。