Active AppearanceModels Revisited

Active AppearanceModels Revisited

ID:39909493

大小:736.85 KB

页数:53页

时间:2019-07-14

Active AppearanceModels Revisited_第1页
Active AppearanceModels Revisited_第2页
Active AppearanceModels Revisited_第3页
Active AppearanceModels Revisited_第4页
Active AppearanceModels Revisited_第5页
资源描述:

《Active AppearanceModels Revisited》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、ActiveAppearanceModelsRevisitedIainMatthewsandSimonBakerTheRoboticsInstituteCarnegieMellonUniversityAbstractActiveAppearanceModels(AAMs)andthecloselyrelatedconceptsofMorphableModelsandActiveBlobsaregenerativemodelsofacertainvisualphenomenon.Althoughlinearinboths

2、hapeandappearance,overall,AAMsarenonlinearparametricmodelsintermsofthepixelintensities.FittinganAAMtoanimageconsistsofminimisingtheerrorbetweentheinputimageandtheclos-estmodelinstance;i.e.solvinganonlinearoptimisationproblem.Weproposeanefficientfittingalgorithmfor

3、AAMsbasedontheinversecompositionalimagealignmentalgorithm.Weshowthattheeffectsofappearancevariationduringfittingcanbeprecomputed(“projectedout”)usingthisalgorithmandhowitcanbeextendedtoincludeaglobalshapenormalisingwarp,typicallya2Dsimilaritytransformation.Weeval

4、uateouralgorithmtodeterminewhichofitsnovelaspectsimproveAAMfittingperformance.Keywords:ActiveAppearanceModels,AAMs,ActiveBlobs,MorphableModels,fitting,effi-ciency,Gauss-Newtongradientdescent,inversecompositionalimagealignment.1IntroductionActiveAppearanceModels(AAM

5、s)[7–11,13],firstproposedin[14],andthecloselyrelatedconceptsofActiveBlobs[21,22]andMorphableModels[6,18,24],arenon-linear,generative,andparametricmodelsofacertainvisualphenomenon.ThemostfrequentapplicationofAAMstodatehasbeenfacemodelling[19].However,AAMsmaybeusef

6、ulforotherphenomenatoo[18,22].Inatypicalapplication,thefirststepistofittheAAMtoaninputimage,i.e.modelparametersarefoundtomaximisethe“match”betweenthemodelinstanceandtheinputimage.Themodelparametersarethenusedinwhatevertheapplicationis.Forexample,theparameterscould

7、bepassedtoaclassifiertoyieldafacerecognitionalgorithm.Manydifferentclassificationtasksarepossible.In[19],forexample,thesamemodelwasusedforfacerecognition,poseestimation,andexpressionrecognition.FittinganAAMtoanimageisanon-linearoptimisationproblem.Theusualapproach

8、[7,10,11]istoiterativelysolveforincrementaladditiveupdatestotheparameters(theshapeandappearancecoefficients.)Giventhecurrentestimatesoftheshapeparameters,itispossiblet

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。