文档下载 文档下载导航
设为首页 | 加入收藏
搜索 请输入内容:  
 导航当前位置: 文档下载 > 所有分类 > 工程科技 > 信息与通信 > Model selection a bootstrap approach
侵权投诉

Model selection a bootstrap approach

Model selection a bootstrap approach

MOTIONESTIMATIONUSINGAJOINTOPTIMISATIONOFTHEMOTIONVECTOR

FIELDANDASUPER-RESOLUTIONREFERENCEIMAGE

ChristianDebes1,ThomasWedi2,ChristopherL.Brown1,andAbdelhakM.Zoubir1SignalProcessingGroupTechnischeUniversit¨atDarmstadt

Darmstadt,Germany

ABSTRACT

Inmanysituations,interdependencybetweenmotionestima-tionandotherestimationtasksisobservable.Thisisforin-stancetrueintheareaofSuper-Resolution(SR)[1].Inor-dertosuccessfullyreconstructaSRimage,accuratemotionvector eldsareneeded.Ontheotherhand,onecanonlygetaccurate(subpixel)motionvectors,ifthereexisthighlyaccuratehigher-resolutionreferenceimages.Neglectingthisinterdependencymayleadtopoorestimationresults-formo-tionestimationaswellasfortheSRimage.Toaddressthisproblem,anewmotionestimationschemeispresentedthatjointlyoptimisesthemotionvector eldandaSRreferenceimage.Forthispurposeandinordertoattenuatealiasingandnoise,whichdeterioratethemotionestimation,anobserva-tionmodelfortheimageacquisitionprocessisappliedandaMaximumAPosteriori(MAP)optimisationisperformed,us-ingMarkovRandomFieldimagemodelsforregularisation.Resultsshowthatthenewmotionestimatorprovidesmoreaccuratemotionvector eldsthanclassicalblockmotiones-timationtechniques.Thejointoptimisationschemeyieldsanestimateofthemotionvector eldaswellasaSRimage.IndexTerms—Motionestimation,ImageReconstruc-tion,Stochastic elds,MAPestimation

1.INTRODUCTION

Motionestimationisanimportantresearchtopicinimageandvideoprocessingaswellasincomputervision.Under-standinghowobjectsmoveisa eldofhighpracticalinter-est,coveringabroadrangeofapplicationsinmanyengineer-ing eldssuchasvideoobjecttracking[2],Super-Resolutionimagereconstruction[1],videocompression[3]andmedicalengineering[4]tonameafew.Anaccurateestimateofthemotionvector eldisthe rstandcrucialstepforsuccessfulresultsintheaforementionedapplications.

Onepopularapproachtomotionvector eldestimationisblockmatching[3].Inblockmatching,imagesofinterestaredividedintoblocks.Matchingblocks(i.e.thosethatrepresentthesamephysicalfeature)aresearchedforinareferenceim-age.Thespatialdisplacementbetweentwomatchingblocks

1

PanasonicFrankfurtLaboratoryPanasonicR&DCenterGermany

Langen,Germany

issaidtocorrespondtothemotionofthephysicalregiontheyrepresent,i.e.themotionvector.Assumingashorttimedis-tancebetweentwoimagesofavideosequence,atranslationalmotionmodelforsmallimageblocksissuf cienttodescribethespatialdisplacement.

Letb1,kbeavectorisedimageblocktakenfromagrayscaleimageofinterest.Itselementsareinlexicographicnotation.Further,letb0,lbeavectorisedimageblockinareferenceim-age.Theaimofblockmatchingisto ndtheblockb0, lthatcorrespondstothesamephysicalregionasb1,k.Makingtheassumptionofintensityconservation,i.e.pixelintensitiesareassumedtobeconstantwithinatrackingperiod,motivatestheuseoftheSumofAbsoluteDifferences(SAD)asameasureofsimilaritybetweenablockb1,kandareferenceblockb0,l.BesidestheSADthereareanumberofotherdistancemea-suressuchastheSSD(SumofSquaredDifferences),SATD(SumofAbsoluteTransformDifferences)andSSTD(SumofSquaredTransformDifferences).WerestrictourselvestotheSADaswefoundtheremainingmeasureshavingcompara-bleperformance.TheSADblockmatcherwithfullsearchworksasfollows:b1,kiscomparedtoallpossible(overlap-ping)blocksb0,linasearchareaofthereferenceimage.Thematchingblockb0, listhencalculatedastheonewhichleadstothelowestSAD.Forhigh-precisionmotionestimationfrac-tionalpixelaccuracyisneeded.

1

Inordertoincreasethemotionresolutiontoppi

2

第1页

你可能喜欢

热门文档

相关文档

站点地图 | 文档上传 | 侵权投诉 | 手机版
本站所有资源均来自互联网,本站只负责收集和整理,均不承担任何法律责任,如有侵权等其它行为请联系我们.
文档下载 Copyright 2013 doc.guandang.net All Rights Reserved.  -  email