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 > 自然科学 > 天文/地理 > Salient Region Detection by UFO_ Uniqueness, Focusness and Objectness
侵权投诉

Salient Region Detection by UFO_ Uniqueness, Focusness and Objectness

显著性检测cvpr14年新方法

SalientRegionDetectionbyUFO:Uniqueness,FocusnessandObjectness

PengJiang1

1

HaibinLing2 JingyiYu3JingliangPeng1

SchoolofComputerScienceandTechnology,ShandongUniversity,Jinan,China2

Computer&InformationScienceDepartment,TempleUniversity,Philadelphia,PAUSA3

DepartmentofComputerandInformationSciences,UniversityofDelaware,Newark,DEUSA

jump@mail.sdu.edu.cn,hbling@temple.edu,yu@cis.udel.edu,jpeng@sdu.edu.cn

Abstract

Thegoalofsaliencydetectionistolocateimportantpix-elsorregionsinanimagewhichattracthumans’visualat-tentionthemost.Thisisafundamentaltaskwhoseoutputmayserveasthebasisforfurthercomputervisiontaskslikesegmentation,resizing,trackingandsoforth.

Inthispaperweproposeanovelsalientregiondetec-tionalgorithmbyintegratingthreeimportantvisualcuesnamelyuniqueness,focusnessandobjectness(UFO).Inparticular,uniquenesscapturestheappearance-derivedvi-sualcontrast;focusnessre ectsthefactthatsalientregionsareoftenphotographedinfocus;andobjectnesshelpskeepcompletenessofdetectedsalientregions.Whileuniquenesshasbeenusedforsaliencydetectionforlong,itisnewtointegratefocusnessandobjectnessforthispurpose.Infact,focusnessandobjectnessbothprovideimportantsaliencyinformationcomplementaryofuniqueness.Inourexperi-mentsusingpublicbenchmarkdatasets,weshowthat,evenwithasimplepixellevelcombinationofthethreecompo-nents,theproposedapproachyieldssigni cantimprove-mentcomparedwithpreviouslyreportedmethods.

Figure1.Fromlefttoright:sourceimages,uniqueness,focusness,objectness,combinedresultsandgroundtruth.

1.Introduction

Humanshavethecapabilitytoquicklyprioritizeexternalvisualstimuliandlocalizetheirmostinterestinascene.Assuch,howtosimulatesuchhumancapabilitywithacom-puter,i.e.,howtoidentifythemostsalientpixelsorregionsinadigitalimagewhichattracthumans’ rstvisualatten-tion,hasbecomeanimportanttaskincomputervision.Fur-ther,resultsofsaliencydetectioncanbeusedtofacilitateothercomputervisiontaskssuchasimageresizing,thumb-nailing,imagesegmentationandobjectdetection.

Duetoitsimportance,saliencydetectionhasreceivedintensiveresearchattentionresultinginmanyrecentlypro-posedalgorithms.Themajorityofthosealgorithmsare

basedonlow-levelfeaturesoftheimagesuchasappear-anceuniquenessinpixelorsuperpixellevel(SeeSec.2).Onebasicideaistoderivethesaliencyvaluefromthelocalcontrastofvariouschannels,suchasintermsofunique-nessde nedin[29].Whileuniquenessoftenhelpsgenerategoodsaliencydetectionresults,itsometimesproduceshighvaluesfornon-salientregions,especiallyforregionswithcomplexstructures.Asaresult,itisdesiredtointegratecomplementarycuestoaddresstheissue.

Inspiredbytheabovediscussion,inthispaperwepro-poseintegratingtwoadditionalcues,focusnessandobject-nesstoimprovesalientregiondetection.First,itiscom-monlyobservedthatobjectsofinterestinanimageareof-tenphotographedinfocus.Thisnaturallyassociatesthefo-cusness(i.e.,degreeoffocus)withthesaliency.Wede-riveanalgorithmforfocusnessestimationbytreatingfo-cusnessasareciprocalofblurriness,whichisinturnes-timatedbythescaleofedgesusingscale-spaceanalysis.Second,intuitively,asalientregionusuallycompletesob-jectsinsteadofcuttingthemintopieces.Thissuggestsustouseobjectcompletenessasacuetoboostthesalientre-giondetection.Therecentlyproposedobjectnessestima-tionmethod[3]serveswellforthispurposebyprovidingthelikelihoodthataregionbelongstoanobject.1

Corresponding

Salient Region Detection by UFO_ Uniqueness, Focusness and Objectness

authors.

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