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 > Tsinghua university at trecvid 2004 Shot boundary detection and high-level feature extracti
侵权投诉

Tsinghua university at trecvid 2004 Shot boundary detection and high-level feature extracti

Our shot boundary detection system mainly consists of three components: fade out/in(FOI)detector, cut(CUT)detector and gradual transition(GT)detector. The key technique of FOI detector is the recognition of monochrome frame. In CUT detector, second order d

TsinghuaUniversityatTRECVID2004:

ShotBoundaryDetectionandHigh-levelFeatureExtraction

JinhuiYuan,WujieZheng,LeChen,DayongDing,DongWang,ZijianTong,HuiyiWang,JunWu

JianminLi,FuzongLin,BoZhang

StateKeyLaboratoryofIntelligentTechnologyandSystem

DepartmentofComputerScienceandTechnologyTsinghuaUniversity

Beijing100084,P.R.China

Abstract

Ourshotboundarydetectionsystemmainlyconsistsofthreecomponents:fadeout/in(FOI)detector,cut(CUT)detectorandgradualtransition(GT)detector.ThekeytechniqueofFOIdetectoristherecognitionofmonochromeframe.InCUTdetector,secondorderderivativemethodisappliedtogetcutcandidates,andthen ashlightdetectorandothergradualtransition ltermoduleareincorporatedtoeliminatefalsepositives.InGTdetector,shortGTandlongGTaretreatedseparately,morespeci cally,twincomparisonalgorithmforshortGTand nitestateautomaton(FSA)modelforlongGT.Tenrunsaresubmitted.Andtheevaluationresultshowsthatoursystemisamongthebest.Thecharacteristicsoftherunsaresummarizedasfollows.

thuai02Baselinesystemwithdefaultarchitectureanddefaultparametersettings.thuai05SecondorderderivativeschemeforshortGT.thuai07LooserconditionsforshortGTdetector.thuai08Withoutgradualtransition ltermodule.thuai10Prolongthedurationofpost-cutmodules,including ashlightdetectorandgradualtransitiondetector.thuai14BaselinesystemDerivedfromthuai10withhighercoef cientsformotionbasedself-adaptivethreshold.thuai15Evenhighercoef cientsformotionbasedself-adaptivethreshold.thuai16IncreasethehigherthresholdofFSAmodel.thuai17FurtherheightenthehigherthresholdofFSAmodel.thuai19ReplacethesecondorderderivativeofCUTdetectorwith rstorderderivative.

InFeatureExtractiontask,thevisualmodaldetectorformsourbaselinesystem.Andseveralkindsofotherdetectorsbasedontextandtimingcueareadded.Thesystemisappliedtothreeconcepts1,namely:“Basketscored”,“BillClinton”and“Beach”.For“Basketscored”,thevisualmodaldetectorachieves0.561inAP.Andtheaddedtextandtiminginfor-mationbrings20%increaseinAPandarrivesthebestrunforthatconcept.Thesubmittedrunsof“Basketscored”are:BthuaiBaselinesystem.AP-basedBordafusionofresultsgeneratedbydifferentvisualmodels.BthuaiBaseline+commercial lterBthuaiBaseline+commercial lter+SRIoperation.BthuaiBaseline+commercial lter+SRIoperation+shotclusteroperation.

1.Introduction

Thisyearweparticipateintwotasksofshotboundarydetection(SBD)andfeatureextraction(FE).

Variousautomaticshotboundarydetectionalgorithmshavebeenproposed(see[7],[11],[13]).Mostofthemseemtofailunderroughlythesameconditionssuchas ashlightorobject/cameramovement.Toboosttheperformance,wedesignandintegratespeci cmodulestoeliminatevariousdisturbance.Theevaluationshowsthatoursystemexhibitsexcellentperformance.

ComparedwiththetraditionalContentBasedImageRetrieval(CBIR),ContentBasedVideoRetrieval(CBVR)hasmuchmoreinformationfrommultiplemodalitiesandvariousinformationfusionmethodstodealwiththesemanticgapbetween

detection”and“featureextraction”areusedinterchangeablyinthispaper.Butasthelatterareoftenconfusedwiththefrequentlyused

“low-levelfeatureextraction”,theformerispreferred.

1“Concept

1

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