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 > IT/计算机 > 计算机软件及应用 > P-Saliency Detection--A Spectral Residual Approach
侵权投诉

P-Saliency Detection--A Spectral Residual Approach

SaliencyDetection:ASpectralResidualApproach

XiaodiHouandLiqingZhang

DepartmentofComputerScience,ShanghaiJiaoTongUniversity

No.800,DongchuanRoad,Shanghai

doc.guandang.net houxiaodi,zhang-lq@cs.sjtu.edu.cn

Abstract

Theabilityofhumanvisualsystemtodetectvisualsaliencyisextraordinarilyfastandreliable.However,com-putationalmodelingofthisbasicintelligentbehaviorstillremainsachallenge.Thispaperpresentsasimplemethodforthevisualsaliencydetection.

Ourmodelisindependentoffeatures,categories,orotherformsofpriorknowledgeoftheobjects.Byanalyz-ingthelog-spectrumofaninputimage,weextractthespec-tralresidualofanimageinspectraldomain,andproposeafastmethodtoconstructthecorrespondingsaliencymapinspatialdomain.

Wetestthismodelonbothnaturalpicturesandarti cialimagessuchaspsychologicalpatterns.Theresultindicatefastandrobustsaliencydetectionofourmethod.

1.Introduction

The rststeptowardsobjectrecognitionisobjectdetec-tion.Objectdetectionaimsatextractinganobjectfromitsbackgroundbeforerecognition.Butbeforeperform-ingrecognitivefeatureanalysis,howcanamachinevisionsystemextractthesalientregionsfromanunknownback-ground?

Traditionalmodels,byrelatingparticularfeatureswithtargets,actuallyconvertthisproblemtothedetectionofspeci ccategoriesofobjects[3].Sincethesemodelsarebasedontraining,theexpansibilitybecomethebottleneckingeneralizedtasks.Facingunpredictableandinnumerablecategoriesofvisualpatterns,ageneralpurposesaliencyde-tectionsystemisrequired.Inotherwords,thesaliencyde-tectorshouldbeimplementedwiththeleastreferenceonstatisticalknowledgeoftheobjects.

Howisthesaliencydetectionprocessachievedinhu-manvisualsystem?Itisbelievedthattwostagesofvisualprocessingareinvolved: rst,theparallel,fast,butsimplepre-attentiveprocess;andthen,theserial,slow,butcom-plexattentionprocess.Propertiesofpre-attentiveprocess-inghavebeendiscussedinliterature[27,24].Inthisstage,

1-4244-1180-7/07/$25.00 ©2007 IEEE

certainlowlevelfeaturessuchasorientation,edges,orin-tensitiescan“popup”automatically.Fromaviewpointofobjectdetection,whatpopsupinthepre-attentivestageisthecandidateofanobject.Inordertoaddressacandidatethathasbeendetectedbutnotyetidenti edasanobject,Rensinkintroducedthenotionofprotoobjectsinhiscoher-encetheory[15,13,14].

To ndthe“protoobjects”inagivenimage,modelshadbeeninventedinthe eldofmachinevision.BasedonTreisman’sintegrationtheory[24],IttiandKochpro-posedasaliencymodelthatsimulatesthevisualsearchpro-cessofhuman[8,6,7].Morerecently,Waltherextendedthesaliencymodel,andsuccessfullyappliedittoobjectrecognitiontasks[26].However,asapre-processingsys-tem,thesemodelsarecomputationallydemanding.

Mostofthedetectionmodelsfocusonsummarizingthepropertiesoftargetobjects.However,generalpropertiessharedbyvariouscategoriesofobjectsarenotlikelytoex-ist.Inthispaper,weposethisprobleminanalternativeway:toexplorethepropertiesofthebackgrounds.

InSection2,thespectralresidualisintroduced.Startingfromtheprincipleofnaturalimagestatistics,weproposeafront-endmethodtosimulatethebehaviorofpre-attentivevisualsearch.Differentfromtraditionalimagestatisticalmodels,weanalyzethelogspectrumofeachimageandobtainthespectralresidual.Thenwetransformthespec-tralresidualtospatialdomaintoobtainthesaliencymap,whichsuggeststhepositionsofproto-objects.InSection3,wealsodemonstratemultipleobjectdetectionbasedonthespectralresidualapproach.

Toevaluatetheperformanceofourmethod,inSection4.1,wecompareourmethodwith[8]andhuman-labeledresults.Theresultindicatesthatourmethodisafastandreliablecomputationalmodelformearlystagevisualpro-cessing.

2.SpectralResidualModel

Ef cientcodingisageneralframeworkunderwhichmanymechanismsofourvisualprocessingcanbeinter-preted.Barlow[1] rstproposedtheef cientcodinghy-

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