信号处理讲稿.pdfVIP

SignalProcessing86(2006)71–83

Anovelapproachtothedetectionofsmallobjects

withlowcontrast$

aba

Feng-YangHsieh,ChhuanHan,Nai-ShenWu,

cFana,

ThomasC.Chuang,Kuo-Chin

aInstituteofComputerScienceandInformationEngineering,NationalCentralUniversity,Chung-Li320,,ROC

bDepartmentofComputerScienceandInformationEngineering,NationalUnitedUniversity,Miao-Li360,,ROC

cDepartmentofComputerScienceandInformationEngineering,VanungUniversity,Chung-Li320,,ROC

Received26September2003;receivedinrevisedform8December2004;accepted24April2005

Availableonline31May2005

Thispaperproposesaneffectiveapproachtothedetectionofsmallobjectsbyemployingwatershed-based

transformation.Inourwork,movingobjectswithsmallsizeandlowcontrastarefirstdetectedfromanimagesequence

whichwascapturedfromacamera.Theproposeddetectionsystemludestwomainmodules,regionofinterest

(ROI)locatingandcontourextraction.Intheformermodule,animagedifferencingtechniqueisfirstemployedontwo

contiguousimageframestogenerateroughcandidateobjectsappearingintheimages.Anovelneighboringencoding

techniquealongwiththeimagedifferencingtechniqueisdevisedheretoeffectivelyreducenoisewhichusuallyaffects

theperformanceofdetectionresults,especiallyforsmallobjects.Next,wefindtheboundedrec