基于形态学的分水岭算法牙科X射线图像分割.docx

基于形态学的分水岭算法牙科X射线图像分割.docx

《基于形态学的分水岭算法牙科X射线图像分割.docx》由会员分享，可在线阅读，更多相关《基于形态学的分水岭算法牙科X射线图像分割.docx（11页珍藏版）》请在冰豆网上搜索。

基于形态学的分水岭算法牙科X射线图像分割

英文原文

WatershedAlgorithmBasedonMorphologyforDentalX-RayImagesSegmentation

Abstract—Anewwatershedalgorithmbasedonmathematicalmorphology,whichcanbeappliedtodentalX-rayimagessegmentation,isproposedinthepaper.Inordertoseparateeachtoothandimprovetheseriousproblemofover-segmentationoftraditionalwatershedalgorithm,weapplytheprocessingbasedonmorphologytogettheimagesenhancedbeforewatershedalgorithm.First,usingthetop-hat-bottom-hattransformationtoamplifythecontrastofforegroundandbackgroundandremovenoises.Andthenerosionprocedureisusedtoweakenthedegreeofadhesionbetweenteeth.Hole-fillingcanhelptoeliminatesomeunnecessarysplit-linewhichcancausethephenomenonofwrongsegmentation.Finallyweapplythewatershedalgorithmusingdistancetransformationofthebinaryimagetogetthesegmentationresult.Fromtheexperimentresults,wecanseethatthisalgorithmcanseparateteethaccuratelyandovercometheover-segmentationefficientlycomparedwithtraditionalmethod.

Keywords-watershedalgorithm;mathematicalmorphology;dentalX-rayimage;imagesegmentation;distancetransformation

I.INTRODUCTION

Dentalbiometricsisanappropriatemethodtoidentifydeceasedindividualsofdisasterssuchascriminalcases,blastsandTsunamis,becauseteethwiththeirdentalworksareveryresistanttomodestforceeffectsandhightemperatureandchemicalcorrosionandalsopossessgoodbiometricproperties.Anditalsoprovidesimportantinformationindentalmedicaltreatmentforfurtheranalysistogetthediagnostic.Imagesegmentationasaprecursorhasbeenseenasoneofthemostimportantaspectoftheentireprocesslikepatternrecognition.Itgivesclarityinformationandthelevelstosubdivisiondependontheproblembeingsolved.

Imagesegmentationasanindispensablemethodisappliedtoextractquantitativeinformationofspecifictissues,alsoasthepremiseandasteptorealizetheprocessingofvisualization.Anditisstillachallengingproblemincomputervisionandimageprocessing.Areasobtainedbysegmentationarebothindependentfromeachother.Andalltheareasbelongtooneregionwillsharethespecialconsistency.TheobjectiveofsegmentationistolocalizetheregionofeachtoothindentalX-rayimage[3].Atpresent,imagesegmentationhasbeenbroadlydividedinto3categories:

clusteranalysis,edgedetectionandregionextraction.Eachofthosemethodshasgotdisadvantagessuchasthefailuretogetthenumberofclustersbeforesegmentationwhenapplyingclusteranalysis.

Watershedalgorithmbasedonmorphologicaltheoryisakindofnonlinearsegmentation.Itpossessesgreatqualitiesinpositioningtheedgerapidlyandaccuratelyandbehavesexcellentatdetectingobjectswithweakedgesetc.Buttraditionalwatershedalgorithmhasseriousproblemofover-segmentationundertheinfluenceofnoiseandthefinetextureoftheobjects.Avarietyofapproachesonhowtoovercomethesplithavebeenproposedfordifferentpurposesindifferentapplications.

WatershedSegmentationAlgorithmbasedonmorphologyhasbeenappliedtocellularimages,asin[1],aimedatreducingtheover-segmentation.Itrequiresfewercomputationandsimplerparametersandgetssegmentationresultsmoreaccuratelycomparedwithothertraditionmethods.

In[2]and[3],NomirandAdbel-Mottalebintroducedafullyautomatedsegmentationtechnique.Itstartsbyapplyingiterativethresholdingfollowedbyadaptivethresholdingtosegmenttheteethfromboththebackgroundandtheboneareas.Andafteradaptivethresholding,horizontalintegralprojectionfollowedbyverticalintegralprojectionareappliedtoseparateeachindividualtooth.Andthismethodcanachievethepositionofeachtoothprecisely.

In[4],analgorithmbasedonwavelettransform（WT）tosegmentdentalX-rayimageswasproposed.Itcontainsthreemajorsteps:

dentalX-raypreparing,panoramicradiographsegmentationusingwavelettransformationandenhancementimagewithmorphologicalimageprocessing.Andthesegmentationresultusingthismethodisbetterthanthresholdingsegmentationandadaptivethresholdingsegmentation.

In[7],awatershedsegmentationalgorithmbasedongrayscalemorphologicalpretreatingispresented.Openingoperationsareappliedtoremovesmalllightdetailsbeforeapplyingwatershedalgorithm.Thereforethephenomenonofover-segmentationwascontrolledandthetouchingobjectsweresegmentedprecisely.

Inthispaper,weintroduceanewmethodbasedontheperformanceofwatershedalgorithmandthecharacteristicsofdentalX-rayimages.Atfirst,weapplythebottom-hat-top-hattransformationtoenhancethedentalradiographs.Thenweusedtheerosionalgorithmtoweakenthedegreeofadhesionbetweenteethandremovethenoises.Andtheimfill（）functioncouldhelptoeliminatethepossibilityofover-segmentationcausedbytheupcomingprocessing.Finallyweutilizethewatershedalgorithmusingdistancetransformofthebinaryimagetogetthesegmentationresult.

II.WATERSHEDALGORITHMBASEDONMORPHOLOGY

A.Top-hat-bottom-hattransformation

Top-hattransformationisdefinedasthedifferenceoftheoriginalimageminustheimageappliedwiththeoperationofimopentoremovethesomepointswiththehighestgrayvalueoftheimage.Andthebottom-hattransformisthedifferenceoftheimageappliedwiththeoperationofimclosetoremovethepointswithsmallestgrayvalueandthenminustheoriginalimage.Thedefinitionsareasfollows:

Top-hattransformation：

Bottom-hattransformation:

Wherefistheoriginalimageandbrepresentsthestructureelements.

Theopeningoffbyb,denotedf○b,is

Theclosingoffbyb,denotedf●b,is

WhereΘistheoperationoferosionandrepresentsdilation.

Openingoperationisusedtoremoveregionssmallerthanthestructureelementswithrelativehighgrayvalueswhiletheclosingoperationcanremovethesmallregionswithrelativelowgrayvalues.Andthesetwooperationspossessgreatqualityinremainingallthelevelsofgrayvalueandkeepingthelargeregionswithhighvaluesrelativelyunchanged.Oneoftheimportantusagesisthatthesetwotransformationsarecapableofcorrectingunevenilluminationeffects.Top-hattransformcanbeusedinsituationofbrightobjectsagainstadarkbackground,whilethebottom-hattransformisappliedintheoppositesituation.

Top-hattransformationhassomecertaincharacteristicsofhighpassfilterthatitcanbeusedtohighlightthegraypeakandenhancetheedgeinformationofthetargets.Andthebottom-hattransformationcanbeappliedtogetthevalleyofthegrayvalueandprominenttheboundariesbetweenconnectedtargetsliketeeth.Therefore,thesetwotransformationscanbeusedincombinationtogettheeffectofimageenhancementfortheforegroundandbackgroundgrayarefurtherstretchedaswellastheobjectivesanddetailsarehighlighted.

Inthispaperwecangettheimagewithitscontrasteffectivelyimprovedbyaddingtheoriginalimagewiththeimageappliedwithtop-hattransformationandthenminustheimageappliedwiththebottom-hattransformation.

Afterenhancementthegrayimagewillbetransformedintoabinaryimagebyusingthresholdingforthenextprocessing.

B.Themorphologicalprocessingtothebinaryimage

Mathematicalmorphologyisveryusefulasatooltoextractimagecomponentsappliedintherepresentationanddescriptionoftheboundary,thebonesandtheconvexhull.Erosionanddilationarethebasicoperationsofmorphology.Thispapermainlyusestheprocessingoferosionwhichactuallymeansusingstructuralelementstofilltheimage.Erosionwouldshrinkorrefinementtheobjections.Themodeanddegreearecontrolledbythestructuralelements.WeapplythestructuralelementsofbtofillthecollectionA.AndwecanassumethatthecollectionAiserodedbybifAstillcontainsthestructureelementsbafterthefillings.Itisdefinedasfollows:

Wherebrepresentsthestructureelements.

Difficultyexistsindentalsegmentationbecauseoftheadhesionbetweenteeth.Wecouldnotgetthepositionofeachtoothpreciselywithoutdeletetheadhesion.Therefore,weneedtoapplytheimagewitherosionoperationtoweakenthedegreeofadhesionbetweenteethforthebenefitofsegmentation.

Hole-fillingreferstotheimagecontourfilling.Usuallytheinnercontourareashouldnotbebiggerthanthemaximumareaoftargets.TheremaybesomeholeswhichshouldnotexistwhenthedentalX-rayimagesaretransformedintobinaryimagesundertheinfluenceofshapecharacteristicsoftoothandthenon-uniformgraydistribution.Andtheseholesmaybetransformedintolittleareasindependentofeachotheraftererosion.Thiswouldleadtothephenomenonofover-segmentafterappliedwatershedalgorithm.Therefore,thestructureelementsbcannotbetoolargeforthepossibilityofholesgenerated.Anditalsocannotbetoosmallfortheadhesionmustbedeletedcompletely.Inthispaper,weapplytheoperationoferosionfortwotimesandduringthesetwoproceduresaddtheoperationofimfill（）toavoidthegenerationofover-segmentationcausedbyholes.

C.Watershedsegmentationalgorithmusingdistancetransformationofbinaryimage

Distancetransformationturnsthebinaryimageintoagrayimage.Andthevalueoflocation（x,y）isthedistanceofpixeltoitsnearestbackgroundpixel.Itaimstodistinguishtheboundarypixelsandtheinnerpixels.

WatershedalgorithmbasedonmorphologicaltheoryisfirstproposedbyS.BeucherandL.Vincentanddevelopedrapidlyinimagesegmentationfieldinrecentyears[6-7].Ingeography,awatershedreferstoadam.Theriversystemsint