数字图像处理-冈萨雷斯-课件(英文)Chapter11 表示与描述

DigitalImageProcessingChapter11:ImageDescriptionandRepresentation12September2007ImageRepresentationandDescription?Objective:Torepresentanddescribeinformationembeddedinanimageinotherformsthataremoresuitablethantheimageitself.Benefits:-Easiertounderstand-Requirefewermemory,fastertobeprocessed-More“readytobeused”Whatkindofinformationwecanuse?-Boundary,shape-Region-Texture-RelationbetweenregionsShapeRepresentationbyUsingChainCodesChaincodes:representanobjectboundarybyaconnectedsequenceofstraightlinesegmentsofspecifiedlengthanddirection.4-directionalchaincode8-directionalchaincodeWhywefocusonaboundary?Theboundaryisagoodrepresentationofanobjectshapeandalsorequiresafewmemory.(ImagesfromRafaelC.GonzalezandRichardE.Wood,DigitalImageProcessing,2ndEdition.ExamplesofChainCodesObjectboundary(resampling)Boundaryvertices4-directionalchaincode8-directionalchaincode(ImagesfromRafaelC.GonzalezandRichardE.Wood,DigitalImageProcessing,2ndEdition.TheFirstDifferenceofaChainCodesProblemofachaincode:achaincodesequencedependsonastartingpoint.Solution:treatachaincodeasacircularsequenceandredefinethestartingpointsothattheresultingsequenceofnumbersformsanintegerofminimummagnitude.Thefirstdifferenceofachaincode:countingthenumberofdirectionchange(incounterclockwise)between2adjacentelementsofthecode.Example:1230Example:-achaincode:10103322-Thefirstdifference=3133030-Treatingachaincodeasacircularsequence,wegetthefirstdifference=33133030Chaincode:Thefirstdifference011022033231202213Thefirstdifferenceisrotationalinvariant.PolygonApproximationObjectboundaryMinimumperimeterpolygonRepresentanobjectboundarybyapolygonMinimumperimeterpolygonconsistsoflinesegmentsthatminimizedistancesbetweenboundarypixels.(ImagesfromRafaelC.GonzalezandRichardE.Wood,DigitalImageProcessing,2ndEdition.PolygonApproximation:SplittingTechniques1.Findthelinejoiningtwoextremepoints0.Objectboundary2.Findthefarthestpointsfromtheline3.Drawapolygon(ImagesfromRafaelC.GonzalezandRichardE.Wood,DigitalImageProcessing,2ndEdition.Distance-Versus-AngleSignaturesRepresentan2-Dobjectboundaryintermofa1-Dfunctionofradialdistancewithrespecttoq.(ImagesfromRafaelC.GonzalezandRichardE.Wood,DigitalImageProcessing,2ndEdition.BoundarySegmentsConcept:Partitioninganobjectboundarybyusingverticesofaconvexhull.Convexhull(graycolor)ObjectboundaryPartitionedboundary(ImagesfromRafaelC.GonzalezandRichardE.Wood,DigitalImageProcessing,2ndEdition.Input:AsetofpointsonacorneaboundaryOutput:Asetofpointsonaboundaryofaconvexhullofacornea1.Sortthepointsbyx-coordinatetogetasequencep1,p2,…,pnFortheuppersideofaconvexhull2.Putthepointsp1andp2inalistLupperwithp1asthefirstpoint3.Fori=3ton4.DoappendpitoLupper5.WhileLuppercontainsmorethan2pointsandthelast3pointsinLupperdonotmakearightturn6.Dodeletethemiddlepointofthelast3pointsfromLupperTurnRightOK!TurnRightOK!TurnLeftNOK!ConvexHullAlgorithmForthelowersideofaconvexhull7.Putthepointspnandpn-1inalistLlowerwithpnasthefirstpoint8.Fori=n-2downto19.DoappendpitoLlower10.WhileLlowercontainsmorethan2pointsandthelast3pointsinLlowerdonotmakearightturn11.Dodeletethemiddlepointofthelast3pointsfromLlower12.RemovethefirstandthelastpointsfromLlower13.AppendLlowertoLupperresultinginthelistL14.ReturnLTurnRightOK!TurnRightOK!TurnLeftNOK!ConvexHullAlgorithm(cont.)SkeletonsObtainedfromthinningorskeletonizingprocessesMedialaxes(dashlines)(ImagesfromRafaelC.GonzalezandRichardE.Wood,DigitalImageProcessing,2ndEdition.ThinningAlgorithmNeighborhoodarrangementforthethinningalgorithmConcept:1.Donotremoveendpoints2.Donotbreakconnectivity3.DonotcauseexcessiveerosionApplyonlytocontourpixels:pixels“1”havingatleastoneofits8neighborpixelsvalued“0”p2p9p3p8p7p1p4p6p5Let98321)(pppppNT(p1)=thenumberoftransition0-1intheorderedsequencep2,p3,…,p8,p9,p2.Notation:=Let00111p1001ExampleN(p1)=4T(p1)=3ThinningAlgorithm(cont.)Step1.Markpixelsfordeletionifthefollowingconditionsaretrue.a)b)T(p1)=1c)d)6)(21pN0642ppp0864pppp2p9p3p8p7p1p4p6p5Step3.Markpixelsfordeletionifthefollowingconditionsaretrue.a)b)T(p1)=1c)d)6)(21pN0842ppp0862pppStep4.DeletemarkedpixelsandrepeatStep1untilnochangeoccurs.(Applytoallborderpixels)Step2.DeletemarkedpixelsandgotoStep3.(Applytoallborderpixels)Example:SkeletonsObtainedfromtheThinningAlg.(ImagesfromRafaelC.GonzalezandRichardE.Wood,DigitalImageProcessing,2ndEdition.SkeletonBoundaryDescriptors1.Simpleboundarydescriptors:wecanuse-Lengthoftheboundary-Thesizeofsmallestcircleorboxthatcantotallyenclosingtheobject2.Shapenumber3.Fourierdescriptor4.StatisticalmomentsShapeNumberShapenumberoftheboundarydefinition:thefirstdifferenceofsmallestmagnitudeTheordernoftheshapenumber:thenumberofdigitsinthesequence1230(ImagesfromRafaelC.GonzalezandRichardE.Wood,DigitalImageProcessing,2ndEdition.ShapeNumber(cont.)Shapenumbersoforder4,6and8(ImagesfromRafaelC.GonzalezandRichardE.Wood,DigitalImageProcessing,2ndEdition.Example:ShapeNumberChaincode:000030032232221211Firstdifference:300031033013003130ShapeNo.0003103301300313031.Originalboundary2.Findthesmallestrectanglethatfitstheshape3.Create