美赛重庆大学特等奖题名论文.docx
《美赛重庆大学特等奖题名论文.docx》由会员分享,可在线阅读,更多相关《美赛重庆大学特等奖题名论文.docx(30页珍藏版)》请在冰豆网上搜索。
美赛重庆大学特等奖题名论文
2014MathematicalContestinModeling(MCM)SummarySheet
Forofficeuseonly
T1
________________
T2
________________
T3
________________
T4
________________
TeamControlNumber
201401
ProblemChosen
B
Forofficeuseonly
F1
________________
F2
________________
F3
________________
F4
________________
Summary
Asisknowntoall,splicingofpaperscrapsisacomplexissue,whichexertsaimportantroleinjudicialevidencerecovery,restorationofhistoricaldocumentsandaccesstomilitaryintelligence.Thispaperfocusesonsplicingproblemofpaperscraps,establishingshreddingdistancemodelandrestorationTSPmodel.Atsametime,wedesignone-dimensionalandmulti-dimensionalpiecesrestorationalgorithmandthensolvesitbyusingMATLAB.
Forquestionone,weextractinformationfromtheAppendix1and2,designingtorecovertheone-dimensionalshreddingalgorithms,whichischaracterizedbyatextcharactersize,linespacingstructure.AndthenwecantransfortheshreddingproblemintorecoveryTSPproblem,thusobtainingthecorrectrecoverygraphicsandsequences.
Forquestiontwo,wefirstlystandardizeshreddingpicturesfromtheAppendix1and2,andthenextractthenormalizedimage-levelfeatures.Forthatpicturescannotbeclassifiedviamachine,weusethedevelopedprogramsofGUItoimprovetheefficiencyoflabor.
Forquestionthree,three-dimensionaldesignshreddingrestorationalgorithm,afirstsurfaceandthesurfaceofAnnex5bintegratepictures,get416piecesofshreddingpictures,thepicturealsostandardizedlevelfeatureextraction,classificationandotheroperations,willreduceddimensionsoftheone-dimensionalproblemandsolvedtoobtainthecorrectrecoveryimagesandsequencesofpositiveandnegativeAnnex5.Takingintoaccounttheproblemofquantitativeevaluationalgorithm,thispaperpresentsminimalinterventionmodeltoimprovethealgorithminplace,thatis,throughthecomputertorecognizetheorderandsequenceinreverseordertorecoverthenumberofmanualinterventiontoachieveaminimumnumberofadvantagesanddisadvantagesofthealgorithmisportrayed.
Keywords:
Reconstructdocuments;TSP;ShreddingDistanceModel;ShreddingRestorationAlgorithm
Content
IIntroduction2
IISymbolDefinitions2
IIIAssumptionsandNotations2
ⅣForquestionone3
4.1ImagePreprocessing3
4.2ShreddingFeatureExtraction3
4.3RecognitionSequenceBasedonTextFeatures4
4.4TheDefinitionofShreddingDistance5
4.5RecoveryofTSP5
4.6SimulateAnneal(SA)Algorithm5
4.7One-DimensionalShreddingRestorationAlgorithm6
4.8TheSolutionofModel6
ⅤForquestiontwo7
5.1ShreddingStandardizationAndLevelFeatureExtraction7
5.2TheClassificationofLevelFeature8
5.3Two—DimensionalShreddingRestorationAlgorithm8
5.4TheSolutionofModel9
ⅥForquestionthree10
6.1DimensionalityReduction10
6.2Three—DimensionalShreddingRestorationAlgorithm11
6.3TheSolutionofModel11
ⅦStrengthsandWeaknesses12
7.1Strengths12
7.2Weaknesses12
ⅧTheRefinementofourModel13
8.1ImprovedApplyforColorfulImages13
8.2MinimalInterventionDegreeAlgorithm13
Reference13
AppendixⅠ15
AppendixⅡ16
AppendixⅢ17
Introduction
Traditionally,reconstructingshreddeddocumentscompletedbyhandiswithhigheraccuracy,butinefficiency,especiallywhenahugeamountofcomplicatedworktocompleteinashorttime.Withthedevelopmentofcomputertechnology,peopleistryingtodevelopautomaticsplicingtechniqueforreconstructingdocuments,astoimprovetherecoveryefficiencyofsplicing.
Inaddition,thisisakindofstaffwhichisrelatedtoourdailylife.Thefactorstobeconsideredinrealityfarmorethanthesubjectitself,andhowtomakethemodelmorerealisticandprovideeffectivesplicinginformationinthisarticleisamajorproblem.Facedbylotofinformationofferedandreasonableassumptionsforshreddingrecovery,weareabletoconducttheresearchforshreddingrecovery.
SymbolDefinitions
SymbolDefinitions
Pixelvaluesbeforebinarization
Pixelvaluesbeforebinarization
ThedistancebetweenshredAandshredB
Leftrecognitionsequence
Rightrecognitionsequence
Widthofcharacters
TotaldistanceofTSP
Thelengthofrecognitionsequence
AssumptionsandNotations
Forthesakeofconvenienceofthefollowingdiscussions,wefirstlyassumethat:
(1)Textdirectionishorizontal
(2)Positiveandnegativeprintmarginsareinthesameformat
(3)Ignoretheefficiencyoflaborproductivity
ⅣForquestionone
4.1Imagepreprocessing
Accordingtotherelevantknowledge,weneedtoprocessthepicturepixels.
Generally,theimagepixelvaluesarepositionedwithin[0,255],andthenaredistinguishedbetweenblankpositionandfontbysettingthethreshold.Asfornon-colorpictures,wejustneedtodistinguishblankandnon-blank.
Tomakethepicturecanclearlydescribetheemptyspaceandthecharacterposition,weuseMATLABforpreprocessingandputtheimageintoMATLABastoobtainthecorrespondingpixelmatrix.Atlast,wemakepixelmatrixbinarizationandthenhave
1,qij=255
Pij=
255,others
4.2Shreddingfeatureextraction
Generallyspeaking,shreddingfeatureextractionisdividedintotwocategories.Oneistoextractshreddingfeaturebysplicingshapefeatures,andtheotherischaracterizedbyextractingtextshreddingbasedonfeatures.Accordingtotheproblem,theshapeofthispaperbelongstothesecondcategory.
Figure1.One-dimensionalshredding
Figure2.Charactersfeatures
Insummary,thetextfeatureextractionasfollows:
Step1:
thepicture’sbinarization.textiswhite,blankisblack.
Step2:
findalllinespacingandemptyplaceofpictures,andmarkitasgray
Step3:
findoutallthekerning,andmarkitasgray
Step4:
calculatethecharacterwidthbyspacing,empty,kerningandotherfeatures.Accordingtotheproblem,thispaperextractstextfeaturebyimportingtheimagepixelsandusingMATLABprogram
4.3Recognitionsequencebasedontextfeatures
ThroughtheanalysisofChinesecharactersandEnglishletters,wesignthecharacterwidthofC.Thewidthisdividedintwoparts,respectivelyC—RandR,andfornotbeingcutcharacter,stillretainsthewidthC.
Figure3.Charactersegmentation
Accordingtothedefinitionofcharacter-basedsegmentation,weconstructrecognitionsequencesbasedonthecharacteristicsofthetext
LeftRight
Figure4.Recognitionsequences
FortherecognitionsequenceinFigure4,theplacewithnocharacterpositionis0,andtheothernodesrepresentthecorrespondingcharacterlength(forthefullCandtheincompleteisC—RorR).
4.4Thedefinitionofshreddingdistance
Accordingtothedefinitionofrecognitionsequence,wedefinethedistancebetweenshreddingAandBandweget
X=0or1
Fromtheseequations,weknowthatthegreaterthedegreeofagreementofthetworecognitionsequences,thesmallerthedistancebetweentwokindsofrecognitionsequence.Undertheconditions,whenthetworecognitionsequencesarefullyconsistent,thedistancewillbe0.
4.5RecoveryofTSP
TSPisoneofthemostfamousproblemsingraphtheory.Ifweseeeachoftheshreddingasapoint,thereisadistancebetweenpoints.Inessence,weneedtofindthesmallesttotaldistancepath,whichistofindanoptimalTSPpath.So,therecoveryofshreddingcanbeabstractedintotherecoveryofTSP.
Therefore,wehavethejunction
S.t
X=0or1
where
DistotaldistanceofTSP
isdistancefromitoi+1
BysolvingTSPproblem,youcangetaccesstoeachpointinthesequence,andfinallyuseMATLABtogetoriginalpaper.
4.6SimulateAnneal(SA)algorithm
Simulatedannealing(SA)algorithmisaniterativesolutionstrategyontherandomsearchalgorithm,itisbasedonthephysicalannealingprocessofsolidmaterialandthegeneralsimilarityofcombinatorialoptimizationproblems.Thenameandinspirationcomefromannealinginmetallurgy,atechniqueinvolvingheatingandcontrolledcoolingofamaterialtoincreasethesizeofitscrystalsandreducetheirdefects.Theheatcausestheatomstobecomeunstuckfromtheirinitialpositionsandwanderrandomlythroughstatesofhigherenergy;theslowcoolinggivesthemmorechancesoffindingconfigurationswithlowerinternalenergythantheinitialone.TheSAcanbedescribedasfollows:
Step1.Initialization.Giventhescopeofmodelforeachparameters,randomlyselectedaninitialsolution
andcalculatethecorrespondingtargetvalueE(
);settheinitialtemperature
finaltemperature
makearandomnumber∈(0,1)asaprobabilitythreshold,setthecoolingfunctionT(
+1)=γ•T(
),inwhich,γisannealingcoefficient,
isthenumberofiterations.
Step2.AtacertainTtemperature,makeaperturbationΔx,thenanewsolutionis
=
+
produced,calculatethedifferenceΔE(
)=E(
)−E(
).
Step3.IfΔE(x)<0,xisaccepted;ifΔE(
)>0,
isacceptedaccordingtoprobabilityp=exp(−ΔE/
•T),
isaconstantandusuallytakenthevalue1.Ifp>ε,
isaccepted.When
accepted,
=
Step4.Inacertaintemperature,repeatsteps3.
Step5.ReducethetemperatureTbyslowcoolingfunction.
Step6.Repeatsteps2tostep5,untiltheconditionismeet.
ByusingSAtosolveTSP,wecanregardeachsequenceaseachsolution,astofindtheoptimalschedulingsequence.
4.7One-dimensionalshreddingrestorationalgorithm
Insummary,throughaone-dimensionalshreddingrecoveryalgorithm,itcanautomaticallyrecovertheone-dimensionalshredding.
Algorithmstepsisasfollows:
Step1:
Extractingimagepixelmatrix