数字图像处理英文文献翻译参考.docx

数字图像处理英文文献翻译参考.docx

《数字图像处理英文文献翻译参考.docx》由会员分享，可在线阅读，更多相关《数字图像处理英文文献翻译参考.docx（30页珍藏版）》请在冰豆网上搜索。

数字图像处理英文文献翻译参考

HybridGeneticAlgorithmBasedImageEnhancement

Technology

MuDongzhouDepartmentoftheInformationEngineeringXuZhouCollegeofIndustrialTechnology

XuZhou,Chinamudzh@

XuChaoandGeHongmeiDepartmentoftheInformationEngineeringXuZhouCollegeofIndustrialTechnology

XuZhou,Chinaxuch@,gehm@

Abstract—inimageenhancement,TubbsproposedanormalizedincompleteBetafunctiontorepresentseveralkindsofcommonlyusednon-lineartransformfunctionstodotheresearchonimageenhancement.ButhowtodefinethecoefficientsoftheBetafunctionisstillaproblem.WeproposedaHybridGeneticAlgorithmwhichcombinestheDifferentialEvolutiontotheGeneticAlgorithmintheimageenhancementprocessandutilizethequicklysearchingabilityofthealgorithmtocarryouttheadaptivemutationandsearches.FinallyweusetheSimulationexperimenttoprovetheeffectivenessofthemethod.

Keywords-Imageenhancement;HybridGeneticAlgorithm;adaptiveenhancement

I.INTRODUCTION

Intheimageformation,transferorconversionprocess,duetootherobjectivefactorssuchassystemnoise,inadequateorexcessiveexposure,relativemotionandsotheimpactwillgettheimageoftenadifferencebetweentheoriginalimage（referredtoasdegradedordegraded）Degradedimageisusuallyblurredoraftertheextractionofinformationthroughthemachinetoreduceorevenwrong,itmusttakesomemeasuresforitsimprovement.

Imageenhancementtechnologyisproposedinthissense,andthepurposeistoimprovetheimagequality.FuzzyImageEnhancementsituationaccordingtotheimageusingavarietyofspecialtechnicalhighlightssomeoftheinformationintheimage,reduceoreliminatetheirrelevantinformation,toemphasizetheimageofthewholeorthepurposeoflocalfeatures.Imageenhancementmethodisstillnounifiedtheory,imageenhancementtechniquescanbedividedintothreecategories:

pointoperations,andspatialfrequencyenhancementmethodsEnhancementAct.Thispaperpresentsanautomaticadjustmentaccordingtotheimagecharacteristicsofadaptiveimageenhancementmethodthatcalledhybridgeneticalgorithm.Itcombinesthedifferentialevolutionalgorithmofadaptivesearchcapabilities,automaticallydeterminesthetransformationfunctionoftheparametervaluesinordertoachieveadaptiveimageenhancement.

II.IMAGEENHANCEMENTTECHNOLOGY

Imageenhancementreferstosomefeaturesoftheimage,suchascontour,contrast,emphasisorhighlightedges,etc.,inordertofacilitatedetectionorfurtheranalysisandprocessing.Enhancementswillnotincreasetheinformationintheimagedata,butwillchoosetheappropriatefeaturesoftheexpansionofdynamicrange,makingthesefeaturesmoreeasilydetectedoridentified,forthedetectionandtreatmentfollow-upanalysisandlayagoodfoundation.

Imageenhancementmethodconsistsofpointoperations,spatialfiltering,andfrequencydomainfilteringcategories.Pointoperations,includingcontraststretching,histogrammodeling,andlimitingnoiseandimagesubtractiontechniques.Spatialfilterincludinglow-passfiltering,medianfiltering,highpassfilter（imagesharpening）.Frequencyfilterincludinghomomorphismfiltering,multi-scalemulti-resolutionimageenhancementapplied[1].

III.DIFFERENTIALEVOLUTIONALGORITHM

DifferentialEvolution（DE）wasfirstproposedbyPriceandStorn,andwithotherevolutionaryalgorithmsarecompared,DEalgorithmhasastrongspatialsearchcapability,andeasytoimplement,easytounderstand.DEalgorithmisanovelsearchalgorithm,itisfirstinthesearchspacerandomlygeneratestheinitialpopulationandthencalculatethedifferencebetweenanytwomembersofthevector,andthedifferenceisaddedtothethirdmemberofthevector,bywhichMethodtoformanewindividual.Ifyoufindthatthefitnessofnewindividualmembersbetterthantheoriginal,thenreplacetheoriginalwiththeformationofindividualself.

TheoperationofDEisthesameasgeneticalgorithm,anditconcludemutation,crossoverandselection,butthemethodsaredifferent.WesupposethatthegroupsizeisP,thevectordimensionisD,andwecanexpresstheobjectvectoras

（1）:

xi=[xi1，xi2，…，xiD]（i=1,…,P）

（1）

Andthemutationvectorcanbeexpressedas

（2）:

i=1,...,P

（2）

arethreerandomlyselectedindividualsfromgroup,andr1

r2

r3

i.Fisarangeof[0,2]betweentheactualtypeconstantfactordifferencevectorisusedtocontroltheinfluence,commonlyreferredtoasscalingfactor.Clearlythedifferencebetweenthevectorandthesmallerthedisturbancealsosmaller,whichmeansthatifgroupsclosetotheoptimumvalue,thedisturbancewillbeautomaticallyreduced.

DEalgorithmselectionoperationisa"greedy"selectionmode,ifandonlyifthenewvectoruithefitnessoftheindividualthanthetargetvectorisbetterwhentheindividualxi,uiwillberetainedtothenextgroup.Otherwise,thetargetvectorxiindividualsremainintheoriginalgroup,onceagainasthenextgenerationoftheparentvector.

IV.HYBRIDGAFORIMAGEENHANCEMENTIMAGE

enhancementisthefoundationtogetthefastobjectdetection,soitisnecessarytofindreal-timeandgoodperformancealgorithm.Forthepracticalrequirementsofdifferentsystems,manyalgorithmsneedtodeterminetheparametersandartificialthresholds.Canuseanon-completeBetafunction,itcancompletelycoverthetypicalimageenhancementtransformtype,buttodeterminetheBetafunctionparametersarestillmanyproblemstobesolved.ThissectionpresentsaBetafunction,sinceaccordingtotheapplicablemethodforimageenhancement,adaptiveHybridgeneticalgorithmsearchcapabilities,automaticallydeterminesthetransformationfunctionoftheparametervaluesinordertoachieveadaptiveimageenhancement.

Thepurposeofimageenhancementistoimproveimagequality,whicharemoreprominentfeaturesofthespecifiedrestorethedegradedimagedetailsandsoon.Inthedegradedimageinacommonfeatureisthecontrastlowersideusuallypresentsbright,dimorgrayconcentrated.Low-contrastdegradedimagecanbestretchedtoachieveadynamichistogramenhancement,suchasgraylevelchange.WeuseIxytoillustratethegraylevelofpoint（x,y）whichcanbeexpressedby（3）.

Ixy=f（x,y）（3）

where:

“f”isalinearornonlinearfunction.Ingeneral,grayimagehavefournonlineartranslations[6][7]thatcanbeshownasFigure1.WeuseanormalizedincompleteBetafunctiontoautomaticallyfitthe4categoriesofimageenhancementtransformationcurve.Itdefinesin（4）:

（4）where:

（5）

Fordifferentvalueofαandβ,wecangetresponsecurvefrom（4）and（5）.

ThehybridGAcanmakeuseoftheprevioussectionadaptivedifferentialevolutionalgorithmtosearchforthebestfunctiontodetermineavalueofBeta,andtheneachpixelgrayscalevaluesintotheBetafunction,thecorrespondingtransformationofFigure1,resultinginidealimageenhancement.Thedetaildescriptionisfollows:

Assumingtheoriginalimagepixel（x,y）ofthepixelgraylevelbytheformula（4）,denotedby

hereΩistheimagedomain.EnhancedimageisdenotedbyIxy.Firstly,theimagegrayvaluenormalizedinto[0,1]by（6）.

（6）

where:

and

expressthemaximumandminimumofimagegrayrelatively.

Definethenonlineartransformationfunctionf（u）（0≤u≤1）totransformsourceimagetoGxy=f（

）,wherethe0≤Gxy≤1.

Finally,weusethehybridgeneticalgorithmtodeterminetheappropriateBetafunctionf（u）theoptimalparametersαandβ.WillenhancetheimageGxytransformedantinormalized.

V.EXPERIMENTANDANALYSIS

Inthesimulation,weusedtwodifferenttypesofgray-scaleimagesdegraded;theprogramperformed50times,populationsizesof30,evolved600times.Theresultsshowthattheproposedmethodcanveryeffectivelyenhancethedifferenttypesofdegradedimage.

Figure2,thesizeoftheoriginalimagea320×320,it'sthecontrasttolow,andsomedetailsofthemoreobscure,inparticular,scarvesandotherdetailsofthetextureisnotobvious,visualeffects,poor,usingthemethodproposedinthissection,toovercometheabovesomeoftheissuesandgetsatisfactoryimageresults,asshowninFigure5（b）shows,thevisualeffectshavebeenwellimproved.Fromthehistogramview,thescopeofthedistributionofimageintensityismoreuniform,andthedistributionoflightanddarkgrayareaismorereasonable.Hybridgeneticalgorithmtoautomaticallyidentifythenonlineartransformationofthefunctioncurve,andthevaluesobtainedbefore9.837,5.7912,fromthecurvecanbedrawn,itisconsistentwithFigure3,c-class,thatstretchacrossthemiddleregioncompressiontransformtheregion,whichwereconsistentwiththehistogram,theoveralloriginalimagelowcontrast,compressionatbothendsofthemiddleregionstretchingregionisconsistentwithhumanvisualsense,enhancedtheeffectofsignificantlyimproved.

Figure3,thesizeoftheoriginalimagea320×256,theoverallintensityislow,theuseofthemethodproposedinthissectionaretheimagesb,wecanseetheground,chairsandclothesandotherdetailsoftheresolutionandcontrastthantheoriginalimagehasImprovedsignificantly,theoriginalimagegraydistributionconcentratedinthelowerregion,andtheenhancedimageofthegrayuniform,graybeforeandaftertransformationandnonlineartransformationofbasicgraph3（a）thesameclass,namely,theimageDimregionstretching,andthevalueswere5.9409,9.5704,nonlineartransformationofimagesdegradedtypeinferenceiscorrect,theenhancedvisualeffectandgoodrobustnessenhancement.

Difficulttoassessthequalityofimageenhance