|
|
|
- Randell Hunter
- 10 years ago
- Views:
Transcription
1 ComprehensiveColourImageNormalization GrahamD.Finlayson1,BerntSchiele2,andJamesL.Crowley3 1TheColour&ImagingInstitute TheUniversityofDerby 2MITMediaLab UnitedKingdom 3INRIARh^onesAlpes 38330Montbonnot CambridgeMA France USA colourbutareplacedatdierentpositionsthencorrespondingrgbpixels ducestwodierentcolourimages.ifthetwoilluminantsarethesame Abstract.Thesamesceneviewedundertwodierentilluminantsin- arerelatedbysimplescalefactors.incontrastifthelightinggeometry colourchannels(e.g.alltheredpixelvaluesorallthegreenpixels)that areascalingapart.itiswellknownthattheimagedependenciesdueto lightinggeometryandilluminantcolourcanberespectivelyremovedby normalizingthemagnitudeofthergbpixeltriplets(e.g.bycalculating isheldxedbutthecolourofthelightchangesthenitistheindividual thernormalizationsucestoaccountforchangesinboththelighting chromaticities)andbynormalizingthelengthsofeachcolourchannel (byrunningthe`grey-world'colourconstancyalgorithm).however,nei- normalizethergbpixels(toremovedependenceonlightinggeometry) geometryandilluminantcolour. andthennormalizether,gandbcolourchannels(toremovedependence colour.ourapproachisdisarminglysimple.wetakethecolourimageand whichremovesimagedependencyonlightinggeometryandillumination Inthispaperwepresentanewcomprehensiveimagenormalization thisprocessuntilwereachastablestate;thatisreachapositionwhere onilluminantcolour).wethenrepeatthisprocess,normalizergbpixels thenr,gandbcolourchannels,andthenrepeatagain.indeedwerepeat procedurealwaysconvergestothesameanswer.moreover,convergence eachnormalizationisidempotent.cruciallythisiterativenormalization thatappearintheliterature:swain'sdatabase,thesimonfraserdatabase, SangWokLee'sdatabase.Inallcases,forrecognitionbycolourdistributioncomparison,thecomprehensivenormalizationimprovesrecognition reportedintheliterature).alsorecognitionforthecompositedatabase rates(theresultsarenearperfectandinallcasesimproveonresults weconsideredtheobjectrecognitionproblemforthreeimagedatabases isveryrapid,typicallytakingjust4or5iterations. Toillustratethevalueofour\comprehensivenormalization"procedure (comprisingalmost100objects)isalsonearperfect.
2 1Introduction Thelightreachingoureyeisafunctionofsurfacereectance,illuminantcolour andlightinggeometry.yet,thecoloursthatweperceivedependalmostexclusivelyonsurfacereectance;thedependenciesduetolightinggeometryand illuminantcolourareremovedthroughsomesortofimagenormalizationprocedure.asanexample,thewhitepageofabooklookswhitewhetherviewed underblueskyorunderarticiallightandremainswhite,independentofthe positionofthelightsource.whileanalogousnormalizationsexistincomputer visionfordiscountinglightinggeometryorilluminantcolourtheredoesnotexistanormalizationwhichcandoboth,togetheratthesametime.yet,sucha comprehensivenormalizationisclearlyneededsincebothlightinggeometryandilluminantcolourcanbeexpectedtochangefromimagetoimage.a comprehensivenormalizationisdevelopedinthispaper. Imagenormalizationresearchincomputervisiongenerallyproceedsintwo stagesandwewilladoptthesamestrategyhere.first,thephysicsofimage formationarecharacterizedandthedependencyduetoagivenphysicalvariableismadeexplicit.inasecondstage,methodsforremovingthisdependency (thatis,cancelingdependentvariables)aredeveloped.asanexampleofthis kindofreasoning,itiswellknownthat,assumingalinearcameraresponse,if lightintensityisscaledbyafactorsthentheimagescalesbythesamefactor: eachcaptured(r;g;b)pixelbecomes(sr;sg;sb).relativetothissimplephysical modelitiseasytoderiveanormalizationprocedurewhichisindependentofthe intensityoftheviewingilluminant: r r+g+b; g r+g+b; b r+g+b (1) Thenormalizedimagecoloursaresometimesrepresentedusingonlythechromaticitiesr r+g+bandg r+g+b(sinceb r+g+b=1?r?g r+g+b). Thenormalizationshownin(1)iswellused,andwellaccepted,inthecomputervisionliterature(e.g.[SW95,CB97,MMK95,FDB91,Hea89])anddoesan admirablejobofrenderingimagecoloursindependentofthepoweroftheviewingilluminant.asweshallseelater,lightinggeometryingeneral(thisincludes thenotionsoflightsourcedirectionandlightsourcepower)aectsonlythe magnitudeofacapturedrgbandsothenormalizationshownin(1)performs wellindiversecircumstances. Dependencyduetoilluminationcolourisalsoverysimpletomodel(subject tocertaincaveatswhichareexploredlater).if(r1;g1;b1)and(r2;g2;b2)denote cameraresponsescorrespondingtotwoscenepointsviewedunderonecolour oflightthen(r1;g1;b1)and(r2;g2;b2)denotetheresponsesinducedby thesamepointsviewedunderadierentcolouroflight[wb86](thered,green andbluecolourchannelsscalebythefactors,and).clearly,itiseasyto derivealgebraicexpressionswhere,andcancel: (2r1 r1+r2;2g1 g1+g2;2b1 b1+b2);(2r2 r1+r2;2g2 g1+g2;2b2 b1+b2) (2)
3 denominatortermbecomesthesumofallpixelsandnumeratorsarescaledby thatis,to`grey'.forthisreason,equation(2)issometimescalled`grey-world' normalization[hun95,gjt88,buc80]. N.Noticethatafternormalization,themeanimagecolourmapsto(1,1,1); Thetwopixelcasesummarizedin(2)naturallyextendstoN-pixels:the itisusefultostepthroughaworkedexample.let(s1r1;s1g2;s1b1)and dencyonbothlightinggeometryandilluminantcolour.toseethatthisisso, (s2r2;s2g2;s2b2)denoteimagecolourscorrespondingtotwoscenepoints where(s1;s2)and(;;)arescalarsthatmodellightinggeometryandilluminantcolourrespectively.underlightinggeometrynormalization,equation(1), Unfortunately,neithernormalization(1)or(2)sucestoremovedepen- thepixelsbecome:( andunderilluminantcolournormalization,equation(2): ( r1+g1+b1; r2+g2+b2; r1+g1+b1; r2+g2+b2; r1+g1+b1); r2+g2+b2) (3) Inbothcasesonlysomeofthedependentvariablescancel.Thisisunsatisfactory (2s1r1 s1r1+s2r2; s1g1+s2g2; 2s1g1 s1b1+s2b2);(2s2r2 2s1b1 s1r1+s2r2; s1g1+s2g2; 2s2g2 s1b1+s2b2) 2s2b2 sincebothlightinggeometryandilluminantcolourwillchangefromimageto image. (4) solvedusingnormalizedcolourdistributionmanifolds.intheirmethodimages ismodelledexplicitly.theyshowthatthelightinggeometrynormalizeddistributionofcoloursinasceneviewedunderallilluminantcoloursoccupiesacontinuousmanifoldindistributionspace.inlaterworkbyberwickandlee[bl98] arenormalizedforlightinggeometryandthevariationduetoilluminantcolour LinanLee[LL97]proposedthatthisproblem(cancelationfailure)couldbe computationaloverhead.ahighdimensionalmanifoldis,atthebestoftimes, themanifold`onthey'.unfortunatelyboththesesolutionsincurasubstantial `matched'byashiftinilluminantcolour;thismatchingeectivelyreconstructs malizedimagecolourdistributionsaredenedtobethesameiftheycanbe thismanifoldisrepresentedimplicitly.hereapairoflightinggeometrynor- agivenmanifold).similarlythedistributionmatchingsolution,whichoperates unwieldyandimpliescostlyindexing(i.e.todiscoverifadistributionbelongsto atthesametime.ourapproachissimplicityitself.wetakeaninputimageand malizeawayvariationduetoilluminantcolourandlightinggeometrytogether byexhaustivedistributioncorrelation,isveryexpensive. normalizeforlightinggeometryusingequation(1).wethennormalizeforilluminantcolourusingequation(2).wetheniterateonthistheme,successively Inthispaperwedevelopanewcomprehensivenormalizationwhichcannor- stepisidempotent. normalizingawaylightinggeometryandlightcolouruntileachnormalization
4 verges.second,thattheconvergentimageisunique:thesamesceneviewed underanylightinggeometryandunderanyilluminantcolourhasthesamecomprehensivelynormalizedimage.wealsofoundthatconvergenceisveryrapid, typicallytakingjust4or5iterations. Weprovetwoveryimportantresults.First,thatthisprocessalwaysconeratedsyntheticimagesofayellow/greywedgeviewedunderwhite,blueand normalforgrey),90o(halfwaybetweenbothsurfaces)and135o(closetothenormalofyellow).theimagecaptureconditionsareillustratedatthetopoffigure1 Toillustratethepowerofthecomprehensivenormalizationprocedurewegen- orangecolouredlightswhichwereplacedatanglesof45o(closetothesurface bottomoftheguretogetherwithcorrespondingnormalizedimages.lighting ingpositionbutnotilluminantcolour.conversely,illuminantcolournormalized geometrynormalization(equation(1))sucestoremovevariationduetolight- images(equation2)areindependentoflightcolourbutdependonlightingposition.onlythecomprehensivenormalizationsucestoremovevariationdueto lightinggeometryandilluminantcolour. (abluelightat80oisshown).the9generatedsyntheticimagesareshownatthe neath,arealmostthesame.thisexperimentsisrepeatedonasecondimage imagesappear.aftercomprehensivenormalizationtheimages,shownunder- apairoflightinggeometriesandilluminantcolours.noticehowdierentthe showninfigure2.thetoptwoimagesareofthesameobjectviewedunder Examplesofthecomprehensivenormalizationactingonrealimagesare pairwithsimilarresults1. distributions(orinthecaseofourexperimentsbythedistributionofcolours e.g.[so95],[nb93]and[ll97])whereobjectsarerepresentedbyimagecolour nitionparadigmsuggestedbyswainandballard[sb91](whichiswidelyused severalobjectrecognitionexperimentsusingrealimages.weadoptedtherecog- Asayetmorerigoroustestofcomprehensivenormalizationwecarriedout comparison:querydistributionsarecomparedtoobjectdistributionsstoredin adatabaseandtheclosestmatchidentiesthequery. incomprehensivelynormalizedimages).recognitionproceedsbydistribution jee(13database,26queries),brewsterandlee(8objectsand9queries)anda compositeset87objectsand67queries,comprehensivenormalizationfacilitated almostperfectrecognition.forthecompositedatabaseallbut6oftheobjects arecorrectlyidentiedandthosethatarenotareidentiedinsecondplace.this FortheimagedatabasesofSwain(66databaseobjects,31queries),Chatter- performanceisquitestartlinginitsownright(itisalargedatabasecompiledby avarietyofresearchgroups).moreover,recognitionperformancesurpasses,by appliedindividually. far,thatsupportedbythelightinggeometryorilluminantcolournormalizations normalizationispresentedinsection3togetherwithproofsofuniquenessand thenormalizationsshownaboveinequations(1)and(2).thecomprehensive 1AllfourinputimagesshowninFigure2weretakenbyBerwickandLee[BL98] Insection2ofthispaperwediscusscolourimageformationandderive
5 Fig.1.Ayellow/greywedge,showntopofgure,isimagedunder3lightinggeometries and3lightcolours.theresulting9imagescomprisethe33gridofimagesshown izationremovestheeectsofbothilluminantcolourandlightinggeometry(thesingle normalizationsuces(top3imagesinthelastcolumn).thecomprehensivenormal- suces(rst3imagesinthelastrow).forxedlightinggeometry,illuminantcolour imageshownbottomright) topleftabove.whenilluminantcolourisheldxed,lightinggeometrynormalization
6 hensivelynormalized(bottompairofimages).noticehoweectivelycomprehensive Fig.2.Apeanutcontainerisimagedundertwodierentlightinggeometriesandilluminantcolours(topofgure).Aftercomprehensivenormalizationtheimagesappear thesame(2ndpairofimages).apairof`split-pea'images(thirdrow)arecompre- normalizationremovesdependenceilluminantcolourandlightinggeometry
7 convergence.insection4theobjectrecognitionexperimentsaredescribedand resultspresented.thepapernisheswithsomeconclusionsinsection5. 2ColourImageFormation Thelightreectedfromasurfacedependsonthespectralpropertiesofthesurfacereectanceandoftheilluminationincidentonthesurface.Inthecaseof Lambertiansurfaces(theonlyonesweconsiderhere),thislightissimplythe productofthespectralpowerdistributionofthelightsourcewiththepercent sensorresponse: spectralreectanceofthesurface.assumingasinglepointsourcelight,illumination,surfacereectionandsensorfunction,combinetogetherinforminga isthe3-vectorofresponsefunctions(red-,greenandblue-sensitive),eisthe whereiswavelength,pisa3-vectorofsensorresponses(rgbpixelvalue),f p^x;e=ex:nxz!sx()e()f()d (5) illuminationstrikingsurfacereectancesxatlocationx.integrationisoverthe visiblespectrum!.here,andthroughoutthispaper,underscoringdenotesvector ofexmodelsthepoweroftheincidentlightatx.notethatthisimpliesthat surfacenormalatxandexisinthedirectionofthelightsource.thelength onto^xonthesensorarray.theprecisepowerofthereectedlightisgoverned bythedot-producttermex:nx.here,nxistheunitvectorcorrespondingtothe quantities.thelightreectedatx,isproportionaltoe()sx()andisprojected R!Sx()E()F()allowsustosimplify(5): thefunctione()isactuallyconstantacrossthescene.substitutingqx;efor lightinggeometry(butdoeschangewithilluminantcolour).equation(6),which Itisnowunderstoodthatqx;Eisthatpartofascenethatdoesnotvarywith p^x;e=qx;eex:nx dealsonlywithpoint-sourcelightsiseasilygeneralizedtomorecomplexlighting (6) thethecameraresponseisequalto: lightswithlightingdirectionvectorsequaltoex;i(i=1;2;;m).inthiscase geometries.supposethelightincidentatxisacombinationofmpointsource Ofcourse,allthelightingvectorscanbecombinedintoasingleeectivedirection vector(andthistakesusbacktoequation(6)): p^x;e=qx;emxi=1ex;i:nx (7) ex=mxi=1ex;i)p^x;e=qx;eex:nx (7a)
8 sucestomodelextendedlightsourcessuchasuorescentlights[pet93]. equalsthesumoftheresponsestoeachindividuallight.simplethough(7)is,it Sincewenowunderstandthedependencybetweencameraresponseandlightinggeometry,itisascalarrelationshipdependentonex:nx,itisastraightforward mattertonormalizeitaway: Equation(7)conveystheintuitiveideathatthecameraresponsetomlight Whenp^x;E=(r;g;b)thenthenormalizationreturns:(r P3i=1p^x;E i = ex:nxp3i=1qx;e qx;eex:nxi = P3i=1qx;E Itisusefultoviewthedynamicsofthisnormalizationintermsofacomplete r+g+b;g i r+g+b;b r+g+b). (8) image.letusplacethenimagergbpixelsinrowsofann3imagematrix I.Itisclearthat(8)scalestherowsofItosumtoone.ThefunctionR() row-normalizesanimagematrixaccordingto(8): Here,andhenceforth,adoublesubscripti;jindexestheijthelementofamatrix. R(I)i;j= P3k=1Ii;k Ii;j Letusnowconsidertheeectofilluminantcolouronthergbsrecordedby (8a) mattersstillfurthercamerasensorsaredeltafunctions:f()=(?i) (i=1;2;3).undere()thecameraresponseisequalto: acamera.hereweholdlightinggeometry,thevectorsex,xed.tosimplify andundere1(): p^x;e i =ex:nxz!sx()e()(?i)d=ex:nxsx(i)e(i) (9) Combining(9)and(10)togetherweseethat: p^x;e1 i =ex:nxz!sx()e1()(?i)d=ex:nxsx(i)e1(i) (10) recordedineachcolourchannelscalebyamultiplicativefactor(onefactorper Equation(11)informsusthat,asthecolourofthelightchanges,thevalues p^x;e1 i =E1(i) E(i)p^x;E scalars).itisasimplemattertoremovedependenceonilluminationcolour: eachofthered,greenandbluecolourchannels)thenunderachangeinlight colourchannel).ifr,gandbdenotethenvaluesrecordedinanimage(for colourthecapturedimagebecomesr,gandb(where,andare
9 PNi=1Gi=N=3G PNi=1Ri=N=3R N=3G N=3R PNi=1Bi=N=3B N=3B (12) columntosumton=3.thisn=3tallyisfarfromarbitrary,butratherensures sameasthetotalimagesumcalculatedpostrownormalization.thus,inprinciple,animagecanbeinbothrow-andcolumn-normalform(andthegoal IntermsoftheN3imagematrixI,thenormalizationactstoscaleeach thatthetotalsumofallpixelspost-columnnormalizationisnwhichisthe ofcomprehensivenormalization,discussedinthenextsection,isfeasible).the functionc()columnnormalizesiaccordingto(12): izationpresentedin(12)deltafunctionsensitivitieswereselectedforourcamera. Itisprudenttoremindthereaderthatinordertoarriveatthesimplenormal- C(I)i;j=N=3Ii;j PNk=1Ik;j (12a) Whilesuchaselectionisnotgenerallyapplicable,studies[FF96,FDF94b,FDF94a] haveshownthatmostcamerasensorsbehave,orcanbemadetobehave,like deltafunctions. 3TheComprehensiveNormalization Thecomprehensivenormalizationprocedureisdenedbelow: 2.Ii+1=C(R(Ii)) 1.I0=I 3.Ii+1=Ii Initialization Iterationstep Thecomprehensiveprocedureiterativelyperformsrowandcolumnnormalizationsuntiltheterminationconditionismet.Inpracticetheprocesswillterminate whenanormalizationstepinducesachangelessthanacriterionamount. vergenceanduniqueness.theprocedureissaidtoconverge,ifforallimages xedscene)thenuniquenessfollows. terminationisguaranteed.iftheconvergentimageisalwaysthesame(forany Obviouslythisiterativeprocedureisusefulifandonlyifwecanshowcon- Terminationcondition (13) thediscussioninsection2,weknowthatviewingthesamesceneunderadierent lightinggeometryandilluminantcolourusingthetoolsofmatrixalgebra.from Asasteptowardsprovinguniquenessitisusefultoexaminetheeectsof
10 lightinggeometryresultsinanimagewherepixels,thatisrowsofi,arescaled. ThisideacanbeexpressedasanNNdiagonalmatrixDrpremultiplyingI: colourchannels;thatis,ascalingofthecolumnsofi.thismaybewrittenas Ipostmultipliedbya33matrixDc: Similarly,achangeinilluminantcolourresultsinascalingofindividual DrI (14) matrices:ianddridc. underapairoflightinggeometriesandilluminantcoloursinducestheimage Equations(14)and(15)takentogetherinformusthatthesamesceneviewed IDc (15) normalizedcounterpartarerelated: scalestherowsandthenthecolumnsofibypre-andpost-multiplyingwith canbecascadedtogetherandsowendthatanimageanditscomprehensively theappropriatediagonalmatrix.theoperationsofsuccessivenormalizations Bydenition,eachiterationofthecomprehensivenormalizationprocedure wherecomprehensive()isafunctionimplementingtheiterativeprocedureshown in(13)andthesymbolconveystheideaofasequenceofdiagonalmatrices. comprehensive(i)=dridc (16) Proof.LetusassumethatC16=C2.By(16),C1=Dr1IDc1andC2=Dr2DrIDcDc2 andc2=comprehensive(dridc)thenc1=c2(proofofuniqueness). forsomediagonalmatricesdr1,dr2,dc1anddc2.itfollowsthat: Theorem1.Assumingtheiterativeprocedureconverges,ifC1=comprehensive(I) Clearly,foranyDaandDbsatisfying(17)sodokDaand1kDbso,withoutloss C16=C2,DaandDbarenotequaltoidentity(orscaled)identitymatrices. whereda=dr2dr[dr1]?1anddb=[dc1]?1dcdc2.bytheassumptionthat C2=DaC1Db (17) ofgeneralityweassumethatdbi;i>1.wealsoassumethatdb1;1>db2;2>db3;3 ofc1).sincec2iscomprehensivelynormalizedwecanexpressthecomponents (sinceifthisisnotthecaseitcanbemadetruebyinterchangingthecolumns ofdaintermsofdbandc1.inparticulartheithdiagonaltermofdais,and mustbe,thereciprocalofthesumoftheithrowofc1db: Fromwhichitfollowsthat: Dai;i= Db1;1C1i;1+Db2;2C1i;2+Db3;3C1i;3 1 (18) C2i;1= Db1;1C1i;1+Db2;2C1i;2+Db3;3C1i;3 (19)
11 SincehaveassumedthatDb1;1>Db2;2>Db3;3,itfollowsthat whichimpliesthat Db1;1C1i;1+Db1;1C1i;2+Db1;1C1i;3< Db1;1C1i;1+Db2;2C1i;2+Db3;3C1i;3 (20) lessthanthecorrespondingelementininc2.however,thiscannotbethecase Equation(21)informsusthateveryelementintherstcolumnofC1isstrictly C1i;1+C1i;2+C1i;3< Db1;1C1i;1+Db2;2C1i;2+Db3;3C1i;3C1i;1<C2i;1 sincebothc1andc2arecomprehensivelynormalizedwhichimpliesthatthesum oftheirrespectedrstcolumnsmustbethesame(andthiscannotbethecase iftheinequalityin(21)holds). Proof.OurprooffollowsdirectlyfromSinkhorn'sthoerem[Sin64]whichwe Wehaveacontradictionandso,C1=C2anduniquenessisproven. Theorem2.Thecomprehensivenormalization,(13),procedurealwaysconverges. ut thattheprocesswheretherowsofbareiterativelyscaledtosumton=3and invokehereasalemma. thenthecolumnsarescaledtosumton=3(inananalogousprocessto(13))is Lemma.LetBdenoteanarbitrarynnallpositivematrix.Sinkhornshowed guaranteedtoconverge2. letmatrixbbea3n3nwherethen3imagematrixiiscopiedntimes inthehorizontaland3timesintheverticaldirection: First,notethatimages,underanyimagingconditions,areallpositive.Now, B=24III sumtonandthecolumnsofbdcsumton(noten=3n=3).fromtheblock SupposethatDrandDcarediagonalmatricessuchthattherowsofDrB 3 5 (22) structureofb,itfollowsthatdri;i=dri+kn;i+kn(i=1;2;;n)(k=2;3). SimilarlybecausecolumnssumtoN,Dci;i=Dci+kN;i+kN(i=1;2;3)andk= canwritedrbandbdcas: (2;;N).SettingDa=Dri;i(i=1;2;;N)andDb=Dci;i(i=1;2;3),we DrB=24DaIDaIDaI DaIDaIDaI 3 5;BDc=24IDbIDbIDb35 3NN=N=3.ThatiseachIinBisnormalizedaccordingtothefunctionsR() NEachrowinDaIsumsto3 3NN=1andeachcolumninIDbsumsto (23) 2Infactwecouldchooseanypositivenumberhere;n=3willworkwellforourpurposes.
12 andc()andsoaftersucientiterations,sinkhorn'siterativeprocessconverges to: Sinkhorn(B)=24comprehensive(I)comprehensive(I)comprehensive(I) Clearly,Sinkhorn'stheoremimpliestheconvergenceofthecomprehensivenor- (24) 3 5 malizationprocedureandourproofiscomplete. veryrapidily:4or5iterationsgenerallysuces. Experimentally,wefoundthatthecomprehensivenormalizationconverges ut 4ObjectRecognitionExperiments WecarriedoutimageindexingexperimentsfortheSwainandBallard[SB91], imagescombined.swainandballard'simagesetcomprises66databaseand31 SimonFraser[Cha95,GFF95],BerwickandLee[BL98]imagesetsandasetofall queryimages.allimagesaretakenunderaxedcolourlightsourceandthere areonlysmallchangesinlightinggeometry.becausethereare,eectively,no confoundingfactorsinswain'simages,weexpectgoodindexingperformance forlightinggeometry,illuminantcolourandcomprehensivenormalizations.the queryimages.queryimagescontainthesameobjectsbutviewedunderlarge SimonFraserdatasetcomprisesasmalldatabaseof13objectimagesand26 setthereare8objectimagesand9queryimages.againqueriesimagesare changesinlightinggeometryandilluminantcolour.inleeandberwick'simage Thecompositesetcomprises87databaseimagesand67queries. capturedunderdierentconditions(viewinggeometryandlightcolourchange). ages,databaseandquery,weseparatelycarriedoutlightinggeometry,illuminant colourandcomprehensivenormalizations.atasecondstagecolourhistograms, structed.thisinvolveshistogrammingonlythe(g;b)tuplesinthenormalized representingthecolourdistributions,ofthevariouslynormalizedimagesarecon- Totesttheecacyofeachnormalizationweproceedasfollows.Forallim- images.thepixelvaluerisdiscardedbecauser+g+b=1afterlightinggeometryandcomprehensivenormalizations,andsoisadependentvariable.after (G;B)colourspace(whichhavevaluesbetween0and1)denethebinsforthe colourhistograms.ifhiandqdenotesthehistogramsfortheithdatabaseand illuminantcolournormalizationonaverager+g+b=1.a1616partionof queryimagesthenthesimilarityoftheithimagetothequeryimageisdened wherejj:jj1denotesthel1(orcity-blockdistance)betweenthecolourdistributions.thisdistanceisequaltothesumofabsolutedierencesofcorresponding jjhi?qjj1 as: (25)
13 spondstosmalldistances. histogrambins.reassuringly,ifhi=qthenjjhi?qjj1=0;closenesscorre- shouldcontainthesameobjectasthequeryimage).tables1,2and3summarizeindexingperformanceforallthreenormalizationsoperatingonallfourdattionsinthedatabase.thesedistancesaresortedintoascendingorderandthe Foreachquerycolourdistribution,wecalculatethedistancetoalldistribu- rankofthecorrectanswerisrecorded(ideallytheimagerankedinrstplace sets.twoperformancemeasuresareshown:the%ofqueriesthatwerecorrectly matched(%in1stplace)andtherankoftheworstcasematch. ImageSet Swain's SimonFraser LeeandBerwick33.33 %correctworstranking thoutof ndoutof66 Table1.Indexingperformanceoflightinggeometrynormalization Composite thoutof87 ImageSet Swain's SimonFraser LeeandBerwick67.7 %correctworstranking thoutof13 5thoutof66 Table2.Indexingperformanceofilluminantnormalization Composite thoutof87 4thoutof9 ImageSet Swain's SimonFraser %correctworstranking 80.62ndoutof66 Table3.Indexingperformanceofcomprehensivenormalization LeeandBerwick Composite 93.12ndoutof stoutof13 1stoutof9 suggestthatlightinggeometrynormalizationworksbestandthecomprehensivenormalizationworst.thisis,infact,notthecase:allthreenormalizationinggeometrynormalizationsplacethecorrectanswersinthetoptworanks workverywell.noticethatonlythecomprehensivenormalizationandlight- AcursorylookatthematchingperformanceforSwain'simagesappearsto
14 andthisisanadmirablelevelofperformancegivensuchalargedatabase. vastlysuperior,100%recognitionissupportedcomparedwith42.3%and80.8% FortheSimonFraserDatabasethecomprehensivenormalizationprocedureis noothercolourdistributioncomparisonmethodhascomeclosetodelivering are13thand6threspectively.thisisquiteunacceptablegiventheverysmall forlightinggeometryandilluminantcolournormalizations.thelatternormalizationsalsoperformverypoorlyintermsoftheworstcaserankingswhich size(just13objects)ofthesimonfraserdatabase.itisworthnotingthat 100%recognitiononthisdataset[FCF96](thesemethodsinclude,colour-angular indexing[fcf96],ane-invariantsofcolourhistograms[hs94]andcolourconstantcolourindexing[ff95]).thesamerecognitionstoryisrepeatedforthnitionandtheothernormalizationperformverypoorly. BerwickandLeedatabase.Comprehensivenormalizationsupports100%recog- thesewerecompiledbythreedierentresearchgroups.alsothemeansofrecognitionisasimplecolourdistributioncomparisonwhichisboundtofailwhen placematchesthatarerecordedhavecolourdistributionswhicharesimilarto performanceisquitestartling.thedatabaseislargecomprising87objectsand normalizationandtheworstcasematchisinsecondplace.suchrecognition teresting.over93%(of67)queriesarecorrectlyidentiedusingcomprehensive Perhapstherecognitionresultsforthecompositedatasetarethemostin- images,orobjects,havethesamemixtureofcolours.indeed,mostofthe,2nd theoverallbestmatch.forexampleanimageof`campbell'schickensoup'is confusedwithanimageof`campbell'sspecialchickensoup'.bothimagesare colournormalizationperformsbetter,arecognitionrateof79%butagainthe andtheworstcaserankingisanincrediblypoor86th(outof87).illuminant predominantlyredandwhite(asweexpectwithcampbell'ssoup). worstcasematchisunacceptable:16thplacedoutof87. usedindividually,performverypoorly.theformersucceedsjust58%ofthetime Incomparisonthelightinggeometryandilluminantcolournormalizations, Thecoloursrecordedinanimagedependonboththelightinggeometryandthe 5Conclusion colouroftheilluminant.unlesstheseconfoundingfactorscanbediscounted, colourscannotbemeaningfullycomparedacrossimages(andsoobjectrecognitionbycolourdistributioncomparisoncannotwork).inthispaperwedeveloped anewcomprehensivenormalizationprocedurewhichcanremovedependency duetolightinggeometryandilluminantcolourandsocanfacilitatecross-image togetheranditeratively.weprovethatthisiterativeprocessalwaysconverges whichdiscounteitherlightinggeometryorilluminantcolourandapplythem toauniquecomprehensivelynormalizedimage. colourcomparison. Ourapproachissimplicityitself.Wesimplyinvokenormalizationprocedures inasettheobjectrecognitionexperiments.forfourimagedatabases:swain's Thepowerofourcomprehensivenormalizationprocedurewasillustrated
15 post-comprehensivenormalizationwasfoundtobenearperfect.importantly, database,thesimonfraserdatabase,sangwoklee'sdatabase,andacomposite set(withalmost100objects),recognitionbycolourdistributioncomparison, References performancegreatlysurpassedthatachievableusingthelightinggeometryor illuminantcolournormalizationsindividually. [BL98]D.BerwickandS.W.Lee.Achromaticityspaceforspecularity-,illumination [CB97]J.L.CrowleyandF.Berard.Multi-modaltrackingoffacesforvideocommunications.InCVPR97,pages640{645,1997. [Buc80]G.Buchsbaum.Aspatialprocessormodelforobjectcolourperception. JournaloftheFranklinInstitute,310:1{26,1980. pagesession2.3,1998. color-andilluminationpose-invariant3-dobjectrecognition.iniccv98, [Cha95]S.S.Chatterjee.Colorinvariantobjectandtexturerecognition,1995.MSc [FCF96]G.D.Finlayson,S.S.Chatterjee,andB.V.Funt.Colorangularindexing.In thesis,simonfraseruniversity,schoolofcomputingscience. TheFourthEuropeanConferenceonComputerVision(VolII),pages16{27. [FDB91]B.V.Funt,M.S.Drew,andM.Brockington.Recoveringshadingincolor [FDF94a]G.D.Finlayson,M.S.Drew,andB.V.Funt.Colorconstancy:Generalized EuropeanVisionSociety,1996. [FDF94b]G.D.Finlayson,M.S.Drew,andB.V.Funt.Spectralsharpening:Sensortransformationsforimprovedcolorconstancy.J.Opt.Soc.Am.A, 11(5):1553{1563,May1994. images,1991.submittedforpublication. [FF95]B.V.FuntandG.D.Finlayson.Colorconstantcolorindexing.IEEEtransactionsonPatternanalysisandMachineIntelligence,1995. diagonaltransformssuce.j.opt.soc.am.a,11:3011{3020,1994. [FF96]G.D.FinlaysonandB.V.Funt.Coecientchannels:Derivationandrelationshiptoothertheoreticalstudies.COLORresearchandapplication, [GFF95]S.S.ChatterjeeG.D.FinlaysonandB.V.Funt.Colorangleinvariantsfor 21(2):87{96,1996. [GJT88]R.Gershon,A.D.Jepson,andJ.K.Tsotsos.From[r;g;b]tosurfacere- objectrecognition.in3rdis&tandsidcolorimagingconference,pages 44{ [HS94]G.HealeyandD.Slater.\Globalcolorconstancy:recognitionofobjectsby [Hea89]G.Healey.Usingcolorforgeometry-insensitivesegmentation.J.Opt.Soc. Am.A,6:920{937,1989. ectance:computingcolorconstantdescriptorsinimages.ininternational useofilluminationinvariantpropertiesofcolordistributions".journalof JointConferenceonArticialIntelligence,pages755{758,1987. [LL97] [Hun95]R.W.G.Hunt.TheReproductionofColor.FountainPress,5thedition,1995. S.LinandS.W.Lee.Usingchromaticitydistributionsandeigenspaceanalysisforpose-,illumination-andspecularity-invariantrecognitionof3dobject. theopticalsocietyofamerica,a,11(11):3003{3010,november1994. [MMK95]J.Matas,R.Marik,andJ.Kittler.Onrepresentationandmatchingof oncomputervision,pages726{732.ieeecomputersociety,june1995. multi-colouredobjects.inproceedingsofthefthinternationalconference InCVPR97,pages426{431,1997.
16 [NB93]W.NiblackandR.Barber.TheQBICproject:Queryingimagesbycontent [Pet93]A.P.Petrov.Onobtainingshapefromcolorshading.COLORresearchand [SB91] usingcolor,textureandshape.instorageandretrievalforimageandvideo DatabasesI,volume1908ofSPIEProceedingsSeries [Sin64]R.Sinkhorn.Arelationshipbetweenarbitrarypositivematricesanddoubly M.J.SwainandD.H..Ballard.Colorindexing.InternationalJournalof application,18(4):236{240,1993. [SO95]M.A.StrickerandM.Orengo.Similarityofcolorimages.InStorageandRetrievalforImageandVideoDatabasesIII,volume2420ofSPIEProceedings stochasticmatrices.annalsofmathematicalstatistics,35:876{879,1964. ComputerVision,7(11):11{32,1991. [WB86]J.A.WortheyandM.H.Brill.HeuristicanalysisofvonKriescolorconstancy. [SW95]B.SchieleandA.Waibel.Gazetrackingbasedonface-color.InInternational Series,pages381{392.Feb WorkshoponAutomaticFace-andGesture-Recognition,June1995. JournalofTheOpticalSocietyofAmericaA,3(10):1708{1712,1986.
23 280 ก ก 2558. ก ก Hydrologic Cycle
ก 23 28 ก ก 2558 ก ก Hydrologic Cycle ก. ก ก ก ก. ก... ก 65-5533-46-5533-47 ก - ก - 8 ก ก ก ก ก 12-46 ก 17 1 ก ก ก ก ก ก ก 1. ก ก 25 2. ก ก ก ก ก 3. ก ก ก ก ก ก 4. ก ก ก ก 5. 1 31 - 1. ก. ก 2 2. P.7A..
COMBINED LIABILITY INSURANCE CERTIFICATE. effected through. Bluefin Insurance Services Limited CI Tower St. George s Square New Malden Surrey KT3 4TP
COMBINED LIABILITY INSURANCE CERTIFICATE effected through Bluefin Insurance Services Limited CI Tower St. George s Square New Malden Surrey KT3 4TP THIS IS TO CERTIFY that in accordance with the authorisation
0vre Haldenveien 14, 3515 H0nefoss Soldiagram
N /f N 1 Haldenveien 1 Haldenveien 1 Var/hstjevndgn kl 12 1 :5 Var/hstjevndgn kl 16 1 : 5 N /f N 1 Haldenveien 1 Sommersolverv kl 12 1 : 5 Sommersolverv kl 16 1 : 5 Soldiagram...,..,. U:\1.ProsJelrtfti\276HalcSenvcien14
What is a piper plot?
What is a piper plot? A piper plot is a way of visualizing the chemistry of a rock, soil, or water sample. It s comprised of three pieces: a ternary diagram in the lower left representing the cations,
B A R M A G - F R I C T I O N U N I T
B A R M A G - F R I C T I O N U N I T F R I C T I O N U N I T T P _ 8 MT - B A R - 1 1 0 1 F R I C T I O N U N I T T P _ 7 MT - B A R - 1 1 0 2 TYP E 7 F R I C T I O N G ж V DE MT - B A R - 1 1 0 2 A TYP
Rotary actuator for 2 and 3-way (control) ball valves Torque 5 Nm Nominal voltage AC 100... 240 V Control: Open/close or 3-point Auxiliary switch
echnical data sheet Rotary actuator LR-S Rotary actuator for and -way (control) ball valves orque 5 Nm Nominal voltage C... 4 V Control: Open/close or -point uxiliary switch echnical data Electrical data
m Future of learning Zehn J a hr e N et A c a d ei n E r f o l g s p r o g r a m Cisco E x p o 2 0 0 7 2 6. J u n i 2 0 0 7, M e sse W ie n C. D or n in g e r, b m u k k 1/ 12 P r e n t t z d e r p u t
School Year 2014-2015. Pathways for Special Education
School Year 2014-2015 Pathways for Special Education Type refers to whether the course is for a student: in a self-contained class, earning a certificate, or earning a diploma Guide for reading the schedule
Enrolled Copy S.B. 159 ASSESSMENT OFFSET FOR DONATIONS PROMOTING OCCUPATIONAL HEALTH AND SAFETY 2005 GENERAL SESSION STATE OF UTAH
ASSESSMENT OFFSET FOR DONATIONS PROMOTING OCCUPATIONAL HEALTH AND SAFETY 2005 GENERAL SESSION STATE OF UTAH Chief Sponsor: Ed Mayne House Sponsor: Roger E. Barrus LONG TITLE General Description: This bill
Math 1314 Lesson 8 Business Applications: Break Even Analysis, Equilibrium Quantity/Price
Math 1314 Lesson 8 Business Applications: Break Even Analysis, Equilibrium Quantity/Price Three functions of importance in business are cost functions, revenue functions and profit functions. Cost functions
E-COMMERCE: AMONG THE FIRST MOVERS. Patrice Taillandier-Thomas Damien Jamet
E-COMMERCE: AMONG THE FIRST MOVERS Patrice Taillandier-Thomas Damien Jamet Société Générale Day - June 20 th, 2000 1 Contents e-commerce main trends Société Générale s strategy Main developments Société
A Guide to the Private Office Asset Management Ltd Wrap Account
A Guide to the Private Office Asset Management Ltd Wrap Account Contents 1. Getting Started 2 2. Main Portfolio Screen 3 3. Producing a Valuation of your Portfolio 4 4. Producing Historic Valuations 4
Appendix B D&B Rating & Score Explanations
Appendix B D&B Rating & Score Explanations 1 D&B Rating & Score Explanations Contents Rating Keys 3-7 PAYDEX Key 8 Commercial Credit/Financial Stress Scores 9-17 Small Business Risk Account Solution 18-19
E. OZCEYLAN /International Journal of Lean Thinking Volume 1, Issue 1 (June 2010)
I J L T k gv m1 I 1( J 2 0 1 0 ) LT k g. jmg: www. k g.m/ j AD S S m C m T M L g E Öz S ç ku D m I E g g C m 4 2 0 3 1 K T k z @ k.. ABSTRACT KEYWORDS L g m g m T m m m M m k g m m g g. F m T A m m x m
Insurance and Banking Supervision XBRL Implementation in France
Insurance and Banking Supervision Implementation in France Eric JARRY Banque de France [email protected] Eurofiling / Europe Roma 2014-05-05 Agenda European System of Financial Reporting and
Deklaracja podatku od nieruchomości DL-1 schemat
Załącznik nr 2 do uchwały Nr XI/99/2015 Rady Gminy Kolbudy z dnia 22 września 2015r. Deklaracja podatku od nieruchomości DL-1 schemat
Chem 115 POGIL Worksheet - Week 4 Moles & Stoichiometry Answers
Key Questions & Exercises Chem 115 POGIL Worksheet - Week 4 Moles & Stoichiometry Answers 1. The atomic weight of carbon is 12.0107 u, so a mole of carbon has a mass of 12.0107 g. Why doesn t a mole of
Reharmonizations and Substitutions
Reharmonizations and Substitutions 7 th chords and scale harmonization Reharmonizing: Substituting chromatic chords within the iii vi ii V I progression For the sake of variety, coloration and creativity,
Green = 0,255,0 (Target Color for E.L. Gray Construction) CIELAB RGB Simulation Result for E.L. Gray Match (43,215,35) Equal Luminance Gray for Green
Red = 255,0,0 (Target Color for E.L. Gray Construction) CIELAB RGB Simulation Result for E.L. Gray Match (184,27,26) Equal Luminance Gray for Red = 255,0,0 (147,147,147) Mean of Observer Matches to Red=255
Checklist for Meeting Regulatory Requirements: Performance Appraisal Systems
Performance Appraisal Systems INTRODUCTION There are a number of regulatory provisions that guide the design and development of performance appraisal systems. These provisions are presented on the following
東 海 大 學 資 訊 工 程 研 究 所 碩 士 論 文
東 海 大 學 資 訊 工 程 研 究 所 碩 士 論 文 指 導 教 授 楊 朝 棟 博 士 以 網 路 功 能 虛 擬 化 實 作 網 路 即 時 流 量 監 控 服 務 研 究 生 楊 曜 佑 中 華 民 國 一 零 四 年 五 月 摘 要 與 的 概 念 一 同 發 展 的, 是 指 利 用 虛 擬 化 的 技 術, 將 現 有 的 網 路 硬 體 設 備, 利 用 軟 體 來 取 代 其
Residual current devices
Residual devices Residual devices Description The devices offer a wide range of product for all of your fault protection needs. & C large offering for standard instantaneous and selective C and types.
Incorporating Innovative Materials for Seismic Resilient Bridge Columns
Incorporating Innovative Materials for Seismic Resilient Bridge Columns WSDOT Including Contributions from: Dr. M. Saiid Saiidi University Nevada, Reno Brain Nakashoji University Nevada, Reno Presentation
RGB - CGA, EGA, HD to VGA Converter ID# 657
RGB - CGA, EGA, HD to VGA Converter ID# 657 Operation Manual Introduction Features The RGB, CGA, EGA, HD to VGA Converter PCB board is designed for application in Industry and Gaming. Used for the easy
PAMANTASAN NG LUNGSOD NG MAYNILA (University of the City of Manila) Intramuros, Manila NOTICE OF VACANT ADMINISTRATIVE POSITIONS
PAMANTASAN NG LUNGSOD NG MAYNILA (University of the City of Manila) Intramuros, Manila NOTICE OF VACANT ADMINISTRATIVE POSITIONS POSITION Electrician Foreman SG OFFICE/COLLEGE 9 Physical 949 High 1 year
Neutralization of Acid Mine Drainage Using Stabilized Flue Gas Desulfurization Material
Neutralization of Acid Mine Drainage Using Stabilized Flue Gas Desulfurization Material W. Wolfe 1, C.-M. Cheng 1, R. Baker 1, T. Butalia 1, J. Massey-Norton 2 1 The Ohio State University, 2 American Electric
8. Relax and do well.
CHEM 1314 3:30 pm Section Exam II ohn II. Gelder October 16, 2002 Name TA's Name Lab Section INSTRUCTIONS: 1. This examination consists of a total of 8 different pages. The last three pages include a periodic
How To Get A Better Price For Your Phone In Orange (European)
Data Centre Networking at Orange Business analysis Roberto Kung vice president research center core network, Orange Labs First Workshop on Data Center - Converged and Virtual Ethernet Switching (DC CAVES),
Zsolt Katona. Last Update: July, 2013
Zsolt Katona Haas School of Business Phone: +1 (510) 643 1426 University of California, Berkeley Fax: +1 (510) 643 1420 Berkeley, CA Email: [email protected] 94720-1900 Web: http://faculty.haas.berkeley.edu/zskatona
Bachelors of Science-Computer System Engineering (Information Assurance)
Bachelors of Science-Computer System (Information Assurance) Program Description The computer systems engineering program is concerned with the analysis, design and evaluation of computer systems, both
Bachelors of Science-Electrical Engineering
Bachelors of Science-Electrical Engineering Program Description The goal of the electrical engineering undergraduate program is to prepare graduates for positions as electrical engineers. Following initial
Site Review Instrument: Outpatient and Outpatient Rehabilitation Treatment Programs Part 822-4
Site Review Instrument: Outpatient and Outpatient Rehabilitation Treatment Programs Part 822-4 I. PATIENT CASE RECORDS Dave Herbert ~ Program Review Unit Manager Laura Higgins ~ Technical Assistance Specialist
The Accounting Cycle. Chapters 4 and 3
The Accounting Cycle Chapters 4 and 3 Accumulate... Accumulate... Communicate! Business Transactions Source Documents Analyzed Journalized Posted The Accounting Cycle 1. Business transactions create source
Mark Scheme (Results) June 2010
Mark Scheme (Results) June 2010 GCE GCE Leisure Studies (6967/01) Unit 2: Working Practices in Leisure Edexcel Limited. Registered in England and Wales No. 4496750 Registered Office: One90 High Holborn,
Aim for Success. Isaac Kohls Financial Services Representative Princor Registered Representative
Aim for Success Financial Services Representative Princor Registered Representative The Principal Financial Group Minnesota Business Center 8530 Eagle Point Blvd., Suite 125 Lake Elmo, MN 55042 (651) 287-5485
DATING YOUR GUILD 1952-1960
DATING YOUR GUILD 1952-1960 YEAR APPROXIMATE LAST SERIAL NUMBER PRODUCED 1953 1000-1500 1954 1500-2200 1955 2200-3000 1956 3000-4000 1957 4000-5700 1958 5700-8300 1959 12035 1960-1969 This chart displays
A Security Protocol Animator Tool for AVISPA
A Security Protocol Animator Tool for AVISPA Yann Glouche 1, Thomas Genet 1, Olivier Heen 2, Olivier Courtay 2 1 IRISA-INRIA, Rennes, France [email protected] [email protected] 2 Thomson R&D France,
Zsolt Katona. Last Update: January, 2015
Zsolt Katona Haas School of Business Phone: +1 (510) 643 1426 University of California, Berkeley Fax: +1 (510) 643 1420 Berkeley, CA Email: [email protected] 94720-1900 Web: http://faculty.haas.berkeley.edu/zskatona
Survey Instrument Requirements Requirements Definition Template
Survey Instrument Requirements Template Version: 1.0 Mike Foregger, Ricky Kaja As of November 17, 2008 Please Note: This is a working document and is changing as we continue to hold discussions and receive
Chem 115 POGIL Worksheet - Week 4 Moles & Stoichiometry
Chem 115 POGIL Worksheet - Week 4 Moles & Stoichiometry Why? Chemists are concerned with mass relationships in chemical reactions, usually run on a macroscopic scale (grams, kilograms, etc.). To deal with
Victims Compensation Claim Status of All Pending Claims and Claims Decided Within the Last Three Years
Claim#:021914-174 Initials: J.T. Last4SSN: 6996 DOB: 5/3/1970 Crime Date: 4/30/2013 Status: Claim is currently under review. Decision expected within 7 days Claim#:041715-334 Initials: M.S. Last4SSN: 2957
STATEMENT OF WORK AND/OR SPECIFICATIONS STATEMENT OF WORK
STATEMENT OF WORK AND/OR SPECIFICATIONS STATEMENT OF WORK LIVE TACTICAL AIRCRAFT TRAINING IN SUPPORT OF (INSERT GOVT ENTITY) GROUP ONE JOINT TERMINAL ATTACK CONTROLLER TRAINING 4.1. BACKGROUND (INSERT
Tutorial 1J: Chords, Keys, and Progressions
Tutorial 1J: Chords, Keys, and Progressions Welcome! In this tutorial you ll learn how to: Other Level 1 Tutorials 1. Relate chords with keys 1A: Virtual Practice 2. Use basic blues progressions & blues
Know Before You Owe Mortgage Disclosure Rule Construction Lending
Know Before You Owe Mortgage Disclosure Rule Construction Lending Outlook Live Webinar March 1, 2016 Nick Hluchyj Senior Counsel Office of Regulations Kristin Switzer Regulatory Implementation Analyst
20.1.1 Every person who is a British citizen is so either "by descent" or "otherwise than by descent".
Part I: British citizenship Chapter 20: British citizenship "by descent" and "otherwise than by descent" section 14 British Nationality Act 1981 section 3 British Nationality (Falkland Islands) Act 1983
Introductory sub-table (I)(a) As a % of total no. of partly paid-up shares. Held by promoter/promoter group 0 0 0 Held by public 0 0 0
Introductory sub-table (I)(a) Name of the Company: MUTHOOT FINANCE LIMITED Scrip Code, Name of the scrip, class of security: MUTHOOTFIN, EQ Quarter ended: 30 TH SEPTEMBER 2011 Partly paid-up shares:- No.
JAN 2 7 2016. continuous struggle for the department of education to place. teachers in schools having high turnover rates or determined as
THE SENATE TWENTY-EIGHTH LEGISLATURE, 0 STATE OF HAWAII JAN 0 A BILL FOR AN ACT S.B. NO. RELATING TO TEACHERS. BE IT ENACTED BY THE LEGISLATURE OF THE STATE OF HAWAII: 0 SECTION. The legislature finds
WELCOME TO UNIVERSIDAD DE LOS ANDES
WELCOME TO UNIVERSIDAD DE LOS ANDES June 2014 Schools Universidad de los Andes has nine schools: Business Administration, Architecture and Design, Arts and Humanities, Science, Social Sciences, OVERVIEW
Roadside Truck Survey in the National Capital Region
1999/ 2000 Interprovincial Roadside Truck Survey in the National Capital Region Establish a comprehensive database of information Provide a basis for future studies and analysis Does not address specific
Bachelors of Science-Civil Engineering Environmental Engineering
Bachelors of Science-Civil Environmental Program Description Civil engineering involves the analysis, planning, design, construction and maintenance of many types of facilities for government, commerce,
KEY FINDINGS. 2 New trends in global shopping habits. Smartphones are increasingly important during all stages of the consumer journey
KEY FINDINGS 1 Smartphones are increasingly important during all stages of the consumer journey 2 New trends in global shopping habits 3 Social media s influence varies widely by country METHODOLOGY To
ELECTRON CONFIGURATION (SHORT FORM) # of electrons in the subshell. valence electrons Valence electrons have the largest value for "n"!
179 ELECTRON CONFIGURATION (SHORT FORM) - We can represent the electron configuration without drawing a diagram or writing down pages of quantum numbers every time. We write the "electron configuration".
PSTN. Gateway. Switch. Supervisor PC. Ethernet LAN. IPCC Express SERVER. CallManager. IP Phone. IP Phone. Cust- DB
M IPCC EXPRESS Product Solution (IPCC - IP Co n t a c t Ce n t e r ) E i n f ü h r u n g Ü b e r h u nd e r t M il l io ne n N u t ze r - P r o g no s e n zu f o l g e w e r d e n e s in d ie s e m J ah
W Cisco Kompetanse eek end 2 0 0 8 SMB = Store Mu ll ii gg hh eter! Nina Gullerud ng ulleru@ c is c o. c o m 1 Vår E n t e r p r i s e e r f a r i n g... 2 S m å o g M e llo m s t o r e B e d r i f t e
BILL ANALYSIS. Senate Research Center S.B. 1238 By: Hinojosa; Huffman Criminal Justice 3/18/2013 As Filed
BILL ANALYSIS Senate Research Center S.B. 1238 By: Hinojosa; Huffman Criminal Justice 3/18/2013 As Filed AUTHOR'S / SPONSOR'S STATEMENT OF INTENT In 2005, the Texas Forensic Science Commission (FSC) was
Inspired LED Price List
Inspired LED Price List To order any product listed here and not on the website, please call us at (480)941-4286 Inspired LED PN# Description List Price Pro Series Kits & Panel Packs (without lenses) KITS
Il/network/italiano/ Risorse digitali e strumenti colaborativi per le Scienze del'antichità/ Venezia'3'o*obre'2014' Emiliano Degl Innocenti
Il/network/italiano/ From the scriptorium to the screentorium/ DARIAH'GENERAL'VCC'MEETING,'Rome'17@19'September' From the scriptorium to the screentorium/ DARIAH'GENERAL'VCC'MEETING,'Rome'17@19'September'
Draft Lambeth Local Plan 2013. Location of Betting Shops, Pawnbrokers and Money Lenders in Lambeth
Draft Lambeth Local Plan 2013 Location of Betting Shops, Pawnbrokers and Money Lenders in Lambeth April 2013 Contents Page No. Map 1 Table 1 1. Introduction 1 Selected A2 Uses in Lambeth 2 Betting Shops,
E-learning and Student Mobility in Higher Education. BEST Symposium on Education, Gothenburg 2 nd June 10 th June; 2007
E-learning and Student Mobility in Higher Education BEST Symposium on Education, Gothenburg 2 nd June 10 th June; 2007 Ta b l e of Contents Board of European Students of Technology TABLE OF CONTENTS...2
How To Use Excel To Compute Compound Interest
Excel has several built in functions for working with compound interest and annuities. To use these functions, we ll start with a standard Excel worksheet. This worksheet contains the variables used throughout
YOUR EQUIFAX CREDIT REPORT
YOUR EQUIFAX CREDIT REPORT Data Protection Act 1998 Your Equifax Credit Report Data Protection Act 1998 Credit reference agencies Credit reference agencies hold public information, such as electoral roll
4 m m 2m 21 K N Am -K 5K E m m m m K S mm m B m V ms S m S E D m V m 1 m m m m m 2 ( m ) 2 m E mm m m mn A m V mm m m E mm m m K m mm m K 3 495 175 B 19 415 16 66 A D ( 1 23 391)1 928 9 337 S G O 18 3
doi: 10.1016/j.jocn.2010.10.005
doi: 10.1016/j.jocn.2010.10.005 A remote desktop-based telemedicine system Yasushi Shibata, MD, PhD Department of Neurosurgery, Mito Medical Center, University of Tsukuba Mito, Ibaraki, 310-0015, Japan
Mark Scheme (Final) January 2012. GCSE ICT 2010 (5IT01) Paper 1 Living in a Digital World
Scheme (Final) January 2012 GCSE ICT 2010 (5IT01) Paper 1 Living in a Digital World Edexcel and BTEC Qualifications Edexcel and BTEC qualifications come from Pearson, the world s leading learning company.
Engineering Economy. Time Value of Money-3
Engineering Economy Time Value of Money-3 Prof. Kwang-Kyu Seo 1 Chapter 2 Time Value of Money Interest: The Cost of Money Economic Equivalence Interest Formulas Single Cash Flows Equal-Payment Series Dealing
UK Register of Expert Witnesses Expert Witness Year Book 2015
UK Register of Expert Witnesses Expert Witness Year Book 2015 Dr Chris Pamplin, Editor UK Register of Expert Witnesses 2015 UK Register of Expert Witnesses J S Publications 11 Kings Court Newmarket Suffolk
Amendment No. 1 to SB0211. Bell Signature of Sponsor. AMEND Senate Bill No. 211* House Bill No. 393
Senate Government Operations Committee 1 Amendment No. 1 to SB0211 Bell Signature of Sponsor AMEND Senate Bill No. 211* House Bill No. 393 by deleting all language after the enacting clause and substituting
