Using Lossy JPEG 2000 Compression For Archival Master Files



UsingLossyJPEG2000Compression

ForArchivalMasterFiles



Prepared

forthe

LibraryofCongress

OfficeofStrategicInitiatives

by

RobertBuckley

NewMarketImaging



Version1.1

March12,2013



PrimeContract/SolicitationNo.:LCOSI10T0057

PrimeTaskOrderNo.OSI12T0010

1



Contents

Abstract......................................................................................................................................................... 3

Introduction .................................................................................................................................................. 3

Specifyingcompressionbysize..................................................................................................................... 4

NationalDigitalNewspaperProgram ................................................................................................... 5

TheBritishNewspaperArchive............................................................................................................. 7

WellcomeDigitalLibrary..................................................................................................................... 10

Specifyingcompressionbyquality..............................................................................................................11

QualityFactor–Luratech,Adobe .......................................................................................................12

Slope–GoogleBooks,InternetArchive .............................................................................................. 13

PSNR–HarvardLibrary ....................................................................................................................... 15

Discussion ................................................................................................................................................... 16

Conclusions .................................................................................................................................................20

References .................................................................................................................................................. 22



2



Abstract

Thisreportdocumentsthestudythatreviewedhowdifferentinstitutionshaveusedthecontrolsettings

andcommandlineparametersavailablewithvariousJPEG2000encoderstocreatevisuallylossless

archivalandproductionmasters.Thestudyhasconfirmedtheimage‐dependentnatureofthecontrols

andconsequentlytheneedforhumanobserverstojudgewhatse

ttingsareappropriateforanimage

typeorcollection.Almostallmemoryinstitutionsthatuse lossycompressionusecon trols th atspecify

compressionbysize,whichgeneratesresultsthatareportablebetweenimplementationsand

institutions.GoogleBooksandtheprojectsassociatedwithitwhichuseslopetospecifycompressionby

quality.HarvardLibraryus

esaPSNR‐basedcontrolprocessforcompressionthatrunsautomaticallybut

mustbemonitoredforknownfailurecasesofaglobalmetricforassessingimagequality.Forthetime

being,thereisnoviablealternativetoimage‐dependentobserver‐basedcontrolstrategies,although 

visualdiscriminationmetri

cscontinuetoofferthepromiseofautomaticandsupervisedmethodsfor

visuallylosslesscompression.Thesemetricsneedtobemindfulofthecharacteristicsofthedisplayused

inthevisualassessmenttaskandtheireffectontheacceptabilitythresholdsforlossycompressionin

archivalmasters.

Introduction

Attheheartofanarchivalsystemisthepreservationorarchivalfile.It“representsthebestcopy

producedbyadigitizingorganization,”where“best”isdefinedwithrespecttotheobjectivesofa

particularprojectorprogram[1].Further,itrepresents“digitalcontentthattheorganizationintendsto

maintainforthelongter

mwithoutlossofessentialfeatures.”Forstillimagecontent,thearchival

masterhastraditionallybeenanuncompressedTIFFfile,asthishasbeenbelievedtobethebestwayto

preserveessentialfeaturesascapturedintheoriginalscan.Howeverformassdigitizationprojects,using

acompressedformatreducescostscomparedtoanuncompressedformatandisthuse

conomically

attractive.

Theusualargumentsagainstusingacompressedimageformatforanarchivalfilearethatitmakesthe

fileanditscontentsmorevulnerabletoerrors,whetherthecompressionislosslessorlossy,andthat

lossycompressionintroducesdifferencesfromthedigitizedoriginalthatmayinterferewithachievinga

project’sobjectivesandthatcoul

dgrowwithsubsequentmigrations.Howeveroncetheseconcerns

havebeenexaminedintermsoftheubiquityoferrorcorrection[2],noiseinthedigitizationprocessand

theresilienceofimage‐basedtaskstocompression,lossycompressionbecomesnotonlyeconomically

attractivebutalsoacce

ptableforarchivalmastersofdigitizedoriginals.Inmostcases,preservingthe

originalscanwithoutcompressionisnotwarrantedfromaqualitystandpoint.Ifsubstitutingalossy

compressedimageforanuncompressedimagedoesnotaffectperformanceofagiventask,thenthe

lossycompressioncanbedescribedas“functionallylossless.”Invisualassess

menttasks,theterm

“visuallylossless”iscommonlyusedandisthevisualequivalentof“functionallylossless.”

3



Theorgani

zationsthathaveadoptedacompressedformatforthearchivalmastersinmassdigitization

projectsforthemostpartareusingJPEG2000asthecompressionmethod

.SomeuselosslessJPEG

2000compression,whichisonestepremovedfromuncompressed,offeringreversiblecompressionso

thatthedigitizedoriginalisrecoverable.OrganizationsareincreasinglyusinglossyJPEG2000

compressionforthearchivalmaster,especiallyformassdigitizationprojects.Whethertheyuselossless

orlossycompressionforthearchivalmast

er,mostalluselossyJPEG2000fortheproductionmasterfile.

ForthosewhouselossyJPEG2000compressionforthearchivalmaster,thismeansthesamefilecan

serveasboththearchivalandproductionmaster.

Whenlossycompressionisacceptedforthearchivalimagemaster,thequestionbecomeshowmuch

lossycompressiontouse(andhowdoyouspecifyit).Thetwobasicmet

hodsofspecifyingcompression

are“bysize”and“byquality.”Bysizemeansspecifyingthesizeofthecompress edfileorthe

compressionratio;byqualitymeansusingacontrolthatismonotonicallyrelatedtothequalityofthe

compressedim

ageasmeasuredusingametricsuchasPSNR

.Therelationshipbetweenthetwo

methodsisnotasimpleoneandvarieswithimagecontent.

Specifyingcompressionbysize

JPEG2000,unlikeJPEG,makesitpossibletocompressafiletoadesiredsize.Thesizecanbespecified

directlyinbytes,orindirectlyintermsofacompressionratiooracompressedbitrate.Thecompression

ratioisthesizeoftheuncompressedfiletothesizeofthecompressedfile,whichcontainsboth

metadataandimageda

ta.(Compressionfactoristhereciprocalofcompressionratio.)Thecompressed

bitrateistheratioofthecompressedimagedatasizetotheproductofthewidthandheightofthe

image

.Inwhateverwaythefilesizeisspecifiedandpassedtothecoder,thecoderachievesthedesired

resultbyretainingonlyasmuchimagedataasisneededtomeetthetargetfilesize.

Ifa24‐bitoriginalimageiscompressedwithacompressionratioof8:1,thenth

ecompressedbitrate

willbeclosetobutnotprecisely3.0.Thisisbecausecompressionratioisbasedonfilesizewhile

compressedbitrateisbasedonthecompressedimagedatasize,whichislessthanthefilesize.

Similarly,ifthecompressedbitrateisspecifiedas3.0,thenthecompressionratiowillno

tbeprecisely

8:1.Thedifferencesareusuallysmallenough thats tatingthecompressionratioasthenumberofbits

perpixelintheuncompressedimagedividedbythecompressedbitrateisconvenientandnot

significantlydifferentfromthevalueobtainedusingthedefinition.

Theres

tofthissectiondescribestheapproachesthatthreeinstitutionshavetakentoestablishinghow

muchcompressiontousewhenspecifyingcompressionbysize.





Some use TIFF with LZW lossless compression and a few use Baseline JPEG, which is a lossy compression

format more often used for access and delivery.

PSNR or Peak Signal to Noise Ratio is 10 times the logarithm of the square of the maximum pixel value (peak

signal) divided by the mean squared error (noise), and is expressed in dB or decibels. The error is the difference

between the image before and after compression/decompression; for lossless compression, the error is zero.

The definition of compressed bit rate is from Kakadu v6.0.

4



NationalDigitalNewspaperProgram

TheNationalDigitalNewspaperProgram(NDNP)isapartnershipbetweentheNationalEndowmentfor

theHumanitiesandtheLibraryofCongresstoprovideonlineaccesstothedigitizedcontentofhistoric

newspapersviahttp://chroniclingamerica.loc.gov/.Newspaperpagesarescannedingrayscaleat8bits

perpixeland300or400dpi.NDNPusesuncompressedTIFFforarchivalmastersofthepagescansand

lossyJPEG2000compressionforproductionmasters;theJPEG2000productionprofileisdescribedin

[3].SincetheuncompressedTIFFfilefromthescanisbeingretainedasth

earchivalmasterduringPhase

1oftheProgram,therewasnoapriorirequirementfortheproductionmastertobelosslessand

“visuallylossless”compressionwasacceptable.Inthiscase,theintendedusesoftheproductionmaster

areviewingandprinting.WhiletheprogramdoesnotuseJPE

G2000forarchivalmasters,theprocess

thatitusedtospecifylossyJPEG2000compressionwasamongthefirsttobepubliclydocu mentedand

canbeadaptedforusewitharchivalmasters

.

Tojudgetheeffectofcompressiononvisualscreenappearance,aseriesofimageswasgeneratedby

applyingdifferentcompress ionratiostoselectedtestimages.Inparticular,representativehalftoned

image,line‐artandtextareaswereselectedfromagreedupontestimages.Foreacharea,tenimages

wereprovidedforviewingandevaluatio

n:anuncompressedoriginalandninevariantscompressedto2,

1.33,1,0.75,0.67,0.5,0.4,0.32and0.25bitsperpixel,correspondingtocompressionratiosof4,6,8,

10.67,12,16,20,25and32to1for8‐bitgrayoriginals.Figure1showstheimagesofthetes

tareas;for

detailsseeAppendixCof[3].

  

(a) (b) (c)

Figure1.Testimageswith(a)halftoned,(b)line‐artand(c)textareas.

ThecompressedimagesamplesweredeliveredtotheLibraryofCongressfortheirreviewandto

establishqualitythresholdsintermsoftheirapplication. The4:1and6:1compressedimageswere

judgedvisuallylosslesswhenviewedonascreen;onlyexperience

dviewerscouldlocatecompression





This description of the process is based on Section 3.1 of [3]

5



artifactsintheseimages.Th

eimagequalityexhibitedby8:1and10.67:1wasjudgedpreferable.

Image

qualityat16:1wasacceptable,althoughitwasfeltthattheartifactsandlossofresolutioncouldmake

extendedreadinguncomfortable.Eventheimagequalityat32:1wasjudged useableforsomepurposes.

Theevaluationsfocusedontextquality;halftonequalitywasnotjudgedtobeasimportant.Verylittle

differencewa

snotedintheprintoutsoverthevaryingcompressionlevels.Overallprintqualitywas

adequatebutlessimportantthanvisualscreenpresentationquality.

Asaresult,NDNPadopted8:1compressionfortheJPEG2000productionmasters.Thiswasjudgeda

goodcompromisebetweenfilesizeandimagequality.Sin cehighercompressio

nratiosmaybe

acceptableforsomepurposes,qualitylayerswereintroducedtomakeitpossibletoeasilyobtainimages

atarangeofcompressionratiosbetween8:1and32:1(bitratesbetween1and0.25bitsperpixel).

Sofarnoissueshavebeenreportedwiththeuseofcompressioninth

eproductionmasters,evenwhen

theuser‐selectedviewsderivedfromthemaretranscodedtoJPEGfordeliverytotheuserbrowser.Asa

rule,scanqualityisamoreimportantfactorinimagequalityandusersatisfactionthancompression

qualityatthislevelofcompression.

Itisworthno

tingthatforafixedcompressedbitrate,theamountofdistortionthatcompression

introducesintoanimagedependsontheimagecontent.Forexample,themorecomplexanimage,the

moredetailorhighspatialfrequencycontentithasandthemorecompresseddata(detail)needstobe

discardedtoachievethetargetcompressedfilesizecomparedtoalesscompl

eximagewithlessdetail.

Thesedifferenceswillbee videntinthePSNRvalues,whichwillbelowerforthemorecompleximages

thanforthelesscomplexoneswhenbotharecompressedtothesamebitrate.However,forthetyp

es

ofimagecontentusedhere—halftoned image,line‐artandtext—andatthesecompressionratios,no

significantdifferenceswer ereportedbetweenthequalityoftheimagecontentatthesamebitrate.

Thesecontenttypesarefairlyhomogeneousinthesetestimages.

AlsoworthnotingintheNDNPuseofJPEG2000encodingisthegeneroususeofqualitylayers.The

LibraryofCongressobservedthathighercompressionratiosmaybeacceptableforsomepurposes.

Usinglayersmakesitpossibletoobtainreduced‐quality,full‐resolutionversionsoftheproduction

masterwithcompressionratiosb

etween8:1and32:1,equivalenttobitratesbetween1and0.25bits

perpixel.

Formaximumqualityatmaximumresolution,theuserwoulddirectthedecodertodecompressall

layersinreconstructingtheimage.However,atlowerresolutionsandforthumbnailimages,theuser

coulddecompressfewerlayerstoobtainadecompressedimagewithoutobje

ctionablevisualartifacts.

Theuseoflayerseffectivelyenableshighercompressionratiostomeettheneedsoflowerresolution

andlowerqualityapplications.Usingfewerlayersmeanslesscompresseddatatoaccessand

decompress,whichspeedsupperformance.[4]

Altogether,theNDNPprofilespecifies25layersthatcoverthera

ngefrom1to0.015625bitsperpixel,

ortheequivalentcompressionratiorangeof8:1to512:1.Thebitratesforthelayerswereselectedso

thatthelogarithmsofthebitrates(orthecompressionratios)areclosetobeinguniformlydistributed

6



betwee

nthemaximumandminimumvalues.Figure2plotsthebitratesandcompressionratiosagainst

layers.Theexactbitratevaluesorcompressionratiosarenotcritical.Whatismoreimportantisthe

rangeofvaluesandtherebeingsufficientvaluestoprovideanadequatesamplingwithin therange.



(a) (b)

Figure2:Plotsof(a)Log

ofBitRateand(b)Log

ofCompressionRatio(CR)vs.Layer

Withinafixedbitbudgetmethodsuchasspecifyingcompressionbysize,themorelayersthereare,the

morebitsareneededtosignalthelayerstructureandthefewerbitsareavailabletorepresentthe

compressedimagedata.Withlesscompressedimagedata,thegreaterthedi

fferencebetweenthe

originalandtheimagethatcanbereconstructedfromthecompressedimagedata.However,forthe

sampleimagesintheNDNPtests,thedifferencesbetweenusing1layerandusing25werenegligible.

Compressingwith25layersinsteadof1layerwasaboutthesameasusingaco

mpressionratioofalittle

lessthan8.1insteadofacompressionratioof8.Thisisarelativelysmallpricetopaytoobtainthe

advantagesofqualityscalability.Futureapplicationsmayusefewerlayersandconsequentlyfewer

qualitysamplingpoints.

TheBritishNews

paperArchive

TheBritishNewspaperArchive(BNA)isapartnershipbetweentheBritishLibraryandbrights olidonline

publishingtodigitizeupto40millionnewspaperpagesfromtheBritishLibrary'scollectionoverthenext

10years[5].Pagesarescannedincolorat24bitsperpixelsand400dpi.Th

epageimagesarethen

compressedusingJPEG2000.

UnlikeNDNP,whichusesuncompressedTIFFforarchivalmastersandlossyJPEG2000forproduction

masters,theBNAuseslossyJPEG2000forbotharchivalandproductionpurposes.Whilerecognizing

thatthereisatrade‐offbetweenthequalityofaco

mpressedfileanditssizeandconsequentlythecost

offilestorage,theBritishLibrarydecidedthatlossycompressionwasnotonlyeconomicallydesirable

butalsoacceptable.Fur ther,theydecidedthatitwasdesirabletohavethesamemasterfilesupport

botharchivalandproductionuses,whichsimplifiesfilemanagement.

Inge

neralterms,aproductionprofileisdesignedforperformanceandfastaccesstotheimagesubsets

thatsupportthetargetedusecases,whileanarchivalprofileisdesignedforimagequalityandfile

integrity.Fortunately,withJPEG2000thetwoarecompatiblesothatoneprofilecanmeetboth

7



purposes.TheprofilethattheBritishLi

braryusesforthemassdigitizationofnewspapersintheBritish

NewspaperArchiveservesforbothproductionandarchivaluse.Forproductionpurposes,thekey

choiceshavetodowiththeuseofprecinctsvs.tiles,progressionorder,theuseofcodestreammarkers

andcodi

ngmodes.Forarchivalpurposes,themainchoiceistheamountofcompressionandhowitis

specified.

TheBritishLibraryprofileuses12qualitylayerswith the9‐7irreversiblewavelettransform[6].The

layerscorrespondtocompressedbitratesof10,8.7,7,5.2,3.4,2.1,1.2,0.6,0.3,0.15,0.075bitsper

pixel(correspondingtoco

mpressionratiosof2.4,2.8,3.4,4.6,7.1,11.4,20,40,80,160,320fora24‐bit

perpixelimage),plusamaximumbitratecorrespondingtotheretentionofalltransformedand

quantizedimagedata.Althoughallthedataisretained,th

ereisstillsomeminimallossbecauseof

round‐offerrorswiththefloatingpointtransformsandfromthequantizationusedinthelo ssy

compressionmodeofJPEG2000:thisistheminimallylossycase.Asinthecaseoflosslesscompression,

thebitratecorrespondingtotheminimallylossycaseisnotknow

nbeforehandanddependsonthe

imagecontent.Thefirstbitratethatisexplicitlyspecifiedis10bitsperpixel,correspondingtoa

compressionratioof2.4:1.Itisconceivablethattheminimallylossycompressionratioforsomeimages

willbehigher,whichmeansthecompressedimagew

illhavefewerthan12layers.Inanycase,the

imagerecoveredbydecompressingalllayerswillcorrespondtotheminimallosswiththeselected

quantizationlevels.

Theminimallylossycasesetsalowerboundontheamountofcompressionlo ss foranimage.Mostif

notallimagescantoleratemoreloss.Usinglayerspe

rmitsincreasingtheamountofcompressionafter

theimageisinitiallycompressedbyeliminatinglayersfromthecompressedimagedata,anoperation

thatcanbeperformedwithoutdecompressing(andrecompressing)theimage.Forexample,eliminating

thetopthreelayerswouldbethesameascompressingtheimagewithacompressionratioof3.4:1.The

alternativeistosettheam

ountofcompressioninadvancebasedonthecharacteristicsoftheimageor

thecollectionfromwhichanimageisdrawn,orinstitutionalfactorssuchasaffordabilityandbudget

constraints.



8



Figure3:DistributionofCompressionRat

iosoverQualityLayers1through12:redsquaresshowa

uniformdistribution;bluediamonds,thedistributionintheBritishLibraryprofile

ThelogarithmsofthecompressionratiosofthelayersintheBri t ishLibrarypr ofile arenotunif

distributedacrosslayers,astheyareintheNDNPpr

ofi



ormly

le.Figure3,whichplotsthelogarithmof

compressionratioontheverticalaxisagainstlayernumberonthehorizontalaxis,showsthe



ing

1,N+2andso

onsincethecompresseddatainagivenlayerisasubsetofthedataintheprece

ding(nextlower

istherangein

whichth

earchivalencoderisexpectedtooperateandarchivalJPEG2000masterswilllikelyhave

is

le

velsofcompression,risingtoa90%costreduction

ata10:1compressionratio.Diminishingreturnssetinatthatpointandthecostsavingscurvebeginsto

leveloff;doublingthecompressionbeyondthatonlyachievesafurther5%incostsavingsattheriskof

significan

tlyincreasingthecostofpoorquality.

differences.Th

eredsquarescorrespondtoauniformdistribution;thebluediamonds,tothe

compressionratiosusedintheBritishLibraryprofile.

Layer12corr

espondstoanimagewithacompressionratioof320:1.Layer11intheBritishLibrary

profilescorrespondstoacompressionratioof160:1.DecompressingLayer11meansdecompress

Layers11and12.Ingeneral,decompressingLayerNmeansdecompressing LayersN,N+

numb

er)layer.Layer2correspondstoacompressionratioof2.4:1.Layer1isshownherewitha

compressionratioof2,althoughasnotedearliertheactualvalueisimagedependent.

Thenon‐uniformdistrib

utionofcompressionratiosintheBritishLibraryprofilewasselectedtoprovide

afinersamplingofcompressionratiosintherange2‐10thanauniformdistribution.This

compressionratiosof10orless.Th

eoptionstillexiststousefewerlayersinsomeproductionusecases;

forexample,smallerimagederivatives,suchasthumbnails,wouldusefewerlayers.

Mostofthecostbenefitsofcompressionareobtainedinthecompressionratiorangefrom2to10.Th

isshowninFi

gure4,whichplotsthesavingsinstoragecostsversuscompressionratio.Theplotisbased

onasimplemodelthatassumescostisproportionaltostoragevolume.Thisisareasonablemodelto

useforthecostofcloudstorage,whichisoftenpricedperunitvolumeperunittime, e.g.TB/

month.

Figure4showsasteepriseincostsavingswithlow



9



Figure4:St

oragecostreductionversuscompressionratio

Pendingadoptionofamethodforsettingthecompressionlevelonanimage‐by‐imageorcollection

basis,theBritishLibraryisusingminimallylossycompression,whichreservestheabilityto“retrofit”or

levelinfuturebydiscardingqualitylaye

rsinarchivedmasters.



ters.Howe

ver,thestorage

requirementforthisvolumeofimagespointedtotheneedforabettercompromisebetweenthecosts

dardsused.

stitutionalpapers,unpublishedworks,mostly20thcentury

ate

rials,mostly

17th,18thand19thcentury

s.

Figure5showssamplematerialsfromtheWellcomeLibrarycollection.

specifyahigh

ercompression

WellcomeDig

italLibrary

TheWellcomeTrustisdevelopingadigitallibrarywiththerequirementtostoreupto30millionimages.

TheypreviouslyuseduncompressedTIFFimagefilesasarchivalmas

ofsecurelong‐termdi

gitalstorageandtheimagestan

Initscollections, theWellcomeDigitalLibraryholds:

 Printedbooks–earlyprintedbooks,modernbooks(monographs),pamphlets,reports

 Archives–personalpapers,in

 Manuscripts–unpublished,handwritte n“manuscriptbooks”andrelatedm

 Artworks–pr

ints,paintings,posters,drawingsandglassslide



  

Figure5.SampleWellcomeLibrarymaterials:Linedrawing,manuscriptpageand pagefromarchive

s

collection.

TheWellcomeTrusthasadoptedlossyJPEG2000compressionforarchivalmastersforlargeproject

andwithitthenotionofusingthesamefileforbothpreservationandproductionina write‐once‐read‐

manyty

peenvironment.JPEG2000waschosenastheimagestorageformatinordertokeeptotal

storagerequirementsatavaluethatrepresentedanacceptablecompromisebetweeneconomicstorage

10



andima

gequality.Inaddition,JPEG2000offersintelligentcompressionforpreservationandintellige

decompressionforproductionandaccess.WhiletheWellcomeLibraryalsousesa

nt

masterJPEG2000file

withmultiplequalitylayers,themainquestioniswhatoveralllevelofcompressionisacceptableto

s,

n,they

ga

imageswer

eviewedat100%onacolorcalibratedmonitor,andsubjectswereaskedto

detectadifferencebetweensuccessivecompressionstepsorbetweenimagesbeforeandafter

im

agedeterminedthecompressionvaluefortheentirecollection.Inresultstodate,a10:1

compressionratiowasfoundtoworkwe

llforbooksandan8:1valueformaterialsfromthearchive

mpression

me

th

e

picallythemostdemandingimagesandthemostdiscerning observersetthe

compressionlevel.Amoredirectwayofachievingacc

eptablequalitylevelswouldbetospecify

Theid

eabehindcompressionbyqualityistosettheamountofcompressionwithamethodthat,unlike



deliverthedesiredbalancebetweenimagequali

tyandreducedstoragefootprint.

SincetheWellcomeDigitalLibrarywantedtousecompressiontomanagetheirstoragerequirement

specifyingcompression“bysize”madesense.Todeterminetheacceptablelevelofcompressio

selectedandviewedrepresentativeimagesfroma“collection,”increasingthecompressionratioin

stepsuntilcompressionartifactsbecom

evisible.Theythenbackedoffonestepandusedthat

compressionratiofortheotherimagesinthecollection[7].Analternateapproachthattheyalsoused

startedwithahighlevelofcompressionthatwasclearlyunacceptableandworkedbackwards,

decreasingthecompressionuntilobtaininganimagethatwas“fitforpurpose.”Theseapproachesare

variationsontheclassicpsychophysicalmethodofli

mitsfordetectingastimulusanddeterminin

threshold.The

compression.

Ifdiffere

ntrepresentativeimagesfromacollectiongavedifferentcompressionthresholds,thenthe

lowestcompressionratiowasusedforalltheimagesinthecollection.Inotherwords,the“worst

performing”

collection.

Anadvantageofasize‐basedapproachtosetti

ngcompressionlevelsisthatitenablesausertoknowor

estimateinadvancehowmuchspaceisneededforstorage,whichisusefulforplanni

ngpurposes.

However,apointthathasemergedfromtheuseofasize‐basedapproachisthatthesameco

levelorratiothatmaybeacceptableforsomeimageswillnotbeforothers.Inotherwords,thesa

compressionlevelcangivedifferentqualitylevelsfordifferentima

ges.Inorderforthesame

compressionleveltoachieveacceptablequalitylevelsacrossasetofimages,itshouldbeapplied

conservativelyortoimageswiththesameorsimilarcontent,whichishowithasbeenappliedin

casesdescribedsofar.Ty

compressionbyquali

ty.

Specifyingcompressionbyquality

specifyingco

mpressionbysize,producesaresultthatdependsonoradaptstotheimagecontent.

CompressionbyqualityisfamiliartoJPEGusers.WhenuserselecttosaveanimageasaJP

EGfile,they

aretypicallypresentedwithaslideralongwhichtheycanadjustapointerfrom“lowquality,most

compression”atoneendto“highquality,leastcompression”attheother.Thehigherthequalityfactor,

thefinerthequantizationofthecoefficientsofthetransformedimageandthesmallerthedifference

11



betwee

nthecompressedandoriginaluncompressedimages.Becauseofcontentdifferences,the

qualityfactorwillgivedifferentcompressionratiosfordifferentimages;specifyingcompressionbysiz

isonlypossiblewithJPEGbyusinganiterativ

same

e

eprocess.Onepotentialdrawbackofquality‐based

approachesisthatthescalesforadjustingqualityha

venotbeenstandardizedandtherearemultiple

Therestofthissectiondescribesdifferentmethodsthatvendorsofferandorganizationshaveusedfor



re

yon

d

00compressio

nofbookpages,theCaliforniaDigital

LibraryandInternetArchivereportedthatonlyaprofessionaldigitalphotographercoulddetect



eswi

ththesame

QualityvaluewillhavethesamevisualqualityorthatthereexistsasingleQualityvaluethatrepresents

thethresholdatwhichcompressionartifactsbecomevisibleacrossasetofimages.

waysofcontr

ollingandmeasuringquality.

specifyingcompressionbyquality.

QualityFact

or–Luratech,Adobe

TheLuratechJP2SDKandcommandlineinterfaceoffersQualityasaparame ter whensavinganimage

asaJP2file.Figure6(a)showstheSaveAsdialogboxfromtheLuraWaveSmartCompresstool,whe

QualityisoneoftheSaveOptions;compressionratioandcompres sedfilesizear

ealsoSaveOptions.

TherangeofQualityvaluesis1to100,with1theminimumand100themaximum.Thelowerthe

Qualityvalue,thecoarserthequantizationofthesubbands ofthewavelet‐transformedimagebe

thesmallest.InthetestsperformedontheJPEG20

compressionartifactsataQualityvalueof70[8].

Figure6(

b)plotsPSNRagainstQualityforthesamplesimagesshowninFigure7

.Forthecolorimagesin

thissample,theplotofPSNRvs.QualityisastraightlineovertherangeofQualityvaluesfrom40to

100;theinterceptandslopeofthelineareimagedependent.Forthegrayscaleimagesinthissample,

theplotwasslightlybowed.Apriorithereisnore

asontobelievethatdifferentimag

 

(a) (b)





The sample images were compressed using LuraWave SmartCompress Version 2.1.05.03 from Luratech.

12



Figure6.Lu

ratechQualitysetting(a)LuraWaveSmartCompressSaveAsDialogBoxwithQuality

Optionsetto70;(b)PlotofPSNRvs.QualityforsampleimagesinFigure7.

AdobePhotoshopalsooffersaQualitysettingwhensavinganimageasaJPFfile,includingaJP2‐

compatibleJPFfile.Howe

ver,thePhotoshopQualitysettingworksdifferentlyfromtheLuratechone. 

Forallimages,theplotofcompressedfilesizevs.Qualityisastraight‐line,withcompressedfilesize 

decreasingasQualitydecreases fromanimage‐dependentmaximumataQualityof100

Save

ty

Q%ofthemaximumfilesize.Thequantiz ationinthePhotoshop

encoderdoesnotchangewith theQualitysetting,sothedifferent‐sizedfilesatdifferentQualitysettings 

areobtainedbydiscardingcompresseddata.Ineffect,PhotoshopQualitycontrolsthecompression

factoronanim

age‐by‐imagebasis.

.ForaQuali

valueofQ,thecompressedfilesizeis

 

 1a34699 card

 

7x5177,24‐bitcolor);card(1542x945,24‐bitcolor);

)

optimiza

tiontoselectasubsetofthecompresseddatathatgivesacompressedf ilewiththespecified



 Woman 691 29516

Figure7.TestImages:1a34699(666

Woman(204

8x2560,24‐bitcolor);691(1306x1294,8‐bitgray);29516(6648x8358,8‐gray

Slope–Goog

leBooks,InternetArchive

Whenspecifyingcompressionbysize,JPEG2000encodersusepost‐compressionrate‐distortion



Selected sample images were compressed using Adobe Photoshop CS3.

13



sizeandth

elowestdistortion[9].Notallcompresseddataequallyeffectsthedistortionorerror,sothat

thecompresseddatatoi ncludefirstinthe



fileisthedatathatgivesthelargestreductionindistortion

forthesmallestincreaseinfilesize.Thisoptimizationproducesaslopevaluethatesti

matesthechange



ope

compresseddatahasbeendiscarded.Theus

eful

rangefortheslopeparameterintheKakaduencoderisbetweenabout49000and53000;the



n

pages.Bycomparison,aslopevalueof52516gaveacce

ptable

qualityratingsonly5%ofthetimewithtextandnon‐textpag

es;theother95%ofthetimethepages



8:1

compressionratiowillhav

edifferentamountsofdistortiondependingontheimageconte

nt,images

hen

en

vironmentthatcan

makesomeclaimtothewayinwhichuserswouldtypicallyviewimages

.KakaduV6.0wasusedto

createthecompressedimageswithslopevaluesbetween49200and52516.



indistortionwiththechangeinfilesize.

Analternativ

etodirectlyspecifyingthecompressedbitrateisdirectlyspecifyingthisslopevalue.Thisis

anoptionintheKakaduandAwarecommandlineinterfaces

.Slopesareinverselyrelatedtobitrates,

sothatthelowertheslope,thehigherthebitrateandthesmallerthedistortion.InKakadu,thesl

valuescanrangefrom0,correspondingtoimagewithallthecompresseddatawithnonediscardedto

65535,correspondingtoanimagewhereallthe

operating

higherth

eslopenumber,thelowerthequality.



AttheJPEG2000SummitinMay2011,GoogledescribedtheiruseofslopevaluesintheJPEG2000

encodingfortheGoogleBooksproject[10].UsingtheKakaduencoder,theyreportedacceptablequality

inday‐to‐dayoperationsataslopevalueof50980;imagesexportedafterimageprocessinguseda

qualityvalueof51492.TheMi

chiganDigitizationProject,whichisr elatedtoGoogleBooks,requiresthat

allcontinuoustoneimagesbedeliveredinJPEG2000formatandthatthedefaultsloperatedistortio

valueforencodinggrayscaleimageswillbe51492[11].InthestudythatChapmanetal.reportedonin

2007[8],aslopevalueof51492gaveperfec

toracceptablequalityratings93%ofthetimewithtext

pagesand87%ofthetimewithnon‐text

werejudgedmarginaloru

nacceptable.

Asamplingofcolorandgray‐scalepageimagesfromtheInternetArchiveshowsthattheywerecreated

withaslopevalueof50995usingKakaduV5.1.Thecorrespondingcompressionratiosrangedfrom

foragrayscalepagewithtex

tandahalftonedillustrationto351:1forablankcolorpagewithsome

show‐through,with52:1foracolorpagewithblack‐and‐whitetext.Justasimageswiththesame

createdusingthesameslo

pevaluewillhavedifferentcompressionratiosdependingontheircontent.

Figure8plotsPSNRagainstslopevalueforthesampleimagesofFigure7.Astheplotclearlyshows,the

relationshipisimage‐dependent.Compressionartifactsarevisibleinallimagescreatedwithaslope

valueof52516.Insomeimag

es,notablyImage691,artifactswerevisibleataslopevalueof50995w

theywereviewedat100%magnificationonalaptopscreeninanuncontrolled



Since the use of visual weights is the default in the Kakadu command line interface, their use must be turned off

when using the slope to minimize distortion in the sense of PSNR or mean squared error.

Uncontrolled here means free form and not measured so the environment may not be able to be replicated. See the

Discussion section for a discussion of the potential for display characteristics to affect compression results.

14



Figure8.PlotsofPSNRvs.SlopeforsampleimagesinFigure7

PSNR–Ha

rvardLibrary

TheplotsinFigures6(b)and8showthattherelationshipbetweenPSNRandtheQualityandSlope

controlsettingsisimagedependent.Foranyoneimage,thisrelationshipcanbecharacteriz edandthen

appliedtoimagesthatsharethesamerelationship,providedtheycanbeidentified,toobtainadesired

targetPSNRva

lue.Analternativeistouseanimagemodelthatallowsananalyticderivationofthe

relationshipbetweencontrolsettings andPSNR(orsomeotherimagemetric).ThedesiredPSNRwould

thenbeobtainedwhencompressingimagesthatthemodeldescribes.Thisleavesasideforthemomen

t

thequestionastohowwellPSNRestimatesvisualqualityacrossdifferentimagessubjecttothesame

treatment,inthiscaseJPEG2000compression.

TheAwareJPEG2000encodercommandlineinterfaceallowsuserstospecifyatargetPSNRvaluewhen

encodinganimage.AttheJPEG2000Su

mmit,HarvardLibraryreportedobtaininghigh‐fidelitycopiesof

pageimagesusingtheAwareJPEG2000encoderwithaPSNRsettingof46db[12].However,theyalso

reportedthatthisapproachcouldproduceoverlycompressed imageswithnoticeablecompression

artifacts.Figure9showsanexamplefrom[12].

15



Figure9.Comparisonof(a)uncompressedand(b)compressedimages;seetextfordescriptionof(b).

Thesefailuresoccurredinimages withlargeuniformareasandabrupttransitionsandthatdepart

significantlyfromtheunderlyingmodel.Toguardagainstthiskindoffailureandover‐compression,

Harvardusesascriptthatcompressesth

efiletoatargetPSNRvalueandthencheckstoseeifthe

compressionratioexceedsapre‐determinedmax imumvalue.Ifitdoes,thenthetargetPSNRvalueis

incrementedbyoneandthefilecompressedagain.Thiscontinuesuntilthecompressedratiofallsbelow

themaximumorun

tilamaximumPSNRvaluereached,inwhichcaseawarningisissued.

ThisbehavioralsohighlightsthelimitationsofaglobalmetricsuchasPSNR—asinglenumberto

describetheeffectofcompressionovertheentireimage,althoughcompressiondoesnotaffectallparts

ofanimageequally.Largerdistortionsan

dveryvisiblecompressionartifactsinsomepartsoftheimage

canbeoffsetbysmallerdistortionsintherestoftheimagetogiveanaveragedistortionfortheentire

imagethatwouldbeacceptableifitoccurreduniformlyovertheimage.

Discussion

Thisstudyhasreviewedhowdifferentinstitutionshaveusedthecontrolsetti ngsandcommandli ne 

parametersavailablewithvariousJPEG2000encoderstocreatevisuallylosslessarchivalandproduction

masters.Thecriterionforjudgingwhetheragivenleveloflossycompressionisacceptableisthatit

introducesnovisibleorfunctionaldiffe

rencesinthedecompressedimagecomparedtotheoriginal

imagebeforecompressionandthatinallthewaysthatmatterthedecompressedisan acceptable 

surrogatefortheoriginal.

16



Thefollowingtablesummarizesth

econtrolsavailableinselectedencodersforspecifyingcompression

bysizeandbyquality.



1. ImageMagickusesJasperforJPEG2000encodingandaddsaquality

parameterthatismappedtothecompressionfactorinJasperbyanon‐linearfunction.

2. GIMPrequirestheOpenJPEGlibraryforJPEG2000encodinganddecoding.

Itisworthnotingthatthedifferentcontrolslistedunder“Compressionbysize”inthetableareall

expressionsofthesameunderlyingmetricandaresimplyrelatedonetoanothersothatCompression

bysizeisastandardizedandeasilyreplicatedapproachtoimagecompression.Different

implementationsproducecomparableandc

ompatibleresults.Bycomparison,thecontrolslistedunder

“Compressionbyquality”inthetablearealldifferent,andthesettingsandresultsforQualityfactorin

particulararenotportablebetweenimplementations.

WhicheverapproachisusedtocontrolJPEG2000compression,theyallgetmappedontotwo

mechanismsthattheencoderuseswh

encreatingcompresseddata:specifyingthesubband

quantizationstepsanddecidingwhichpartsofthecompresseddatastreamareretainedandwrittento

theoutputfile.

Ashasbeenobserved,themappingfromthesettingsorvaluesofanyofthesecontrolstothevisibility

thresholdforcompressionartifactsisimag

edependent.Asaresult,mappingssuchastheonesthat

NDNPortheWellcomeLibraryusesforcompressionratioandthatGoogleBooksusesforslopeare

appliedconservativelyandtoacollectionortypeofimage.Sofartherehasnotbeenamappingthat

appliestoallimagesnoristherelikelytobeonewhenth

ecriterionisdetecting compressionartifacts

sincethesecontrolsandtheassociatedmetricsareglobal.Theyuseasinglenumberthatappliestothe

entireimagewhenvisualdiscriminationisusuallybasedonlocalfeatures.

PSNRisadifferentcontrolfromtheothersinthetab

leinthesensethatone cannotapplyitdirectlyto

theencodingprocessandexpecttogetbackacompressedimagewiththespecifiedPSNRvalue.The

exceptioniswhentheimagefitsthemodelthatwasassumedinthemappingfromPSNRvaluetothe

compressioncontrols.Otherwisetoobtainaco

mpressedimagewithaspecificPSNRvalue,aniterative

processisneededthatgoesthroughmultiplecompress‐comparecyclesuntilthedesiredPSNRvalueis

obtained.Bycomparison,ifyouspecifythecompressionratioorslopevalue,thenyouwillgeta

compressedimagewithth

especifiedcompressionratioorslopevalue.

17



Evenwh

enaspecificPSNRvalueisachieved,itiswellknownthatPSNRisnota reliablepredictorof

imagequalityortheimagequalitydiffer encesacrossimagesandimagetreatments,suchas

compression[13].AmetricthatshowsabetterrelationshiptoimagequalityistheSSIM(Structural

Similarity

)index[14].Besidesbeingaglobalmetric,itcanbeappliedlocally,whichmaybeusefulin

identifyingregionsinanimagewhosestructuralcontenthasbeenmostdisturbedbycompression.A

localmetricwouldbesuperiorinmimickingtheprocessthatobserversusetoevaluateimage

compressionandde

terminethevisibilitythresholdforcompressionartifacts.However,theSSIMindex

ismorechallenginganalyticallythanPSNRandachievingatargetvaluewillrequireaniterativeprocess.

TheglobalSSIMIndexhasbeenusedtoassesstheeffectsofJPEG2000compressiononthescalefrom

visuallylosslesstovisuallylossycompression[14,15].Ourinteresthereisinfindin

gtheboundaryor

thresholdbetweenthetwoforaclassorcollectionofimages.



Figure10.Blockdiagramofsystemforassessingcompressionartifacts

Figure10showsthestepsinvolvedinthevisualassessmentsreviewedinthisstudy.Anoriginalis

captured,compressedusingspecified controlparametersandthendecompressed.Theresulting

decompressedimageandtheoriginalcapturedimagearedisplayedtoanobserverwhoco

mparesthem,

lookingfordifferencesandartifacts.Becausetheobservercanonlybaseanassessmentontheimages

presentedbythedisplay,it hasbeenimplicitthedisplayfaithfullypresentstotheobserverthe

compressionartifactsthatmattertotheassessment.Thisassumptionneedstobeexamined.

Thespatialfrequencyresponse(SF

R)ofthecapturedevicecanaffectthecompressibility oftheimage.

Forexample,decreasingtheSFRorsamplingefficiencywill increasethelosslessorminimallylossy

compressionratioforanimage,allotherthingsbeingequal.Also,whentwoimagesarecompressedto

thesamelossycompressionratio,theim

agewiththelowerSFRwillhavealowermeansquareerror.

TheSFRofthedisplaydevicecanalsoaffectthecompressibilityofanimage,ormorepreciselythevisual

assessmentofthecompressibilityofanimage.Forexample,twoimages,onecompressedandtheother

uncompressed,thataredi

stinguishableononedisplaymaybeindistinguishable onadisplaywitha

lowerSFR.Inotherwords,apoordisplaymaymakeanimagelookmorecompressiblethanitreallyisor

shouldbe.

18



Figure11.Contrast‐ResolutionTestTarget

Figure11showstheContrast‐ResolutionTestTargetthatFranzSig gofRITdesig nedandcreatedfor

assessingthecontrast‐resolutioncharacteristicsofaprinter[16].Thetargetcontains10rowsand10

columnsofconcentriccirclesofvaryingsizeandcontrast.Usingconcentriccirclesmeanstha

tedgesare

availableatallorientations.Thecirclesineachcolumnhavethesamecontrastratio,startingwith100%

inthelefthandcolumnanddecreasinginequal‐sizedlogstepsto1%intheright‐handcolumn.The

averagetonevalueoftheconcentriccirclesis50%,w

hichisalsothetonevalueofthegraystrips

separatingthecolumns.Thecirclesineachrowhavethesameresolution,startingwith0.63linepairs

permm(about32dpi)inthetoprowandincreasinginequallogstepsto6.25linepairspermminthe

bottomrow.

Whenthetar

getresolutionexceedsthedeviceresolution,thecirclescannolongerberesolvedandthe

Moirépatternslikethoseinthelowerleftcornerofthetargetbecomevisible.WheretheSFRofthe

devicefallsoff,low‐contrast,high‐frequencydetailcannotbereproducedandthecirclesble

ndinto

uniformgraypatcheslikethoseinthelowerrightcornerofthetarget.Thisisalsotheregionwhere

19



transformco

derslikeJPEGandJPEG2000operate,coarselyquantizing thehighspatialfrequency

contentoftheimageandeliminatinglowcontrast,highfrequencyedges.

Whatmakesthissignificantisthatinsettingthresholdsforthevisibilityofcompressionartifacts,itis

importanttoknowifartifactsarenotvisiblebe

causethehumanvisualsystemcannotdetectthemor

becausethedisplaycan’tresolvethem.Otherthanreportingwhatdisplay wasusedintesting,the

displayanditscharacteristicsarenotmentionedinreportsofthetestingofcompressionse ttings.This

meansthattheresults,besidesbeingimagedependent,mayalsobedisplayd

ependent.Itisworth

testingdifferencesbetweendisplays(andviewingconditions)toseehowgreatasourceofvariability

theymaybeintheresultsreportedbydifferentinstitutionsandindifferenttests.

Finally,thecapturedevi cecanalsobeasourceofvariabilityintheimagepathofFigure10.Afocuson

compressionasalossyornoisyprocessshouldbea

ccompaniedbyarecognitionthatcaptureisalsoa

noisyprocess.

Conclusions

Asarule,theamountofcompressionthatisacceptabledependsonthetask,theimageandthe

compressionmethod.Thisreporthasreviewedandpresentedresultsbasedonthevisualassessmentof

digitizedimagesfromculturalheritageinstitutionsthatwerecompressedusingJPEG2000tocreate

visuallylosslessarchivalandproductionmasters.Inparticular,ithasreviewedan

danalyzedthe

differentcontrolsettingsandcommandlineparametersthatwereusedbyvariousJPEG2000encoders

tocreatemasterfiles.

Thecontrolscanbeclassifiedasspecifyingcompressionbysizeorbyquality.“Bysize”contro lsare

easiertoshareandreplicatebe

tweenimplementationsandinstituti onsandtheyarethemostwidely

usedmethodforspecifyingcompression.However,theycannotbegeneralizedbetweenimagesor

acrossimageclassesbecausetherelationshipbetweencompressionratioandqualityiscontent

dependent.The“byquality”slopecontrolisrelatedtothe“bysize”controlsbecauseoftheme

chanism

thatJPEG2000encodersusetoobtainacompressedfilewithaspecifiedsize.TheGoogleBooksproject

usestheslopecontrolinapplicationsthatfocusonbooksandtextualmaterialsandwherethesize‐

qualitytradeoffusuallytiltstowardsize(andspeed). HarvardLibraryusesa“byquality”PSNRcontrolin

acompressionprocesstha

t,oncethePSNRvaluehasbeenset,canruna utomatically,althoughitmust

checkforthefailurecasestowhichglobalmetricssuchasPSNRarevulnerable.Theissueisusing

controlsbasedonglobalmetricswhenvisualtaskperformanceisbasedonlocalfeatures.

Thestudyha

sconfirmedtheimage‐dependentnatureofthecontrolscurrentlyavailableand

consequentlytheneedforhumanobserverstojudgethecontrolsettingsthatareappropriateforan

imagetypeorcollection.Thecriterionforanacceptableleveloflossycompressionhasbeenthatthere

isnofunctionaldifferencebetwee

ntheimagesbeforeandaftercompression,where“functi onal”

translatesto“visual“whenthetaskisacomparisonmadebyahumanobserver,asitishere.Forthe

timebeing,thereissimplynosubstituteforahumanobserverwhenthecriterionforanacceptablelevel

ofcompressionisnovisibleorsignifi

cantdifferencebetweenoriginalandcompressedimages.

20



Whiletheco

ntrolsettingstomeettheneedsofthetaskareimagedependent,theycanandhavebeen

appliedtoimagesofthesamegeneraltypeorwithinthesamecollection.Untilanimage typeorclassis

morecloselydefinedandcircumscribedforthepurposesofcompression,thecontrolshavebeen

appliedcons

ervatively.Usingqualitylayersinthecompresseddatadoesnotaffecttheoverallquali ty

whileprovidingoptionsforreducedqualityselectionsthatcansatisfyotherapplicationsorcriteria.

Analternativetoanimage‐dependentobserver‐basedcontrolstrategywouldreducetherelianceon

observers.Thiswoulden

abletheprocessfordeterminingthethresholdofacceptablecompressiontobe

automatedandtoscalemoregracefullyandlessexpensivelyascollectionsandthedemandfor

compressedmastersgrow.Theplacetolookforanalternativetoobserversisavisualdiscrimination

metricthatcanmimichowhumanobserversdetectcompressionartifactsan

ddeterminenear‐threshold

differencesbetweenimages.Ageneralmetricwouldbecostlytodevelop.Fortunatelythemetricfor

thisapplicationcanbeconstrainedbyJPEG2000compressionartifacts,whichholdsthepromiseof

lowerdevelopmentcostsandsuperiorperformance throughtuning.

Whiletheimagedependenceoftheco

mpressionthresholdsisunavoidable,itcanbebetterunderstood

andcharacterizedthroughtheuseofasynthetictarget.Likethemetric,thetargetcanbedeveloped

withJPEG2000compressioninmind.Inaddition,atargetthatincorporatedelementsofthetargetin

Figure11wouldalsomakeitpossibletoassessthedisplayde

pendenceofcompressionthresholds.

Addinganeducationcomponenttothesedevelopmentsoffersthebenefitofincreasingconfidencein

theuseoflossyJPEG2000compressionforarchivalmasters.Thiswouldbeanimportantadditiontothe

costbenefitsofcompression,whicharealreadywellunderstood.  

21



22



References

[1] FederalAgenciesDigitizationGuideline,http://www.digitizationguidelines.gov/glossary.php

[2] RobertBuckley,“ErrorCorrectionandRecoveryofJPEG2000CompressedImagesinDigital

Preservation,”Proc.9thIS&TArchivingConference,Copenhagen,pp.270‐274,June2012

[3] RobertBuckleyandRogerSam,“JPEGProfilefortheNationalDigitalNewspaperProgram,”

availableat

http://www.loc.gov/ndnp/guidelines/docs/NDNP_JP2HistNewsProfile.pdf

[4] RobertBuckley,WilliamStumboandJimReid,“TheUseofJPEG2 000intheInformationPackages

oftheOAISReferenceModel,”Proc.IS&T2007ArchivingConference,pp.24‐28,Arlington,VA,

May2007

[5] http://www.britishnewspaperarchive.co.uk/help/about



[6] SeanMartinandRobertBuckley,“BritishLibraryJPEG2000ProfileforBulkDigitisation,”

presentationattheLibraryofCongressJPEG2000SummitWorkshop,availableat

http://www.digitizationguidelines.gov/still‐im

age/documents/Martin.pdf

[7] ChristyHenshaw,“ReportfromtheWellcomeLibraryandtheJP2K‐UKGroup,”presentationat

theLibraryofCongressJPEG2000SummitWorkshop,availableat

http://www.digitizationguidelines.gov/still‐image/d

ocuments/Henshaw.pdf

[8] StephenChapman,LaurentDuplouy,JohnKunze,StuartBlair,StephenAbrams,Catherine

Lupovici,AnnJensen,andDanJohnston,“PageImageCompressionforMassDigitization,”Proc.

IS&T2007ArchivingConference,pp.37‐42,Arlington,VA,May2007

[9] DavidS.Taubman,“HighperformancescalableimagecompressionwithEBCOT,”IEEE

TransactionsonImageProcessing,vol.9,no.7,pp

.1158‐1170,2000

[10] JeffBreidenbach,“JPEG2000andGoogleBooks,”presentationattheLibraryofCongressJPEG

2000SummitWorkshop,availableathttp://www.digitizationguidelines.gov/still‐

image/documents/Breidenbach.pdf

[11] TheUniversityofMichigan,UniversityLibraryDigitalLibraryProductionServices,Digital

ConversionUnit,http://www.hathitrust.org/documents/UMDigitizationSpecs2

0100827.pdf

[12] BillComstock,“UsingPSNRThresholdstoModulatetheDegreeofLossyCompressioninJPEG

2000Files,”presentationattheLibraryofCongressJPEG2000SummitWorkshop,availableat

http://www.digitizationguidelines.gov/resources/jpeg2000.html



[13] Z.WangandA.C.Bovik,“Meansquarederror:loveitorleaveit?‐Anewlookatsignalfidelity

measures,”IEEESignalProcessingMagazine,vol.26,no.1,pp.98‐117,Jan.2009

[14] Z.Wang,A.C.Bovik,H.R.SheikhandE.P.Simoncelli,"Imagequalityassessment:Fromerro

r

visibilitytostructuralsimilarity,"IEEETransactionsonImageProcessing,vol.13,no.4,pp.600‐

612,Apr.2004

[15] RichardClark,“JPEG2000standardsindigitalpreser vation,” J.Phys.:Conf.Ser.231012015,

2010

[16] FranzSigg,“TestingforResolutionandContrast,”pp.1‐5inTestTargets6.0,RITSchoolofPrint

Media,Nov.2006