(Luận án) Research and development of digital watermarking methods for authentication and copyright protection

This thesis focuses on watermarking and proposes robust andfragilewatermarkingschemesusingpublickeysinordertoenhancethesecurityandimprovetheaccuracyofimagevalidationaswellasenhancerobust

MINISTRYOFEDUCATIONANDTRAINING MINISTRYOF NATIONALDEFENCE

ACADEMYOFMILITARYSCIENCEANDTECHNOLOGY

Cao Thi Luyen

RESEARCH AND DEVELOPMENT OF

… … …, 2018

Forfurtherdetails:

- Libraryof AcademyofMilitaryScienceandTechnology

- VietnamNationalLibrary

INTRODUCTION

1 Necessities

Nowadays,thestorageandtransmissionofdataismadepopularthroughthe Internetenvironment The problem of digital product infringementcauses a lot of negative impacts to

theotherhand,thedigitalproductistransformedbeforereachingtherecipient,if thereceived data has been transformed previously, then which area ischanged.Thatistheproblemofdigitaldataauthentication.Findingeffectivemethodstosolvetheproblemsofcopyrightprotectionandcontentauthenticationbecomesanurgentneedandreceivesmuchattentionsfromboth academia and society Watermarking is one of the mosteffectiveremediestodealwiththeseproblems

Currentwatermarkingstudiesaimtosuccessfullydevelopawatermarkingsystemthatbestsatisfiesthebasicrequirementsofawatermarkingsystemsuchaswatermarkingimagequality,computationalcomplexity,safety.Theresultsofpublickeywatermarkingarestilllimited

This thesis focuses on watermarking and proposes robust andfragilewatermarkingschemesusingpublickeysinordertoenhancethesecurityandimprovetheaccuracyofimagevalidationaswellasenhancerobustnessof the watermarking schemebesides enhancing the watermarking imagequality

• Developanumberofsharedlibrarymodulesthatembedandextractwatermarkusingtheproposedmethods

robustness,confidentiality,fragilityandlatencyofwatermarkingmethods.Exploit the limitations of several widely usedwatermarkingschemesandproposenewwaystoimprovethem.

•

Designdatasetsandexperimentalmannerstocomparedifferentwatermarkingmethods

5 Researchcontents

• Makingresearch,analysis,evaluationofalgorithms,algorithms,systemsofwatermarkinganddatahidingonexistingdigitalimage,especiallytheSVDdomain,reversibleand publickeywatermarking

• ProposingsomepublickeyrobustwatermarkingmethodsbasedonSVDtransformations

• StudyingwatermarkingschemesforJPEGimages,makeevaluationandenhancementstodevelopahigh-embeddingfracturepublickeywatermarkingscheme

• Research and proposing a fragile, public-key, reversiblewatermarkingapproachbasedonthedifferenceextensioncandetectanysmallchange

• Developing a number of library modules using proposed algorithmstosolvecopyrightprotectionandimagevalidationproblems

6 Scientificandpracticalvalueoftheresearch

The study of public key watermarking methods has implications onbothscientificand practicalaspects:

commonlyshare a secret key in both the embedded process and thewatermarkingtest This method has the disadvantage of beingvulnerable to exposureduring the exchange between the submerger and theuser,t h e r e f o r e , the safety is not high In the public key

usetwokeys:thesecretkeytoembedandthepublickeytotestwatermarking.Thus,Noneedtoexchangekeyswillhelpthewatermarkingsystemmoresecure.Thewatermarkingm e t h o d proposedbytheresearcherwillbebotheasytouseandsafe.Inaddition,theproposedwatermarkingschemewillhavegoodwatermarking quality as well as reduced computationalcomplexitycompared tothecurrentmethod

On the practical aspect: Based on current technology research,

NCSwillproposesomenewpublickeywatermarkingmethods.Thesemethods will be expressed on a mathematical basis and implementedbylibrary mudules, software that performs embedded functionsandseparatesthewatermarkingsignonthedigitalimagedata.Thesemodulesandsoftwarecanbeappliedtoanyinformationservice

provider to protect the integrity of information, restrict distortion,orotherwise misuse them in a public exchange environment as wellascontributetothe preventionof fake photos.

7 Structureofthethesis

Thethesisconsistsofthreechapters,includingintroduction,conclusion,listsofwor ks andappendic es Theproduct of thethesisis07 scientific articles which have beenpublished in the seminars andmagazines in as well as internationally, of which 01 haveSCOPUSindex

CHAPTER1:OVERVIEWOFDIGITALWATERMARKING 1.1 Conceptsofwatermarkinganddatahiding

Data hiding [36] is a technique for embedding information AintoenvironmentB.AandBcanbetext,image,audio,orvideofiles.InformationembeddedAmaycontainconfidentialmessagesthatneedtobeexchangedorinformationaboutcopyright,multimediaproducts

Digitalwatermarkingisthetechniqueofembeddinginformation(watermark)intoanobject(adatahidingenvironment)beforeitisdistributed over the network to protect the messageobjects Messageobjectscanalsobedigitalphotos,textfiles,soundfiles, Watermarkcanthen berestored as evidence to prove legal rights as well as to test digitalproductswhichareinintegrityornotorthereisunauthorizedaccessbeforereachingtheuser

1.2 Importantcharacteristicsofthewatermarkingscheme

Accordingto[36],thewatermarkingsystemshouldmeetthefollowingbasicrequirements:

- Robustness: Watermark must be robust that means, it can bedetectedaftercommonsignalprocessingoperations.Therobustnessrequirementoften is used for selective authentication systems such as semi-fragilewatermarkings y s t e m s [ 1 3 ] , [ 2 3 ] , [ 3 2 ] , [ 6 5 ] , [ 6 9 ] ,[ 7 3 ] o r r o b u s t

watermarking[3],[6],[62],[72],[83]

- Sensitivity: The authentication system should be able to detectanymodifications to the contents or to detect any operation on theimageeven if it is small For sensitive watermarking systems [4], [5],[19],[34], [45], the watermark will be lost after any signalprocessingapplication

- Capacity:Thewatermarksareusuallysmallinsize,butthewatermarkingschememustbeabletodiptheidentifiedwatermark.Themoredataareembedded,thebetterwatermarkingschemeis

- Localization: The watermarking system should be able to identify thelocationoftheimagethathaschanged.

- Abilitytorestore:Thewatermarkingsystemmustbeabletopartiallyorcompletelyrestoretheimagethathasbeentamperedwith

- Security:Thewatermarkingsystemmusthavetheabilitytoprotecttheauthenticdataagainstanyattemptedtampering

- Complexity:Thewatermarkingsystemshouldusetimetoimplementalgorithmsbutnottoocomplicatedandslow

Depending on the type of watermarking system, which featureareconcerned For example, for robust watermarking, the robustness oftheschemeisimportantbutsensitivityisignored,whileforfragilewatermarkingschemes,itsensitivityisimportantandtherobustnessisnotconsideredinthistypeofwatermarking

1.3 Principlesofdevelopingwatermarkingscheme

A watermarking scheme consists of two processes [36]:watermarkembedding process and watermark sign test During the embeddingofwatermark,theownerembedsthewatermarkWintotheoriginalimage𝐹toreceiveawatermarkedimage𝐹’.Forextrasecurity,itispossibletouseK1 key The exchange of

procedure to verify the copyright orverify the integrity of the image𝐹*may

orrestoretheoriginalimage.TestingmayalsorequiretheuseofK 2 key.Ifthetesting

image𝐹*is restored to its original image𝐹, it is called

reversiblewatermarking.IftheK 1 keyusedintheembeddingprocessisasecretkey,theK 2 ke

keywatermarkingmodel.Adetailedwatermarkingclassificationismentionedbelow

1.4 ApplycationofWatermarking

1.5 Overviewofnationalandinternationalresearchondigitalwaterma rking

Inthecountry,thereareanumberofarticlesanddoctoralthesesondatahiding andwatermarkingbut mostly robust watermarking, watermarkingschemes using secret keys

annualInternationalWorkshoponDigitalForensicsandWatermarking(IWDW),whicharepublishedbySpringerPublishingHousingonLectureNotesinComputerSciences

watermarkingschemesinthecurrentresearchdirections

1.5.1 Datahidingandwatermarkingonthespatialdomain

This group of methods, presented on binary images, iscompletelyapplicabletograyandcolorimagesbyseparatinglowbitstoformbinarymatrices Its advantage is low computational speed and fragile, whichisoftenappliedtotheproblemofdatahidingordataintegrity

1.4.1.1 Datahiding,watermarkingbasedontheparity

In 1998, M.Wu and J.Lee proposed a simple data hiding basedbinary(WLalgorithm)andwascitedbymanyworks[54].In2000, Y Cheng et al [80] expandedthe WL algorithm to embed amaximum of r bits per element block𝑚 × 𝑛that onlychanges 2 elements(𝑟=⌊𝑙𝑜𝑔2(𝑚× 𝑛+ 1)⌋)

schemeforparity-1.4.1.2 DatahidingandwatermarkingbasedonLSBtechnique

This technique is based on the observation that if only low bit ischanged,change of image can not be detected visually Thistechniqueinsertsconfidentialinformationintothelowbitof8-

bitpixels.Thetypicalscheme using this technique F5 was proposed by Andress Westfeldin1999,JstegproposedbybyJ.Fridrichetal.[37]proposedin2003,andetc.The advantage ofthis method is high capacity(3-bits embedded per oneimage colour) Its biggest limitation is that it

imagescontainingstatisticalinformationbecausetheimagecontainsahistogramchanges,aswellasthismethodisnotrobustpriortogeometrictransformation.Dataembeddingandwatermarkingschemesbasedonlowbit inserting or parity are fragile so this technique is often used infragilewatermarkingschemes

1.5.2 Datahidingandwatermarkingonthefrequencydomain

Formorerobustwatermarkingscheme,DCT,DWT,DFT,SVD,QR areusedintherobustwatermarkingscheme

1.5.3 Fragilewatermarking

Fragilewatermarking[4],[5],[34],[36],

[45]isatypeofwatermarkingthatthewatermarkiseasilydistortediftheimageisattackedevenbysmallchanges Based on that variation of the watermark to conclude thattheimagehasbeenaccessedillegally,aswellastheintegrityoftheimage.Ifthe image is altered,

Thisusuallyusesimagingtechniquesembeddedinspatialdomainandisappliedtotheintegrityproblemoftheimage

1.5.4 Robustwatermarking

Contrarytothefragilewatermarking,therobustwatermarkingistheonethat requires a

conventionalattackstoe li mi na t e wa t e r m a r ki ng (JPEGcom pre ssion,noise,

f il t er i ng,

[83] commonly embed watermark on the frequency domain.Frequencydomains are usually performed through transformations such asDCT(DiscreteCosineTransform)[4],[33],

[56],SVD(singularvaluedecomposition)[8],[40],[54],[58],[54],[60],

[73],DWT(discretewavelettransform), [6], QR (quick responsion) [31] or a combination ofsaidtransformations [9], [15] , [23], [31], [54], [72] to create newschemes.Next,SVDisnotonlybeusedinlinearalgebra,butalsoinmanydifferentapplications,suchasinimageprocessing,ispresentasbelow

The main benefits of SVD in image processing are that it provides amethod

to decompose a large matrix to smaller and moremanageablematricesandthesingularvalues(SVs)ofananalyzedmatriceshaveagoodstability.Thus,whenamanipulationisappliedtoanimage,itsSVsdonotsignificantlychanged

Concretely,anm×nimageblock(canbeconsideredasamatrix)isdecomposedintothre ematricesU,DandV[28,pp448]:

𝐴= 𝑈 ×𝐷×𝑉𝑇= 𝑈1𝐷(1,1)𝑉𝑇+ 𝑈2𝐷(2,2)𝑉𝑇+ ⋯+𝑈𝑠𝐷(𝑠,𝑠)𝑉𝑇,

wheres = min(m,n),𝑈 ∈ 𝑅 𝑚×𝑚,𝑉 ∈ 𝑅 𝑛×𝑛are normalizedorthogonalmatrices and𝐷 ∈ 𝑅𝑚×𝑛is a diagonal matrix The maindiagonal

ofinformationaboutthekeyisnotdifficult.However,thislimitationwillbesolvedusingapublickeywatermarkingmodel.Forthepublickeywatermarking [33], [59], [78], [79], [81], two keys will

secretkeyusedinthedipstickdipprocess,arestillintestingofthewatermarking,usethepublickeyK2

For robust watermarking scheme, public keys have spreadspectrumdirections, using pseudorandom numbers; For the sensitive publickeywatermarking,theschemeproposedbyKimin2004willbeapplied

1.5.6 Reversiblewatermarking

All most above watermarking schemes are irreversible.Reversiblewatermarkingisatypeofwatermarkingwhich,besidesrestoringawatermarking,a l so al l ows t h e or i gi na l i m a g e tobe r e st or e d Re c e ntl y,

reversiblewatermarkinghasbeenstudiedextensivelyforitsapplicabilityinpractice,particularly inhealth,military,security andeducation.Thereversible watermarking

inversetransformexists.Somereversiblewatermarkingmethodsare:usingcompressionfeatures[11],[17],[67],histogramshifting[30],[68],[85]

differenceexpansion[5],[22],[39],[42],[47],[51],[78],predictions[24],[25]orcombinationsofthosemethods[10],[20]

1.5.7 Zerowatermarking

Traditionalwatermarkingimplementsimportantinformationintotheimage toprotect the image or validate the integrity of the image Thisaffects the image quality Often one must

inrobustness,orpayapriceforsensitivityinordertoobtaintheappropriatewatermarkingimage quality The zero watermarkingwill solve the problemthanks to the watermarking butnot embedding any information in theimage Therefore, how to secure the copyright of theowner as well asverify that the image has been accessed illegally The feature

zerowatermarkingistofindtherobustnessoftheimage,whichcanbestoredorcombinedwithconfidentialinformationtomakewatermarkingevidenceaspartofacopyrightverificationoridentificationprocessfortheintegrityoftheimage.Thedisadvantageofthistypeofwatermarkingistherobustnessof the scheme, which depends on the feature extraction methods.SomeselectiveextractionmethodsaresuchasDCT,SVD,QR,DWTorcombinationsofthesetransformations[6],[67],[68],[82],[83]

1.6 Parametersforevaluation

PSNR:Hệsốnàyđểđánhgiáchấtlượngảnhthủyvânđượctínhtheocôngthức:

𝑃𝑆𝑁𝑅=20𝑙𝑜𝑔10(𝑀𝐴K )

√𝑀𝑆 𝐸

thatb e s t s a t i s f i e s thebasicrequirementsofthewatermarkingsystem,suchaswatermarkingimagequality,highembeddingcapability,lowcomputationalcomplexity, high safety mentioned in Section 1.2.Thethesisfocusesoninvestigatingcurrentwatermarkingschemestoimprove,develop public key watermarking schemes, and newreversiblewatremarkingschemewithgoodimperceptibility,lowcalculationcomplexity as well as enhance embedding capacity of scheme This is atwo-wayresearchthatiscurrent,scientificandhighly

applicableinpractice.ThenextsectionofthisthesiswillresearchtherobustwatermarkingbasedontheSVDtransformationinChapter2,thereversiblewatermarkingbasedondifferenceexpansionofthepixelvectorseriesandthepublickey fragilewatermarkinginthechapter 3

CHAPTER 2 ROBUST WATERMAKRING BASED ON

SVDDEVELOPMENT

Robustwatermarkingisaneffectivesolutionforcopyrightprotection.Thewatermarkembeddingprocessandlicensingofrobustwatermarkingschemesareoftenperformedontransformeddomainsbytransformations.Thethesisselects

theSingularValueDecomposition(SVD)forthefollowing reasons: SVD is a

amatrixofsquareorrectangularmatriceswhileDCTandDWTonlyallowtodividetheimageintosquarematrices;SVDisenergy-centric,therefore,they can be used for a variety of purposes:information encapsulation,image compression, feature extraction, watermarking, etc.This chapterdiscusses SVD; Application of SVD in robust features extractionand

itsapplicationinwatermarking;Improveextractionmethodtogetmorerobustwatermarkin

g scheme; The robust public key watermarking proposalhasgoodimagequalityandlowcomputationalcomplexity

2.1 WatermarkingschemesbasedonSVD

SVD transformation is commonly used to construct watermarkingschemes[8], [15], [16], [40], [57], [59] Each watermarking schemeconsists of twoprocesses: the watermark embedding process and thewatermark extractionprocess At the process of watermark embedding:firstly, the original image ispartitioned into sub-blocks Each block willbeappliedSVDandonewatermarkbitisinserted intoit.Next, theinverse SVD transformation isimplemented to obtain a new sub-imageblock Finally, watermarked image isreceivee by gatherring sub-blocks.At the watermark extraction stage, people

intosub-blocksastheydointhewatermarkembeddingprocess.Subsequently,eachblockwillbeappliedSVDtransformationandwatermark bit will be extracted out Based

on the sequences of extractedbitstodetermine ifthereceived imageis ownedbytheauthor

TherearetwomainresearchdirectionsforSVDbasedimagewatermarkingschemes The first direction: watermarks are embedded inthe diagonal matrix D, namely

on the element D (1,1) as in [59], or onsome elements of D as in [72], [74], orb a s e d o n

t h e e n t i r e m a t r i x D a s in [15], [57] The second direction:watermarks can be embedded in thefirst column of matrix U or V as in [16],[40] In addition, some authorscombine SVD analysis with other transformssuch as DCT, QR, DWT tocreatenewwatremarking schemesasin [9],[23],[31],[ 5 2 ] , [ 6 7 ] , [69], [82] Among the SVD schemes where embedding - extracting isdone in D,the scheme suggested by Sun et al (SunSVD) was heavilyreferenced [59], orembedded - extracted in the first column of U orV,thes c h e m e s p r o p o s e d b y C h u n g e t a l [ 4 0 ]

( C h u n g S V D ) a n d C C L a i

[16](LaiSVD)arequitetypical

AccordingtoChungSVD,theoriginalimageissplitintonon-overlapped

blocks, then SVD is expanded and then one bit per blockbased on U (2,1) and

L a i S V D hasimprovedbyintroducingtheentropyselectioncriteriaandembeddingontheselectedblocks.Inaddition,theschemealsoincorporates SVD with DCTtransform to enhance robustness Whenembedding, LaiSVD relies on U (3,1)and U (4,1) instead of U (2,1) andU(3,1)as ChungSVD

Comments

- Wherew a t e r m a r k b i t s n e e d t o b e e m b e d d e d i s l e s s t h a n e m b e

d d i n g ability,selectwhichblockstoembedsothatthehiddingabilityisgood

i=1 j=1

j=1 i=1

LaiSVDhasintroducedtheentropyselectioncriteriabuttheeffectisnotclear

- Most schemes use only D (1,1) or U (1) to embed and to extractthewatermark,buttheSVDdecompositionproblemstillhastofindallD(i,i),U(i)and V(i) correspondingly, i = 1,2 , s Therefore, there is largecomputationalcomplexity

2.2 Proposedalgorithms

2.2.1 BlockComplexity

Thevaluesofimagepixelscanbechangedafterembeddingwatermarks In thecases of binary or few color images, the change caneasily be recognized byhuman eyes Conversely, in texture regions, it isdifficult to see the change Therefore, hidinginformation in complexregions will be hard to discover In this section, we propose asimple buteffectivemethodtocomputethecomplexityforimageblocks

Assuming thatAis a block with the size ofm × n.It is noted thatneighboring

image pixels are distinguishable if the Euclideandistancebetweenthepixelsarelargerthanapredefinedthreshold(calledδ).)

Assumingm1andm2are the values of two neighboring pixels Wedefine:

Step 1:Dividing the original image I into non-overlappedm×n

blocksIi.Step2:Inthisscheme,inspiredthemethodforfindingt h e largesteigenval

{𝑈i(1,1), 𝑈i(2,1)}are selected for embedding bit wiand inthe other case,the

pair of {𝑉i(1,1),𝑉i(2,1)} is used For embedding

i

+Computing𝑥′