SteganographyinaVideoConferencingSystem? AndreasWestfeld1andGrittaWolf2 2InstituteforOperatingSystems,DatabasesandComputerNetworks 1InstituteforTheoreticalComputerScience DresdenUniversityofTechnology fwestfeld,g.wolfg@inf.tu-dresden.de D-01062Dresden,Germany Abstract.Wedescribeasteganographicsystemwhichembedssecret includesdiscretecosinetransformation(dct)based,lossycompression. securityisestablishedbyindeterminismwithinthesignalpath. Resultisthetechnicalrealisationofasteganographicalgorithmwhose messagesintoavideostream.weexaminethesignalpathwhichtypically 1Introduction Theescalationofcommunicationviacomputernetworkhasbeenlinkedtothe informationhidingexistfordigitalimageles,audioles,andinbackground soundsofphonecalls[2].therearemorethan20programsontheinternet(for increasinguseofcomputeraidedsteganography[5],[6].steganographicmethods examplesseethelistbelow). thatathirdpersoncannotdetectorevenprovethisprocess.examplesfor usuallyhidecipheredmessagesinother,harmless-lookingdatainsuchaway {S-ToolsbyAndyBrownembedsdataasleastsignicantbitsinaudioles {JstegbyDerekUphamembedsdatainJFIFimages.Itoverwritestheleast (.wav)orasleastsignicantbitsofthergbcolorvaluesingraphicles (.bmp).athirdmethodhidesdatainfreesectorsofdiskettes.severalsymmetricencryptionmethods(des,idea,...)areoeredforadditionalencryptionofthesecretdata.[9] signicantbitsofthecoecients.[8],[10] {HideandSeekbyColinMaroneyhidesdata(encryptedwithIDEA)inGIF {MandelstegbyHenryHasturcalculatesaGIFfractalfromale.Theresultingimagesareverysimilar.Dierencescanonlybeseenwhencomparing {PGE(PrettyGoodEnvelope)byRoche-CryptpacksdatainGIForJPEG les.theuseofanadditionalsecureencryptionmethodisrecommended.[12] les.[11]?thisworkissponsoredbythegermanfederalministryofeducation,science,researchandtechnology(bmbf). {StegobyJohnWalkertransformsanyletoanonsensicaltextbymeansof theircolorvalues.[13] afreechoosabledictionary.
2{TextobyKevinMahertransferslesintopoeticEnglishsentences(comparablewithstego,whichproducesnonsensicaltexts).[14] stereophony,colortv,videotext,traccontrolsystem(tcs),andradiodata system(rds)atfmradio. Datacamouageisalsousedforcompatibleenlargementofnorms,suchas 2VideoConferencingSystems Videoconferencesusecompressionalgorithmstoensureanacceptablequality Thispaperdoesnotdealwithwatermarkingsystemsatall[1]. evenonlowdataratesystemslikeisdn.usually,compressionmethodsarelossy whichmeansthatthereconstructedimageisnotidenticalwiththeoriginal. tempresentedinthispaperworksontheh.261standard.thisisthemost commonstandardforcompressioninvideoconferencesandisrecommendedby [3].InFig.4,wecanseethepointsforembeddingandextractingwithinthe thecomiteconsultatifinternationaltelegraphiqueettelephonique1(ccitt) Thevideoconferenceusedfortheimplementationofthesteganographicsys- H.261informationow. weneedacarrierthatallowsthepossibilityofunnoticeablemodications.signal possibilitiesfordataembeddingwehave.inchapter5weinvestigateatypical noiseandirrelevancearecommonexamplesforit.compressionmethodstryto removesignalnoiseandirrelevance.thebetterasignaliscompressed,theless Compressionanddataembeddinghavecontrarygoals.Fordataembedding signalpathfordataembedding. ThesteganographicalgorithmdescribedinChapter6embedsdataintransformedblocks.Thereforewedescribethetransformationprocessusedinthis videoconferencingsystem. image.thesubsequentquantizationremovestheinsignicantpartsoftheimage. forthehumaneye)frommarginalparts(invisibleforthehumaneye)ofthe Thetransformationemployedhastobeinvertibleinordertoregaintheessential Asuitabletransformationisameanstoseparateessentialinformation(visible 3DiscreteCosineTransformation (DCT).Ittransformesanimageof88pixelswith88=64brightnessvalues H.261,M-JPEG,MPEG,usethetwo-dimensionaldiscretecosinetransformation F(0;0):::F(7;7)into64values(so-calledDCTcoecients)f(0;0):::f(7;7) (seeequation1).thetransformationcausesnosignicantloss(roundingerrors Manydigitalvideoconferencingsystems,forinstancebasedonthestandards only).theretransformedimageresultsfrombacktransformationofthedct coecients(seeequation2).itcanalsobeunderstoodaslinearcombination Bk;n(seeFig.1). ofthedctcoecients(seeequation3andfig.2)withthedctbaseimages 1theformerCCITTisnowtheInternationalTelecommunicationUnion(ITU)
increasingverticalfrequency @@ kn 0 1 2 3 4 5 6 7 3 01234567????88 image increasinghorizontalfrequency Fig.1.DCTbaseimagesBk;n - =f(0;0) +f(0;1) +:::+f(7;7) Fig.2.Presentationofan88imageby64baseimageparts?
4 F(x;y)=7Xk=07Xn=0C(k) f(k;n)=c(k) 2C(n) 22C(n) 7Xx=07Xy=0F(x;y)cos(2x+1)k 2f(k;n)cos(2x+1)k 16 16 (1) 16 cos(2y+1)n 16 (2) with F(x;y)=7Xk=07Xn=0f(k;n)Bk;n(x;y) C(z)=12p2forz=0 (3) 4AnExample 1else InordertoillustratetheDCT,wetransformadotoverthei.Fig.3showsit stronglyenlarged.aspresentedinfig.3b)thedotovertheihasagratingof 64brightnessvalues.Let'slookatitstransformedmatrix: 264?840?292150100?31015128?110?1?40 50?4?84?2?31?2?12?1 00 00?12010010?2?20?10 00?10?20 00?10?10 0?1 3 Thequantizationcausesanaccumulationofzerosbyapplyingastepfunction 75 (dividingandrounding)tothedctcoecients: 264?30?80?20 20?80?30?10 20 10?10 00 10 00 10 00 00 3 Theexampledemonstratesthatonly15coecientsdierentfromzeroareleftof 75 coded.[3]and[4]encloseadescriptionofrunandlevelandhumancoding.the theinitial64brightnessvalues.thecoecientsarearrangedinlinearordering andthenarerunandlevelcoded.thenewcreatedsequencewillbehuman- signicantfactisthattheyareloss-freecodings.fig.3c)showstheresultof thebacktransformation.
5 a) b) c) Fig.3.\Dotoverthei":a)original,b)rasteredc)afterdecompression Input Motion Estimationand Motion Compensation 6Output Compensation Discrete Cosine Transform Cosine Transform InverseDiscrete Quantization Quantization Inverse Coding Model EmbedExtract? Model Decoder Entropy Coder Entropy SS??SSS??HHHHH Decoder Fig.4.InformationowintheH.261Codec[4] Transmission 666 -
65SignalPath Apreciseknowledgeofthesignalpathisimportantinordertobeabletoestimate thesafetyofasteganographictechnique.fromthecameratothecodedsequence ofpictures,thesignalpathissubjecttolossesbytransformationsaswellasto inuencesandtodisturbances.inthefollowing,sometransformationpointson thepatharedesignated,andinparentheses,thealteredquantity. imageoftheoriginalispreprocessedopticallybythelensofthecamera.additionally,attitudeaperturesetting(depthoffocus),thefocuses(partofhigh videofrequencies),thefocallength(detail,videodepth)andthequalityofthe Theappearanceoftheoriginalisinuencedbythelightingconditions.The lens(distortion)contributeessentially.throughdispersion,thefocusesaredependentonthecoloroflight.thelightisusuallytransformedintoanelectrical signalinthecamerabyachargecoupleddevice(ccd).thetinyccdsare characterizedthroughtheirhighsensitivitytolight.thelightinfrontofthe about380000photosensitivepointsislteredbymanycolored,narrow,vertical stripes.eachthreeadjacentsensors,receivingrespectivelyred,green,andblue lteredlight,makeonepixel.thehorizontaldistanceofthethreesensorsisonly apartialpixeldistanceandthus,isneglected.accdhasatemporalinertness (thereaderpossiblyobservedthe\tracing"inthecaseofacamerapanshot) andisoperatedwithaspecicsamplingfrequency.afterwards,thealteredand runsthroughacircuitwhichcontainssemiconductors(temperaturedependence, rasteredimageisavailableintheformofanelectricalsignal.inthisform,it whichcontainstheimagewhichisnowevenmorecoarselyrasteredthaninthe transmittedtothedevicedriver.2adatastructureinthergbformatresults, camera.onthevideocardinthecomputerthepictureislocked,digitizedand noise),anditischangedintoantscorpalsignal.thesignalpathnowleads tothecomputeroveracoaxialcable(spectralphaseshift,attenuation)fromthe boundarytodeterminism(seefig.5).allfurtherprocessingstepsaredigitaland NTSCorPALsignal.Withthelaststeponthesignalpath,weexceededthe deterministic.withthetransformationandquantization,desirableroundingerrorsoccur.aloss-freeentropycodingcompressesthedatabetweenquantization andtransfer. -non-deterministic Processing - lossy Processing deterministic,embedding?-deterministic, Fig.5.Sectionsofthesignalpath loss-free Processing 2Oftenonlytheinterfaceofthedevicedriverisdocumented.Theprogrammeris unabletoseparatetheactivitiesofthevideocardandthoseofthedevicedriver.
meansthatresultingoutputsignalsdierfromeachotherwithhighprobability Thesignalpathcanbedividedintothreeparts.Non-deterministicprocessing7 inthecaseofidenticalinputsignals. noiselevelisdependentoftheconsideredbandwidth.theimageincorporated linespersecond(for25fullframes).thevoltagewithinalinevariesduringhorizontalcolorchange,withverticalfromlinetoline.therefore,uprightbrightness Anexampleisthenoiseofthesemiconductordevicesmentionedalready.The bythecameraisledinthepalsignallinebylinefromtoptobottom,15625 modicationsareputinthesignalatabandwidthamaximumof8khz(oneline iswhite,onelineisblack,alternating).upto800imageelementsperlinecan beplacedincomparisontothis,whichcorrespondstoabandwidthof6mhz. Thebandwidthforhorizontalvideofrequencies{andthereforethenoiselevel{ isupto800timesaslargeasthebandwidthforverticalvideofrequencies. whichisincludedintheccvssignal.lineinterlacingdividesafullframeinto twohalfframes.ifonenumbersthelinesofafullframefromtoptobottom,the rstframecontainsalloddlinesandthesecondframeallevenlines.therefore, twostraightadjacentlinesofthefullframeare1 Twoscanlinesarealignedbymeansofthehorizontalsynchronizationpulse other.sincethehalfframesarealsoregisteredattheframerate,thescreen contentcanalreadyhavechanged. mightbeplacedinformat720by540.threeadjacentsensorswillbesummarized Herefollowsanexamplewithnumbers.Theabout380000sensorsofaCCD 50frameseparatedfromeach asoneimagepoint(rgb),althoughthesensorshaveadistanceof1 spectrum. theimagecontentof110pixelsallowsconsiderablemodicationsofthefrequency length.thishorizontaldistanceis,referringtothesmallesth.261format(176by 144),14pixeldistance.Table1shows,thatalreadyahorizontaldisplacementof 720line generatedwithhorizontaldephasingandtransformed.thedephasingresults ifterm2y+1isreplacedby2y+1:2inequation1.throughconsideration offig.1,itisobviousthatahorizontal\dephasing"ofthebaseimagesbk;0 bringsnochange.underexclusionofthecoecientsf(0;0)(basebrightness) WedevelopedalittleprogramwhichcreatesTable1whenbaseimagesare andf(k;n)(appropriateforthebaseimage),thecoecientf(k0;n0)withthe truncatedvaluesduringcomputation.inthelineforb0;1,thecoecientwiththe inbracketsarenottobetracedbacktodisplacementandcanbeexplainedby strongestdieringamountprobablyisf(0;0).sincethiscoecientisexcluded largestabsolutevaluewasalwayssearchedinthetransformedmatrix.thevalues fromconsideration,thesmallernextappears.otherwise,forn>0,thefollowing patternisvalid: thatcoecientf(k0;n0)currentlychangesby1.sincethecoecientsareinteger, Columnfmincontainstheminimumamountforthecoecientf(k;n)from n0=n?1andf(k0;n0) k0=k f(k;n)n3% amodicationlessthan1isnotpossible.
8 Table1.RelativechangeofDCTcoecientsf(k0;n0)whilehorizontaldephasingof baseimagesbk;nby1 knk0n0f(k0;n0) 00{{(0.00)%(1) 10pixel 0102 0201 f(k;n)fmin 0302 1.51% 6.75% 9.19% 67 15 knk0n0f(k0;n0) 0403 0504 0605 12.27% 15.16% 18.90% 119 4036(0.00)%(1) 0706 1030(0.27)%(369) 1110 24.60% 3.61% 287 4140 1211 6.79% 156 4241 f(k;n)fmin 1312 9.28% 115 4342 4443 4544 3.57% 29 4645 6.75% 15 4746 9.19% 11 1413 1514 1615 1716 2060(0.13)%(780) 12.28% 15.13% 18.98% 24.55% 9 5010(0.14)%(736) 12.29% 9 2120 3.58% 287 5150 15.16% 7 2221 6.75% 156 5251 18.90% 6 2322 9.29% 115 5352 24.60% 5 5453 5554 3.62% 28 5655 6.80% 15 5756 9.28% 11 2423 2524 2625 2726 3070(0.14)%(737) 12.18% 15.19% 18.82% 24.62% 9 6020(0.26)%(390) 12.30% 9 3130 3.61% 287 6160 15.16% 7 3231 6.79% 156 6261 18.98% 6 3332 9.28% 115 6362 24.55% 5 6463 6564 3.59% 28 6665 6.76% 15 6766 12.20% 9.29% 11 9 3433 3534 3635 3736 12.30% 15.16% 18.98% 24.55% 9 7030(0.14)%(736) 15.21% 18.82% 76 7 7170 24.67% 5 6 7271 3.62% 6.80% 28 15 5 7372 7473 7574 7675 7776 12.30% 15.16% 18.98% 24.60% 9.28% 11 9765
6Algorithm 9 time,wewanttoshowanunveriablemodication,unveriableinthesenseof Inthischapterwewilldiscusstheimage(dotoverthei)ofChapter4again.This changescauseanimperceptiblehorizontaldislocationoftheimage,thealgorithmdoesnotinuenceso-calledmotionvectors,atleastnotinadirectway.a motioncompensationstep(seefig.4)isnotnecessaryfortheimplementation ofthissteganographicvideoconferencingsystem.anattackercouldgetmore changethecarriersignal.theheartofasteganographicalgorithmisaprocess thecalculatedexampleinchapter5.whenembeddingsomething,wehaveto thatchangesthesignal.inourcaseitchangesdctcoecients.althoughthe onlyonceasakeyframe.hence,itistransmittedandusedforsteganography easierforsteganographers.anunchangingpictureinfrontofthecameracomes forembedding.however,deltaframecodingincaseofstillimagesmakeslife whichhecouldmatchagainsttheactualframe.thiswouldreducethespace preciseimagedatabyinterpolatingconsecutiveframesofanunchangingpicture steganographicprocessing. agecontainabigcoecient(seefminintable1)makingthemsuitablefor onlyonce.itisveryunlikelythatthedierence(ordelta)framesofastillim- a) b) c) d) e) f) (unchanged),c)afterdecompression(changedbyalgorithm),d)by115pixelshifted Fig.6.\Dotoverthei"andhorizontalshifting:a)original,b)afterdecompression original,e)movedimageaftergrating,f)movedimageafterdecompression result,theabsolutevalueofoneofthedctcoecientsaccordingtotable1is highercontrast. largeenoughtoallowamodication.(refertocoecientf(0;2)intheexample.) Thefollowingmatrixincludesthe64brightnessvaluesoftheoriginalimagewith Thecontrastoftheoriginalimagehasbeenincreased(seeFig.6a)).Asa
10 2 6 4 2020537980562120 20821101101101108622 5311011011011011011059 7911011011011011011085 8011011011011011011086 5611011011011011011062 21861101101101109123 2022598586622320 3 ThefollowingleftmatrixincludestheDCTcoecientsafterquantization. 75 Theboldhighlightedcoecientf(0;2)allowsamodicationof6.75%,which means160:0675=1:08.therightmatrixshowsthismodicationforf(0;1). 264?160?503020?60 40?160?60?20 00 205020 000000?20 00 202000 000000 3 2 4?1?160?60?20 64?160?503020?20?60 00 00 202000 205020 000000 000000 3 Afterrecovery,thefollowingmatrixesofbrightnessvaluesresult,presented 75 infig.6b)andc),too.themodicationleadstoaslightshiftingtotheright. 264 26831051041041058326 2626537676532626 7310510610410410610573 5410710310410410310754 26831051041041058326 2626537676532626 3 75 2 64 2626527677552626 26811041041051078526 7110310410310410710875 5210510210310410510957 26811041041051078526 2626527677552626 3 The\natural"shiftingasacomparison:Fig.6d)showstheoriginal,shifted 75 tionofthecoecient. Theshiftingoftheoriginalimagewouldhavecausedamoreintensivemodica- transformationandquantizationoftheleftfollowingmatrix(seealsofig.6e). by115pixel.thecoecientspresentedintherightfollowingmatrixresultfrom 264 2020507881582220 20761101101101109123 4711011011011011011065 7311011011011011011091 7411011011011011011092 5011011011011011011068 20821101101101109525 2021568487642420 3 75 2 4?2?160?60?20 64?160?513020?20?60 00 00 000000 202000 205020 000000 3 75
Finally,Fig.6f)showsthefollowing,recoveredmatrix. 11 264 2626487379592626 26791021031061098726 6710210710510310510880 4910310210310410411060 26791021031061098726 2626487379592626 3 Astheexampleshows,early,non-deterministiceectsatthebeginningof 75 7Implementation thesignalpathcanbereproducedinalaterpart(seefig.5). coecients.blocksare\suitable"iftheyincludeacoecientwhichislarger exploitstheeectdescribedinchapter6:thefrequencyspectrumchangesconsiderablyalreadyatminorchangingsofthephasingoftheimage. Theimplementedsteganographicfunction\Embedding"(seethemodelin[7]) thanitsminimumamountfmin(seetable1).inthesourcecode,allminimum amountsarerepresentedbydelta[].allotherblocksare\unsuitable"andwill betransmittedwithoutsteganographicmodication. Atrst,wedistinguishbetween\suitable"and\unsuitable"blocksofDCT */ /*Tobeclassifiedas"suitable",ablockmustcontainone unsignedintdelta[64]={ coefficientgreaterorequaltoitscorrespondentvalue inthefollowingmatrix. -1,-1,-1,-1,-1,-1,-1,-1, -1,28,28,28,29,28,28,28, 15,15,15,15,15,15,15,15, 11,11,11,11,11,11,11,11, 9,9,9,9,9,9,9,9, /*-1meansinfinity*/ }; 5,5,5,5,5,5,5,5 6,6,6,6,6,6,6,6, 7,7,7,7,7,7,7,7, /*stego_in(p)isthesteganographicfunction"embedding". */ Theparameterppointstoamatrixof64coefficients.
12 voidstego_in(int*p) { inti,most_suitable,sum_of_block,is_stego,*steg_ptr; is_stego=0; for(i=1;i<64;i++){/*skipdccoefficientp[0]*/ sum_of_block=0;/*forsum(mod2)*/ if(p[i]){ sum_of_block+=abs(p[i]); /*1means"suitable"block*/ if(abs(p[i])>=delta[i]){ /*coefficientlargeenough?*/ is_stego=1; /*considernon-zerocoefficients*/ /*moresuitable?thenkeepthepointer*/ /*"suitable"block*/ /*sumup*/ if(abs(p[i])-delta[i]>=most_suitable){ steg_ptr=&p[i-8];/*thisisf(k',n')*/ } } } } most_suitable=abs(p[i])-delta[i]; embedding,theblockwillbetransmittedunchanged.iftheparityisnotequal, 2sumofitscoecients(akindofparity).Iftheparityisequaltothenextbitfor...Forablockclassiedassuitable,itsfurthertreatmentdependsonthemodulo- ithastobechanged....if(is_stego) /*comparethemodulo-2sumwiththenextbittoembed*/ if((sum_of_block&1)!=get_bit_to_embed()){ if(*steg_ptr>0) /*decrementabs(*steg_ptr),thecoefficient*/ /*suitableblock?*/ elseif(*steg_ptr<0) (*steg_ptr)--; }Letf(k;n)bethecoecientofasuitableblockwhich,correspondingto } else/*0==>1*/ (*steg_ptr)++; *steg_ptr=1; Table1,allowsthemaximummodication.Inthiscase,theabsolutevalueof thecoecientf(k;n?1)willbedecreasedby1orifitiszero,setto1.this way,acoecientofablockischangedby1anditsparityips.
necessary.thewholescenarioisshowninfig.7.therecipientreceivessuitable Allchangedblocksaretransmittedaswellasthose,wherenochangewas 13 andunsuitableblockswhichareseparatedbythesamecriteriaasatthesender. Ithastoberemarkedthatchangedsuitableblocks(suitableblockswithipped parity)willalwaysstaysuitableblocksbecausethecoecientf(k;n)hasnot beenchanged,butfulllsthecriterion\suitable". imagereconstruction.thesteganographicalgorithmpresentedhereactslikea bitsofthesuitableblockssequentially.therecipientsystemusesallblocksfor quantizationwithahigherdivisor.itincreasesthecompressionratesothatthe introducederrorlooksnatural.alowerquantizershouldequalizetheeectof Therecipientcanextracttheembeddeddatathroughreadingouttheparity thealgorithm. voidstego_out(int*p)/*steganographicfunction"extraction"*/ { inti,sum_of_block,is_stego; sum_of_block=0; is_stego=0; for(i=1;i<64;i++){ if(p[i]){ sum_of_block+=abs(p[i]);/*sumup*/ /*1means"suitable"block*/ /*considernon-zerocoefficients*/ /*formod-2sum*/ /*coefficientlargeenough?*/ if(abs(p[i])>=delta[i]) }InFig.8weshowthesurfaceoftheapplication.ivsdisthedaemonwhich }if(is_stego)put_embedded_bit(sum_of_block&1); } is_stego=1; /*YES!suitableblock!*/ otherdaemons(single-ormulticast-addresses).afterestablishingthevideoconferenceconnection,theuseroftheconferencewithsteganographicenhancement receivesconferencecalls.permouseclickitispossibletoopenawindowtocall andonefordisplayingtheembeddedmessagesofthecommunicatingpartner. Theapplicationiscomparabletoacombinationofavideoconferenceandthe Unixstandardcommandtalk3. 8Conclusion Throughcompression,asusedwithvideoconferencing,theleastsignicantbits gainimportance,soeverybitofthecompressedsignalcontributesasignicant parttothepicture.thedetectionofarandomreplacementofthesebitsis hastwoadditionalwindows:onefortheinputoftext(thesecretmessagetohide) 3talkisacommunicationprogramforterminals.
14 Embedding digital brightness values Extracting compressed video signal (embedded data) Transformation Entropy Decoding unsuitable measure the distance to quantized value detect whether block is suitable secret message secret message calculate modulo-2 sum 01001010011 01001010011 Quantization lossy 11010011101 11010011101 01011101001 01011101001 01101000100 0110 unsuitable Inverse detect whether block is suitable Quantization calculate modulo-2 sum sum=bit compare sum with the bit to embed change one coefficient Fig.7.EmbeddingandExtracting by 1, if necessary Entropy Coding loss-free Inverse Transformation compressed video signal (embedded data) digital brightness values totheunchangedcarrier,whichshouldneverleavethesecuritydomain.we carrier,makingitimpossibletodetectthesechangeswithoutdirectcomparison possibleasshownforjstegin[8].however,itispossibletochangepartsofthe Transmission raiseanysuspicionforapossibleattacker.forthisreason,wescrutinisedthe picturereceptionclosely. analysisoftheinputdevicesshowsfreespacespermittingembeddeddata.if steganographictechniquessimulatepeculiaritiesofacamera,thechangesdonot usespecialfeaturesoftheinputdevices,suchasacameraorscanner.the ceptibly.adirectcomparisonwiththeoriginalallowsdierentiation,butthis stilldoesnotenabletheobservertodiscernbetweentheoriginalandthealtered signals.furthermore,thesendermerelytransmitsthechangedframes.inthis manner,asecretmessagecanbeembedded.theslighthorizontaldephasingis Ouralgorithmreproducestheseeectsarticially;thesignalchangesimper- unnoticeable. thisitisnecessarytoseparatethealgorithmfromthesecrets.thesimplest receiverneedthesamekeyandproceduretogeneratethesebitsandusethem possibilityisthegenerationofpseudorandombits.boththesenderandthe asapseudoone-timepad.becausethedistributionofthesebitshasthesame Algorithmsareonlytrustworthy,whentheyareopentopublicscrutiny.For randomuniformityasbitsextractedfromanyvideoconference,theattackercan notdiscernbetweenanormalvideoconferenceandoneinwhichsecretdatahas beenembeddedafterencryption. picture,becauseitisimpossibletoembeddatainblackframes. phoneconversation(upto8kbit/s).thisdependsuponthetextureofthe InanISDNvideoconferencingsystemitispossibletoembedaGSMtele-
15 References 1.IngemarJ.Cox,JoeKilian,TomLeighton,TatalShamoon,ASecure,RobustWa- 2.ElkeFranz,AnjaJerichow,SteenMoller,AndreasPtzmann,IngoStierand: termarkformultimedia,in:proceedings:informationhiding.workshop,cam- Fig.8.Userinterfaceoftheimplementedapplication 3.CCITTRecommendationH.261,VideoCodecForAudiovisualServicesAtp64 Cambridge,U.K.,May/June,1996,LNCS1174. ComputerBasedSteganography.In:Proceedings:InformationHiding.Workshop, 7.BirgitPtzmann:InformationHidingTerminology.In:Proceedings:Information 6.MaritKohntopp,SteganographiealsVerschlusselungstechnik,iX4/1996 4.AndyC.Hung,PVRG-P64Codec1.1,StanfordUniversity,1993 5.NeilF.Johnson,Steganography,GeorgeMasonUniversity,1996 kbit/s,genf,1990 8.RobertTinsley,SteganographyandJPEGCompression,FinalYearProjectReport,UniversityofWarwick,1996 Hiding.Workshop,Cambridge,U.K.,May/June,1996,LNCS.
169.ftp://idea.sec.dsi.unimi.it/pub/security/crypt/code/s-tools4.zip 10.ftp://ftp.funet.fi/pub/crypt/steganography/ 14.http://ftp.giga.or.at/pub/hacker/stego/texto_os2.zip 11.http://www.rugeley.demon.co.uk/security/hdsk50.zip 12.http://www.rugeley.demon.co.uk/security/ 13.ftp://ftp.funet.fi/pub/crypt/mirrors/idea.sec.dsi.unimi.it/ cypherpunks/steganography/mandelsteg1.0.tar.gz