Spread Spectrum Image Watermarking for Secured Multimedia Data Communication

Transcription

1 Word Academy of Scence, Engneerng and Technoogy Internatona Journa of Computer and Informaton Engneerng Vo:, No:6, 007 Spread Spectrum Image Watermarkng for Secured utmeda Data Communcaton Trtha S. Das, Ayan K. Sau, and Subr K. Sarkar Internatona Scence Index, Computer and Informaton Engneerng Vo:, No:6, 007 waset.org/pubcaton/840 Abstract Dgta watermarkng s a way to provde the facty of secure mutmeda data communcaton besdes ts copyrght protecton approach. The Spread Spectrum moduaton prncpe s wdey used n dgta watermarkng to satsfy the robustness of mutmeda sgnas aganst varous sgna-processng operatons. Severa SS watermarkng agorthms have been proposed for mutmeda sgnas but very few works have dscussed on the ssues responsbe for secure data communcaton and ts robustness mprovement. The current paper has crtcay anayzed few such factors namey propertes of spreadng codes, proper sgna decomposton sutabe for data embeddng, securty provded by the key, successve bt canceaton method apped at decoder whch have greater mpact on the detecton reabty, secure communcaton of sgnfcant sgna under camoufage of nsgnfcant sgnas etc. Based on the anayss, robust SS watermarkng scheme for secure data communcaton s proposed n waveet doman and mprovement n secure communcaton and robustness performance s reported through expermenta resuts. The reported resut aso shows mprovement n vsua and statstca nvsbty of the hdden data. Keywords Spread spectrum moduaton, spreadng code, sgna decomposton, securty, successve bt canceaton D I. INTRODUCTION IGITAL watermarkng agorthms can be thought as dgta communcaton scheme where an auxary message s embedded n dgta mutmeda sgnas and are avaabe where ever the atter sgnas move. The decoded message atter on serves the purpose of securty n communcaton, copyrght protecton, data authentcaton, broadcast montorng etc. Robustness s an essenta crteron n dgta mutmeda watermarkng schemes aong wth vsua transparency, hgh data embeddng rate, ow computaton cost and compexty of the agorthms needed for data embeddng and recovery purpose. * A these requrements are reated n confctng manner and the partcuar agorthmc deveopment emphaszes to a greater extent on one or more such requrements dependng on the type of appcaton. In dgta communcaton spread spectrum moduaton fnds wde usages n hoste envronment as we as n mutpe access communcaton system due to ts nherent ant-jammng anuscrpt receved on January, 006. T. S. Das s wth the Gurunanak Insttute of Technoogy, Panhat, 7004, Inda (e-ma: trthasankardas@yahoo.com. A. K. Sau s wth the Jadavpur Unversty, Jadavpur-70003, Inda (e-ma: ayankumarsau@yahoo.com. S. K. Sarkar s wth the Jadavpur Unversty, Jadavpur-70003, Inda (ema: sksarkar@etce.jdvu.ac.n. and nterference rejecton property [-3]. These attrbutes motvate researchers to deveop severa SS watermarkng schemes for mutmeda sgnas. The SS watermarkng that uses dstnct pseudo nose (PN spreadng codes for embeddng each bnary dgt s popuar and s proven to be effcent, robust and cryptographcay secure. The motvaton of the present work stems from fndng out factors that have greater mpact on mprovement n secure data communcaton and ts detecton reabty of SS watermarkng scheme. II. SPREAD SPECTRU WATERARKING AND DETECTION Spread Spectrum watermarkng and detecton s accompshed by embeddng every watermark bt over many sampes of cover mage usng a moduated pseudo random spreadng sequence [4-6]. SS watermark embeddng and detecton are anayzed mathematcay as foows: A. Watermark nserton Let B denotes the bnary vaued watermark bts strng as a sequence of N bts. B { b, b, b3,..., b3}, b ε{,0} ( If I denotes the mage bock of ength L.e. mage transformaton coeffcents of ength L, a bnary vaued code pattern of ength s used to spread each watermark bt. Therefore, a set P of N code patterns, each of ength are generated to form watermark sequence W by performng the foowng operaton. N ( [ W ] b [ P ] The watermarked mage I w can be obtaned by embeddng watermark nformaton W nto the mage bock I. The data embeddng operaton can be mathematcay expressed as foows: [( IW ] [ I ] + α.[ W ] (3 Where α s the moduaton ndex and ts proper choce w optmze the maxmum amount of aowed dstorton and mnmum watermark energy needed for a reabe detecton. α may or may not be a functon of mage coeffcents. Accordngy SS watermarkng schemes can be caed as sgna adaptve or non-adaptve SS watermarkng. B. Watermark detecton In SS watermarkng, the detecton reabty for bnary vaued watermark depends on decson varabe t obtaned by computng the zero ag spata cross-covarance functon Internatona Schoary and Scentfc Research & Innovaton ( schoar.waset.org/ /840

2 Word Academy of Scence, Engneerng and Technoogy Internatona Journa of Computer and Informaton Engneerng Vo:, No:6, 007 Internatona Scence Index, Computer and Informaton Engneerng Vo:, No:6, 007 waset.org/pubcaton/840 between the watermarked mage I w and every code pattern P. The decson varabe t can be mathematcay expressed as foows: t P m( P, IW m( IW (0 (4 where m (S represents the average of the sequence S. If S k represents the eements of S wth k,,3,, then m (S can be mathematcay expressed as foows: (5 m( S s k k The symbo (0 n (4 ndcates the zero ag cross-correaton and for two sequences S and R, the zero ag cross correaton s gven by: (6 S, R (0 s k r k k Where s k and r k are the eements of sequence S and R respectvey wth k,,3,,. The bt b s detected as zero f t >0 and as one otherwse. If the code pattern P s chosen n such a way so that m (P 0, the computaton of t becomes P, I + α. N t P, I + α. N b P m ( I b P, P P, m ( I (7 (8 P, I (9 W III. DESIRABLE PROPERTIES FOR IPROVED WATERARK DETECTION Ths anayss ndcates that the code patterns used for spread spectrum watermarkng shoud posses some specfc propertes. Watermark detecton s mproved f the foowng condtons are satsfed:. P,,, 3,, shoud be dstnct sequences wth zero average.. The spata correatons <P, P j >, j shoud be mnmzed. Ideay sequences P and P j shoud be orthogona whenever j. 3. Each P for I,,3,, shoud be uncorreated wth the mage bock I when mage predcton (for estmatng mage dstorton s not used before evauatng the cross-correaton. A. Transform for data hdng Over the years waveet transform (WT attracts attenton n varous mage processng appcatons ncudng dgta mage watermarkng, detecton, denosng and the upcomng mage compresson standard JPEG-000 due to ts cojont representaton n the form of space-frequency resouton, whch provdes both oca and goba nformaton of mage sgna [7-0]. The DWT system decomposes the mage sgna nto LL, LH, HL and HH sub bands whe band waveet system decomposes the same nto ( X channes, correspondng to dfferent drecton and resouton. Such decomposton aows data embeddng n dfferent channes that offers resency aganst varous mage mparments. In context of SS watermarkng, data embeddng s recommended n waveet coeffcents of LL, LH, HL and HH sub bands (n case of DWT and n group of channes wth few hghest and few owest varance (n case of -band WT, varance s cacuated for a the x channes and paced n descendng order; one group s formed by takng the channes of extremey hgh varance vaues and other group s formed by takng the channes of extremey ow varance vaues vaues of coeffcents [-4]. Because these above-mentoned sub bands/channes are very much dssmar n nature compared to other combnaton of sub bands/channes mpyng that they jonty provde wde range of frequency components of cover mage. To justfy the seecton of data embeddng regons on the bass of better spectrum spreadng, varance of dfferent channes are computed. The varance for coeffcents of channes H, H 3, H 4, H 4 and the channes H 4, H 4, H 43 and H 3 for ( 4 band system are aways the few owest and few hghest vaues. B. Code Propertes Hgh detecton reabty can be acheved f code pattern sequences possess very ow zero ag cross correaton among each other as we as wth mage bock when mage predcton s not used to evauate the cross correaton [5, 6]. The code patterns P used frequenty for SS moduatons are (a Fg. (a Sub bands correspondng to DWT decomposton Cross covarance among dfferent DWT sub bands Internatona Schoary and Scentfc Research & Innovaton ( schoar.waset.org/ /840

3 Word Academy of Scence, Engneerng and Technoogy Internatona Journa of Computer and Informaton Engneerng Vo:, No:6, 007 Internatona Scence Index, Computer and Informaton Engneerng Vo:, No:6, 007 waset.org/pubcaton/840 pseudorandom or pseudo nose (PN sequences. The rand ( and srand ( functons of ANSI C brary can be used to generate PN sequences by near congruentua generators. Here atab s PN generator can aso be used. But n both (a Fg. (a Varance of dfferent channes; Frequency bands correspondng to 4 band decomposton cases the desred propertes of m (P 0 and <P, P j > 0 for j I can be theoretcay guaranteed f nfnte ength sequences, whch of course s not feasbe for practca mage processng operaton. The wdey used PN codes n spread spectrum moduaton s the maxmum ength sequence or m-sequences that exhbt some correaton among each other and does not provde good detecton. ayer et a proposed determnstc sequences from Wash and Hadamard bass and generaton by Gram-Schmdt orthogonazaton of pseudorandom sequences, whch woud sgnfcanty ncrease the detecton reabty compared to the tradtona PN generaton when the mage bock sze s sma [7]. On the other hand, for better spectrum spreadng of the embedded data the mage bock wth arge sze s desrabe. oreover f Wash Hadamard bass s used as spreadng codes the embedded data may be extracted by thrd party due to the determnstc nature of the bass functon. To sove the probems Wash/Hadamard bass functon s used here to moduate the code patterns. Each PN code s XOR-ed wth a row of Hadamard or Wash matrx of proper dmenson, whch s anaogous to the Wash coverng n DS-CDA dgta ceuar system (IS-95 deveoped by Qua comm. Inc [8-30]. C. Choce of moduaton functons The man dea behnd the mprovement n detecton reabty s to expot the knowedge about the sgna I (or more respectvey, the projecton of I on the watermark and accordngy to moduate the energy of the nserted watermark n order to compensate for the sgna nterference. To accompsh ths sgna adaptve SS watermarkng, we vary the amptude of the nserted chp sequence by usng near and power-aw transformaton functons. We show under, a fxed attack dstorton, better detecton reabty n such cases s possbe compared to conventona sgna non-adaptve SS scheme. We now have three dfferent forms of watermarked mage as foows: I I ± k.[ P ] (0 I I ± k. I.[ P ] ( µ I I ± I.[ P ] ( Equaton (0 represents conventona sgna non-adaptve SS watermarkng scheme where as ( and ( are sgna adaptve SS watermarkng scheme usng near and power-aw moduaton functons respectvey. TABLE I IPROVEENT IN ZERO AVERAGING FOR ATALB GENERATED PN SEQUENCE USING WALSH/HADAARD BASIS 0Code Patterns (PN sze (8 x 8 No. of 0s before moduaton No. of s before moduaton No. of 0s after moduaton No. of s after moduaton PN PN PN PN PN In a the above equatons I represents watermarked mage coeffcents, I are mage coeffcents, P s the code pattern wth ength equa to the mage bock, k s the gan factor or moduaton ndex, k s the rato of mnmum mage coeffcent to maxmum mage coeffcent, a negatve quantty, µ s the moduaton ndex n power ow moduaton functon. The moduaton ndex µ s negatve numerc quantty wth vaue 0 to -. Detecton reabty n each case s determned by the stabty of the decson varabe t wth respect to a gven attack dstorton. The expresson of t for a partcuar P s rewrtten here for convenence of our further anayss. t < I', P > < I', P > (3 Internatona Schoary and Scentfc Research & Innovaton ( schoar.waset.org/ /840

4 Word Academy of Scence, Engneerng and Technoogy Internatona Journa of Computer and Informaton Engneerng Vo:, No:6, 007 Internatona Scence Index, Computer and Informaton Engneerng Vo:, No:6, 007 waset.org/pubcaton/840 where s the ength of the sequence. If we substtute the vaue of I from (0, ( and ( nto (3, the expressons of t becomes as foows respectvey: t / ( I ± k.p ( P t t / ( I. P ± k. P / (I ± k.i.p ( P / ( I. P ± k. I. P µ / ( I ± I.P( P (4 (5 (a (c Fg. 3 Reducton n correaton usng Wash/Hadamard bass; hm ndcates the correaton after moduaton / ( I. P ± I µ. P (6 It s aready shown that mproved detecton requres zero averagng sequence.e. n the code pattern number of zero shoud be equa to the number of one. Dfferentatng (4, (5 and (6 wth respect to I and nvokng the above property of the code pattern, we do have the expressons of dt / di as foows: dt dt / di (0. / di P +. (0. +. ( + k dt + / di ( P + kp µ. I k + ( ( µ (0. P + µ. P I +. ( µ P (7 (8 (9 We have defned k I (mn / I (max n (8, a negatve quantty and et k - k, the expresson of dt / di for the equaton becomes as foows: dt / di ( k / < dt / di (0 In (7, (9 and (0 dt / di denote the change of decson varabe t wth respect to the change of I.e. a measure of nose mmunty n the detecton process. Lower vaue of dt / di ndcates better detecton reabty. IV. PROPOSED WATERARKING SCHEE Internatona Schoary and Scentfc Research & Innovaton ( schoar.waset.org/ /840

5 Word Academy of Scence, Engneerng and Technoogy Internatona Journa of Computer and Informaton Engneerng Vo:, No:6, 007 Internatona Scence Index, Computer and Informaton Engneerng Vo:, No:6, 007 waset.org/pubcaton/840 A. Basc spread spectrum watermarkng Watermark nformaton (ogo/data, audo data and vdeo data s embedded accordng to (3 n channes H, H 3, H 4, H 4 and H 4, H 4, H 43 and H 3 of -band ( 4 decomposton of the cover mage. For each watermark bt, one PN sequence of ength equa to the sze of each channe s generated. If the code PN s used for data embeddng n H 4, H 4, H 43 or H 3 channe, then the orthogona code (PN Fg. 4 Bock dagram of SS watermarkng scheme obtaned by compementng the bts of PN code are used for data embeddng n H, H 3, H 4 or H 4 channe. Here the data/ogo s redundanty embedded n H 4 and H channes, audo nformaton n H 4 and H 3 channes and vdeo data n H 43, H 3 and H 4, H 4 channes. Here our objectve s to transmt mutmeda data. Therefore the sgnfcant nformaton s encoded wth dstnct set of keys n four channes and other two types of nsgnfcant nformaton s embedded usng the same set of keys n ony two channes Fg. 5 Bock dagram of watermark decodng each. Snce data embeddng regons are dfferent n each type of nformaton, so there woud a tte nterference despte the same set of keys for nsgnfcant messages. In ths way we can transmt our sgnfcant mutmeda nformaton under the camoufage of nsgnfcant data. B. -ary moduaton and demoduaton n SS watermarkng In dgta communcaton the smaest nformaton entty s caed as symbo where n bnary sgnang there are two dfferent symbos and n - ary ( > sgnang there are more than two numbers of dfferent symbos [, ]. The n -ary refers to the number of symbos used n communcaton scheme. From communcaton theory we know that for a certan moduaton scheme an ncrease n the number of the symbo decreases the symbo error probabty. Now the use of -ary moduaton n dgta watermarkng s dscussed n the foowng paragraph. In -ary sgnang f more number of bts are used to represent a symbo of fxed ength bnary message, the fewer number of symbos decrease the error probabty by provdng the more ocatons of embeddng for each symbo [3, 4]. In transform doman SS watermarkng scheme, the mprovement n performance due to the usages of -ary moduaton can be expaned as foows. When more number of bts are used to from a symbo.e. the arger the vaue of, ess be the numbers of symbos need to be hdden for a gven ength of bnary message and there s the scope of choosng of hgher moduaton ndex vaue. In other words for a gven embeddng dstorton, hgher moduaton ndex vaue can be chosen n -ary sgnang compared to bnary sgnang. The hgher moduaton ndex vaues mprove robustness performance.e. reabty of transmsson through nosy channe. Fg. shows the bock dagram of -ary moduaton based SS watermark embeddng scheme. -ary moduaton scheme frst maps a L bt ong message (say of two symbos sgna to dfferent symbos message by groupng og ( bts of the orgna message to one symbo. Each symbo s represented by a b-eve spread spectrum moduaton functon (PN matrx and the tota number of functons n the bass set S (where N shoud be equa to the tota number of symbo message. So f ( m, where each symbo s represented by m bts and N represent the number of dfferent dstnct symbos and tota number of symbos n the symbo message respectvey, we need dfferent sets of code patterns each havng N number of b-eve moduaton functon. Fg. 7 shows the bock dagram of watermark decodng n -ary system. To decode a symbo at one partcuar poston, correaton between the embedded mage bock and the moduaton functons of that partcuar poston for a the sets of keys are cacuated. The ndex of the argest correaton coeffcent.e. the partcuar set of key whose moduaton functon of respectve poston yeds the maxmum correaton vaue determnes the decoded symbo. C. Desgn of experment Internatona Schoary and Scentfc Research & Innovaton ( schoar.waset.org/ /840

6 Word Academy of Scence, Engneerng and Technoogy Internatona Journa of Computer and Informaton Engneerng Vo:, No:6, 007 Internatona Scence Index, Computer and Informaton Engneerng Vo:, No:6, 007 waset.org/pubcaton/840 We empoy -ary moduaton n SS data embeddng for dgta mages athough the experments, whch can aso be done for other knd of data, ke audo, musc, vdeo etc. The Fg. 6 Bock dagram of -ary based SS watermarkng scheme cover mage I s a gray eve mage of sze (N*N where N p and the reference watermark w s a bnary mage of sze (* where n. The vaues of p and n, ndcate the sze of the cover and the watermark mage where p > n, typcay (p/n > 4. The proposed work consders a bnary mage sze (6 *6 as watermark and (56, 8 bts/pxe gray mage as cover mage. In order to show the better comparson of -ary moduaton scheme, data s embedded separatey n spata (drecty n pxe vaues doman as we as n transform doman. Athough there are many transformatons such as DCT, DFT, Fourer- en, DHT etc. but we use here dscrete waveet transform (DWT as t attracts attenton n varous mage processng appcaton ncudng dgta mage watermarkng, detecton, de nosng and the upcomng mage compresson standard JPEG-000 due to ts co-jont representaton of the mage sgna [5]. The norma DWT ( Fg. 7 Bock dagram of watermark decodng n -ary system band system decomposes an mage sgna nto LL (ow- ow, LH (ow- hgh, HL (hgh-ow and HH (hgh-hgh sub bands whe the -band dscrete waveet transformaton (bdwt system decomposes the same nto (* channes, correspondng to dfferent drecton and resouton. In case of DWT decomposton these sub bands correspond to the coarse approxmaton, horzonta, vertca and dagona deta of the mage sgna respectvey. In order to accompsh better spectrum spreadng we embedded data n LL and HH sub bands ( band system and few seected channes (-band system wth ow and hgh varance vaues. In the case of - band waveet decomposton (where 4 resuts obtaned from the arge number of mages show that the varance vaues for the coeffcents of sub bands H, H 3, H 4, H 4 are aways n the ower range and for the sub bands H 4, H 4, H 43, H 3, are n the upper range. A set of bnary vaued Pseudo Number (PN equa to the sze of mage bock are generated and for each PN the orthogona code PN s obtaned by compementng the bts of PN code. Each PN and PN matrx are moduated by one row of Hadamard matrx wth proper dmenson. Hadamard bass s used to decrease crosscorreaton among PN codes. Ths s possbe snce rows and coumns are orthogona to each other. If the code PN s used for data embeddng n LL sub band (H, H 3, H 4, H 4, the orthogona code PN obtaned by compementng the bts of PN are used for data embeddng n HH sub band (H 4, H 4, H 43, H 3. The use of PN and PN ndcates ow correspondng mage bocks.e. sub bans or channes and property (3 of the code pattern s thus fufed. V. RESULTS AND DISCUSSION We test the effect of severa above-mentoned factors n robustness mprovement over arge number of bench marked mages ke Fshng Boat, Lena, Ps, US ar force etc. We use peak sgna to nose rato (PSNR and structura smarty ndex measurement (SSI as representatve objectve measure of data mperceptbty where as reatve entropy dstance (Kuback Leber dstance as measure of securty (ε. Numerca resuts for these measures are shown n tabe II and III. Varous non macous as we as deberate mage degradaton n the form of near & nonnear fterng, sharpenng, hstogram equazaton, change n dynamc range, couson, corrupton by addtve nose and specke and random nose, ossy mage compresson ke JPEG & JPEG-000 etc. have been smuated over the watermarked mages as proof of the mprovement n detecton reabty. However we ony show the mprovement n detecton aganst addtve whte gaussan nose and JPEG & JPEG-000 compresson. We treat dgta watermarkng ads a probem of dgta communcaton and mutua nformaton s consdered as objectve measure to quantfy the robustness effcency. Internatona Schoary and Scentfc Research & Innovaton ( schoar.waset.org/ /840

7 Word Academy of Scence, Engneerng and Technoogy Internatona Journa of Computer and Informaton Engneerng Vo:, No:6, 007 Expermenta resuts show that data embeddng n -band coeffcents show better robustness compared to DWT doman ncreased varance vaue s the reason of ow correaton between the mage bocks and code patterns (9. Low correaton vaue reduces the vaue of dt / di for a gven attack dstorton. In case of -ary moduaton, SS watermarkng scheme s TABLE II RESULTS OF PSNR, SSI SECURITY VALUE FOR FISHING BOAT IAGE USING DWT AND -BAND DECOPOSITION Decomposton Technque PSNR vaue (db SSI vaue Securty (ε vaue DWT Band Internatona Scence Index, Computer and Informaton Engneerng Vo:, No:6, 007 waset.org/pubcaton/840 (a Fg. 8 Robustness resuts aganst (a JPEG; JPEG-000 compresson; hm ndcates moduaton usng Hadamard bass embeddng athough the performance of the atter s much better wth respect to other SS watermarkng schemes mpemented by usng DCT, Wash-Hadamard transformaton. Fg. 8 and 9 show graphcay the mprovement n detecton reabty aganst JPEG, JPEG 000 compresson for -band over DWT decomposton. The fgures aso hghght the effect of Wash-Hadamard bass n mprovng the robustness effcency. The roe of usng dfferent moduaton functons n robustness mprovement s shown n tabe aganst addtve whte gaussan nose. The fgure shows that wth the ncrease of depth of nose addton near moduaton functon show better detecton compared to tradtona SS watermarkng scheme. Ths s qute cear from (7 and (0 where for a gven attack dstorton change n decson varabe t n case of near moduaton functon I ess compared to conventona moduaton functon. Under the same crcumstance a better detecton s acheved n power-aw functon s used nstead of near moduaton functon. Ths s due to the fact that wth the decrease n numerca vaue of µ, the varance of the embedded coeffcents of the sub bands s ncreased. The apped over arge number of bench marked mages. Data s embedded separatey n spata doman, LL and HH sub bands obtaned after DWT decomposton and few seectve sub channes of ow and hgh varance bocks after -band decomposton. The performance of -ary moduaton s shown n Fg. 0 where t s qute cear that wth the ncrease of vaue robustness effcency s aso mproved but at the same tme computaton cost of decodng s aso ncreased. The reason for the atter pont s that for decodng a partcuar symbo the embedded channe s projected onto the a moduaton functons of that partcuar poston for each set of the key. It s aso cear from the fgures that robustness performance for any vaue of n -ary moduaton s better for -band waveet decomposton compared to DWT decomposton whe the performance of the atter s comparatvey better than spata doman SS embeddng scheme. The resut s reported for JPEG-000 compresson operaton; however the resut s aso vad for other types of sgna processng operaton such as near & nonnear fterng, sharpenng, hstogram equazaton, change n Fg. 9 Robustness resut of dfferent moduaton functons for fxed gaussan nose dstorton Internatona Schoary and Scentfc Research & Innovaton ( schoar.waset.org/ /840

8 Word Academy of Scence, Engneerng and Technoogy Internatona Journa of Computer and Informaton Engneerng Vo:, No:6, 007 Internatona Scence Index, Computer and Informaton Engneerng Vo:, No:6, 007 waset.org/pubcaton/840 (a Fg. 0 -ary moduaton for (a JPEG compresson; JPEG-000 compresson dynamc range, corrupton by addtve nose as we as other popuar ossy mage compresson ke JPEG. A. Increased Capacty usng mutpe watermarkng & borthogona waveet. The capacty of the watermarked mage can be ncreased usng Spread Spectrum utpe watermarkng scheme. Here, the mutpe watermarks are embedded over the same cover mage usng the same basc watermarkng technque but wth dstnct set of keys. Therefore more than one authentcate user can easy use the same cover mage for data embeddng and compete secrecy of ther nformaton (.e. watermark can be mantaned among those users. Snce Spread Spectrum technque s used for data embeddng and t s popuar for ts ant-jammng and nterference rejecton property, so ths scheme shows better mperceptbty as we as robustness compared to the other non Spread Spectrum technques. The dsadvantage of ths mutpe watermarkng s that, athough Spread Spectrum s used for watermarkng but embeddng dstorton can t be gnored. Snce a the watermarks are embedded n the same drecton of the waveet decomposton of the cover mage so naturay adjacent channe nterference woud be the maxmum. Therefore the securty of the dfferent hdden messages aso woud be affected. In order to ncrease the capacty of the watermarked mage wth mnmum embeddng dstorton, the borthogona waveet transform s used here. Its speca feature s that t provdes two scang functons nstead of one and therefore just ke quadrature phase shft keyng ths transform gves four possbe drecton of decomposton. If the correatons among a these drectons by takng two at a tme are seem to be mnmum then we can nfer that a the co-channe and adjacent channe nterference w be mnmum. So we can expect a better mperceptbty, robustness as we as securty of the hdden data. Another advantage s that snce borthogona waveet decomposton provdes four data embeddng drecton, so the four dfferent watermarks can be embedded usng the same set of key nstead of dfferent sets of key. Therefore, the tme compexty of the key rather code generaton woud be much smpfed. B. Successve bt/watermark canceaton method (a Fg. (a JPEG compresson (wth orthogonaty; JPEG compresson (wthout orthogonaty Internatona Schoary and Scentfc Research & Innovaton ( schoar.waset.org/ /840

9 Word Academy of Scence, Engneerng and Technoogy Internatona Journa of Computer and Informaton Engneerng Vo:, No:6, 007 Each bt of the watermark s retreved on the bass of crosscorreaton threshod vaue. Snce mutpe watermarks (text, audo and vdeo.e. mutmeda data are smutaneousy embedded on the same cover mage, therefore the nterference among them woud greaty mpact on the decodng process.e. Detecton reabty s aso mproved by empoyng the successve watermark/bt canceaton method resutng n reduced nterference among the embedded watermarks. The data embeddng capacty s greaty ncreased by usng borthogona waveets. Proposed SS watermarkng scheme aso offers vsua and statstca nvsbty of the hdden data. Here, the embedded watermarks are not ony mage but aso audo or vdeo cps etc. So, the reported resuts are based on mutmeda data as watermarks, whch are smutaneousy embedded over the same mage. ACKNOWLEDGEENTS Dr. Subr Kumar Sarkar acknowedges the fna assstance obtaned from the Center for obe Computng and Communcaton, Jadavpur Unversty. Internatona Scence Index, Computer and Informaton Engneerng Vo:, No:6, 007 waset.org/pubcaton/840 (a Fg. (a JPEG-000 compresson (wth orthogonaty; Securty (wth & wthout orthogonaty for mutpe messages on the stabty of the threshod vaues. In order to dehde each type of watermark propery, ths nterference has to be mnmzed. Therefore, here successve watermark canceaton as we as successve bt canceaton methods are empoyed durng the watermark recovers process for ths purpose. The resuts show that successve bt canceaton method s far better than successve watermark canceaton and norma decodng process as t greaty decreases ths nterference. VI. CONCLUSION In ths paper, we have crtcay anayzed few ssues that have sgnfcant mpact on the detecton reabty n SS watermarkng for the purpose of secure data communcaton. It s found that data embeddng n the channes (H, H 3, H 4, H 4 or H 4, H 4, H 43, H 3 of -band decomposton offers hgher resency aganst varous types of mage dstorton. Detecton reabty s mproved by ncreasng orthogonaty among code patterns usng Wash-Hadamard bass functons. (a Fg. 3 (a Successve bt and watermark canceaton on JPEG 000 compressed mage; Successve watermark canceaton (Four dfferent watermarks, N, W & E are embedded usng bor 6.8 fter and form a JPEG000 compressed mage Internatona Schoary and Scentfc Research & Innovaton ( schoar.waset.org/ /840

10 Word Academy of Scence, Engneerng and Technoogy Internatona Journa of Computer and Informaton Engneerng Vo:, No:6, 007 Internatona Scence Index, Computer and Informaton Engneerng Vo:, No:6, 007 waset.org/pubcaton/840 TABLE III RESULTS OF PSNR, SSI AND SECURITY VALUE USING BI- ORTHOGONAL WAVELET DECOPOSITION AFTER EBEDDING FOUR WATERARKS Name of the Image Fshng Boat Name of the borthogona Fter PSNR (db REFERENCES SSI Securty Vaue bor bor bor [] I. J. Cox and. L. er, A Revew of Watermarkng and the Importance of perceptua odeng, Proc of Eectronc Images, Feb [] N. Nkoads, I. Ptas, Robust Image Watermarkng n the Spata Doman, Esevs Sgna Processng, Sept. 997, pp [3] G.W. Braudaway, Protectng pubcy avaabe mages wth an nvsbe mage Watermark, Int. Conf. On Image Processng, vo. Oct [4] C.T. Hsu, J. L. Wu, Hdden sgnatures n mages, Int. Conf. On Image Process, vo. 3, Lausanne, Swtzerand, Sept [5] P. aun and A. Ivanovc, IEEE Trans on sgna processng, Apr. 003, vo. 5, pp [6] X. Nu, S. Sun, W. Xang, ut-resouton Watermarkng for Vdeo Based on Gray-eve Dgta Watermark, IEEE Trans. Consumer Eectroncs vo. 46, ay 000. [7] W. Bender, D. Gruh, N ormoto and A. Lu, Technques for Data Hdng, IB systems journa, vo. 35, pp 3-4, 996. [8] G. azzn, G. Sett and R. Rovatt, Chaotc compex spreadng sequences for asynchronous DS-CDA, IEEE Trans on Crcuts & Systems 97. [9] A. B. Watson, G. Y. Yang, J. A. Soomon, J. Vasenor, Vsbty of Waveet Quantzaton Nose, IEEE Trans on Image Processng vo. 6, August 997. [0] B. L. W. Zeng, S. Le, Eextracton of mutresouton watermark mages for resovng rghtfu ownershp, Proc. of SPIE, Securty and Watermarkng of utmeda Contents, January 999. [] I. J. Cox, J. Kan, F. T. Leghton, T. Shamoon, Secured Spread Spectrum Watermarkng for utmeda, IEEE Trans on Image Processng, vo. 6, pp 60-60, June 000. []. Ban, F. Barton and A. De, Rosa, on the performance of mutpcatve spread spectrum watermarkng, proc. on IEEE work on SP 0 [3] I. J. Cox and. L. er, and A. L. ckeps, Watermarkng as communcatons wth sde nformaton, Proc. of IEEE 99. [4] S. Craver, N. emon, B. L. Yeo and.. Yeung, Resovng Rghtfu Ownershp wth Invsbe Watermarkng Technques: Lmtatons, Attacks and Impcatons, IEEE Journa of Seected Areas n Communcatons 98. [5] J.R.Hernandez and F.Perez- Gonzaes, Statstca anayss of watermarkng schemes for copyrght protecton of mages, Proc. of IEEE, 99. [6] N. erhav, Random codng error exponents of watermarkng systems, IEEE Trans on Informaton Theory 000. [7]. Barn, F. Barton, A. Pva, Improved waveet-based watermarkng through pxe-wse maskng, IEEE Trans. on Image Processng, pp , 00. [8] W. Det, P. eerwad, A. Uh, Protecton of waveet based watermarkng systems usng fter parameterzatons, Sgna Processng. October 003. [9] D. Kundur, D. Hatznakos, Towards Robust ogo watermarkng usng mutresouton mage fuson, IEEE Trans. utmeda 6, Feb [0] P. eerwad, A. Uh, A survey of waveet-doman watermarkng agorthms, In Proc. of SPIE, Eectronc Imagng, Securty And Watermarkng of utmeda ContentsIII, CA, USA434, Jan. 00. [] Wang, Y., Doherty, J. F., VanDyck, R.E., A waveet based watermarkng agorthm for ownershp verfcaton of dgta mages, IEEE Trans. Image Process, February 00. []. K. Kundu and. Acharya, -Band Waveets: Appcaton to Texture Segmentaton for Rea Lfe Image Anayss, Proc. Internatona journa of waveets, utresouton and Informaton proceedng, vo., pp5-49, 003. [3] Phppe D., Benot., and Drk T.. S., Sgna adapted mutresouton transform for mage codng, IEEE Trans on Informaton Theory, 38(, arch 99. [4] H. S. avar and D. A. F Forenco, Improved Spread Spectrum: A new oduaton Technque for Robust Watermarkng, IEEE Trans on sgna processng, Apr 003, pp [5] A. Chouy, A Braja and S. Jourdan, Orthogona mutcarrer technques apped to drect sequence spread spectrum CDA systems, Proc of IEEE GLOBECO 93. [6] S. Hara and R. Prasad, Overvew of mutcarrer CDA, IEEE communcaton ag 97. [7] V.. Dasva and E. S. Sousa, Performance of orthogona CDA codes for quas-synchronous communcaton systems, Proc. of IEEE ICUPC 93. [8] S. Kaser, OFD-CDA vs. DS-CDA: Performance evauaton for fadng channes, Proc. of IEEE ICC 95. [9] L. Vandendorpe, uttone Spread Spectrum utpe Access Communcatons System n a utpath Rcan Fadng Channe, IEEE Trans Vehc 95. [30] N. Yee, L. Lnnartz and G. Fettwes, utcarrer CDA n Indoor Wreess Rado Networks, Proc of IEEE PIRC 93. Trtha Sankar Das became member of IEEE n 005. He receved hs B. Tech degree n Eectroncs and Teecommuncaton Engneerng from Vdyasagar Unversty, WB, Inda n 00 and.e. degree from Benga Engneerng & Scence Unversty, Shbpore n 004. At present he s a Lecturer n Eectroncs and Communcaton at Gurunanak Insttute of Technoogy, Panhat, Kokata, Inda. He s currenty dong hs Ph.D under Prof. S.K.Sarkar n Jadavpur Unversty. Hs fed of nterest spans dgta mage processng, sgna processng, communcaton and VLSI. Ayan Kumar Sau receved hs B. Tech degree n Eectroncs and Teecommuncaton from Kayan Unversty, WB, Inda n 00. At present dong hs.e. degree n Eectroncs and Tee-communcaton from Jadavpur Unversty, Inda. Hs area of nterest spans dgta mage processng, watermarkng of mutmeda data and many other chaengng probems n data hdng data and data encrypton. Dr. Subr Kumar Sarkar receved hs B. Tech. from Unversty of Cacutta, Inda n 98,. Tech. from Unversty of Cacutta, Inda n 983 and Ph. D (Tech from Insttute of Rado Physcs and Eectroncs, Unversty of Cacutta n999. From 99 to 999 he was ecturer n Benga Engneerng and Scence Unversty, Shbpore, Howrah. Currenty he s Reader n Jadavpur Unversty, Kokata, Inda, snce 999. Hs present fed of nterest s appcaton of soft computng toos n smuatons of devce modes and aso n the fed of Hgh Frequency and Low Power Consumng Devces and ther parameter optmzaton and wreess mobe communcaton. Hs research nterests aso ncude snge eectron devces and next generaton dgta eectroncs and mage processng. Internatona Schoary and Scentfc Research & Innovaton ( schoar.waset.org/ /840