Keywords Semi-Fragile, Digital watermarking, Toeplitz matrix, Image Authentication, Content based watermarking.

Transcription

1 Volume 5, Issue 1, Jnury 2015 ISSN: X Interntionl Journl of Advnced Reserch in Computer Science nd Softwre Engineering Reserch Pper Avilble online t: Exploittion of Toeplitz Mtrix in ttining Semi-Frgile Wtermrking Technique S. S. Sujth Associte Professor in Computer Science South Trvncore Hindu College, Ngercoil, Indi Abstrct Digitl Wtermrking is known s good solution for chieving copyright protection nd content integrity protection by embedding the secret dt into the content. The wtermrk cn lter be detected to verify the ownership when dispute over the copyright or integrity of the digitl content rises. Semi-frgile methods re robust to incidentl modifictions but fgile to mlicious ttcks. The im of this pper is to propose semi-frgile techniqe which is ble to discriminte intentionl mnipultions from dmissible mnipultions.the binry wtermrk is designed in the form of Toeplitz mtrix nd is embedded in the high frequency sub-bnd of Discrete Wvelet Trnsform. The robustness of the lgorithm ws ddressed in [22]. The frgility of the method is identified under mlicious ttcks nd is mesured with metric Similrity Rtio.Experimentl results show tht the technique preserves imperceptbility nd indictes the mlicious ttcks. Keywords Semi-Frgile, Digitl wtermrking, Toeplitz mtrix, Imge Authentiction, Content bsed wtermrking. I. INTRODUCTION Informtion security plys very importnt role in modern computing informtion systems. The need for securing informtion rises from the use of computers in sensitive fields such s government, militry, bnking nd industries. With the globl cceptnce of Internet, virtully every computer in the world is connected to every other. Nowdys, millions of people use Internet for immedite communiction nd trnsformtion of vst informtion. Aprt from these opportunities, Internet hs creted new risks for the users. Since copying nd editing n imge is crried out esily with powerful imge editing softwre, uthentiction nd copyright protection from unuthorized mnipultion of digitl imge becomes n importnt issue in the field of digitl medi. Authentiction of digitl document ssures the recipient tht the document hs been received from the right person. The im of uthentiction is to provide method to uthenticte the imge nd ssure the integrity of tht imge but not to protect the contents from being copied or stolen. Imge uthentiction is chieved by technique clled Digitl Wtermrking. Wtermrking for imge uthentiction hs four types [23] s mentioned below: Exct uthentiction - verifies tht n imge hs not been ltered t ll since it hs left trusted prty. Even if single bit chnges, the imge is regrded s inuthentic. Selective uthentiction - verifies tht n imge hs not been ltered significntly. Therefore, if single or even smll group of bits hve been chnged, the imge is regrded s uthentic. However, distortions leding to chnge in the perceptul qulity of the imge re regrded s inuthentic. Locliztion - identifies the regions of the imge tht hve been corrupted, nd identifies the remining regions tht re unltered. Restortion - identifies the regions of the imge tht hve been corrupted, nd restores the ltered imge. Visible digitl imge wtermrking nd Invisible imge wtermrking techniques re the tw ctegories of digitl imge wtermrking scheme. In visible wtermrking, the informtion is visible in the picture or video. Typiclly, the informtion is text or logo which identifies the owner of the originl document. In invisible wtermrking, informtion is dded s digitl dt to udio, picture or video, but it cnnot be perceived s such[21]. Further, the invisible wtermrks re ctegorized into wtermrking techniques s robust, frgile nd semi-frgile. Robust Robust techniques re insensitive to incidentl distortions nd ccept n imge tht hs undergone such mnipultions. These lgorithms ensure tht the imge processing opertions do not erse the embedded wtermrk signl. Frgile In frgile techniques [2], even one bit chnge in imge is not llowble. They re minly pplied to content uthentiction nd integrity ttesttion, becuse they re sensitive to lmost ll modifictions. Semi-frgile Semi-frgile methods [3] re robust to incidentl modifictions such s JPEG compression, but frgile to other modifictions such s high impct dditive noises. Tht is, some incidentl imge mnipultions hve to be considered llowble during the process of medi trnsmission nd storge, while other mlicious modifictions (e.g. ltertion of content) from ttckers should be rejected. Intentionl distortion 2015, IJARCSSE All Rights Reserved Pge 331

2 A detiled survey on severl wtermrking methods hve been provided in [4][5]. Two ctegories of Digitl wtermrking lgorithms re sptil-domin techniques nd frequency-domin techniques. Lest Significnt Bit (LSB) is the simplest technique in the sptil domin techniques [6] which directly modifies the intensities of some selected pixels. The frequency domin technique trnsforms n imge into set of frequency domin coefficients [7]. The trnsformtion dopted my be discrete cosine trnsform (DCT), discrete Fourier trnsforms (DFT) nd discrete wvelet trnsforms (DWT) etc. After pplying trnsformtion, wtermrk is embedded in the trnsformed coefficients of the imge such tht wtermrk is not visible. Finlly, the wtermrked imge is obtined by cquiring inverse trnsformtion of the coefficients. A detil survey on wvelet bsed wtermrking techniques nd semi-frgile wtermrking lgorithms. cn be found in [8] nd [11] respectively. Tsi nd Chein [10] designed semi-frgile wtermrk from low frequency bnd nd embedding the wtermrk into the high frequency bnd by the help of HVS. The robustness of the lgorithm is nlyzed for mild modifictions like JPEG compression nd frgility to mlicious ttcks. proposed n integer wvelet bsed Multiple logo wtermrking scheme, the wtermrk is permuted using Arnold trnsform nd is embedded by modifying the coefficients of the HH nd LL subbnds. Qiwei et l.[11] put forwrd DWT bsed blind wtermrking scheme by scrmbling the wtermrk using chos sequence. Mny of the lgorithms proposed meet the imperceptibility requirement quite esily but robustness to different imge processing ttcks is the key chllenge nd the lgorithms in literture ddressed only subset of ttcks. A semi-frgile wtermrking technique presented by Zhou X. et l. [13] constructs wtermrk from the originl imge nd inserts this wtermrk bck into the imge. Thus, it voids dditionl signture files. This scheme is tolernt of lossy compression such s JPEG, but mlicious chnges of the imge will result in the violtion of the wtermrk detection. A survey on Lin et.l.[4] proposed n imge uthentiction method tht cn differentite the prcticl JPEG lossy bseline compression with predefined LAJQ from mlicious mnipultion. Hung et l. [15] uses the block vector quntiztion indices for uthentiction dt. The shortcomings of the reported wtermrking schemes re insecurity nd low robustness to incidentl opertions. This pper proposes novel DWT bsed blind semi-frgile wtermrking scheme which ssures security by designing binry wtermrk pttern in the form of Toeplitz mtrix. The rest of this pper is orgnized s follows: Section 2 gives n overview of Discrete Wvelet Trnsform, nd Toeplitz mtrix. The detils of wtermrk genertion, embedding nd extrction processes re explined in Section 3. Section 4 presents experimentl results nd discussion. The pper is concluded in section 5. II. RELATED BACKGROUND A. Discrete Wvelet Trnsform The DWT use wvelet filters like Hr, Dubechies Orthogonl nd Dubechies Bi-Orthogonl nd decomposes input imge into four components nmely LL, HL, LH nd HH. The lowest resolution level LL consists of the pproximtion prt of the originl imge. The remining three resolution levels consist of the detil prts nd give the verticl high (LH), horizontl high (HL) nd high (HH) frequencies. In the proposed lgorithm, wtermrk is embedded into the host imge by modifying the coefficients of high-frequency bnds i.e. HH subbnd. B. Toeplitz Mtrix In mthemticl discipline of liner lgebr, Toeplitz mtrix or digonl constnt mtrix is mtrix in which ech descending digonl from left to right is constnt. It ws nmed fter Otto Toeplitz nd is frequently encountered in pplictions where mtrix computtion is exploited in order to devise very effective numericl solution lgorithm. For instnce, Toeplitz mtrix is squre mtrix with constnt digonls nd for constructing Toeplitz mtrix of order NxN, we need 2N-1 elements. Let P= {, b, c, d, e, f, g, h, i}. Since there re 9 elements in this vector, Toeplitz mtrix of size 5x5 my be constructed so tht its first row is first 5 elements nd first column except the first element is lst 4 elements, which is given in eqution (1). The remining elements re clculted ccording to eqution (2). b c d e f b c d T = g f h g i h g f f b c b A ( i, j) A( i 1, j 1) (2) III. PROPOSED METHOD The successive stges of the proposed semi-frgile wtermrking lgorithm re provided in figure 1. A. Wtermrk genertion The wtermrk genertion lgorithm mkes use of the Toeplitz mtrix to design wtermrk in the binry form. The procedure is described in detil s follows: 2015, IJARCSSE All Rights Reserved Pge 332 (1)

3 Host imge Wtermrk genertion Wtermrk Wtermrk embedding Host imge with wtermrk Secured prt Publishing Attcks Host imge with wtermrk Secured or insecured prt Detecting wtermrk Result of checking Fig. 1 Frmework of proposed wtermrking process Input: The host imge P of size M x M Output: The wtermrk W of size M/2 x M/2 ) Consider the originl imge P of size MxM. b) Rndomly select M-1 elements from P so tht ll the pixels hve vlid 3x3 neighborhoods. c) Let the neighborhoods of the selected pixel P(x, y) re P(x-1, y-1), P(x-1, y), P(x-, y+1), P(x,y+1), P(x+1,y+1), P(x+1,y),P(x+1,y-1), nd P(x, y-1) d) Find verge vlue of those neighborhoods. Let it be P (x, y). e) A binry sequence B cn be obtined by pplying the following constrint. 0 if P( x, y) P ( x, y) B i 1 otherwise where i=1, 2, 3,, M-1. f) A Toeplitz mtrix of size M/2 x M/2 is constructed from the binry sequence B using the following procedure. First row elements re clculted ccording to eqution (3), first column elements except the first re clculted ccording to eqution (4) nd remining elements re computed by using eqution (2). T ( 1, j) B( i j 1) (3) T ( i,1) B( i M / 2 1) (4) where 2 i M / 2, nd 1 j M / 2 Now T is the wtermrk pttern to be embedded within host imge. B. Wtermrk embedding The lgorithm embeds the wtermrk in the high frequency subbnd of host imge. The detiled steps re listed s follows: Input : The host imge of size nd wtermrk. Output: The wtermrked imge ) Use hr wvelet nd pply 1-level DWT on originl imge. b) It decomposes the imge into four sub-bnds nmely LL1, HL1, LH1 nd HH1. c) The wtermrk is embedded in the high frequency component HH1 of DWT by replcing it with the wtermrk. d) Perform inverse wvelet trnsform to obtin the wtermrked imge. C. Wtermrk Detection Proposed wtermrking scheme extrcts the embedded wtermrk nd reconstructs wtermrk informtion from wtermrked imge. Thus the lgorithm does not require the originl imge in the detection phse nd hence it is referred s blind wtermrking. The uthentiction process includes the following steps: 2015, IJARCSSE All Rights Reserved Pge 333

4 Input : The wtermrked imge. Output : The extrcted nd reconstructed wtermrk ) Wtermrk is derived form the content of wtermrked imge using the steps described under wtermrk genertion in section III.A. b) Apply 1-level DWT to the wtermrked imge nd extrct the embedded wtermrk from HH1 subbnd. c) Compre the two wtermrks (derived nd extrcted). If two vlues mtch, uthenticity is preserved. Otherwise the uthenticity is suspected. d) Qulity of wtermrked imge nd the wtermrk is found out ccording to eqution (5) nd (7). IV. EXPERIMENTAL RESULTS The vlidity of the proposed lgorithm is studied in this section by conducting mny experiments under different cover imges nd wtermrks. For testing, the size of the originl imge is tken s 512x512. Subsequently, the wtermrk is constructed from the perceptul informtion of originl imge in the form of binry Toeplitz mtrix pttern. In the experiment, the pek signl to noise rtio (PSNR) s defined in (5) is used to mesure the embedding distortion, nd Similrity Rtio (SR) s defined in (7) is used to mesure the robustness nd frgility PSNR 10 log (5) 10 MSE where MSE is Men Squred Error between originl nd distorted imges, which is defined in eqution (4). M 1N 1 i0 j0 [ OI ( i, j) DI ( i, j)] MSE (6) MxN where OI is originl imge nd DI is the distorted imge. S SR (7) S D where S denotes number of mtching pixel vlues nd D denotes number of different pixel vlues. 2 () (b) Fig 2. Input nd Processed Imges () Originl imge, (b) Wtermrked imge The wtermrked imge is shown in Fig.2, nd its PSNR is db which indictes tht the proposed lgorithm is highly impercetible. Also, the high vlue (0.9801) obtined for SR between extrcted nd clculted wtermrks shows tht the number of mtching pixels is high nd hence uthenticity is preserved. To evlute the performnce of the proposed wtermrking scheme, experiments hve been conducted on the cover imge under some common imge processing ttcks which preserve the content of the imge nd the results re tbulted in Tble 1. The ttcks chosen were medin filtering, liner filtering, blurring, JPEG compression, intensity djustment nd histogrm equliztion. The Similrity Rtio obtined revels tht the proposed method is ble to withstnd ginst incidentl ttcks nd hence prove the robustness of the lgorithm. Tble 1 sr ginst common ttcks Attcks SR No Medin filtering 3x Liner filtering 3x Blurring JPEG compression 50% Imge djustment Histogrm Equliztion , IJARCSSE All Rights Reserved Pge 334

5 To determine the frgile nture of the lgorithm, the wtermrked imge hs been subjected to intentionl ttcks such s Gussin nd Slt&pepper dditive noises nd the results re tbulted in Tble 2 nd Tble 3. Tble 2 sr under gussin noise Vrince SR Tble 3 sr under slt&pepper noise Density SR An observtion on the Similrity Rtio provided in Tble 2 revels tht the proposed lgorithm is frgile to dditive Gussin noises t vrious vlues for vrince. Similrly, experimentl results provided in Tble 3 show tht the proposed lgorithm is ble to identify slt&pepper noisy ttcks for higher vlues of density. V. CONCLUSION Aprt from robust nd frgile wtermrking, the im of semi-frgile wtermrking is to tolerte the incidentl ttcks while indicting the intentionl ttcks if the imge is hevily processed. This study hs proposed semi-frgile wtermrking which designs binry wtermrk from the host imge in the form of Toeplitz mtrix nd hides the wtermrk imperceptibly in the digitl imge inorder to secure it. Wtermk is embedded in the high frequency sub-bnd of DWT nd is extrcted in the bsence of originl imge, which results in blind wtermrking scheme. The performnce of the wtermrking scheme is evluted with content preserving common imge processing ttcks nd content ltering intentionl ttcks. Experimentl results demonstrte tht the proposed scheme gurntee the sfety of the wtermrk, nd identifies mlicious ttcks while tolerting Filtering opertions, JPEG compression, Imge djustment nd histogrm equliztion. Hence the proposed technique is effective for imge uthentiction. REFERENCES [1] Tmne S.C., Mnz R. nd Deshmukh R.R., Digitl Wtermrking using Imge Fusion Method, Interntionl Journl of Recent Trends in Engineering, Vol.1, No.2, pp , [2] Megh Knsl, Sukhjeet K. Rnde nd Amndeep Kur, Frgile Wtermrking For Imge Authentiction Using Hierrchicl Mechnism, Interntionl Journl of Engineering Reserch nd Applictions, Vol. 2, No.4, pp , [3] Lin.C, Su.T nd Hsieh.W, Semi-Frgile Wtermrking Scheme for Authentiction of JPEG Imges, Tmkng Journl of Science nd Engineering, Vol.10, No.1, pp (2007). [4] Liu T. nd Qiu Z.D., The Survey of Digitl Wtermrking bsed Imge Authentiction Techniques, 6 th Interntionl Conference, pp , [5] C. Rey, J.Dugely: A survey of wtermrking lgorithm for Imge uthentiction. In: Journl on Applied Signl Processing, Vol.6, pp , [6] C.I.Podilchuk, E.J.Delp: Digitl wtermrking: lgorithms nd pplictions. In: IEEE Signl Processing Mgzine, pp , July [7] Arvind kumr Prthsrthy, Subhsh Kk: An Improved Method of Content Bsed Imge Wtermrking. In: IEEE Trnsction on brodcsting, Vol.53, no.2, June 2007, pp [8] Q.Ying nd W.Ying, A survey of wvelet-domin bsed digitl imge wtermrking lgorithm, Computer Engineering nd Applictions, Vol.11, pp.46-49, [9] Yun Yun, Deci Hung, nd Dunyng Liu, An Integer Wvelet Bsed Multiple Logo-wtermrking Scheme, IEEE, Vol.2 pp , [10] Tsi M.J. nd Chein C.C., A Wvelet bsed Frgile Wtermrking with Recovery Mechnism, Proceedings of IEEE, pp , [11] Qiwei Lin, Zhenhui Liu, nd Gui Feng, DWT bsed on wtermrking lgorihthm nd its implementing with DSP, IEEE Xplore, pp , [12] Ekiei O., Snkur B., Coskun B., et. l, Comprtive evlution of semifrgile wtermrking lgorithms, Journl of Electronic Imging, Vol.13(1), pp (2004) [13] Zhou X., Dun X. nd Wng D., A Semi-frgile Wtermrk Scheme for Imge Authentiction, IEEE Interntionl Conference on Multimedi Modeling, pp , [14] C. Y. Lin nd S. F. Chng, Semifrgile wtermrking for uthentiction JPEG visul content, Proc. SPIE 3971, (2000). [15] K.L.Hung, C.C.Cheng, nd T.S.Chen, Secure Discrete Cosine Trnsform Bsed Technique for Recovereble Tmper Proofing, Opt Eng. 40(9), pp (2001). [16] Xing-Gen Xi, Chrles G.Boncelet, Gonzlo: Wvelet Trnsform bsed wtermrk for digitl imges. In: OPTICS EXPRESS, 1998 Vol.3, No.12, pp , IJARCSSE All Rights Reserved Pge 335

6 [17] Snjeev Kumr, Blsubrmnin Rmn, Mnoj Thkur: Rel Coded Genetic Algorithm bsed Stereo imge Wtermrking. In: IJSDIA, 2009, Vol. 1 No.1 pp [18] Hongmei Liu, Junhui Ro, Xinzhi Yo: Feture Bsed Wtermrking Scheme for Imge Authentiction. In: IEEE, 2008, pp [19] J.Dittmnn: Content-frgile Wtermrking for Imge Authentiction. In: Proc. of SPIE, Security nd Wtermrking of Multimedi Contents III, vol.4314, pp , [20] Rfel C.Gonzlez, R.E.Woods,, Steven L. Eddins : Digitl Imge Processing Using MATLAB, Indi (2008). [21] Mohmed Sthik M., Sujth S. S., Authentiction of Digitl Imges by using Semi-frgile Wtermrking Technique, Interntionl Journl of Advnced Reserch in Computer Science nd Softwre Engineering, Vol.2, No.11,pp.39-44, Nov [22] Dr. M. Mohmed Sthik nd S.S.Sujth, Appliction of Toeplitz Mtrix in Wtermrking for Imge Authentiction, Proceedings of the IEEE Interntionl Conference on Computer, Communiction nd Electricl Technology (ICCCET-2011), Tirunelveli, Mrch [23] Cox I.J., Miller M. nd Bloom J., Digitl Wtermrking Principles nd Prctice, Morgn Kufmnn Publishers, , IJARCSSE All Rights Reserved Pge 336