CONTOURLET TRANSFORM BASED WATERMARKING FOR COLOUR IMAGES

Transcription

1 he Internatonal Journal of Multmeda & Its Applcatons (IJMA) Vol.3, No.1, February 211 CONOURLE RANSFORM BASED WAERMARKING FOR COLOUR IMAGES Dnesh Kumar 1 and Vjay Kumar 2 1 Department of Computer Scence & Engneerng, GJUS&, Hsar, Haryana dnesh_chutan@yahoo.com 2 Department of Computer Scence & Engneerng, JCDMCOE, Srsa, Haryana vjaykumarchahar@gmal.com ABSRAC hs paper presents a contourlet transform based atermarkng scheme for colour mages. It also makes use of dscrete cosne transform and sngular value decomposton. Dscrete cosne transform s appled on three colour components (Y, Cb and Cr), obtaned from the colour mage, to produce DC coeffcents. hese coeffcents are decomposed nto drectonal subbands usng 2- level contourlet transform. he atermark s also decomposed usng contourlet transform. he fourth drectonal subbands of cover mage and atermark as ell are further subjected to Laplacan pyramd decomposton. As a result, the lo pass subbands are obtaned. he sngular values of lopass subband of atermark are embedded nto sngular values of lopass subband of cover mage. he expermental results reveal that the proposed atermarkng scheme s robust aganst common mage processng attacks. he results also demonstrate good mperceptblty. KEYWORDS Contourlet ransform (C), Dscrete Cosne ransform (DC), Sngular Value Decomposton (SVD) 1. INRODUCION he technologcal advancement n the feld of multmeda and the avalablty of hgh speed netorks have enabled us to obtan, reproduce and dstrbute data th ease and th almost no loss n qualty. Under such a scenaro, the protecton of dgtal data has become ncreasngly mportant. Dgtal atermarkng s one among a number of solutons that can help provde copyrght protecton. It s beng frequently used n applcatons such as medcal magng, dgtal camera and vdeo-on-demand systems etc. [1]. he atermarkng technque ncorporates embeddng of atermark nto dgtal data that can later be extracted n order to clam the onershp of data. Such technques are broadly categorzed nto to groups, the spatal doman and frequency doman atermarkng technques [2, 4, 5]. he former embeds atermarks n spatal characterstcs of cover mage drectly hereas the latter transforms the cover mage nto frequency doman coeffcents before embeddng atermarks n t. he commonly used transforms n atermarkng technques are Dscrete Wavelet ransform (DW), Dscrete Fourer ransform (DF), Dscrete Cosne ransform (DC), and Sngular Value Decomposton (SVD). DW and SVD are more popular as compared to others. But these technques do not possess the drectonal nformaton such as drectonal edges of the mage. Dscrete Contourlet ransform (C) s capable of capturng the drectonal nformaton th multresoluton representaton [6, 11]. It makes use of Laplacan Pyramd for multresoluton representaton of the mage [9, 1] folloed by a drectonal decomposton on every band pass mage usng drectonal flters [8, 1]. hs paper explores a colour mage atermarkng scheme heren Contourlet ransform has been used along th DC and SVD. he remander of ths paper s organzed as follos. Secton 2 brefs about SVD, DC and C. he next secton 3 covers the proposed colour mage DOI : /jma

2 he Internatonal Journal of Multmeda & Its Applcatons (IJMA) Vol.3, No.1, February 211 atermarkng scheme. Secton 4 demonstrates the expermental results folloed by conclusons n secton BACKGROUND he follong contans a bref descrpton of Sngular Value Decomposton, Dscrete Cosne ransform and Contourlet ransform Sngular Value Decomposton he Sngular Value Decomposton (SVD) technque has been dely used n large number of applcatons such as pattern recognton, mage compresson and atermarkng [2, 15, 16]. Suppose e have an mage represented by matrx X of sze K ros and M columns; applyng SVD on X ll produce three matrces and X can be rtten as X = USV (1) here matrces U and V are K K and M M orthogonal matrces respectvely. S s a K M dagonal matrx th sngular values (SV) on the dagonal. he columns of U and V are the egenvectors of X X and X X respectvely. he Sngular Values (SVs) n S matrx are the square roots of the egenvalues of X X or X X [17]. he man property of SVD n respect of atermarkng s that larger sngular values of an mage do not change sgnfcantly hen mage processng attacks are performed on the mage [2] Dscrete Cosne ransform he Dscrete Cosne ransform (DC) s a technque for convertng a sgnal nto elementary frequency components [2, 3]. It s appled on an mage I, havng K M pxels to transform the mage accordng to equaton 2 [2]: v m= k= ( 2 + 1) π ( 2 + 1) M ( ) K m u k vπ y u, v = αuα I ( m, k ) cos cos (2) M K 2M 2k he values y ( u, v ) are called the DC coeffcents of the mage matrx. he values of α u and α v are both set to 1 2 hen u, v =, otherse 1. he mage can be reconstructed by applyng IDC accordng to equaton 3 [2]: I 2 M 2 K M 1K 1 ( m, k) α α y( u, v) = u= v= u v cos ( 2m + 1) uπ ( 2k + 1) 2M cos vπ 2k (3) 2.3. Contourlet ransform he Contourlet ransform (C) as proposed by Do and Vetterln [6, 11]. he man feature of ths transform s ts effcency n capturng to dmensonal sngulartes lke edges hereas avelet transform capture only pont sngulartes [7, 12]. It provdes a multresoluton and drectonal expanson of mages usng Pyramdal Drectonal Flter Bank (PDFB). he PDFB s constructed by to flter-bank stages, a Lapalacan Pyramd (LP) folloed by a Drectonal Flter Bank (DFB) [12]. he former captures pont dscontnutes and latter lnks these nto lner structures. he LP decomposes the mage nto frequency band to obtan sngular ponts. he DFB decomposes each LP detal band to capture drectonalty. 123

3 he Internatonal Journal of Multmeda & Its Applcatons (IJMA) Vol.3, No.1, February PROPOSED WAERMARKING SCHEME he scheme has embeddng and extracton procedures. he steps for embeddng and extracton are as follos: 3.1. Embeddng Procedure 1. Apply 2-level C on the atermark to obtan drectonal sub bands. 2. Apply LP decomposton on fourth drectonal subband to obtan bandpass mage W. 3. Perform SVD on bandpass mage obtaned from Step2. W = U S V (4) here W s the bandpass mage of atermark. 4. ransform the cover mage from RGB colour space nto YcbCr colour space. 5. For each colour component (a) Perform DC. (b) Perform 2-level C to obtaned drectonal sub bands. (c) Perform LP decomposton on fourth drectonal subband. (d) Perform SVD on bandpass mage obtaned from Step 5-(c). CBP = UBP SBP VBP (5) here CB represent the fourth drecton subband of cover mage. P (e) Modfy the sngular values of bandpass mage. SB P = SBP + α S (6) here SB P s modfed sngular matrx, SBP s sngular matrx of drectonal subband of cover mage, S s sngular matrx of atermark and α s the amplfcaton factor. (f) Obtan the modfed band pass mage. * CBP = UBP SBP VBP (7) here * CBP s modfed bandpass of cover mage, UBP and matrces of bandpass of cover mage. (g) Apply LP reconstructon on modfed bandpass mage. (h) Perform nverse C (IC). () Perform IDC. 6. ransform the mage nto RGB colour space to get the atermarked mage Extracton Procedure VBP are orthogonal 1. ransform the cover mage and atermarked mage from RGB colour space nto YCbCr Space. 2. For each colour component of cover and atermarked mages (a) Perform DC. (b) Perform 2-level C to obtaned drectonal subbands. (c) Perform LP decomposton on fourth drectonal subband. (d) Perform SVD on bandpass mage of cover and that of atermarked mage. CBW = UBW SBW VBW (8) CBP = UBP SBP VBP 124

4 he Internatonal Journal of Multmeda & Its Applcatons (IJMA) Vol.3, No.1, February 211 here CBW andcbp are bandpass of atermarked and cover mages respectvely. (e) Extract sngular values of atermark as ext SBW SBP S = (9) α here SBW and SBP are sngular matrces of atermarked and cover mages ext respectvely. S s extracted sngular matrx of atermark. 3. Perform SVD on all bandpass mages of atermark ext ext W = U S (1) V ext here W s extracted atermark mage. 4. Apply LP reconstructon on results obtaned from Step Perform IC to get the atermarks. 4. EXPERIMENAION AND RESULS 4.1. Experment 1 and Results he expermentaton as done usng MALAB platform. As many as eght dfferent cover mages ere taken. hese mages are: Lena, Barbara, Goldhll, Redfort, Boat, aj Mahal, Statue of Lberty and Nuclear Plant, each of sze One gray scale mage named Sprnger logo of sze as taken as atermark. he C uses pkva for both Pyramd flter and Drectonal flter to obtan a to level decomposton. Amplfcaton factor has been set to.1 for embeddng atermark. he Sprnger logo s embedded nto all the cover mages. Fgure 1 shos atermarked mages and extracted atermark mage usng proposed method. (a) (b) (c) (d) (e) (f) (g) (h) () Fgure 1. Watermarked mages; (a) Lena (b) Barbara (c) Goldhll (d) Redfort (e) Boat (f) aj (g) Statue (h) Nuclear and Extracted atermark mage; () Sprnger Logo he Peak Sgnal to Nose Rato (PSNR), Structural Smlarty Index (SSIM) [13] and 125

5 he Internatonal Journal of Multmeda & Its Applcatons (IJMA) Vol.3, No.1, February 211 correlaton coeffcent values of each atermarked mage are tabulated n table 1. able 1. Qualty metrcs of atermarked mages and correlaton coeffcent usng Proposed Method. Lena Barbara Goldhll Redfort Boat aj Status Nuclear PSNR SSIM Cor. Coef able 1 shos that the proposed method yelds much better results so far as the qualty of mages s concerned. We also obtan better correlaton coeffcent usng the proposed scheme Experment 2 and Results he next experment as performed to see the effect of dfferent mage processng attacks such as Sharpenng, Gamma Correcton, Affne ransform, Hstogram Equalzaton, Reszng, Croppng, Rotaton and Gaussan blur. Fgure 2 shos the Nuclear Plant atermarked mages after applyng above sad attacks. he best extracted atermarks are also shon. Sharpenng Gamma Correcton Affne ransform Hstogram Equalzaton Reszng ( ) Croppng Rotate 3 Gaussan Blurrng Rotate 75 Fgure 2. Extracted Sprnger logo from atermarked mage nuclear Plant after varous mage Processng attacks. he correlaton coeffcents of best extracted atermark mage after dfferent mage processng attacks are n table 2. he results reveal that extracted atermark mages have good correlaton coeffcents. 126

6 he Internatonal Journal of Multmeda & Its Applcatons (IJMA) Vol.3, No.1, February 211 able 2. Correlaton coeffcents of the best extracted atermark mages under dfferent mage processng attacks. Attacks Lena Barbara Goldhll Redfort Boat aj Status Nuclear Sharpenng Gamma Correcton ransform Hstogram Equalzaton Reszng Croppng(left half) Croppng(rght half) Rotate Rotate Gaussan Blur Experment 3 and Results he next experment as performed to see the effect of addng dfferent types of noses such as Salt & Peppers and Speckle Nose on atermarked mages th varyng densty/varance thn the nterval [.1,.1]. Amplfcaton factor s set to.1 for embeddng atermark. ables 3 and 4 depct the correlaton coeffcents of best extracted atermark mage after applyng salt & peppers and speckle nose attacks respectvely. he results sho that correlaton coeffcent of extracted atermark decreases as the nose densty/varance ncreases. able 3. Correlaton coeffcents of the best extracted atermark mages under Salt & Peppers nose Densty Lena Barbara Goldhll Redfort Boat aj Statue Nuclear

7 he Internatonal Journal of Multmeda & Its Applcatons (IJMA) Vol.3, No.1, February 211 able 4. Correlaton coeffcents of the best extracted atermark mages under Speckle nose Varance Lena Barbara Goldhll Redfort Boat aj Statue Nuclear Another experment as performed to see the effect of decomposton flter on atermarked and extracted atermark mage. We use three dfferent flters pkva, 9/7 and 5/3 for expermentaton. able 5 shos the correlaton coeffcents of best extracted atermark mage for dfferent types of decomposton flters. he results reveal that correlaton coeffcents of extracted atermark n all cover mages for flter pkva s better as compared to that obtaned usng 5/3 and 9/7 flters. ables 6 and 7 demonstrate the PSNR and SSIM values of atermarked mage under dfferent flters. he result shos that the atermarked mage s less dstorted hen e use flter pkva nstead of 9/7 and 5/3 flters. able 5. Effect of decomposton flter on extracted atermark mages n terms of correlaton coeffcents. Flter Name Lena Barbara Goldhll Redfort Boat aj Status Nuclear pkva / / able 6. Effect of decomposton flter on atermarked mages n terms of PSNR. Flter Name Lena Barbara Goldhll Redfort Boat aj Status Nuclear pkva / /

8 he Internatonal Journal of Multmeda & Its Applcatons (IJMA) Vol.3, No.1, February 211 able 7. Effect of decomposton flter on atermarked mages n terms of SSIM. Flter Name Lena Barbara Goldhll Redfort Boat aj Status Nuclear pkva / / he next experment as performed to sho the effect of level of decomposton flter used on atermark mage. Fgure 3 shos the effect of level of decomposton on drectonal flter bank (DFB). he results reveal that thrd-level decomposton gves good qualty of extracted atermark mage. 2 Level 3 Level 4 Level 5 Level Correlaton Coeffcent Lena Barbara Goldhll Redfort Boat aj Statue Nuclear Fgure 3. Effect of level of decomposton on DFB Further smulaton as done to see the effect of scalng factor on atermarked mage. ables 8 and 9 gve SSIM and PSNR for dfferent values of scalng factor on atermarked mage respectvely. A careful look at the tables 8 and 9 reveal that the ncrease n value of scalng factor results n decrease n values of both PSNR and SSIM. he best values are obtaned hen the scalng factor s n the range.3.1. able 8. Structural Smlarty metrc of the atermarked mages under dfferent scalng factor Scalng Factor Nuclear aj Statue Goldhll Boat

9 he Internatonal Journal of Multmeda & Its Applcatons (IJMA) Vol.3, No.1, February 211 able 9. PSNR of the atermarked mages under dfferent scalng factor Scalng Factor Nuclear aj Statue Goldhll Boat CONCLUSIONS hs paper has presented a colour mage atermarkng scheme for embeddng a gray scale atermark nto colour cover mage usng C together th DC and SVD. he atermarked mages ere subjected to common mage processng attacks. he smulaton results depct the effcency of the scheme n terms of qualty of atermarked mage and the robustness aganst the attacks. he extracted atermarks also have better correlaton coeffcents. REFERENCES [1] M., Arnolad, M., Schumucker, & A., Wolthusen, (23) echnques and Applcatons of Dgtal Watermarkng and Content Protecton, Artech House, USA. [2] Al Al Haj, (27) Combned DW-DC Dgtal Image Watermarkng, J. Computer Scence, Vol. 3, pp [3] K., Rao, & P.,Yp, (199) Dscrete Cosne ransform: algorthms, advantages, applcatons, Academc Press, USA. [4] V., Potdar, S., Han, & E., Chang, (25) A Survey of Dgtal Image Watermarkng echnques, IEEE Internatonal Conference on Industral Informatcs, pp [5] A., Pva, M., Barn, & F., Bartoln, (1998) Copyrght protecton of dgtal mages by means of frequency doman atermarkng, SPIE Conference on Mathematcs of Data/Image Codng, Compresson and Encrypton, pp [6] M., N., Do, & M., Vtterl, (25) he Contourlet ransform: An effcent drectonal multresoluton mage representaton, IEEE ranscatons on Image Processng, Vol. 14, No. 12, pp [7] B., Chandra Mohan, & S., Srnvas Kumar, (28) Robust Dgtal Watermarkng Scheme usng Contourlet ransform, J. Computer Scence and Netork Securty, Vol. 4, No. 2, pp [8] R., H., Bamerger, & M., J.,., B., Smth, (1992) A flter bank for the drectonal decomposton of mages: theory and desgn, IEEE ranscatons on Sgnal Processng, Vol. 4, pp [9] P., J., Burt, & E., H., Adelson, (1983) he Laplacan pyramd as a compact mage codes, IEEE ranscatons on Communcatons, Vol. 31, pp [1] M., Jaylakshm, S., N., Merchant, & U., B, Desa, (26) Blnd Watermarkng n Contourlet Doman th Improved Detecton, Internatonal Conference on Intellgent Informaton Hdng and Multmeda Sgnal Processng,. [11] M., N., Do, & M., Vtterl, (26) Contourlets: a drectonal multresoluatonal mage representaton, Internatonal Conference on Image Processng, pp [12] S., Ghannam, & F., E., Z., Abou-Chad, (29) Enhancng Performance of Image Watermarks Usng Contourlet ransform, Natonal Rado Scence Conference, pp

10 he Internatonal Journal of Multmeda & Its Applcatons (IJMA) Vol.3, No.1, February 211 [13] Z., Wang, A., C., Bovk, H., R., Shekh, & E., P. Smoncell, (24) Image Qualty assessment: From error measurement to structural, IEEE ranscatons on Image Processng, Vol. 13, No. 1, pp [14] Sanjay, Raat, & Balasubramanan Raman, (21) A Ne Robust Watermarkng Scheme for Color Images, IEEE Internatonal Advance Computng Conference, pp [15] Z., Q., Hong, (1991) Algebrac feature extracton of mage for recognton, J. Pattern Recognton, Vol. 24, pp [16] A., Shnayderman, A., Gusev & A., Eskcoglu, (24) A Multdmensonal Image Qualty Measure Usng Sngular Value Decomposton, SPIE Image Qualty and System Performance, pp [17] A. Basso, F., Bergadano, D., Cavagnno, V., Pomponu & A., Vernone, (29) A Novel Blockbased Watermarkng Scheme Usng the SVD ransform, J. Algorthms, Vol.2, pp