Hybrd Non-Blnd Color Image Watermarkng Ms C.N.Sujatha 1, Dr. P. Satyanarayana 2 1 Assocate Professor, Dept. of ECE, SNIST, Yamnampet, Ghatkesar Hyderabad-501301, Telangana 2 Professor, Dept. of ECE, AITS, Trupat-, Andhrapradesh ABSTRACT Dgtal multmeda content has progressvely becomng an mportant ssue because mage watermarkng s dentfed as a foremost technology used n copyrght protecton. The man requrement for mage watermarkng s ts mperceptblty and robustness. To acheve these requrements, a new approach for mage watermarkng based on sngular value decomposton (SVD) and dscrete wavelet transform (DWT) s proposed n ths paper. Modfcaton of the approprate sub-bands leads to a watermarkng scheme whch postvely preserves the qualty. The addtonal advantage of the proposed technque s securty, mperceptablty and ts robustness aganst the most of common attacks. In ths method, the sngular values of DWT coeffcents of B-plane are modfed usng dfferent scalng factors to nsert the sngular values of watermark. The sngular values of R, G, B planes of watermark are embedded nto the three sub- bands LH, HL, HH of B-plane n host mage. The proposed watermarkng s realzed by usng MATLAB and evaluated by usng performance metrcs MSE, PSNR. Keywords:- Watermark, DWT, SVD, PSNR and MSE 1. INTRODUCTION Dgtal data utlzaton along wth the ncreased popularty of the nternet has facltated nformaton sharng and dstrbuton. Watermark, whch s usually some nformaton related to orgnal data or the owner, s embedded n the orgnal data; then the watermarked data s dstrbuted throughout computer networks. Consderng the applcatons of such systems, the watermark can be extracted from the meda. Watermarkng s the process of embeddng a watermark nto the host mage, vdeo or audo. To get more effcent output the watermark should be perceptually nvsble & dffcult to remove wthout affectng the mage qualty and should robustly resst to mage dstortons caused by attacks. The watermarkng technques can be classfed nto Spatal doman technques & Frequency doman technques [1]- [2]. Spatal doman technque embeds the data by drectly modfyng the pxel values of the orgnal mage [3]-[4]. In Frequency doman technque, data wll be embedded by modulatng the coeffcents of preferred transform. The most commonly used transforms are SVD, DWT, DCT & DFT. The transform doman technques always gve most robust output [5]. In few schemes, both watermark and host mages are preprocessed n transformed doman to acheve hgh rgdty. In the lterature, many schemes uses the SVD-DWT based embeddng for gray scale mage watermarkng. The proposed scheme embeds the color watermark nto color cover mage. The color mage s represented by Red (R), Green (G) and Blue (B) channels. Out of these three channels, change n the ntensty of R channel s the most senstve to human eyes whereas for B channel t s least senstve [6]-[7]. Hence, n the proposed scheme the blue channel s consdered for embeddng. The wavelet transform of mage gves four frequency sub-band coeffcents. In mage processng each subband s resstant to dfferent types of attacks or transformatons. For example, the low frequency subband coeffcents are less robust to geometrcal dstortons and hstogram equalzaton. In the proposed scheme the copy of the watermark s embedded nto three subband coeffcents whch s hard to destroy the watermark even after the dfferent types of attacks on the watermarked mages. To mprove the robustness of the scheme the watermark s embedded nto dfferent sub-band coeffcents obtaned from B channel of the color mage. 2. DWT AND SVD TRANSFORMATIONS In ths secton we dscuss n bref the Dscrete Wavelet Transform and Sngular Value Decomposton of mages [8]- [9]. 1. Dscrete Wavelet Transform As one of the most glorous achevement n 20 th century, DWT has become the most useful tool for mage compresson, processng and analyss. The basc dea s to mult-frequency decompose the mage nto subbands at dfferent frequency and dfferent space, then to process the coeffcents of the subbands [10]. A DWT decomposes Image nto four subbands: low frequency band (LL), horzontal detal band (HL), vertcal detal band (LH), and dagonal detal band (HH). The mage s energy s manly focused on the low frequency band. The other three bands characterze the margnal nformaton of the correspondng drecton and have lttle energy. Volume 3, Issue 11, November 2014 Page 381
2. Sngular Value Decomposton If a m x n mage s represented as a real matrx A, t can be decomposed as: A = U S V T It s called a sngular value decomposton of A. Where U s a m x m untary matrx, S s a m x n matrx wth nonnegatve numbers on the dagonal and zeros on the off dagonal, and VT denotes the conjugate transpose of V, an n x n untary matrx [6]. The nonnegatve components of S represent the lumnance value of the mage. Changng them slghtly does not affect the mage qualty and they also don t change much after attacks, watermarkng algorthms make use of these two propertes [7]. 3. PROPOSED SCHEME The proposed scheme s based on the dea of replacng the frequency coeffcents of LH, HL, and HH wth sngular values of R, G, B planes of watermark mage [10]. The watermarkng procedures can be descrbed as follows: Watermark Embeddng: The proposed method uses the color mage I of sze m x n as the host mage and the color/ monochrome mage W of sze m/2 x n/2 as the watermark. The color mage s transformed nto R, G and B planes of sze m x n. Human eyes are less senstve to change n the ntensty of the B plane. So the B plane s chosen for watermark embeddng. Then twolevel DWT s appled on the B plane to generate subband coeffcents LL, LH, HL, HH of sze m/2 x n/2. The SVD decomposton s appled on detal subband coeffcents and R, G, B planes of watermark. The sngular values of three planes of watermark are added to the subbands ( LH, HL, HH) of the DWT transformed B plane usng scalng factor a. l a (1) Now the nverse DWT s appled on the modfed subband coeffcents of B plane to acheve the embedded B plane. The embedded B plane s combned wth R and G planes of host mage to acheve watermarked color mage. w Fgure 1: Flowchart for Watermark Embeddng Watermark Extracton: As non-blnd watermarkng technque uses the host mage and watermark mage to extract the Watermark, ths extracton process uses the host mage I, watermarked color mage I each of sze m x n and the watermark mage W of sze m/2 x n/2. The orgnal and the watermarked color mages are transformed nto R, G and B planes. On the B plane of the mage I, DWT s appled to create subband coeffcents LL, LH, HL, HH of sze m/2 x n/2. The SVD decomposton s appled to LH, HL, HH subbands. The sngular values of R, G, B planes of watermark are extracted Volume 3, Issue 11, November 2014 Page 382
from the (LH, HL, HH ) of the DWT transformed B plane of color mage and watermarked color mage usng scalng factor a. w l (2) a The extracted sngular values of watermark are combned wth other matrces of watermark to generate the watermark mage. In ths paper, we are usng Mean square error (MSE) and Peak Sgnal to Nose Rato (PSNR) to estmate the watermark mperceptblty [10]. 4. EXPERIMENTAL RESULTS Fgure 2: Flowchart for Watermark Extracton The proposed watermark scheme s tested on dfferent types of mages for varous scalng factors. In our experments 256x256 color mages namely Lena, Peppers, Sunset, Balloon, autumn were used as shown n fgures 3 7. 128x128 colour SNIST logo s used as a sample watermark. MATLAB s used for the mplementaton of the proposed algorthm. The followng fgures show nput (host) mage, extracted B plane, DWT doman of B plane, logo (watermark) mage, watermarked B plane, watermarked mage and extracted logo mage for scalng factor of 0.01. Fgure 3 Results of peppers mage Fgure 4 Results of Lena mage Volume 3, Issue 11, November 2014 Page 383
Fgure 5 Results of sunset mage Fgure 6 Results of Balloon mage Fgure 7 Results of autumn mage To evaluate the performance of the proposed method, PSNR and MSE are used. From the observed results, PSNR and MSE are evaluated between host and watermarked mages at varous scalng factors as shown n Table 1,where we observe that the PSNR value decreases as scalng factor ncreases. Increased scalng factor gves large error between host and watermarked mages. Table 1: Performance results n terms of PSNR AND MSE Scalng Images MSE PSNR Parameter 0.01 peppers 0.1097 57.7293 Lena 0.1589 56.1184 sunset 0.1500 56.3707 balloon 0.1586 56.1282 autumn 0.0909 58.5457 0.03 peppers 0.5238 50.9390 Lena 0.9475 48.3652 sunset 0.9204 48.4911 balloon 1.0393 47.9633 autumn 0.4471 51.6270 0.05 peppers 0.9221 48.4831 Lena 1.8017 45.5740 sunset 1.7653 45.6625 balloon 2.1232 44.8608 autumn 0.7736 49.2458 0.07 peppers 1.2989 46.9951 Volume 3, Issue 11, November 2014 Page 384
Lena 2.5606 44.0473 sunset 2.5939 43.9913 balloon 3.1734 43.1155 autumn 1.0429 47.9482 0.10 peppers 1.8155 45.5408 Lena 3.5505 42.6279 sunset 3.7821 42.3535 balloon 4.5513 41.5494 autumn 1.3688 46.7673 0.50 peppers 5.0941 41.0601 Lena 8.9300 38.6223 sunset 10.6495 37.8575 balloon 10.9652 37.7306 autumn 3.2872 42.962 1.00 peppers 6.2729 40.1561 Lena 10.8436 37.7791 sunset 12.3519 37.2135 balloon 12.4881 37.1658 autumn 4.0283 42.0796 5. CONCLUSION In ths paper, A Hybrd technque has been ntroduced for colour mage watermarkng based on DWT and SVD. The sngular values of watermark are added to the sngular values of detaled coeffcents of DWT doman of the orgnal mage. In smulaton, we go through the dfferent scalng factors to embed the watermark n dfferent mages of sze 256x256. Here we observe that PSNR decreases and MSE ncreases wth ncreased scalng factors. It has been observed that PSNR and MSE values of our proposed algorthm are better than those of SVD algorthm. The characterstcs of SVD lead to the best performance of ths algorthm n both securty and robustness. Further research can be carred out wth the sngular values of watermark to be embedded n sngular values of DCT coeffcents of DWT doman. References [1] Potdar VM, Han S, Chang E, A Survey of dgtal mage watermarkng technques, Proceedngs of IEEE nternatonal Conference on ndustral nformatcs, 2005, pp. 709-716 [2] Chandra D.V.S.; Dgtal mage watermarkng usng sngular value decomposton, Crcuts and Systems 2002. MWSCAS-2002, vol.3, 4-7Aug 2002, pp. 264-267 [3] N. Nkolads and I. Ptas, Robust mage watermarkng n the spatal doman, Sgnal Processng, Vol.66, No.3,pp.385403, 1998. [4] R. Lu, T. Tan, An SVD based watermarkng scheme for protectng rghtful ownershp, IEEE Transacton on Multmeda Volume 4, ssue 1, March 2002 pp121-128. [5] S. Mallat, The theory for multres\oluton sngnal decomposton: the wavelet represntaton, IEEE Trans. Pattern Anal. Mach. Intell., vol. 11, no. 7, pp.654-693, Jul.1989. [6] Q. L, C. Yuan and Y.Z. Zhong, Adaptve DWT SVD doman mage watermarkng usng human vsual model, n Proc. 9 th Inst. Conf. Adv. Commun. Technol., Gangwon-Do, South Korea, 2007, pp. 1947-1951. [7] X.P. Zhang, K.L comments on An SVD-Based Watermarkng scheme for Protectng Rghtful Ownershp, IEEE Transacton on multmeda Vol.7,no.2,2005, pp.593-594. [8] Chn La and Cheng-Chh Tsa, Dgtal Image Watermarkng Usng Dscrete Wavelet Transform and Sngular Value Decomposton, IEEE Transactons n Instrumentaton and Measurement., Vol.59, no. 11, pp.3060-3063, Nov. 2010. [9] G. Bhatnagar and B. Raman, A new robust reference watermarkng scheme based on DWT-SVD, Comput. Standards Interfaces, Vol.31, no. 5,pp.1002-1013,Sep.2009. [10] E.Ganc and A.M.Eskcoglu, Robust DWT-SVD doman mage watermarkng: Embeddng data n all frequences,n Proc. Workshop Multmeda Securty, Magdeburg, Germany, 2004, pp. 166-174. Volume 3, Issue 11, November 2014 Page 385