Journal of Information Security, 2011, 2, 169184 doi:10.4236/jis.2011.24017 Published Online October 2011 (http://www.SciRP.org/journal/jis) Copyright © 2011 SciRes. JIS 169 On Secure Digital Image Watermarking Techniques Manjit Thapa1, Sandeep Kumar Sood2* 1Department of Computer Science, Sri Sai College of Engineering & Technonogy, Badhani, Pa thankot, India 2Department of Computer Science and Engineering, G.N.D.U.R.C., Gurdaspur, Punjab, India Email: manjit.thapa@ya hoo.co.in, *san1198@gmail.com Received July 29, 2011; revised September 9, 2011; accepted September 20, 2011 Abstract Digital watermarking is used to hide the information inside a signal, which can not be easily extracted by the third party. Its widely used application is copyright protection of digital information. It is different from the encryption in the sense that it allows the user to access, view and interpret the signal but protect the owner ship of the content. One of the current research areas is to protect digital watermark inside the information so that ownership of the information cannot be claimed by third party. With a lot of information available on various search engines, to protect the ownership of information is a crucial area of research. In latest years, several digital watermarking techniques are presented based on discrete cosine transform (DCT), discrete wavelets transform (DWT) and discrete fourier transforms (DFT). In this paper, we propose an algorithm for digital image watermarking technique based on singular value decomposition; both of the L and U compo nents are explored for watermarking algorithm. This technique refers to the watermark embedding algorithm and watermark extracting algorithm. The experimental results prove that the quality of the watermarked im age is excellent and there is strong resistant against many geometrical attacks. Keywords: Digital Image Watermarking, Singular Value Decomposition, Watermark Embedding Algorithm, Watermark Extracting Algorithm, Ratio Analysis, Security Analysis 1. Introduction Digital watermarking is a technique that embeds data called watermark into a multimedia object so that wa termark can be detected to make an assertion about the objects. It can be categorized as visible or invisible. Ex ample of visible watermarking is the logo visible super imposed on the corner of television channel in a televi sion picture. On the other hand, invisible watermark is hidden in the object, which can be detected by an au thorized person. Such watermarks are used for suit the author authentication and detecting unauthorized copying. The novel technology of digital watermarking has been sponsored by many consultants as the best method for such multimedia copyright protection problem [1,2]. Digital watermarking is having a variety of useful appli cations such as digital cameras, medical imaging, image databases, video on demand systems, and many others. In recent years, many digital image watermarking tech niques have been proposed in the literature which is based on spatial domain technique and frequency domain technique. These techniques are used in watermark em bedding algorithm and watermark extracting algorithm. In 2002, Ali [3] proposed an approach based on DWT and DCT to improve the performance of the DWTbased watermarking algorithms. In this method, watermarking is done by embedding the watermark in first and second level of DWT subbands of the host image, followed by the application of DCT on the selected DWT subbands. The combination of these two transforms improved the watermarking performance considerably in comparison with only watermarking approaches. They showed that the quality of watermark image is very good. In 2005, Chen [4] proposed a singular value decomposition scheme based on components of D and U without using DWT, DCT and DFT transforms. They showed that quality of watermarked image is good on their schemes. In 2007, Seed [5] introduced a novel digital watermark ing method based on single key image for extracting dif ferent watermarks. In this method, they used Arnold transform technique in watermark embedding and ex traction, which is based on DWT and DCT algorithm. With the popularity of internet and availability of large storage devices, storing and transferring an image is
M. THAPA ET AL. 170 i simple and feasible. They showed that robustness of the algorithm against many signal processing operations. In 2010, Lamma and Ali [6] suggested two blind, imper ceptible and robust video watermarking algorithms that are based on singular value decomposition. Each algo rithm integrates the watermark in the transform domain. They used the components of matrices such as U and V. Their schemes are shown to provide very good perform ance in watermarked video as compared to Chan [4]. Most of the domain transformation watermarking tech niques works with DCT and DWT. However singular value decomposition (SVD) is one of the most powerful numeric analysis techniques and used in various re quirements. These requirements can be organized and described as follows [710]. Undeletable: An embedded watermark is difficult to detect and cannot be removed by an illegal person. Also the algorithm must resist different attacks. Perceptually visible: The original images and water marked images cannot be distinguished by the human eye. This means that there is not enough alteration of a watermarked image to prevent motivation to an illegal person. Unambiguous: An embedded watermark selected from a watermarked image that must be clear enough for ownership to be determined. In this way, the extracted watermark cannot be distorted to such an extent that the original watermark cannot be recognized. In this paper, we will describe a digital image water marking algorithm based on singular value decomposi tion technique. This paper is organized as follows. In Section 2, we introduce the SVD transformation and SVD based watermarking techniques briefly. In Section 3, we propose the embedding and extracting algorithm. In Section 4, we evaluate the performance of watermark image. In section 5, we show the experimental results and Section 6 conclude the paper. 2. A Review of Related Work Singular value decomposition (SVD) is a mathematical technique based on linear algebra and used by factoriza tion of a real matrix or complex matrix, with many useful applications in signal processing and statistics. 2.1. Singular Value Decomposition (SVD) Singular value decomposition is one of a number of valuable numerical analysis tools which is used to ana lyze matrices. It can be appeared from three jointly compatible points of view. On the other hand, we can see it as a method for transforming correlated variables into a set of uncorrelated ones that better expose the various relationships among the original data items. At the same time, SVD is a method for identifying and ordering the dimensions along which data points demonstrate the most variation. This attach the third way of viewing sin gular value decomposition, which accepted the most variation, it’s possible to find the best approximation of the original data points using less dimensions. Hence, SVD can be seen as a method for data reduction. In SVD transformation, a matrix can be decayed into three ma trices that are having the same size as the original matrix. It is useful to establish a contrast with Gaussian elimina tion and its equation. Given A is a n × n square matrix, this matrix can be decomposed into three components, L, D and U, respectively such that [L D U] = SVD (A), A’ = LDUT, L–1 where A = LDU. 1,1 1,21,1,11,21, 2,1 2,22,2,12,22, 3,1 3,23,3,13,23, 1,1 1,21, T 2,1 2,22, 1 3,1 3,23, = nn nn nn nn nii i n ll l lll lll uuu uuu lu uuu (1) where the L and U components are real unitary matrices or complex matrices with small singular values, and the D component is an n × n diagonal matrix with larger singular value or eigen vector values entries which spec ify σ1,1 >> σ2,2 >> … σk,k,k+1 = … σn,n = 0. ∑ are non zero matrix by diagonals of A. SVD is nonlinear because the orthogonal matrices L and U depend on A and shown in Equation (1). A’ is the reconstructed matrix after the in verse SVD transformation. Reduced singular value de composition is the mathematical technique underlying a type of document retrieval and word semblance method. These are also known as Latent Semantic Indexing or Latent Semantic Analysis. In this way, the three compo nents of matrices L, D, and U specify Aui = σili and µiTA = σiuiT. SVD Example A matrix is said to be square if it has the same number of rows as columns. To designate the size of a square matrix with n rows and columns, it is called nsquare matrix. For example, the matrix below is 3square. As an example to simplify SVD transformation, suppose A = 1021 15 30 923 185329 If SVD operation is useful on this matrix, then the ma trix A will be decomposed into equivalent three matrices as follows: Copyright © 2011 SciRes. JIS
M. THAPA ET AL. 171 L = , D = , U = 0.4019 0.12020.9079 0.47490.8749 0.9417 0.7830 0.46900.4083 68.5399 00 024.2485 0 00 0.5342 0.44920.7233 0.5242 0.6995 0.64980.2972 0.5557 0.2332 0.7979 Here diagonal elements of matrix D are singular val ues and we observe that these values satisfy the non in creasing order: 68.5399 ≥ 24.2485 ≥ 0.5342. Digital image watermarking techniques has several advantages that used singular value decomposition. Firstly, SVD transformation from the size of memory is not fixed and can be represented by a rectangle or square matrices. Secondly, SVD increase accuracy and decrease the memory requirement. Thirdly, digital images in sin gular values are less affected if general image watermark is executed. Fourth, singular value decomposition in clude by algebraic properties. 2.2. SVDTransformation We will describe a digital image watermarking technique which is based on singular value decomposition trans form, such as DWT and DCT. Watermarking is estab lished by the wavelet coefficient of selected sub bands and followed by the requirements of DCT transform on the selected subbands [1114]. In this section, we will introduce the transformation of digital watermarking technique. The DCT Transform: The discrete cosine transform is a transformation technique based on digital water marking algorithm and spatial domain technique. The discrete cosine transform is derived from discrete Fourier transforms and represents data in terms of frequency space rather than an amplitude space. The spatial domain technique can be transformed into the frequency domain, and the frequency domain technique can be transformed back to the spatial domain by using inverse discrete co sine transform. The discrete cosine transform (DCT) is a technique for converting a signal into effortless fre quency components. It represents an image as a sum of sinusoids of varying magnitudes and frequencies. With an input image, k and the DCT coefficients for the trans formed output image, L is computed according to Equa tion (2) as shown below. In the equation, k, is the input image having N × M pixels, k(m, n) is the intensity of the pixel in row m and column n of the image and L(u, v) is the DCT coefficient in row u and column v of the DCT matrix. The DCT formulas are as follows. The general equation for a one dimension (N data items) DCT is defined by the following equation: M1 0 2cos(21)π L( M2M x mu um k )m (2) and the corresponding inverse 1D DCT transform is sim ple L–1(u) where 10 =2 11,2, otherwi u m u se The m function computes the twodimensional discrete cosine transform (DCT) of an image. The DCT has the property that, for a typical image, most of the visually significant information about the image is con centrated in just a few coefficients of the DCT. For this reason, the DCT is often used in image compression ap plications. The general equation for a 2D (N by M image) DCT is defined by the following Equation (3): M1N1 00 22 L, MN cos(2 1)πcos(2 1)π (,) 2M 2N xy uvm n mu nu kmn (3) and the corresponding inverse 2D DCT transform is sim ple L–1(u, v), i.e.: where 10 =2 11,2,,M u m u 1 10 =2 11,2,,N v n v 1 The image is reconstructed by applying inverse DCT operation according to Equation (4): M1N1 00 22 K(,)MN cos(2 1)πcos(2 1)π , 22 uv mnm n mu nu luv MN (4) Examples for Transformation: The grayscale Lena image of 256 × 256 pixels, with 4bit representation for each pixel is used as the test input. The input image is divided into 4by4 or 8by8 blocks, and the twodi mensional DCT is computed for each block. The DCT coefficients are then quantized, coded, and transmitted. Copyright © 2011 SciRes. JIS
M. THAPA ET AL. Copyright © 2011 SciRes. JIS 172 changing component of the signal. In most cases, high pass component is not so rich with data offering good property for compression. In some cases, such as audio or video signal, it is possible to contend with some of the samples of the high pass component without noticing any significant changes in signal. Filters from the filter bank are called wavelets and as shown in Figure 2. The JPEG receiver (or JPEG file reader) decodes the quantized DCT coefficients, computes the inverse two dimensional DCT of each block, and then puts the blocks back together into a single image. For typical images, many of the DCT coefficients have values close to zero; these coefficients can be discarded without seriously affecting the quality of the reconstructed image as shown in Figure 1. The transform matrix computation method is used. The test image was compressed to different scales, from one to three and the compression ratio as well as the mean square root error of the reconstructed image were calculated for minimal error case and quan tised case. For 2D images, applying DWT corresponds to proc essing the image by 2D filters in each dimension. The filters divide the input image into four nonoverlapping multiresolution subbands LL1, LH1, HL1 and HH1. The subband LL1 represents the coarsescale DWT coeffi cients while the subbands LH1, HL1 and HH1 represent the finescale of DWT coefficients. To obtain the next coarser scale of wavelet coefficients, the subband LL1 is further processed until some final scale N is reached. When N is reached we will have 3N + 1 subbands con sisting of the multiresolution subbands LLy and LHy, HLy and HHy where y ranges from 1 until N. It has the following steps in digital image watermarking transfor mation such as The popular blockbased DCT transform segments is an image nonoverlapping blocks and applies DCT to each block. This result in giving three frequency coeffi cient sets: low frequency subband, mid frequency sub band, and high frequency sub band. The digital water marking based on two facts. The first fact is that much of the signal energy lies at low frequency subband which includes the most important visual parts of the image. The second fact is that high frequency components of the image are generally detached through compression and noise attacks. The watermark is surrounded by modify ing the coefficients of the middle frequency subband so that the visibility of the image will not be overstated and the watermark will not be removed by compression. Step 1: Present DWT on the original image to de compose it into four nonoverlapping multiresolution coefficient sets, such as LL1, HL1, LH1 and HH1. Step 2: Present DWT again on two HL1 and LH1 sub bands to get eight smaller subbands and prefer four co efficient sets: HL12, LH12, HL22 and LH22 as shown in Figure 1. Discrete Wavelet Transform: The transformation pro duct is a set of coefficients organized in the way that enables not only spectrum analyses of the signal, but also spectral behavior of the signal in time. This is achieved by decomposing signal, breaking it into two components, each concerned information about source signal. Filters from the filter bank used for decomposition come in pairs: low pass and high pass. Low pass filtered signal contains information about slow changing component of the signal, looking very similar to the original signal, only two times shorter in term of number of samples. High pass filtered signal be full of information about fast Figure 1. Reconstructed image, compression image and quan tization image by using DCT. Figure 2. Sketch map of DWT and DCT decomposed subbands.
173 M. THAPA ET AL. Step 3: Present DWT again on four subbands, such as HL12, LH12, HL22 and LH22 to get sixteen smaller coeffi cient sets and prefer four coefficient sets, such as HL13, LH13, HL13 and LH13 as shown in Figure 1. Step 4: Divide four coefficient sets such as HL13, LH13, HL23 and LH23 into 4 × 4 blocks. Step 5: Present DCT to each block in the chosen sub bands (HL13, LH13, HL23 and LH23). These coefficient sets are chosen to inquire both of imperceptibility and strength of algorithm equally. The DWT is very suitable to identify the areas in the original image where a watermark can be embedded ef fectively. This property allows the utilization of the masking effect of the human visual system such that if a DWT coefficient is modified, only the region corre sponding to that coefficient will be modified. In general most of the image energy is concentrated at the lower frequency subbands LLx and therefore embedding wa termarks in these subbands may humiliate the image appreciably. Embedding in the low frequency subbands, however, could increase robustness appreciably. On the other hand, the high frequency subbands HHx include the edges and textures of the image and the human eye is not generally sensitive to changes in such subbands. This allows the watermark to be embedded without being superficial by the human eye. The compromise accepted by many DWTbased watermarking algorithm, is to em bed the watermark in the middle frequency subbands LHy and HLy where good enough performance of im perceptibility and robustness could be achieved. 3. Proposed Watermarking Techniques We proposed a singular value decomposition technique and quantization based watermarking technique. The watermarking techniques can be represented by three components, L, D and U. It relies on row and column operations. Row operations involve premultiplying ma trix and column operations involve postmultiplying ma trix. The D component can be explored with a diagonal matrix. These techniques depend upon the watermark embedding algorithm and watermark extracting algo rithm. 3.1. Watermark Embedding Algorithm The digital image watermarking algorithm can be fol lowed by singular value decomposition techniques, which involve the characteristics of the D and U components. In the embedding algorithm, the largest coefficients in D component were customized and used to embed a wa termark. The adaptation was determined by the quantiza tion method. We will start the algorithm by applying SVD transformation on original image and to reconstruct the watermarked image. Because the largest coefficients in the D component can oppose with general image processing, the embedded watermark was not really af fected. In this way, the quality of the watermarked image can be decomposed by quantization method. In our in spection, two important features of the D and U compo nents are found. In the first feature, the number of non zero coefficients in the D component could be used to determine the complexity of a matrix. Commonly, the greater number of the nonzero coefficient can be speci fied by greater complexity. In the second feature, the relationship between the coefficients in the first column of the L component could be sealed, when usually image processing was presented as shown in Figure 3. The watermarks embedding algorithm can be described as follows. Step 1: Read the original host image. Step 2: Partition the image into blocks of n × n pixels. Examples: Perform combination of two filters as pre filtering operation. The first filter is 3 × 3 sharpening filter which is defined as Equation (5). 010 13 1 010 (5) This filter enhances contrast of watermarked image. The second filter is designed by Laplacian of Gaussian filter and defined by general equation as 6. 2 2 2 2 1 G 2π e x (6) The Gaussian blur is types of imageblurring filter that uses a Gaussian function (which also expresses the normal distribution in statistics) for calculating the transformation to apply to each pixel in the image as shown by Equation (7). 22 2 2 2 1 G, e 2π y xy (7) where x is the distance from the origin in the horizontal axis, y is the distance from the origin in the vertical axis, and σ is the standard deviation of the Gaussian distribu tion. The default value for them in g = 4 and = 0.5. Performing these two filters on watermarked image could caused details of image become more visible, its means that watermark information which is different from image background become recognizable uncompli catedly. Step 3: Perform singular value decomposition (SVD) transformation. Copyright © 2011 SciRes. JIS
M. THAPA ET AL. 174 Step 4: Extract the greater coefficient D(1, 1) from each D component and quantize by using a predefined quantization coefficients A. Suppose that S = D(1, 1) mod A. Step 5: Perform embed the two pseudorandom se quences PN0, PN1, that is applied to the midband coef ficients. If A is the matrix of the mid band coefficients of SVD transformed block, then embedding is done as fol lows: If the watermark bit is 0 then, D′(1, 1) = D(1, 1) + K/4 – A, so that [A < 3K/4] Otherwise, if the watermark bit is 1 then, D′(1, 1) = D(1, 1) –K/4 + A, so that [A < K/4] Step 6: Perform the inverse of singular value decom position transformation to reform the watermarked im age. 3.2. Watermark Extracting Algorithm The watermark extracting algorithm is similar to the wa termark embedding algorithm. Extraction algorithm is the same as embedding and prefiltering is used before applying SVD transform to superior split watermark in formation from original image. The watermark extraction algorithm is performed as described by the following steps. The first three steps of the watermark extracting algorithms are same as the watermark embedding algo rithm except that the original image is replaced with the watermarked image. Previously, an embedded block is detected according to the feature of the D component and PRNG, the relationship of the U component coefficients is observed. If a positive relationship is detected, the ex tracted watermark has assigned a bit value of 1. Other wise, the extracted watermark has assigned a bit value of 0. These extracted bit values convert the original image SVD from the extracted watermark. The extracted wa termark can be specified by original watermarked image and as shown in Figure 4. Step 1: Read the watermarked image. Step 2: watermarked it into blocks of n × n pixels. Step 3: Perform the SVD transformation. Step 4: Extract the greater coefficients D''(1, 1) from each D component and quantize by using a predefined quantization coefficients A. Suppose that S = D'(1, 1) mod A. Step 5: Regenerate the two pseudo random sequences number using the same key, which is used in the water mark embedding algorithm. Step 6: For an extraction watermark bit valued of zero, if A < K/2. On the other hand, the extraction watermark bit value of one, if A > K/2. Step 7: The watermark is restructured using the ex Figure 3. Flow chart for watermark embedding algorithm. Figure 4. Flow chart for watermark embedding algorithm. tracted watermark bits, and compute the similarity be tween the original watermark and extracted watermarks. 4. Performance Evaluation We evaluated the performance of the SVD image wa termarking algorithms. The performance of the water marking methods can be measured by imperceptibility and robust capabilities. Imperceptibility means that the superficial quality of the original image should not be distorted by the presence of watermark image and as shown by Equation (8). On the other hand, the robustness is a measure of the intentionally attacks and unintention Copyright © 2011 SciRes. JIS
M. THAPA ET AL. Copyright © 2011 SciRes. JIS 175 ally attacks. It was found that the image quality meas ured by peak signal to noise ratio among the water marked images was larger than 42 db [15]. This peak signal to noise ratio is defined as 1 10 2 1 11 255 255 PSNR( ,)10log1 mn ij oo oo xy (8) of 4 × 4 pixels. Each block can be transformed into L, D, and U components by singular value decomposition. And then, a set of blocks with the same size as the watermark was selected, according to the feature of the D compo nent. For an embedding watermark block, the relation ship between the L component coefficients can be ex amined and the coefficients were modified, according to the watermark to be embedded. In our experiment, the original image and watermarked image quality is shown in Figure 5. The PSNR is employed to evaluate the difference be tween an original image o and watermarked image o1. For the robust capability, mean absolute error (MSE) measures the difference between an original watermark W and corresponding extracted watermark W1 as shown by Equation (9). We claimed the embedding algorithm and extracting algorithm to identify the ownership of the original wa termarked image as shown in Tables 1 and 2. We can see that the performance of our algorithm against the different attacks. Further, the proposed watermarking algorithm can be used for protecting the copyright of digital images. It can be observed from Tables 1 and 2 that the future method provides excellent results in the geometrical attacks. 01 01 0 MSE( , )() d i ww ww w (9) Generally, if PSNR value is larger than 40 db the wa termarked image is within acceptable degradation levels, i.e. the watermarked is almost invisible to human visual system. A lower mean absolute error reveals that the extracted watermark w0 resembles the w1 more closely. The strength of digital watermarking method is accessed from the watermarked image o1, which is further de graded by attacks and the digital watermarking perform ance of proposed method is compared with that of Chen [4]. If a method has a lower MSE(w0, w1), it is more ro bust. Simulation results suggest that this digital watermark ing algorithm is robust against many common different types of attacks such as cropping attacks, pyramid at (a) (b) (c) 5. Experimental Results Figure 5. Three original images of 256 × 256 pixels (a) The original lena image (b) The original facial image (c) The original moon image. The experimental results are simulated with the software MATLAB 7.10 version. It provides a single platform for computation, visualization, programming and software development. All problems and solutions in Matlab are expressed in notation used in linear algebra and essen tially involve operations using matrices and vectors. We are using a 256 × 256 “Lena”, “facial”, and “Moon” as the gray scale original host image, and a 256 × 256 grey scale image of the watermark image. The three images are shown in Figures 5 and 6 respectively. In the pro posed method, we select the largest complexity of blocks; the original images can be separated into blocks (a) (b) (c) Figure 6. Three watermarked images of 256 × 256 pixels (a) The watermarked lena image; (b) The watermarked facial image; (c) The watermarked moon image. Table 1. The experimental results of the error ratio of the embedded watermark after different attacks by proposed method. Images Type Without Attacks Cropping Attacks Pyramid AttacksRotation AttacksNoise AttacksBlurring Attacks PSNR (DB) Lena Image 29.2752 38.7569 45.3102 30.0444 26.8879 29.2752 α = 0.2 Facial Image 29.9337 35.2695 46.3784 31.3159 26.5601 29.9337 α = 0.2 Moon Image 24.0375 36.3759 42.8253 26.2853 24.0375 24.0375 α = 0.2
M. THAPA ET AL. 176 Table 2. The experimental results of the error ratio of the extracted watermark after different attacks by the proposed method. Images Type Without Attacks Cropping Attacks Pyramid AttacksRotation AttacksNoise AttacksBlurring Attacks PSNR (DB) Lena Image 272.2211 279.3277 282.3816 273.8656 283.2165 272.2217 α = 0.2 Facial Image 284.1641 283.9997 280.3740 285.8853 279.7679 284.1641 α = 0.2 Moon Image 275.1305 271.2818 287.8321 278.3293 275.1305 275.1305 α = 0.2 tacks, rotation attacks, and noise attacks and blurring at tacks Figures 712. However, cropping is a geometrical manipulation and rotation is a geometrical distortion in practical application. If alpha’s value is more than 0.2 then quality of original image and watermarked image is not good. So we are using the dumpy value in these techniques. We are using the different coefficent of parameters by proposed method based on digital image watermarking embedding algorithm and extracting algorithm as a shown by Tables 3 and 4. Its depends upon the security analysis and as shown in Figures 1330. The parameters are to be satisfying by E1, E2 and E3. E1 is the parameter of “Lena Image”, E2 is the parameter of “Facial Image” and E3 is the parameter of “Moon Image”. Figure 7. Grey level of original image and watermarked image under without attacks and the corresponding qualities: (a) Lena (29.2752 db), (272.2211 db), (b) Facial (29.9337 db), (284.1641 db), (c) Moon (24.0375 db), (275.1305 db). Figure 8. Grey level of original image and watermarked image under cropping attacks and the corresponding qualities: (a) Lena (38.7569 db), (279.3277 db), (b) Facial (35.2695 db), (283.9997 db), (c) Moon (36.3759 db), (271.2718 db). Figure 9. Grey level of original image and watermarked image under pyramid attacks and the corresponding qualities: (a) Lena (45.3102 db), (288.3216 db), (b) Facial (46.3784 db), (280.3740 db) (c) Moon (42.8253 db), (287.8721 db). Figure 10. Grey level of original image and wtaremarked image under rotation attacks and the corresponding qualities: (a) Lena (30.0444 db), 273.8656 (db), (b) Facial (31.3159 db), 285.8853 (db) (c) Moon (26.2853 db), 278.3293(db). Copyright © 2011 SciRes. JIS
177 M. THAPA ET AL. Figure 11. Grey level of original image and watermarked image under noise attacks and the corresponding qualities: (a) Lena (26.8879 db), 283.2165 (db) (b) Facial (26.5601 db), 279.7678 (db) (c) Moon (24.0375 db), 275.1305 (db). Figure 12. Grey level of original image and watermarked image under blurred attacks and the corresponding qualities: (a) Lena (26.8879 db), 272.2217 (db), (b) Facial (26.5601 db), 284.1641 (db), (c) Moon (24.0375 db), 275.1305 (db). Table 3. A similarity between coefficients of original image and watermarked image using different parameters by proposed method under embedded algorithm. Coefficent of Parameters Without Attacks Cropping Attacks Pyramid AttacksRotation AttacksNoise AttacksBlurring Attacks PSNR (DB) E1 29.2753 38.7691 29.2752 30.4487 27.4361 23.5507 α = 0.2 E2 29.9334 35.3072 38.4104 10.0200 27.5376 29.9347 α = 0.2 E3 24.0375 36.3290 24.0375 11.1151 12.4233 23.5507 α = 0.2 Table 4. A similarity between coefficients of original image and watermarked image using different parameters by proposed method under extracted algorithm. Coefficent of Parameters Without Attacks Cropping Attacks Pyramid AttacksRotation AttacksNoise AttacksBlurring Attacks PSNR (DB) E1 272.2212 279.3657 272.2212 273.8651 271.7452 272.4247 α = 0.2 E2 241.1641 285.0083 278.8258 284.1675 284.8783 284.1667 α = 0.2 E3 275.1305 291.1681 275.5309 275.1345 273.5305 276.7801 α = 0.2 Figure 13. A similarity between coefficents of original watermarked image under without attacks: (a) Lena without attacks (29.2753 db), (272.2212 db). Copyright © 2011 SciRes. JIS
M. THAPA ET AL. 178 Figure 14. A similarity between coefficents of original watermarked image under cropping attacks: (a) Lropped attacks (38.7691 db, 279.3657db). ena c Figure 15. A similarity between coefficents of original watermarked image under pyramid attacks: (a) Lena py ramid attacks (29.2752), (272.2212 db). Figure 16. A similarity between coefficents of original watermarked image under noise attacks: (a) Noise tacks (27.4361 db), (271.7452 db). at Copyright © 2011 SciRes. JIS
179 M. THAPA ET AL. Fgure 17. A similarity between coefficents of original watermarked image under rotated attacks: (a) Lena rotated attacks (30.4487 db), (273.865 db). Figure 18. A similarity between coefficents of original watermarked image under blurring attacks: (a) Lea blurring attacks (23.5507 db), (272.4247 db). n Figure 19. A similarity between coefficents of original watermarked image under without attacks: (a) Faal without attacks (29.9337 db), (284.1641 db). ci Copyright © 2011 SciRes. JIS
M. THAPA ET AL. 180 Figure 20. A similarity between coefficents of original watermarked image under cropped attacks: (a) Facial cr opped attacks (35.3072 db), (285.0083 db). Figure 21. A similarity between coefficents of original watermarked image under pyramid attacks: (a) Facial pyramid attacks (38.4104 db), (278.8258 db). Figure 22. A similarity between coefficents of original watermarked image under rotated attacks: (a) Facial otated attacks (10.0200 db), (284.1675 db). r Copyright © 2011 SciRes. JIS
181 M. THAPA ET AL. Figure 23. A similarity between coefficents of original watermarked image under noise attacks: (a) Facial noise attacks (27.5376 db), (284.8483 db). Figure 24. A similarity between coefficents of original watermarked image under blurring attacks: (a) Facial brred attacks (29.9347 db), (284.1667 db). lu Figure 25. A similarity between coefficents of original watermarked image under without attacks (a) Mooithout attacks (24.0375 db), (275.1305 db). n w Copyright © 2011 SciRes. JIS
M. THAPA ET AL. 182 Figure 26. A similarity between coefficents of original watermarked image under cropped attacks: (a) Moon cropped attacks (36.329 db), (291.1681 db). Figure 27. A similarity between coefficents of original watermarked image under pyramid attacks: (a) Moonyramid attacks (24.0375 db), (275.5309 db). p Figure 28. A similarity between coefficents of original watermarked image under rotated attacks: (a) Mo rotated attacks (12.4233 db), (273.5305 db). on Copyright © 2011 SciRes. JIS
M. THAPA ET AL. Copyright © 2011 SciRes. JIS 183 Figure 29. A similarity between coefficents of original watermarked image under noise attacks: (a) Moon nois attacks (11.1151 db), (274.1345 db). e Figure 30. A similarity between coefficents of original watermarked image under without and blurred attacks: (a) Moon blurred attacks (23.5507 db), (276.7801 db). decomposition s one of emerging area of research. were explored in the proposed technique that provide In this way, we designed a singular value algorithm based on digital image watermarking tech nique and are to be following by this Figure 31. 6. Conclusions Digital watermarking i In this paper, we proposed a digital image watermarking algorithm based on singular value decomposition. The algorithm is used for watermarking embedding and wa termark extraction. The feautre of the D component and the realation ship between the U Component coefficents stronger robustness against different attacks and better image quality. So, Digital image watermarking tech niques are secure on this algorithm. If alpha has a less than 0.2 value then quality of the original image and wa termarked image is good. The experimental results also recognized the effectiveness of the proposed technique. Because of these properties, SVD is used for DCT, DFT, and DWT transformations, and oneway nonsymmetri cal decomposition. These provide the advantages of various sizes of transformation and more security. That is a good performance of the proposed scheme both in
M. THAPA ET AL. 184 Figure 31. SVD algorithm terms of robustness and security. d D. Hatzinakos, “Digital Watermarking Wavelet Decomposition,” Speech 7. References [1] D. Kundur an Using Multiresolution and Signal Processing Proceedings, Acoustics, 1997, pp. 29692972. [2] P. Zeng and C. Jin, “Image Adaptive Watermarking Us ing Visual Models,” IEEE Journal on Selected Areas in Communications, Vol. 16, No. 4, 1998, pp. 525539. doi:10.1109/49.668975 [3] L. Rajab, T. Khatib and A. Haj, “Combined DWTDCT Digital Image Watermarking,” Journal of Computer Sci ence, Vol. 3, 2002, pp. 740749. [4] C. C. Chang and P. Tsai, “SVDbased Digital Image Wa termarking Scheme,” Pattern Recognition Letters, Vol. 26, No. 10, 2005, pp. 15771586. doi:10.1016/j.patrec.2005.01.004 [5] T. V. Nguyen and J. C. Patra, “A Simple ICA Based Di gital Image Watermarking Scheme,” Digital Signal Pro cessing, Vol. 18, No. 5, 2007, pp. 762776. doi:10.1016/j.dsp.2007.10.004 [6] A. H. Ali and M. Ahmad, “Digital Audio Watermarking Based on the Discrete Wavelets Transform and Singular ine, Vol. 18, No. 4, 2001, pp. 3346. Value Decomposition,” Europe Journal of Science Re search, Vol. 39, No. 1, 2010, pp. 621. [7] C. I. Podilchuk and E. J. Delp, “Digital Watermarking: Algorithms and Applications,” IEEE Signal Processing Magaz doi:10.1109/79.939835 [8] B. Kim, J. G. Choi and D. Min, “Robust Digital Water marking Method Against Geometric Attacks,” Real Time Imaging Processing, Vol. 9, No. 2, 2003, pp. 139149. doi:10.1016/S10772014(03)000202 [9] H. Tina, W. Lu, R. prawn and Y. Ming, “A Fragile Wa termarking Scheme for 3D Meshes,” MMSEC’05, ACM, pp. 117123, 2008. [10] W. Loo and X. Kingsbury, “Digital Watermarking using . Sukang, “A Digital Watermarking Complex Wavelets,” International Conference on Image Processing, Vol. 3, 1999, pp. 2932. [11] M. Jiansheng and L Algorithm Based on DCT and DWT,” International Sym posium on Web Information System and Application (WISA), 2009, pp. 104107. [12] A. H. Ali and M. Ahmad, “Digital Audio Watermarking Based on the Discrete Wavelets Transform and Singular Value Decomposition,” Europe Journal of Science Re search, Vol. 39, No. 1, 2010, pp. 621. [13] W. Lu, H. Lu and F. L. Chung, “Feature Based Water marking Using Watermark Template Match,” Applied Ma thematics and Computation, Vol. 177, No. 1, 2011, pp. 886893. [14] Y. Lu, K. Uehira and K. Yanaka, “Practical Evaluation of Illumination Watermarking Technique Using Orthogonal Transforms,” Journal of Display Technology, Vol. 6, No. 9, 2010, pp. 351358. doi:10.1109/JDT.2010.2049336 Copyright © 2011 SciRes. JIS
