水印与加密Joint watermarking and encryption is an upcom

文件名称: 水印与加密

所属分类: 网络安全

开发工具:

文件大小: 499kb

下载次数: 0

上传时间: 2019-03-04

提供者: jmu***

下载 (499kb)

不能下载？报告错误

详细说明：Joint watermarking and encryption is an upcoming security solution that combines leading but complementary techniques to achieve an enhanced security level.Multimed Tools Appl(2013)67: 593-606 595 work should provide robustness to the embedded watermark against compression without deteriorating the compression efficiency. The present work, thus, intends to develop a jWE framework that achieves data confidentiality, proves content ownership and offers high compression ratio. In the proposed framework, watermarking and encryption are implemented at content owner and content distributor end, respectively. To achieve the desired objectives watermark is embedded using singular value decomposition of the wavelet packet transformed image and encryption is performed during SPIhT encoding. Security attained by the proposed JWE framework is ascertained by detailed experimental analysIs 2 Singular value decomposition Singular value decomposition(SvD)is a powerful technique in many matrix compu tations and analyses [4. Use of SVD in matrix computations provides robustness provides robustness against numerical errors. SVD of a square or a rectangular matrix of size M x N can be expressed as A=U米S*V7 where U and v are orthogonal(unitary) matrices, and s is a diagonal matrix given by s=diag(o1, 02, ., Or).Here, oi denotes singular value of matrix A, and 202> σr,1≤i≤ r and r=min(M,N). The first r columns of v and u are termed as right and left singular vectors, respectively. The main motivation for using the SvD is its energy compaction property and its ability to adapt to the variations in local statistics of an image. Each singular value of the image matrix specifies luminance of the image layer, while corresponding pair of singular vectors specify geometry of the image layer. Therefore slight variations of singular values does not affect visual perception of the cover image Also, storing the approximation of a matrix using SVD often results in a significant savings over storing the whole matrix Singular values of a matrix possess algebraic and geometric invariance to some extent, due to which it has certain distinct advan tages in digital image processing. For instance, singular values of an image matrix remain same, irrespective of the transposition, rotation or translation performed on the original matrix. Further, singular values of an image are less effected in case of general image processing operations on the image matrix 3 Proposed joint watermarking and encryption framework In the proposed JWE framework, the original image X is initially watermarked using key Kw. This watermarked image is then partially encrypted with key Ke. The final image obtained by implementing these two processes in a sequential manner is mathematically expressed as Y= W(XB K Z=E(Y, Ke ger 596 Multimed Tools Appl(2013)67: 593-606 Here, Y, B, Kw, wo,Z, Ke and EO are the watermarked copy of original image X, watermark, watermark key, watermark embedding algorithm, encrypted copy of watermarked component Y, encryption key and encryption algorithm, respectively The watermarking key, Kw comprises of watermark strength, a as the main key component. This controls perceptibility of the embedded watermark; higher the value of a, more observable is the watermark. An optimal value can be chosen as per the desired visibility of embedded watermark. In contrast, the encryption key, Ke consist of the compression ratio, number of Arnold iterations for scrambling and a seed value for generating a random vector a block diagram depicting the proposed JWE framework is indicated in Fig. 1 and the two processes controlled by the independent keys, Ku and Ke, are explained as follows The watermarking process initially transforms the host image into wavelet packet transform(WPT) domain. SVd is then performed on all subbands of the transformed image and the watermark image For watermark embedding, the obtained singular values are modified using(4) where oge gives original singular values of the subband,(of, e )*denotes modified singular values of the subband, ow denotes singular values of the watermark image and a is the watermark strength After replacing original singular values by the modified values, inverse SVD s taken. This is followed by inverse WPT to retrieve the watermarked image This watermarked image is transmitted to the content distributor end. Thereafter encryption is performed on this watermarked image during SPIHT compression [17] In sPiht compression, an image is initially transformed into wavelet domain, and a tree structure is formed. The tree structure is then encoded to obtain a spiht compressed bitstream. In the proposed framework, encryption is implemented in wavelet domain, just before the formation of tree structure. Encryption is achieved by scrambling the approximation subband using Arnold cat map[15]. This is followed by sign bit encryption of the scrambled transform coefficients using a stream cipher, Watermarking Perform WPt and Modify Singular value Perform Inverse SVD Original Image then Svd for watermark and IWPt embedding Watermarked Image Encryption Perform dWt Sign bit encryption of Watermarked and scrambled SPIHT Compression HI> Encrypted Compressed Approx imation band ig. 1 Block diagram for the proposed framework Springer Multimed Tools Appl(2013)67: 593-606 597 generated from a seed value. The original approximation subband is replaced with the scrambled and sign bit encrypted subband. Afterwards the modified transformed image is used to generate the compressed bitstream using SPIhT At the receiver end, compressed bitstream generates the transformed image and inverse dWt of this transformed image provides the reconstructed image An unauthorized receiver, not having the security keys, would only retrieve an incomprehensible image. Contrary to this, an authorized receiver would perform sign bit decryption and arnold descrambling of the approximation subband before IDWT. This provides a correctly decrypted output to an authorized receiver. To verify achieved security level of the proposed framework, several experiments have been performed, and are discussed in the next section 4 Results and discussion The proposed framework provides twin layer of protection to digital images by combining watermarking and encryption. To substantiate performance of the pro- posed framework, different subjective and objective evaluation parameters are used Diverse watermarking and encryption related security attacks are also launched to assess robustness of the proposed framework. Simulations have been performed on various grayscale images, however, results for only 'Barbara' image are illus trated here 4.1 Subjective and objective evaluation To examine the quantum of detail actually lost, or retained by the proposed JWE framework, a visual inspection of the watermarked and encrypted images is per formed. These images are illustrated in Fig. 2. It is observed that the watermarked Fig 2 Results for the proposed framework UIT Original image Original watermark Watermarked image (a) (b) (c) Encrypted image Decrypted image Extracted watermark Springer 598 Multimed Tools Appl(2013)67: 593-606 Table 1 PSnr(dB)obtained for various images Image Ba arbara Lena PI lane Crowd Bridge Lake Watermarked 40.1201 38.7121 38.9503 39.5840 37.5025 37.2499 Encrypted 3.2341 3.1203 3.1842 3.0001 3.1981 2.9064 image is similar to the original image, and the encrypted image is completely incomprehensible. The embedding of watermark in an imperceptible manner has not resulted into loss of any detail from the original image. In contrast, an unintelligible encrypted image reflects that the developed encryption technique provides high data confidentiality and does not leak any information of the original image. To further verify the results, objective evaluation is performed using peak signal to noise ratio (PSNR) The obtained PsNR values for the watermarked and the encrypted output with reference to the original image are indicated in table 1.a high psnr value after watermarking indicates perceptual similarity between the original and the watermarked image, while a low Psnr value of the encrypted output indicates sufficient dissimilarity between the original and the encrypted image. This indicates more computational effort required by an intruder to retrieve the correct image without the knowledge of security keys. pSnR values and visual inspection of results depicts that the proposed technique satisfies the subjective and objective evaluation metric for an acceptable watermarking and encryption technique 4.2 Key sensitivity analysis As per the Kerckhoffs principle, security keys are the most important part of any cryptosystem, and decryption using an incorrect key or an approximately correct key should not reveal any details of the original image [18. To determine key sensitivity of the proposed encryption technique, wrong decryption keys are generated by introducing slight modifications in Arnold scrambling iterations and the seed value. Decryption is then performed by using these slightly modified keys. Figure 3a and demonstrates the decrypted results when incorrect descrambling iterations or ncorrect seed value is considered. It is observed that the images decrypted with wrong decryption keys do not give a clear view of the original image. This reflects high key sensitivity of the developed encryption algorithm Thereafter, watermark is extracted from these incorrectly decrypted images The extracted watermarks are shown in Fig. 3b and d. It is observed that despite unintelligible decrypted images, meaningful watermarks are extracted. This reflects robustness of the watermark embedding technique. To further verify strength of em bedding technique, watermark is extracted from the encrypted image. The extracted watermark is indicated in Fig. 3f, and can easily be related to the original watermark Strength of the encryption technique is evident from achieved data confidentiality and high key sensitivity. In addition to this, extraction of watermark from the encrypted and incorrectly decrypted image illustrates strength of the watermarking technique. This depicts that content ownership can be proved in a scenario, where a Springer Multimed Tools Appl(2013)67: 593-606 599 Fig 3 Key sensitivity results for the encryption technique HIT Wrong Arnold descrambled image Wrong sign bit decryption (a) (b) fIT Encrypted image Approximation attacked image (e) g) pirate captures an unclear but watermarked copy of the original image. The above analysis corroborates strength of the developed JWE framework 4.3 Compression performance analysis In the proposed jwe framework, compression ratio achieved by the employed SPIHT encoder is analyzed. To experimentally evaluate the effect on compression efficiency, original and encrypted images are compressed with 0. 8 bits per pixel. As the output bit rate is equal for both the images, length of the compressed bitstream is observed to be same for all the original and the encrypted image. This indicates that the proposed framework does not adversely effect compression efficiency of the SPIHT encoder 4.4 Approximation attack Security of the proposed technique is also verified against approximation attack [14. In this attack, part of the encrypted data is replaced by random data and reconstruction is performed using this partially assumed data. In the present case few transform coefficients of approximation subband are replaced by a constant value 0, before IDWT. Figure 3g and h shows the reconstructed image and the extracted watermark for this case. It is observed that a clear view of the original mage is not obtained. However, watermark extracted from this approximate image has perceptual resemblance with the original watermark. This indicates resistance of the proposed framework for approximation attack An approximated copy of the test image is used to measure block-based Lu minance Similarity Score (LSS), which captures the coarse luminance information 14. LSS measures the perception-oriented distance between the clear-text copy of ger 600 Multimed Tools Appl(2013)67: 593-606 multimedia and attacker's recovered copy from the encrypted media. It was assumed that two given images are pre-processed to be aligned and scaled to the same size These two images are first divided into blocks in the same way, using 8 x 8or 16x 16 non-overlapping blocks. Average luminance values of ith block is then calculated from both images to measure LSS using LSSE ∑ f(x1;,x2) (5) Here, the function f(x1, x2)for each pair of average luminance values is defined as if x1. a round otherwise where the parameters a and B control sensitivity of Lss and set to 0.1 and 3, respectively. For the proposed framework, negative LSS is obtained. This indicates a substantial dissimilarity in luminance of the two images 4. 5 Attack analysis After performing key sensitivity and approximation attack analysis for the proposed encryption technique, this section discusses performance of the proposed watermark ing technique in different attack scenarios. Robustness of the proposed watermark ing technique is investigated by launching various attacks on the watermarked image and observing the quality of extracted watermarks from the attacked images. Visual inspection of the extracted watermark is performed to assess its perceptual similarity with the original watermark. For an objective evaluation of similarity, correlation coefficient is calculated between the extracted and actual singular values, using >(w(i)-Wmean)(w(i)-Wmean 0(U,) (v(i) mean mean where w, W, Wmean and Wmean are the original singular values, extracted singular values, mean of the original singular values and mean of the extracted singular values Here, r= min(M, M), and(m, m denote size of the image. Among the various attacks launched on the watermarked image, basic attack includes(a)a 13 13 Gaussian blurring on the watermarked image, (b) rotation of the watermarked image by 500, and (c) addition of 80% Gaussian noise to the watermarked image. Watermarks are extracted from these three attacked images Figure 4c-e indicates the attacked watermarked images and their corresponding extracted watermarks. It is observed that the extracted watermarks are recognizable and can be assumed as a degraded version of the original watermark Correlation coefficient values for the extracted watermarks is indicated in table 2 As cropping is a frequently used operation in image applications, watermarked image is also tested for cropping attack. Process of selecting and removing a portion of an image is generally performed to create focus or strengthen the composition. In the present test case, the watermarked image is cropped to only 2.5% of the actual Springer Multimed Tools Appl(2013)67: 593-606 601 Fig 4 Image ar extracted watermark Original (a Gaussian blurred image Rotated image (d) 【 Noise attacked image Cropped image e) 回回 Resized image JPEG compressed image size, and watermark extraction is performed. Figure 4f indicates the cropped image and the extracted watermark lt is to be noted that the watermark could be extracted even from an image equal to 2.5% of the actual image size Another frequently used image processing operation is resizing, wherein the image is reduced or enlarged to a desired size. This leads to data loss of the original Table 2 Correlation coefficient(CC)for extracted watermark from attacked Barbara image Attack Gaussian Rotation Noise R ropping Blur compression CC 0.6885 0.9402 0.3732 0.9656 0.6832 06079 ger 602 Multimed Tools Appl(2013)67: 593-606 image and the watermark embedded within it. In the present test case, the image is reduced to 16 16 and again carried back to the original size 256X 256. Figure 4g depicts the resized image and its corresponding extracted watermark. It is observed that the extracted watermark is still recognizable and is similar to the origina watermark Another potential attack for a watermarking technique is compression, that is generally performed owing to the large data size and limited channel bandwidth As image compression techniques are lossy in nature, they lead to data loss from the entire image and the watermark embedded in it. Despite the losses, a secure system requires that the watermark is extractable, even from a compressed image To assess the proposed technique against compression attack, lossy JPEG com pression, with a compression ratio of 80: 1, is performed on the watermarked image Watermark is extracted from this JPEG compressed image. Figure 4h illustrates the jPeg compressed image and its extracted watermark. It is observed that the extracted watermark is of very high quality, and almost an exact replica of the original watermark. Further, it is to be noted that the proposed framework is based on SPIhT compression of the watermarked data. Hence, the watermark indicated in Fig. 4b is actually the watermark extracted from a spiht compressed image. This demonstrates robustness of the proposed watermarking technique against SPl compression. This reflects that the proposed watermarking technique can withstand lossy JPEG and SPIhT compression attack In all the above-mentioned attack scenarios it is observed that the extracted and the original watermark are perceptually similar. This, along with the correlation coefficient values indicated in Table 2, reflects the resistance of the proposed watermarking technique against various image processing operations. The above ownership, even from attacked and compressed images amev discussed analysis reflects the ability of the proposed framework to prove content 5 Comparative analysis This section discusses comparative analysis of the proposed framework, with an existing JWE framework [10. The existing framework utilizes the entire image data to provide data confidentiality It provides aes encryption to all the coefficients of low level subband, and sign bit encryption to all the coefficients of remaining subbands. In contrast, the proposed framework performs only sign bit encryption Table 3 Scheme of existing and proposed JWE Technique Watermarking Encryption Existing 10 All coefficients of middle level subband are AES in approximation subband and used,i.e,(MIX N1) coefficients used for sign bit encryption in remaining a subband of size (Mi x Nu) subbands P d Uses only singular values for watermarking Scrambling and sign bit encryption i.e., min(MI, N1) coefficients used for of only approximation subband a subband of size (MI x Nu) Springer

(系统自动生成,下载前可以参看下载内容)