VIEWS: 170 PAGES: 7 CATEGORY: Emerging Technologies POSTED ON: 3/14/2011 Public Domain
(IJCSIS) International Journal of Computer Science and Information Security, Vol. 9, No. 2, February 2011 A Quantization based blind and Robust Image Watermarking Algorithm Mohamed M. Fouad Electronics and Communication Department- Faculty of Engineering- Zagazig University- Egypt fouadzu@hotmail.com Abstract—Security and privacy issues of the transmitted data multi-media object. The embedding process is guided by use have become an important concern in multimedia technology. of a secret key, which decides the locations within the Watermarking which belong to the field of information hiding multimedia object (image) where the watermark would be has seen a lot of research interest recently. Watermarking is used embedded. Once the watermark is embedded it can experience for a variety of reasons including security, content protection, several attacks because the multimedia object can be digitally copyright management, trust management, content authentication, tamper detection and privacy. Recently many processed. The attacks can be unintentional (in the case of watermarking techniques have been proposed to support these images, low pass filtering or gamma correction or applications but one major issue with most of the watermarking compression) or intentional (like cropping). Hence, the techniques is that these techniques fail in the presence of severe watermark has to be very robust against all these possible attacks. This has been a major threat to content providers attacks. When the owner wants to check the watermarks in the because if the digital content is dramatically changed then it possibly attacked and distorted multimedia object, s/he relies would be difficult to prove the existence of a watermark in it and on the secret key that was used to embed the watermark. Using consequently its ownership. To tackle this security threat towards the secret key, the embedded watermark sequence can be ownership issues in this paper, we propose a computationally extracted. This extracted watermark may or may not resemble efficient and secure two quantization based watermarking algorithms which offer incredible performance in presence of the original watermark, because the object might have been malicious attacks which try to remove ownership information. attacked. The performance of the proposed techniques is compared with Hence, to validate the existence of a watermark, either the that of other watermarking techniques and it gives a very good original object is used to compare and ascertain the watermark perceptual quality especially at lower bit rates. We present experimental results which show that the proposed techniques signal (non-blind watermarking), or a correlation measure is outperform many techniques for multimedia over wireless used to detect the strength of the watermark signal from the applications. The proposed schemes are backed up with excellent extracted watermark (blind watermarking). In correlation results. based detection, the original watermark sequence is compared with the extracted watermark sequence, and a statistical Keywords-component; Watermark Detection; Watermarking; DCT; DWT; Quantization correlation test is used to determine the existence of the watermark. I. INTRODUCTION A. Requirements of Digital Watermarking Watermarking is a method of hiding proprietary There are three main requirements of digital information in digital media like photographs, digital music, or watermarking. They are transparency, robustness and digital video. The ease with which digital content can be capacity. exchanged over the Internet has created copyright infringement issues. Copyrighted material can be easily Transparency or Fidelity, The digital watermark should exchanged over peer-to-peer networks, and this has caused not affect the quality of the original image after it is major concerns for those content providers who produce these watermarked. Cox et al. (2002) defines transparency or fidelity digital contents. In order to protect the interest of the content as ‘perceptual similarity between the original and the providers these digital contents can be watermarked. watermarked versions of the cover work’ [1]. Watermarking should not introduce visible distortions because if such The process of embedding a watermark in a multimedia distortions are introduced it reduces the commercial value of object is termed as watermarking. A Watermark can be considered as a kind of a signature, which reveals the owner of the image. the multimedia object. Content providers want to embed watermarks in their multimedia objects (digital content) for several reasons like copyright protection, content authentication, tamper detection etc. A watermarking algorithm embeds a visible or invisible watermark in a given 241 http://sites.google.com/site/ijcsis/ ISSN 1947-5500 (IJCSIS) International Journal of Computer Science and Information Security, Vol. 9, No. 2, February 2011 Robustness, Cox et al. (2002) defines robustness as the slice has a width that is inversely proportional to its height. ‘ability to detect the watermark after common signal The number of these slices is equal to the number of processing operations’ [1]. Watermarks could be removed quantization levels. intentionally or unintentionally by simple image processing 2. On the horizontal axis of the sliced histogram, each slice operations like contrast or brightness enhancement, gamma has start and end points. The midpoint value (on the width) correction etc. Hence watermarks should be robust against a of each slice is considered as a quantization level. variety of such attacks into four basic categories, attacks that try to remove watermarks totally, attacks that try to remove 3. In this way, we get a non-uniform quantization in which the synchronization between the embedder and the detector, the density of the quantization levels increases in cryptographic attacks and protocol attacks. proportion to the probability of occurrence of the pixel value. Capacity or Data Payload, Cox et al. (2002) define capacity or data payload as ‘the number of bits a watermark 4. All the pixel values that lie within the width of a slice are encodes within a unit of time or work’ [1]. This property mapped to the quantization level that is represented by the describes how much data should be embedded as a watermark midpoint of this slice. to successfully detect during extraction. Watermark should be The resultant compression ratio and signal-to-noise ratio able to carry enough information to represent the uniqueness vary depending on the chosen number of quantization levels. of the image. Different applications have different payload requirements [1]. This technique is irreversible, i.e. the quantized values can’t be converted back to their original values leading to Security, according to Kerckhoff’s principle the security information loss. of a cryptosystem depends on the secrecy of the key and not on the cryptographic algorithm. Same rule applies to water- III. DCT PROPOSED WATERMARKING TECHNIQUE marking algorithms, i.e. the watermarking algorithms must be public but watermark embedding should base on a secret key The first proposed watermarking scheme is a blind [2]. quantization based scheme [4]. A block diagram detailing its steps is shown in Fig. 1. The input N*M image; an image To prevent image manipulations and fraudulent use of assumed to be a matrix has length of N rows and width of M modiﬁed images, the watermark should survive modiﬁcations columns, is first converted into single vector by concatenating introduced by random noise or compression, but should not be successive rows beside each other to form a long row that detectable from non-authentic regions of the image. The contains all the image pixels using matrix to vector converter. original image cannot be used by the watermark detect or to This vector is exposed to DCT [5]-[7] to transform the image verify the authenticity of the image. In this paper, we from spatial domain into frequency domain in which energy of investigate the application of a recently developed the image information is concentrated in a few number of quantization based watermarking scheme to image coefficients. The output of the DCT process is a vector that authentication. The two proposed watermarking techniques has the same length of the image) number of pixels in the allow reliable blind watermark detection from a small number image), but with many values approximated to zeros. After of pixels, and thus enable the detection of local modiﬁcations applying the DCT the output coefficients are arranged in a to the image content. descending order according to the pixels probabilities. The output vector of the DCT is now ready to be processed by the II. HISTOGRAM EQUAL AREA DIVISION histogram equal area quantization technique to choose the QUANTIZATION TECHNIQUE appropriate values used in the watermark embedding process, quantization levels. The watermarked coefficients vector is The technique calculates the quantization levels using a reshaped and returned back to the spatial domain using IDCT. method that is dependent on the image content (hence the word "adaptive") and then round off the pixels values to the nearest quantization level. In this way, the number of transmitted values is reduced. The quantization scheme provides a wide range of compression ratios (CRs) with a very slight degradation of the signal-to-noise ratio (SNR). HEAD is a quantization technique in which the transmitted values are reduced by mapping the values of Figure 1. The first proposed image watermarking scheme. image pixels to a finite number of quantization levels. A. Watermark Embedding The HEAD quantization procedure can be listed as follows: 1. The area under the histogram of the image pixels is divided The steps of watermark embedding can be summarized as into a number of vertical slices with equal areas. Thus each follows: 242 http://sites.google.com/site/ijcsis/ ISSN 1947-5500 (IJCSIS) International Journal of Computer Science and Information Security, Vol. 9, No. 2, February 2011 1. The host image is transformed into the DCT domain; the No watermark was inserted into the low-pass sub-band. Unlike transformed coefficients are watermarked using HEAD some non-blind watermarking schemes [9][10], this scheme quantization using 4 quantization levels t0, t1, t2, and t3. allows a watermark to be detected without access to the 2. A binary watermark of the same size as the image of original image. It performs an implicit visual masking as only interest is created using a secret key, which is a seed of a wavelet coefficients with large magnitude are selected for random number generator. watermark insertion. These coefficients correspond to regions 3. Each s wij of the selected DCT Coefficients is quantized. of texture and edges in an image. This scheme makes it difficult for a human viewer to perceive any degradation in the The quantization process can be summarized as follows: watermarked image. Also, because wavelet coefficients of 's If xij = 1 and s wij > 0, then wij = t2, large magnitude are perceptually significant, it is difficult to remove the watermark without severely distorting the 's If xij = 0 and s wij > 0, then wij = t1, watermarked image. The most novel aspect of this scheme was 's the introduction of a watermark consisting of pseudorandom s If xij = 1 and wij < 0, then wij = -t3, real numbers. Since watermark detection typically consists of s 's a process of correlation estimation, in which the watermark If xij = 0 and wij < 0, then wij = -t0. (1) coefficients are placed in the image, changes in the location of ' the watermarked coefficients are unacceptable. The Where xij the watermark is bit corresponding to wij , and s wijs watermarking scheme proposed by Dugad et al. is based on is the watermarked coefficient. After all the selected adding the watermark in selected coefficients with significant coefficients are quantized, the inverse discrete cosine energy in the transform domain in order to ensure the non- transform (IDCT) is applied and the watermarked image is erasability of the watermark. This scheme has overcome the obtained. problem of “order sensitivity”. B. Watermark Detection Unfortunately, this scheme has also some disadvantages. It embeds the watermark in an additive fashion. It is known that 1. The possibly corrupted watermarked image is transformed blind detectors for additive watermarking schemes must into the DCT domain as in the embedding process. correlate the possibly watermarked image coefficients with the 2. The extraction is performed on the coefficients. known watermark in order to determine if the image has or has 3. All the coefficients of magnitude equal to t1, t2, - t3 and - t0 not been marked. Thus, the image itself must be treated as 's are selected; these are denoted wij .The watermark bits noise, which makes the detection of the watermark exceedingly difficult [8]. In order to overcome this problem, it are extracted from each of the selected DCT coefficients is necessary to correlate a very large number of coefficients, with Eq.2. Fig. 2 illustrates the watermark detection which in turn requires the watermark to be embedded into process. several image coefficients at the insertion stage. As a result, the degradation in the watermarked image increases. Another drawback is that the detector can only tell if the watermark is present or not. It cannot recover the actual watermark. The scheme in [11] is another example of wavelet-based watermarking schemes. A noise-like Gaussian sequence is Figure 2. Watermark detection in the proposed scheme. used as a watermark. To embed the watermark robustly and 's imperceptibly, watermark components are added to the If wij = t2 or t3, then the recovered watermark bit is a 1. significant coefficients of each selected sub-band by 's considering the human visual system (HVS) characteristics. If wij = t0 or t1, then the recovered watermark bit is a 0 Some small modifications are performed to improve the HVS (2) model. The host image is needed in the watermark extraction 4. The recovered watermark is then correlated with the procedure. original watermark in the watermark file, obtained via the secret key. This allows a confidence measure to be V. PROPOSED DWT WATERMARKING TECHNIQUE ascertained for the presence or absence of a watermark in an image. Discrete wavelet transform is a technique using which a 2D image can be transferred from spatial domain to frequency IV. DWT WATERMARKING TECHNIQUE domain. The input N*M image; an image assumed to be a matrix has length of N rows and width of M columns, is Dugad et al. presented a blind additive watermarking exposed to wavelet transform. After one level DWT an image scheme operating in the wavelet domain [8]. Three-level I is decomposed into four subbands LL, HL, LH, and HH. LL wavelet decomposition with Daubechies 8-tap filters was used. is called the approximate band and it contains most of the 243 http://sites.google.com/site/ijcsis/ ISSN 1947-5500 (IJCSIS) International Journal of Computer Science and Information Security, Vol. 9, No. 2, February 2011 energy. In the algorithm we decompose the image into four 4. After all the selected coefficients are quantized, the levels and embed the watermark in HL, LH sub-bands. Here inverse discrete wavelet transform (IDWT) is applied and we assume the size of the watermark logo is in multiple of the the watermarked image is obtained. sub-band size. In the second proposed a quantization based B. Watermark Detection watermarking algorithm, we incorporate implicit visual masking by embedding the watermark in the LH, HL sub- bands. The output vector of the wavelet is now ready to be 1. The possibly corrupted watermarked image is processed by the histogram equal area quantization technique transformed into the wavelet domain using the same to choose the appropriate values used in the watermark wavelet transform as in the embedding process. embedding process, quantization levels. The watermarked 2. The extraction is performed on the coefficients in the first coefficients vector is reshaped and returned back to the spatial level wavelet transform (excluding the LL1 subband). domain using IDWT. 3. All the coefficients of magnitude equal to t1, t2, - t3 and - t0 's are selected; these are denoted wij .The watermark bits are extracted from each of the selected DCT coefficients with Eq.4. Fig. 4 illustrates the watermark detection process. Figure 3. The proposed image watermarking scheme. Figure 4. Watermark detection in the proposed scheme. 's If wij = t2 or t3, then the recovered watermark bit is a 1. A. Watermark Embedding 's If wij = t0 or t1, then the recovered watermark bit is a 0 The steps of watermark embedding can be summarized as (4) follows: 1. The host image is transformed into the wavelet domain; one 4. The recovered watermark is then correlated with the level Daubechies wavelet with filters of length 4 is used. original watermark in the watermark file, obtained via the The coefficients (excluding the LL1 and HH1) secret key. This allows a confidence measure to be coefficients are watermarked using HEAD quantization ascertained for the presence or absence of a watermark in using 4 quantization levels t0, t1, t2, and t3. an image. 2. A binary watermark of the same size as the subbands of 5. The recovered watermark is then correlated with the interest is created using a secret key, which is a seed of a original watermark in the watermark file, obtained via the random number generator. secret key, only in the locations of the selected 3. Each s wij of the selected wavelet coefficients is quantized. coefficients. This allows a confidence measure to be ascertained for the presence or absence of a watermark in The quantization process can be summarized as follows: an image. s 's If xij = 1 and wij > 0, then wij = t2, VI. PERCEPTUAL QUALITY METRICS s 's If xij = 0 and w > 0, then wij = t1, ij Two metrics for ascertaining the quality of a watermarked s 's image are highlighted in this section. These metrics are the If xij = 1 and w < 0, then w = -t3, ij ij Mean Square Error (MSE), and the Peak Signal to Noise Ratio ' If xij = 0 and s wij < 0, then wijs = -t0. (PSNR). The MSE measures the average pixel-by-pixel difference between the original image (I) and the watermarked (3) ˆ image ( I ) [12]. 1 (5) s Where xij the watermark is bit corresponding to wij , and MSE = ∑ (I m,n − Iˆm,n )2 MN m,n ' wijs is the watermarked wavelet coefficient. Figure (3) 2 I (6) shows the watermark embedding in a positive wavelet PSNR ( dB ) = 10 log 10 peak coefficient. MSE 244 http://sites.google.com/site/ijcsis/ ISSN 1947-5500 (IJCSIS) International Journal of Computer Science and Information Security, Vol. 9, No. 2, February 2011 Where Ipeak is the peak intensity level in the original image (most commonly 255 for an 8-bit grayscale image), M and N are the dimensions of the image. The original and recovered messages or watermarks can be compared by computing the Normalized Correlation (NC) [12]: m * .m (7) NC = * m . m * Where m is the original message and m is the recovered message. For unipolar vectors, m ∈ {0, 1}, and for bipolar vectors, m ∈ {−1, 1}. VII. SIMULATION RESULTS For all the tests in this paper, MATLAB is used. All tests are performed upon the 8-bit grayscale 256 × 256 cameraman image. To simulate the watermarking schemes on the cameraman image, the four quantization levels are T0=113; T1=124; T2=156; T3=159. Results of the two schemes for the cameraman image are shown in Fig. 5 and Fig. 6, respectively. The comparison of fidelity is shown in Table I. The numerical evaluation metrics for all schemes in the absence and presence of attacks are tabulated in Tables II. From Table II, we notice that the proposed watermarking scheme achieves the lowest distortion in the watermarked image in the absence of attacks we find that the proposed using wavelet give the image with fidelity better than the tech using DCT. From Table II it gives the comparison between our technique using DCT and wavelets, we notice also that a percentage of around 50% of the input watermark bits can be extracted in the proposed scheme with most of the attacks. In the case of DCT we find that we can detect watermark at the presence of blurring, Gaussian or compression attack, in the case of wavelet we can detect the watermark at the presence of Gaussian, resizing, blurring or compression attack. Figure 5. Watermarked image using proposed technique with DCT We compare our results to daugads [8], LSB technique [9] and with and without attacks. the technique in [4]. In the case of LSB technique, we find it is difficult to detect the watermark at the case of attacks applied to the watermarked image. The technique in [4] gives better result than the existed technique and the proposed one in the case of compression. 245 http://sites.google.com/site/ijcsis/ ISSN 1947-5500 (IJCSIS) International Journal of Computer Science and Information Security, Vol. 9, No. 2, February 2011 TABLE II. COMPARISON OF NC OF THE EXTRACTED WATERMARKS FOR OUR SCHEME FOR THE CAMERAMAN IMAGE AND THE OTHER EXISTING TECHNIQUES. VIII. CONCLUSION This paper presented a blind DCT –DWT based image watermarking schemes. These schemes depend on the quantization of coefficients within certain amplitude ranges in a binary manner to embed meaningful information in the image. Experimental results have shown the superiority of the proposed schemes from the host image quality point of view and the blindness point of view. References [1] Cox, IJ, Miller, ML & Bloom, JA 2002, Digital Watermarking, Morgan Kaufmann Publisher, San Francisco, CA, USA. nd Figure 6. Watermarked image using the proposed DWT technique [2] Schneier, B., ‘Applied Cryptography’, WILEY, 2 Edition. with and without attacks. [3] Shaimaa A. El-said, Khalid F. A. Hussein, and Mohamed M. TABLE I. EVALUATION METRICS VALUES FOR ALL Fouad, “Adaptive Lossy Image Compression Technique,” SCHEMES FOR THE CAMERAMAN IMAGE. Electrical and Computer Systems Engineering Conference (ECSE’10), 2010. [4] Mohiy Mohammed hadhoud , Abdalhameed shaalan, hanaa Scheme PSNR NC abdalaziz abdallah “A Modified Image Watermarking Using DCT proposed technique 42 1 Scalar Quantization in Wavelet Domain” UbiCC Journal, DWT Proposed technique 52.7 1 Volume 4, Number 3, August 2009 LSB scheme blind 51.64 1 [5] A. S. Khayam, The Discrete Cosine Transform :Theory and daugad Scheme blind 38.42 0.39 Application, Michigan State University ,March 10th 2003 . Quantization Tech in [4] 47.29 1 [6] A. B. Watson, Image Compression Using the Discrete Cosine Transform, Mathematica Journal, 4(1), 1994 ,p. 81-88. [7] D.A. Huffman, A method for the construction of minimum- redundancy codes. Proc. Inst. Radio Eng. 40(9), pp.1098-1101, 1952. [8] K. Dugad, R. Ratakonda, and N. Ahuja, “A New Wavelet-Based Scheme for Watermarking Images,” Proceedings of 1998 International Conference on Image Processing (ICIP 1998), Vol. 2, Chicago, IL, October 4-7, 1998, pp. 419-423. [9] M. Corvi and G. Nicchiotti, “Wavelet-based image watermarking for copyright protection, Scandinavian Conference on Image Analysis,” SCIA ’97, Lappeenranta, Finland, June 1997, 157- 163. 246 http://sites.google.com/site/ijcsis/ ISSN 1947-5500 (IJCSIS) International Journal of Computer Science and Information Security, Vol. 9, No. 2, February 2011 [10] P. Meerwald, Digital image watermarking in the wavelet [11] S. Voloshynovskiy, S. Pereira, V. Iquise, and T. Pun. “Attack transform domain, Master thesis, Department of Scientific modeling: Towards a second generation watermarking Computing, University of Salzburg, Austria, 2001. benchmark” Journal of Signal Processing,80 (6) , May 2001. http://www.cosy.sbg.ac.at/˜pmeerw/Watermarking/ [12] C. Shoemaker, Rudko, “Hidden Bits: A Survey of Techniques for Digital Watermarking” Independent StudyEER-290 Prof Rudko, Spring 2002. 247 http://sites.google.com/site/ijcsis/ ISSN 1947-5500