doi:10.1016/S0020-0255(02)00291-8 A comparative study of digital watermarking in JPEG and JPEG 2000 environments M.A. Suhail a , M.S. Obaidat b,*, S.S. Ipson b , B. Sadoun c a Monmouth University, UK b University of Bradford, UK c Al-Balqa’ Applied University (BAU), Jordan Received 1 June 2001; accepted 27 April 2002 Abstract JPEG 2000 is a new compression technology that achieves very high compression rate and maintains visual quality. Digital watermarking techniques have been developed to protect the copyright of media signals. The goal of this paper is to put into perspective joint photographic experts group (JPEG) and JPEG 2000 concepts a long with water- marking principle. It provides evaluation of the compatibility aspects of JPEG 2000 versus JPEG standard with watermarking. Various experiments have been conducted to compare the performance of both standards under various conditions. An outlook on the future of digital image watermarking within JPEG 2000 is introduced. � 2002 Elsevier Science Inc. All rights reserved. Keywords: JPEG; JPEG 2000; Watermarking; Image processing; DCT; DWT; Image compression 1. Introduction Many watermarking schemes have been suggested for images and some for audio and video streams. A large number of these schemes address the problems Information Sciences 151 (2003) 93–105 www.elsevier.com/locate/ins * Corresponding author. Address: Department of Computer Science, Monmouth University, W. Long Branch, NJ 07764, USA. Tel.: +1-732-571-4482; fax: +1-732-263-5202. E-mail address: obaidat@monmouth.edu (M.S. Obaidat). 0020-0255/02/$ - see front matter � 2002 Elsevier Science Inc. All rights reserved. doi:10.1016/S0020-0255(02)00291-8 mail to: obaidat@monmouth.edu of implementing invisible watermarks. Basic watermarking concepts are dis- cussed in [1,2]. Researchers define a digital watermark as identification code carrying information (an author�s signature, a company�s log, etc.) about the copyright owner, the creator of the work, the authorized consumer and so on. It is permanently embedded into digital data for copyright protection and may be used for checking whether the data have been modified. Visible and invisible watermarking are the two categories of digital watermarking. The concept of the visible watermarking is very simple; it is analogous to stamping a mark on a paper. The data is said to be digitally stamped. An example of visible water- marking is seen in television channels when the station�s logo is visibly super- imposed in the corner of the screen. Invisible watermarking, on the other hand, is a far more complex concept. It is most often used to identify copyright data, like author, distributor, etc. On the other hand, image compression of digital images is the process of reducing the size of an image while retaining the highest possible visual quality. JPEG is the very well known ISO/ITU-T standard for image compression. Several modes are defined in JPEG [3]. It is released in late 1980s. JPEG 2000 is a new compression technology that achieves very high compression rate and maintains visual quality. The JPEG 2000 is issued now to become an Inter- national Standard (IS) [3,4]. This paper will give an answer to a common question about the robustness performance of the watermarking techniques against attacks of JPEG and JPEG 2000. An outlook on the future of digital image watermarking within JPEG 2000 will be given. Section 2 introduces an overview of watermarking concept. Section 3 provides a background about JPEG and JPEG 2000. Also, it compares both standards in terms of technology, performance, and appli- cations. Section 4 presents the experimental work and the results. Section 5 provides a look into the future of image coding and data security. Section 4 concludes this paper. 2. Watermarking concept and techniques 2.1. Watermarking applications and properties The two major applications for watermarking are protecting copyrights and authenticating photographs. The main reason for protecting copyrights is to prevent image piracy when the provider distributes the image on the Internet [2]. One method used to authenticate digital images is to embed a digital watermark that breaks or changes as the image is tampered with. This informs the authenticator that the image has been manipulated. Watermarking techniques that are intended to be widely used must satisfy several requirements. The type of application decides which watermarking 94 M.A. Suhail et al. / Information Sciences 151 (2003) 93–105 technique to be used. However, three requirements have been found to be common to most practical applications. These are robustness, invisibility and delectability. Some of watermarking requirements competes with each other. Also other requirements [1,2] may be significant. 2.2. Digital watermarking approaches There are two main generations of watermarking: first generation water- marking and second generation watermarking [5]. Both approaches can be achieved via spatial or transform techniques such as discrete cosine transform (DCT) and discrete wavelet transform (DWT). First generation watermarking (1GW) methods have been mainly focused on applying the watermarking on the entire image/video domain. However, this approach is not compatible with novel approaches for still image and video compression. JPEG 2000 and MPEG4/7 standards are the new techniques for image and video compression. They are region- or object-based, as can be seen in the compression process. By contrast, second generation watermarking (2GW) was developed in order to increase the robustness and invisibility and to overcome the 1GW weakness. The 2GW takes into account region, boundary and object characteristics. They give additional advantages in terms of detection and recovery from geometric attacks as com- pared to the 1GW [5]. This can be achieved by exploiting salient region or object features and characteristics of the image. Such watermarking methods may present additional advantages in terms of detection and recovery from geometric attacks [6]. They may be designed so that selective robustness to different classes of attacks is obtained. This will improve the watermark flexibility. 3. Joint photographic experts group standards This section will introduce some discussion about JPEG and JPEG 2000 standards. JPEG is the most widely used standard. However, JPEG 2000 is a new standard, which will appear in various applications in the near future. It represents the state-of-the-art in image coding. This section will give an ex- planation of the principles behind the algorithms used in both standards. 3.1. JPEG standard JPEG is the very well known ISO/ITU-T standard. Several modes are de- fined in JPEG. Baseline and lossless modes are the most popular ones [7,8]. It was released in late 1980s. The baseline mode supports lossy coding. The lossless mode is created for lossless coding only. In the baseline mode, the image is subdivided into pixels of size 8 � 8 (64 pixels). Then, each pixel of this subimage is level shifted by M.A. Suhail et al. / Information Sciences 151 (2003) 93–105 95 subtracting the quantity 2ð̂ð n)1). The DCT of each block is then computed. After that, the block is quantized and reordered using zig-zag pattern to form 1-D sequence. The AC coefficients of this 1-D sequence are coded using a variable-length code. The DC coefficient is coded relative to the DC coefficient of the previous subimage. An excellent background and examples of this are given in [8]. The transformation and normalization process produces a large number of zero-valued coefficients. These coefficients that remain after the normalization process will be discarded. Then, entropy coded with Huffman coding is performed. The quantization step size for each of the 64 DCT co- efficients is specified in a quantization table, which remains the same for all blocks. To decompress a JPEG compressed subimage, the decoder must first recreate the normalized transform coefficients that led to the compressed bit stream. Because a Huffman coded binary sequence is instantaneous and uniquely decodable, this step is easily accomplished using a lookup table. Any difference between the original and reconstructed subimage is the result of the lossy nature of the JPEG compression and decompression processes [3,8]. Fig. 1 shows a JPEG block diagram for lossy compression. However, the lossless mode is based on a completely different algorithm. It relies on a pre- dictive scheme which is based on the nearest three causal neighbors and seven different predictors are defined (the same one is used for all samples). The prediction error is entropy coded with Huffman coding. The other modes de- fined in JPEG provide variants of the previous two basic modes. This is like progressive bit streams and arithmetic entropy coding [8,9]. 3.2. JPEG 2000 standard JPEG 2000 was also developed by the International Standards Organization (ISO). It is the new image compression standard. The JPEG 2000 code handles both lossy and lossless compression using the same transform-based frame- work [4]. It is based on the DWT. The latter provides a number of benefits over Fig. 1. Block diagram of JPEG algorithm (lossy mode encoder). 96 M.A. Suhail et al. / Information Sciences 151 (2003) 93–105 the previous JPEG compression techniques that are based on DCT. DWT encodes the image in a continuous stream. So, this will avoid the tendency toward visible artifacts that sometimes result from DCT�s division of an image into discrete compression blocks [3,4]. Also, its model relies on scalar quanti- zation, context modeling, and arithmetic coding and post-compression rate allocation. The DWT used in JPEG 2000 is dyadic. It can be performed with a reversible filter (Le Gall (5,3) taps filter 9) [9], which provides for lossless coding. Also, a non-reversible filters (Daubechies (9,7) taps BI-orthogonal one 10) can be used for higher compression to do lossy compression but not lossless. The quantizer follows an embedded dead-zone scalar approach. It is independent for each subband. Each subband is divided into block ð64 � 64Þ. These subbands are entropy coded using context modeling and bit-plane arithmetic coding. The coded data is organized in layers. They are quality levels, using the post-compression rate allocation and output to the code- stream in packets [3]. The basic scheme of JPEG 2000 can be seen in Fig. 2. The above is part 1 description of JPEG 2000 standard, which defines the core system. Part 2 is still in preparation [9]. 3.2.1. JPEG 2000 functionality and features JPEG 2000 includes many-advanced features and supports a number of functionalities. Many of these functionalities are inherent from the algorithm itself [4,9,10]. These feature and functionalities are: • High compression ratio. • Lossy and lossless compression. • Progressive recovery by fidelity or resolution. • Visual (fixed and progressive) coding. • Good error resilience. • Arbitrarily shaped region of interest coding. • Random access to specific regions in an image. • Security • Multiple component images • Palletized images • It also can support images in width and height from 1 up to 232)1. 3.2.2. JPEG 2000 applications The following are examples of potential application that will benefit directly from JPEG 2000; see Table 1. Fig. 2. Basic encoding scheme of JPEG 2000. M.A. Suhail et al. / Information Sciences 151 (2003) 93–105 97 3.3. Comparison between JPEG and JPEG 2000 In this section, we will present some comparative experimental results to show the difference between JPEG and JPEG 2000. A boat image is com- pressed at very low bit rates using JPEG, at the same time, the image is compressed to the same degree using JPEG 2000, see Figs. 3–5. The com- pression ratio is 30:1. The images compressed using JPEG degrades signifi- cantly. Also, the images compressed using the JPEG 2000 algorithms and at the same compression rates do not suffer from the same degree of degradation as JPEG images. The noise artifacts, such as blockiness, that are clearly evident with JPEG are reduced with JPEG 2000. At very high compression rates the image content is easily recognizable with JPEG 2000 but not with JPEG. This shows that JPEG 2000 outperforms JPEG at higher compression ratios. Table 2 shows the main difference between the JPEG and JPEG 2000. Table 1 Examples of potential applications of JPEG 2000 Document imaging Digital photography Scanning Color facsimile Medical imaging Internet Web browsing E-Commerce Image archiving Remote sensing Digital library Mobile Fig. 3. Original boat image. 98 M.A. Suhail et al. / Information Sciences 151 (2003) 93–105 4. Experimental work 4.1. DCT-based watermarking algorithms The two-dimensional forward DCT kernel is used here. It is defined as Fig. 4. Compressed image using JPEG standard. Fig. 5. Compressed image using JPEG 2000 with same compression ratio as in Fig. 4. M.A. Suhail et al. / Information Sciences 151 (2003) 93–105 99 gðx; y; 0; 0Þ ¼ 1 N ; ð4aÞ gðx; y; u; vÞ ¼ 1 2N 3 ½cosð2x þ 1Þup�½cosð2y þ 1Þvp� ð4bÞ for x; y ¼ 0; 1; . . . ; N � 1, and u; v ¼ 1; 2; . . . ; N � 1 [8]. The DCT scheme relies on some of the ideas proposed by Cox et al. [11]. They propose a watermark that consists of a sequence of randomly generated real numbers. These num- bers have a normal distribution with zero mean and unity variance: W ¼ fw1; w2; . . . ; wNg: ð5Þ Then, the DCT of the whole image is computed. The DCT coefficients are chosen to be watermarked. After that, the watermark is added by modifying the DCT coefficients: C ¼ fc1; c2; . . . ; cNg: ð6Þ According to c0i ¼ ci þ aciwi; ð7Þ where i ¼ 1; 2; . . . ; N; and a ¼ 0:1. If we denote the original image by I0 and the watermarked possibly distorted image I�w, then, a possibly corrupted water- mark W � can be extracted. Reversing the embedding procedure can do this extracting. This is done using the inverse DCT. Table 2 Major differences between JPEG and JPEG 2000 Standard Technologies and features Applications JPEG by ISO/IEC DCT Internet imaging Perceptual quantization Digital photography Zig-zag reordering Image and video editing Huffman coding Arithmetic coding JPEG 2000 by ISO/IEC DWT Digital libraries New functionalities E-Commerce Reversible integer-to-integer and nonreversible real-to-real DWT Internet ROI Digital photography Error resilience Image and video editing Progression orders Printing Lossy to lossless in one system Medical imaging Better compression at low bit-rates Mobile Better at compound images and graphics (palletized) Color facsimile Satellite imaging Scanning Remote sensing 100 M.A. Suhail et al. / Information Sciences 151 (2003) 93–105 The watermark is embedded in subimages of the image. Therefore, the N � M image I is subdivided into pixels of size 16 � 16 (256 pixels). The DCT of the block is then computed. After that, the DCT coefficients are reordered into a zig-zag scan. This reordering is similar to the JPEG compression algo- rithm [8]. Then the coefficients in the zig-zag ordering of the DCT spectrum are selected. These selected coefficients are modified, according to (6), where ci is the original DCT coefficient, wi is the watermark coefficient and ci; w is the modified coefficient. To tune the watermark energy, the term a is used. The higher the a value, the more robust and visible the watermark. Finally, we need to reverse the above procedure to get our watermarked image. The modified DCT coefficients are reinserted in the zig-zag scan. Then, the inverse DCT is applied. Finally, the blocks are merged to obtain the watermarked image Iw. 4.2. DWT-based watermarking algorithms A DWT-based approach is used. The image (I) and watermark (W) are transformed into the DWT. The host image is transformed into three levels ðL ¼ 3Þ of DWT. Each of these levels (l:1 to 3) will produce a sequence of three levels detail images ðj ¼ 1; 2; 3Þ. Also, a gross approximation of the image at the coarsest resolution level will be generated at level three (l ¼ 3 and j ¼ 4) [12]. The resulting coefficients are then watermarked according to Iwj;lðx; yÞ ¼ Ij;lðx; yÞ þ bðf1; f2ÞWj;1ðx; yÞ; where Iðx; yÞ is the DWT of the host image, Iwðx; yÞ is the watermarked image, W is the watermark, l is the DWT resolution level and j is the DWT frequency orientation. The watermarking algorithm is adaptive by making use of human visual system (HVS) characteristics, which increase robustness and invisibility at the same time. The HVS ½bðf1;f2Þ� can be represented by [13,14]: bðf1; f2Þ ¼ 5:05e�0:178ðf1þf2Þ e0:1ðf1þf2Þ � � 1 � ; where, f1 andf2 are the spatial frequencies (cycles/visual angle). However, the watermark will be spatially localized at high-resolution levels of the host image. By this, the watermark will be more robust. At the end, the inverse DWT is applied to form the watermarked image. Fig. 3 shows the block diagram of the proposed method. 4.3. Watermarking detection The embedding watermark function makes small modifications to Iorig. For example, if W ¼ ðw1; w2; . . .Þ ¼ ð1; 0; 1; 1; 0; . . .Þ, the embedding operation may involve adding or subtracting a small quantity a from each pixel or sample of Iorig when wi is 1 or 0, respectively. During the second stage of the water- marking system, the detecting function D uses knowledge of W, and possibly Iorig, to extract a sequence W � from the signal R undergoing testing: M.A. Suhail et al. / Information Sciences 151 (2003) 93–105 101 DðR; IorigÞ ¼ W �: The signal R may be the watermarked signal Iw. It may be a distorted version of Iw resulting from attempts to remove the watermark, or it may be an unrelated signal. The extracted sequence W � is compared with the watermark W to de- termine whether R is watermarked. The comparison is usually based on a correlation measure q, and a threshold c0 used to make the binary decision (Z) on whether the signal is watermarked or not. To check the similarity between W (the embedded watermark), and W � (the extracted watermark), the corre- lation measure between them can be found using qðW ; W �Þ ¼ W W �ffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi W � W � p ; where W W � is the scalar product between these two vectors. However the decision function is ZðW �; W Þ ¼ 1; c P c0 0 otherwise; � where c is the value of the correlation and c0 is a threshold. A �1� indicates a watermark has been detected, while a �0� indicates that a watermark has not been detected. In other words, if W and W � are sufficiently correlated (greater than some threshold c0), the signal R has been verified to contain the water- mark, which confirms the author�s ownership rights to the signal. Otherwise, the owner of the watermark W has no rights over the signal R. The detection threshold c0 is considered empirically to be 0.1 in our experiments. This was decided based on the examination of the correlation of random sequences. 4.4. Results and discussion Figs. 6–9 show the experimental results. DCT- and wavelet-based water- marking algorithms described in this paper are implemented in Matlab envi- ronment. Different watermarked images are exposed to JPEG and JPEG 2000 for different compression ratios. The results are recorded in Table 3, which shows the correlation coefficients of the watermarking detector after com- pression using JPEG and JPEG 2000. The compression ratio was varied from 5 to 45. The recorded data is for wavelet and DCT techniques. Fig. 6 shows a DCT-based watermarking algorithms exposed to both JPEG and JPEG 2000. It is clear from the figure that the robustness of the water- marking algorithm against JPEG is better than JPEG 2000. On the other hand, the wavelet-based watermarking technique is less robust when exposed to JPEG attack than when exposed to JPEG 2000. This is shown in Fig. 7. On another display of the recorded data, Fig. 8 shows the results of JPEG attacks on DCT and wavelet watermarking techniques. The robustness of DCT algorithm compared to wavelet algorithms is better. However, the robustness of wavelet 102 M.A. Suhail et al. / Information Sciences 151 (2003) 93–105 Fig. 6. Comparison between JPEG and JPEG 2000 on a DCT-based watermarking technique. Fig. 7. Comparison between JPEG and JPEG 2000 on a wavelet-based watermarking technique. Fig. 8. Results of JPEG-DCT attacks on DCT and wavelet techniques. M.A. Suhail et al. / Information Sciences 151 (2003) 93–105 103 algorithm against JPEG 2000 outperform the same attack when applied on DCT algorithm. One can conclude from this that the compatibly issues between the watermarking algorithms and the compression standard may play an important role in the robustness of the watermarking. When the images are compressed at a very high compression ratio, the images compressed using JPEG-DCT degrades significantly. However, the images compressed using the JPEG 2000 algorithms and at the same compression rates do not suffer from the same degree of degra- dation. Also, the study needs more investigation on wide range of watermarking algorithms either using DCT or wavelet domain to generalize this result. 5. Concluding remarks The goal of this paper is to evaluate the performance of DCT and wavelet watermarking techniques against JPEG and JPEG 2000 attacks. The paper Fig. 9. Results of JPEG 2000 attacks on DCT and wavelet techniques. Table 3 Experimental results Tech DCT-based Wavelet-based CR JPEG-DCT JPEG 2K JPEG-DCT JPEG 2K 5 1 1 1 1 10 0.99 0.99 0.98 1 15 0.99 0.975 0.97 1 20 0.98 0.960 0.95 1 25 0.96 0.945 0.93 0.985 30 0.95 0.940 0.92 0.985 35 0.94 0.92 0.90 0.985 40 0.93 0.91 0.89 0.985 45 0.925 0.9 0.79 0.970 104 M.A. Suhail et al. / Information Sciences 151 (2003) 93–105 shows that there are compatibly issues between the robustness of the water- marking algorithms and the compression standards. More investigation is needed on a wide range of DCT- and wavelet-based watermarking algorithms to investigate the generalization of this conclusion. There was a discussion about how and whether watermarking should form part of the standard during the standardization process of JPEG 2000. The requirements regarding security have been identified in the framework of JPEG 2000. However, there has been neither in depth clarification nor a harmonized effort to address watermarking issues. The initial drafts of the JPEG 2000 standard did not mention the issue of watermarking. However, there is a plan to examine how watermarking might be best applied within the JPEG 2000. Therefore the potential is that watermarking technology will be used in con- junction with JPEG 2000. References [1] Busch, Christopher, Olfgang Funk, Stephen Wolthusen, Digital watermarking: from concepts to real-time video applications, IEEE Computer Graphics and Applications 19 (1) (1999) 25–35. [2] F. Hartung, M. Kutter, Multimedia watermarking techniques, Proceedings of the IEEE 87 (7) (1999) 1079–1106. [3] D. Santa-Cruz, T. Ebrahimi, A study of JPEG 2000 still image coding versus other standards, in: Proceeding of the X European Signal Processing Conference (EUSIPCO), Tampere, Finland, September 5–8, vol. 2, 2000, pp. 673–676. [4] ISO/IEC JTC 1/SC 29/WG 1, ISO/IEC FCD 15444-1: Information technology JPEG 2000 image coding system: core coding system, WG 1 N 1646, pp. 1–205, March 2000. Refer to: http://www.jpeg.org/FCD15444-1.htm. [5] http://www.tsi.enst.fr/~maitre/tatouage//icip2000.html. [6] M. Kutter, S.K. Bhattacharjee, T. Ebrahimi, Towards second generation watermarking schemes, in: Proceedings of the International Conference on Image Processing, 1999, ICIP 99, 1 (October) (1999) 320–323. [7] http://www.jpeg.org/. [8] R. Gonalez, P. Wintz, Digital Image Processing, second ed., Addison Wesley, Reading, MA, 1987. [9] D. Santa-Cruz, T. Ebrahimi, J. Askel€oof, M. Larsson, C. Christopoulos, JPEG 2000 still image coding versus other standards, in: Proceedings of the SPIE�s 45th annual meeting, Applications of Digital Image Processing XXIII, V 4115, San Diego, California, July 30–August 4, 2000, pp. 1–10. [10] D. Santa-Cruz, T. Ebrahimi, An analytical study of JPEG 2000 functionalities, in: Proceedings of the IEEE International Conference on Image Processing (ICIP), vol. 2, Vancouver, Canada, September 10–13, 2000, pp. 49–52. [11] Cox, J. Kilian, T. Leighton, T. Shamoon, Secure Spread Spectrum Watermarking for Multimedia. Tech. Rep. 95-10, NEC Research Institute, 1995. [12] C. Burrus, H. Guo, Introduction to Wavelets and Wavelet Transforms: A Primer, first ed., Prentice-Hall, Englewood Cliffs, NJ, 1998. [13] M. Levine, Vision in Man and Machine, McGraw-Hill, New York, 1985. [14] R. Clark, An introduction to JPEG 2000 and Watermarking, IEE Seminar on Secure Images & Image Authentication, 2000, pp. 3/1–3/6 http://ltswww.epfl.ch/~ebrahimi/JPEG2000Seminar. M.A. Suhail et al. / Information Sciences 151 (2003) 93–105 105 http://www.jpeg.org/FCD15444-1.htm http://www.tsi.enst.fr/~maitre/tatouage//icip2000.html http://www.jpeg.org/ http://ltswww.epfl.ch/~ebrahimi/JPEG2000Seminar A comparative study of digital watermarking in JPEG and JPEG 2000 environments Introduction Watermarking concept and techniques Watermarking applications and properties Digital watermarking approaches Joint photographic experts group standards JPEG standard JPEG 2000 standard JPEG 2000 functionality and features JPEG 2000 applications Comparison between JPEG and JPEG 2000 Experimental work DCT-based watermarking algorithms DWT-based watermarking algorithms Watermarking detection Results and discussion Concluding remarks References