Perfectly Secure Steganography Capacity, Error Exponents, and Code Constructions

A New Robust and Secure SteganographicSystem for Greyscale ImagesHesham A. El ZoukaComputer Engineering Department. Arab Academy for Science and Technology,Alexandria - Egypthelzouka@, hae@Abstract:The research work in this paper shows that the currently available stenographic methods might be quite easily detected by using sufficiently careful analysis of the transmitted data. In order to minimize the error introduced due to hiding foreign message carrier into the cover image, a robust and secure algorithm will be introduced, which might be used efficiently to protect the embedded message against attacks. The idea in the new approach depends on spreading the secret message over the cover image using both a pseudo random number generator and a hash function that drive one bit from each block of pixels in a random sequence manner. The intensity of each random pixel is predicted based on the average weighted sum value of the surrounding pixels. The difference between each random pixel is calculated causing a deviation error to be introduced. An error correction function will then be used to minimize the error introduced due to hiding foreign message carrier into the cover image and hence minimizing degradation of the embedded images.Keywords: Steganography; Information hiding; Secure communications; Stego-analysis1. IntroductionSteganography is the art of secret communication. Its purpose is to hide the very presence of communication as opposed to cryptography whose goal is to make communication unintelligible to those who do not posses the right keys. By embedding a secret message into a digital image, a stego-image is obtained. It is important that the stego-image does not contain any easily detectable artifacts due to message embedding that could be detected by eavesdropper. There are many different steganographic methods that have been overviewed and analysed by many researchers over the last few years, for e.g. hiding files in the least significant bits of digital images [1]. However, one common drawback of all current data embedding methods is the fact that the original image is distorted by some small amount of noise due the data embedding itself. This noise could reveal the existence of secret message and hence, weakness the security value of the covert channel. In this paper we investigated most of the steganography techniques, which have been proposed in the last few years. After analysing the drawbacks in the analysed systems, we proposed an approach, which maintains such noise in a way that makes the transmitted signal undetectable. In addition, we built the software programs that provide the simulation results including the histograms of both the cover images and the stego images, and the variance between them. Also, the simulation results for all equivalent substitution techniques, which covered in this paper, are used here for comparisons purpose.2. Previous WorkThere are many different steganographic methods that have been proposed over the last few years. Most of the simple techniques can be broken by careful analysis of statistical properties of the channel’s noise [2]. In substitution steganography techniques for example, one can observe that: these methods substitute insignificant parts of the image, e.g. the noise component of the cover with the secret message. These parts have specific statistical properties and the embedding process usually does not pay attention to them, and change the statistical profile of the cover significantly. A simple attack such as “laplace filtering” [3] can exploit this fact and detect the use of the steganographic system. In addition, these systems are extremely sensitive to cover modification and the attacker, who is not able to extract or prove the existence of a secure message can add a random noise to the transmitted cover or simply convert the image into another file format in attempt to destroy the secret message. With transform domain hiding techniques [4], more substantial processing is required to disable the readability of the embedded information. Meanwhile, most of these stenography systems have a vital drawback, which is that the system doesn’t discard image blocks where the desired relation of DCT coefficients [5] cannot be enforced without severely damaging the image data contained in this specific block. TCP/IP packet headers [6], [7] can also be reviewed easily. For e.g., firewall filters are set to test the validity of the source and destination IP addresses. Those filters can also be configured to catch packets that have information in supposed unused or reserved space. Based on the analysis of spread spectrum techniques [8], it can be observed that phase coding provide robustness against resembling of the carrier signal, but at the same time it has a low data transmission rate. These techniques have a problem with the absolute phase of all following segment that followed the first modified one, since all of them will have a change that could be noticeable to the attacker. Moreover, at the receiver end; the embedding process is reversed and image restoration technique such as adaptive wiener filter [9] is needed to estimate the original image.3. A Hybrid System using Greyscale ImagesThe proposed technique in this paper distorts the image insignificantly by making small modifications over a large number of pixels. Therefore, we will spread the secret message over a large area of cover image to produce a small modification on the carrier media. The new approach combines both cryptography and steganography to exchange secret messages in a way that it’s impossible to discover without the knowledge of the cover image and the secret key that have been used. Firstly, the parity bits from pixels c1, c2,..,c i are computed and encoded with the corresponding bits in the text file, which contains the secret message. The process is repeated for the whole stream of bits. If the computed parity bit c i and the secret bit m i are equal, then the encoded bit is zero and if the 2 input bits are different, then the output is one. Finally, the encoded bits are lined up to reconstruct the encoded file. Now, the file is ready to be encrypted and sent in any insecure channel to the receiver who had both, the secret key and a copy of the cover image which has been used. Therefore, the receiver of the encoded message will decipher the message using his secret key and the shared cover image.3.1 Spreading Secret Message over an Untraceable Cover AreaAfter the encoding process had taken place, the output file was encrypted and sent directly to the other party as a cipher file. However, instead of sending an encrypted stream of bits, an alternative scheme can be adopted by injecting the stream of bits back to the cover image with a probability of 50% of changing the LSB of embedded pixels in the image. Our goal here is to embed one bit of the secret message m i into one pixel of the cover image c i, where C is composed of all the pixels {c1, c2,….,c i}, and since L(m) < L(C) the rest of theimage can left unchanged. Moreover it's possible to select only some message pixels c i in a rather random manner according to a secret key and leave the other unchanged. Therefore, the idea depends on spreading the secret message over the cover image using both a pseudo random number generator and a hash function [10] that specifies one bit from each block of pixels randomly as follows:∑∈=ij j c LSB I P 2mod )()(If the parity bit of one cover block c i doesn’t match with the secret bit m i , the program will flip the LSB of one pixel in the block in a way to that makes p(Ii) equals to m i . Studying the properties of pixels surrounding the target pixel, we could invoke a statistical command that will fix up the number of 1's or 0's inside the chosen block in a way that conceals any statistical existence of a hidden information inside the stego image. This is done by studying the neighbouring pixels surrounding the chosen bit and changing its value to match the adjacent one in a way that prevent any statistical tracing. The Gibbon cover image in figure 1 provides special features and will be used in this research work as a test image.Figure1. Gibbon Cover Image [11]4.Implementation and DesignBefore communication starts, both sender and receiver have to agree on the location of the pixels c i . These pixels will be used as subjected pixels, from which the parity bits p(c i ) are computed. Hence, the computed parity bits are xored with the corresponding bits in the secret message file to produce the encoded message. A parity bit of a given pixel is computed from the following hash function:At this stage, the output file could be encrypted using standard encryption technique such as triple DES or IDEA [10]. The file is then compressed and sent directly to the other parity or injected back to the transmitted image with the encoded parity bits as mentioned before. The embedding process is preceded by leaving or altering the parity bit according to the value of the embedded bits. In the decoding process, a reverse process is launched to reconstruct the secret message.4.1 Error Correction Function This function is used to minimize the error due to hiding the foreign message carrier into cover images. The method is based on statistical analysis of images and it is very robust to changes that happen due to file format conversions or blurring filter such as a Gaussian convolution. The error correction algorithm proceeds by dividing the image into blocks of 4*4 pixels and chooses the blocks in sequence according to a given seed number. The intensity of each pixel x[i][j] within the cover area is predicted according to the valueof∑∈=ij j i s c p 2mod )(pixels in a specific neighborhood. Hence, the difference between the intensity of each pixel and its adjacent pixels is calculated as follows:Where x(i,j) represents the pixel coordinates in the selected cover region c. For each tested pixel in the block, the average weighted sum of the surrounding pixels is computed and compared with the target pixel.5. Simulation Results and Comparisons with Related SystemsThe simulation results showed that the text message could be embedded without any degradation of the image. Figure 2 shows the stego image after an edge detection algorithm has been run.Figure 2. Stego Image with Sharpen EdgesStudying the histogram of Laplace filter in figure 4 for the provided image, we notice that on average, the amount by which the image is modified is smaller than some known substitution embedding systems that we investigated.∇2 p(x,y) ∇2 p(x,y)Distance in pixels Distance in pixelsFigure 3. Laplace Filter of the Cover Image Figure 4. Laplace Filter of the new techniqueFor example comparing the distortion introduced by PGMStealth [12], and the new technique, we can clearly see that the new technique provides visibly fewer and less peaks than PGMStealth filtered histogram which has a wider band and many peaks clustered aroundzero as seen in figure 5.∑∑−−=−=41:41]][[161]][[]][[i j j i x j i in j i out∇2 p(x,y)Distance in pixelsFigure 5. Laplace Filter of PGMStealth Stego ImageIn hide and seek version 4.1 [13] the same cover image of a Gibbon was used to embed the same message. The original image of Gibbon is 256 x 256 pixels and 256 shades of grey. However, the resulting image was forced to 320 x 480 pixels as shown in figure 6. Instead of "stretching" the image to fit, large black areas were added to the image making it 320 x 480.Figure 6. Stego Image Using Hide and Seek Technique Black Wolfs Picture Encoder version 0.90a [14], uses the LSB method to embed the secret messages. The program consists of series of programs and it only works with 256 x 256 images with 256 grey levels. Despite the size restrictions, which are necessary for this program to run efficiently, the stego image is obviously distorted. The stego image was cropped and padded to a 320 x 200-pixel image. Figure 7illustrates this distortion when the text message is embedded.Figure 7. The Black Wolf Encoder Stego ImageOut of all the stenography techniques have been analyzed and tested, the one which gave the best results was the Snow steganography system [15] since there is no visual change in sight of the image. The “Whitespace” system as it is called in some literature hides the secret messages into an image in a way that nearly indistinguishable from normal random noise. However, using some other powerful steganalysis techniques such as the discrete Laplace filter, its possible to detect secret message in the image. Also giving the histogram of the repeated pixels within the greyscale Gibbon image, as shown in figure 8, one can notice that: the histogram of the original image has a limited peak height at certain locations, while the histogram of stego image has higher peaks at many points.Pixel Frequency Pixel FrequencyPixe l code Pixe l code Figure 8. (A) Histogram of the Cover Image (B) Histogram of Snow Stego ImageHowever, studying the statistical properties of neighbouring pixels in the proposed steganographic system, the embedding process gave us better results. That is because our new system has an algorithm that determines whether a candidate pixel can be used or not by checking the variance in luminosity of the neighbouring pixels. After looking to the pixel repetition histogram of the stego image in figure 9 and compare it with histogram of original image, it can be observed that there is very little difference between them and there are a few thin lines distributed in various part of the histogram.Pixel FrequencyPixe l codeFigure 9. Histogram of the New TechniqueFor some one who doesn’t hold a copy of the original cover image, it's hard for him to discover the differences between the normal and embedded one. For example, relative to the histogram of white noise, it's obviously seen that the white noise histogram contains many high peaks in the middle of the histogram, which could give the attacker the implication that the picture was subjected to some modifications.6. Robustness of the Proposed SystemOur first approach which does not change the cover is perfect in the sense above, but it does have the disadvantage of needing a key to be communicated separately. On the other hand our parity encoding scheme does introduce small changes to the image, but as shown in the results, it is very hard to detect. This method also achieves a very low-cost of embedding foreign data into the image. This benefit is due to the fact that the original function model isnon-linear. The changes introduced by this model are nonlinear and hence are not easily noticeable or detectable and almost impossible to model unless great sophistication is used by the attacker. Figure 5 showed the histograms of Laplace filtered greyscale image printed in one coordinate system for the public domain program PGMStealth, which hides secret information in the LSB of every cover-pixel, while Figure 4 showed the histogram of Laplace filtered for the proposed new model. Since the embedding process adds noise to the picture, which is statistically quite different from true random noise, the histogram of PGMStealth program differs extremely. On the other hand, the histogram of our model does not prove the existence of a secret message which minimises the chance that the embedded information will be detected. The histogram and analysis tests show that our method is more robust than any simple bit substitution model in common use. First the sender can choose which element should be modified and according to both the random number in the sequence and the calculated parity bit. Studying the error function and changing the seed or the width of the pixel on which the parity bit is calculated could change the cover statistics least.7.Conclusion and Future Work7.1ConclusionThe results of the proposed approach showed that the encoding process distorted the image insignificantly by making small modifications over large numbers of pixels. The algorithm divides the image into small blocks that are analysed for parity check values equivalent to the embedded bit. The technique was designed with the intent of maximising the quality of the stego image by the aid of error correction function that introduced extremely small modification to the cover images. Initial investigations showed that this modification was difficult to detect visually, and there is no tell-tale artifact could be picked up during the investigation process. In order to compare the provided approach with other established methods, many stego-analysis techniques were investigated and applied to the stego image. Testing our proposed steganographic algorithm's using an edge detection method, we found that the stego image does not show any artifacts and thus, it gives no indication that the image contains any hidden information. Comparing the Laplace filter histogram for the provided cover image with the one which contains the embedded message, we noticed that on average, the amount by which the image is modified is smaller than other known substitution steganographic systems that we investigated. Looking at the pixel repetition histogram of the stego image and comparing it with the histogram of the original image, it can be observed that there are only very small differences between them, and there are a few fine lines distributed over some parts of the histogram.7.2 Future WorkWe accepted that it is hard to build a model that adapts to all of the parameters needed to define the statistical properties of a steganographic model. We noticed also that there is always a trade-off between robustness and data rate, which may prevent any embedding process from meeting the needs of all applications. However for the future work of this work we recommend that the cryptography methods should be taken more seriously into account in order to design a more successful steganographic system and in an attempt to provide a secure function to the steganography process. In addition to make the communication even more secure, we recommend that the secret message should be compressed or encoded before the encryption process takes place. This is important because in this way we will minimise the amount of information that is sent, and hence minimizing the chance of degrading the image. We recommend also using efficient error correction coders and programs that could run in parallel with the embedding process in order to detect any suspect pixel with distinctive features that could lead the attacker for further investigation.8.References[1]Asha, A. Information hiding - The Art of Steganography, GSEC practical. SANSInstitute, 2005.[2]Elke ; Fraz. Steganography preserving statistical properties, proceeding of the 5thinternationally Workshop on information Hiding, Noordwijkerhout, The Netherlands, October 2002, LNCS 2578, pp. 278-294, Springer 2003.[3]Petitcolas, F. A. P., R. J. Anderson ; Kuhn, M. G. Information Hiding- A Survey,Proceedings of the IEEE, vol. 87, no.7, Jul. 1999, pp. 1062-1078.[4]Stefan Katzenbeisser ; F. A.P. Petitcolas ;. Information Hiding Techniques forSteganography and Digital Watermarking . Artech House ; ISBN: 1580530354, 2000. [5]Andres, A.; Huertas ; Gerard Medioni. Detection of Intensity Changes with SubpixelAccuracy Using Laplacian- Gaussian Masks, IEEE Transactions on pattern analysis and machine intelligence, vol. pami-8, no. 5, pp. 651-664, September 1986.[6]Anderson, R. J.; Needham, R. M. ; Shamir, A.. The Steganographic File System, inProceedings of the Second International Workshop on Information Hiding, vol. 1525 of Lecture Notes in Computer Science, Springer, 1998, pp. 73-82.[7]Piscitello, D. and Chapin, A, L., Open Systems Networking: TCP/IP and OSI,Addison-Wesley, Reading, Mass., 1997[8]Petitcolas, J., Gabriella, C. A Bayesian Approach to Spread Spectrum WatermarkDetection and Secure copyright protection for Digital Libraries. IEEE Conference on Computer vision and pattern recognition (CVPR'99, Colorado, USA, June , 1999.[9]Westfield, A. and Pfitzmann, A., “Attacks on seteganographic Systems”. In: Proc. 3rdInformation Hiding Workshop, Dresden, Germany, September (1999) 61-75.[10]Mohammad Peyravian ; Nev Zunic. Hash-Based Encryption System. Computers &Security Vol. 18, No.4, pp.345-350 , 2003.[11]USC – SIPI. Image database, Signal & image processing Institute. University ofSouthern California. /publications.html[12]Kwan, M. How gifshuffle works, Technical report, Helsinki University of Technology,June 2004.[13] Money, C. Hide and Seek. <ftp:///pub/cypherpunks/st~ganography/hdsk41b.zip> ,1994-1997.[14]Black Wolf's Picture Encoder vO.90B,ftp:///pub/cypherpunks/Steganography[15]. Arachelian, R. White Noise Storm. ftp:///pub/cypherpunks/steganography/wns210.zip.[16]Cover, T.M.; Thomas J. A. Elements of Information Theory. New York, Chichester:John Wiley & Sons, 1991.。

小学上册英语第一单元全练全测(含答案)英语试题一、综合题(本题有100小题，每小题1分，共100分.每小题不选、错误，均不给分)1.Galaxies can collide and merge with ______.2.The scientific study of matter and its changes is called _______.3.What is the name of the famous mountain range in Europe?A. RockiesB. HimalayasC. AlpsD. Andes4.An endothermic reaction requires _____ from its surroundings.5.What do we call a shape with four equal sides?A. RectangleB. SquareC. TriangleD. Pentagon答案:B6.His favorite sport is ________.7.I like to help my dad in the ____.8.The _______ of an object can affect its movement.9.My teacher helps me with ____.10.Which of these colors is made by mixing blue and yellow?A. RedB. GreenC. PurpleD. Orange答案:B11. A _____ (植物网络) can connect enthusiasts globally.12.An ecosystem includes living and non-living ______.13.I enjoy playing with my ______ (玩具车) in the living room. It goes ______ (快).14. A pendulum swings back and ______ (forth).15.Which language is spoken in Brazil?A. SpanishB. PortugueseC. FrenchD. Italian答案:B16.The garden is ________ (美丽).17.中国的________ (legends) 经常包含神话与历史交织的故事。

STEGANOGRAPHY WITH TWO JPEGS OF THE SAME SCENE TomášDenemark,Student Member,IEEE,and Jessica Fridrich,Fellow,IEEEBinghamton UniversityDepartment of ECEBinghamton,NYABSTRACTIt is widely recognized that incorporating side-information at the sender can significantly improve steganographic security in practice.Currently,most side-informed schemes for digital images utilize a high quality“precover”image that is subsequently processed and then jointly quantized and embedded with a secret. In this paper,we investigate an alternative form of side-information in the form of two JPEG images of the same scene.The second JPEG image is used to determine the preferred polarity of embedding changes and to modu-late their costs.Tests on real imagery show a very sig-nificant improvement in empirical security with respect to steganography utilizing a single JPEG image.Index Terms—Steganography,side-information, precover,security,steganalysis,JPEG,UNIWARD1.INTRODUCTION Steganography is a private communication tool in which secrets are embedded in cover objects to hide the pres-ence of the message itself.In side-informed steganogra-phy,the sender utilizes information that is unavailable to the steganalyst(and the recipient)to improve secu-rity.For example,the embedding can take place while processing(compressing)a higher quality representation of the cover image called precover[1].The most common example of this type of steganography uses non-rounded DCT coefficients when saving an uncompressed image as JPEG[2,3,4,5,6,7,8].Most consumer-end electronic devices,such as cell phones,tablets,and low-end digital cameras,however, can save images only in the JPEG format and thus do not The work on this paper was partially supported by NSF grant No.1561446and by Air Force Office of Scientific Research under the research grant number FA9950-12-1-0124.The ern-ment is authorized to reproduce and distribute reprints for Gov-ernmental purposes notwithstanding any copyright notation there on.The views and conclusions contained herein are those of the authors and should not be interpreted as necessarily representing the official policies,either expressed or implied of AFOSR or the ernment.give the user access to the uncompressed image.In this case,one can utilize a different type of side-information –multiple JPEG images of the same scene.This re-search direction has not been developed much mostly due to the difficulty of acquiring the required imagery and modeling the differences between acquisitions.The first work on this topic includes[9,10,11]where the au-thors made multiple scans of the same printed image and then modeled the differences between scans and among neighboring pixels.Unfortunately,this requires acquir-ing a potentially large number of scans,which makes this approach rather labor intensive.Moreover,differences in the movement of the scanner head between scans lead to misalignment that complicates using this type of side-information properly.In this paper,we work with multiple images acquired in the JPEG format as we expect quantized DCT coef-ficients to be naturally more robust to small imperfec-tions during acquisition.Since our intention is to design a practical method,we avoid the difficult and poten-tially extremely time consuming task of modeling the differences between acquisitions[9,10,11]and make the approach work well even when mere two images are avail-able to the sender.In particular,we modulate the em-bedding costs of J-UNIWARD[7]based on the preferred direction inferred from two JPEG images of the same scene.The method is tested on real-life multiple expo-sures obtained using a tripod-mounted digital camera. The proposed embedding with two JPEG images is sub-stantially more secure than when only a single JPEG is available to the steganographer.In the next section,we review existing side-informed steganography with a high quality precover.The new steganographic method for embedding with two JPEGs is detailed in Section3.In Section4,we describe and analyze the image source used for experiments in Sec-tion5.The same section contains a comparison with J-UNIWARD and SI-UNIWARD as well as a study of how the security gain due to the second JPEG changes with differences between exposures.The paper is con-cluded in Section6.1020Burst index kM S EFig.1.MSE between z (1)and z (k ),k =2,...,7from each burst averaged over all 9,310bursts from BURST-base.See the main text for notation.2.STEGANOGRAPHY WITH PRECOVER Virtually all modern embedding schemes for JPEG im-ages,whether or not they use side-information,are im-plemented within the paradigm of distortion minimiza-tion.The sender first defines the cost of modifying each cover element (DCT coefficient)and then embeds the payload so that the expected value of the total cost is as small as possible.Syndrome-trellis codes[12]can be used to implement the embedding in practice.For simplicity,we work with 8-bit M ×N grayscale images with M and N multiples of 8.The non-rounded and rounded values of DCT coefficients from (u,v )th8×8block will be denoted c (u,v )ij ∈R and x (u,v )ij ∈{−1023,...,1024},1≤i,j ≤8,1≤u ≤M/8,1≤v ≤N/8,respectively.The cost of changing x (u,v )ij by 1and −1is ρ(u,v )ij (1)and ρ(u,v )ij (−1),respectively.When the symbols x ij ,c ij ,ρij are used without super-scripts,the range of i,j spans the entire M ×N image.The total cost (distortion)of embedding is D (x ,y )= x ij =y ij ρij (y ij −x ij ),where y ij ∈{x ij −1,x ij ,x ij +1}is the stego image.An embedding scheme operating at the rate–distortion bound (with minimal D )would embed a payload of R bits by modifying the DCT coefficients with probabilities:β±ij=P {y ij =x ij ±1}=e −λρij (±1)1+e −λρij (1)+e −λρij (−1),(1)where λis determined from the payload constraint R = ij h 3(β+ij ,β−ij ),with h 3(x,y )=−x log 2x −y log 2y −(1−x −y )log 2(1−x −y )the ternary entropy func-tion.One of the most secure schemes for JPEG im-ages called J-UNIWARD [7]computes the costs from the decompressed JPEG image.The costs are symmetric ρij (1)=ρij (−1)for all i,j .While it is currently an open problem how to use side-information (c ij )in an optimal fashion for embed-ding [13],numerous heuristic schemes have been pro-0.20.40.6Quality factor QM o d u l a t i o n m (Q )Fig.2.Modulation factor m (Q )as a function of the JPEG quality factor Q for images from BURSTbase.posed in the past [3,5,6,7,8,4].In a nut shell,these schemes use the rounding error e ij =c ij −x ij ,−1/2≤e ij ≤1/2,to modulate the embedding costs ρij by 1−2|e ij |∈[0,1].In SI-UNIWARD [7],for example,the costs are:ρij (sign(e ij ))=(1−2|e ij |)ρ(J)ij (2)ρij (−sign(e ij ))=C wet ,(3)where ρ(J)ij are J-UNIWARD costs and C wet is some large number (“wet cost”).In [8],a ternary version of SI-UNIWARD was studied where the authors argued that,as the rounding error e ij becomes small,the embed-ding rule should be allowed to change the coefficient both ways.This ternary version of SI-UNIWARD usesρij (−sign(e ij ))=ρ(J)ij instead of (3).3.STEGANOGRAPHY WITH TWO JPEGS Let us consider a situation when the sender acquires twoJPEG images of the same scene,x (1)ij and x (2)ij ,while pro-nouncing,e.g.,the first image as cover JPEG and consid-ering x (2)ij as side-information.The value x (2)ij can only be useful to the sender when x (2)ij =x (1)ij ,which happens increasingly more often with smaller quantization steps (larger JPEG quality).This type of side-information isdifferent from the non-rounded values c (1)ij .In particu-lar,it informs the sender more about the direction along which the costs should be modulated and less about themagnitude of the rounding error e (1)ij =c (1)ij −x (1)ij .The proposed embedding scheme,which we call J2-UNIWARD,uses J-UNIWARD costs [7]when x (1)ij =x (2)ij and modulated costs otherwise:ρij (sign(x (2)ij −x (1)ij ))= ρ(J)ij if x (1)ij =x (2)ij m (Q )ρ(J)ij if x (1)ij =x (2)ij ,(4)with the modulation factors m (Q )∈[0,1]to be de-termined experimentally for each JPEG quality factor 1≤Q ≤100.0.10.20.30.40.5Quality factor QD e t e c t i o n e r r o r P E0.10.20.30.40.5Quality factor QD e t e c t i o n e r r o r P EFig.3.Empirical security of J2-UNIWARD as a function of the JPEG quality factor Q with the merger of GFR,SRM,and ccJRM features.Left:Comparison with previous art for R =0.2bpnzac.Right:P E for R ∈{0.1,0.2,0.3,0.4,0.5}bpnzac,embedding simulated at rate–distortion bound.4.THE BURSTBASE DATASETIt is generally difficult to acquire two images of the ex-act same scene because the camera position may slightly change between the exposures even when mounted on a tripod due to vibrations caused by the shutter.Another potential source of differences is slightly varying exposure time and changing light conditions between exposures.To eliminate possible impact of flicker of artificial lights,all images were acquired in daylight,both indoor and outdoor,and without a flash.Canon 6D,a DSLR camera with a full-frame 20MP CMOS sensor,set to a fixed ISO of 200was used in a burst mode.The shutter was operated using a cable release with a two-second self-timer to further minimize vibrations due to operating the camera.To prevent the camera from changing the set-tings during the burst,it was used in manual mode.All images were acquired in the RAW CR2format and then exported from Lightroom 5.7to 24-bit TIFF format with no other processing applied.A total of 133bursts were acquired,each containing 7images.To increase the number of images for experi-ments,the 5472×3648TIFF images were cropped into 10×7equidistantly positioned tiles with 512×512pixels.This required a slight overlap between neighboring tiles (7pixels horizontally and 35pixels vertically).These 70×133=9,310smaller images were then converted to grayscale in Matlab using ’rgb2gray ’and saved in a lossless raster format to facilitate experiments with a range of JPEG quality factors.We call this database of 7×9,310uncompressed grayscale images ’BURST-base’.The images were further JPEG compressed with different quality factors for all experiments in this paper.For each pair of different images from each burst,we computed the mean square error (MSE)between them and then selected the pair with the smallest MSE,ran-domly denoting one as z (1)ij and the other z (2)ij .The re-maining five images from the burst were denoted z (k )ij ,k =3,...,7,so that the MSE between z (1)ij and z (k )ij forms a non-decreasing sequence in k .Next,we ana-lyzed images from BURSTbase sorted in this manner to determine how much the differences between the imagesare due to acquisition noise.The MSE between z (1)ij and z (k )ij ,k =2,...,7,averaged over the BURSTbase is plot-ted in Figure 1.For the closest pair,MSE(z (1),z (2))≈5,which would correspond to σ2a =5if the differences weresolely due to AWG noise with variance σ2a .This closely matches the variance estimated from a single image of content-less scenes,such as blue sky.This reasoning in-dicates that z (2)and z (3)are on average reasonably well aligned with z (1)while z (k ),k ≥4,are affected by small spatial shifts.5.EXPERIMENTSIn this section,the security of J2-UNIWARD is studied across a range of quality factors and payloads and con-trasted with the same scheme utilizing a single JPEG im-age and a scheme utilizing a single high-quality precover.We also investigate the security boost of the second ex-posure with increased differences between exposures.The modulation factor m (Q )(4)was determined for each quality factor Q to minimize P E =min P F A (P MD +P F A )/2,the minimal total probability of error on the training set,where P MD ,P F A are missed-detection and false-alarm rates of a detector implemented using the0.20.4Quality factor QP EFig.4.Security of J2-UNIWARD when k th closest image from each burst is used as side-information,0.4bpnzac.ensemble classifier [14]with GFR (Gabor Filter Resid-ual)features [15]when splitting BURSTbase into equally sized training and testing set.The GFR features were selected because they are known to be highly effec-tive against modern JPEG steganography,including J-UNIWARD and SI-UNIWARD.The optimal modulation factor determined experimentally and shown in Figure 2can be well approximated by a ramp function:m (Q )=max {0.075,0.02167×Q −1.55}.(5)The ramp function can be justified when adopting a generalized Gaussian model of precover JPEG DCT co-efficients and an AWG model of the acquisition noise.This argument is omitted here due to space limitations and will appear in the journal version of this paper [16].The largest observed loss in P E due to replacing optimal values of m (Q )with the ramp function was about 0.01.Because the feedback from detection with GFR fea-tures was used to design the embedding scheme,all detectors in this section were implemented with a diverse feature set that is a merger of the spatial rich model (SRM)[17],Cartesian-calibrate JPEG Rich Model (ccJRM)[18],and GFR to make sure the em-bedding does not have a fatal weakness with respect to older features.Figure 3left shows P E averaged over ten splits of BURSTbase into training and testing sets (denoted P E )as a function of the JPEG quality fac-tor for payload 0.2bpnzac together with the results forJ-UNIWARD (with x (1)k as covers)and SI-UNIWARD(with c (1)k as side-information).The side-information in the form of two JPEG images significantly increases em-pirical security w.r.t.embedding with a single JPEG (J-UNIWARD)especially for large payloads and small qual-ity factors.The empirical security is however not bet-ter than when non-rounded DCT coefficients are used as side-information (SI-UNIWARD).Figure 3right shows the detection error as a function of the quality factor for five payloads.Since the statistical spread of P E over the splits ranged between 0.0010−0.0075,we do not show the error bars in the figure as it would be hard to discern them visually.To assess how sensitive J2-UNIWARD is w.r.t.smalldifferences between exposures,we implemented thescheme with x (1)ij as cover and x (k )ij ,k =3,...,7as side-information,essentially using the second closest (k =3),the third closest (k =4),etc.,image instead of the clos-est image.As apparent from Figure 1,with increasing k ,the boost should start decreasing.Figure 4shows P E as a function of the quality factor across k =2,...,7to-gether with the value of J-UNIWARD (JUNI).While the gain of the second image indeed decreases with increased MSE,this decrease is gradual and rather small for higher quality factors.This experiment proves that the sec-ond exposure provides useful side-information even when small spatial shifts are present opening thus the possibil-ity to improve steganography even when the multiple exposures are acquired with a hand-held camera rather than mounted on a tripod.This possibility is left as part of future research.6.CONCLUSIONSWe study steganography with side-information at the sender in the form of a second JPEG image of the same scene that is used to infer the preferred direction of steganographic embedding changes.This information is incorporated into the embedding algorithm by decreas-ing (modulating)the embedding costs of such preferred changes.Experiments with real multiple acquisitions show a quite significant increase in empirical security of with respect to steganography with a single cover image (J-UNIWARD).The boost in empirical security appears fairly insensitive to small differences between the two ac-quisitions,which makes the proposed method practical and opens up the possibility to use multiple exposures obtained using a hand-held camera or acquiring multiple exposures from short video clips.Further improvement is likely possible by optimizing the embedding cost modulation for each DCT mode,quantization step,and the average grayscale of the DCT block because the acquisition noise amplitude depends on luminance.Finally,we plan to study how to utilize more than two (quantized and unquantized)acquisitions.7.REFERENCES[1]A.D.Ker,“A fusion of maximal likelihood andstructural steganalysis,”in Information Hiding, 9th International Workshop,T.Furon, F.Cayre,G.Doërr,and P.Bas,Eds.,Saint Malo,France,June11–13,2007,vol.4567of LNCS,pp.204–219, Springer-Verlag,Berlin.[2]J.Fridrich,M.Goljan,and D.Soukal,“Perturbedquantization steganography using wet paper codes,”in Proceedings of the6th ACM Multimedia&Secu-rity Workshop,J.Dittmann and J.Fridrich,Eds., Magdeburg,Germany,September20–21,2004,pp.4–15.[3]Y.Kim,Z.Duric,and D.Richards,“Modifiedmatrix encoding technique for minimal distortion steganography,”in Information Hiding,8th Inter-national Workshop,J.L.Camenisch,C.S.Collberg, N.F.Johnson,and P.Sallee,Eds.,Alexandria,VA, July10–12,2006,vol.4437of LNCS,pp.314–327, Springer-Verlag,New York.[4]V.Sachnev,H.J.Kim,and R.Zhang,“Lessdetectable JPEG steganography method based on heuristic optimization and BCH syndrome cod-ing,”in Proceedings of the11th ACM Multimedia &Security Workshop,J.Dittmann,S.Craver,and J.Fridrich,Eds.,Princeton,NJ,September7–8, 2009,pp.131–140.[5]F.Huang,J.Huang,and Y.-Q.Shi,“New chan-nel selection rule for JPEG steganography,”IEEE Transactions on Information Forensics and Secu-rity,vol.7,no.4,pp.1181–1191,August2012. [6]L.Guo,J.Ni,and Y.Q.Shi,“Uniform embeddingfor efficient JPEG steganography,”IEEE Transac-tions on Information Forensics and Security,vol.9, no.5,pp.814–825,May2014.[7]V.Holub,J.Fridrich,and T.Denemark,“Universaldistortion design for steganography in an arbitrary domain,”EURASIP Journal on Information Secu-rity,Special Issue on Revised Selected Papers of the 1st ACM IH and MMS Workshop,vol.2014:1,2014.[8]T.Denemark and J.Fridrich,“Side-informedsteganography with additive distortion,”in IEEE International Workshop on Information Forensics and Security,Rome,Italy,November16–192015.[9]E.Franz,“Steganography preserving statisticalproperties,”in Information Hiding,5th Interna-tional Workshop,F.A.P.Petitcolas,Ed.,Noordwi-jkerhout,The Netherlands,October7–9,2002,vol.2578of LNCS,pp.278–294,Springer-Verlag,New York.[10]E.Franz and A.Schneidewind,“Pre-processing foradding noise steganography,”in Information Hid-ing,7th International Workshop,M.Barni,J.Her-rera,S.Katzenbeisser,and F.Pérez-González,Eds., Barcelona,Spain,June6–8,2005,vol.3727of LNCS,pp.189–203,Springer-Verlag,Berlin. [11]E.Franz,“Embedding considering dependenciesbetween pixels,”in Proceedings SPIE,Electronic Imaging,Security,Forensics,Steganography,and Watermarking of Multimedia Contents X, E.J.Delp,P.W.Wong,J.Dittmann,and N.D.Memon, Eds.,San Jose,CA,January27–31,2008,vol.6819, pp.D1–12.[12]T.Filler,J.Judas,and J.Fridrich,“Minimizing ad-ditive distortion in steganography using syndrome-trellis codes,”IEEE Transactions on Information Forensics and Security,vol.6,no.3,pp.920–935, September2011.[13]J.Fridrich,“On the role of side-information insteganography in empirical covers,”in Proceed-ings SPIE,Electronic Imaging,Media Watermark-ing,Security,and Forensics2013,A.Alattar,N.D.Memon,and C.Heitzenrater,Eds.,San Francisco, CA,February5–7,2013,vol.8665,pp.0I1–11. [14]J.Kodovský,J.Fridrich,and V.Holub,“Ensembleclassifiers for steganalysis of digital media,”IEEE Transactions on Information Forensics and Secu-rity,vol.7,no.2,pp.432–444,April2012.[15]X.Song,F.Liu,C.Yang,X.Luo,and Y.Zhang,“Steganalysis of adaptive JPEG steganography us-ing2D Gaborfilters,”in3rd ACM IH&MMSec.Workshop,esana,J.Fridrich,and A.Alat-tar,Eds.,Portland,Oregon,June17–19,2015. [16]T.Denemark and J.Fridrich,“Steganography withmultiple JPEG images of the same scene,”IEEE Transactions on Information Forensics and Secu-rity,2016,in preparation.[17]J.Fridrich and J.Kodovský,“Rich models for ste-ganalysis of digital images,”IEEE Transactions on Information Forensics and Security,vol.7,no.3, pp.868–882,June2011.[18]J.Kodovskýand J.Fridrich,“Steganalysis of JPEGimages using rich models,”in Proceedings SPIE, Electronic Imaging,Media Watermarking,Security, and Forensics2012,A.Alattar,N.D.Memon,andE.J.Delp,Eds.,San Francisco,CA,January23–26,2012,vol.8303,pp.0A1–13.。

Today I’ll tell you something about Information Hiding.As we all know, with the rapid development of communication technology, information security is becoming an important research topic. Information hiding is an emerging research hotspot in the field of information security, involving a number of disciplines such as perceptual science, information theory, cryptography and so on, and covers the research direction of signal processing, spread spectrum communication and so on. It’s emergence and development to expand a new field for information security research and application.Today I will introduce the following four aspects:Information hiding is a way to hide secret information into the general non-secret digital media carrier, such as images, sounds, videos, and documents, so that opponents can not find it.In short, information hiding is to hide the important information into the ordinary digital media.It is generally considered that It has the following three characteristics：①IntangibilityDo not affect the subjective quality of the original carrier and not easy to be detected by observer，except the watermarking。

Secure Spread SpectrumWatermarking for MultimediaIngemar J.Cox,Senior Member,IEEE,Joe Kilian,F.Thomson Leighton,and Talal Shamoon,Member,IEEEAbstract—This paper presents a secure(tamper-resistant)al-gorithm for watermarking images,and a methodology for digital watermarking that may be generalized to audio,video,and multimedia data.We advocate that a watermark should be constructed as an independent and identically distributed(i.i.d.) Gaussian random vector that is imperceptibly inserted in a spread-spectrum-like fashion into the perceptually most signifi-cant spectral components of the data.We argue that insertion of a watermark under this regime makes the watermark robust to signal processing operations(such as lossy compression,filtering, digital-analog and analog-digital conversion,requantization,etc.), and common geometric transformations(such as cropping,scal-ing,translation,and rotation)provided that the original image is available and that it can be succesfully registered against the transformed watermarked image.In these cases,the watermark detector unambiguously identifies the owner.Further,the use of Gaussian noise,ensures strong resilience to multiple-document,or collusional,attacks.Experimental results are provided to support these claims,along with an exposition of pending open problems. Index Terms—Intellectual property,fingerprinting,multime-dia,security,steganography,watermarking.I.I NTRODUCTIONT HE PROLIFERATION of digitized media(audio,image, and video)is creating a pressing need for copyright enforcement schemes that protect copyright ownership.Con-ventional cryptographic systems permit only valid keyholders access to encrypted data,but once such data is decrypted there is no way to track its reproduction or retransmission. Therefore,conventional cryptography provides little protection against data piracy,in which a publisher is confronted with unauthorized reproduction of information.A digital watermark is intended to complement cryptographic processes.It is a visible,or preferably invisible,identification code that is permanently embedded in the data and remains present withinManuscript received January14,1996;revised January24,1997.Portions of this work were reprinted,with permission,from the Proceedings of the IEEE Conference on Image Processing,1996,and from the Proceedings of the First International Conference on Data Hiding(Springer-Verlag,1996). The associate editor coordinating the reivew of this manuscript and approving it for publication was Prof.Sarah Rajala.I.J.Cox and J.Kilian are with NEC Research Institute,Princeton,NJ08540 USA(e-mail:ingemar@;joe@).F.T.Leighton is with the Mathematics Department and Laboratory for Computer Science,The Massachusetts Institute of Technology,Cambridge, MA02139USA(e-mail:ftl@).T.Shamoon is with InterTrust STAR Laboratory,Sunnyvale,CA94086 USA(e-mail:talal@).Publisher Item Identifier S1057-7149(97)08460-1.the data after any decryption process.In the context of this work,data refers to audio(speech and music),images (photographs and graphics),and video(movies).It does not include ASCII representations of text,but does include text represented as an image.Many of the properties of the scheme presented in this work may be adapted to accommodate audio and video implementations,but the algorithms here specifically apply to images.A simple example of a digital watermark would be a visible“seal”placed over an image to identify the copyright owner(e.g.,[2]).A visible watermark is limited in many ways.It marrs the imagefidelity and is susceptible to attack through direct image processing.A watermark may contain additional information,including the identity of the purchaser of a particular copy of the material.In order to be effective,a watermark should have the characteristics outlined below. Unobtrusiveness:The watermark should be perceptually invisible,or its presence should not interfere with the work being protected.Robustness:The watermark must be difficult(hopefully impossible)to remove.If only partial knowledge is available (for example,the exact location of the watermark in an image is unknown),then attempts to remove or destroy a watermark should result in severe degradation infidelity before the watermark is lost.In particular,the watermark should be robust in the following areas.•Common signal processing:The watermark should still be retrievable even if common signal processing oper-ations are applied to the data.These include,digital-to-analog and analog-to-digital conversion,resampling, requantization(including dithering and recompression), and common signal enhancements to image contrast and color,or audio bass and treble,for example.•Common geometric distortions(image and video data): Watermarks in image and video data should also be im-mune from geometric image operations such as rotation, translation,cropping and scaling.•Subterfuge attacks(collusion and forgery):In addition, the watermark should be robust to collusion by multiple individuals who each possess a watermarked copy of the data.That is,the watermark should be robust to combining copies of the same data set to destroy the watermarks.Further,if a digital watermark is to be used in litigation,it must be impossible for colluders to combine their images to generate a different valid watermark with the intention of framing a third party.1057–7149/97$10.00©1997IEEEUniversality:The same digital watermarking algorithm should apply to all three media under consideration.This is potentially helpful in the watermarking of multimedia products.Also,this feature is conducive to implementation of audio and image/video watermarking algorithms on common hardware.Unambiguousness:Retrieval of the watermark should un-ambiguously identify the owner.Furthermore,the accuracy of owner identification should degrade gracefully in the face of attack.There are two parts to building a strong watermark:the watermark structure and the insertion strategy.In order for a watermark to be robust and secure,these two components must be designed correctly.We provide two key insights that make our watermark both robust and secure:We argue that the watermark be placed explicitly in the perceptually most significant components of the data,and that the watermark be composed of random numbers drawn from a Gaussian distribution.The stipulation that the watermark be placed in the per-ceptually significant components means that an attacker must target the fundamental structural components of the data, thereby heightening the chances offidelity degradation.While this strategy may seem counterintuitive from the point of view of steganography(how can these components hide any signal?),we discovered that the significant components have a perceptual capacity that allows watermark insertion without perceptual degradation.Further,most processing techniques applied to media data tend to leave the perceptually significant components intact.While one may choose from a variety of such components,in this paper,we focus on the perceptually significant spectral components of the data.This simultane-ously yields high perceptual capacity and achieves a uniform spread of watermark energy in the pixel domain.The principle underlying our watermark structuring strategy is that the mark be constructed from independent,identically distributed(i.i.d.)samples drawn from a Gaussian distribu-tion.Once the significant components are located,Gaussian noise is injected therein.The choice of this distribution gives resilient performance against collusion attacks.The Gaussian watermark also gives our scheme strong performance in the face of quantization,and may be structured to provide low false positive and false negative detection.This is discussed below,and elaborated on in[13].Finally,note that the techniques presented herein do not provide proof of content ownership on their own.The focus of this paper are algorithms that insert messages into content in an extremely secure and robust fashion.Nothing prevents someone from inserting another message and claiming owner-ship.However,it is possible to couple our methods with strong authentication and other cryptographic techniques in order to provide complete,secure and robust owner identification and authentication.Section III begins with a discussion of how common sig-nal transformations,such as compression,quantization,and manipulation,affect the frequency spectrum of a signal.This discussion motivates our belief that a watermark should be embedded in the data’s perceptually significant frequency components.Of course,the major problem then becomes how to imperceptibly insert a watermark into perceptually significant components of the frequency spectrum.Section III-A proposes a solution based on ideas from spread spectrum communications.In particular,we present a watermarking algorithm that relies on the use of the original image to extract the watermark.Section IV provides an analysis based on pos-sible collusion attacks that indicates that a binary watermark is not as robust as a continuous one.Furthermore,we show that a watermark structure based on sampling drawn from multiple i.i.d Gaussian random variables offers good protection against collusion.Ultimately,no watermarking system can be made perfect.For example,a watermark placed in a textual image may be eliminated by using optical character recogni-tion technology.However,for common signal and geometric distortions,the experimental results of Section V suggest that our system satisfies most of the properties discussed in the introduction,and displays strong immunity to a variety of attacks in a collusion resistant manner.Finally,Section VI discusses possible weaknesses and potential enhancements to the system and describes open problems and subsequent work.II.P REVIOUS W ORKSeveral previous digital watermarking methods have been proposed.Turner[25]proposed a method for inserting an identification string into a digital audio signal by substituting the“insignificant”bits of randomly selected audio samples with the bits of an identification code.Bits are deemed “insignificant”if their alteration is inaudible.Such a system is also appropriate for two-dimensional(2-D)data such as images,as discussed in[26].Unfortunately,Turner’s method may easily be circumvented.For example,if it is known that the algorithm only affects the least significant two bits of a word,then it is possible to randomlyflip all such bits,thereby destroying any existing identification code.Caronni[6]suggests adding tags—small geometric pat-terns—to digitized images at brightness levels that are imper-ceptible.While the idea of hiding a spatial watermark in an image is fundamentally sound,this scheme may be susceptible to attack byfiltering and redigitization.The fainter such watermarks are,the more susceptible they are such attacks and geometric shapes provide only a limited alphabet with which to encode information.Moreover,the scheme is not applicable to audio data and may not be robust to common geometric distortions,especially cropping.Brassil et al.[4]propose three methods appropriate for document images in which text is common.Digital watermarks are coded by1)vertically shifting text lines,2)horizontally shifting words,or3)altering text features such as the vertical endlines of individual characters.Unfortunately,all three proposals are easily defeated,as discussed by the authors. Moreover,these techniques are restricted exclusively to images containing text.Tanaka et al.[19],[24]describe several watermarking schemes that rely on embedding watermarks that resemble quantization noise.Their ideas hinge on the notion that quan-tization noise is typically imperceptible to viewers.TheirCOX et al.:SPREAD SPECTRUM WATERMARKING 1675first scheme injects a watermark into an image by using a predetermined data stream to guide level selection in a predictive quantizer.The data stream is chosen so that the resulting image looks like quantization noise.A variation on this scheme is also presented,where a watermark in the form of a dithering matrix is used to dither an image in a certain way.There are several drawbacks to these schemes.The most important is that they are susceptible to signal processing,especially requantization,and geometric attacks such as cropping.Furthermore,they degrade an image in the same way that predictive coding and dithering can.In [24],the authors also propose a scheme for watermarking facsimile data.This scheme shortens or lengthens certain runs of data in the run length code used to generate the coded fax image.This proposal is susceptible to digital-to-analog and analog-to-digital attacks.In particular,randomizing the least significant bit (LSB)of each pixel’s intensity will completely alter the resulting run length encoding.Tanaka et al.also propose a watermarking method for “color-scaled picture and video sequences”.This method applies the same signal transform as the Joint Photographers Expert Group (JPEG)(discrete cosine transform of8pairs of image points,,and increases thebrightnessat by one unit while correspondingly decreasing the brightnessofpairs of points isthenbits with theLSB of each pixel.If the LSB is equal to the corresponding mask bit,then the random quantity is added,otherwise it is subtracted.The watermark is subtracted by first computing the difference between the original and watermarked imagesand then by examining the sign of the difference,pixel by pixel,to determine if it corresponds to the original sequence of additions and subtractions.This method does not make use of perceptual relevance,but it is proposed that the high frequency noise be prefiltered to provide some robustness to lowpass filtering.This scheme does not consider the problem of collusion attacks.Koch,Rindfrey,and Zhao [14]propose two general methods for watermarking images.The first method,attributed to Scott Burgett,breaks up an image into88DCT block.The choice ofthe eight frequencies to be altered within the DCT block is based on a belief that the “middle frequencies...have moderate variance,”i.e.they have similar magnitude.This property is needed in order to allow the relative strength of the frequency triples to be altered without requiring a modification that would be perceptually noticeable.Superficially,this scheme is similar to our own proposal,also drawing an analogy to spread spectrum communications.However,the structure of their watermark is different from ours,and the set of frequencies is not chosen based on any direct perceptual significance,or relative energy considerations.Further,because the variance between the eight frequency coefficients is small,one would expect that their technique may be sensitive to noise or distortions.This is supported by the experimental results that report that the “embedded labels are robust against JPEG compression for a quality factor as low as about 50%.”By comparison,we demonstrate that our method performs well with compression quality factors as low as 5%.An earlier proposal by Koch and Zhao [15]used not triples of frequencies but pairs of frequencies,and was again designed specifically for robustness to JPEG compression.Nevertheless,they state that “a lower quality factor will increase the likelihood that the changes necessary to superimpose the embedded code on the signal will be noticeably visible.”In a second method,designed for black and white images,no frequency transform is employed.Instead,the selected blocks are modified so that the relative frequency of white and black pixels encodes the final value.Both watermarking procedures are particularly vulnerable to multiple document attacks.To protect against this,Zhao and Koch propose a distributed8,for example)thatare selected based on the binary digit to be transmitted.Thus,1676IEEE TRANSACTIONS ON IMAGE PROCESSING,VOL.6,NO.12,DECEMBER1997 Adelson’s method is equivalent to watermark schemes thatencode information into the LSB’s of the data or its transformcoefficients.Adelson recognizes that the method is susceptibleto noise and therefore proposes an alternative scheme whereina2l e v e l s a n d t h e h i g h f r e q u e n c y c o e ffic i e n t s,COX et al.:SPREAD SPECTRUM WATERMARKING1677be immune to intentional manipulation by malicious parties. These manipulations can include combinations of the above distortions,and can also include collusion and forgery attacks, which are discussed in Section IV-E.A.Spread Spectrum Coding of a WatermarkThe above discussion illustrates that the watermark should not be placed in perceptually insignificant regions of the image (or its spectrum),since many common signal and geometric processes affect these components.For example,a watermark placed in the high-frequency spectrum of an image can be easily eliminated with little degradation to the image by any process that directly or indirectly performs lowpassfiltering. The problem then becomes how to insert a watermark into the most perceptually significant regions of the spectrum in afidelity preserving fashion.Clearly,any spectral coefficient may be altered,provided such modification is small.However, very small changes are very susceptible to noise.To solve this problem,the frequency domain of the image or sound at hand is viewed as a communication channel, and correspondingly,the watermark is viewed as a signal that is transmitted through it.Attacks and unintentional signal distortions are thus treated as noise that the immersed signal must be immune to.While we use this methodology to hide watermarks in data,the same rationale can be applied to sending any type of message through media data.We originally conceived our approach by analogy to spread spectrum communications[20].In spread spectrum commu-nications,one transmits a narrowband signal over a much larger bandwidth such that the signal energy present in any single frequency is undetectable.Similarly,the watermark is spread over very many frequency bins so that the energy in any one bin is very small and certainly undetectable.Nevertheless, because the watermark verification process knows the location and content of the watermark,it is possible to concentrate these many weak signals into a single output with high signal-to-noise ratio(SNR).However,to destroy such a watermark would require noise of high amplitude to be added to all frequency bins.Spreading the watermark throughout the spectrum of an image ensures a large measure of security against unintentional or intentional attack:First,the location of the watermark is not obvious.Furthermore,frequency regions should be selected in a fashion that ensures severe degradation of the original data following any attack on the watermark.A watermark that is well placed in the frequency domain of an image or a sound track will be practically impossible to see or hear.This will always be the case if the energy in the watermark is sufficiently small in any single frequency coefficient.Moreover,it is possible to increase the energy present in particular frequencies by exploiting knowledge of masking phenomena in the human auditory and visual systems. Perceptual masking refers to any situation where information in certain regions of an image or a sound is occluded by perceptually more prominent information in another part of the scene.In digital waveform coding,this frequency domain (and,in some cases,time/pixel domain)masking isexploitedFig.2.Stages of watermark insertion process. extensively to achieve low bit rate encoding of data[9],[12].It is known that both the auditory and visual systems attach more resolution to the high-energy,low-frequency,spectral regions of an auditory or visual scene[12].Further,spectrum analysis of images and sounds reveals that most of the information in such data is located in the low-frequency regions.Fig.2illustrates the general procedure for frequency domain watermarking.Upon applying a frequency transformation to the data,a perceptual mask is computed that highlights per-ceptually significant regions in the spectrum that can support the watermark without affecting perceptualfidelity.The wa-termark signal is then inserted into these regions in a manner described in Section IV-B.The precise magnitude of each modification is only known to the owner.By contrast,an attacker may only have knowledge of the possible range of modification.To be confident of eliminating a watermark,an attacker must assume that each modification was at the limit of this range,despite the fact that few such modifications are typically this large.As a result,an attack creates visible(or audible)defects in the data.Similarly,unintentional signal distortions due to compression or image manipulation,must leave the perceptually significant spectral components intact, otherwise the resulting image will be severely degraded.This is why the watermark is robust.In principle,any frequency domain transform can be used. However,in the experimental results of Section VI we use a Fourier domain method based on the DCT[16],although we are currently exploring the use of wavelet-based schemes as a variation.In our view,each coefficient in the frequency domain has a perceptual capacity,that is,a quantity of additional1678IEEE TRANSACTIONS ON IMAGE PROCESSING,VOL.6,NO.12,DECEMBER 1997information can be added without any (or with minimal)impact to the perceptual fidelity of the data.To determine the perceptual capacity of each frequency,one can use models for the appropriate perceptual system or simple experimentation.In practice,in order to place alengthimage,we computedthe DCT of the image andplaced the watermark intothe.In practice,we create a watermark where eachvalue is chosen independentlyaccordingtodenotes a normaldistribution withmeana sequence ofvaluesto obtain an adjusted sequence ofvalues.toobtain a watermarkeddocumentandfor statistical significance.Weextract)and thengenerating.Frequency-domain based methods forextracting andinsertinginto,which determines the extent towhich.Three natural formulae forcomputing(2),which holds in all of our experiments.Given.Equation (1)may not be appropriate whenthe values vary widely.Ifadding 100will distortthis value unacceptably.Insertion based on (2)or (3)are more robust against such differences in scale.We note that (2)and (3)give similar resultswhenmay not be applicable for perturbingall of thevalues ,since different spectral components may exhibit more or less tolerance to modification.More generally one can have multiple scalingparametersCOX et al.:SPREAD SPECTRUM WATERMARKING 1679can perceptually “get away”withaltering by a large factor without degrading the document.There remains the problem of selecting the multiple scaling values.In some cases,the choiceof may be based on some general assumption.For example,(2)is a special case of the generalized(1),extract the correspondingvalueswhenever.One way to combine this constraint with the empirical approach would be toset accordingto.When we computed JPEG-based distortionsof the original image,we observed that the higher energy frequency components were not altered proportional to their magnitude [the implicit assumption of (2)].We suspect that we could make a less obtrusive mark of equal strength by attenuating our alterations of the high-energy components and amplifying our alterations of the lower energy components.However,we have not yet performed this experiment.C.Choosing theLength,dictates the degree to which the watermarkis spread out among the relevant components of the image.In general,as the number of altered components are increased the extent to which they must be altered decreases.For a more quantitative assessment of this tradeoff,we consider watermarks of theformwhere are chosen accordingto independent normal distributions with standarddeviationis proportionalto.Even the act ofrequantizing the watermarked document for delivery willcause.We measure the similarityofbysim(4)Many other measures are possible,including the standard correlation coefficient.Further variations on this basic metric are discussed in IV-D2.To decidewhether match,one determines whethersim ,whereand(eitherthrough the seller or through a watermarked document).Then,even conditioned on any fixed valueforisindependentofis a ing the well-knownformula for the distribution of a linear combination of variables that are independent and normallydistributed,will be distributed accordingtois distributed accordingtothen the probability thatsimequal to six will cause spurious matchings tobe extremely rare.Of course,the number of tests to be performed must be considered in determining what false positive probability is acceptable.For example,if one tests an extracted watermark watermarks,then the probability of a false positive is increased by a multiplicative factor of 10,the size of thewatermark.However,is generatedin the prescribed manner.As a rule of thumb,larger valuesofand),1680IEEE TRANSACTIONS ON IMAGE PROCESSING,VOL.6,NO.12,DECEMBER1997Fig.4.Bavarian couple image courtesy of Corel Stock Photo Library.without causing larger similarity valueswhenare independent.This benefit must be balanced against the tendency for the document to be more distortedwhenis generated with only finite pre-cisions.However,this effect is caused only by the arithmetic precision,and not on the constraints imposed by the document.Ifeach is stored as a double-precision real number,the difference between the calculated value of sim and its “ideal”value will be quite small for anyreasonablefrom,denoted,differed substantially from zero,due to the effectsof a dithering procedure.While this artifact could be easily eliminated as part of the extraction process,it provides a motivation for postprocessing extracted watermarks.We found that the simpletransformation yielded superior values of sim .The improved performance resulted from the decreased valueofcouldbe greatly distorted for some valuesofiftoleranceAgain,the goal of such a transformation is tolowerorCOX et al.:SPREAD SPECTRUM WATERMARKING1681(a)(b)Fig.7.(a)Lowpass filtered,0.5scaled image of Bavarian couple.(b)Rescaled image showing noticeable loss of fine detail.procedure forextractingandmultiple wa-termarkedcopiesto produce anunwatermarkeddocumentthwatermark is the same forallis generated byeitheraddingat randomto .Then as soon as one finds two documents with unequal valuesfordocuments one can,on average,eliminate all buta,whereas determin-ing a fidelity/undetectability tradeoff curve and the valueofby a linear update rule,thenan averaging attack,whichsetswill resultinwill beroughlywill be1682IEEE TRANSACTIONS ON IMAGE PROCESSING,VOL.6,NO.12,DECEMBER1997Fig.8.JPEG encoded version of Bavarian couple with 10%quality and 0%smoothing.roughly .Thus,the similarity measure can be shrunk bya factor ofreduction in the similarity measure.V.E XPERIMENTAL R ESULTSIn order to evaluate the proposed watermarking scheme,we took the Bavarian couple 2image of Fig.4and produced the watermarked version of Fig.5.We then subjected the watermarked image to a series of image processing and collusion style attacks.These experiments are preliminary,but show resilience to certain types of common processing.Of note is our method’s resistance to compression such as JPEG,and data conversion (printing,xeroxing and scanning).Note that in the case of affine transforms,registration to the original image is crucial to successful extraction.In all experiments,a watermark length of 1000was used.We added the watermark to the image by modifying 1000of the more perceptually significant components of the image spectrum using (2).More specifically,the 1000largest coeffi-cients of the DCT (excluding the DC term)were used.A fixed scale factor of 0.1was used throughout.A.Experiment 1:Uniqueness of WatermarkFig.6shows the response of the watermark detector to 1000randomly generated watermarks of which only one matches the watermark present in Fig.5.The positive response due to the correct watermark is very much stronger that the response to2Thecommon test image Lenna was originally used in our experiments,and similar results were obtained.However,Playboy Inc.refused to grant copyright permission for electronicdistribution.Fig.9.JPEG encoded version of Bavarian couple with 5%quality and 0%smoothing.Fig.10.Dithered version of the Bavarian couple image.incorrect watermarks,suggesting that the algorithm has very low false positive response rates.B.Experiment 2:Image ScalingWe scaled the watermarked image to half of its original size,as shown in Fig.7(a).In order to recover the watermark,the quarter-sized image was rescaled to its original dimensions,as shown in Fig.7(b),in which it is clear that considerable fine detail has been lost in the scaling process.This is to be expected since subsampling of the image requires a lowpass spatial filtering operation.The response of the watermark detector to the original watermarked image of Fig.5was 32.0,which compares to a response of 13.4for the rescaled version of Fig.7(b).While the detector response is down by over 50%,the response is still well above random chance。

Assignment SteganographyIf you are concerned about installing any of the software in these projects on your regular computer, you can instead install the software in the Windows virtual machine, ask the instructor for details.Part 1 Using OpenPuff SteganographyUnlike cryptography that scrambles a message so that it cannot be viewed, steganography hides the existence of the data. In this project, you will use OpenPuff to create a hiddenmessage.e your web browser to go to/OpenPuff_Steganography_Home.htmlIt is not unusual for websites to change the location of where files are stored. If the URL above no longer functions, open a search engine and search for “OpenPuff.”2.Click Manual to open the OpenPuff manual. Save this file to your computer. Readthrough the manual to see the different features available.3.Click OpenPuff to download the program.4.Click Screenshot to view a screen capture of OpenPuff. Right-click on this imageand save this image OpenPuff_Screenshot.jpg to your computer. This will be thecarrier file that will contain the secret message.For added security OpenPuff allows a message to be spread across several carrierfiles.5.Navigate to the location of the download and uncompress the Zip file on yourcomputer.6.Now create the secret message to be hidden. Open Notepad and enter This is asecret message.7.Save this file as Message.txt and close Notepad.8.Create a Zip file from the Message file. Navigate to the location of this file throughWindows Explorer and click the right mouse button.9.Click Send to and select Compressed (zipped) folder to create the Zip file.10.Navigate to the OpenPuff directory and double-click OpenPuff.exe.11.Click Hide.Under Bit selection options,note the wide variety of file types that can be used to hide a message.12.Under (1) create three unrelated passwords and enter them into Cryptography (A),(B), and (C). Be sure that the Scrambing (C) password is long enough to turn thePassword check bar from red to green.13.Under (2) locate the message to be hidden. Click Browse and navigate to the fileMessage.zip. Click Open.14.Under (3) select the carrier file. Click Add and navigate toOpenPuff_Screenshot.jpg as shown in Figure 3-11.15.Click Hide Data!16.Navigate to a different location than that of the carrier files and click OK.17.After the processing has completed, navigate to the location of the carrier file thatcontains the message and open the file. Can you detect anything different with the filenow that it contains the message?18.Now uncover the message. Close the OpenPuff Data Hiding screen to return to themain menu.19.Click Unhide.20.Enter the three passwords.21.Click Add Carriers and navigate to the location of Carrier1 that contains the hiddenmessage.22.Click Unhide! and navigate to a location to deposit the hidden message. When it hasfinished processing click OK.23.Click Done after reading the report.24.Go to that location and you will see Message.zip.25.Close OpenPuff and close all windows.Deliverable1.Submit a screenshot showing the images bytes without the hidden file and show the bytesof the file with the hidden file. You can use Windows Explore.2.Submit a screenshot showing successful hiding and unhiding the document.Part 2 Hiding an image in multiple images or video.Using what you have learned in part 1. Identify a word document file that is not more than 15Kb. Using sample images or videos as Carriers. Hide the Word document into a group of images or videos.Use only 1 password to secure the files and deactivate the other passwords.3.Submit a screenshot showing the images bytes without the hidden file and show the bytesof the file with the hidden file. You can use Windows Explore.4.Submit a screenshot showing successful hiding and unhiding the document.Part 3 Questions1.Which areas of a file cannot be used by steganography to hide data?2.What are the advantages and disadvantages of using steganography versus symmetric orasymmetric encryption?3.Which of the 5 security fundamentals principals for defense can be found usingsteganography,4.Explain how each principal is used from questions 3.Grading RubricCorrect submission of Part 1 30 pointsCorrect submission of Part 2 30 pointsCorrect submission of Part 3 40 pointsTotal 100 points。

小学上册英语第1单元真题试卷(有答案)英语试题一、综合题(本题有100小题，每小题1分，共100分.每小题不选、错误，均不给分)1.The flowers in the garden are _______ and cheerful, spreading happiness.2. A _______ (海鸥) flies near the shore.3.I can sing songs with my musical ________ (玩具名称).4.The density of a substance tells us how much mass is in a _______.5.What do we call the main ingredient in salad?A. LettuceB. DressingC. VegetablesD. All of the above答案:D. All of the above6.What is the capital of Portugal?A. LisbonB. MadridC. RomeD. Athens答案: A. Lisbon7.I want to _____ (read/write) a story.8.The capital of Brazil is ________ (巴西的首都是________).9.The chemical symbol for chromium is ______.10.I have learned a lot about __________ in school this year.11.The ice cream is ________ (冷).12.This boy, ______ (这个男孩), is working on a science project.13.I like to ride ______ with my friends. (horses)14.The ______ (小鱼) swims gracefully in the aquarium.15. A _______ is used to measure electrical current.16.The ancient Egyptians used _____ for mummification.17.He is a mechanic, ______ (他是一名机械师), who fixes cars.18.The main gas emitted from vehicles is __________.19.The ancient Greeks held _____ to commemorate their gods.20.The _______ plays an important role in the ecosystem.21.What is the opposite of "hot"?A. ColdB. WarmC. HighD. Tall答案:A Cold22.The first successful vaccine was developed for _______.23.My teacher helps me with ____.24. A neutral solution has a pH of _____.25.My aunt, ______ (我的阿姨), loves to travel and explore.26.What do we call a story that is made up?A. FictionB. Non-fictionC. BiographyD. History答案:A27.In 1776, the Declaration of Independence was signed in _____.28.The Titanic was a famous _______ ship.29.I can use my toy ________ (玩具名称) to inspire creativity.30.The __________ can reveal the geological timeline of Earth.31.I want to learn how to ______ (skate) on ice.32.What is 9 2?A. 5B. 6C. 7D. 8答案: C33.The ancient civilizations of the Americas built ________ for religious ceremonies.34.The _______ of sound is perceived as volume.35.The teacher promotes _____ (学习) in the classroom.36.I enjoy planting ________ in different colors.37.ts can _____ (存活) in extreme conditions. Some pla38.The chemical formula for water is ______.39.The __________ (历史的背景) shapes individual experiences.40.We enjoy _____ (hiking) trails.41. A reaction that results in a change of state is called a ______ reaction.42.The chemical formula for lithium hydroxide is _______.43.The __________ (历史的洞察) opens eyes.44.My favorite animal is a _______ (海豚).45.I have a favorite ________ that I bring everywhere.46.The car is parked _____ (in front/behind) the house.47.The ______ teaches us about international relations.48.The capital of Oman is __________.49.My sister and I have fun ____.50. A _____ (鱼) swims happily in the water. It has shiny scales. 一条鱼在水中快乐地游泳。

Expert Systems With Applications46(2016)293–306Contents lists available at ScienceDirectExpert Systems With Applicationsjournal homepage:/locate/eswaA steganography embedding method based on edge identification andXOR codingHayat Al-Dmour∗,Ahmed Al-AniSchool of Electrical,Mechanical and Mechatronic Systems,Faculty of Engineering and Information Technology,University of Technology,Sydney,NSW2007,Australiaa r t i c l e i n f oKeywords:SteganographyEdge detectionHuman Visual System(HVS)XOR codingIWTa b s t r a c tIn this paper,we present a novel image steganography algorithm that combines the strengths of edge detec-tion and XOR coding,to conceal a secret message either in the spatial domain or an Integer Wavelet Transform(IWT)based transform domain of the cover image.Edge detection enables the identification of sharp edgesin the cover image that when embedding in would cause less degradation to the image quality compared toembedding in a pre-specified set of pixels that do not differentiate between sharp and smooth areas.This ismotivated by the fact that the human visual system(HVS)is less sensitive to changes in sharp contrast areascompared to uniform areas of the image.The edge detection method presented here is capable of estimatingthe exact edge intensities for both the cover and stego images(before and after embedding the message),which is essential when extracting the message.The XOR coding,on the other hand,is a simple,yet effective,process that helps in reducing differences between the cover and stego images.In order to embed three se-cret message bits,the algorithm requires four bits of the cover image,but due to the coding mechanism,nomore than two of the four bits will be changed when producing the stego image.The proposed method uti-lizes the sharpest regions of the imagefirst and then gradually moves to the less sharp regions.Experimentalresults demonstrate that the proposed method has achieved better imperceptibility results than other pop-ular steganography methods.Furthermore,when applying a textural feature steganalytic algorithm to dif-ferentiate between cover and stego images produced using various embedding rates,the proposed methodmaintained a good level of security compared to other steganography methods.©2015Elsevier Ltd.All rights reserved.1.IntroductionSince the early stages of the human civilization,there has been anincreased interest in information security,particularly the protectionand privacy of communications(Pal&Pramanik,2013).In modern so-cieties,the excessive use of electronic data has made protection frommalicious users more difficult(Grover&Mohapatra,2013).Informa-tion hiding has emerged as an effective solution to this problem(Wu&Tsai,2003;Wu,Lee,Tsai,Chu,&Chen,2009).Steganography is a kind of information hiding,in which a secretmessage is concealed within digital media(image,audio,video ortext data)(Bassil,2012;Cheddad,Condell,Curran,&Mc Kevitt,2010).This property distinguishes steganography from other informationsecurity techniques(Modi,Islam,&Gupta,2013).For instance,incryptography,the message that needs to be transferred is encryptedto prevent intruders from understanding it.Hence,people can∗Corresponding author.Tel.:+61-426401478.E-mail addresses:HayatShahir.T.Al-Dmour@.au,HayatDmour@(H.Al-Dmour),Ahmed.Al-Ani@.au(A.Al-Ani).recognize the existence of the message,however it cannot be under-stood without decryption(Bassil,2012;Cheddad et al.,2010;Verma,2011).As opposed to data concealing,steganalysis was initially designedto distinguish whether a given digital media has a secret message em-bedded in it.Moreover,some steganalysis methods may determinethe type of steganography technique or estimate the length of the se-cret message(Li,He,Huang,&Shi,2011).In term of security mea-surement,steganalysis has been utilized to evaluate the efficiencyof steganography techniques from a security point of view(Geetha,Ishwarya,&Kamaraj,2010).Steganalysis methods can be performedeither by using image processing operation or by implementingmethods that analyze the statistical features of the stego image struc-ture,such asfirst order statistics(histogram)or second order statis-tics(correlations between pixels)(Cheddad et al.,2010).Ziou and Ja-fari suggestedfive requirements for steganalysis methods:(1)detec-tion of the existence or absence of an embedded message in a givenimage,(2)identification of the steganographic method that have beenused to hide the secret message,(3)approximation of the hiddenmessage length or location and(4)extraction of the secret message(Ziou&Jafari,2014)./10.1016/j.eswa.2015.10.0240957-4174/©2015Elsevier Ltd.All rights reserved.294H.Al-Dmour,A.Al-Ani/Expert Systems With Applications46(2016)293–306Table1Differentiation between image steganography schemes in spatial and transform doamins.Domain Advantages DisadvantagesHigh embedding capacitySpatial Shorter computational time Vulnerable to geometric attacks.Domain High controllable imperceptibilityHigh computational timeTransform Robustness against attacks such as Limited embedding capacityDomain Geometric attacks and compression Lower controllable imperceptibilityRecently,steganography has grown in popularity(Grover& Mohapatra,2013).Digital images have particularly been the focus of many researchers because of their high degree of redundancy(stored with an accuracy far greater than necessary for the data’s use and dis-play(Morkel,Eloff,&Olivier,2005)).Also,using images as cover will not create any suspicion due to their widespread use on the Inter-net(Cheddad et al.,2010).Three major requirements should be con-sidered when evaluating a steganography scheme:data embedding rate,imperceptibility,and robustness(Bassil,2012;Chen,Chang,& Le,2010).These three evaluation factors are needed as it is important for steganography techniques to have a high capacity and to be un-detectable(Chan&Chang,2010;Wu et al.,2009).The steganography terminology is listed below.•Cover image(C):carrier medium used to hide the message.•Stego image(S):output of the embedding process.•Message(M):secret message to be hidden within the cover image.•Key(K):used to encrypt the message before embedding(op-tional).•Embedding Process(Em):the process of generating S by hiding M into C.•Extraction Process(Ex):the process of retrieving M from S.Mathematically,the embedding(or concealing)process can be represented as:S=Em(C,M,K),and the extraction process as:´M= Ex(S,K).The extraction process should be reversible to the embed-ding process.Hence,Ex(Em(C,M,K),K)should be equal to M(or ´M=M).Numerous steganography methods have been proposed in the lit-erature(Cheddad et al.,2010;Luo,Huang,&Huang,2010;Verma, 2011).These methods can be partitioned based on the embedding domain;spatial and transform(Bassil,2012;Grover&Mohapatra, 2013;Ioannidou,Halkidis,&Stephanides,2012).In spatial domain steganography methods,M is directly embedded in the pixels of C. Two of the most famous steganography techniques are the Least Significant Bit(LSB)and Pixel Value Differencing(PVD).Transform domain steganography methods,on the other hand,transform C into another domain by performing one or more transforms,such as Discrete Transform Domain(DCT),Discrete Wavelet Transform (DWT),or Singular Value Decomposition(SVD),and then embed M by modifying the transformed coefficient values(Cheddad et al.,2010; Kanan&Nazeri,2014).Table1presents the difference between im-age steganography in spatial and transform domains in term of em-bedding capacity,imperceptibility and robustness(Bandyopadhyay, Dasgupta,Mandal,&Dutta,2014;Ghebleh&Kanso,2014;Hussain& Hussain,2013).The Least Significant Bit(LSB)hiding is the most common methodology to implement steganography(Morkel et al.,2005).LSB-based steganography is based on manipulating the LSBs of some or all pixels of the cover image to embed the message and can be clas-sified into two main types;LSB replacement(LSBR)and LSB match-ing(LSBM)(Luo et al.,2010;Zhu,Zhang,&Wan,2013).LSB-based steganography methods allow the concealment of a large amount of data.Another advantage of LSB steganography is the simple extrac-tion process(Luo et al.,2010).Usually,a pseudo random number generator is used to improve security of LSB steganography by spreading the message on the cover randomly(Luo et al.,2010).In order to enhance the embedding efficiency,coding methods (mainly matrix encoding)have been introduced with the aim of minimizing the modifications created by embedding the message (Crandall,1998;Hou,Lu,Tsai,&Tzeng,2011).In this paper,we propose a functional and simple image steganog-raphy method that is based on identifying edge locations on the cover image and incorporates an XOR coding function.The XOR function, which has a lower computation complexity compared to other ma-trix encoding methods,adds some security and reduces the distor-tion caused by embedding the message.Embedding in both spatial and wavelet transformed domains has been implemented.The rest of this paper is organized as follows.Section2provides a brief introduction to wavelet transform.Section3describes some of the existing steganography methods and examines their strengths and weaknesses.Details of the proposed method are presented in Section4.Section5presents the experimental results,and the con-clusion is given in section6.2.Wavelet transformTransform domain embedding methods provide a higher level of robustness,particularly when applying some image processing oper-ations,compared to spatial domain methods.One of the most popular transforms is the Discrete Wavelet Transform(DWT)(Baby,Thomas, Augustine,George,&Michael,2015;Thanikaiselvan et al.,2014).The Wavelet transform requires less computational cost compared to DCT and FFT(Fourier Transform)and offers sub-representations of the im-age that can be considered related to how the human visual system (HVS)perceives images.Generally,the wavelet transform allows em-bedding data in high frequency regions where the HVS cannot distin-guish modifications compared to uniform regions with low frequency (Sharma&Swami,2013).When DWT is performed to an image it is divided into4sub-bands:Low–Low(LL),Low–High(LH),High–Low (HL)and High–High(HH)frequency sub-bands,as shown in Fig.1. The low frequency sub-band represents coarse information of pixels, while the high frequency sub-bands represent the edge information (Sharma&Swami,2013).Hiding Information in the high frequency sub-bands(LH,HL,and HH)increases the robustness and ensures the visual quality,where the HVS is less sensitive to modifications in these sub-bands.The Integer Wavelet Transform(IWT)maps inte-gers to integers and allows the construction of lossless compression to exactly retrieve the original data(Thanikaiselvan et al.,2014).3.Related workImperceptibility is an essential requirement for steganography techniques,which reflects the ability of these techniques in main-taining the visual quality of the produced stego images.It is well-known that the HVS is less sensitive to changes in sharp areas of images compared to smooth areas.Thefirst steganography method designed based on this fact was the Pixel-value differencing(PVD), which attempts to embed into sharp areas.The original PVD algo-rithm introduced by Wu and Tsai(2003)converts the2D image into aH.Al-Dmour,A.Al-Ani/Expert Systems With Applications46(2016)293–306295Fig.1.(a)DWT sub-bands and(b)An example of the First Level DWT decomposition.1D vector.The number of bits that can be used for embedding in each pixel is calculated based on the difference between that pixel and its neighbour.Thus,more bits are to be embedded in a pixel if its grey level is noticeably different from that of its neighboring pixel.This method,however,only considers differences in one dimension(ei-ther horizontal or vertical),which does not guarantee that all edges are identified.This implementation can also increase the possibility of detecting the message by tracing pixels in every block.As it will create noticeable change in the adjacent pins in the histogram(Zhang &Wang,2004).In other words,it changes the relationship between consecutive pixels,and hence may cause distortion in the histogram of the stego-image.Because of this limitation,many revisions have been introduced to the original PVD algorithm,such as(Luo et al., 2010).An alternative approach is the utilization of edge detection al-gorithms,which has recently received an increased interest from the steganography community(Zhang&Wang,2004).Since the intensity of edge pixels is either higher or lower than their neighboring pixels, edge pixels might be distinguished as noisy pixels.Because of their high variations in statistical characteristics,edge regions can be con-sidered as the best regions to conceal the secret message compared to any other region of the image.3.1.Utilization of edge detection in steganographyThe utilization of edge detection in image steganography has been considered by a number of researchers.Due to sensitivity of the hu-man eye to changes in smooth areas of the image compared to sharp contrast areas,it is logical to focus on sharp edges when embedding the secret message.However,the main obstacle to applying tradi-tional edge detection methods in image steganography is the correct identification of edge pixels in the stego image S that need to exactly match the original edge pixels in the cover image C.This problem arises from the fact that the embedding process introduces minor changes to the stego image,which may make the produced stego im-age not identical with the cover image,and this can affect the mes-sage extraction process.Some of the existing edge-based steganog-raphy methods suggested certain solutions to overcome this prob-lem.Chen et al.(2010)applied a hybrid edge detection method to conceal the message.An edge image is created by performing Canny and fuzzy edge detection methods.The cover is then distributed into blocks of n pixels.Thefirst pixel of each block is changed to represent the status of(n−1)pixels if it is considered as edge pixel.LSB tech-nique is used to embed x bits into non-edge pixels and y bits into edge pixels.The main drawback of this method is the unwanted modifica-tion that are created in the stego image because the method replaces (n−1)bits from thefirst pixel of each block.Li,Luo,Li,and Fang(2009)introduced a spatial color image steganography based on Sobel operators.Sobel edge detection was performed on one R,G or B channel of the cover image.Embed-ding locations are chosen based on the largest number of gradients among R,G and B planes.The LSB of corresponding pixels in differ-ent planes are altered to conceal data.Embedding capacity is im-proved by repeating these phases many times until the secret mes-sage is embedded.Finally,the separate planes are integrated to form the stego-image.However,this process does not guarantee a high embedding rate and the data extraction is sometimes incorrect be-cause the method may embed data into the LSB pixels more than one time.Bassil(2012)proposed a color image steganography method based on canny edge detection to select the embedding location and LSB techniques to hide the message into the cover.The embedding rate is increased by this method because it hides3LSBs in every pixel detected by the canny method.However,this method does not introduce any solution to correctly identifying the same edge pixels.Luo et al.(2010)designed an edge adaptive LSB Matching Revis-ited(EALSB-MR)algorithm.The method discovers vertical and hori-zontal edges in an adaptive way.It searches for edge regions by cal-culating the difference between consecutive pixels.The selection of regions depends on the secret message length and is verified by a threshold value.The method uses horizontal and vertical edges by dividing the image into blocks then rotating each block by a random angle,however,this process could destroy the relationship between vertical/horizontal pixels(Modi et al.,2013).Modi et al.(2013)introduced a color image steganography based on the Canny edge detection method and LSBM to hide2LSBs on the edge pixels.Canny method is applied on one channel only(R,G,or B). Then the pixels in the other two channels corresponding to the edge pixels are selected for embedding.The main limitation of this method is the low embedding capacity,where the payload is0.083bpp of the edge pixel of the color image.Also,it cannot be utilized for grayscale images.3.2.Utilization of the coding theoryEnhancing the embedding efficiency has been the focus of many steganography algorithms,as minimizing the amount of changes in the image when embedding(embedding rate)will enable the em-bedding of bigger messages.Matrix embedding was introduced by Crandall(1998)to enhance the hiding efficiency through minimiz-ing the difference between the cover image and the stego image. Crandall’s method utilizes the XOR function to conceal2bits of mes-sage into a block of3pixels.This procedure has an embedding rate296H.Al-Dmour,A.Al-Ani /Expert Systems With Applications 46(2016)293–306Fig.2.(a)Cover image,(b)Edge pixels in a cover image using Canny method,(c)Edge pixels in a stego image using Canny method,and (d)Difference between edge pixels in the cover and stego images.of 67%and a change rate of 25%.The F5steganography algorithm,proposed by Westfeld (2001),is the first execution of matrix encod-ing to increase the capacity of embedding data as well as to mini-mize the change of DCT coefficients.This method has become well-known because it integrated the Hamming code with the transform domain implementation,which can embed k bits of the secret data in 2k −1cover bits by changing at most one bit only.As a result,this method has a limited embedding capacity,for example when k =3,the method only embeds 3bits in every 7bits of the cover image.An-other limitation of matrix embedding is the computational cost,as it requires matrix multiplication (Westfeld,2001).Hou et al.(2011)introduced an approach called tree based parity check (TBPC)that uses a tree structure to enhance the embedding ef-ficiency by reducing deformation on the cover object.They proposed a strategy of majority voting for TBPC and argued that this strategy inherited the efficiency of the TBPC method and produced the least deformation.Similar to some of the other coding methods,the draw-back of this method is the high computational cost,especially for trees that have multiple levels.The method can hide 2n bits in a bi-nary tree of n levels.For example,if the binary tree has 2levels,then it hides 4secret bits into 7pixels.3.3.Utilization of integer wavelet transformReddy and Raja (2011)proposed a wavelet based on LSB steganog-raphy method in which the original image is divided into 4×4pixels and DWT/IWT is performed on each block to form 2×2sub-bands block.The embedding process is performed on the HH sub-band of DWT/IWT.The 2×2HH coefficients are modified to carry bit pairs of the secret data using identity matrix to produce the stego image.For the extraction process,the key and the stego image are used to re-trieve payload.The authors claimed that this method cannot be dis-covered by some steganalysis methods such as Chi-square and pair of values methods.Kanan and Nazeri (2014),proposed a steganographic technique based on IWT and genetic algorithm where IWT is utilized to avoid floating point precision problem of the wavelet filter.It embeds the secret data in the IWT coefficients by using a mapping function based on a Genetic Algorithm implementation of an 8×8block of the cover image.The Optimal Pixel Adjustment Process (OPAP)is performed af-ter hiding the secret data to reduce the difference error between the cover and stego images and to improve the embedding payload with good image quality.The next section presents our proposed steganography embed-ding algorithm with its two implementations;one for the spa-tial domain while the other for the wavelet transform domain.We also present a simple,yet effective,coding process that is not as computationally demanding as most of the existing coding methods.4.The proposed method 4.1.The spatial domain algorithm4.1.1.Identification of edgesAs mentioned earlier,the human visual system is less tangible to changes in image areas that contain edges and sharp transitions in comparison to smooth areas.Accordingly,it is logical to conceal the message in edge areas in order for the steganography algorithm to have a good imperceptibility.The edge image generated by traditional edge detection methods is usually sensitive to changes in the original gray image,even if the changes are minor or not significant.This property limits the utiliza-tion of edge detection in steganography,as concealing the message would introduce some changes to the original image.Thus,embed-ding in pixels identified by one of the existing edge detection meth-ods,such as Canny,cannot guarantee the identification of the ex-act edge intensities for the cover and stego images.Fig.2(a)and (b)shows the cover image and the corresponding edge pixels which are identified by applying Canny edge detection.Edge pixels of the stego image produced after embedding a message of length 26214bits is shown in Fig.2(c).Fig.2(d)shows the difference between the two edge images,which indicate that edge pixels in the cover and stego images are not identical.We propose here a simple new way to discover the edge (sharp)regions of the cover image,such that the two edge images generated using the original cover image and the stego image are identical.This will enable the correct extraction of the concealed message from the stego image.The algorithm starts by dividing the image into non-overlapping blocks that would be individually evaluated for inclusion of edges.The key idea behind preserving the same edge image is not to embed in the pixels that are used to calculate the edge strength,which are the outer pixels of the block.We will first explain how to embed one bit per each of the center pixels,and then expand that to n bits per pixel.Below are the detailed implementation steps.Inputs:Cover image (C ),block size (n ×n ,which is expected here to be 3×3),threshold (Th ranges between 4and 96)Output:edge image with edge magnitude (E )Step 1:Divide the image C into non-overlapping blocks of the sizen ×n .Step 2:Compute the absolute mean difference between the left andright columns of the block (magnitude of vertical edge).Re-peat for horizontal,first diagonal and second diagonal edges.Fig.3(a)shows the specific pixels used to calculate the edges for the 3×3block.Step 3:Find the maximum of the four values and assign it to e .If e >Th ,then the block is considered to be an edge block,otherwiseH.Al-Dmour,A.Al-Ani /Expert Systems With Applications 46(2016)293–306297Fig.3.(a)3×3block edges and (b)Selected pixels for embedding 3×3block.Fig.4.(a)Input image,(b)edge image using Th =70,(c)edge image using T h =60,(d)edge image using Th =50,(e)edge image using Th =40,(f)edge image using Th =30,(g)edge image using T h =20,(h)edge image using T h =10.it is not an edge block.Construct E that contains the calculated e value of each of the edge blocks,and 0for non-edge blocks.A binary edge image can also be constructed,which contains 1for edge blocks and 0for non-edge blocks.Step 4:For the edge blocks,embed in the shaded 5pixels as shownin Fig.3(b).In order to evaluate the obtained binary edge image of the pro-posed algorithm,we considered the gray image shown in Fig.4(a)and used different values of threshold in constructing binary edge images using a block size of 3×3,as shown in Fig.4(b)−(h).The edge images indicate the ability of this method in detecting edges with an accept-able accuracy.Out of the nine pixels of the block,the five pixels shown in Fig.3(b)will be used for embedding if the block is identified as anedge block.Thus,the four corner pixels that are used for estimating the edge strength will remain unchanged after embedding.This guar-antees each block in the cover image to have the same edge strength as its counter part in the stego image.4.1.2.Message embeddingThe flow diagram of our proposed method is illustrated in Fig.5.The data embedding process begins with reading the cover image and the secret message.A high threshold (96)is initially considered,which is then adjusted based on the number of pix-els needed for embedding (identified by the generated binary edge image)and the message length,according to the following condition:298H.Al-Dmour,A.Al-Ani /Expert Systems With Applications 46(2016)293–306Fig.5.Data embedding process in the spatialdomain.Fig.6.Data extraction process in the spatial domain.Table 2Embedding conditions.ConditionAction to be taken m 1=k 1m 2=k 2m 3=k 3No change required m 1=k 1m 2=k 2m 3=k 3Complement p 3and p 4m 1=k 1m 2=k 2m 3=k 3Complement p 4m 1=k 1m 2=k 2m 3=k 3Complement p 3m 1=k 1m 2=k 2m 3=k 3Complement p 2m 1=k 1m 2=k 2m 3=k 3Complement p 1m 1=k 1m 2=k 2m 3=k 3Complement p 2and p 4m 1=k 1m 2=k 2m 3=k 3Complement p 1and p 4For the given threshold value,if (no.of edge pixels ≥(4∗Message Length)/3))then the discovered area is enough to embed the secret mes-sage.The embedding process is performed on the detected edge loca-tions using the proposed XOR coding.This method partitions the in-dex table into groups of four pixels and encodes three message bits into the pixels of each group.The XOR operation ensures that the se-cret message is concealed into the cover with minimum number of pixel changes.Thus,the three secret bits m 1,m 2,and m 3are embed-ded in the four LSBs p 1,p 2,p 3,and p 4(one bit for each edge pixel)according to the following procedure:1.Perform the following three XOR operations k 1=p 1⊕p 2k 2=p 3⊕p 4k 3=p 1⊕p 32.To embed the three secret bits m 1,m 2,and m 3,the three calcu-lated bits k 1,k 2and k 3are compared with the secret message bits m 1,m 2,and m3.The result of this comparison,which can take one of eight possibilities,determines which of the four bits p 1,p 2,p 3,and p 4have to be modified,as shown in Table 2.We will refer to the new four bits of the stego image as q 1,q 2,q 3,and q4.The ta-ble indicates that embedding 3message bits into 4cover bits will cause an average modification of 1.25bits.3.The threshold value should also be embedded,as it is needed by the extraction process.In this algorithm,the threshold value is embedded into the last pixel of the cover image.4.1.3.Message extractionThe extraction process is easier and faster than the embedding process.Fig.6represents the flow diagram of the extraction process.It starts by retrieving the threshold value.The edge blocks of the stego image are then identified using the retrieved threshold,which will return the same edge image as the one obtained using the cover im-age.This will be followed by dividing the LSBs of the edge pixels into groups of four.Finally,for each of the four stego edge bits q 1,q 2,q 3,and q 4the XOR operations listed below are used to retrieve three message bits m 1,m 2,and m 3m 1=q 1⊕q 2m 2=q 3⊕q 4m 3=q 1⊕q 3When considering any combination of m 1,m 2,m 3,p 1,p 2,p 3,and p 4to verify the embedding and extraction processes,one can find that the extraction process truly restores the original message.4.1.4.Embedding and extraction of n bits per pixelIn order to improve the embedding capacity,we present here an extension of our 1bpp algorithm to embed n bits in each edge pixel.The value of n is to be determined based on the edge mean value of each block.Thus,strong edges will enable the embedding of more bits than the less strong ones.Hence,unlike the embedding of one bit per pixel that only considers the existence of an edge in a block,this algorithm utilizes the edge strength of each block,e .Embedding n bits per pixel,where n varies from one block to another,may also improve the security of the message,as in this case n needs to be correctly calculated for each block in order to successfully reveal the message.The data hiding process begins with reading the cover image and the secret message.The new edge detection is then applied to pro-duce the edge strength,e ,of each block.In order to specify the num-ber of bits to embed,n ,the edge pixels are classified into five groups (G 1,G 2,G 3,G 4and G 5)based on the edge strength,e ,of the block.。

Good evening everyone, t oday I’ll tell you something about Information Hiding.As we all know, with the rapid development of communication technology, information security is becoming an important research topic. Information hiding is an emerging research hotspot in the field of information security. The emergence and development of Information hiding expanded a new field for the application of information security research.Information hiding is a way to hide secret information into the general non-secret digital media carrier, such as images, sounds, videos, and documents, so that opponents can’t find it.In short, information hiding is to hide the important information into the ordinary digital media.It is generally considered that It has the following three characteristics：①IntangibilityIt means do not affect the subjective quality of the original carrier and not easy to be detected by observer，except the watermarking。