On the correlation analysis of sequences designed for spread spectrum watermarking

In this paper, some concepts of correlation analysis which is mandatory in spread spectrum (SS) watermarking are reviewed. Next, the link between two analysis methods named CDS and D-transform, which look quite different is shown for the first time, is studied.

Nghiên cứu khoa học công nghệ ON THE CORRELATION ANALYSIS OF SEQUENCES DESIGNED FOR SPREAD SPECTRUM WATERMARKING Nguyen Le Cuong* Abstract: In this paper, some concepts of correlation analysis which is mandatory in spread spectrum (SS) watermarking are reviewed Next, the link between two analysis methods named CDS and D-transform, which look quite different is shown for the first time, is studied Their application for checking the auto-correlation function of pseudo-noise sequences (PN) employed in SS watermarking is investigated A comparison between them is made to give a broader view of mathematical tools employed in sequence design in SS watermarking Keywords: Watermarking, Spread spectrum, Correlation analysis INTRODUCTION Along with the digitalization of media assets, the rapid growth of the Internet, and the speed of file transfers, the danger of intellectual right violation become obvious Therefore, it is necessary to have mechanisms to protect these digital assets and associated rights In this regard, digital watermarking is considered as an effective measure against the illegal copies of images, music titles, and video films In digital watermarking process the information such as hidden copyright notices or verification messages are added to the cover medium like digital image, audio/video, or documents signals, to protect the ownership rights These hidden messages consist of a group of bits giving information about the author of the signal or the signal itself In order to extract or detect the hidden messages, the correlation analysis technique is used Since, in the last decade the spread spectrum communications (CDMA, HSPA) and cryptography techniques [5] had developed rapidly, this concept has been borrowed not only successfully for watermarking [14, 6-11] but also for other applications like: RFID, EPC, BARCODE [12, 13] and anti-jamming, protective jamming [14-16] In the above mentioned applications the following merits of the sequences are most desired: sharp auto – correlation function (ACF), large linear complexity (LC), larger length L, normal distribution assumption, bi-polar… This fact is obviously an encouraging motive for carry out an in depth study on nonlinear sequences satisfying those demand As we already know, in spread spectrum communication, a narrow-band signal is spread and then transmitted over a much larger bandwidth such that the signal energy presented in any signal frequency is undetectable The spreading signal is generated from a PN sequence running periodically at a much higher rate than the original data signal On the receiving end, the receiver first performs a correlation process on the incoming signal so that the original data signal is recovered in original bandwidth Similarly, a watermark is spread over many frequency bins so that the energy in one bin is very small and certainly undetectable, if the spreading sequence is not known Since the the location, the spreading sequence and the content of the watermark are known to the watermark verification process, it is possible to concentrate these weak signals into a single output with high SNR thank to the so Tạp chí Nghiên cứu KH&CN quân sự, Số 48, 04 - 2017 Kỹ thuật điều khiển & Điện tử called processing gain In technical terms we can express this process as [17] adding a modified maximal-length linear shift register sequence (m-sequence) to the pixel data The identification of the watermark is implemented by correlation techniques (two types of sequences may be formed from an m-sequence: unipolar and bipolar The elements of a bipolar sequence are {-1,1} and the elements of a unipolar sequence are{0,1}) It is worth to make the remark that the correlation process is not only applied in spread spectrum watermarking (SS watermarking) but widely used in other watermarking schemes [16] Without going into the details of watermarking process we concentrate ourself on the issue of appropriate sequences selection The paper is organized as follows: at the end of this section (I) we will review the related work in spread spectrum watermarking process The preliminaries are represented in next section (II) In the next section, attentions are paid to the: design and analysis issues However, due to the constraint scope of the paper, only the correlation analysis, which is the most important property in watermarking, is discussed The remaining requirements (robustness against attacks, balance…) will be discussed later in other contributions In the last section some comparisons and conclusions are given The sequences with above mentioned properties have found wide applications in watermarking process (image as well as audio watermarking) to improve its attributes of robustness, fidelity, capacity, detection [18] To improve the above attributes, the Pseudo-Random Sequence: with good random criteria (uniformly distributed, sharf ACF) are required In order to serve as secret key they should have larger linear complexity (LC) As for audio watermarking according to [19], it is pointed out that by applying redundant-chip coding to spread-spectrum watermarking, the system can effectively resist the geometric distortion such as time scaling and frequency scaling of up to 4% by performing multiple correlation Beside the spreading PN sequences, a so-called tracking sequence is also employed as a key sequence Later on, in the receiving end, one can locate the watermark embedding position by extensive computation the correlation to find out the maximum value According to [11] spread spectrum watermarking technique offers robustness Spread spectrum embedding is resistant against number of attacks, especially highly resistant to collusion attacks, since the watermarks have a component-wise Gaussian distribution and are statistically independent The randomness inherent in such watermarks makes the probability of accusing an innocent user very unlikely Spread spectrum embeds the watermark in overlapped regions and this spreading makes it challenging to change even a single bit at will The popular direct-sequence spread-spectrum watermarking approach is employed in the DCT domain because such spreading gives a robust but invisible watermark and it allows various types of detectors for blind watermark extraction [7] In [20], the need for introducing PN sequence for watermark signal spreading and robustness improving are highlighted The PN sequence is generated by a pseudo-random noise generator which has been initialized with a seed that depends on the secret-key, which significantly improve the robustness of the system This N L Cuong, “On the correlation analysis of… spread spectrum watermarking.” Nghiên cứu khoa học công nghệ key is known only to the legal owner of the watermarked document and without it the generation of the watermark at the receiver is impossible Furthermore, the spreading sequence must have noise-like properties in order to spread the spectrum of the input signal In other words, the mean of the sequence should be precisely zero and its auto-correlation should approach the delta function Consequently, a bi-polar pseudo-random sequence which takes the values, with relative frequencies 1/2 each will be the suitable candidate for this choice To survive the attack [20] proposed to combine PN sequences and concatenated turbo codes In [10], a PN of length L=296-1 bits is used to ensure larger LC PRELIMINARY In order to make sure that the generated sequences meet the above mentioned requirements, many methods have been proposed for analyzing the the properties of nonlinear sequences where LC and ACF, the two most important sequence criteria, have attracted a lot of attentions Recently, in [21] the methods for generating and analyzing LC of nonlinear sequences are thoroughly reviewed but ACF analysis is only briefly mentioned due to scope of that paper Since correlation analysis is the most crucial tool not only in every watermarking scheme but in frequency hopping, anti-jamming techniques [14,15] and security mechanism in digital watermarking [16], it deserves a special attention and needs to be separately treated When designing a watermark detection system, we need to consider the desired performance and robustness of the system The watermark should be able to be detected under common signal processing operations, such as digital-to-analog and analog-to-digital conversion, linear and nonlinear filtering, compression, and scaling The watermark detection is based on the correlation analysis [22] The correlation coefficients can be defined via the related variable [6] or via the statistic parameters of the variables In [6], the concept of SS watermarking was first introduced (1997) and the correlation coefficients as a measure of similarity between the original X and the extracted signals X* was used Furthermore [23], the concept of linear and normalized correlation of two vectors are employed The linear correlation between two vectors is the average product of their elements To normalize the linear correlation by the magnitudes of the two vectors, the detection is unaffected if all elements of either vector are multiplied by a constant A system using the correlation coefficient (CC) is robust to changes in image brightness and contrast In some papers, correlation coefficients are defined via the statistic parameters of the variables [24] The watermark energy resides in all frequency bands Compression and other degradation may remove signal energy from certain parts of the spectrum, but since the energy is distributed all over the spectrum, some of the watermark remains Co-variance is a measure of the joint variability of two random variables (In probability theory and statistics) Tạp chí Nghiên cứu KH&CN quân sự, Số 48, 04 - 2017 Kỹ thuật điều khiển & Điện tử Variance is a statistical measure that tells us how measured data vary from the average value of the set of data In other words, variance is the mean of the squares of the deviations from the arithmetic mean of a data set In case, the signal is expressed as a sequence of samples, the sample correlation statistic is introduced [25] Given a received signal, the watermark detector makes a (possibly incorrect) decision about the presence or absence of watermark We assume that the detector is synchronized with the embedded watermark W[n] A popular detection method is correlation detection, in which the detector computes the sample correlation statistic, and then compares to a threshold to decide whether W[n] is present or not A larger value of corresponds to increasing confidence that is indeed present In order to evaluate the performance of this decision a false alarm criterion is introduced [22] In some watermarking schemes [10, 24], two dimension processing technique is applied, therefore the 2D pseudo sequences (array) and the 2D correlation is defined correspondingly In these schemes, a very long m-sequence is employed to increase the LC and some bits are discarded to resist the correlation attack (like in mobile communications cryptography) Recently a more complicated and sophisticated watermarking schemes, which considers the effect of quantizing noise and the nonlinear correlations on the performance of spread spectrum watermark are presented [7] Now we can summarize the section as follows: correlation analysis is a crucial operation in watermark detection and extraction The performances of these actions depend closely upon the correlation properties of the employed PN sequences (best possible ACF) In the next section we will investigate the methods to check whether the designed sequences ensure these properties or not? CORRELATION ANALYSIS IN PN SEQUENCES DESIGNED FOR WATERMARK In this section we will review the two well known methods for ACF analysis and point out the link between them, which is still missing in the literature (to the best of our knowledge).They are namely combinatorial method and D-transform method 3.1 The mathematical (combinatorial) approach Since there is a one-to-one correspondence between cyclic difference sets and almost balanced binary sequences with the auto-correlation property [26-28] the constructing all cyclic difference sets is equivalent to finding all almost-balanced binary sequences with the desired auto-correlation property This problem has been thoroughly discussed and reported in literature so that we just give short reference here Definition [26-28] - difference set: A set of distinct integers D = {d1, d2, …, dk} modulo an integer υ is called integer difference set or difference set denoted by (υ, k, λ) if every integer b ≠ (mod υ) can be expressed in exactly λ way in the form di - dj ≡ b (mod υ), where di, dj belong to the integer set D Example 3.1: N L Cuong, “On the correlation analysis of… spread spectrum watermarking.” Nghiên cứu khoa học công nghệ 9 10 5 10 D = {1, 3, 4, 5, 9} is a (11, 5, 2) – difference set λ=2 It is well known that [27-29] CDS characteristic sequence of period υ defined by: 0 for t  D s(t)=  (1) 1 for t  D Has the two-level autocorrelation function: for   0(mod )  Rs ( )   (2) otherwise   4(k   ) Example 3.2: Consider a CDS (15, 7, 3), and D = {0 5-7 10,11-13,14} The corresponding sequence s(t) determined by (1) is: 011110101100100 and has a two-levelled ACF Definition 2- CDS with Singer parameters: Cyclic difference sets in GF(2n) with Singer parameters are those with parameters (2n – 1, 2n – – 1, 2n − − 1) for some integer n or their complements Sets of sequences with Singer parameters are: qary m-sequences, q-ary GMW sequences, and q-ary cascaded GMW sequences and they are having interleaved structure All of them are having two-levelled ACF For details please see [30,31] 3.2 The (technical oriented) D-transform method [21, 30, 31] This method is first proposed as early as 1985, much earlier than [29] but it is more hardware oriented and less well–known In technical term interleaving is nothing but time multiplexing, which is very well known to telecommunication engineers and is traditionally represented via delay operation (D-transform) Definition [21,30,31]: The D-transform of a sequence {bi} over GF(p) is denoted by D[bi] or F and designed by: (3) For example, let {bi} = 010111, D-transform of bn is D(bi) = D + D3 + D4 + D5 The inverse transform of D is D-1 = {bi} The D-transform of the generator sequence {bi} of a linear feedback shift register (LFSR) is then given by: (4) Where G(D) of degree n is the generating polynomial of a LFSR and S(D) of degree ≤ n -1 specifies the initial condition corresponding to a particular shifted version of {bi} - Shift sequences (interleaving orders) by D-transform Tạp chí Nghiên cứu KH&CN quân sự, Số 48, 04 - 2017 Kỹ thuật điều khiển & Điện tử Since interleaving process and D-transform are both sort of time multiplexing [8, 10] one can easily derive the interleaving order ITp straightforwardly In fact, there are two methods for derive ITp namely: expanding and decomposition (decimation) For the sake of simplicity we just show the result by decimation as in Example Example 3.3: Let m = 3, n = and let α be a primitive element of GF(26) with primitive polynomial b(D) = D6 + D5 + over GF(2) Let {bi} denote the msequence generated by b(D): {bi} = {0 1 1 1 1 1 0 1 1 1 1 0 0 1 0 1 1 0 1 0 1 0 0 0 0} Decimation of {bi} by T = 9, we obtain {ai} = {bi9} and rearrange is as a [9x7] matrix: 0000000 1110010 1011100 1110010 1100101 1011100 1011100 0101110 1110010 We can see that the rows are shift equivalent and = {z,5,3,5,6,3,3,2,5}, where z represents Null sequence It can be seen obviously that represents the relative phase shifts of sub-sequence {ai} In D-transform method the ACF of interleaved sequence is explained in a different way (without any concept of CDS) The two-level ACF is ensured by the phase shift relations between the sub-sequences and proved via the coincidence matrix: The digits in both row and column represent the relative phase shift between sub-sequence (or ACF matrix ) z z 5 3 6 5 3 + z 6 4 + + + + + + + + + + + + + + + + Note: The key remark here is: in each diagonal I≠0, there is exactly one place where the digits in both column and row are the same (marked by +)! In other N L Cuong, “On the correlation analysis of… spread spectrum watermarking.” Nghiên cứu khoa học công nghệ words: they form the one coincidence sequences and ensure that ACF is almost ideal! For details please see [21][30][31] 3.3 The missing links between two methods Even though the two methods are independently developed by different mathematical tools and parallel used in sequence design for a long time, there has not been any link between them shown so far Since one can not find the common language for them, it is best to illustrate the link via numerical examples Example 3.4: Step I: from the list of CDS, we pick up some CDS corresponding to sequences with interleaved structure (definition 2) - 1st: CDS {7, 3, 1}, D =1, 2, 4, λ=1 - 2nd: CDS {15, 7, 3}, D = {0, 5, 7, 10, 11, 13, 14} - 3rd: CDS {63, 31, 15}, D={0 ,1 ,2 ,3, 4, 6, 7, 8, 9, 12, 13, 14, 16, 18, 19, 24, 26, 27, 28, 32, 33, 35, 36, 38, 41, 45, 48, 49, 52, 54, 56} λ=15 - 4th: CDS {255, 127, 63}, D = {0, 7, 11, 13, 14, 17, 19, 22, 23, 26, 27, 28, 34, 38, 39, 43, 44, 46, 47, 49, 51, 52, 53, 54, 55, 56, 57, 63, 67, 68, 76, 77, 78, 83, 85, 86, 88, 89, 92, 94, 95, 97, 98, 99, 101, 102, 104, 106, 108, 110, 111, 112, 113, 114, 115, 119, 121, 123, 125, 126, 131, 133, 134, 136, 137, 139, 141, 147, 149, 151, 152, 153, 154, 155, 156, 159, 161, 166, 169, 170, 172, 175, 176, 177, 178, 183, 184, 185, 187, 188, 189, 190, 193, 194, 196, 197, 198, 201, 202, 203, 204, 205, 207, 208, 212, 215, 216, 219, 220, 221, 222, 224, 226, 228, 229, 230, 231, 235, 237, 238, 242, 243, 245, 246, 249, 250, 252}… Step II: conversion CDS into Binary sequences: Using s(t), which is defined as binary incidence indication sequence (or characteristic sequence) of the CDS st - binary incidence indication sequence: {bn1} = {1,0,0,1,0,1,1} (as a basic sub-sequence) nd - binary incidence indication sequence: {bn2} = {0,1,1,1,0,1,0,1,1,0,0,1,0, 0} rd binary incidence indication sequence: {bn3}= {0,0,0,0,0,1,0,0,0-,0,1,1,0, - 0,0,1,0,1-,0,0,1,1,1,1,0,1,0-,0,0,1,1,1,0,0,1,0-,0,1,0,1,1,0,1,1,1,0,1,1,0,0,1,1,0,1-,0, 1,0,1,1,1,1,1,1} th - binary incidence indication sequence: 011111101110/100110101100/110001111101/110011100100/101000000011/ 111011100111/111100011110/100100110100/100010010101/010000001110/ 101010011110/100100101011/111010100000/011010111101/100101100001/ 111000100001/100100011000/001001110110/011000010101/000011101001/ 110010011001/011-3 Step III: Decomposition of binary indication sequences into sub-sequences (like 1st and 2nd sequences) This operation is useful when applied for sequences with great composite length L = N.T, with N being the length of sub-sequences [29-31] We now decompose the 3rd and the 4th sequences into corresponding sub-sequences and arrange them into matrices [9x7] as follows: (rows represent sub-sequences) - {bn3} as: Tạp chí Nghiên cứu KH&CN quân sự, Số 48, 04 - 2017 Kỹ thuật điều khiển & Điện tử 0000000 0100111 0111010 0011101 0011101 1010011 0100111 0011101 0100111 With Shift sequence: = {z 5 0} - {bn4} as [17x15] matrix 000000000000000 111010110010001 101011001000111 111010110010001 110010001111010 100011110101100 101011001000111 011110101100100 111101011001000 100100011110101 100100011110101 001111010110010 110010001111010 010110010001111 011110101100100 110101100100011 111010110010001 With shift sequence: = {z, 0, 2, 0, 6, 10, 2, 13, 14, 7, 7, 12, 6, 3, 13, 1, 0} Note: There are another algorithms to calculate the shift sequences Step IV: Checking the ACF by correlation matrix [30][31] We create the ACF matrix by writing = {z, 0, 2, 0, 6, 10, 2, 13, 14, 7, 7, 12, 6, 3, 13, 1, 0} into row and column like below and remark the coincidences between them by + z 10 13 14 7 12 13 z 11 14 8 13 14 2 z + + + + + + + + + + + + + + 10 + + + + 13 + + + 14 + + + 10 N L Cuong, “On the correlation analysis of… spread spectrum watermarking.” Nghiên cứu khoa học công nghệ 7 12 13 ++ + + + + + + + + + + + + + + + + + Similarly, for the CDS {1023 511 255} corresponds to the binary sequences generated by GF(210) = + d3 + d10 we can decompose it into sub-sequences corresponding to GF(25) Then, we get the = { -, 23, 15, 20, 30, 10, 9, 17, 29, 13, 20, 21, 18, 21, 3, 9, 27, 12, 26, 21, 9, 7, 11, 11, 5, 22, 11, 4, 6, 27, 18, 14, 23} and the ACF matrix as: z 2315203010 17291320211821 27122621 1111 2211 27181423 z + 23 + + 15 + 20 + + 30 + 10 + + + + 17 + 29 + 13 + 20 + + 21 + + + 18 + + 21 + + + + + + + 27 + + 12 + 26 + 21 + + + + + + + 11 + + + 11 + + + + 22 + 11 + + + + Tạp chí Nghiên cứu KH&CN quân sự, Số 48, 04 - 2017 11 Kỹ thuật điều khiển & Điện tử 27 18 14 23 + + + + + + + + In each diagonal there is exactly one place where the digits in both column and row are the same (marked by +).The ACF is exactly as that one given by CDS and that clearly gives the evidence that both CDS and D-transform methods are equivalent This fact throws a new insight into the relation between two popular mathematical tools used in correlation analysis, which has not been given in the literature so far In this regard, we would like to emphasize that checking the coincidences between rows and columns of the matrix TxT is very much faster and far more intuitive than comparison of two sequence shifts, since the length of the sequence is L=T.N, especially when L is very large in practice Below is an examples for illustration For the sequence length L = 218 – = 262143 bits we have: - Sub-sequence length N = 29 – = 511 bits - T = L/N = 513 = {, 356, 201, 66, 402, 410, 132, 221, 293, 312, 309, 157, 264, 39, 442, 178, 75, 328, 113, 56, 107, 476, 314, 165, 17, 136, 78, 139, 373, 394, 356, 275, 150, 107, 145, 12, 226, 438, 112, 396, 214, 346, 441, 139, 117, 506, 330, 191, 34, 220, 272, 446, 156, 280, 278, 115, 235, 357, 277, 166, 201, 14, 39, 38, 300, 411, 214, 404, 290, 369, 24, 270, 452, 481, 365, 299, 224, 265, 281, 364, 428, 260, 181, 244, 371, 173, 278, 347, 234, 284, 501, 138, 149, 472, 382, 258, 68, 322, 440, 447, 33, 400, 381, 75, 312, 95, 49, 150, 45, 175, 230, 203, 470, 244, 203, 57, 43, 371, 332, 356, 402, 347, 28, 114, 78, 390, 76, 357, 89, 400, 311, 295, 428, 142, 297, 35, 69, 131, 227, 35, 48, 495, 29, 425, 393, 229, 451, 75, 219, 189, 87, 418, 448, 112, 19, 421, 51, 270, 217, 412, 345, 74, 9, 9, 362, 19, 488, 361, 231, 258, 346, 341, 45, 353, 183, 179, 468, 127, 57, 389, 491, 134, 276, 220, 298, 397, 433, 204, 253, 206, 5, 338, 136, 371, 133, 507, 369, 119, 383, 227, 66, 302, 289, 22, 251, 242, 150, 430, 113, 82, 190, 24, 98, 386, 300, 81, 90, 143, 350, 368, 460, 100, 406, 169, 429, 499, 488, 54, 406, 286, 114, 198, 86, 328, 231, 304, 153, 179, 201, 500, 293, 378, 183, 312, 56, 327, 228, 19, 156, 82, 269, 329, 152, 363, 203, 32, 178, 434, 289, 461, 111, 412, 79, 20, 345, 420, 284, 494, 83, 324, 70, 453, 138, 54, 262, 475, 454, 429, 70, 509, 96, 366, 479, 396, 58, 179, 339, 263, 275, 206, 458, 185, 391, 241, 150, 133, 438, 386, 378, 87, 174, 398, 325, 492, 385, 417, 224, 456, 38, 48, 331, 88, 102, 378, 29, 186, 434, 174, 313, 451, 179, 456, 148, 327, 18, 66, 18, 248, 213, 115, 38, 95, 465, 312, 211, 391, 462, 293, 5, 10, 181, 385, 171, 177, 90, 64, 195, 323, 366, 199, 358, 359, 425, 186, 254, 60, 114, 407, 267, 377, 471, 297, 268, 449, 41, 72, 440, 306, 85, 250, 283, 399, 355, 420, 408, 90, 506, 445, 412, 164, 10, 297, 165, 182, 272, 473, 231, 462, 266, 466, 503, 418, 227, 283, 238, 215, 255, 439, 454, 90, 132, 359, 93, 121, 67, 449, 44, 455, 502, 209, 484, 280, 300, 445, 349, 343, 226, 484, 164, 289, 380, 58, 48, 412, 196, 147, 261, 193, 89, 288, 162, 100, 180, 349, 286, 204, 189, 149, 225, 292, 409, 381, 200, 466, 301, 223, 338, 149, 12 N L Cuong, “On the correlation analysis of… spread spectrum watermarking.” Nghiên cứu khoa học công nghệ 347, 237, 487, 489, 465, 142, 108, 220, 301, 180, 61, 225, 228, 105, 396, 223, 172, 498, 145, 285, 462, 74, 97, 400, 306, 175, 358, 75, 402, 191, 489, 112, 75, 119, 245, 327, 366, 346, 113, 53, 112, 505, 143, 293, 456, 88, 38, 96, 312, 60, 164, 429, 27, 253, 147, 300, 304, 286, 215, 127, 406, 399, 64, 200, 356} Using conventional method the correlation function of the binary sequences can be calculated as: R(s)=(A-D)/(A+D) Where A is the number of Agreements and D is the number of Disagreements between {bn5} and its shift version Ts{bn5} ( shifted by bits) and A+D = L = 262143 Since there are 262143 values of shift in each R(s),this conventional approach is laborious and difficult to follow With the proposed approach ,only T=513 elements need to be checked! Taking into the fact that L maybe far greater in the practive (e.g 296-1) the convetional approach becomes unbearable DISCUSSION AND FUTURE WORK In this contribution we first review some concepts of correlation analysis, which is mandatory in SS watermarking Next, the link between two analysis methods, namely CDS and D-transform, which look quite different is shown for the first time, is studied The numerical example illustrates the steps to converse one method to another The following comparison can be made: - The CDS concept is a compact mathematical tool and powerful in describing the sequence structure However, being a pure mathematical tool, it gives no indication for hardware implementation Furthermore, it cannot be used for LC analysis - The D-transform is a hardware oriented tool (multiplexing) and can be used for LC analysis also even though it is not as compact as CDS We hope that this contribution and [21] will give a broader view of mathematical tools employed in sequence design in SS watermarking Due to the limited scope of the paper, some interesting issues could not be addressed, such as: multi-dimensional correlation analysis, bipolar sequences and experimental result REFERENCES [1] Peterson R.L, Zeemer R.E & Both D.B, “Introduction to spread spectrum”, Prentice Hall Int Inc, 1995 [2] Bambang Harjito PhD thesis, “Copyright Protection of Scalar and Multimedia Sensor Network Data Using Digital Watermarking”, Curtin University Australia, 2013 [3] Kaur, S., “Digital Watermarking of ECG Data for Secure Wireless Communication International Conference on Recent Trends in Information, Telecommunication and Computing”, 2010 [4] Pingping, Y.S., Yao Jiangtao, Xu Yu, Zhang Ye, Chang, “Copyright Protection for Digital Image in Wireless Sensor Network in Wireless Tạp chí Nghiên cứu KH&CN quân sự, Số 48, 04 - 2017 13 Kỹ thuật điều khiển & Điện tử [5] [6] [7] [8] [9] [10] [11] [12] [13] [14] [15] [16] [17] [18] 14 Communications, Networking and Mobile Computing,” WiCom '09 5th International Conference, 2009 Thomas Johansson, Fre Jonsson, “Theoretical Analysis of a correlation Attack based on Convolutional code”, IEEE Trans On information theory, Vol 48, No 8, pp 2173-2181, August 2002 I J Cox, J Kilian, T Leighton, and T Shamoon, “Secure spread spectrum watermarking for multimedia,” IEEE Trans Image Processing, Vol 6, no 12, pp 1673–1687, Dec 1997 Ashok Patel, “A Bart Kosko Noise Benefits in Quantizer-Array Correlation Detection and Watermark Decoding,” IEEE TRANSACTIONS ON SIGNAL PROCESSING, VOL 59, NO 2, pp 448, February 2011 Kang X Member, Yang R and J Huang R, “Geometric Invariant Audio Watermarking Based on an LCM Feature,” IEEE TRANSACTIONS ON MULTIMEDIA, VOL 13, NO 2, pp 181-190, APRIL 2011 Al-Rawi1.S.S, Sadiq.A.T, Farhan B.G, “Digital Video Quality Metric Based on Watermarking Technique with Geffe Generator Computer Science and Engineering,” 2(7): 138-146, 2012 Raymond B W, Edward J., “A watermarking technique for digital imagery: further studies cerias,” Tech Report 2007-44 Delp Video and Image Processing Laboratory (VIPER) USA, 2007 Shahid Z Chaumont M and Puech W., “Spread spectrum-based watermarking for tardos code-based fingerprinting for H.264/AVC video,” University of Montpellier II, VOODDO (2008-2011) project of the french Agence Nationale pour la Recherche Joan Melia-Segu et al, “Multiple-Polynomial LFSR based Pseudorandom Number Generator for EPC Gen2 RFID Tags,” Universitat Oberta de Catalunya Spain, 2010 A.Mitrokotsa, M.R.T.Beye, P.PerisLope, “Classification of RFID Threats based on Security Princip,” TU Delft 2011 Christina Popper et al, “Jamming-resistant Broadcast Communication without Shared Keys,” Usenix Security Symp 2009 Christina Popper et al, “Anti-jamming Broadcast Communication using Uncoordinated Spread Spectrum Techniques,” IEEE JOURNAL ON SELECTED AREAS IN COMMUNICATIONS, VOL.28, NO.5, pp1-13, JUNE 2010 Taeho Kang, “Survey of Security Mechanisms with Direct Sequence Spread Spectrum Signals,” Journal of Computing Science and Engineering, Vol.7, No.3, pp 187-197, September 2013 “A watermarking technique for digital imagery: Further studies,” Tech Report 2007-44, USA Charles Way Hun Fung, Antonio Gortan, Avenida Sete de Setembro, “The Tenth International Conference on Networks A Review Study on Image Digital Watermarking,” ICN 2011 N L Cuong, “On the correlation analysis of… spread spectrum watermarking.” Nghiên cứu khoa học công nghệ [19] Xiangui Kang, Rui Yang and Jiwu Huang, “Geometric Invariant Audio Watermarking Based on an LCM Feature,” IEEE Transactions on multimedia, Vol.13, No.2, pp.181-190, April 2011 [20] Kang X, Huang J and Zeng.W, “Improving Robustness of QuantizationBased Image Watermarking via Adaptive Receiver,” IEEE Transactions on multimedia, Vol.10, No 6, pp 953-959, October 2008 [21] Cuong N L., “A comparative study on some mathematical tools used in design and analysis of interleaved sequences,” Journal of Science and Technology, Nov 2016 [22] Jongwon Seok, Jinwoo Hong and Jinwoong Kim, “A Novel Audio Watermarking Algorithm for Copyright Protection of Digital Audio,” Inetri journal, pp 181-188, 2002 [23] cmlab.csie.ntu.edu.tw [24] Gerrit C Langelaar, Reginald L Lagendijk, Jan Biemond, “Removing Spatial Spread Spectrum Watermarks by Non-linear Filtering,” Information and Communication Theory Group, Delft University of Technology, The Netherlands 1997 [25] Jonathan K Su, Bernd Girod, “Power-Spectrum Condition for EnergyEfficient Watermarking,” IEEE Transactions on multimedia, Vol.4, No 4, pp 551-560, December 2002 [26] B Gordon, W H Mills and L R Welch, “Some new difference sets,” Canad J Math., Vol.14, pp 614–625, 1962 [27] L D Baumert, “Cyclic Difference Sets,” Lecture Notes in Mathematics, Springer Verlag, 1971 [28] J Kim and H.Y.Song, “Existence of Cyclic Hadamard Difference Sets and its Relation to Binary Sequences with Ideal Autocorrelation,” Journal of Communications and Networks, Vol.1, No.1, pp 14-18, MARCH 1999 [29] S W Golomb and G Gong, “Signal Design for Good Correlation - for Wireless Communication, Cryptography and Radar,” Cambridge University Press, 2005 [30] Quynh L C., S Prasad, “A class of binary cipher sequences with best possible correlation function,” IEEE Proceeding Part F., Vol.132, pp 560, Dec 1985 [31] Hieu L M., Quynh L C., “Design and Analysis of Sequences with Interleaved Structure by d-Transform,” IETE Journal of Research, vol 51, no l, pp 61-67, Jan-Feb 2005 [32] M.Natarajan et al, “Performance Comparison of Single and Multiple Watermarking Techniques”, I.J Computer Network and Information Security, 7, 28-34, 2014 [33] Z Huang et al, “Digital Watermarking Algorithm Based on Spread Spectrum and Fourier Transform”, 3rd International Conference on Management, Education, Information and Control (MEICI 2015) pp-1348-1352, 2015 [34] H A Zörlein, “Channel Coding Inspired Contributions to Compressed Sensing”, PhD thesis Universität Ulm Germany, 2015 Tạp chí Nghiên cứu KH&CN quân sự, Số 48, 04 - 2017 15 Kỹ thuật điều khiển & Điện tử TĨM TẮT PHÂN TÍCH TƯƠNG QUAN CÁC DÃY ĐƯỢC SỬ DỤNG TRONG THỦY VÂN TRẢI PHỔ Trong báo này, trước hết, số khái niệm phân tích tương quan, vấn đề có tính chất bắt buộc thủy vân trải phổ (SS) trình bày Tiếp theo, báo nghiên cứu mối quan hệ hai phương pháp phân tích khác CDS D-transform Ứng dụng chúng để kiểm tra hàm tự tương quan chuỗi giả ngẫu nhiên (PN) dùng thủy vân trải phổ khảo sát Việc so sánh cung cấp góc nhìn sâu rộng cơng cụ tốn học dùng để thiết kế chuỗi giả ngẫu nhiên dùng thủy vân trải phổ Từ khóa: Thủy vân, Trải phổ, Phân tích tương quan Nhận ngày 01 tháng năm 2017 Hoàn thiện ngày 04 tháng năm 2017 Chấp nhận đăng ngày 05 tháng năm 2017 Address: Electric Power University, 235 – Hoang Quoc Viet, Hanoi, Vietnam * Email: cuongnl@epu.edu.vn 16 N L Cuong, “On the correlation analysis of… spread spectrum watermarking.” ... section as follows: correlation analysis is a crucial operation in watermark detection and extraction The performances of these actions depend closely upon the correlation properties of the employed... compression, and scaling The watermark detection is based on the correlation analysis [22] The correlation coefficients can be defined via the related variable [6] or via the statistic parameters of the. .. sophisticated watermarking schemes, which considers the effect of quantizing noise and the nonlinear correlations on the performance of spread spectrum watermark are presented [7] Now we can summarize the