WO2008060115A1

WO2008060115A1 - Method of detecting watermark using sub-block and apparatus therefor

Info

Publication number: WO2008060115A1
Application number: PCT/KR2007/005767
Authority: WO
Inventors: Hyuk Choi; Hyun Wook Kim; Hae Kyung Jung; Tae Jeong Kim
Original assignee: Hyuk Choi; Hyun Wook Kim; Hae Kyung Jung; Tae Jeong Kim
Priority date: 2006-11-17
Filing date: 2007-11-16
Publication date: 2008-05-22
Also published as: KR20080044680A; KR100862101B1

Abstract

A method and an apparatus for detecting a watermark using a sub-block are provided. In the method of detecting the watermark using the sub-block, it is possible to improve watermark detection performance by selecting sub-blocks having higher watermark-to-noise ratio (WNR) and then performing correlation-detection with the selected sub-blocks. In addition, the method is robust with respect to existing attacks.

Description

METHOD OF DETECTING WATERMARK USING SUB-BLOCK

AND APPARATUS THEREFOR

Technical Field

[1] The present invention relates to a watermark, and more particularly, to a method and an apparatus for detecting a watermark using a sub-block. Background Art

[2] Recently, high speed internet has been rapidly developed. Contents in addition to analogue media become digitalized as the analogue era is changed into the digital era.

[3] Unlike an analogue signal, a digital signal can be easily copied and stored, and users conveniently use the digital signal. However, since the digital signal can be copied as the same signal as the original signal, the digital signal is always exposed to an illegal copy with respect to the copyrighted digital contents. The copyright holder cannot be compensated for the copyrighted digital contents due to the illegal copy. Accordingly, since the copyright holder loses a creative desire, the illegal copy becomes a very serious problem. In order to solve the aforementioned problem, there is provided a watermarking technique.

[4] The watermarking technique is designed to protect intellectual property with respect to contents by inserting copyright information, which is unrecognizable for people, into the copyrighted digital content of the copyright holder, when intellectual property right is legally disputed.

[5] Watermark systems are classified into a type of a watermark system in which the watermark can be detected by using an original and a type of a watermark system in which the watermark can be detected without an original. The latter is referred to as a blind watermark system.

[6] Advantages of the blind watermark system will be described in the following. In an unblinded watermark system, an original into which the watermark is not inserted is needed, in addition to media into which the watermark is inserted, in order to detect the watermark. Accordingly, since the original is separately stored in the apparatus for detecting a watermark, a storage space is wasted. However, in the blind watermark system, since the original into which the watermark is inserted is not needed, a waste of the storage space decrease.

[7] In a generally used correlation based watermark, the watermark is added to the original signal or multiplied by the original signal. It is important to know how to find the inserted watermark more than to insert the watermark.

[8] However, an existing correlation-detection apparatus cannot be easily used in various application fields, since a variance of a degree of correlation between the cover signal and the watermark is generally large due to a characteristic of a cover signal (original source not including the watermark). Disclosure of Invention

Technical Problem

[9] The present invention provides a method and an apparatus for detecting a watermark using a sub-block for optimizing correlation-detection performance. Technical Solution

[10] According to an aspect of the present invention, there is provided a method of detecting a watermark using sub-blocks, the method comprising: selecting a sub-block in which the watermark-to-noise ratio (WNR) is higher than a predetermined value; and correlation-detecting the sub-block, when the sub-block in which the WNR is higher than the predetermined value is selected.

[11] In the above aspect of the present invention, the selecting of the sub-block in which the WNR is higher then a predetermined value may comprises: dividing a block into a plurality of sub-blocks; calculating a WNR for each sub-block; and aligning WNRs in descending order and selecting a sub-block in which a WNR is equal to or greater than a predetermined value in which a variance of a degree of correlation among cover or stego signal and the watermark is minimized.

[12] In addition, in the correlation-detecting of the sub-block, correlation-detection may be performed by using only the sub-blocks in which WNRs are higher than the predetermined value obtained until the lowest variance is detected.

[13] According to another aspect of the present invention, there is provided an apparatus for detecting a watermark using sub-blocks, which detect the watermark from a system including an encoder for encoding a signal by inserting a watermark signal into the signal, the apparatus comprises a detector which detects the watermark signal by selecting a part having higher WNR among signals output from the encoder and correlation-detecting the part.

[14] In the above aspect of the present invention, the apparatus may further comprise: a sub-block processing unit which divides a block into a plurality of sub-blocks; a sub- WNR calculation unit which calculates a WNR for each sub-block obtained by dividing the block in the sub-block processing unit; a sub-WNR alignment unit which aligns WNRs in descending order according to the result calculated by the sub-WNR calculation unit; a correlation-detection entry selection unit which selects a part in which the WNR is equal to or greater than a predetermined value in which the degree z of correlation among the cover or stego signals and the watermark can be minimized, according to the alignment order obtained from the sub-WNR alignment unit; and a correlation-detection unit which detects watermark using sub-blocks selected by the correlation-detection entry selection unit.

[15] In addition, the sub watermark-to-noise ratio alignment unit may align the WNRs in ascending order. When the WNRs are aligned in ascending order, a correlation- detection entry may be selected from the back side of the sub-WNR alignment unit.

[16] In the present invention, a watermark inserting device is used as in the past, and only a part having higher WNR according to characteristics of a cover signal (an original signal) is selected and watermark is detected using the selected part.

[17] This algorithm reduces a variance of correlation between the cover signal and the watermark. In addition, this algorithm is robust with respect to attacks such as compression, down sampling, jittering, and recording.

Advantageous Effects

[18] As described above, in a method and an apparatus for detecting a watermark using a sub-block according to the present invention, the watermark can be detected without errors and the detection performance is improved by selecting sub-blocks in which WNRs are high until the estimated variance of the degree of correlation is the minimum and performing correlation-detection using only the selected sub-blocks.

[19] In addition, the watermark is robust with respect to various attacks such as compression, down sampling, DA/AD conversion, jittering, and recording by decreasing the variance between the cover signal and the watermark. Brief Description of the Drawings

[20] The above and other features and advantages of the present invention will become more apparent by describing in detail exemplary embodiments thereof with reference to the attached drawings in which:

[21] FIG. 1 illustrates an apparatus for detecting a watermark to which the present invention is applied;

[22] FIG. 2 illustrates a structure obtained by dividing a block into sub-blocks in an apparatus for detecting a watermark according to an embodiment of the present invention;

[23] FIG. 3 is a detailed block diagram illustrating a detector shown in FIG. 1;

[24] FIG. 4 is a flowchart illustrating a method of detecting a watermark according to an embodiment of the present invention;

[25] FIG. 5 is a graph of a case where variance is large; and

[26] FIG. 6 is a graph of a case where variance is small.

Mode for the Invention

[27] Now, preferred embodiments of the present invention will be described in detail with reference to the attached drawings. [28] In order to apply a watermarking technique to contents, typically, transparency and robustness has to be considered. The transparency indicates a characteristic that the content into which the watermark is inserted is not easily distinguished from an original content. The robustness indicates a characteristic that the inserted watermark can be protected with respect to various attacks. Watermarking techniques, which are currently introduced in order to satisfy the two characteristics, include a spread spectrum technique, a technique of inserting watermark into phase, a quantization watermarking technique, a patchwork technique, and a method of inserting an echo signal into an audio signal.

[29] A correlation-detection technique is the most widely used watermark technique in the spread spectrum technique. Hereinafter, the present invention will be described in detail with reference to the attached drawings. FIG. 1 illustrates an apparatus for detecting a watermark to which the present invention is applied. As shown in FIG. 1, an original signal x[k] is not used. Although noise n[k] is added due to an attack, watermark noise w[k] can be detected. In the correlation-detection technique, the watermark noise w[k] is added to a signal component x[k], like s[k] = x[k] + aw[k], when the watermark is inserted. On the contrary, the correlation between w[k] and s[k] is obtained when the watermark is desired to be detected.

[30] The watermark signal w[k] is noise which damages the original signal. Although the watermark w[k] is the noise-like signal, it has to be under the following rules which are suitable to maintain signal quality. The rules will be described in the following.

[31] 1. The watermark signal w[k] to be concealed has to have a characteristic of a random number. Like the random number, irregularity of the watermark signal w[k] prevents attackers from estimating the watermark signal w[k]. Otherwise, the watermark signal w[k] can be easily estimated. The attackers attack the watermark signal w[k], thereby paralyzing the watermark signal w[k].

[32] 2. The random number has to be regenerated as it is. The random number that satisfies the aforementioned characteristics is referred to as a pseudo-random number or PN sequence. When the pseudo-random number is used, the same pseudo-random number as the pseudo-random number used by an encoder can be regenerated by the detector, and the watermark can be reconstructed with the pseudo-random number.

[33] 3. When the noise is represented as a number which ranges from -1 to 1, the mean value of the noise may approach 0. When the noise leans close to 1 or -1, the attacker can more easily estimate the random number, and a direct current (DC) component increases. Accordingly, detection performance deteriorates.

[34] In the present invention, in order to mathematically represent the result obtained by a correlation-detector, a sample x(n) of each cover signal is assumed to be an independent Gaussian random variable of which mean value is 0. [35] A degree of correlation between the cover signal x and a PN-sequence w is assumed to be a random number c. [36] MathFigure 1

[Math.l]

Λ-l c =< x, w >= ^ x(n)w(n) , where w(n) <≡ (-1,1) •

[37] Here, <,> indicates an inner product (scalar product).

[38] A mean value and a variance of the random number c are calculated by Math Figure

2 as follows: [39] MathFigure 2

[Math.2]

AIc] = O₅VaT[C] = JVP₁ , where P ' =

— ²^ .

N

[40] Here, N is a length of a signal x, P is a mean power of the signal x, E[c] is a mean value of the random number c, and var[c] is a variance of the random number c. Like an existing technique, the watermark is inserted by generating a stego signal s by adding a watermark aw to the cover signal x according to Math Figure 3. The stego signal indicates a source obtained by inserting the watermark into the cover signal.

[41] MathFigure 3

[Math.3] s = x + aw ,

[42] where is αscaling factor.

[43] The following Equations are used to evaluate an estimated variance of a degree of correlation and performance of a correlation-detector. When a degree z of correlation is represented by using the random number c of Math Figure 1, the degree z of correlation is calculated by Math Figure 4 as follows: [44] MathFigure 4

[Math.4] μ <x, αw> _ c πo.z - _{<αWj αw>} - _αN,

■ i ._ _ <s, αw> _ c Π^{1IZ "} <αw, αw> ^" αN

[45] where, E[c] = 0.

[46] Here, H is the abbreviation of hypothesis. In the present invention, H indicates an index for representing a case. H indicates a case where there is no watermark in the original signal. H indicates a case where a watermark is inserted into the original signal.

[47] MathFigure 5

[Math.5]

E[Z [Ho] = O₁ E[Z [H₁I =I,

[48] Since variance of the degree z of correlation is var[c] = NP , a relation in Math

Figure 6 is satisfied with respect to both cases H (a case where there is no watermark) and H (a case where there is a watermark). [49] MathFigure 6

[Math.6]

[50] When a mean value of the degree z of correlation is fixed with respect to the cases H and H , the performance of the correlation-detector increases, as the variance becomes small.

[51] Accordingly, since the mean value of the degree z of correlation among sub-blocks of the correlation-detector is fixed as in Math Figure 5, it is important to reduce the variance of the degree z of correlation.

[52] FIG. 2 illustrates a structure obtained by dividing a block into sub-blocks in an apparatus for detecting a watermark according to an embodiment of the present invention. As shown in FIG. 2, blocks B of any data are divided into sub-blocks.

[53] FIG. 3 is a detailed block diagram illustrating a detector shown in FIG. 1.

[54] In the present invention, a watermark detector 200 is designed to detect a watermark signal by selecting a part of a signal output from the encoder 100 shown in FIG. 1, in which a watermark-to-noise ratio (hereinafter, abbreviated to WNR ) is high and correlation-detecting the part of the signal. As shown in FIG. 3, the watermark detector 200 includes a sub-block processing unit 210, a sub- WNR calculation unit 220, a sub- WNR alignment unit 230, a correlation-detection entry selection unit 240, and a correlation-detection unit 250.

[55] The sub-block processing unit 210 serves to divide blocks into a plurality of sub- blocks. The sub- WNR calculation unit 220 serves to calculate a WNR for each sub- block obtained by dividing the blocks in the sub-block processing unit.

[56] The sub-WNR alignment unit 230 serves to align WNRs in descending order, according to the result calculated by the sub-WNR calculation unit 220. The correlation-detection entry selection unit 240 serves to select a part in which the WNR is equal to or greater than a predetermined value in which the degree z of correlation among the cover and stego signals and the watermark can be minimized, according to the alignment order obtained from the sub-WNR alignment unit 230.

[57] The correlation-detection unit 250 serves to detect the watermark by correlation- detecting the sub-blocks selected by the correlation-detection entry selection unit 240.

[58] At this time, since the detected watermark is obtained by selecting the sub-blocks having higher WNRs and correlation-detecting the selected sub-blocks, an error is small. The detected watermark is robust with respect to existing attacks.

[59] On the other hand, when the WNRs are aligned in ascending order by the sub-WNR alignment unit 230, the part in which WNR is high can be correlation-detected by correlation-detecting the back side of the sub-WNR alignment unit 230.

[60] Procedures of detecting the watermark by using the aforementioned apparatus for detecting the watermark will be described in the following.

[61] FIG. 4 is a flowchart illustrating a method of detecting a watermark according to an embodiment of the present invention.

[62] Referring to FIG. 4, first, a block of a watermark and a stego signal is divided into sub-blocks (operation SlO). Sub-WNRs are calculated for the sub-blocks obtained by dividing the blocks of the watermark and the stego signal in operation SlO (operation S 12). The sub- WNRs calculated in operation S 12 are aligned in descending order (operation S 14). That is, the sub-WNRs are aligned in ascending order of the WNRs. When the WNR is large, the variance is small.

[63] Variance of the watermark detector is estimated by adding sub-blocks in the order of alignment.

[64] The variance of the watermark detected by the watermark detector is calculated by

Math Figure 7 as follows:

[65] MathFigure 7

[Math.7]

, where P _k : power of first k number of sub-

blocks . [66] Here, N k indicates length of selected k number of sub-blocks.

[67] When the variance is estimated in operation S 16, correlation-detection is performed by using only sub-blocks when the variance is smallest (S 18). When the correlation- detection of each sub-block is completed, the watermark is detected (operation S20).

[68] A real audio signal is tested by using the aforementioned apparatus and method.

[69] Even in the same piece of music according to genres and features of audios, charac- teristics of audio signals are different with respect to blocks of the piece of music. Performance for detecting the watermark is improved by calculating the number of the sub-blocks in which the variance is lowest and in which the WNR is high and performing correlation-detection by using the sub-blocks.

[70] FIG. 5 is a graph of a case where a variance is large. Referring to FIG. 5, since two histograms have wide widths, the two histograms overlap with each other.

[71] FIG. 6 is a graph of a case where a variance is small. In FIG. 6, since two histograms have narrow widths, the two histograms do not overlap with each other. That is, when the variance is large, the histograms have wide widths as shown in FIG. 5. When the variance is small, the histograms have narrow widths as shown in FIG. 6.

[72] When the mean value of the degree of correlation is a predetermined value, as the va riance decreases, the error in detecting the watermark decreases.

[73] As described above, in a method and an apparatus for detecting a watermark using a sub-block according to an embodiment of the present invention, the watermark can be detected without errors and the detection performance is improved by selecting sub- blocks in which WNRs are high until the estimated variance of the degree of correlation is the minimum and performing correlation-detection using only the selected sub-blocks.

[74] In addition, the watermark is robust with respect to various attacks such as compression, down sampling, DA/AD conversion, jittering, and recording by decreasing the variance between the cover signal and the watermark.

[75] While the present invention has been particularly shown and described with reference to exemplary embodiments thereof, it will be understood by those skilled in the art that various changes in form and details may be made therein without departing from the spirit and scope of the present invention as defined by the appended claims.

Claims

[1] A method of detecting a watermark using sub-blocks, the method comprising: selecting a sub-block in which the watermark-to-noise ratio (WNR) is higher than a predetermined value; and correlation-detecting the sub-block, when the sub-block in which the WNR is higher than the predetermined value is selected.

[2] The method of claim 1, wherein the selecting of the sub-block in which the WNR is higher then a predetermined value comprises: dividing a block into a plurality of sub-blocks; calculating a WNR for each sub-block; and aligning WNRs in descending order and selecting a sub-block in which a WNR is equal to or greater than a predetermined value in which a variance of a degree of correlation among cover or stego signal and the watermark is minimized.

[3] The method of claim 1, wherein in the correlation-detecting of the sub-block, correlation-detection is performed by using only the sub-blocks in which WNRs are higher than the predetermined value obtained until the lowest variance is detected.

[4] An apparatus for detecting a watermark using sub-blocks, which detect the watermark from a system including an encoder for encoding a signal by inserting a watermark signal into the signal, the apparatus comprising a detector which detects the watermark signal by selecting a part of a signal output from the encoder, in which a WNR is high, and correlation-detecting the part.

[5] The apparatus of claim 4, further comprising: a sub-block processing unit which divides a block into a plurality of sub-blocks; a sub-WNR calculation unit which calculates a WNR for each sub-block obtained by dividing the block in the sub-block processing unit; a sub-WNR alignment unit which aligns WNRs in descending order according to the result calculated by the sub-WNR calculation unit; a correlation-detection entry selection unit which selects a part in which the

WNR is equal to or greater than a predetermined value in which the degree z of correlation among the cover and stego signals and the watermark is minimized, according to the alignment order obtained from the sub-WNR alignment unit; and a correlation-detection unit which correlation-detects the sub-blocks selected by the correlation-detection entry selection unit.

[6] The apparatus of claim 4, wherein the sub-WNR alignment unit aligns the WNRs in ascending order. [7] The apparatus of claim 6, wherein when the WNRs are aligned in ascending order, a correlation-detection entry is selected from the back side of the sub- WNR alignment unit.