JP2002244685A - Embedding and detection of digital watermark - Google Patents

Abstract

PROBLEM TO BE SOLVED: To provide digital watermark technology which actualizes high secrecy and sufficiently withstand an attack destroying signal synchronism like a jitter attack. SOLUTION: In a step S102, an MDCT(modified discrete cosine transformation) part 42 prepares source sound data and a key random number sequence. In step S104, the MDCT part 42 performs MDCT of the source sound data to fine MDCT coefficients. In step S106, selection and arithmetic part 44 selects a plurality of couples of MDCT coefficients out of the MDCT coefficients according to random number values extracted from the key random number sequences and adds two separately generated control random number values to two MDCT coefficients for each selected pairs to obtain MDCT coefficients having been embedded. In step S108, an IMDCT part 46 applies IMDCT to a plurality of the already embedded IMDCT coefficients to put the embedded MDCT coefficients back into the sound data, so that the embedded sound data can be obtained.

Description

DETAILED DESCRIPTION OF THE INVENTION

[0001]

BACKGROUND OF THE INVENTION 1. Field of the Invention The present invention relates to a digital watermarking technique for embedding a digital watermark in digital data such as audio data and image data.

[0002]

2. Description of the Related Art With the development of computer networks such as the Internet, digitization of information has progressed, and many users can easily access necessary information. On the other hand, digital content in which copyright is generated in digital information is becoming an environment where data can be easily copied without asking the author, and the problem of copyright infringement due to unauthorized copying has been attracting attention. ing. Therefore, digital watermark technology for embedding watermark information such as copyright information in audio data or image data has attracted attention for the purpose of preventing copyright infringement on audio and images as main information of digital contents.

[0003] Among such digital watermarking techniques, for example, in a method using spread spectrum, high confidentiality can be realized by using a spread code string as a key. Also, in a technique called patchwork, by controlling two randomly selected time-series sample values and shifting the peak position of the distribution obtained from those values, the selectivity of the sample values can be randomized. High confidentiality similar to the diffusion method can be realized.

[0004]

However, in a method of randomly controlling a time-series sample value, such as a spread spectrum method or a patchwork method, the synchronization of random numbers when detecting a watermark must be accurate. Although this can realize high confidentiality, it has a problem that it is vulnerable to an attack that breaks signal synchronization such as a Jitter attack known as an attack method against digital watermarking.

[0005] Therefore, an object of the present invention is to solve the above-mentioned problems of the prior art, to realize high secrecy, and
An object of the present invention is to provide a digital watermarking technique that can sufficiently withstand an attack such as an itter attack that breaks signal synchronization.

[0006]

In order to achieve at least a part of the above objects, a digital watermark embedding method according to the present invention is a digital watermark embedding method for embedding a digital watermark in digital data. (A) preparing target data for embedding the digital watermark and a key random number sequence used for embedding the digital watermark; and (b) performing a predetermined orthogonal transformation on the target data. And (c) selecting n (n) from among the plurality of transform coefficients in accordance with random numbers sequentially extracted from the key random number sequence.
(D is an integer of 2 or more), and (d) selecting at least one transform coefficient among at least one of the n transform coefficients for each selected set. Performing a predetermined operation with a random number value; and (e) performing a reverse operation of the orthogonal transformation on the plurality of transformation coefficients including the transformation coefficient on which the operation has been performed, thereby returning the transformation data to target data. , Is provided.

As described above, according to the digital watermark embedding method of the present invention, a predetermined orthogonal transformation is performed on the target data to obtain a plurality of transform coefficients, and according to a random number value extracted from a key random number sequence prepared in advance, From these conversion coefficients, n is 1
Select a set of conversion coefficients to be set. Then, for each of the selected sets, a predetermined operation is performed on at least one of the n transform coefficients using a control random value obtained separately. Further, a plurality of transform coefficients including the calculated transform coefficients are subjected to a transform reverse to the orthogonal transform to return to the target data.

According to the digital watermark embedding method of the present invention, the digital watermark can be detected only when the same key random number sequence as that used for embedding the digital watermark is used. It is extremely difficult for a third party who does not know the random number sequence to identify the presence of the digital watermark by unauthorized means and remove the embedded digital watermark without deteriorating the sound quality. Therefore, high confidentiality can be realized.

According to the digital watermark embedding method of the present invention, a digital watermark is embedded in a frequency region which is not easily affected by minute modulation in a time series. Therefore, signal synchronization such as Jitter attack can be performed. It can withstand even breaking attacks. In addition, it has sufficient resistance to level modulation, band-pass filters, high-efficiency compression coding, and the like.

In the digital watermark embedding method according to the present invention, it is preferable that the step (b) includes a step of performing a modified discrete cosine transform as the predetermined orthogonal transform.

As described above, by using the modified discrete cosine transform as the orthogonal transform, even when the target data is divided into a plurality of frames and subjected to the orthogonal transform, spike noise and distortion which are likely to occur between frames can be reduced. Can be.

[0012] In the digital watermark embedding method of the present invention, the step (c) includes the step of:
Selecting a set of transform coefficients that makes the two a set,
Preferably, the step (d) includes a step of adding or subtracting two separately obtained control random numbers to the two transform coefficients for each selected set.

With this configuration, the digital watermark can be reliably embedded by a simpler operation.

In the digital watermark embedding method of the present invention, the step (a) includes a step of preparing a plurality of the key random number sequences, and the step (c) includes the step of: According to the random value sequentially extracted from the column,
N (n is an integer of 2 or more) among the plurality of conversion coefficients
Preferably, the method includes a step of selecting a set of transform coefficients each of which is a set.

As described above, by embedding an electronic watermark using a plurality of key random number sequences, it becomes possible to embed an electronic watermark in the same content multiple times.

The digital watermark detection method according to the present invention is a digital watermark detection method for detecting the digital watermark from digital data in which a digital watermark is embedded, wherein: (a) the digital watermark detection target data; Preparing a key random number sequence used for detecting the digital watermark, (b) performing a predetermined orthogonal transform on the target data to obtain a plurality of transform coefficients, and (c) determining a plurality of transform coefficients. (D) selecting a set of transform coefficients in which n (n is an integer of 2 or more) one set from the plurality of transform coefficients in accordance with random numbers sequentially extracted from the key random number sequence; A) performing a predetermined operation between n transform coefficients for each selected set; and (e) detecting the presence of the digital watermark from the appearance frequency of the obtained operation value. Make a summary.

As described above, according to the digital watermark detection method of the present invention, a predetermined orthogonal transform is performed on the target data to obtain a plurality of transform coefficients, and according to a random number value extracted from a key random number sequence prepared in advance, From these transform coefficients, a set of transform coefficients of n sets is selected. Then, for each of the selected sets, a predetermined operation is performed between n transform coefficients, and the presence of a digital watermark is detected from the appearance frequency of the obtained operation value.

According to the digital watermark detection method of the present invention, if only the same key random number sequence as the key random number sequence used for embedding the digital watermark is prepared, the digital watermark embedded by the above-described digital watermark embedding method of the present invention can be used. , Can be easily detected.

In the digital watermark detection method according to the present invention, it is preferable that the step (b) includes a step of performing a modified discrete cosine transform as the predetermined orthogonal transform. In addition, the step (c) includes a step of selecting a set of transform coefficients of which two are one set from the plurality of transform coefficients, and the step (d) includes the step of: It is preferable to include a step of adding or subtracting two transform coefficients.

This is because the orthogonal transform and the operation method used when the digital watermark is embedded are to be used.

It should be noted that the present invention is not limited to the aspects of the method invention such as the above-described digital watermark embedding method and the digital watermark detection method. It can also be realized in a form as a recording medium on which a computer program for recording is recorded. Further, the present invention can be realized in various modes, such as a data signal including the computer program and embodied in a carrier wave or the computer program itself.

[0022]

DESCRIPTION OF THE PREFERRED EMBODIMENTS Embodiments of the present invention will be described below in the following order based on examples. A. Principle of digital watermark embedding: Digital watermark embedding processing: Digital watermark detection processing: Specific example: D-1. Effect on sound quality: D-2. Digital watermark detection: D-3. Resistance to level modulation: D-4. Resistance to bandpass filter: D-5. Effect on High Efficiency Compression Coding: D-6. Jitter attack resistance: F. Configuration and processing procedure of entire device: Modification: F-1. Modification Example 1: F-2. Modification Example 2: F-3. Modification 3 F-4. Modification Example 4: F-5. Modification 5:

A. Principle of Digital Watermark Embedding: In embedding a digital watermark according to the present invention, first, a target data is subjected to a predetermined orthogonal transformation to obtain, for example, M
Frequency coefficients X _i (k), which are transform coefficients, 0 ≦ k ≦ M
-1 is obtained. Next, two coefficients X _i (a _i ) and X _i (b _i ) are randomly selected from the M frequency coefficients X _i (k).
Is selected, and the difference value d _i is calculated according to equation (1).

[0024]

(Equation 1)

Next, the two frequency coefficients X
a _{i (a i), X i} (b i), respectively, by adding the random number value u _i, v _i, by embedding an electronic watermark, the frequency coefficient after embedding _{X i '(a i),} X _i ′ (b _i ),
These difference values d _i ′ are as shown in Expression (2).

[0026]

(Equation 2)

As is apparent from the equation (2), the difference value d _i 'can be regarded as one in which the difference value d _i is affected by the acceleration of the random-number value u _i, v _i. Therefore, the statistical distribution of the difference value d _i ′ in the state where the digital watermark is embedded is more than the statistical distribution of the difference value d _{i in} the state where the digital watermark is not embedded. It should be more random. That is, two coefficients subjected to addition and subtraction of random numbers are extracted, and
Only when the difference statistics are examined, a characteristic distribution can be obtained.

In this embodiment, the digital watermark is embedded by diffusing the statistical distribution of the difference value of the frequency coefficient by using such a principle. Less than,
The digital watermark embedding process and its detection process in the present embodiment will be described in detail.

B. Digital Watermark Embedding Process: FIG. 1 is an explanatory diagram showing an outline of a digital watermark embedding process as one embodiment of the present invention. Here, a digital watermark embedding process for audio data will be described. First, the original voice data and a random number sequence serving as a detection key (hereinafter referred to as a key random number sequence)
And prepare. Here, the original audio data at the sampling time t is s (t), and the key random number sequence is R.

Next, the original audio data s (t) is divided into a plurality of audio frames of a fixed length. Here, the i-th audio frame is assumed to be _Si . Then, the audio frame S _i
Is subjected to the modified discrete cosine transform (MDCT),
MDCT coefficients X _i that are M frequency coefficients (transform coefficients)
Get.

The MDCT, as shown in FIG.
To obtain DCT coefficients, as a speech frame S _i, 2
M time series samples are used. This is a technique in which the inter-frame distortion can be suppressed by increasing the degree of frequency separation and making adjacent frames overlap each other.

Therefore, i in the original audio data s (t)
For each audio frame S _i , each MDCT coefficient X
_i (k) can be obtained by equation (3).

[0033]

(Equation 3)

Here, the window function w (n) and MD
The CT basis c (k, n) is given by Equations (4) and (5), respectively.
As shown in FIG.

[0035]

(Equation 4)

[0036]

(Equation 5)

Next, from the key random number sequence R, the audio frame S _i
For each time, a different random value a _i (0 ≦ a _i ≦ M−1) is extracted to generate a sequence A _i having M elements A _i (k) as shown in Expression (6).

[0038]

(Equation 6)

For example, when M = 10, if the extracted random number value a _i = 3, A _i = {0,0,0,1,0,0,0,0,0,0} is generated. Is done.

Next, each element A _i (k) of the generated sequence A _i
Is multiplied by a separately generated control random value u _i (| u _i | ≦ p) to generate a series C _i . Here, p is a parameter for controlling the strength of digital watermark embedding.

For example, if the generated control random number value u _i = 5 for the above-mentioned sequence A _i , C _i = {0,0,0,5,0,0,0,0,0, 0｝ is generated.

Next, using the same procedure as described above,
A random number value b _i (0 ≦ b _i ≦ M−1) is extracted from the key random number sequence R to generate a sequence B _i having M elements B _i (k), and each element B of this sequence B _i a _i (k), separately generated control random number _{v i (| u i | ≦} p) by multiplying the, generating a sequence D _i.

For example, when M = 10, if the extracted random value b _i = 7 and the generated control random value v _i = 3, then D _i = {0,0,0,0,0,0 , 0, -3,0,0} are generated.

Next, by adding the elements to each other in series C _i and series D _i, to generate the embedding control sequence W _i.

For example, in the case of the series C _i and D _i described above, W _i = {0,0,0,5,0,0,0, −3,0,0} is generated.

Next, the i-th speech frame S _i, added to the MDCT coefficients of the M previously obtained X _i (k), the embedded control sequence W _i the elements W _i of the (k), respectively By doing so, the digital watermark is embedded, and M embedded MDCs as shown in Expression (7) are obtained.
Obtain the T coefficient X _i ′ (k).

[0047]

(Equation 7)

FIG. 3 shows the processing from generation of such sequences A _i and B _i to addition of the embedded control sequence W _i . In FIG. 3, the graph in each block represents a frequency spectrum.

The processing from the generation of the sequences A _i , B _i to the addition of the embedding control sequence W _i as described above results in M M
From the DCT coefficients X _i (k), 0 ≦ k ≦ M−1, two MDCT coefficients X are randomly selected according to the key random number sequence R.
_{i (a i), X i} (b i) is selected, the set of MDCT coefficients X _i (a _i), with respect to X _i (b _i), respectively, the control random number value u _i, v _i Are added to form a set of embedded MDCT coefficients X _i ′ as shown in equation (8).
(A _i ) and X _i ′ (b _i ) are obtained.

[0050]

(Equation 8)

Thus, a set of MDCT coefficients X
_An electronic watermark is embedded in _i (a _i ) and X _i (b _i ).

Next, in the same procedure as described above, the process from generation of the sequences A _i , B _i to addition of the embedded control sequence W _i is repeated f times using the key random number sequence R. , M M
Of the DCT coefficients X _i (k), embedding of a digital watermark is performed for each of f sets of MDCT coefficients.

The M embedded MDs thus obtained
Next, an inverse modified discrete cosine transform (IMDCT) transform is performed on the CT coefficient X _i ′ (k) to generate an i-th embedded speech frame S _i ′.

The embedding of the digital watermark as described above is performed in the same manner for the other plurality of audio frames S _i , and after obtaining a plurality of embedded audio frames S _i ′, these embedded audio frames S _i are obtained. Are synthesized to generate embedded voice data s ′ (t).

C. Digital Watermark Detection Processing: FIG. 4 is an explanatory diagram showing an outline of a digital watermark detection processing as one embodiment of the present invention. First, embedded audio data s ′ (t) generated by the digital watermark embedding process of FIG.
A key random number sequence R that is the same as the key random number sequence R used in the digital watermark embedding process is prepared.

Next, the embedded voice data s'
(T) is divided into a plurality of embedded voice frames of a fixed length, and the i-th embedded voice frame S _i ′ is subjected to MDCT conversion to obtain M embedded MDCTs.
The coefficient X _i ′ (k), 0 ≦ k ≦ M−1, is obtained.

Next, from the key random number sequence R, a random number value a _i (0 ≦
a _i ≦ M−1), and according to the above equation (6),
Generate a sequence A _i having M elements A _i (k).

Next, the obtained M embedded MDCs
T coefficient X _i ′ (k) and each element A of the generated sequence A _i
By performing convolution integration on _i (k) as shown in Expression (9), one MDCT coefficient X _i ′ (a _i ) is obtained from a set of MDCT coefficients in which a digital watermark is embedded.

[0059]

(Equation 9)

Next, using the same procedure as described above,
Random number value b from key random number sequence R_i(0 ≦ b_i≤M-1)
And M elements B_iSequence B with (k)_iGenerate
M embedded MDCT coefficients X_i’(K)
And the generated series B_iEach element B of _i(K) and the equation (1)
0), as shown in FIG.
The other MDC of the set of embedded MDCT coefficients
T coefficient X_i’(B_i) Is obtained.

[0061]

(Equation 10)

Next, the obtained set of MDCT coefficients X _i ′
The difference value d _i ′ between (a _i ) and X _i ′ (b _i ) is obtained according to equation (11).

[0063]

[Equation 11]

FIG. 5 shows processing from generation of such series A _i , B _i to calculation of the difference value d _i ′. In FIG.
The graph in each block represents a frequency spectrum.

Next, in the same procedure as described above, using the key random number sequence R, the generation of the series A _i , B _{i and} the difference value d _i ′
The process up to the calculation is repeated f times, and a difference value d _i ′ is obtained for each of the f sets of MDCT coefficients in which the digital watermark is embedded.

The above-described calculation of the difference value d _i ′ is similarly performed for the other plurality of speech frames S _i , and a statistical distribution of the difference value d _i ′ is obtained.

Next, the digital watermark embedding process shown in FIG.
Random number sequence R different from the key random number sequence R _z(Z = 1, 2,
…) Are prepared.

Then, using these random number sequences _Rz , in the same procedure as described above, for each random number sequence _Rz , the difference value d _i
Find the statistical distribution of.

[0069] Figure 6 in the case of using a key the random number sequence R and the key number sequence R is a graph showing respectively the statistical distribution of the differential values d _i when using different random number sequence R _z. In FIG.
(A) shows the statistical distribution of the differential values d _i in the case of using a key the random number sequence R, shows the statistical distribution of the difference values d _i in the case of using (b) the random number sequence R _z. The horizontal axis difference value d _i 'are each a value of, and the vertical axis the difference value d _i'
Is the appearance frequency y (d _i ').

When the key random number sequence R is used, the difference value d _i '
Can be expressed from Expression (11) as shown in Expression (12).

[0071]

(Equation 12)

That is, as described above, the embedded MD
CT coefficient X_i’(A_i), X_i’(B_i) Difference value d_i’
Is the MDCT coefficient X before embedding_i(A_i), X_i(B_i)
Difference value d_iIs the control random value u_i, V_iImpact of adjustment
Can be considered to have received Because of this,
Difference value d in a state where the digital watermark is embedded
_i′, As shown in FIG.
The randomness is increased. In other words, randomness increases
As a result, the difference value d_i´ = 0
Is greatly reduced.

On the other hand, a random number sequence R _z (z = 1, 2,
…) Is used, the random number value extracted from the random number sequence is different from that of the key random number sequence R.
The CT coefficient is also the MDCT without digital watermark embedded.
It becomes the coefficient X _i (k). Therefore, the difference value of the electronic watermark with embedded non MDCT coefficients, because not affected by the acceleration of the control random number, the statistical distribution of the difference values d _i is shown in FIG. 6 (b) Thus, the statistical distribution of the difference values of the MDCT coefficients obtained from the original audio data s (t) is substantially the same.

The difference value d _i ′ between the statistical distribution of the difference value d _i ′ when the key random number sequence R is used and the statistical distribution of the difference value d _i when the random number sequence R _z is used is calculated. _When only the appearance frequency y (0) when _i ′ and d _i are 0 is extracted and graphed, the result is as shown in FIG. 7, the horizontal axis is the random number sequence R _z of the number (type) z, the vertical axis represents the difference value by the random number sequence R _z is frequency y _z where 0 (0). In FIG. 7, a random number sequence R ₁₂₈ of z = 128, is used as a key the random number sequence R.

As described above, when the key random number sequence R is used, the frequency of appearance of the value near the difference value d _i ′ = 0 is greatly reduced, so that the key random number sequence R is used as shown in FIG. The appearance frequency of the difference value 0 in the case is significantly lower than the appearance frequency of the difference value 0 when the other random number sequence _Rz is used, and this fact makes it possible to add the key random number sequence to the embedded voice data s ′ (t). It can be proved that the digital watermark based on R has been embedded.

D. Specific example: When embedding an electronic watermark in high-quality music data such as a CD as audio data, there is little deterioration in sound quality, and even if the data amount is highly efficiently compressed, the embedded electronic watermark is not easily lost. desired. Therefore, in a specific example described below, the effect on the sound quality when a digital watermark is embedded in high-quality music data and the case where digital signal processing such as high-efficiency compression using so-called MP3 (MPEG Audio Layer III) is performed are performed. ,
Consider the effects of digital watermarking.

In this specific example, the target audio data is music data (Classic, Jazz, Danc).
e) (3 kinds of reproduced sounds) were digitized at 44, 1 kHz and 16 bits. Usually,
Although the music data is a stereo sound, for the sake of simplicity of discussion, only one-sided components are used in this specific example. Further, as the random number sequence _Rz , 256 kinds of pseudo random number sequences (R ₁
To R ₂₅₆ ). These random number sequences are usually generated by setting a secret key in a pseudo random number generator.

As a sound quality evaluation method, the most basic signal-to-quantization noise ratio S is used as an objective evaluation scale of sound quality.
NR (Signal to quantization Noise Ratio) is known. The SNR evaluation expression is defined as Expression (13) using the input speech So (m) and its quantization error Er (m).

[0079]

(Equation 13)

In this specific example, the SNR is improved (Segmental SNR) to improve the correspondence with the subjective evaluation.
Was used. SNRseg is obtained as shown in equation (14).

[0081]

[Equation 14]

Here, N _h represents the number of frames in the measurement section, and SNR _h is the SNR in h frames. In this specific example, the length of one frame is set to 32 ms. Also, a speech frame without error, that is, SNR _h =
The measurement was performed excluding the voice frame of ∞.

D-1. Effect on sound quality: When examining the sound quality when a digital watermark is embedded in the target audio data,
The following results were obtained.

First, when p = 5 and f = 1 and the frame length M was changed, the SNRseg changed as shown in FIG. From this result, it can be seen that setting the frame length M to a large value does not significantly improve the sound quality even though the embedding frequency is reduced. This is considered because the influence of the embedding control is distributed over the entire frame.

Next, when M = 1024 and f = 1, and the embedding strength parameter p is changed, the SNRseg
Changed as shown in FIG. From this result, it is understood that the influence on the sound quality can be reduced as the embedding strength parameter p is smaller. This is because as the embedding strength parameter p is small, the control random number used for embedding (u _i, v _i) caused by is generally small.

Further, when M = 1024 and p = 5, and the embedding frequency f in the audio frame is changed, SN
Rseg changed as shown in FIG. From this result, it can be seen that as the embedding frequency f increases, the sound quality decreases, but the decrease rate gradually decreases.

In the digital watermark embedding method of this embodiment,
Since different frequency components are operated for each voice frame, it is predicted that unpredictable distortion will occur between voice frames. The effect of the embedding on the audio waveform by comparing the audio waveform of the original audio data with the audio waveform of the audio data in which M = 1024 and f = 1 and embedding a digital watermark is examined as shown in FIG. The result was obtained. In FIG. 11, (a) shows Class
An audio waveform extracted from a part (8000 samples) of the ic violin player, (b) is an audio waveform in which a digital watermark is embedded, and (c) is a difference waveform between them. From this result, it can be seen that unnatural waveform distortion does not occur between audio frames. This is the frequency conversion for embedding, MD
It is considered that the use of CT caused adjacent speech frames to interfere with each other. Further, even if a person actually listens, no unnatural distortion can be felt from the reproduced voice in which the digital watermark is embedded.

D-2. Watermark Detection: Next, as the audio data to be, using the Classic, as a key the random number sequence, using the random number sequence R ₁₂₈ of z = 128, M = 10
When embedding of a digital watermark was performed with 24, p = 5, f = 5, and it was checked whether or not the digital watermark could be correctly detected from the audio data, the results shown in FIGS. 12 and 13 were obtained.

In FIG. 12, (a) shows a difference value d _i obtained from audio data in which a digital watermark is not embedded.
Shows the statistical distribution of, (b) show a statistical distribution of the differential values d _i obtained from the audio data embedded digital watermark. From this result, it can be seen that the distribution of FIG. 12B in which the digital watermark is embedded has clearly increased randomness as compared with the distribution of FIG. 12A.

[0090] FIG. 13 is the difference value d _i obtained for random number sequence R ₁ to R ₂₅₆ is a graph showing the frequency y _z (0) in case of 0. In FIG. 13, the horizontal axis is a random number sequence R _z
The vertical axis represents the appearance frequency y _z (0) when the difference value of the random number sequence R _z is 0. From this result, the appearance frequency y _z (0) of the difference value 0 according to the random number sequence R _z other than the key random number sequence R (that is, the random number sequence R ₁₂₈ ) is
It turns out that it becomes about 800. This indicates that the statistical distribution of the difference values obtained from the set of MDCT coefficients in which the digital watermark is not embedded is biased toward zero. on the other hand,
When the digital watermark is detected using the correct key random number sequence R (that is, the random number sequence R ₁₂₈ ), the appearance frequency y in which the difference value is 0 is obtained.
_It can be seen that _z (0) has decreased to about 250. Thus, the position of the key random number sequence R (that is, z = 128)
) And the position where the appearance frequency _yz (0) of the difference value of 0 is low, it can be proved that the digital watermark exists.

The digital watermark embedding method according to the present embodiment embeds a digital watermark into different f sets of MDCT coefficients (frequency coefficients) for each audio frame. Therefore, in order to illegally extract the digital watermark embedded in this way, it is necessary to reliably identify the combination of random numbers (a _i , b _i ) in order to derive the set of MDCT coefficients. In this embodiment, since M MDCT coefficients are generated from one voice frame, there are M ² combinations of MDCT coefficients that can be embedded. Therefore, assuming that there are I voice frames, if a digital watermark is embedded in a set of MDCT coefficients for each voice frame, random numbers (a _i , b _i) necessary to extract the digital watermark ) ^Are M ^2I combinations. For example, when M = 1024 and I = 50, the total number of combinations is about ¹⁰³⁰⁰ . Therefore, it is difficult for a third party to accurately specify the MDCT coefficient in which the digital watermark is embedded.

That is, according to the digital watermark embedding method of the present embodiment, the digital watermark can be detected only when the key random number sequence used for embedding matches, so that the key random number sequence used for embedding is known. It is extremely difficult for an unauthorized third party to identify the presence of the digital watermark by unauthorized means and remove the digital watermark information without substantially deteriorating the sound quality. Therefore, the use of the digital watermarking technique of the present embodiment can achieve high confidentiality.

D-3. Immunity to level modulation:
Using Classic as the target audio data,
As a key the random number sequence, using the random number sequence R ₁₂₈ of z = 128, M
= 1024, p = 5, f = 5, embedding of a digital watermark is performed, and the waveform spectrum of the audio data is converted into a constant
When the digital watermark was amplified or attenuated at 0.5 and 1.5 to check whether the digital watermark disappeared, the result shown in FIG. 14 was obtained. From this result, when γ = 0.5, the value of the frequency spectrum is small, so that the appearance frequency y of the difference value of 0 is obtained.
_It can be seen that _z (0) generally increases and the entire graph goes up. On the other hand, when γ = 1.5, the value of the frequency spectrum increases and the appearance frequency _yz (0) decreases, so that it can be seen that the entire graph falls. However, in each case, since the entire graph was affected, the existence of the digital watermark could be determined. Therefore, according to the digital watermark embedding method of this embodiment, if the level modulation is such that the randomness of the difference value obtained from the set of MDCT coefficients is maintained, the digital watermark does not disappear and the audio waveform It has resistance to amplitude level modulation.

D-4. Immunity to bandpass filters:
In the present embodiment, the information of the digital watermark exists as a set of MDCT coefficients distributed over the entire band. Even if attacked by a band-pass filter, in principle, information embedded outside the limited band is not affected by the filter. Generally, the pass band width of the band-pass filter is set wider than the limit band in order to maintain high sound quality. Therefore, most of the information of the digital watermark is not affected by the band-pass filter.

Then, as audio data to be processed, C
used Lassic, as a key the random number sequence, using the random number sequence R ₁₂₈ of z = 128, performs embedding of watermark as M = 1024, p = 5, f = 5, according to the discrete Fourier transform, which is commonly used Using a bandpass filter,
When the bandwidth of the reproduced sound was limited to 5 kHz to 17 kHz and it was examined whether or not the digital watermark disappeared, the result shown in FIG. 15 was obtained. From this result, it can be seen that the digital watermark embedding method of this embodiment has resistance to band-pass filter processing.

However, if both sets of MDCT coefficients are included in the restricted band, the difference value is close to 0, so that the appearance frequency of the difference value 0 obtained using the key random number sequence R , A difference value 0 obtained by using a random number sequence other than the key random number sequence R
It is expected that the difference between the occurrence frequency of and will decrease. Therefore, if an attack is performed by setting the pass band of the band-pass filter narrower in exchange for the deterioration of the sound quality, it becomes difficult to detect the digital watermark. However, if it is possible to determine the band whose transmission is restricted by the band-pass filter from the intensity of the frequency component, it is possible to cope with such an attack.

D-5. Impact on high efficiency compression coding:
Next, MP3 (MPEG Audio Layer), which is a high-efficiency compression coding method based on frequency conversion and sub-band coding
The resistance to advanced digital signal processing was examined using the algorithm of III). Code compression by MP3 is widely used as a high-quality music distribution tool using the Internet.

More specifically, Classic is used as the target audio data, and z = 12 is used as the key random number sequence.
With 8 random number sequence _{R 128, M = 1024, p} = 5, f =
In the case of embedding a digital watermark as No. 5 and compressing / expanding the audio data by MP3 and detecting a digital watermark from the reproduced audio, a result as shown in FIG. 16 was obtained. As the MP3, an MP3 having a bit rate of about 1/10, which is a standard compression rate used on the Internet, was used. From this result, although the detection rate of the digital watermark is slightly reduced due to the removal of the redundant component by the compression encoding of MP3, the appearance frequency y in which the difference value by the key random number sequence is 0 is y.
It was confirmed that _z (0) generally shows a low value. Therefore, the digital watermark embedding method of the present embodiment is considered to be resistant to some degree of information compression.

D-6. Resistance to Jitter attacks:
Next, it was examined whether the digital watermark was lost by a Jitter attack known as an attack for destroying the digital watermark embedded in the audio data. The Jitter attack is an attack method that unspecifiedly increases or decreases the frequency band of music by about 0.1%, and uses a feature that human hearing cannot perceive minute pitch fluctuations. It is considered to be effective for attacking the digital watermark by the used patchwork method or the like.

More specifically, Classic is used as the target audio data, and z = 12 is used as the key random number sequence.
With 8 random number sequence _{R 128, M = 1024, p} = 5, f =
In the case of embedding a digital watermark as No. 5 and performing a Jitter attack on the audio data and detecting an electronic watermark from the audio data, a result as shown in FIG. 17 was obtained. From this result, it can be seen that the digital watermark embedding method of this embodiment can withstand Jitter attacks.

E. Configuration and Processing Procedure of Entire Apparatus: The configuration of the digital watermark apparatus 10 that executes the above-described digital watermark embedding processing and digital watermark detection processing will be described with reference to FIG. FIG. 18 is a block diagram showing a configuration of the digital watermark device 10 as one embodiment of the present invention. The digital watermarking device 10 includes a CPU 22, a RAM 24, an RO,
M26, a keyboard 30, a mouse 32, a display device 34 composed of a CRT or the like, and a hard disk device 36
And a communication device 38 including a network card and a modem, and a bus 40 connecting these components. In FIG. 1, various interface circuits are omitted. The communication device 38 is connected to a computer network via a communication line (not shown). A server (not shown) of the computer network has a function as a program supply device that supplies a computer program to the digital watermark device 10 via a communication line.

The RAM 24 has an MDCT converter 42,
A computer program for realizing each function of the selection / calculation unit 44, the IMDCT conversion unit 46, and the detection unit 48 is stored. Each of these parts 42, 44, 46,
The 48 functions are as already described in detail.
That is, the MDCT conversion unit 42 performs MDCT conversion on the audio data to obtain MDCT coefficients. Selection / calculation unit 44
Selects a set of two MDCT coefficients from the MDCT coefficients according to the random number value extracted from the key random number sequence, and assigns two MDCT coefficients to the two MDCT coefficients for each selected set.
Adds the three control random numbers to the embedded MD
Obtain CT coefficients. The IMDCT transform unit 46 performs an IMDCT transform on the embedded MDCT coefficients to return to the audio data. Also, when detecting a digital watermark,
The MDCT conversion unit 42 performs an MDCT conversion on the audio data to obtain an MDCT coefficient, and the selection / calculation unit 44 sets two of the MDCT coefficients as one set according to the random number value extracted from the key random number sequence. A set of MDCT coefficients to be selected is selected, and two MDCT coefficients are subtracted from each other for each selected set to obtain a difference value. The detection unit 48 detects the presence of a digital watermark from the obtained appearance frequency of the difference value.

A computer program for realizing the functions of the units 42, 44, 46 and 48 is provided in a form recorded on a computer-readable recording medium such as a flexible disk or a CD-ROM. The computer reads the computer program from the recording medium and transfers it to an internal storage device or an external storage device. Alternatively, a computer program may be supplied to a computer via a communication path. When implementing the functions of the computer program, the computer program stored in the internal storage device is executed by the microprocessor of the computer. Further, a computer may read a computer program recorded on a recording medium and directly execute the computer program.

In this specification, a computer is
The concept includes a hardware device and an operation system, and means a hardware device that operates under the control of the operation system. In the case where an operation system is unnecessary and a hardware device is operated by an application program alone or firmware alone, the hardware device itself corresponds to a computer. The hardware device includes at least a microprocessor such as a CPU and means for reading a computer program recorded on a recording medium. For example, a CD-RAM drive or DVD-R
When an electronic device such as an AM drive incorporates a CPU, a ROM, and the like, and the electronic device has a function as a computer, the electronic device is naturally included in the concept of a computer. The computer program includes a program code that causes such a computer to realize the functions of the above-described units. Some of the above functions are not application programs,
It may be realized by an operation system. Further, the program for performing the digital watermark embedding processing and the detection processing may be added in a plug-in format to the audio processing program. Also,
The above-described pseudo-random number generator that generates the random number sequence and the control random value may be configured by hardware,
It may be configured by software.

The "recording medium" in the present invention includes a flexible disk, a CD-ROM, a magneto-optical disk, an IC card, a ROM cartridge, a punched card, a printed matter on which a code such as a bar code is printed, and an internal storage of a computer. Device (RAM, ROM, etc.)
Various computer-readable media such as an external storage device and the like can be used.

FIG. 19 is a flowchart showing the procedure of the digital watermark embedding process performed by the MDCT conversion unit 42, the selection / calculation unit 44, and the IMDCT conversion unit 46.

In step S102, the MDCT converter 4
2 prepares original voice data and a key random number sequence. Step S
At 104, the MDCT conversion unit 42 converts the original audio data into M
A plurality of MDCT coefficients are obtained by performing a DCT transform. In step S106, the selection / calculation unit 44 selects a set of two MDCT coefficients from the plurality of MDCT coefficients according to the random number value extracted from the key random number sequence, and selects the selected set of MDCT coefficients. Each time, two separately generated control random numbers are added to the two MDCT coefficients to obtain embedded MDCT coefficients. In step S108, the IMD
The CT conversion unit 46 performs the IMDCT transform on the plurality of embedded MDCT coefficients and returns to the audio data, thereby obtaining the embedded audio data.

FIG. 20 is a flowchart showing the procedure of the digital watermark detection process performed by the MDCT conversion unit 42, the selection / calculation unit 44, and the detection unit 48.

In step S202, the MDCT converter 4
2 prepares the embedded voice data and the same key random number sequence as at the time of embedding. In step S204, the MDCT conversion unit 42 performs MDCT conversion on the embedded audio data to obtain a plurality of MDCT coefficients. Step S206
Then, the selection / calculation unit 44 selects a set of two MDCT coefficients from a plurality of MDCT coefficients in accordance with the random number value extracted from the key random number sequence, and also selects a set of MDCT coefficients for each selected set. And subtracting each of the two MDCT coefficients to obtain a difference value. In step S208, the detection unit 48 detects the presence of a digital watermark from the obtained appearance frequency of the difference value.

F. Modifications: The present invention is not limited to the above-described examples and embodiments, and can be carried out in various modes without departing from the gist thereof. For example, the following modifications are also possible. is there.

F-1. Modified Example 1: In the embodiment described above,
Although the key random number sequence used when embedding the digital watermark is one type, the digital watermark is multiplexed into the same content, that is, the same audio data, using two or more types of key random number sequences. You may do it. According to the digital watermark embedding method of the present invention, a set of two transform coefficients is selected at random, and a difference value between the two transform coefficients is calculated as follows.
This is to spread the distribution bias. Therefore, even if digital watermarks are multiplexed and embedded in the same content, they do not affect each other if the combination of transform coefficients (that is, the combination of random numbers (a _i , b _i )) is different. Even if a digital watermark is accidentally embedded in the same set of transform coefficients, the values added to the difference values are random, so that they rarely disappear completely. Therefore, even if the embedded digital watermark partially overlaps, it is considered that there is almost no influence on the detection result of the digital watermark.

FIG. 21 shows the same contents (that is, the same contents).
Two kinds of key random number sequences R ₁, R_TwoUsing electron
FIG. 9 is an explanatory diagram showing an example of a case where a watermark is embedded multiple times.
You. In FIG. 21, (a) shows a state in which a digital watermark is embedded.
(B) shows a key random number sequence R₁To use
Difference value for each frame when child watermark is embedded
And (c) shows a key random number sequence R_TwoUsing
Difference value for each frame when child watermark is embedded
2 shows the statistical distribution of. In (b) and (c), the frame
The horizontal axis of each system is the difference value, and the vertical axis is the difference.
This is the frequency of occurrence of the minute value.

In the example shown in FIG. 21, when embedding a digital watermark in the same content using two types of key random number sequences R ₁ and R ₂ , the presence or absence of embedding is controlled for each frame. In this way, a plurality of pieces of bit information are embedded as rectangular waves.

It should be noted, however, that embedding a digital watermark in a multiplex manner in this manner also increases the effect on sound quality.

As a specific example in which such a digital watermark is multiplexed, for example, as audio data to be processed,
Using Classic, two types of a random number sequence R ₆₄ of z = ₆₄ and a random number sequence R ₁₉₂ of z = 192 as a key random number sequence,
Assuming that M = 1024, p = 5 and f = 5, a digital watermark is multiplexed and the key data random number sequence R ₆₄ , R
_When the digital watermark was detected using ₁₉₂ , the result as shown in FIG. 22 was obtained. From this result, it is understood that the digital watermark can be multiplexed by using the key random number sequence related to the embedding of the digital watermark.

F-2. Modification 2 In the above embodiment,
Although the MDCT transform was used as the orthogonal transform, DCT, F
Other types of orthogonal transforms such as FT and wavelet transform can be employed.

F-3. Modification 3 In the above-described embodiment,
In embedding the digital watermark, two transform coefficients are selected as one set. However, three or more transform coefficients may be selected as one set and the digital watermark may be embedded.

In this case, when detecting a digital watermark, a desired operation is performed between three or more variable coefficients, and the presence of the digital watermark is detected based on the appearance frequency of the calculated value. You may do it.

F-4. Modification 4: In the above embodiment,
In embedding the digital watermark, the control random number value is added to the transform coefficient, but may be subtracted. Further, multiplication, division, and other various operations may be performed on the control random number.

F-5. Modification 5: In the above embodiment,
When embedding the digital watermark, the control random value is added to each of the two transform coefficients in the set, but the control random value may be added to only one transform coefficient. As is clear from equation (12), for example, MDCT
The coefficients X _i (a _i), by adding the value of a control random value (u _i -v _i), without anything added to control random value to the MDCT coefficients X _i (b _i), pre-embedded X _i '(a _i ), X _i '
This is because the difference value d _i ′ of (b _i ) becomes d _i + u _i −v _i , and the same value as in equation (12) is obtained. The same applies to the case where three or more transform coefficients are set as one set.

F-6. Modification 6: In the above embodiment,
Although a digital watermark is embedded in audio data, the present invention is not limited to this, and can be applied to a case where a digital watermark is embedded in image data. That is, the present invention is generally applicable to a case where a digital watermark is embedded in digital data. Note that the original audio data and original image data in which the digital watermark is embedded can be simply referred to as “target data”.

[Brief description of the drawings]

FIG. 1 is an explanatory diagram showing an outline of a digital watermark embedding process as one embodiment of the present invention.

FIG. 2 is an explanatory diagram showing a window function in MDCT.

3 is an explanatory diagram showing the processing contents to the addition of series A _i, the control sequence W _i embedding the generation of B _i in FIG.

FIG. 4 is an explanatory diagram showing an outline of a digital watermark detection process as one embodiment of the present invention.

5 is a diagram showing a difference value d from the generation of the sequences A _i and B _i in FIG. 4.
FIG. 9 is an explanatory diagram showing the contents of processing up to the calculation of _i ′.

[6] When using key random number sequence R and the key number sequence R is a graph showing respectively the statistical distribution of the differential values d _i when using different random number sequence R _z.

FIG. 7 is a graph showing an appearance frequency of a difference value 0 for each random number sequence.

FIG. 8 is a graph showing a change in sound quality with respect to a frame length M;

FIG. 9 is a graph showing a change in sound quality with respect to an embedding strength parameter p.

FIG. 10 is a graph showing a change in sound quality with respect to a frequency f of embedding in an audio frame.

FIG. 11 is an explanatory diagram for explaining an effect on an audio waveform due to embedding of a digital watermark.

FIG. 12 is an explanatory diagram showing a comparison between statistical values of difference values obtained from audio data in which an electronic watermark is not embedded and difference values obtained from audio data in which an electronic watermark is embedded.

FIG. 13 is a graph showing an appearance frequency of a difference value 0 obtained for random number sequences R _{1 to} R ₂₅₆ from audio data in which a digital watermark is embedded.

FIG. 14 is a graph showing an appearance frequency of a difference value 0 for each random number sequence when level modulation is performed.

FIG. 15 is a graph showing the appearance frequency of a difference value 0 for each random number sequence when band limitation is performed by a band-pass filter.

FIG. 16 is a graph showing the appearance frequency of a difference value 0 for each random number sequence when the data is compressed and expanded by MP3.

FIG. 17 is a graph showing an appearance frequency of a difference value 0 for each random number sequence when a Jitter attack is performed.

FIG. 18 is a digital watermarking apparatus 1 as one embodiment of the present invention.
FIG. 3 is a block diagram showing a configuration of a 0.

FIG. 19 is a flowchart showing a procedure of a digital watermark embedding process performed by the MDCT conversion unit 42, the selection / calculation unit 44, and the IMDCT conversion unit 46.

20 is a flowchart illustrating a procedure of a digital watermark detection process performed by the MDCT conversion unit 42, the selection / calculation unit 44, and the detection unit 48. FIG.

FIG. 21 shows two types of key random number sequences R ₁ ,
FIG. 9 is an explanatory diagram showing an example of a case where a digital watermark is multiplexed and embedded using R ₂ .

FIG. 22 is a graph showing an appearance frequency of a difference value 0 for each random number sequence when multiple embedding of a digital watermark is performed.

[Explanation of symbols]

 DESCRIPTION OF SYMBOLS 10 ... Digital watermarking device 22 ... CPU 24 ... RAM 26 ... ROM 30 ... Keyboard 32 ... Mouse 34 ... Display device 36 ... Hard disk device 38 ... Communication device 40 ... Bus 42 ... MDCT conversion part 44 ... Selection / calculation part 46 ... IMDCT conversion Unit 48 Detector

Claims (12)

[Claims] 1. A digital watermark embedding method for embedding a digital watermark in digital data, comprising: (a) target data for embedding the digital watermark; and a key random number sequence used for embedding the digital watermark. (B) performing a predetermined orthogonal transform on the target data to obtain a plurality of transform coefficients; and (c) responding to random number values sequentially extracted from the key random number sequence. Selecting a set of n (n is an integer of 2 or more) transform coefficients from the plurality of transform coefficients; and (d) selecting n transform coefficients for each selected set. Performing a predetermined operation on at least one conversion coefficient using a separately obtained control random number value; and (e) applying a plurality of conversion coefficients including the operation-performed conversion coefficient to the plurality of conversion coefficients opposite to the orthogonal transformation. By performing the conversion of The electronic watermark embedding method characterized by comprising the step of returning to the target data. 2. The digital watermark embedding method according to claim 1, wherein the step (b) includes a step of performing a modified discrete cosine transform as the predetermined orthogonal transform. 3. The digital watermark embedding method according to claim 1, wherein the step (c) includes a step of selecting a set of two transform coefficients from the plurality of transform coefficients. The digital watermark embedding method, wherein the step (d) includes a step of adding or subtracting two separately obtained control random numbers to the two transform coefficients for each selected set. 4. The digital watermark embedding method according to claim 1, wherein the step (a) includes a step of preparing a plurality of the key random number sequences, and the step (c) includes the steps of: A step of selecting a set of transform coefficients of n (n is an integer of 2 or more) from the plurality of transform coefficients in accordance with random numbers sequentially extracted from the key random number sequence. A digital watermark embedding method characterized by including: 5. A digital watermark detection method for detecting the digital watermark from digital data in which a digital watermark is embedded, comprising: (a) target data for which the digital watermark is to be detected; A key random number sequence used for
(B) performing a predetermined orthogonal transform on the target data to obtain a plurality of transform coefficients; and (c) determining a plurality of transform coefficients in accordance with random numbers sequentially extracted from the key random number sequence. A step of selecting a set of transform coefficients in which n (n is an integer of 2 or more) one set from a plurality of transform coefficients; and (d) for each of the selected sets, n transform coefficients are selected. A digital watermark detection method, comprising: performing a predetermined calculation; and (e) detecting the presence of the digital watermark from the appearance frequency of the obtained calculation value. 6. The digital watermark detection method according to claim 5, wherein the step (b) includes a step of performing a modified discrete cosine transform as the predetermined orthogonal transform. 7. The digital watermark embedding method according to claim 5, wherein the step (c) includes a step of selecting a set of two transform coefficients from the plurality of transform coefficients. The digital watermark embedding method, wherein the step (d) includes a step of adding or subtracting the two transform coefficients for each selected set. 8. An electronic watermarking apparatus for embedding an electronic watermark in digital data, wherein an orthogonal transformation for obtaining a plurality of transformation coefficients is performed by performing a predetermined orthogonal transformation on target data for embedding the electronic watermark. And selecting a set of n (n is an integer of 2 or more) transform coefficients from the plurality of transform coefficients according to random numbers sequentially extracted from a key random number sequence prepared in advance. A selecting unit to perform, and for each selected set, at least one of n conversion coefficients.
An operation unit that performs a predetermined operation on two transform coefficients with a control random value obtained separately, by performing the inverse transform of the orthogonal transform on the plurality of transform coefficients including the transformed transform coefficient And an inverse orthogonal transform unit for returning to the target data. 9. A digital watermarking apparatus for detecting the digital watermark from digital data in which a digital watermark is embedded, wherein a predetermined orthogonal transformation is performed on target data for which the digital watermark is to be detected, thereby obtaining a plurality of digital watermarks. A set of n (n is an integer of 2 or more) out of the plurality of transform coefficients according to an orthogonal transform unit for obtaining a transform coefficient and a random number value sequentially extracted from a key random number sequence prepared in advance. A selecting unit for selecting a set of transform coefficients; a calculating unit for performing a predetermined calculation between n transform coefficients for each of the selected sets; An electronic watermarking device, comprising: a detecting unit that detects. 10. A computer program for embedding a digital watermark in digital data, wherein a function of obtaining a plurality of transform coefficients by performing a predetermined orthogonal transformation on target data for embedding the digital watermark. And selecting a set of transform coefficients in which n (n is an integer of 2 or more) is set from the plurality of transform coefficients in accordance with random numbers sequentially extracted from a key random number sequence prepared in advance. Function and, for each selected set, at least one of the n transform coefficients
A function of performing a predetermined operation with a control random number value obtained separately for one of the conversion coefficients, and by performing the inverse conversion to the orthogonal conversion on the plurality of conversion coefficients including the conversion coefficient on which the operation has been performed, A computer program that allows a computer to implement the function of returning the target data and 11. A computer program for detecting a digital watermark from digital data in which a digital watermark is embedded, wherein the computer program performs a predetermined orthogonal transformation on target data for which the digital watermark is to be detected. And a function of converting n (n is an integer of 2 or more) from the plurality of transform coefficients into one set according to a random number value sequentially extracted from a key random number sequence prepared in advance. A function of selecting a set of coefficients, a function of performing a predetermined operation between n transform coefficients for each selected set, and a function of detecting the presence of the digital watermark from the appearance frequency of the obtained operation value And a computer program for realizing the following on the computer. 12. A computer-readable recording medium on which the computer program according to claim 10 is recorded.