KR20020079242A - Audio watermark embedding and extracting method robust against time axis attack and apparatus thereof - Google Patents

Abstract

PURPOSE: A method for inserting and extracting an audio watermark resistant to a time base attack, and apparatus thereof are provided to perfectly protect a CONSTITUTION: A method for inserting an audio watermark resistant to a time base attack includes the steps of transforming an audio signal into the audio signal of a time-invariant domain not affected by signal transformation at a time base, inserting a watermark signal into the audio signal, transforming the audio signal having been inserted with the watermark signal into the audio signal of a time domain through a time-invariant invert transformation. A method for extracting the audio watermark includes the steps of transforming the audio signal having been inserted with the watermark into the audio signal of the time-invariant domain not affected by signal transformation at the time base, XOR-operating an estimated diffusion watermark component calculated from the transformed audio signal with a pseudo random sequence, reconstructing a bit stream by level discrimination applying a threshold to the transformed signal.

Description

AUDIO WATERMARK EMBEDDING AND EXTRACTING METHOD ROBUST AGAINST TIME AXIS ATTACK AND APPARATUS THEREOF}

The present invention relates to a method and apparatus for embedding and extracting an audio watermark that is robust against time-base attacks, and more particularly, to converting watermarks into time-invariant regions in order to allow watermarks to remain in various signal processing attacks on the time-base. A watermark embedding and extraction method and apparatus therefor.

Digital watermarking technology is a method of inserting copyright information of digital content into a watermark and using it for future claims, a method of extracting a watermark of the content during piracy and distribution, and using it to detect illegal distributors. It has various application fields such as the method used to verify the integrity and sender of contents by checking whether there is any modification or manipulation of the contents, and the method used to prevent reproduction and control of copying of contents.

The results of research on digital watermarking technology began to be published in 1990, when researchers from K. Tanaka and others published a technique for embedding watermarks into digital images. Actively progressed. While the early technology has a very simple form as a watermark embedding technology in space or time plane, the spread spectrum method developed by researchers such as IJ Cox has the potential to survive watermark information in digital content compression technology. It is recognized in a robust way. Because watermark embedding in space or temporal planes is vulnerable to nonlinear attacks or compression, many researchers have developed watermark embedding techniques in other transformation planes. Audio watermarking has also been developing watermark embedding technologies in the time and conversion planes, and a variety of technologies have been developed in accordance with the organization of SDMI (Secure Digital Music Initiative). These various techniques can be summarized as follows according to the insertion method.

Low bit coding

This method is a method of arbitrarily manipulating the least significant bit (LSB) of a digital audio signal according to watermark information. This method has the advantage of preserving the sound quality of the audio close to the original sound, but has the disadvantage that it is easily broken in general signal processing such as noise addition, sample frequency variation, compression and filtering.

2. Phase coding

This method converts the digital audio signal into the frequency plane and embeds a watermark in the phase spectrum. The phase coding method uses the property that the auditory system is insensitive to the phase change. The signal-to-noise ratio is good, but the watermark can be detected only by synchronizing correctly.

Spread spectrum coding

This method spreads and inserts narrowband information so that a watermark signal exists over the entire band of frequencies. It utilizes the technology used in the communication system and generates a pseudo-random signal with an arbitrary key value and utilizes it as a watermark. This method has very strong characteristics such as compression and filtering, but it requires the accumulation of considerable digital data in order to obtain accurate results, and it can be detected from various attacks only when the exact insertion position is known.

4. Echo embedding

This method uses the watermark by adding them to the original signal with a certain delay time. However, this method has a disadvantage in that the watermark can be easily removed by the sound effect and adversely affects the sound quality.

5. Patchwork

This method inserts a watermark by adding a certain data set to the original signal so that the average value is +1, and adds another data set to the original signal so that the average value is -1. Although this method is effective as a probabilistic method, it requires a large number of iterative operations in order to secure reliability, and has a disadvantage in that the position must be accurately identified. This method can be used in both time and conversion planes.

The above-described prior arts have not been developed as a technique capable of defending all the various attacks attempted to remove the watermark, and attempted to solve the problem by using two or more algorithms to solve the above problems. However, as the reason for selecting the standard of SDMI is to apply to PD (Portable Device), the algorithm is required to be simplified, and to satisfy this problem, it has to be vulnerable to any part of attack.

In addition, most of the above-described prior arts have disadvantages that are very vulnerable to time-base deformation attacks, and techniques that can solve these problems have disadvantages that are vulnerable to attacks such as compression. Examples of time-base deformation attacks include linear speed change and time scaling attacks. The linear velocity change causes the tape to stretch or run fast, making most of the watermarked signals spread in the spread spectrum method. A temporal scaling attack is one that increases or decreases the length of music without changing the timbre.

SUMMARY OF THE INVENTION The present invention has been made to solve the above-mentioned problems, and an object of the present invention is to create a watermark in the form of a spread spectrum and insert the copyright or control information into an audio signal transferred to an area which is not affected by a time-base deformation attack, thereby providing a time-base. It is to implement a watermark insertion and extraction method that is not affected by the deformation attack.

Another object of the present invention is to defend against time-base strain attack, which is a weak point of spread spectrum coding technology, while maintaining the strengths of compression, nonlinear filtering, etc. of spread spectrum coding technology in watermarking.

1 is a block diagram showing the configuration of a watermark embedding apparatus according to the present invention.

FIG. 2 is a block diagram illustrating an internal configuration of the time invariant converter of FIG. 1.

3 is a block diagram showing an internal configuration of the spread watermark generator of FIG.

FIG. 4 is a block diagram illustrating an internal configuration of the time invariant inverse converter of FIG. 1.

5 is a block diagram showing the configuration of a watermark extraction apparatus according to the present invention.

FIG. 6 is a block diagram showing an internal configuration of the watermark extractor of FIG.

Explanation of symbols on the main parts of the drawings

10: time invariant converter 12: time axis scaling unit

14: interpolation filter 16: time invariant conversion unit

20: diffuse watermark generator 22: time base expander

24: pseudo random sequence generator 26: XOR operator

30: adder 40: time-invariant inverse converter

42: time invariant inverse transform unit 44: extrapolation filter

46: time axis scaling unit 110: time invariant converter

120: pseudo random sequence generator 130: multiplier

140: watermark extractor 142: threshold setting unit

144: level determining unit 146: time base systolic

In order to achieve the object of the present invention, the present invention provides a method for embedding a watermark signal into an audio signal, comprising the steps of: (a) converting the audio signal into a time-invariant region which is not affected by signal distortion in the time axis; (b) embedding a watermark signal in the audio signal converted into the time invariant region; And (c) converting the audio signal into which the watermark signal is inserted is converted into an audio signal in the time domain by inversely inversely converting the watermark signal.

Here, step (a) includes (a1) converting an audio signal from a linear scale to a logarithmic scale; And (a2) Fourier transforming a signal converted to a logarithmic scale to obtain an amplitude spectrum, and (c) comprises: (c1) Fourier inversely converting an audio signal into which a watermark signal is inserted; And (c2) converting the Fourier inverse signal from a logarithmic scale to a linear scale.

In order to achieve another object of the present invention, the present invention provides a method for extracting a watermark from an audio signal having a watermark embedded therein, wherein (a) the watermark-embedded audio signal is not affected by signal distortion on the time axis. Converting to an invariant time invariant region; Performing an XOR operation on the estimated spread watermark component and the pseudo random sequence obtained from the signal converted into the time invariant region; and (c) reconstructing the bit stream as a level discrimination applying a threshold to the resultant signal in step (b).

Here, the step (a) may include: (a1) converting the audio signal having the watermark embedded from a linear scale to a logarithmic scale; And (a2) Fourier transforming the signal converted to the logarithmic scale to obtain an amplitude spectrum.

Prior to describing the watermark embedding and extracting apparatus according to the present invention in detail, first, an area which is not affected by time-base deformation and a conversion to this area will be described.

The original audio signal is called s (t), and the signal with timebase distortion t). here Is A value in the range of zero. In this way, the signal s ( t) acts as a factor that makes it difficult to detect the embedded watermark. Therefore, it is necessary to introduce an area that is not affected by the time axis deformation. Various kinds of transformations may be used for the transformation to the region that is not affected by the time-base deformation, but the Fourier transformation will be used here as an example.

Explain the definition of the common Fourier transform and its transformation characteristics.

The Fourier transform is defined as Equation 1 below.

The definition of the Fourier inverse transform is shown in Equation 2 below.

Meanwhile, the time axis shift of a signal in the time domain is a linear shift of phase components in the frequency domain, which is represented by Equation 3 below.

On the other hand, if the scaling of the signal in the time domain is inverse scaling in the frequency domain, it is represented by Equation 4 below.

In the present invention, a characteristic to be used to implement a region that is not affected by the time axis strain attack is the characteristic of Equation 3, in which the time axis movement of the signal in the time domain does not affect the amplitude spectrum in the frequency domain. Therefore, if the scaling change of the signal in the time domain can also be changed to time axis shift, it will not affect the amplitude spectrum in the frequency domain.

To achieve this, it is useful to change the linear scale in the time domain to a logarithmic scale. Since the form of the product is represented in the form of the sum of the logarithmic functions, the logarithmic function in the time domain is represented by Equation 5 below.

By using the logarithm function as shown in Equation 5, the scaling change of the signal in the time domain is recorded on the time axis of the log scale. You can switch to as many moves. Since the movement in the time axis does not affect the amplitude spectrum as described in Equation 3, the log scale transformation in the time domain and the amplitude spectrum of the signal create a time-invariant region that is not affected by the time-base deformation attack. I can make it.

Hereinafter, exemplary embodiments of the present invention will be described in detail with reference to the accompanying drawings.

1 is a block diagram showing the configuration of a watermark embedding apparatus according to the present invention.

Referring to FIG. 1, the watermark embedding apparatus includes a time-invariant converter 10 for converting an audio signal into a time-invariant region, a diffusion watermark generator 20 for spreading copyright or control information on a time axis. An adder 30 which adds the audio signal converted into the invariant region and the spread copyright or control information signal and a time invariant inverter 40 which returns the audio signal with the watermark in the time invariant back to the time domain. .

FIG. 2 is a block diagram showing an internal configuration of the time invariant converter 10 of FIG.

Referring to FIG. 2, the time invariant converter 10 is composed of a time axis scaling unit 12, an interpolation filter 14, and a time invariant converter 16.

The time axis scaling unit 12 performs a function of converting an input audio signal from a linear scale to a logarithmic scale as shown in Equation 5 above. The audio signal converted to the logarithmic scale is interpolated by the interpolation filter 14 and then input to the time-invariant converter 16. The time-invariant transform section 16 performs an FFT (Fast Fourier Transform) on the signal to obtain an amplitude spectrum, thereby completing the process of converting the audio signal into the time-invariant region.

As a process of calculating the signal to the log scale and amplitude spectrum, the amplitude spectrum is not affected by the scaling deformation attack in the time domain as described above, resulting in a time-invariant region that is not affected by the time-axis deformation attack. The original audio signal is changed.

3 is a block diagram showing an internal configuration of the spread watermark generator 20 of FIG.

Referring to FIG. 3, the spread watermark generator 20 is composed of a time base expander 22, a pseudo random sequence generator 24, and an XOR operator 26. As shown in FIG.

Prior to inserting the copyright or control information into the audio signal, the copyright or control information is made of a signal extended on the time axis in the time axis expander 22.

That is, assuming that the bit string of copyright or control information to be inserted is I (n), the time axis extension of copyright or control information is expressed as in Equation 6 below.

N: number of information bits

k = 0,1,2, ..., RN-1

Here R denotes a magnification for extending the time axis.

The pseudo random sequence generator 24 generates a pseudorandom sequence (hereafter referred to as PN (k)) existing over the entire band of frequencies.

Next, in the XOR operator 26, an XOR operation of the time axis extended copyright or control information I _expanded (k) and the pseudo random sequence PN (k) is performed to spread the watermark signal W (k). Is generated, which is expressed by Equation 7 below.

Referring again to FIG. 1, the audio signal converted from the time invariant converter 10 into the time invariant region and the spread watermark signal generated by the spread watermark generator 20 are added in the adder 30.

That is, the spread watermark signal is inserted in the time-invariant region into the audio signal converted into the time-invariant region, which is expressed as Equation 8 below.

Here, S _ti (k) represents the Fourier transform coefficient of the audio signal converted into the time-invariant region, Is a constant value that determines the embedding strength of the watermark, Multiply by to reduce the noise effect where the signal is weak when the watermark is inserted at a uniform level.

The audio signal in which the watermark in the form of an amplitude spectrum such as Equation 8 is inserted is input to the time invariant converter 40.

FIG. 4 is a block diagram showing the internal configuration of the time-invariant inverse transformer 40 of FIG.

Referring to FIG. 4, the time invariant inverse transformer 40 includes a time invariant inverse transformer 42, an extrapolation filter 44, and a time axis scaling unit 46.

The time invariant inverse transform unit 42 performs a function of converting an audio signal having an amplitude spectrum type watermark embedded therein, that is, an audio signal having a watermark embedded in the time invariant region into an audio signal in the time domain. That is, for example, an Inverse Fast Fourier Transform (IFFT) is performed, which is represented by Equation 9 below.

The watermarked signal s _w (t) in the time domain is extrapolated by the extrapolation filter section 44 and then input to the time axis scaling section 46. The time base scaling unit 46 in the time invariant transformer 40 performs an inverse process of the process performed by the time base scaling unit 12 in the time invariant converter 10. In other words, it converts a log scale signal back to a linear scale signal.

Next, a watermark extraction apparatus for extracting copyright or control information from the watermarked signal will be described.

5 is a block diagram showing the configuration of a watermark extraction apparatus according to the present invention.

Referring to FIG. 5, the apparatus for extracting a watermark includes a time-invariant converter 110 for converting a watermarked audio signal into a time-invariant region, a pseudo-random sequence generator 120 for generating a pseudo-random sequence, and a watermark inserted therein. A multiplier 130 for obtaining a correlation value between the audio signal and a pseudo-random sequence, and a watermark extractor 140 for extracting watermark information (ie, copyright or control information) from the watermark-embedded audio signal.

The configuration of the time-invariant converter 110 of the watermark extraction apparatus is the same as that of the time-invariant converter 10 of the watermark embedding apparatus, so that the signal S _w (t) in which the watermark of the time domain is embedded is It is converted into a signal in the form of an amplitude spectrum of the constant region, which is represented by Equation 10 below.

In the pseudo random sequence generator 120, a pseudo random sequence PN (k) is generated, and a correlation between the pseudo random sequence PN (k) and a signal with a watermark is obtained in the multiplier 130.

FIG. 6 is a block diagram showing an internal configuration of the watermark extractor 140 of FIG.

Referring to FIG. 6, the watermark extractor 140 includes a threshold setting unit 142, a level discriminating unit 144, and a time axis shrinker 146.

The watermark-inserted signal converted into the time-invariant region by the time-invariant converter 110 is repeatedly converted into the same time-invariant region by the watermark extractor 140 for a long time to obtain a harmonic mean, which is represented by the following equation. Represented as 11 and 12.

By the process of Equation 12, the irregular signal is averaged, and the amplitude spectrum remains only for the signal existing throughout the audio signal. Thus, in W _mean (k) Removing only the component for leaves only the component for the watermark.

To actually get the watermark component Is not possible to use the curve fitting method Is an estimate of Is calculated and used, and is represented by Equation 13 below.

Here, e (k) means an error signal.

Contains a spread watermark component, and in order to extract copyright or control information therefrom, the pseudorandom sequence PN (k) generated by the pseudo-random sequence generator 120 is estimated with the estimated spread watermark component and exclusive. An OR is taken, the threshold set by the threshold setting unit 142 is applied to the result, and the level discriminating unit 144 determines the level to reconstruct the bit string. The bit string obtained at this time becomes copyright or control information spread on the time axis, and this process is represented by Equation 14 below.

Here, I ' _expanded (k) represents a reconstruction into a bit string by applying a threshold.

Copyright or control information obtained from equation (14) Is spread over the time base. Accordingly, this spread copyright or control information is subjected to a process of being restored to the original copyright or control information in the time base systolic machine 146, which is represented by the following equation (15).

From here, Denotes a bit value represented by an average value of every R pieces of copyright or control information in a spread form.

Thus, the watermark extraction process of extracting copyright or control information from the watermarked signal is terminated.

Due to the insertion and extraction including the conversion of the watermark into the time-invariant region as described above, the watermark is not removed by the time-base deformation attack.

Although the invention has been particularly shown and described with reference to the above embodiments, it has been used for the purpose of illustration and those of ordinary skill in the art to which the invention pertains have the spirit and scope of the invention as defined in the appended claims. Various modifications can be made without departing.

As described above, according to the present invention, it is effective to implement a watermark embedding and extraction method that is robust against time-base strain attack while maintaining strengths such as spread spectrum compression and nonlinear filtering.

In addition, by utilizing the digital watermarking technology according to the present invention, effective reproduction prevention, copying and distribution tracking of illegally copied or illegally distributed digital audio contents can be effectively performed, thereby providing more complete copyright protection of digital audio contents.

Claims (16)

A method of embedding a watermark signal into an audio signal, (a) converting the audio signal into a time-invariant region which is not affected by signal distortion in the time axis; (b) inserting a watermark signal into the audio signal converted into the time invariant region; And and (c) converting the audio signal into which the watermark signal is inserted is inversely inversely converted into an audio signal in a time domain. The method of claim 1, Step (a) is (a1) converting the audio signal from a linear scale to a logarithmic scale; And (a2) Fourier transforming the signal converted to the logarithmic scale to obtain an amplitude spectrum, Step (c) is (c1) inversely transforming an audio signal into which the watermark signal is inserted; And (c2) converting the Fourier inverse signal from a logarithmic scale to a linear scale. The method of claim 2, The step (a) further comprises the step of interpolating the audio signal, The step (c) further comprises the step of extrapolating the audio signal embedded in the watermark. The method according to any one of claims 1 to 3, And the watermark signal to be inserted is a spread watermark signal. A method of extracting a watermark from an audio signal having a watermark embedded therein, converting the watermark-embedded audio signal into a time-invariant region which is not affected by signal distortion on a time axis; (b) XORing an estimated spread watermark component and a pseudo random sequence obtained from the signal converted into the time-invariant region; and (c) reconstructing the bit stream as a level discrimination applying a threshold to the resultant signal in step (b). The method of claim 5, Step (a) is converting the watermark-embedded audio signal from a linear scale to a logarithmic scale; And and (a2) obtaining an amplitude spectrum by Fourier transforming the signal converted to the logarithmic scale. The method of claim 6, The step (a) further comprises the step of interpolating the audio signal embedded with the watermark. The method according to any one of claims 5 to 7, The watermark extraction method is performed after the step (c). (d) restoring the time axis to restore the extended bit string of the watermark signal to the original watermark signal. The method according to any one of claims 5 to 7, And the embedded watermark signal is a spread watermark signal. A time invariant converter for converting an audio signal into a time invariant region which is not affected by signal distortion in the time axis; A spreading watermark generator for generating copyright or control information as a spreading watermark signal; And a time invariant inverse converter for converting an audio signal having a spread watermark signal inserted in the time invariant into an invariant time inverse to an audio signal in the time domain. The method of claim 10, The time invariant converter A time base scaling unit for converting the audio signal from a linear scale to a logarithmic scale; And A time-invariant converter configured to obtain an amplitude spectrum by Fourier transforming the signal converted to the logarithmic scale, The time invariant inverse transform unit A time invariant inverse transform unit which performs Fourier inverse transform on the audio signal into which the watermark signal is inserted; And And a time axis scaling unit for converting the Fourier inverse transformed signal from a logarithmic scale to a linear scale. The method of claim 11, The time invariant converter further includes an interpolation filter for interpolating the audio signal, And the time invariant inverse transform unit further comprises an extrapolation filter for extrapolating the audio signal into which the watermark is embedded. A time invariant converter for converting an audio signal having a watermark embedded into a time invariant region which is not affected by signal distortion in a time axis; A pseudo random sequence generator for generating a pseudo random sequence; And a watermark extractor for performing an XOR operation on the estimated spread watermark component obtained from the signal converted into the time-invariant region and the pseudo-random sequence, and reconstructing the bit stream as a level discrimination applying a threshold to the resulting signal. Watermark extraction apparatus, characterized in that. The method of claim 13, The time invariant converter A time axis scaling unit for converting the watermark-embedded audio signal from a linear scale to a logarithmic scale; And And a time-invariant converter configured to obtain an amplitude spectrum by Fourier transforming the signal converted to the logarithmic scale. The method of claim 14, The time invariant converter further comprises an interpolation filter for interpolating the audio signal into which the watermark is inserted. The method according to any one of claims 13 to 15, The watermark extractor And a time axis shrinker for restoring the time axis to restore the extended bit string of the watermark signal to the original watermark signal.