KR20110014871A - Apparatus and method for embedding audio watermark, and apparatus and method for detecting audio watermark - Google Patents

Abstract

PURPOSE: An apparatus and a method for embedding audio watermark, and an apparatus and a method for detecting audio watermark, capable of easily detecting watermark inserted in a digital audio signal are provided to maintain the voice quality of the audio signal after inserting the watermark in the digital audio signal. CONSTITUTION: An audio watermark inserting apparatus(100) includes a band selector(120), a peak coefficient selector(130), a watermark inserting unit(140) and a reconfiguration unit(150). The band selector selects a specific frequency band by applying DWT(Discrete Wavelet Transform). The peak coefficient selector divides the wavelet coefficient values of the selected specific frequency band and a peak coefficient. The watermark inserting inserts the watermark in the peak coefficient. The reconfiguration unit reconfigures the audio signal with the watermark by applying a IDWT(Inverse Discrete Wavelet Transform).

Description

Apparatus and method for embedding audio watermark, and apparatus and method for detecting audio watermark}

The present invention relates to an audio watermark embedding apparatus and method, and an audio watermark detecting apparatus and method, and more particularly, to an audio watermark embedding apparatus and method for embedding a watermark at a main frequency peak, and an audio watermark. A detection apparatus and method are provided.

Robust watermarking technology is one of the techniques for protecting the copyright of the original audio signal, and inserts copyright information, that is, a watermark into the work, and the inserted watermark does not disappear even by various modifications of the work. Used as evidence for claiming ownership. Audio watermarking must be robust to signal processing such as compression and sampling, while at the same time maintaining audio quality so that it is inaudible to human hearing.

However, conventional watermarking techniques do not show sufficient robustness in signal processing. This is because it is difficult to select the best transform domain coefficient to insert the watermark. Therefore, there is a need for a method that is robust to signal processing and that is easy to insert and detect a watermark.

Accordingly, an object of the present invention is an audio watermark embedding apparatus, which is robust to signal processing while maintaining the sound quality of an audio signal even after embedding a watermark in a digital audio signal, and which can easily detect a watermark embedded in the digital audio signal, and the same. A method, and a watermark detection apparatus and method thereof are provided.

An audio watermark embedding apparatus according to the present invention for achieving the above object comprises: a band selector for selecting a specific frequency band from a plurality of frequency bands separated by applying a DWT (Discrete Wavelet Transform) to the original audio signal; A peak coefficient selector for separating the wavelet coefficients of the selected specific frequency band into a plurality of subsets and selecting a peak coefficient that is a wavelet coefficient corresponding to the peak of each subset; A watermark embedding unit for embedding a watermark in the peak coefficients selected for each subset; And a reconstruction unit reconstructing the watermark-embedded audio signal by applying an IDWT (Inverse Discrete Wavelet Transform) to the peak coefficient having the watermark embedded therein and frequency bands not selected by the band selection unit.

And a band estimator for estimating a major frequency band including a maximum energy of the spectrum of the original audio signal, wherein the band selector is configured to identify a band including the estimated major frequency band among the plurality of frequency bands. Select the frequency band.

The band estimator calculates the spectrum using a Wigner Distribution method and a Short-time Fourier transform (STFT) method.

The band selector includes: a DWT unit for separating the plurality of frequency bands by applying DWT to the original audio signal; A selection unit for selecting the specific frequency band from the plurality of frequency bands; And an IDWT unit for generating reconstructed wavelet coefficients by applying IDWT to the wavelet coefficients of the selected specific frequency band.

The peak coefficient selector may include: an FFT unit configured to generate FFT coefficients by applying a fast fourier transform (FFT) to the reconstructed wavelet coefficients; And a selector for separating the obtained FFT coefficients into the plurality of subsets each including n coefficients, identifying peaks for each subset, and then selecting peak coefficients corresponding to the identified peaks. Include.

The reconstruction unit may include: an IFFT unit reconstructing each subset by applying an inverse fast fourier transform (IFFT) after combining the watermarked peak coefficients and the watermarked wavelet coefficients; A DWT unit generating a new wavelet coefficient by applying the DWT to each of the restored subsets; And an IDWT unit for outputting a watermark-embedded audio signal by applying the IDWT to wavelet coefficients of the frequency bands not selected in the plurality of frequency bands and the new wavelet coefficients.

An audio watermark embedding method according to the present invention for achieving the above object comprises the steps of selecting a specific frequency band from a plurality of frequency bands separated by applying a DWT (Discrete Wavelet Transform) to the original audio signal; Separating the wavelet coefficients of the selected specific frequency band into a plurality of subsets, and selecting peak coefficients that are wavelet coefficients corresponding to each peak of each subset; Inserting a watermark into the peak coefficients selected for each subset; And outputting a watermark-embedded audio signal by applying an IDWT (Inverse Discrete Wavelet Transform) to the watermark-embedded peak coefficients and frequency bands not selected by the band selector.

Estimating a major frequency band including a maximum energy of the spectrum of the original audio signal; wherein selecting the band includes the estimated major frequency band of the plurality of separated frequency bands; A band is selected as the specific frequency band.

The estimating step. The spectrum is calculated using the Wigner Distribution method and the Short-time Fourier transform (STFT) method.

The selecting of the specific frequency band may include separating the plurality of frequency bands by applying DWT to the original audio signal; Selecting the specific frequency band from the plurality of frequency bands; And generating reconstructed wavelet coefficients by applying IDWT to the wavelet coefficients of the selected specific frequency band.

The selecting of the peak coefficients may include: generating FFT coefficients by applying a fast fourier transform (FFT) to the reconstructed wavelet coefficients; And dividing the obtained FFT coefficients into the plurality of subsets each including n coefficients, identifying peaks for each subset, and then selecting peak coefficients corresponding to the identified peaks. do.

The outputting may include: reconstructing each subset by applying an inverse fast fourier transform (IFFT) after combining the watermarked peak coefficients and the watermarked wavelet coefficients; Generating a new wavelet coefficient by applying the DWT to each reconstructed subset; And applying the IDWT (Inverse Discrete Wavelet Transform) to the wavelet coefficients of the frequency bands not selected in the plurality of frequency bands and the new wavelet coefficients to output a watermarked audio signal. do.

In accordance with another aspect of the present invention, an audio watermark detection apparatus includes a first specific frequency band from a plurality of frequency bands separated by applying a discrete wavelet transform (DWT) to each of an original audio signal and a watermarked audio signal. A band selector which selects a second specific frequency band respectively; An FFT unit configured to obtain first and second FFT coefficients by applying a fast fourier transform (FFT) to the selected first and second specific frequency bands; Separating the first FFT coefficients into a plurality of first subsets, selecting a first FFT peak coefficient corresponding to the peak of each first subset, separating the second FFT coefficients into a plurality of second subsets, and A peak coefficient selector selecting a second FFT peak coefficient corresponding to the peak of each second sub set; And a watermark detector configured to detect the watermark by comparing the selected first and second FFT peak coefficients.

The watermark detection unit detects the watermark under the following conditions.

W _ij '= 1, if C _ij ' -C _ij ≥ 0, and

W _ij '= 0, if C _ij ' -C _ij <0

(i = 1, 2,…, I, and j = 1, 2,…, J)

Where W _ij 'is the watermark, C _ij is the first FFT peak coefficient selected from the i th subset of the first subset, and C _ij ' is the second FFT peak selected from the i th subset of the second subset. The coefficient, j, is the location of each of the first and second FFT pig coefficients in the i th subset.

The audio watermark detection method according to the present invention for achieving the above object comprises a first specific frequency band from a plurality of frequency bands separated by applying a DWT (Discrete Wavelet Transform) to each of the original audio signal and the watermarked audio signal; Selecting a second specific frequency band, respectively; Obtaining first and second FFT coefficients by applying a fast fourier transform (FFT) to the selected first and second specific frequency bands; Separating the first FFT coefficients into a plurality of first subsets, selecting a first FFT peak coefficient corresponding to the peak of each first subset, separating the second FFT coefficients into a plurality of second subsets, and Selecting a second FFT peak coefficient corresponding to the peak of each second sub set; and comparing the selected first and second FFT peak coefficients to detect the watermark.

Hereinafter, an embodiment of the present invention will be described in detail with reference to the accompanying drawings. However, in describing the present invention, when it is determined that a detailed description of a related known function or configuration may unnecessarily obscure the subject matter of the present invention, a detailed description thereof will be omitted.

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

Referring to FIG. 1, the audio watermark embedding apparatus 100 includes a band estimator 110, a band selector 120, a peak coefficient selector 130, a watermark inserter 140, and a reconstruction unit 150. It includes.

The band estimator 110 estimates the important frequency band A including the maximum energy of the spectrum of the original audio signal S.

The band estimator 110 calculates a spectrum of the original audio signal S by using the Wigner Distribution method and the Advanced Wigner Distribution method generated by using the Short-time Fourier transform (STFT) method.

First, a method of providing a spectrum of an audio signal is various, and examples thereof include a fast fourier transform (FFT) method, a short-time fourier transform (STFT) method, a Wigner distribution method, and a wavelet method. The FFT and STFT methods provide accurate spectra but are rougher than others, making it difficult to select when the signal is smooth. The Wigner Distribution method provides a smooth spectrum, but the cross-term appearance makes the spectrum inaccurate. Therefore, in order to obtain the best results, the present invention uses the advantages of the Wigner Distribution method and the STFT method, thereby suggesting a new Advanced Wigner Distribution method.

The Advanced Wigner Distribution method is calculated by the following process.

First, STFT is defined as follows.

The spectrogram according to [Equation 1] is defined as the absolute value of the STFT transform and is expressed by [Equation 2].

과 동일한 허수 단위이다. In Equations 1 and 2, F _SP is a STFT function, h (t) is a window, s (t) is an analog signal, and STFT (t, w) is a spectrogram of F _SP . In an embodiment of the present invention, w, which is an angular frequency, is used, and j is a square root of −1.

Is the same imaginary unit.

The Wigner distribution function may be expressed in Equation 3 in the time domain and the frequency domain.

Where s (t) and S (w) are signals in the time and frequency domain, respectively, and S ^* (w) is a complex conjugate of S (w).

Pseudo-Wigner Distribution (PWD) is described in the time domain as follows.

Advanced wigner distribution using the above-described STFT, PWD and a narrow window P (Θ) is defined as follows.

The discrete form of the spectrogram is shown in Equation 6 below.

Where s (n) is a discrete signal, h (i) is a discrete form of window h (t), and N is the number of Fourier Transform windows in signal s (n).

The calculation of the Discrete advanced Wigner method is shown in [Equation 7].

Assuming p (i) is a rectangular window, ADWD (n, k) is defined as in Equation (8).

2L + 1 is the length of the discrete window p (i). In the above equation, if p (i) is a discrete delta function, ADWD (n, k) decreases to STFT, and if p (i) has all values of 1 (for all i), ADWD ( n, k) will be the Wigner distribution spectrum. 2, p (i) is a rectangular window, and 2Q + 1 is the width of the window.

In the Standard Wigner distribution, the cross term represents the result of cross computation of two frequency components. Cross terms are real and unused values that make it difficult to examine the results of the analysis. However, by adjusting the size of the window, the cross term can be completely eliminated.

In the experiment, the FFT, STFT, Wigner distribution, and Advanced Wigner distribution were applied to a signal of a music file having a sampling frequency of 44,100 samples per second. 3A-3D show the spectral results of the four schemes. 3A to 3D, magnitude represents the magnitude of spectral energy, frequency represents frequency, and time represents time. Advanced Wigner Distribution shows a smoother and more visible spectrum than other methods.

When the spectral result of the original audio signal S is as shown in FIG. 3D, the band estimator 110 estimates a band including the largest energy as the main frequency band A. FIG. Here, the main frequency band A may include only the largest energy or may be a band to which the largest energy belongs. The main frequency band (A) containing the greatest energy can be adjusted by the engineer at design time.

The band selector 120 separates the original audio signal S into a plurality of frequency bands by applying a DWT (Discrete Wavelet Transform) to the original audio signal S to separate the original audio signal S at the level L. The most significant frequency band is selected from among a plurality of frequency bands to be separated. In particular, the band selector 120 selects a band including the main frequency band A estimated by the band estimator 110 among the plurality of frequency bands separated as a specific frequency band A '. To this end, the band selector 120 includes a DWT unit 121, a selector 122, and an IDWT unit 123 as shown in FIG. 4.

The DWT unit 121 applies DWT to the original audio signal S to separate the original audio signal S into multiple frequency bands, that is, a high frequency band and a low frequency band, as shown in FIG. 5. In the case of FIG. 5, the original audio signal S is divided into six frequency bands.

The selector 122 selects a specific frequency band A 'from the plurality of separated frequency bands. The specific frequency band A ′ refers to a band covering, that is, including, the main frequency band A estimated by the band estimator 110 among the plurality of separated frequency bands.

The IDWT unit 123 applies an inverse discrete wavelet transform (IDWT) to the wavelet coefficients of the selected specific frequency band A ', that is, the DWT coefficients, and as a result, the wavelet coefficients B reconstructed from the DWT coefficients. Create

Referring back to FIG. 1, the peak coefficient selector 130 separates wavelet coefficients B of a specific frequency band A ′ selected by the band selector 120 into a plurality of subsets C _i , The wavelet coefficient corresponding to the peak of each subset C _i is selected. Hereinafter, the wavelet coefficient corresponding to the peak (ie, the important frequency peak) of each subset C _i is referred to as a peak coefficient.

To this end, the peak coefficient selector 130 includes an FFT unit 131 and a selector 132 as shown in FIG. 6. The FFT unit 131 generates FFT coefficients by applying a Fast Fourier transform (FFT) to the wavelet coefficients B reconstructed by the IDWT unit 123. Hereinafter, the FFT coefficients are referred to as FFT coefficient set (C).

The selector 132 divides the obtained FFT coefficient set C into a plurality of subsets C _i each including n coefficients (samples), and a peak for each subset C _i (the important frequency). After the peak), the peak coefficients C _ij corresponding to each identified peak are selected.

For example, the selector 132 separates the FFT coefficient set C into 200-sample subsets, thereby obtaining I subsets. C _i (i = 1, 2, ..., I) can be assumed to be a set of coefficients in the i th subset of the total I subsets. The selector 132 identifies a peak among a plurality of coefficients included in the subset C _i , and selects a coefficient corresponding to the identified peak as the peak coefficient C _ij . Where j (j = 1, 2, ..., J) means that the selected peak coefficient among the coefficients in the i-th subset is located in the j-th.

The selector 132 may select a coefficient having a magnitude greater than or equal to a threshold value for each subset C _i as a pig coefficient. Accordingly, the selector 132 may select a plurality of peak coefficients C _ij for each subset C _i , and the threshold value is stored in a memory (not shown).

The watermark inserting unit 140 inserts a watermark into each peak coefficient C _ij selected for each subset C _i . The inserted watermark is a multibit watermark and has a value of 0 or 1. The watermark inserting unit 140 inserts a multibit watermark into each peak coefficient C _ij according to the following conditions.

If W _ij = 1, C _ij '= C _ij + λ × abs (C _ij )

otherwise, if W _ij = 0, C _ij '= C _ij -λ × abs (C _ij )

In Equation 9 and Equation 10, C _ij 'is the peak coefficient to which the watermark is inserted, C _ij is the peak coefficient before the watermark is inserted, and λ is a constant selectable by the designer, abs (C _ij ) Is the absolute value of C _ij .

In detail, the watermark inserting unit 140 inserts the multibit watermark into the wavelet coefficient of the main band in the frequency domain. The important frequency peaks are extracted in a specific frequency band (A '). For each peak, that is, a wavelet coefficient corresponding to each peak, a one-bit watermark is inserted by changing the magnitude of the peak as in Equations 9 and 10. The watermarked audio signal is obtained by reconstructing the changed coefficients and the unchanged coefficients in the reconstruction unit 150.

The reconstructor 150 applies the IDWT to the peak coefficients C _ij ′ with the watermark embedded therein and other frequency bands not selected by the band selector 120, thereby inserting the watermarked audio signal S Reconstruct with '). Referring to FIG. 7, the reconstruction unit 150 includes an IFFT unit 151, a DWT unit 152, and an IDWT unit 153.

The IFFT unit 151 combines the watermarked peak coefficients C _ij 'and the watermarked wavelet coefficients, and then applies an inverse fast fourier transform (IFFT) on the combined signal to each subset. Reconstruction The wavelet coefficients without the watermark are wavelet coefficients that are not peaks in the subset Ci, and the signal B corresponding to each of the subsets output by the reconstruction is a signal forming a specific frequency band A '. B).

The DWT unit 152 generates a new wavelet coefficient C 'by applying the DWT to the signal B output by the reconstruction.

The IDWT unit 153 performs wavelet coefficients and new wavelet coefficients of frequency bands not selected by the band selector 120, that is, the remaining frequency bands except a specific frequency band A 'among the plurality of frequency bands. IDWT is applied to ') to output the audio signal S' with the watermark embedded therein.

FIG. 8 is a flowchart illustrating an audio watermark embedding method of FIG. 1.

Referring to FIG. 8, the band estimator 110 estimates the main frequency band A from the spectrum of the original audio signal S (S810). The band estimator 110 calculates a spectrum of the original audio signal S by using the above-described Advanced Wigner Distribution method.

The band selector 120 separates the original audio signal S into a plurality of frequency bands by applying DWT to the original audio signal S (S820), and selects a major frequency band A of the plurality of frequency bands. A specific frequency band A 'is included (S830).

In operation S840, the band selector 120 generates reconstructed wavelet coefficients B by applying IDWT to the wavelet coefficients of the selected specific frequency band A '.

The peak coefficient selector 130 generates an FFT coefficient set C by applying an FFT to wavelet coefficients B generated by IDWT (S850).

The peak coefficient selector 130 separates the obtained FFT coefficient set C into a plurality of subsets C _i , and the peak coefficients corresponding to the important frequency peaks for each subset C _i ( C _ij ) is selected (S860).

The watermark inserting unit 140 inserts a watermark into each peak coefficient C _ij selected for each subset C _i using [Equation 9] or [Equation 10] (S870).

The reconstruction unit 150 combines the peak coefficients C _ij 'with the watermark embedded therein and wavelet coefficients without the watermark embedded therein, and then reconstructs each subset by applying an IFFT to the combined signal. (S880).

The reconstruction unit 150 generates and outputs a watermark-embedded audio signal S 'using the reconstructed signal B, that is, the plurality of reconstructed subsets and the remaining frequency bands not selected in step S830. (S890). This is because the plurality of subsets are signals that make up a particular frequency band A '.

9 is a block diagram showing an audio watermark detection apparatus according to an embodiment of the present invention.

Extracting the watermark from the watermark-embedded audio signal S 'corresponds to reverse processing of the embedded watermark algorithm. In the present invention, the original audio signal is used to detect the watermark. After the attack, after an attack such as illegal copying, the audio signal with the watermark embedded there is S ', the original audio signal is S, and the original multibit watermark is W. In the algorithm proposed by the present invention, a sequence of bits (W ') is extracted at S' and continuously compared with the one watermark W. The bit is detected when W and W 'are the same.

To this end, the audio watermark detection apparatus 900 includes a band selector 910, an FFT unit 920, a peak coefficient selector 930, and a watermark detector 940.

In order to separate the original audio signal S at the level L, the band selector 910 divides the original audio signal S into a plurality of frequency bands by applying DWT to the original audio signal S, and then separates the first audio signal S into a plurality of frequency bands. Select a specific frequency band (B). In addition, the band selector 120 separates the watermarked audio signal S 'at a level L, divides the watermarked audio signal S' into a plurality of frequency bands by applying DWT, and then separates the watermarked audio signal S '. A second specific frequency band B 'is selected from the plurality of frequency bands.

The algorithm for selecting the first and second specific frequency bands B and B 'may include the first and second specific frequency bands B and B, each of which includes the most significant frequency band, as described with reference to FIG. ') Is selected and Advanced Wigner Distribution is used for this. Detailed description thereof will be omitted.

The FFT unit 920 applies FFTs to the selected first and second specific frequency bands B and B 'to obtain first and second FFT coefficients C and C', respectively.

The peak coefficient selector 930 separates the first FFT coefficients C obtained from the first specific frequency band B into a plurality of first subsets, and first FFT peak coefficients corresponding to the peaks of each first subset. Selects (C _ij ), separates the second FFT coefficient C ′ obtained from the second specific frequency band B ′ into a plurality of second subsets, and a second FFT corresponding to the peak of each second subset. Select the peak coefficient (C _ij '). J peaks are generated for this subset. The j-th peak is defined by C _ij and C _ij '(i = 1, 2,…, I, j = 1, 2,…, J).

Thus, the first and second FFT peak coefficients are selected by the number of first and second subsets, respectively. Alternatively, when m peaks are selected for each subset in the watermark embedding apparatus 100 of FIG. 1, the peak coefficient selector 930 selects m first and second FFT peak coefficients, respectively.

For example, the peak coefficient selector 930 separates C and C 'into 200-sample subsets to obtain I subsets, respectively. This is because C and C 'have the same length and the same subset of numbers is returned). i = 1, 2,... , I and C _i and C _i 'are assumed, and I is in each case a set of coefficients in the i th subset.

The watermark detector 940 detects the watermark W 'by comparing the selected first and second FFT coefficients. The watermark detection unit 940 detects a watermark (or watermark bit) under the following conditions.

W _ij '= 1, if C _ij ' -C _ij ≥ 0,

(i = 1, 2,…, I, and j = 1, 2,…, J)

W _ij '= 0, if C _ij ' -C _ij <0

(i = 1, 2,…, I, and j = 1, 2,…, J)

Where W _ij 'is the watermark, C _ij is the first FFT peak coefficient selected from the i th subset of each first subset, and C _ij ' is the second FFT peak coefficient selected from the i th subset of each second subset , j is the position of each of the first and second FFT pig coefficients in the i th subset.

10 is a flowchart for explaining an audio watermark detection method of FIG. 9.

Referring to FIG. 10, the band selector 910 divides a plurality of frequency bands by applying DWT to an original audio signal S and an audio signal S ′ into which a watermark is inserted, and then separates a plurality of frequency bands. The first and second specific frequency bands B and B 'are selected from the bands (S1010).

The FFT unit 920 applies the FFT to the selected first and second specific frequency bands B and B 'to obtain first and second FFT coefficients C and C', respectively (S1020).

The peak coefficient selector 930 separates the first FFT coefficient C into a plurality of first subsets, selects a first FFT peak coefficient corresponding to the peak of each first subset, and selects a second FFT coefficient C '. Are separated into a plurality of second sub sets, and second FFT peak coefficients corresponding to the peaks of the second sub sets are selected (S1030 and S1040).

The watermark detection unit 940 detects the watermark W 'by comparing the selected first FFT coefficient and the second FFT peak coefficient based on [Equation 12] and [Equation 13] (S1050).

The above process is repeated until all hidden watermarks are detected. The detected watermark W 'is used to compare with the original watermark W in order to estimate the robustness of the watermark.

Table 1 shows the relationship between the constant λ and the probability of detected watermark bits.

λ Total bits embedded Number of bits detecged 0.05 30,138 30,138 0.1 30,138 30,136 0.2 30,138 30,130 0.3 30,138 30,074 0.4 30,138 30,058 0.5 30,138 30,050 0.6 30,138 30,048 0.7 30,138 30,042 0.8 30,138 30,040

Referring to [Table 1], it can be seen that the smaller the λ, the higher the detection probability and the improved the imperceptibility. [lambda] = 0.1, the detection of the multi-bit watermark is more accurately detected, thereby satisfying [Equation 11]. λ may vary within 0.05-0.1 depending on the algorithm and application applied as well as the result of the performance.

The multi-bit watermark may be inserted into the important frequency peaks of the original audio signal by using the above-described algorithm to obtain an audio signal having the watermark embedded therein. Although there is a small difference between each spectrum of the original audio signal and the watermarked audio signal, the sounds of the two signals are almost the same as shown in Figs. 11A and 11B. In other words, the watermarking algorithm of the present application is quite robust. 11A and 11B show the spectrum of the original audio signal and the watermarked audio signal using Advanced Wigner Distribution.

When an audio file is broadcast, the audio file may be subject to various dispersions or powerful attacks (eg, piracy). The main purpose of watermarking technology is the robustness to attack, and the concept of robustness is clear. In other words, the watermark is robust unless the watermark is impaired without rendering unused attacked data. Watermark impairment can be measured by criteria such as miss probability, probability of bit error or channel capacity. In multimedia, the usefulness of an attacked data can be assessed taking into account its perceptual quality or distortion. Therefore, the robustness can be evaluated simultaneously considering the damage of the watermark and the distortion of the attacked data. If the watermark is compromised within acceptable limits while maintaining the perceptual quality of the attacked data, the attack is successful. Thus, a good approach for watermarking must deal with such a strong attack in order to extract the high embedded watermark message.

The robustness of a watermark message is evaluated by attacking the watermark message in various ways (MP3 compression, resampling, low pass filtering, noise addition, etc.). The number of bits detected after the attack is directly proportional to the robustness of the watermark.

12 to 15 are diagrams for describing a test embodiment according to various attacks.

12 shows a model for testing an MP3 compression attack. Referring to Fig. 12, in this model, the watermarked audio signal S 'in the form of a * .wav file is compressed by MP3 compression (128 kbps) to generate an MP3 file. The generated MP3 file is decoded into a wave file. Thus, if the file with the watermark embedded is converted into a wave file, it will face an MP3 compression attack.

13 shows a model for testing a noise add attack. Referring to FIG. 13, in the case of a noise addition attack, white Gaussian noise is added to the watermark-embedded audio signal S 'at a low noise level so that the original audio signal and the changed signal cannot be distinguished. Therefore, in the present invention, Gaussian noise addition with SNR = 30 is selected.

14 shows a model for testing a filtering attack. Referring to FIG. 14, a high or low frequency band is separated using a low pass filter (LPF) or a high pass filter (HPF). Due to the sound quality, the most significant frequency band cannot be cut off, but the sound is changed. In the present invention, a Butterworth LPF is used for the watermarked audio file S ', where the cutoff frequency Wn is 0.9.

15 shows a model for testing a resampling attack. Referring to FIG. 15, up or down sampling may change a sampling frequency of an original audio signal. In the test of the present invention, the watermarked audio file S 'is resampled from 44100 to 22100 Hz.

The wave file obtained from the attacks described with reference to FIGS. 12 to 15 is applied to the detection algorithm proposed in the present invention in order to detect a multibit watermark message. For each attack, each detected bit is compared with the one watermark message. If the detected bit is the same value (0 or 1) as the one watermark bit, it means that the bit is detected.

Table 2 below shows the results of the tests.

Attacks Total bits embedded Number of bits detected Percentage of bits detected MP3 30,138 29,466 97.77 Noise addition 30,138 30,136 99.99 Low Pass Filtering 30,138 30,136 99.99 Re-sampling 30,138 30,134 99.98

Referring to Table 2, since the number of bits detected from each attack is large, the results show that the watermarks have good robustness. In other words, it can be seen that the watermarking scheme proposed herein is robust to strong attacks such as noise addition, filtering, resampling, and MP3 compression attacks.

On the other hand, with the development of technology, music performance is directly propagated through broadcasting stations, the Internet, and the like. The quality of performance and protection of intellectual property must be established. For this audio watermarking technique used for this, efficient computational complexity for real-time operation is required. Thus, the present application estimates the efficiency of watermarking techniques for real-time operation by considering computational complexity relative to other techniques.

The present application contemplates the Advanced Wigner distribution process and the embedded watermark process. For the Advanced Wigner process, a comparison between the Advanced Wigner distribution and the original Wigner distribution is provided in terms of the number of arithmetic operations required. Discrete forms of the Advanced Wigner distribution are shown in Equations 7 and 8. In order to calculate the Advanced Wigner method, it is necessary to calculate the STFT at time n. This can be done by a resursive formula, from the value of the previous STFT obtained at time n-1.

In the present example, it is assumed that the window h (i) defined in Equation 1 is a raised consine window and includes N samples. The regression equation for the STFT coefficient is defined as in Equation 14 below.

From Equation 8 and Equation 14 (Papoulis, 1977), complex multiplication and complex addition of the Advanced Wigner method are 'N (3 + Q) / 2' and 'N, respectively. (6 + Q) / 2 '. Here, Q is defined by [Equation 8]. However, the complex multiplication and complex addition of the Original Wigner Distribution are 'N (4_log ₂ N) / 2' and 'N (log ₂ 2N)' (Mecklenbrauker and Hlawatsch, 1997). Q is less than N, so the Advanced Wigner approach provides more efficient complexity. In the experiment of the present invention, N = 256 and Q = 4.

In order to consider the computational complexity of the proposed watermarking insertion process, we compare the proposed watermarking technique with standard watermark-embedding practices. In a typical watermark-embedding technique, a watermark message is inserted into each FFT coefficient of the original signal. However, in our algorithm, the watermark message is inserted at the importane peaks. To find these peaks, we use at least three samples (ie coefficients) that are adjacent to each other. Thus, less than half of the sample is used to insert a watermark message.

In addition, since the most imporant frequency band is used after using the wavelet of level L, if the original signal includes P samples, the most imporant frequency band includes P / 2 ^L samples (Strand and Nguyen, 1997). . In the present invention, instead of using the P watermark message embedded in the original signal, only the P / 2 ^{L + 1} watermark message is used. Thus, the Computational Time can be shortened.

In the above description, the watermark, the multibit watermark, the watermark bit, the watermark message, and the like are the same, and the expressions of the Adavnced Wigner Distribution, the Adavnced Wigner method, etc. are also the same, and various descriptions are used for convenience of description.

On the other hand, as a result, the present invention provides an audio watermarking technique that uses a feature that does not change in the frequency domain of the original audio signal. The constant features used are the peak coefficients, ie the important frequency peaks. The Advanced Wigner Distribution is also used to estimate the most significant frequency band that characterizes the sound. In the proposed watermarking technique, a multi-bit watermark is embedded in the important frequency peaks by fine-tuning the magnitude of the important frequency peaks. Test results by the algorithm proposed by the present invention show that the watermarking technique is robust against strong attacks such as noise addition, filtering, resampling, and MP3 compression attacks. In addition, the efficient computational complexity of the present application is very effective for real-time applications.

While the above has been shown and described with respect to preferred embodiments of the present invention, the present invention is not limited to the specific embodiments described above, it is usually in the technical field to which the invention belongs without departing from the spirit of the invention claimed in the claims. Various modifications may be made by those skilled in the art, and these modifications should not be individually understood from the technical spirit or the prospect of the present invention.

1 is a block diagram showing an audio watermark embedding apparatus according to an embodiment of the present invention;

2 shows an example of a rectangular window p (i),

3a to 3d show spectral results of the four schemes,

4 is a view illustrating in detail the band selector of FIG. 1;

5 is a view for explaining a process of separating into multiple frequency bands by the DWT unit of FIG.

6 is a view illustrating in detail the peak coefficient selector of FIG.

7 is a detailed view of the restoration unit of FIG. 1;

8 is a flowchart for explaining an audio watermark embedding method according to FIG. 1;

9 is a block diagram showing an audio watermark detection apparatus according to an embodiment of the present invention;

10 is a flowchart for explaining an audio watermark detection method according to FIG. 9;

11A and 11B show spectra of an original audio signal and a watermarked audio signal using Advanced Wigner Distribution, and

12 to 15 are diagrams for describing a test embodiment according to various attacks.

Explanation of symbols on the main parts of the drawings

100: audio watermark insertion device 110: estimator

120: band selector 130: peak coefficient selector

140: watermark insertion unit 150: reconstruction unit

Claims (16)

A band selector configured to select a specific frequency band from a plurality of frequency bands separated by applying a discrete wavelet transform (DWT) to the original audio signal; A peak coefficient selector for separating the wavelet coefficients of the selected specific frequency band into a plurality of subsets and selecting a peak coefficient that is a wavelet coefficient corresponding to the peak of each subset; A watermark embedding unit for embedding a watermark in the peak coefficients selected for each subset; And And a reconstruction unit for restoring the watermark-embedded audio signal by applying IDWT (Inverse Discrete Wavelet Transform) to the peak coefficient having the watermark embedded therein and frequency bands not selected by the band selection unit. Audio watermark inserting device. The method of claim 1, And a band estimator for estimating a main frequency band including the maximum energy of the spectrum of the original audio signal. And the band selector selects a band including the estimated main frequency band among the plurality of frequency bands as the specific frequency band. 3. The method of claim 2, And the band estimating unit calculates the spectrum using a Wigner distribution method and a short-time fourier transform (STFT) method. The method of claim 1, The band selector, A DWT unit for applying the DWT to the original audio signal and separating the plurality of frequency bands; A selection unit for selecting the specific frequency band from the plurality of frequency bands; And And an IDWT unit for generating reconstructed wavelet coefficients by applying IDWT to the selected wavelet coefficients of the specific frequency band. The method of claim 1, The peak coefficient selector, An FFT unit generating FFT coefficients by applying a fast fourier transform (FFT) to the reconstructed wavelet coefficients; And A selector for separating the obtained FFT coefficients into a plurality of subsets each including n coefficients, identifying peaks for each subset, and then selecting peak coefficients corresponding to the identified peaks; Audio watermark insertion apparatus, characterized in that. The method of claim 5, The reconstruction unit, An IFFT unit for reconstructing each subset by applying an inverse fast fourier transform (IFFT) after combining the watermarked peak coefficients and the watermarked wavelet coefficients; A DWT unit generating a new wavelet coefficient by applying the DWT to each of the restored subsets; And And an IDWT unit for outputting a watermark-embedded audio signal by applying the IDWT to wavelet coefficients of the frequency bands not selected in the plurality of frequency bands and the new wavelet coefficients. Insertion device. Selecting a specific frequency band from a plurality of frequency bands separated by applying a discrete wavelet transform (DWT) to the original audio signal; Separating the wavelet coefficients of the selected specific frequency band into a plurality of subsets, and selecting a peak coefficient that is a wavelet coefficient corresponding to each peak of each subset; Inserting a watermark into the peak coefficients selected for each subset; And And outputting a watermark-embedded audio signal by applying an IDWT (Inverse Discrete Wavelet Transform) to the peak coefficient to which the watermark is inserted and frequency bands not selected by the band selector. How to embed an audio watermark. The method of claim 7, wherein Estimating a major frequency band including the maximum energy of the spectrum of the original audio signal; The selecting of the band comprises: selecting a band including the estimated main frequency band among the plurality of separated frequency bands as the specific frequency band. The method of claim 8, The estimating step. An audio watermark embedding method comprising calculating a spectrum using a Wigner distribution method and a short-time fourier transform (STFT) method. The method of claim 7, wherein Selecting the specific frequency band, Applying DWT to the original audio signal and separating the plurality of frequency bands; Selecting the specific frequency band from the plurality of frequency bands; And And generating reconstructed wavelet coefficients by applying IDWT to the selected wavelet coefficients of the specific frequency band. The method of claim 7, wherein Selecting the peak coefficients, Generating FFT coefficients by applying a fast fourier transform (FFT) to the reconstructed wavelet coefficients; And Dividing the obtained FFT coefficients into the plurality of subsets each including n coefficients, identifying peaks for each subset, and then selecting peak coefficients corresponding to the identified peaks; Audio watermark insertion method, characterized in that. The method of claim 11, The outputting step, Restoring each subset by combining an inverse fast fourier transform (IFFT) after combining the watermarked peak coefficients with the watermarked wavelet coefficients; Generating a new wavelet coefficient by applying the DWT to each reconstructed subset; And Outputting a watermark-embedded audio signal by applying the IDWT (Inverse Discrete Wavelet Transform) to the wavelet coefficients of the frequency bands not selected in the plurality of frequency bands and the new wavelet coefficients; Audio watermark insertion method, characterized in that. A band selector configured to select a first specific frequency band and a second specific frequency band from a plurality of frequency bands separated by applying a discrete wavelet transform (DWT) to each of the original audio signal and the watermarked audio signal; An FFT unit configured to obtain first and second FFT coefficients by applying a fast fourier transform (FFT) to the selected first and second specific frequency bands; Separating the first FFT coefficients into a plurality of first subsets, selecting a first FFT peak coefficient corresponding to the peak of each first subset, separating the second FFT coefficients into a plurality of second subsets, and A peak coefficient selector selecting a second FFT peak coefficient corresponding to the peak of each second sub set; And And a watermark detector for comparing the selected first and second FFT peak coefficients to detect the watermark. The method of claim 13, The watermark detection unit detects the watermark under the following conditions: W _ij '= 1, if C _ij ' -C _ij ≥ 0, and W _ij '= 0, if C _ij ' -C _ij <0 (i = 1, 2,…, I, and j = 1, 2,…, J) Where W _ij 'is the watermark, C _ij is the first FFT peak coefficient selected from the i th subset of the first subset, and C _ij ' is the second FFT peak selected from the i th subset of the second subset. Coefficient, j is the position of each first and second FFT pig coefficients in the i th subset. Selecting a first specific frequency band and a second specific frequency band from a plurality of frequency bands separated by applying a discrete wavelet transform (DWT) to each of the original audio signal and the watermarked audio signal; Obtaining first and second FFT coefficients by applying a fast fourier transform (FFT) to the selected first and second specific frequency bands; Separating the first FFT coefficients into a plurality of first subsets, selecting a first FFT peak coefficient corresponding to each peak of the first subset, separating the second FFT coefficients into a plurality of second subsets, Selecting a second FFT peak coefficient corresponding to the peak of each second sub set; and And detecting the watermark by comparing the selected first and second FFT peak coefficients. 15. The method of claim 14, The detecting of the watermark may include detecting the watermark under the following conditions: W _ij '= 1, if C _ij ' -C _ij ≥ 0, and W _ij '= 0, if C _ij ' -C _ij <0 (i = 1, 2,…, I, and j = 1, 2,…, J) Where W _ij 'is the watermark, C _ij is the first FFT peak coefficient selected from the i th subset of the first subset, and C _ij ' is the second FFT peak selected from the i th subset of the second subset. Coefficient, j is the position of each first and second FFT pig coefficients in the i th subset.

Images