Digital watermarking method and apparatus for audio data

[00002] 1. Technical Field

[00003] The present invention relates to digital watermarking of data, including audio, video, and multimedia data. Specifically, the invention relates to embedding a watermark signal into digital audio data.

[00004] 2. Description of the Related Art

[00005] The proliferation of digitized media such as image, video and multimedia is creating a need for security system which facilitates the identification of the source of the material. Particularly, the internet is increasingly used for transmitting recorded music in a digitized format. Content providers, i.e., owners of such recorded music in digital form, have a need to embed into multimedia data a predetermined mark which can subsequently be detected by software and/or hardware devices for purposes of authenticating copyright ownership, control and management of the multimedia data. Digital watermarking has been developed as a technique for embedding an identifiable data into multimedia data.

[00006] Conventionally, a watermark signal used for watermarking audio signal has been relatively simple signals such as a sequence of code symbols because, unlike image or video, inserting a large watermark signal would affect original audio perceptibility. Therefore, a watermarking technique employing a large image as a watermark signal has been proposed. However, prior arts watermarking techniques involving an image watermark are susceptible to unauthorized removal of watermarks, thereby making hard to trace the origin of a copyright protected material.

[00007] An objective of the present invention is to provide a digital watermarking technique that does not allow easy removal by an unauthorized person of a watermark signal embedded in digital data, particularly audio signal data and yet minimize distortion of original data. The objective is achieved in part by correlating the coefficients of wavelet transformation of magnitudes of Fourier transformed audio signal with the coefficients of discrete cosine transformed watermark signal. The coefficients of transformed audio signal data and scaled-down coefficients of watermark signal are added, inverse wavelet transformed and inverse Fourier transformed to produce watermarked audio signal data.

[00008] In accordance with one aspect of the present invention, a method for inserting a watermark signal into audio signal data comprises the steps of: Fourier transforming audio signal data in the frequency domain in a form of first components and second components; wavelet transforming absolute values of the first components to generate first spectral coefficients; discrete cosine transforming a watermark signal to generate second spectral coefficients; combining the first spectral coefficients and the second spectral coefficients; and Inverse wavelet transforming the combined coefficients.

[00009] The first components and second components may be the magnitudes and phases of coefficients respectively. Preferably, the step of combining includes a step of performing a weighted addition of the first and second spectral coefficients. It is preferable for the method to further comprise a step of inverse Fourier transforming the output of the inverse wavelet transforming by using the phases of coefficients. Also, it is preferable for the method to further comprise a step of multiplying information from the first spectral coefficients to the second spectral coefficients prior to the combining step. Further, the method may comprise a step of multiplying a scaling factor to the second spectral coefficients prior to said combining step. The scaling factor may be in the range of 0.01-0.05. Preferably, the information is a function of the sign of the first spectral coefficients.

[00010] In accordance with another aspect of the present invention, a method for extracting a watermark from a watermark-embedded audio data comprises the steps of Fourier transforming a watermark-embedded audio data and an original audio data to generate the first components and the second components respectively; Wavelet transforming the absolute magnitudes of the first components of the watermark-embedded audio data and the original audio data, respectively; taking the differences between wavelet-transform coefficients of the watermark-embedded audio data and the original audio data; and inverse-discrete cosine transforming the differences.

[00011] Preferably, the method further comprise a step of multiplying the sign of the wavelet-transform coefficients associated with the original audio data to wavelet-transform coefficients associated with the watermark-embedded audio data. Further, the multiplying step may comprise a step of multiplying a scaling factor to wavelet coefficients associated with the watermark-embedded audio data. The sign may be obtained by using a signum function. The scaling factor may be in the range of 20-100.

[00015] Referring to FIG. 1, a digital watermarking method and system according to the present invention will be described.

[00016] When a watermark-signal is transformed using a transformation scheme, the shape of the original watermark is not preserved. The present invention is based on the idea that a watermark of an impulse type is hard to delete because the watermark, after inventive transformations, would be distributed over the whole transform plane. Thus it helps to prevent unauthorized copying of a legitimate data.

[00017] Among many transformation schemes, the present invention employs DCT to transform a watermark, because coefficients of DCT transformed plane are real values, whereas coefficients of Fourier-transformed plane have complex components, making it more difficult to match with original image data.

[00018] When inserting a watermark (W) into original audio data (S) to form a watermark-embedded audio data (S), the quality of the watermark embedded audio data (S′) can be controlled by adjusting the interval between the original audio data (S) and the watermark (W) using a scaling parameter α, as shown in Eq. 1.

[heading-00019] [Equation 1]
S′_i=S_i+aW_i  Eq. 1a
S′_i=S_i(1+aW_i)  Eq. 1b
Si′_i=S_i(e^aW,)   Eq. 1c

[00023] Eq. 1a is always invertible. Eqs. 1b and 1c are invertible when Wi±0. If Eqs. 1b and 1c are employed, the security of watermarks may not be maintained for various processes in multimedia applications. Thus, the present invention utilizes Eq. 1a.

[00024] FIGS. 1 and 2 show processes of watermarking original digital data and extracting the watermarks, in accordance with the present invention Referring to FIG. 1, a process of watermarking original digital data will be described.

[00025] When original audio data to embed a watermark is inputted to processing means (not shown in the figure), the processing means Fourier-transforms the original audio data by using a predetermined algorithm to generate amplitude and phase components. A Fourier Series is used for the Fourier transform, as follows: $\begin{array}{cc}\begin{array}{c}{X}_n=\frac{1}{T_o}{\int }_{T_o}^{}x\left(t\right)e^{-\mathrm{j2\pi }n{\int }_{n}t}dt\\ x\left(t\right)=\sum _{n=-\infty }^{\infty }{X}_n{e}^{\mathrm{j2\pi }n{\int }_{n}t}\end{array}& \left[\mathrm{Equation}2\right]\end{array}$

[00026] The process of Fourier-transforming a continuous function f(x) using the infinite series of Eq. 2 may be defined as Eq. 3. $\begin{array}{cc}\begin{array}{c}X\left(f\right)={\int }_{-\infty }^{\infty }x\left(t\right)e^{-\mathrm{j2\pi }\int l}dt\\ x\left(t\right)={\int }_{-\infty }^{\infty }X\left(f\right)e^{-\mathrm{j2\pi }\int l}df\end{array}& \left[\mathrm{Equation}3\right]\end{array}$

[00027] For Example $\begin{array}{c}X\left(t\right)=\sum _{n=-\infty }^{\infty }\delta \left(t-{\mathrm{nT}}_o\right)\\ X_n=\frac{1}{T_o}{\int }_{4.8T_0}^{6.8T_0}\delta \left(t-5T_o\right)e^{-\mathrm{j2\pi }n{\int }_{0}t}dt\\ =\frac{1}{T_o}\leftarrow f_o{T}_o=1\\ x\left(t\right)=\sum _{n=-\infty }^{\infty }{X}_n{e}^{-\mathrm{j2\pi }n{\int }_{o}t}\\ =\frac{1}{T_o}\sum _{n=-\infty }^{\infty }{e}^{-\mathrm{j2\pi }n{\int }_{o}t}\\ X\left(f\right)={\int }_{-\infty }^{\infty }\sum _{n=-\infty }^{\infty }\delta \left(t-{\mathrm{nT}}_o\right)e^{-\mathrm{j2\pi }\mathrm{fl}}dt\\ =\sum _{n=-\infty }^{\infty }{\int }_{-\infty }^{\infty }\delta \left(t-{\mathrm{nT}}_o\right)e^{-\mathrm{j2\pi }\mathrm{fl}}dt\\ =\sum _{n=-\infty }^{\infty }{e}^{-\mathrm{j2\pi }{\mathrm{nfT}}_o}\\ ={\int }_{-\infty }^{\infty }\frac{1}{T_o}\sum _{n=-\infty }^{\infty }{e}^{-\mathrm{j2\pi }n{\int }_{o}t}{e}^{-\mathrm{j2\pi }\mathrm{fl}}dt\\ =\frac{1}{T_o}\sum _{n=-\infty }^{\infty }{\int }_{-\infty }^{\infty }{e}^{-\mathrm{j2\pi }\left(f+{\mathrm{nf}}_{n}t\right)}dt\\ =\frac{1}{T_o}\sum _{n=-\infty }^{\infty }\delta \left(f+{\mathrm{nf}}_o\right)\end{array}$
where $\sum _{n=-\infty }^{\infty }\delta \left(t-{\mathrm{nT}}_o\right)\leftarrow \to f_o\sum _{n=-\infty }^{\infty }\delta \left(f+{\mathrm{nf}}_o\right)$ $\sum _{n=-\infty }^{\infty }\delta \left(t-{\mathrm{nT}}_o\right)=\frac{1}{T_o}\sum _{n=-\infty }^{\infty }{e}^{-\mathrm{j2\pi }{\mathrm{nf}}_{o}t}$ $\sum _{n=-\infty }^{\infty }\delta \left(f-{\mathrm{nf}}_o\right)=\frac{1}{f_0}\sum _{n=-\infty }^{\infty }{e}^{-\mathrm{j2\pi }{\mathrm{nfT}}_p}$ $\sum _{n=-\infty }^{\infty }\delta \left(f-n\right)=\sum _{n=-\infty }^{\infty }{e}^{-\mathrm{j2\pi }\mathrm{nf}}\leftarrow T_0=f_o=1$

[00029] In the Fourier transform as defines in Eq. 3, it is prefereable to use complex values, since a complex value may represent both the amplitude at a time, as shown below.

[heading-00030] [Equation 4]
F(u)=R(u)+jl(u)
F(u)=|F(u)|e^jφ(u)

[00033] In Eq. 4, Fourier spectrum is expressed as:
|F(u)|=[R²(u)+I²(u)]^1/2
the phase is expressed as: $\varphi \left(u\right)={\tan }^{-1}\left[\frac{I\left(u\right)}{R\left(u\right)}\right]$
the power spectrum is expressed as:
P(u)=|F(u)²=R²(u)+2(u)
where u represents a variable for frequency.

[00039] By employing Euler's equation, i.e., exp[−j2πux]=cos2πx, the Fourier transform can be represented by the equation defined in Eq, 5. $\begin{array}{cc}T\left\{f\left(x,y\right)\right\}=F\left(u,v\right)={\int }_{-\infty }^{\infty }{\int }_{-\infty }^{\infty }f\left(x,y\right)\exp \left[-\mathrm{j2}\pi \left(\mathrm{ux}+\mathrm{vy}\right)\right]dxdy{T}^{-1}\left\{f\left(u,v\right)\right\}=F\left(x,y\right)={\int }_{-\infty }^{\infty }{\int }_{-\infty }^{\infty }f\left(u,v\right)\exp \left[-\mathrm{j2}\pi \left(\mathrm{ux}+\mathrm{vy}\right)\right]dudv& \left[\mathrm{Equation}5\right]\end{array}$

[00040] Therefore, the Fourier spectrum, phase, and power spectrum can be given as follows:

[heading-00041] Fourier spectrum:
|P(u,v)=|=[R²(u,v)+I²(u,v)]^1/2
Phase: $\Phi \left(u,v\right)={\tan }^{-1}\left[\frac{I\left(u,v\right)}{R\left(u,v\right)}\right]$
Power spectrum:
P(u,v)=|F(u,v)²=R²(u,v)+I²(u,v)

[00046] As shown above, Fourier transform employs infinite series to transform analog signals to sampled digital signals. However, in order to implement Fourier transform by a computer, modified Fourier transform for sampled data, i.e., Discrete Fourier Transform (DFT) is used on behalf of Fourier transform. If DFT is employed, f(x) can be given as Eq. 6. $\begin{array}{cc}\begin{array}{c}X\left[m\right]=\sum _{n=0}^{N-1}x\left[n\right]e^{-j\frac{2\pi \mathrm{mn}}{N}},m\left[0,N-1\right]\\ =X\left(e^{-j\frac{2\pi m}{N}}\right),\\ x\left[n\right]=\sum _{m=0}^{N-1}X\left[m\right]e^{j\frac{2\pi \mathrm{mn}}{N}},n\left[0,N-1\right]\end{array}& \left[\mathrm{Equation}6\right]\end{array}$

[00047] Also, inverse $X\left(e^{\frac{2\pi m}{N}}\right)$
is defined as Eq. 7, when the rotational initial and maximum value is X[n]. $\begin{array}{cc}\begin{array}{c}{x\left[n\right]}_N=\sum _{m=0}^{N-1}X\left(e^{\frac{2\pi m}{N}}\right)e^{j\frac{2\pi \mathrm{mn}}{N}}\\ =\sum _{k=-\infty }^{\infty }x\left[n-\mathrm{kN}\right]\text{:}\mathrm{Period}N\end{array}& \left[\mathrm{Equation}7\right]\end{array}$

[00049] Digital audio data is Fourier transformed at a Fourier transformer 10 as described above while a watermark signal is discrete cosine transformed at a discrete cosine transformer 14. Next, the magnitudes of the coefficients of Fourier transformed audio data, obtained by a magnitude extractor 11, are wavelet transformed at a wavelet transformer 13. Now the signs (+, −, 0) of the audio's coefficients are respectively multiplied to the spectral coefficients of the watermark signal at the first multiplier 31 in order to correlate the audio signal and the watermark signal to certain extent. The sign can be easily obtained by using the signum function unit 15, which outputs 1, −1 or 0 depending on the sign/polarity of an input value disregarding the magnitude. The spectral coefficients of the watermark signal are further multiplied by a scaling factor α at the second multiplier 32 so as not change the audio signal's quality as perceived by the listener. The scaling factor is preferably in the range of 0.01 to 0.05. In other words the influence of the scaled watermark signal's coefficients on the spectral shape of the audio data is minimized so that watermark-embedded audio signal is perceptively no different from the original audio signal from the perspective of the listener. The scaled coefficients are then added to the coefficients of wavelet transformed audio signal data at an adder 30. The added coefficients are inverse wavelet transformed at an inverse wavelet transformer 16 to generate adjusted coefficient magnitudes. Finally, the adjusted magnitudes, generated by the inverse wavelet transformer, and the phase component of the audio signal data, obtained by a phase extractor 12, are input to an inverse Fourier transformer 11 to finally generate watermark-embedded audio data.

[00050] Next a watermark extraction from a watermark-embedded audio data will be described referring to FIG. 2. A watermark-embedded audio data undergoes a Fourier transform at a Fourier transformer 20 to generate a first set of coefficients in the frequency domain. Simultaneously or independently an original audio data is also Fourier transformed at a Fourier transformer 23 to generate a second set of coefficients in the frequency domain. The magnitudes of the two set of coefficients, obtained by magnitude extractors 21 and 24 respectively, are further wavelet transformed at wavelet transformers 22 and 25 respectively. The wavelet coefficients associated with the original audio data are subtracted from those with the watermark-embedded audio signal at a subtracter 33. The differences in the coefficients are multiplied by a scaling factor (1/α) and the sign (1 for positive, 0 for none and −1 for negative) of the wavelet transform coefficients associated with the original audio data at a multiplier 34. The sign can be obtained by using a signum function unit 26. Finally, the scaled coefficients, multiplied by the output of the signum function unit 26 is inverse discrete cosine transformed at an inverse discrete cosine transformer 27 to produce a watermark which had been embedded in the original audio data.

[00051] The watermarking method described above can be implemented on a single chip integrated circuit or discrete components. Specifically, a digital signal processor may be programmed to perform the steps in the inventive watermarking.

[00052] While there has been described and illustrated a method and system for inserting a watermark data by discrete cosine transforming the watermark signal and Fourier/wavelet transforming an original audio data, it will be apparent to those skill in the art that variations and modifications are possible without deviating from the broad principles and teachings of the present invention which shall be limited solely by the scope of the claims appended hereto.