KR20020003251A - Audio Watermark Using Wavelet Transform Decomposition Property - Google Patents

Abstract

PURPOSE: An audio watermarking method using decomposition characteristic of wavelet transformation, which generates a watermark that has no noise and is strong to attack using wavelet transformation and psychoacoustic model and inserts the watermark into audio data, is provided to prevent illegal reproduction of the audio data. CONSTITUTION: An audio signal is wavelet-transformed using wavelet transformation to be decomposed into an approximation value and a detail value. The original signal is compressed using CODEC that employs psychoacoustic model and then reproduced again to perform wavelet transformation, to thereby obtain S64. Werr that is the difference between the original signal and the approximation value of S64 is obtained. The value of the approximation region of the original signal is multiplied by a PN-sequence value, and samples are multiplied by different scale factors using pseudo random noise as the scale factors, to generate random watermark. The Werr and random watermark are added up to generate watermark. In order to detect a signal into which watermark was inserted, wavelet transformation is performed and a difference between two signals to extract a watermark value. Correlation between the extracted value and the original watermark value is inspected to verify the ownership of watermark.

Description

Audio Watermark Using Wavelet Transform Decomposition Property

The present invention relates to an audio watermark method using the decomposition characteristics of a wavelet transform, and more particularly, using a wavelet transform and a psychoacoustic model in a watermarking technique, which is a technique for preventing illegal copying of audio data, The present invention relates to an audio watermark method using a decomposition characteristic of a wavelet transform that prevents illegal copying of audio data by generating and inserting a watermark satisfying an attack-resistant characteristic.

Recently, with the development of digital technology, multimedia information is rapidly increasing, and as the network develops, the multimedia information has been easily obtained.

In particular, audio is compressed by technologies such as MPEG, AAC, VQF, and so on, and is easily propagated through the Internet.

This high quality audio compressed data at the CD level enables AOD (Audio On Demand) services on a network.

However, due to the nature of digital data, compressed audio data is easy to copy and copy, and it is frequently illegally distributed on the network.

To prevent such piracy, the US Recording Industry Association of America Inc. is working on how to form and protect the Secure Digital Music Initiative under its umbrella.

In addition, various institutions around the world are making various proposals for the standardization of digital copyright.

In general, a copy protection system uses a method of embedding a watermark in the data or encrypting the data using a public key algorithm, a third party authentication system, or an encryption algorithm in order to protect digital data.

However, due to the development of digital signal processing technology and the limitation of the application sound output method on the PC, the original data can be obtained from the data encrypted by D / A, A / D conversion or output data capture. There is a problem.

Watermarks are used to compensate for the shortcomings of such systems, which can prevent illegal use or redistribution by inserting the copyright holder's or distributor's information into the data.

However, the existing watermarking techniques are not yet complete, and there is a problem in that the watermark does not function as the watermark is deformed or lost during data conversion.

The present invention is to solve the problems of the conventional watermarking technology as described above, using the wavelet transform and psychoacoustic model audio using the decomposition characteristics of the wavelet transform that exhibits a strong characteristic against attack without noise Its purpose is to provide a watermark method.

1 is a block diagram showing a process of transformation and synthesis by wavelet transformation;

2 is an exploded view of the wavelet used in the present invention,

3 is a graph illustrating a difference between an encoded value and an axonation value of an original signal;

4 is a block diagram illustrating a conversion process according to the present invention.

5 is a block diagram showing a watermark extraction process according to the present invention;

In order to achieve the above object, the present invention includes the steps of wavelet transforming an audio signal using a wavelet transform method and decomposing it into an apposition value and a detail value; S64 is obtained by compressing the original signal using a codec using a psychoacoustic model to generate a watermark, such as a characteristic of noise generated during compression, and then restoring it again to perform wavelet transformation as in the original signal. Wow; Obtaining Werr, which is a difference between the original signal and the A64's a proxy value; Each sample is multiplied by the original signal's approximation region (Soriginal) and the PN-sequence (pnseq) value, and pseudo random noise equal to the length of the proxy is used as the scale factor (α). Generating a random watermark (Wwat) by multiplying different scale factors for each time; Generating a watermark (W) by adding Werr and Wwat generated in the above step to each other; Extracting the watermark value through the difference between the two signals after performing wavelet transform to detect the signal with the watermark embedded therein; An audio watermark method using a decomposition characteristic of a wavelet transform, comprising verifying ownership of a watermark by investigating correlation between the value extracted in the step and an original watermark value.

(Example)

A detailed watermarking and decomposition method of an audio watermark method using the decomposition characteristic of the wavelet transform according to the present invention will be described.

1 is a block diagram illustrating a process of transforming and synthesizing by a wavelet transform, FIG. 2 is a wavelet resolution diagram used in the present invention, and FIG. 4 is a block diagram showing a conversion process according to the present invention. 5 is a block diagram illustrating a watermark extraction process according to the present invention.

First, the psychoacoustic model and watermark will be described before explaining the present invention.

Perceptual Coding is an audio coding method that compresses data without degrading the quality of data. There are methods such as MPEG, AAC, and VQF, which use psychoacoustic models. To increase the compression ratio.

The psychoacoustic model uses the masking characteristics of the human auditory characteristics, and does not compress the data of the part which is not heard by the masking.

In perceptual coding, quantization error and components below the masking curve are matched to the masking curve to generate noise during data decoding.

However, this is not audible to the human ear due to the masking nature.

When a watermark for protecting copyright is inserted into audio data, when the watermarked data is compressed and converted, the watermark is easily removed due to the application of a psychoacoustic model and noise due to quantization.

Therefore, in order to make the watermark strong during compression, if the watermark is similar to the noise of the inserted noise during compression, the watermark is not seen by the masking property and has a strong characteristic against deformation by compression.

In addition, the quantization error should also be taken into account in the transformation of the data by compression.

The present invention inserts a watermark using the characteristics of the psychoacoustic model as described above, and a detailed description thereof is as follows.

In the present invention, a watermark is inserted using a spread spectrum technique proposed by I. J. Cox.

The spread spectrum technique sends a narrowband signal over a larger bandwidth so that the energy of the signal is very small at one frequency.

There are various methods for converting a signal to frequency, such as FFT and DCT, but the wavelet transform is used in the present invention.

The wavelet transform is a kind of subband filtering that decomposes the signal by the wavelet basis.

The wavelet transform can multi-level resolve the input signal. As level L increases, the resolution for frequency improves while the resolution for time decreases.

In addition, the decomposition ability varies depending on the wavelet basis function used at this time. In the present invention, the dbN wavelet basis function is used.

Enough regu- larity must be ensured for faithful division of the frequency band during wavelet conversion.

In order to maintain sufficient regulation, the wavelet with the largest vanishing moment should be selected.

For dbN wavelets, N represents the vanishing moment.

However, as N becomes longer, the filtering is performed proportionally, while the amount of calculation increases.

Here, FIG. 1 illustrates a process of transforming and synthesizing by wavelet transform, and passing through a low pass filter and a high pass filter to decompose them into applications and details. do.

2 is an exploded view of the wavelets used in the present invention.

3 is a graph showing the difference between the encoded signal and the agglomeration value of the original signal.

In order to generate watermarks such as the characteristics of noise generated during compression, the original signal is compressed to 64 kbps using a codec using a psychoacoustic model, and then restored again to perform wavelet transformation like the original signal. Obtain

Then we obtain Werr (see Equation 1), which is the difference between the approximation values of the two signals.

In order to generate a random watermark (Wwat) as shown in Equation 2, multiply the value (Soriginal) and the PN-sequence (pnseq) value of the proxy region by the original signal, and the pseudo randomness equal to the length of the proxy. Pseudo Random Noise is used as the scale factor α to multiply each sample with a different scale factor.

Werr and Wwat generated as described above are added as shown in Equation 3 to generate a watermark (W).

The above process is shown in block diagram in FIG.

In order to detect the signal with the watermark embedded therein, after performing wavelet transform as shown in FIG. 5, the watermark value is extracted through the difference between the two signals.

As shown in Equation 4 below, the extracted value is examined for correlation with the original watermark value to verify ownership of the watermark.

(Where w is the extracted watermark and y is the watermark for detection)

Using the present invention as described above, the audio signal is divided into blocks of 1024 samples, each block is taken, and wavelet transform is performed on the wavelet basis of db 10, db 20, db 30, and db 40, respectively.

In addition, the experiment was performed by changing the number of levels to 4, 5, 6, and 7, respectively, and the results are shown in Tables 1 to 4 below.

Low pass Echo MPEG db 10 0.6498 0.8554 0.7575 db 20 0.6531 0.8554 0.7568 db 30 0.6542 0.8549 0.7584 db 40 0.6554 0.8548 0.7613

Low pass Echo MPEG db 10 0.6760 0.8406 0.7967 db 20 0.6773 0.8369 0.8066 db 30 0.6705 0.8390 0.8097 db 40 0.6732 0.8388 0.8114

Low pass Echo MPEG db 10 0.7879 0.7839 0.7826 db 20 0.8244 0.8167 0.8336 db 30 0.7819 0.7838 0.7866 db 40 0.8250 0.8189 0.8350

Low pass Echo MPEG db 10 0.8497 0.7735 0.7206 db 20 0.8553 0.8005 0.7783 db 30 0.8538 0.7975 0.7702 db 40 0.8431 0.8122 0.7748

In addition, as an attack to remove the watermark, the following three methods were used.

Attack 1: low pass filtering

Cutoff frequency: 3500 Hz

RESULTS: From Table 1 to Table 4, the overall robustness to low pass filter attack is shown. The higher the level, the more robust.

Attack 2: Echo

Echo is typically implemented using a Comb filter or an all pass filter.

This can be seen as a method of attacking over the entire frequency range.

Unlike white noise, it has an attack characteristic that bursts at a specific frequency according to the original signal. (Delay 0.01s, Volume 0.4)

Result: Unlike a low pass filter attack, the higher the level, the weaker the result.

Attack 3: MPEG Compression

MPEG compression does not compress data that is inaudible due to human auditory characteristics using a psychoacoustic model. (MPEG 1 Layer-3, 64kbps, mono, 44.1kHz)

Results: When the psychoacoustic model was not applied, a low detection rate of 0.1 to 0.3 was obtained after the MPEG compression attack. The psychoacoustic model showed excellent performance of 0.7 or more. Levels 5 and 6 show relatively strong characteristics.

The present invention relates to the generation of a watermark signal resistant to MPEG compression using a psychoacoustic model. I verified the effect.

The test results were examined only for the 'four seasons' of the classical music Vivaldi, so the results of music data with different characteristics should be examined.

The audio watermark method using the decomposition characteristic of the wavelet transform according to the present invention made as described above has no noise by using the wavelet transform and the psychoacoustic model in the watermarking technique, which is a technique for preventing illegal copying of the audio data. By generating and inserting a watermark that satisfies the strong characteristic, the illegal copying of audio data is prevented.

In the above, the present invention has been described with reference to specific preferred embodiments, but the present invention is not limited to the above-described embodiments, and the present invention is not limited to the scope of the present invention. Various changes and modifications can be made by the user.

Claims (1)

In the method of embedding and extracting a watermark in the audio data, Wavelet transforming the audio signal using the wavelet transform method and decomposing it into an approximation value and a detail value; S64 is obtained by compressing the original signal using a codec using a psychoacoustic model to generate a watermark, such as a characteristic of noise generated during compression, and then restoring it again to perform wavelet transformation as in the original signal. Wow; Obtaining Werr, which is a difference between the original signal and the A64's a proxy value; Each sample is multiplied by the original signal's approximation region (Soriginal) and the PN-sequence (pnseq) value, and pseudo random noise equal to the length of the proxy is used as the scale factor (α). Generating a random watermark (Wwat) by multiplying different scale factors for each time; Generating a watermark (W) by adding Werr and Wwat generated in the above step to each other; Extracting the watermark value through the difference between the two signals after performing wavelet transform to detect the signal with the watermark embedded therein; And verifying ownership of the watermark by investigating the correlation extracted from the extracted value with the original watermark value.