CN102324234A - Audio watermarking method based on MP3 encoding principle - Google Patents

Abstract

The invention discloses an audio watermarking method based on the MP3 encoding principle, which belongs to the technical field of multimedia digital watermarks. The method includes two processes, i.e. watermark embedment and watermark extraction; a watermark is embedded into a low-frequency MDCT (Modified Discrete Cosine Transform) coefficient as an encoding process is carried out synchronously; in order to enhance the robustness of the watermark, an appropriate audio segment is chosen in combination with the frequency domain masking effect in a human auditory system to be embedded; and in order to resist desynchronization attacks, a synchronization mechanism is introduced. By processing the encoding process, the method fulfills the injection of the watermark, and the method can effectively resist common audio attacks, and also has good robustness on desynchronization attacks (such as shearing attacks, time scaling and the like). While taking audio watermark robustness into consideration, the method guarantees the auditory invisibility of audio contents, the computational complexity of the algorithm is low, and the method is easy to implement.

Description

A kind of audio-frequency water mark method based on the MP3 coding principle

Technical field

The invention belongs to multimedia digital digital watermark field, be specifically related to a kind of audio-frequency water mark method based on the MP3 coding principle.

Background technology

In recent years, popular along with a large amount of making and the distribution, particularly MP3 of digital music goods, the importance of DAB product copyright protection becomes increasingly conspicuous, and digital watermark technology also arises at the historic moment.

Various audio edited software makes the research of audio frequency watermark have bigger challenge on human auditory system's (HAS) susceptibility and the network.In recent years, to audio frequency watermark multiple algorithm based on time domain, frequency domain and compression domain has been proposed.Low and the easy realization of time-domain audio watermarking algorithm computation complexity, but relatively poor to the resistivity of attacking.The basic thought of frequency domain algorithm is to hide watermark through the frequency coefficient of technology such as discrete Fourier transformation, discrete cosine transform and wavelet transform change sound signal, compares with Time-Domain algorithm to have higher robustness.Along with compressed format music popular that with MP3 is representative, the compression domain watermark is arisen at the historic moment.Though the compressed domain audio digital watermark technology has been obtained certain progress; But because it is directly the audio content of compressed format to be handled; Dependence to scrambler etc. is stronger, therefore also needs fully to combine MP3 coding principle and new method to promote the development of compression domain watermark.

Present most of Audio Watermarking Algorithm is that watermark is embedded into ad-hoc location and detects from this position; Attack the extraction that the change in location that causes will directly have influence on watermark and desynchronize, therefore in algorithm, introduce synchronization mechanism and seem particularly important by shear, add or deletion sample etc.Desynchronizing to attack for opposing has proposed multiple synchronous code, like exhaustive search, display synchronization and constant watermark etc., attacks the influence that watermark extracting is caused though these methods have reduced to desynchronize to a certain extent, but still has limitation.The algorithm of design can be resisted certain specific attack of desynchronizing, but still can not effectively resist other calculated attack.

Summary of the invention

For overcoming the deficiency of prior art; The present invention provides a kind of audio-frequency water mark method based on the MP3 coding principle; With attacks such as effective opposing compression, resampling, weightization, filtering, and can accurately locate synchronizing signal, thereby correct extraction goes out watermark information; For reaching this purpose, the present invention adopts following technical scheme:

A kind of audio-frequency water mark method based on the MP3 coding principle may further comprise the steps:

Watermark embed step is following:

Watermarking images to be embedded is carried out scramble and does dimension-reduction treatment, obtain the binary sequence of one dimension;

The synchronous code information of certain-length is embedded in the audio file;

Behind the embedded location that synchronous code is confirmed, the original audio signal after decompressing is carried out segmentation according to masking effect, as candidate's audio section of watermark embedding;

In conjunction with the MP3 coding principle, choose the MDCT low frequency coefficient of the original audio data signal of decoding back candidate's audio section and proceed segmentation, the little piecemeal of coefficient magnitude is handled with embed watermark information;

Carrying out simultaneously with audio-frequency signal coding of embed watermark information obtains the MP3 audio frequency of embed watermark information.

The step of watermark extracting:

Detect synchronous code information, watermark information is extracted in the correct back of verification;

Decoded audio signal identical principle when embedding is carried out segmentation, confirm the sound signal segmentation that is used to embed according to masking effect;

Principle when embedding is carried out segmentation to the low frequency MDCT coefficient of the sound signal segmentation confirmed and is confirmed to embed the MDCT coefficient of watermark, and the size through adjacent two coefficients before and after relatively extracts watermark information;

The watermark information that extracts is carried out inverted disorderly, and after rising dimension, obtain the watermarking images that extracts.

The detailed step that above-mentioned watermark embeds is:

Watermarking images is carried out the Arnold scramble, obtain one-dimensional sequence W={wi|wi=0 or 1} after the dimension-reduction treatment, 1≤i≤Len wherein, Len=W * H, W, H represent the height and the width of watermarking images respectively;

In time domain synchronous code is embedded in the original audio signal, characterizes the starting position that watermark embeds;

To the complete voice data after decompressing,, calculate the short-time energy of each segmentation confirming to carry out segmentation behind the embedded location.According to human auditory system's time domain masking effect, a small amount of variation that the signal that energy is high took place in the short time of front and back is difficult to discovered.Therefore choose Len segmentation with big short-time energy candidate segment as the watermark embedding;

MDCT low frequency coefficient to the original audio data signal of candidate's audio section is proceeded segmentation; Since the change of MDCT coefficient affects auditory effect and the invisibility of watermark impacted, therefore choose the less relatively segmentation of M section magnitude and adopt quantification mechanism to carry out the embedding of watermark; Concrete quantification principle is:

Wherein,

When embedding, watermark carries out the MP3 coding, the MP3 audio file that contains watermark information after obtaining compressing.

The detailed step of above-mentioned watermark extracting is:

Detect synchronous code information, correctly locate the extracting position that extracts watermark through synchronous code;

Decoded audio signal identical principle when embedding is carried out segmentation, calculate the short-time energy of each segmentation, choose the sound signal segmentation that Len segmentation with big short-time energy is used to extract according to masking effect;

Rule during according to embedding is determined the MDCT low frequency coefficient that extracts the watermark segmentation, and extracts watermark information according to the corresponding quantization method;

The information that extracts is carried out the inverted unrest and risen the watermarking images that obtains extracting after the dimension.

In the above-mentioned watermark extracting step, M the watermark information that each stage extraction is gone out adopts the method for statistical study finally to confirm watermark information.Each bit of all watermark informations that extracts is carried out statistical, be 1 above half and think that this watermark information is 1, think that this watermark information is 0 if be 0 above half.Repeat this process by bit, finally obtain complete watermark information.

The beneficial effect of this programme is: the first, introduced synchronous code mechanism, effectively raised the robustness that the audio frequency watermark opposing is desynchronized and attacked.The second, algorithm watermark embed process and cataloged procedure are synchronous, can accomplish to load watermark in real time, have satisfied the requirement of audio frequency watermark real-time, help through engineering approaches and realize.The 3rd, confirm the candidate MDCT low frequency coefficient that is used to embed in conjunction with time domain masking effect of human auditory system etc., improved the robustness of audio frequency watermark.

Description of drawings

Fig. 1 is the process flow diagram that watermark embeds;

Fig. 2 is the sound signal block plan;

Fig. 3 is the process flow diagram of watermark extracting;

Fig. 4 is MP3 encoding and decoding principle flow charts;

Fig. 5 (a) is original array sound signal synoptic diagram; Fig. 5 (b) is the digital audio and video signals synoptic diagram behind the embed watermark.

Embodiment

Further specify technical scheme of the present invention below in conjunction with accompanying drawing and through embodiment.

A kind of audio-frequency watermark processing method based on the MP3 coding principle of technical scheme of the present invention comprises watermark embedding and two processes of watermark extracting.

1, watermark embeds

(1) watermark information pre-service.Watermarking images is carried out the Arnold scramble, obtain one-dimensional sequence W={wi|wi=0 or 1} after the dimension-reduction treatment, 1≤i≤Len wherein, Len=W * H, W, H represent the height and the width of watermarking images respectively;

(2) synchronous code embeds.In Fig. 1, K is image scrambling and handles required key.Use the image scrambling technology that pre-service is carried out in watermark, can improve the security of watermark information, strengthen the ability of watermark opposing malicious attack.Watermark information through behind the scramble will become a series of mixed and disorderly and nonsensical sequences by raw information, destroy the attribute of image self and the correlativity between information.Like this, though illegal person intercept and capture still cannot be correct behind the watermark information the original watermark information that gets access to, thereby can't reach the purpose of attack.

(3) embedded location is chosen.To the complete voice data after decompressing, confirming to carry out segmentation behind the embedded location, the expression formula of every section sound signal that is used for embed watermark information is as follows:

S _k＝{s _k(i)＝s((k-1)N+i)|k＝1，2，...，Len；i＝1，2，...，N}

The interior a small amount of variation that takes place of short time was difficult to discovered before and after the time domain masking effect of human auditory's model was meant the high signal of energy.Release the bigger audio section of short-time energy thus and can better hide watermark information.In order to make algorithm can reach better not sentience, in audio parsing, choose sound signal that the Len piece has big short-time energy and carry out watermark and embed.Wherein the definition of the short-time energy computing formula of signal is as follows:

$E\left(k\right)=\underset{i=1}{\overset{N}{\mathrm{\Σ}}}{s}^2(\mathrm{kN}+i)$

The MDCT low frequency coefficient is chosen.MDCT low frequency coefficient to the original audio data signal of candidate's audio section is proceeded segmentation; Since the change of MDCT coefficient affects auditory effect and the invisibility of watermark impacted, therefore choose the less relatively segmentation of M section magnitude and adopt quantification mechanism to carry out the embedding of watermark.Concrete quantitative formula is following:

Wherein,

(5) MP3 cataloged procedure as shown in Figure 1 carries out the MP3 coding, the MP3 audio file that contains watermark information after obtaining compressing when watermark embeds.

2. watermark extracting

(1) confirms the extraction reference position of watermark.After extracting synchronous code information, carry out verification, and confirm to begin to extract the position of watermark information.

(2) confirm to extract watermark location.As shown in Figure 2 with the sound signal segmentation, confirm the sound signal piecemeal and the segmentation of low frequency MDCT coefficient of extraction watermark.

(3) extract watermark information.Extract watermark information by the following formula principle.

$w_k=\left\{\begin{array}{c}1,\underset{i=1}{\overset{M}{\mathrm{\Σ}}}{w^{\′}}_k\left(i\right)>\frac{M}{2}\\ 0,\underset{i=1}{\overset{M}{\mathrm{\Σ}}}{w^{\′}}_k\left(i\right)\≤\frac{M}{2}\end{array}\right.$

1≤k≤Len wherein, Len w _kBe reassembled as watermark information w.

(4) watermark information that extracts is carried out inverted and disorderly handle and obtain watermark information, and after rising dimension, obtain the watermarking images that extracts.

The invisibility of audio frequency watermark is except that the human auditory system evaluation of subjectivity, and Y-PSNR SNR is the important indicator of the objective evaluation watermark transparency.If original audio B has N sampled point, the audio file that embedded watermark is A, and the calculating of audio frequency signal to noise ratio snr can be calculated through following formula:

$\mathrm{SNR}(B,A)=10\lg (\underset{n=1}{\overset{N}{\mathrm{\Σ}}}{B}^2\left(N\right)/\underset{n=1}{\overset{N}{\mathrm{\Σ}}}{[B\left(N\right)-A\left(N\right)]}^2)$

In order to weigh the watermark that extracts and the similarity degree of original watermark, this paper adopts the normalization calculating formula of similarity to come the performance of objective judge watermarking algorithm, shown in the following formula of normalization calculating formula of similarity of two watermarking images:

$\mathrm{NC}=\frac{\underset{i}{\mathrm{\Σ}}\underset{j}{\mathrm{\Σ}}w{(i,j)}^{\′}\×w(i,j)}{\underset{i}{\mathrm{\Σ}}\underset{j}{\mathrm{\Σ}}w{(i,j)}^2}$

Wherein, and w (i, j) ' be the watermark information that extracts, (i j) is the information of original watermark to w.The criterion whether watermark exists is: if NC＞τ can judge that watermark exists, otherwise watermark does not exist.The selection of τ has determined false-alarm probability and false dismissal probability, and the selection of τ needs the statistical information decision of a large amount of experiments, similarity thresholding τ=0.5 in the experiment.

Influence according to the sound signal synchronization structure generally is divided into two big types to attack:

(1) do not influence the attack of sound signal synchronization structure: the MP3 compression, adds echo, resampling and weightization etc. at low pass/bandpass filtering.

(2) sound signal is caused the attack of synchronization structure influence: (Time Scale Modification TSM) attacks etc. like jitter attack, modified tone (pitch-shift warping), time scale distortion.

The watermark extracting test is all carried out emulation experiment two big types of different attacks respectively, and comes the extraction effect of objective evaluation watermark through the Y-PSNR PSNR of sound signal with the error rate BER that extracts watermarking images.

The effect of the inventive method can be through reflecting like following table 1.

Watermark extracting accuracy under table 1 various attack

Attack	BER	NC	SNR(dB)
				Do not attack	0％	1	25.568
Add the attack of making an uproar	2.10％	0.979	25.418
				Compression attack	2.33％	0.9767	25.698
High-pass filtering	3.04％	0.9696	26.145
				LPF	3.10％	0.969	25.647
Signal strengthens	3.06％	0.9694	26.301

The above; Be merely the preferable embodiment of the present invention, but protection scope of the present invention is not limited thereto, anyly is familiar with this technological people in the technical scope that the present invention disclosed; The variation that can expect easily or replacement all should be encompassed within protection scope of the present invention.Therefore, protection scope of the present invention should be as the criterion with the protection domain of claim.

Claims (5)

1. the audio-frequency water mark method based on the MP3 coding principle is characterized in that, this method comprises watermark embed process:

Watermark embed step is following:

Watermarking images to be embedded is carried out scramble and does dimension-reduction treatment, obtain the binary sequence of one dimension;

The synchronous code information of certain-length is embedded in the audio file;

Behind the embedded location that synchronous code is confirmed, the original audio signal after decompressing is carried out segmentation according to masking effect, as candidate's audio section of watermark embedding;

In conjunction with the MP3 coding principle, choose the MDCT low frequency coefficient of the original audio data signal of decoding back candidate's audio section and proceed segmentation, the little piecemeal of coefficient magnitude is handled with embed watermark information;

Carrying out simultaneously with audio-frequency signal coding of embed watermark information obtains the MP3 audio frequency of embed watermark information.

2. audio-frequency water mark method based on the MP3 coding principle is characterized in that this method comprises watermark extraction process:

The watermark extracting step is following:

Detect synchronous code information, watermark information is extracted in the correct back of verification;

Decoded audio signal identical principle when embedding is carried out segmentation, confirm the sound signal segmentation that is used to embed according to masking effect;

Principle when embedding is carried out segmentation to the low frequency MDCT coefficient of the sound signal segmentation confirmed and is confirmed to embed the MDCT coefficient of watermark, and the size through adjacent two coefficients before and after relatively extracts watermark information;

The watermark information that extracts is carried out inverted disorderly, and after rising dimension, obtain the watermarking images that extracts.

3. a kind of audio-frequency water mark method based on the MP3 coding principle according to claim 1 is characterized in that, and is further comprising the steps of:

Watermarking images is carried out the Arnold scramble, obtain one-dimensional sequence W={wi|wi=0 or 1} after the dimension-reduction treatment, 1≤i≤Len wherein, Len=W * H, W, H represent the height and the width of watermarking images respectively;

In time domain synchronous code is embedded in the original audio signal, characterizes the starting position that watermark embeds;

To the complete voice data after decompressing,, calculate the short-time energy of each segmentation confirming to carry out segmentation behind the embedded location.According to human auditory system's time domain masking effect, a small amount of variation that the signal that energy is high took place in the short time of front and back is difficult to discovered.Therefore choose Len segmentation with big short-time energy candidate segment as the watermark embedding;

MDCT low frequency coefficient to the original audio data signal of candidate's audio section is proceeded segmentation; Since the change of MDCT coefficient affects auditory effect and the invisibility of watermark impacted, therefore choose the less relatively segmentation of M section magnitude and adopt quantification mechanism to carry out the embedding of watermark;

When embedding, watermark carries out the MP3 coding, the MP3 audio file that contains watermark information after obtaining compressing.

4. a kind of audio-frequency water mark method based on the MP3 coding principle according to claim 1 is characterized in that the quantification mechanism in the said telescopiny is specially:

Wherein,

5. a kind of audio-frequency water mark method based on the MP3 coding principle according to claim 2 is characterized in that the detailed step of said extraction is:

Detect synchronous code information, correctly locate the extracting position that extracts watermark through synchronous code;

Decoded audio signal identical principle when embedding is carried out segmentation, calculate the short-time energy of each segmentation, choose the sound signal segmentation that Len segmentation with big short-time energy is used to extract according to masking effect;

Rule during according to embedding is determined the MDCT low frequency coefficient that extracts the watermark segmentation, and extracts watermark information according to the corresponding quantization method;

The information that extracts is carried out the inverted unrest and risen the watermarking images that obtains extracting after the dimension.

Images