CN102664014A - Blind audio watermark implementing method based on logarithmic quantization index modulation - Google Patents

Abstract

The invention discloses a blind audio watermark implementing method based on logarithmic quantization index modulation, and belongs to the technical field of audio watermarks. The blind audio watermark implementing method is characterized by including sufficiently utilizing robustness of vector norms and imperceptibility of the logarithmic quantization index modulation based on mu-law companding; improving safety and robustness of a watermark by the aid of a chaos sequence encryption watermark image; converting the sectioned vector norms of wavelet approximate components into a transform domain via the mu-law companding; then embedding an encryption binary watermark image; and extracting the binary watermark image after the watermark is attacked. The audio watermark algorithm has the advantages of high volume, fine tone, imperceptibility and robustness and low complexity, the watermark still can be correctly extracted by the aid of the method under various watermark attack conditions, and accordingly amplitude attack can be effectively resisted by the method.

Description

A kind of blind audio frequency watermark implementation method based on the logarithm quantization index modulation

Technical field

The present invention relates to a kind of blind audio frequency watermark implementation method based on the logarithm quantization index modulation; Adopt this method; Not only have capacity height, characteristics that tonequality is good, also have the advantage that sentience is not good, robustness is good, complexity is low, under the situation that various watermarks are attacked; Still can utilize this method correct extract watermark, and effectively opposing amplitude of the method is attacked.

Background technology

Digital watermarking (Digital Watermarking) is that the redundancy of the consciousness system (like vision or auditory system) that utilizes the people embeds some identification informations in carriers such as image, audio frequency, but does not influence the use value of original vector.

According to the difference of watermark carrier, can digital watermark be divided into several big type of video watermark, image watermark, audio frequency watermark etc.The comparative maturity that present video, Image Watermarking Technique have developed, owing to following reason, audio frequency watermark becomes the emphasis and the focus of watermark research gradually: one of which, voice applications is extensive, is the human important tool that exchanges; Its two, have bulk redundancy in the audio frequency, make things convenient for embed watermark information.But audio frequency watermark has with image watermark very big difference is arranged, main because: one, audio frequency is an one-dimensional signal; Two, human auditory system is much sensitiveer than the mankind's vision system, implements more difficult than vision on the not sentience acoustically; Three, amount of audio data is bigger, and is mainly used in environment such as broadcasting, online distribution, seeks very difficulty of original audio, so the Detection and Extraction of watermark promptly realize blind Detecting not in requisition for original audio in principle.

Generally, audio frequency watermark should satisfy following basic demand: (1) is sentience not: be meant to add the degree that influences the original audio quality after the watermark; (2) robustness: be meant that the audio frequency that has added watermark can resist the degree of various attack; (3) capacity: be meant that per second is embedded into the quantity of the watermark of original audio; (4) complexity: the time and the space complexity that are meant algorithm.Usually above the requirement is conflicting, so we often select a balance according to the needs of practical application.

One of method of the audio frequency watermark that adopts at present is: based on the audio frequency watermark of spread spectrum.Major advantage is: can make signal under the low power transmissions situation, improve antijamming capability, and disguised high, even the dropout of several frequency ranges, but restoring signal still, and guarantee to have only the receiver of known extensions function could detection signal, can realize blind Detecting.Major defect is: take too many frequency spectrum, embedded quantity is little; The blind Detecting prerequisite is to contain between sound signal and the watermark random signal of watermark must reach fully synchronously.For obtaining less watermark error extraction rate, watermark length must be enough big, but can increase detection complexity like this and increase time delay.

Two of the method for the audio frequency watermark that adopts at present is: based on the audio frequency watermark of quantization index modulation.Major advantage is: it is simple to have an algorithm, and complexity is low, and containing much information of embedding realizes blind extraction easily, and not sentience, between robustness and capacity, reach balance.Major defect is: responsive to noise ratio, this algorithm uses fixed quantization step, has introduced bigger quantizing noise, the data fluctuations that therefore causes for quantizing noise and the extraction error code that causes is higher.

Summary of the invention

At present; Based on the audio frequency watermark of quantization index modulation because it is more superior and can reach balance preferably at each aspect of performance; Obtain extensive concern and fast development; But what its quantizing process was all taked at present is uniform quantization, fixed quantization step, though its be easy to realize, to noise-sensitive and its robustness and not the performance aspect the sentience be not enough good.Analyze more top two kinds of watermarking algorithms, be not difficult to find that audio frequency watermark contradiction is both need improve the robustness of audio frequency watermark, need improve the not sentience of audio frequency watermark again.For this reason; This research is considered from this two aspect, has proposed a kind of new Audio Watermarking Algorithm, solves existing algorithm in robustness and the shortcoming aspect the sentience not; And overcome the high difficulty of complexity; The present invention effectively combines the robustness of vector norm with to the not sentience of quantification, is transformed into transform domain to the vector norm of the wavelet approximation coefficients of segmentation through the mu-law companding, embeds the binary watermark that upsets then; After the watermark attack, extract binary watermark.The Audio Watermarking Algorithm that the present invention proposes; Not only have capacity height, characteristics that tonequality is good; Also has the advantage that sentience is not good, robustness is good, complexity is low; Under the situation that various watermarks are attacked, still can utilize this method correct extract watermark, and effectively opposing amplitude of the method is attacked.

The invention is characterized in that said method contains following steps successively:

Step (1) is at transmitting terminal, the telescopiny of audio frequency watermark, and steps in sequence is following;

The original audio signal X={x of step (1.1) to setting _i, 1≤i≤L} carries out 2 grades of wavelet transform DWT, and the wavelet basis of employing is Db4, and wherein L is the number of original audio sampled point, must be similar to component B={b then _i, 1≤i≤L/4} is divided into M * M mutual nonoverlapping frame S to approximate component _i, 1≤i≤M * M, M are positive integers, every frame S _iLength be

Every like this frame is hidden 1 bit-binary watermark information, and wherein the size of binary watermark is M * M;

Step (1.2) is according to formula:

σ _i＝||S _i||，

$S_i={\mathrm{\σ}}_i{u}_i^T,$

$u_i^T=S_i/{\mathrm{\σ}}_i,$ Be 1 * p dimensional vector,

Calculate every frame S _iVector norm σ _i, 1≤i≤M * M, and let σ _Max=max (σ _i), 1≤i≤M * M;

Step (1.3) is to the vector norm σ of every frame _iDo the mu-law companding, concrete calculating as follows:

c _i＝σ _i/σ _max，

λ _i＝ln(1+μc _i)/ln(1+μ)，

d＝λ _imodΔ，

Δ is predefined quantization step, 0≤Δ≤1, and μ is the parameter of mu-law companding, 0≤μ≤255;

Step (1.4) is utilized binary watermark W={w _Ij, 1≤i≤M, 1≤j≤M} and chaos sequence

The watermarking images W={w that obtains upsetting _Ij, 1≤i≤M, 1≤j≤M}:

Be xor operation, W={w _Ij, 1≤i≤M, 1≤j≤M} will be embedded among the original audio signal X, and the i of i, j difference representative image matrix is capable, the j row,

If w (i, j)=1, then by following rules modification λ _i:

${\mathrm{\&lambda;}}_i^{\&prime;}=\left\{\begin{array}{cc}{\mathrm{\&lambda;}}_i-\left({\mathrm{\&lambda;}}_i\mathrm{mod}\mathrm{\&Delta;}\right),& \mathrm{if}(d<\mathrm{\&Delta;}/2)\\ {\mathrm{\&lambda;}}_i+\mathrm{\&Delta;}-\left({\mathrm{\&lambda;}}_i\mathrm{mod}\mathrm{\&Delta;}\right),& \mathrm{if}(d>\mathrm{\&Delta;}/2)\end{array}\right.$

If w (i, j)=0, then by following rules modification λ _i:

λ′ _i＝λ _i+Δ/2-(λ _imodΔ)；

Step (1.5) is utilized formula:

${\mathrm{\&sigma;}}_i^{\&prime;}=({(1+\mathrm{\&mu;})}^{{\mathrm{\&lambda;}}_i^{\&prime;}}-1)*{\mathrm{\&sigma;}}_{\max }/\mathrm{\&mu;},$

is 1 * p dimensional vector; Definition as mentioned above

Calculate the vector norm σ ' that revised _iWith frame S ' _i, 1≤i≤M * M, and utilize frame S ' _iReconstruct be similar to component B '=b ' _i, 1≤i≤L/4};

Step (1.6) pairing approximation component B '=b ' _i, 1≤i≤L/4} carries out 2 grades discrete wavelet inverse transformation IDWT, obtains containing the sound signal X ' of watermark;

Step (2) is at receiving end, the leaching process of audio frequency watermark, and steps in sequence is following;

Sound signal X '={ the x of step (2.1) to containing watermark _i, 1≤i≤L} carries out 2 grades wavelet transform DWT, the wavelet basis of employing is Db4, wherein L definition as stated, must be similar to then component B '=b ' _i, 1≤i≤L/4} is divided into M * M mutual nonoverlapping frame S ' to approximate component _i, 1≤i≤M * M, every frame S ' _iLength be

Step (2.2) is according to formula:

σ′ _i＝||S′ _i||，

Calculate every frame S ' _iVector norm σ ' _i, 1≤i≤M * M;

Step (2.3) is to the vector norm σ ' of every frame _iDo the mu-law companding, concrete calculating as follows:

c′ _i＝σ′ _i/σ _max，

λ′ _i＝ln(1+μc′ _i)/ln(1+μ)，

d＝λ′ _imodΔ，

Δ is predefined quantization step, defines as stated, and μ is the parameter of mu-law companding, and definition as stated;

Step (2.4) basis:

$w(i,j)=\left\{\begin{array}{cc}0,& \mathrm{if}(d>\mathrm{\&Delta;}/4\mathrm{andd}<3*\mathrm{\&Delta;}/4)\\ 1,& \mathrm{if}\left(\mathrm{others}\right)\end{array}\right.$

$W=W\&CirclePlus;E,$

Obtain original watermark image W={w _Ij, 1≤i≤M, 1≤j≤M}.

The blind audio frequency watermark implementation method that the present invention proposes based on the logarithm quantization index modulation; Its advantage mainly comprises: adopt the not sentience that quantification has been improved algorithm; Adopt vector norm to improve the robustness of algorithm; Adopt chaos sequence encrypted watermark image to improve the security and the robustness of watermark, be transformed into transform domain to the vector norm of the wavelet approximation coefficients of segmentation, embed the binary watermark of encrypting then through the mu-law companding; After the watermark attack, extract binary watermark.The Audio Watermarking Algorithm that the present invention proposes; Not only have capacity height, characteristics that tonequality is good; Also has the advantage that sentience is not good, robustness is good, complexity is low; Under the situation that various watermarks are attacked, still can utilize this method correct extract watermark, and effectively opposing amplitude of the method is attacked.

Description of drawings

Fig. 1 is the merge module of Audio Watermarking Algorithm.

Fig. 2 is the extraction module of Audio Watermarking Algorithm.

Fig. 3 is the system chart of Audio Watermarking Algorithm.

Embodiment

The blind audio frequency watermark implementation method based on the logarithm quantization index modulation that the present invention proposes comprises two parts of watermark merge module and watermark extracting module:

1) watermark merge module: be used for embed watermark information.To original audio signal X={x _i, 1≤i≤L} carries out 2 grades wavelet transform DWT, and wherein L is the number of original audio sampled point, must be similar to component B={b _i, 1≤i≤L/4} is divided into M * M mutual nonoverlapping frame S to approximate component _i, 1≤i≤M * M calculates every frame S _iVector norm σ _i, 1≤i≤M * M, and let σ _Max=max (σ _i), 1≤i≤M * M, and to the vector norm σ of every frame _iMake following calculating: c _i=σ _i/ σ _Max, λ _i=ln (1+ μ c _i)/ln (1+ μ), d=λ _iMod Δ, Δ are predefined quantization steps, and μ is the parameter of mu-law companding.Utilize binary watermark W={w _Ij, 1≤i≤M, 1≤j≤M} and chaos sequence

And formula

The watermarking images W={w that we can obtain upsetting _Ij, 1≤i≤M, 1≤j≤M}, if w (i, j)=1, then by following rules modification λ _i: ${\mathrm{\&lambda;}}_i^{\&prime;}=\left\{\begin{array}{cc}{\mathrm{\&lambda;}}_i-\left({\mathrm{\&lambda;}}_i\mathrm{Mod}\mathrm{\&Delta;}\right),& \mathrm{If}(d<\mathrm{\&Delta;}/2)\\ {\mathrm{\&lambda;}}_i+\mathrm{\&Delta;}-\left({\mathrm{\&lambda;}}_i\mathrm{Mod}\mathrm{\&Delta;}\right),& \mathrm{If}(d>\mathrm{\&Delta;}/2)\end{array}\right.,$ If w (i, j)=0, then by following rules modification λ _i: λ ' _i=λ _i+ Δ/2-(λ _iThe mod Δ).Utilize formula: ${\mathrm{\&sigma;}}_i^{\&prime;}=({(1+\mathrm{\&mu;})}^{{\mathrm{\&lambda;}}_i^{\&prime;}}-1)*{\mathrm{\&sigma;}}_{\mathrm{Max}}/\mathrm{\&mu;},$

Calculate the vector norm σ ' that revised _iWith frame S ' _i, 1≤i≤M * M, and utilize frame S ' _iReconstruct be similar to component B '=b ' _i, 1≤i≤L/4}, the pairing approximation coefficient B '=b ' _i, 1≤i≤L/4} carries out 2 grades discrete wavelet inverse transformation IDWT, and we can obtain containing the sound signal X ' of watermark;

2) watermark extracting module: be used to extract watermark information.To audio frequency watermark signal X '={ x _i, 1≤i≤L} carries out 2 grades wavelet transform DWT, must be similar to component B '=b ' _i, 1≤i≤L/4} is divided into M * M mutual nonoverlapping frame S ' to approximate component _i, 1≤i≤M * M is according to formula: σ ' _i=|| S ' _i||, calculate every frame S ' _iVector norm σ ' _i, 1≤i≤M * M is to the vector norm σ ' of every frame _iMake following calculating: c ' _i=σ ' _i/ σ _Max, λ ' _i=ln (1+ μ c ' _i)/ln (1+ μ), d=λ ' _iMod Δ, Δ are predefined quantization steps, and μ is the parameter of mu-law companding, according to

With We can obtain original watermark image W={w _Ij, 1≤i≤M, 1≤j≤M}.

Below in conjunction with accompanying drawing, specify content of the present invention:

Fig. 1 is based on the audio frequency watermark merge module of logarithm quantization index modulation.As shown in Figure 1, the wavelet transform DWT to original audio signal X carries out 2 grades is divided into M * M mutual nonoverlapping frame to the approximate component that obtains; Calculate the vector norm of every frame, carry out the mu-law companding then, embed the binary watermark W that is encrypted by chaos sequence; And then carry out the mu-law companding; Calculate the vector norm and this frame revised, repeat said process till all watermark bit all embed, and utilize the approximate component of the frame reconstruct of revising; And the pairing approximation component carries out 2 grades discrete wavelet inverse transformation IDWT, and we can obtain containing the sound signal X ' of watermark.

Fig. 2 is based on the audio frequency watermark extraction module of logarithm quantization index modulation.As shown in Figure 2; The sound signal X ' that contains watermark is carried out 2 grades wavelet transform DWT, be divided into M * M mutual nonoverlapping frame to the approximate component that obtains, calculate the vector norm of every frame; Carry out the mu-law companding then; Extraction with the chaos sequence deciphering, is obtained original binary watermark W by the scale-of-two watermark bit.

Fig. 3 is the system chart of Audio Watermarking Algorithm.As shown in Figure 3, watermark information is embedded in the initial carrier through the watermark merge module, through after the various attack, extracts watermark information through the watermark extracting module again.

Claims (1)

1. blind audio frequency watermark implementation method based on the logarithm quantization index modulation the invention is characterized in that said method contains following steps successively:

Step (1) is at transmitting terminal, the telescopiny of audio frequency watermark, and steps in sequence is following;

The original audio signal X={x of step (1.1) to setting _i, 1≤i≤L} carries out 2 grades of wavelet transform DWT, and the wavelet basis of employing is Db4, and wherein L is the number of original audio sampled point, must be similar to component B={b then _i, 1≤i≤L/4} is divided into M * M mutual nonoverlapping frame S to approximate component _i, 1≤i≤M * M, M are positive integers, every frame S _iLength be

Every like this frame is hidden 1 bit-binary watermark information, and wherein the size of binary watermark is M * M;

Step (1.2) is according to formula:

σ _i＝||S _i||，

$S_i={\mathrm{\&sigma;}}_i{u}_i^T,$

$u_i^T=S_i/{\mathrm{\&sigma;}}_i,$ Be 1 * p dimensional vector,

Calculate every frame S _iVector norm σ _i, 1≤i≤M * M, and let σ _Max=max (σ _i), 1≤i≤M * M;

Step (1.3) is to the vector norm σ of every frame _iDo the mu-law companding, concrete calculating as follows:

c _i＝σ _i/σ _max，

λ _i＝ln(1+μc _i)/ln(1+μ)，

d＝λ _imodΔ，

Δ is predefined quantization step, 0≤Δ≤1, and μ is the parameter of mu-law companding, 0≤μ≤255; Step (1.4) is utilized binary watermark W={w _Ij, 1≤i≤M, 1≤j≤M} and chaos sequence

The watermarking images W={w that obtains upsetting _Ij, 1≤i≤M, 1≤j≤M}:

Be xor operation, W={w _Ij, 1≤i≤M, 1≤j≤M} will be embedded among the original audio signal X, and the i of i, j difference representative image matrix is capable, the j row,

If w (i, j)=1, then by following rules modification λ _i:

${\mathrm{\&lambda;}}_i^{\&prime;}=\left\{\begin{array}{cc}{\mathrm{\&lambda;}}_i-\left({\mathrm{\&lambda;}}_i\mathrm{mod}\mathrm{\&Delta;}\right),& \mathrm{if}(d<\mathrm{\&Delta;}/2)\\ {\mathrm{\&lambda;}}_i+\mathrm{\&Delta;}-\left({\mathrm{\&lambda;}}_i\mathrm{mod}\mathrm{\&Delta;}\right),& \mathrm{if}(d>\mathrm{\&Delta;}/2)\end{array}\right.$

If w (i, j)=0, then by following rules modification λ _i:

λ′ _i＝λ _i+Δ/2-(λ _imodΔ)；

Step (1.5) is utilized formula:

${\mathrm{\&sigma;}}_i^{\&prime;}=({(1+\mathrm{\&mu;})}^{{\mathrm{\&lambda;}}_i^{\&prime;}}-1)*{\mathrm{\&sigma;}}_{\max }/\mathrm{\&mu;},$

is 1 * p dimensional vector; Definition as mentioned above

Calculate the vector norm σ ' that revised _iWith frame S ' _i, 1≤i≤M * M, and utilize frame S ' _iReconstruct be similar to component B '=b ' _i, 1≤i≤L/4};

Step (1.6) pairing approximation component B '=b ' _i, 1≤i≤L/4} carries out 2 grades discrete wavelet inverse transformation IDWT, obtains containing the sound signal X ' of watermark;

Step (2) is at receiving end, the leaching process of audio frequency watermark, and steps in sequence is following;

Sound signal X '={ the x of step (2.1) to containing watermark _i, 1≤i≤L} carries out 2 grades wavelet transform DWT, the wavelet basis of employing is Db4, wherein L definition as stated, must be similar to then component B '=b ' _i, 1≤i≤L/4} is divided into M * M mutual nonoverlapping frame S ' to approximate component _i, 1≤i≤M * M, every frame S ' _iLength be

Step (2.2) is according to formula:

σ′ _i＝||S′ _i||，

Calculate every frame S ' _iVector norm σ ' _i, 1≤i≤M * M;

Step (2.3) is to the vector norm σ ' of every frame _iDo the mu-law companding, concrete calculating as follows:

c′ _i＝σ′ _i/σ _max，

λ′ _i＝ln(1+μc′ _i)/ln(1+μ)，

d＝λ′ _imodΔ，

Δ is predefined quantization step, defines as stated, and μ is the parameter of mu-law companding, and definition as stated;

Step (2.4) basis:

$w(i,j)=\left\{\begin{array}{cc}0,& \mathrm{if}(d>\mathrm{\&Delta;}/4\mathrm{andd}<3*\mathrm{\&Delta;}/4)\\ 1,& \mathrm{if}\left(\mathrm{others}\right)\end{array}\right.$

$W=W\&CirclePlus;E,$

Obtain original watermark image W={w _Ij, 1≤i≤M, 1≤j≤M}.

Images