CN103208288A - Dual encryption based discrete wavelet transform-discrete cosine transform (DWT-DCT) domain audio public watermarking algorithm - Google Patents

Abstract

The invention discloses a dual encryption based DWT-DCT domain audio public watermarking algorithm. The algorithm comprises a watermark embedding process and a watermark detection process. The watermark embedding process sequentially comprises dual encryption of a binary watermark image, segment treatment on an original voice frequency, DWT and DCT of each segment coefficient, quantization embedding, inversed discrete cosine transform (IDCT) and inversed discrete wavelet transform (IDWT) of a quantization coefficient, reorganization and obtaining of a voice frequency containing watermark information. The watermark detection process sequentially comprises segment treatment on an audio signal to be tested, DWT and DCT of each segment coefficient, summing of watermark reference values at corresponding positions of n watermark coefficient series, reconfiguration of a two-dimensional image, inverted scrambling treatment and obtaining of a restored watermark image. By the aid of the algorithm, the transparency of watermarks is effectively guaranteed, the miss ratio of watermarks is greatly reduced, and the robustness and the accuracy of the watermark detection are improved.

Description

Based on the open watermarking algorithm of the DWT-DCT territory audio frequency of double-encryption

Technical field

The present invention relates to image information safety technique field, relate in particular to the open watermarking algorithm of a kind of DWT based on double-encryption-DCT territory audio frequency.

Background technology

In recent years, the argumentation of relevant authenticating water-mark increases day by day both at home and abroad.The pre-service of watermark is more common assurance watermark security method menu; usually it is disorderly and unsystematic to adopt the method for image scrambling to make that image becomes; play the encryption effect; but the image scrambling method has periodically; continue by former method conversion; will restore original watermark information, so the security of this kind encrypted watermark is relative, certain defective be arranged.

Chaos phenomenon is similar at random, the deterministic process that occurs in the nonlinear dynamic system.The principal feature of chaos sequence: (1) form is simple.Only need chaotic maps parameter and starting condition just can conveniently generate, needn't store whole sequence by wasting space.(2) starting condition susceptibility.General different initial value, even quite approaching, the chaos locus sequence that iteration obtains is all inequality.(3) determinacy.As long as given identical initial value is arranged, its corresponding track is identical certainly.Generally provide the chaos sequence of one section finite length, illegal person is difficult to therefrom infer its starting condition.Therefore, digital picture and chaos sequence are carried out computing, with making digital picture also have the characteristics of chaos sequence, reach the purpose of image encryption.Yet the chaos encryption technology only changes the pixel value of image, does not change locations of pixels, can not resist the shearing attack of watermark effectively.

Summary of the invention

In order to overcome the deficiencies in the prior art, the objective of the invention is to propose a kind of shearing attack that has robustness, can effectively resist watermark, to the damage of original image less and the open watermarking algorithm of the transparent good DWT based on double-encryption of watermark-DCT territory audio frequency.

For achieving the above object, technical scheme of the present invention is: the open watermarking algorithm of a kind of DWT based on double-encryption-DCT territory audio frequency may further comprise the steps:

One, the telescopiny of watermark:

The definition original audio signal is A={a (i), 0≤i＜Length}, and wherein, Length is the number of voice data, a (i) is the range value of i voice data;

The definition binary bitmap be W={W (i, j), 0≤i＜M, 0≤j＜N}, wherein, M, N represent the row, column number of watermark, W (i, j) { 0,1} represents the capable j row of the i grey scale pixel value of binary bitmap to ∈;

Its concrete telescopiny is as follows:

1) binary bitmap W _{M * N}Carry out scramble and dimension-reduction treatment with the Arnold conversion, realize once encrypting; Carry out XOR with isometric Logistic chaos sequence then, realize superencipher, namely obtain the watermark W' behind the double-encryption _{M * N}

2) original audio A is carried out staging treating, described segmentation process is:

Choosing element formation n segment length by great-jump-forward is L * M * N * 2 coefficients series IO (i), and wherein, L is the DWT number of transitions, and n * L * M * N * 2＜Length;

3) every section coefficient is carried out L level DWT conversion, get CA ^L, CD ^L, CD ^L-1... CD ¹, wherein, CA ^LBe the approximate component of L level, CD ⁱIt is the details component;

4) to CA ^LCarry out dct transform, the Low Medium Frequency coefficient x (j) that chooses then after the conversion quantizes, and embeds the position with it as watermark and embed watermark information after encrypting, and described coefficient quantization process is as follows:

Make z (j)=[x (j)/Da]; Da is predefined quantization parameter,

If z (j) %2==W'(i), x (j)=z (j) * Da-Da/2 then;

Otherwise x (j)=z (j) * Da+Da/2;

Wherein, the computing of [] expression round, % represents complementation;

5) the coefficient x after quantizing is done idct transform and obtain CA ', and then carry out L level IDWT conversion, at last each section of moisture official seal breath replaced former coefficient of correspondence, namely obtain moisture official seal breath audio frequency;

Two, the testing process of watermark:

Step1: sound signal to be detected is carried out staging treating, and described segmentation process constitutes n section coefficient series for choose element by great-jump-forward;

Step2: each section coefficient is carried out the L level DWT conversion identical with watermark embed process, obtain the approximate component Ca of L level ^L

Step3: to Ca ^LCarry out dct transform, the Low Medium Frequency coefficient x ' that chooses after the conversion (i) calculates and obtains a watermark reference information value, and the formula of described calculating watermark reference information value is:

Wherein, Represent to round up computing, % represents complementation;

Step4: the watermark reference value addition on the corresponding position of n watermark coefficient series, determine a corresponding watermark information position according to " ballot principle " then, described " ballot principle " is as follows:

If W ' is arranged ₁(i)+W ' ₂(i)+W ' ₃(i)+...+W ' _n(i)〉n/2 can judge that then corresponding watermark information is W " (i)=1, otherwise W " (i)=0;

Step5: repeat the operation of Step4, longly be M*N information series W up to obtaining one ", identical Logistic chaos sequence carries out XOR, the reconstruct two dimensional image when embedding then; re-use scramble template identical when the embedding random processing that is inverted, namely recover watermarking images w '.

Further optimization algorithm, the telescopiny of described watermark also comprises the step that the moisture official seal breath audio frequency that obtains is estimated, this evaluation procedure adopts the difference between signal to noise ratio snr evaluation original audio signal and the moisture official seal breath sound signal, and its judgement schematics is as follows:

$\mathrm{SNR}=10{\log }_{10}\frac{{\mathrm{\δ}}^2}{D}$

Wherein, ${\mathrm{\δ}}^2=\frac{1}{N}\underset{i=0}{\overset{N-1}{\mathrm{\Σ}}}{(x_i-\frac{1}{N}\underset{i=0}{\overset{N-1}{\mathrm{\Σ}}}{x}_i)}^2,$ $D=\frac{1}{N}\underset{i=0}{\overset{N-1}{\mathrm{\Σ}}}{(x_i-{\hat{x}}_i)}^2,$ x _iBe original audio, Moisture official seal breath audio frequency.

Further optimization algorithm comprises also in the testing process of described watermark that the watermarking images to recovering carries out the normalized crosscorrelation calculation procedure, and the formula that this normalized crosscorrelation calculates is as follows:

$\mathrm{NC}(W^{\′},W)=\frac{\underset{i=0}{\overset{P-1}{\mathrm{\Σ}}}\underset{j=0}{\overset{Q-1}{\mathrm{\Σ}}}w(i,j)\·w^{\′}(i,j)}{\sqrt{\underset{i=0}{\overset{P-1}{\mathrm{\Σ}}}\underset{j=0}{\overset{Q-1}{\mathrm{\Σ}}}{w}^2(i,j)}\sqrt{\underset{i=0}{\overset{P-1}{\mathrm{\Σ}}}\underset{j=0}{\overset{Q-1}{\mathrm{\Σ}}}{w}^{\′2}(i,j)}}$

When being 1, the NC value represents that wherein both are identical, extraction effect the best; The NC value is more little, and extraction effect is just more poor.

Adopt above technical scheme, the present invention utilizes the approximate component coefficient of DWT and DCT in the telescopiny of watermark Low Medium Frequency coefficient diminishes various that to attack be to have stable performance most, it organically combines the multiresolution characteristic of wavelet transform and the cumulative characteristic of discrete cosine transform, behind DWT repeatedly, choose approximate component, carry out dct transform again, choose the Low Medium Frequency coefficient of DCT at last as the position that watermark embeds, guaranteed the robustness of embed watermark information;

The present invention takes repeatedly embed watermark information of segmentation in the telescopiny of watermark in addition, has increased the embedding quantity of information of watermark with respect to prior art, and it effectively improves the robustness of watermark.

The method that watermark at present embeds is directly to add watermark information in original vector, and it is difficult to realize the blind Detecting of watermark.Watermark embed process of the present invention is realized by quantization parameter, and quantization parameter is chosen according to the audio carrier self-adaptation, has so both guaranteed the transparency of watermark, not needing to have realized original audio to participate in simultaneously in detecting the watermark process.

Be to determine every watermark signal by " ballot principle " when the present invention detects watermark, this also greatly reduces the loss of watermark, improves robustness and the accuracy rate of watermark detection.

Description of drawings

The present invention is further detailed explanation below in conjunction with the drawings and specific embodiments:

Fig. 1 is the schematic flow sheet of the watermark embed process of the open watermarking algorithm of DWT-DCT territory audio frequency that the present invention is based on double-encryption;

Fig. 2 is the schematic flow sheet of the watermark detection process of the open watermarking algorithm of DWT-DCT territory audio frequency that the present invention is based on double-encryption;

Fig. 3 is the original audio synoptic diagram of the embodiment of the invention 1;

Fig. 4 is the moisture official seal breath audio frequency synoptic diagram that the original audio of embodiment 1 Fig. 3 obtains behind double-encryption of the present invention;

Fig. 5 is the original watermark synoptic diagram of the embodiment of the invention 1;

Fig. 6 is the watermark synoptic diagram of original watermark after once encrypting of the embodiment of the invention 1;

Fig. 7 is the watermark synoptic diagram of original watermark behind superencipher of the embodiment of the invention 1;

Fig. 8 for the original watermark of embodiment 1 after being embedded into original audio after the algorithm for encryption of the present invention, through attacking, directly to containing the watermark synoptic diagram that the watermark audio detection goes out;

Fig. 9 carries out shear treatment to containing the watermark audio frequency, more detected watermark synoptic diagram for after general's original watermark similarly to Example 1 directly embeds original audio same among the embodiment 1 in the prior art;

Figure 10 be will the original watermark identical with embodiment 1 after algorithm for encryption of the present invention, embed with embodiment 1 in behind the same original audio, also the watermark audio frequency that contains that obtains is carried out identical shear treatment, more detected watermark synoptic diagram;

Figure 11 be will the original watermark identical with embodiment 1 after algorithm for encryption of the present invention, embed with embodiment 1 in behind the same original audio, the watermark audio frequency that contains that obtains is carried out low-pass filtering treatment, more detected watermark synoptic diagram;

Figure 12 be will the original watermark identical with embodiment 1 after algorithm for encryption of the present invention, embed with embodiment 1 in behind the same original audio, the watermark audio frequency that contains that obtains is added Gauss's white noise and handles, more detected watermark synoptic diagram;

Figure 13 be will the original watermark identical with embodiment 1 after algorithm for encryption of the present invention, embed with embodiment 1 in behind the same original audio, the watermark audio frequency that contains that obtains is added down-sampling and handles, more detected watermark synoptic diagram;

Figure 14 be will the original watermark identical with embodiment 1 after algorithm for encryption of the present invention, embed with embodiment 1 in behind the same original audio, the watermark audio frequency that contains that obtains is carried out the weight processing, more detected watermark synoptic diagram;

Figure 15 be will the original watermark identical with embodiment 1 after algorithm for encryption of the present invention, embed with embodiment 1 in behind the same original audio, the watermark audio frequency that contains that obtains is carried out the Mp3 compression and handles, more detected watermark synoptic diagram.

Embodiment

Shown in one of Fig. 1 to Figure 15, the open watermarking algorithm of a kind of DWT based on double-encryption of the present invention-DCT territory audio frequency may further comprise the steps:

One, the telescopiny of watermark:

The definition original audio signal is A={a (i), 0≤i＜Length}, and wherein, Length is the number of voice data, a (i) is the range value of i voice data;

The definition binary bitmap be W={W (i, j), 0≤i＜M, 0≤j＜N}, wherein, M, N represent the row, column number of watermark, W (i, j) { 0,1} represents the capable j row of the i grey scale pixel value of binary bitmap to ∈;

Its concrete telescopiny is as follows:

1) binary bitmap W _{M * N}Carry out scramble and dimension-reduction treatment with the Arnold conversion, realize once encrypting; Carry out XOR with isometric Logistic chaos sequence then, realize superencipher, namely obtain the watermark W' behind the double-encryption _{M * N}

2) original audio A is carried out staging treating, described segmentation process is:

Choosing element formation n segment length by great-jump-forward is L * M * N * 2 coefficients series IO (i), and wherein, L is the DWT number of transitions, and n * L * M * N * 2＜Length;

3) every section coefficient is carried out L level DWT conversion, get CA ^L, CD ^L, CD ^L-1... CD ¹, wherein, CA ^LBe the approximate component of L level, CD ⁱIt is the details component;

4) to CA ^LCarry out dct transform, the Low Medium Frequency coefficient x (j) that chooses then after the conversion quantizes, and embeds the position with it as watermark and embed watermark information after encrypting, and described coefficient quantization process is as follows:

Make z (j)=[x (j)/Da]; Da is predefined quantization parameter,

If z (j) %2==W'(i), x (j)=z (j) * Da-Da/2 then;

Otherwise x (j)=z (j) * Da+Da/2;

Wherein, the computing of [] expression round, % represents complementation;

5) the coefficient x after quantizing is done idct transform and obtain CA ', and then carry out L level IDWT conversion, recombinate at last, be about to the former coefficient of correspondence of each section replacement of moisture official seal breath, just obtained moisture official seal breath audio frequency;

Two, the testing process of watermark:

Step1: sound signal to be detected is carried out staging treating, and described segmentation process constitutes n section coefficient series for choose element by great-jump-forward;

Step2: each section coefficient is carried out the L level DWT conversion identical with watermark embed process, obtain the approximate component Ca of L level ^L

Step3: to Ca ^LCarry out dct transform, the Low Medium Frequency coefficient x ' that chooses after the conversion (i) calculates and obtains a watermark reference information value, and the formula of described calculating watermark reference information value is:

Wherein,

Represent to round up computing, % represents complementation;

Step4: the watermark reference value addition on the corresponding position of n watermark coefficient series, determine a corresponding watermark information position according to " ballot principle " then, described " ballot principle " is as follows:

If W ' is arranged ₁(i)+W ' ₂(i)+W ' ₃(i)+...+W ' _n(i)〉n/2 can judge that then corresponding watermark information is W " (i)=1, otherwise W " (i)=0;

Step5: repeat the operation of Step4, longly be M*N information series W up to obtaining one ", identical Logistic chaos sequence carries out XOR, the reconstruct two dimensional image when embedding then; re-use scramble template identical when the embedding random processing that is inverted, namely recover watermarking images w '.

Further optimization algorithm, the telescopiny of described watermark also comprises the step that the moisture official seal breath audio frequency that obtains is estimated, this evaluation procedure adopts the difference between signal to noise ratio snr evaluation original audio signal and the moisture official seal breath sound signal, and its judgement schematics is as follows:

$\mathrm{SNR}=10{\log }_{10}\frac{{\mathrm{\δ}}^2}{D}$

Wherein, ${\mathrm{\δ}}^2=\frac{1}{N}\underset{i=0}{\overset{N-1}{\mathrm{\Σ}}}{(x_i-\frac{1}{N}\underset{i=0}{\overset{N-1}{\mathrm{\Σ}}}{x}_i)}^2,$ $D=\frac{1}{N}\underset{i=0}{\overset{N-1}{\mathrm{\Σ}}}{(x_i-{\hat{x}}_i)}^2,$ x _iBe original audio, Moisture official seal breath audio frequency.

Further optimization algorithm comprises also in the testing process of described watermark that the watermarking images to recovering carries out the normalized crosscorrelation calculation procedure, and the formula that this normalized crosscorrelation calculates is as follows:

$\mathrm{NC}(W^{\′},W)=\frac{\underset{i=0}{\overset{P-1}{\mathrm{\Σ}}}\underset{j=0}{\overset{Q-1}{\mathrm{\Σ}}}w(i,j)\·w^{\′}(i,j)}{\sqrt{\underset{i=0}{\overset{P-1}{\mathrm{\Σ}}}\underset{j=0}{\overset{Q-1}{\mathrm{\Σ}}}{w}^2(i,j)}\sqrt{\underset{i=0}{\overset{P-1}{\mathrm{\Σ}}}\underset{j=0}{\overset{Q-1}{\mathrm{\Σ}}}{w}^{\′2}(i,j)}}$

When being 1, the NC value represents that wherein both are identical, extraction effect the best; The NC value is more little, and extraction effect is just more poor.

Embodiment 1

In order to verify the validity of watermarking algorithm of the present invention, we carry out emulation experiment under the MATLAB7 environment.It is 44.1KHZ that the original audio signal of experiment has been selected sample frequency for use, and resolution is 16 bits, and length is 9.75 seconds monophony digital audio and video signals, and the bianry image with 32 * 32 is original watermark image; Wherein, the coefficient hop count n that chooses is 5, DWT(wavelet transform) employing Daubechie-4 is wavelet basis, DWT progression L is that 2, Arnold scramble number of times is that the λ of 9, Logistic chaos sequence is 2, X ₀Be 0.6.

Its simulation result is as follows:

One, watermarking algorithm of the present invention has improved the security of watermark.Be binary bitmap as shown in Figure 2 through the combine result of modulation of over-sampling and chaos, visually can't see any Useful Information, image information becomes disorderly and unsystematic, therefore, even it is watermarking algorithm is open, allow have a mind to or unintentionally the assailant detect the watermark information that contains, also can't obtain real watermarking images.

Two, the present invention is highly resistant to the sampling point cutting, guarantees the vision integrality of watermark.With the continuous audio frequency to be measured of 15000 sampled points of deletion of the watermark information after the present invention encrypts, adopt algorithm of the present invention that it is detected watermark to it then, its testing result and records NC=0.81704 as shown in figure 10.To be audio frequency to be measured 15000 sampled points of deletion continuously of the watermark information of not encrypting through the present invention that embeds equally when embed watermark, extraction result as Fig. 9, and record NC=0.63768, as seen adopt the encryption of algorithm of the present invention to strengthen the robustness of watermark.This is owing to used Arnold scramble and chaos method with watermark pretreatment in the algorithm, and the location of pixels of watermark and pixel value are changed, and when being subjected to local sampling point cutting and attacking, the impaired number of the information of the relevant position in the watermark just significantly reduces.

Be the original audio synoptic diagram as Fig. 3, as the moisture official seal breath audio frequency synoptic diagram of Fig. 4 for behind the double-encryption of algorithm of the present invention, obtaining among Fig. 3.Its as can be seen the waveform of signal after using this algorithm embed watermark, significantly do not change, and in also not significantly difference of audio frequency watermark and original audio acoustically, be difficult to feel the existence of watermark, it is transparent preferably that this illustrates that this watermark has.And audio frequency to be measured can accurately extract watermark under nothing is attacked, and it extracts result such as Fig. 8, and records NC=1.

Handle carry out following common signal through the sound signal of watermarking algorithm of the present invention, wherein

Carry out the design sketch of the watermark that extracts again after the low-pass filtering treatment as Figure 11, this low-pass filter is that length is that 8 rank cut-off frequency rates are the Butterworth low-pass filter of 2k Hz.

As the design sketch that adds the watermark that extracts again after Gauss's white noise is handled of Figure 12, the average of this white Gaussian noise is 0, and mean square deviation is 0.01.

The design sketch of the watermark that extracts again behind the down-sampling as Figure 13: original audio is resampled with 22.050kHz, use the 44.100kHz resampling again, recover audio frequency.

Design sketch as the watermark that extracts again after the weightization of Figure 14: original audio is quantified as 8bit from 16bit, is quantified as 16bit again.

The design sketch of the watermark that extracts again after the Mp3 compression as Figure 15: the Mp3 file with original audio boil down to position speed 56kbps (ratio of compression 12:1) reverts to the audio frequency identical with original audio frequency format then.

In sum, watermarking algorithm disclosed by the invention except having preferably the transparency, also has following three characteristics: one, security, adopt the method for image scrambling and chaos encryption that watermarking images is carried out double-encryption, guaranteed the security of watermark well; Two, robustness, taking full advantage of the multiresolution characteristic of DWT and the cumulative characteristic of DCT handles the audio frequency coefficient series of choosing, select DCT Low Medium Frequency coefficient to quantize embed watermark, guaranteed the robustness of watermark, three, publicity, watermark adopts coefficient quantization to embed, and the ballot principle is adopted in multiple embedding during extraction, do not need original audio to participate in, the real blind Detecting that realizes has good practical values.

Claims (3)

1. based on the open watermarking algorithm of the DWT-DCT territory audio frequency of double-encryption, it is characterized in that: it may further comprise the steps:

One, the telescopiny of watermark:

The definition original audio signal is A={a (i), 0≤i＜Length}, and wherein, Length is the number of voice data, a (i) is the range value of i voice data;

The definition binary bitmap be W={W (i, j), 0≤i＜M, 0≤j＜N}, wherein, M, N represent the row, column number of watermark, W (i, j) { 0,1} represents the capable j row of the i grey scale pixel value of binary bitmap to ∈;

Its concrete telescopiny is as follows:

1) binary bitmap W _{M * N}Carry out scramble and dimension-reduction treatment with the Arnold conversion, realize once encrypting; Carry out XOR with isometric Logistic chaos sequence then, realize superencipher, namely obtain the watermark W' behind the double-encryption _{M * N}

2) original audio A is carried out staging treating, described segmentation process is:

Choosing element formation n segment length by great-jump-forward is L * M * N * 2 coefficients series IO (i), and wherein, L is the DWT number of transitions, and n * L * M * N * 2＜Length;

3) every section coefficient is carried out L level DWT conversion, get CA ^L, CD ^L, CD ^L-1... CD ¹, wherein, CA ^LBe the approximate component of L level, CD ⁱIt is the details component;

4) to CA ^LCarry out dct transform, the Low Medium Frequency coefficient x (j) that chooses then after the conversion quantizes, and embeds the position with it as watermark and embed watermark information after encrypting, and described coefficient quantization process is as follows:

Make z (j)=[x (j)/Da]; Da is predefined quantization parameter,

If z (j) %2==W'(i), x (j)=z (j) * Da-Da/2 then;

Otherwise x (j)=z (j) * Da+Da/2;

Wherein, the computing of [] expression round, % represents complementation;

5) the coefficient x after quantizing is done idct transform and obtain CA ', and then carry out L level IDWT conversion, at last each section of moisture official seal breath replaced former coefficient of correspondence, namely obtain moisture official seal breath audio frequency;

Two, the testing process of watermark:

Step1: sound signal to be detected is carried out staging treating, and described segmentation process constitutes n section coefficient series for choose element by great-jump-forward;

Step2: each section coefficient is carried out the L level DWT conversion identical with watermark embed process, obtain the approximate component Ca of L level ^L

Step3: to Ca ^LCarry out dct transform, the Low Medium Frequency coefficient x ' that chooses after the conversion (i) calculates and obtains a watermark reference information value, and the formula of described calculating watermark reference information value is:

Wherein,

Represent to round up computing, % represents complementation;

Step4: the watermark reference value addition on the corresponding position of n watermark coefficient series, determine a corresponding watermark information position according to " ballot principle " then, described " ballot principle " is as follows:

If W ' is arranged ₁(i)+W ' ₂(i)+W ' ₃(i)+...+W ' _n(i)〉n/2 can judge that then corresponding watermark information is W " (i)=1, otherwise W " (i)=0;

Step5: repeat the operation of Step4, longly be M*N information series W up to obtaining one ", identical Logistic chaos sequence carries out XOR, the reconstruct two dimensional image when embedding then; re-use scramble template identical when the embedding random processing that is inverted, namely recover watermarking images w '.

2. the DWT based on double-encryption according to claim 1-DCT territory audio frequency discloses watermarking algorithm, it is characterized in that: the telescopiny of described watermark also comprises the step that the moisture official seal breath audio frequency that obtains is estimated, this evaluation procedure adopts the difference between signal to noise ratio snr evaluation original audio signal and the moisture official seal breath sound signal, and its judgement schematics is as follows:

$\mathrm{SNR}=10{\log }_{10}\frac{{\mathrm{\δ}}^2}{D}$

Wherein, ${\mathrm{\δ}}^2=\frac{1}{N}\underset{i=0}{\overset{N-1}{\mathrm{\Σ}}}{(x_i-\frac{1}{N}\underset{i=0}{\overset{N-1}{\mathrm{\Σ}}}{x}_i)}^2,$ $D=\frac{1}{N}\underset{i=0}{\overset{N-1}{\mathrm{\Σ}}}{(x_i-{\hat{x}}_i)}^2,$ x _iBe original audio,

Moisture official seal breath audio frequency.

3. the DWT based on double-encryption according to claim 1-DCT territory audio frequency discloses watermarking algorithm, it is characterized in that: comprise also in the testing process of described watermark that the watermarking images to recovering carries out the normalized crosscorrelation calculation procedure, the formula that this normalized crosscorrelation calculates is as follows:

$\mathrm{NC}(W^{\′},W)=\frac{\underset{i=0}{\overset{P-1}{\mathrm{\Σ}}}\underset{j=0}{\overset{Q-1}{\mathrm{\Σ}}}w(i,j)\·w^{\′}(i,j)}{\sqrt{\underset{i=0}{\overset{P-1}{\mathrm{\Σ}}}\underset{j=0}{\overset{Q-1}{\mathrm{\Σ}}}{w}^2(i,j)}\sqrt{\underset{i=0}{\overset{P-1}{\mathrm{\Σ}}}\underset{j=0}{\overset{Q-1}{\mathrm{\Σ}}}{w}^{\′2}(i,j)}}$

When being 1, the NC value represents that wherein both are identical, extraction effect the best; The NC value is more little, and extraction effect is just more poor.

Images