CN108010532B - Digital watermark detection method based on multivariate generalized Gaussian distribution - Google Patents
Digital watermark detection method based on multivariate generalized Gaussian distribution Download PDFInfo
- Publication number
- CN108010532B CN108010532B CN201711363551.4A CN201711363551A CN108010532B CN 108010532 B CN108010532 B CN 108010532B CN 201711363551 A CN201711363551 A CN 201711363551A CN 108010532 B CN108010532 B CN 108010532B
- Authority
- CN
- China
- Prior art keywords
- watermark
- coefficient
- swt
- embedding
- carrying
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Active
Links
Images
Classifications
-
- G—PHYSICS
- G10—MUSICAL INSTRUMENTS; ACOUSTICS
- G10L—SPEECH ANALYSIS OR SYNTHESIS; SPEECH RECOGNITION; SPEECH OR VOICE PROCESSING; SPEECH OR AUDIO CODING OR DECODING
- G10L19/00—Speech or audio signals analysis-synthesis techniques for redundancy reduction, e.g. in vocoders; Coding or decoding of speech or audio signals, using source filter models or psychoacoustic analysis
- G10L19/018—Audio watermarking, i.e. embedding inaudible data in the audio signal
Landscapes
- Engineering & Computer Science (AREA)
- Computational Linguistics (AREA)
- Signal Processing (AREA)
- Health & Medical Sciences (AREA)
- Audiology, Speech & Language Pathology (AREA)
- Human Computer Interaction (AREA)
- Physics & Mathematics (AREA)
- Acoustics & Sound (AREA)
- Multimedia (AREA)
- Editing Of Facsimile Originals (AREA)
Abstract
The invention discloses a digital audio watermark detection method based on multivariate generalized Gaussian distribution, which comprises the steps of firstly carrying out segmentation processing on original audio, constructing an SWT (single-tone transform) important coefficient section with higher local energy according to a short-time energy calculation method, modifying the coefficient of the SWT important coefficient section in a multiplicative embedding mode, and carrying out inverse SWT (single-tone transform) conversion on a sub-band of the modified coefficient and other sub-bands to obtain digital audio containing watermarks; then, carrying out statistical modeling on the SWT coefficient by using the correlation existing between the SWT high-frequency sub-bands and using multivariate generalized Gaussian distribution, and carrying out parameter estimation by combining an MLE (maximum likelihood estimation) method and a Newton-Raphson iteration method; and finally, constructing a maximum likelihood detector by using a log-likelihood ratio method to extract watermark information.
Description
Technical Field
The invention belongs to the technical field of copyright protection of digital audio, relates to a digital audio watermarking method based on a statistical model, and particularly relates to a digital watermark detection method based on multivariate generalized Gaussian distribution.
Background
With the rapid development of multimedia technology and the popularization of internet technology, the way for people to acquire various digital multimedia resources becomes more and more convenient, so that the behaviors of illegally copying, tampering and spreading the multimedia resources become more and more rampant, and therefore, how to safely and effectively store and transmit multimedia information becomes the focus of attention of people. The digital audio watermarking technology is produced as an important branch in the field of information security, provides an effective means for solving the problem of audio information intellectual property protection, and has great research and application values in the fields of digital audio content authentication and copyright protection.
The digital audio watermarking technology is characterized in that watermarking information is embedded into audio by using a data embedding strategy, so that the watermarking information is hidden in original audio data and cannot be perceived, the integrity of the audio is kept, legal protection for an audio property owner is realized, and the ownership of an audio work can be proved by using the watermarking technology even if the digital audio is attacked or decrypted. In recent years, the digital audio watermarking technology has been developed dramatically, and many excellent audio watermarking algorithms appear successively, which have great significance in both theoretical research and practical application, but still have many shortages to be solved, wherein the balance problem between robustness and imperceptibility is still a difficult problem in the field.
The transform domain audio watermarking method based on the statistical model can simultaneously optimize the imperceptibility and the robustness, the watermark can be embedded into the transform domain coefficient to improve the robustness of the watermark, and the multiplicative embedding algorithm can enable the embedding strength of the watermark to change in proportion with the strength of a carrier signal, so that the balance problem between the robustness and the imperceptibility can be effectively solved by combining the multiplicative embedding method in the transform domain, and the transform domain audio watermarking method is widely concerned by researchers. Watermark detection methods based on transform domain statistical modeling are proposed in succession, and mainly comprise two types: one method is to detect whether the watermark exists only at the detection end, and the other method is to extract specific watermark information at the detection end, so that the second method has practical value obviously. However, the statistical model-based watermarking method still has the following disadvantages: firstly, the distributions selected by the current statistical modeling are unit distributions, and only intra-scale correlation is utilized without fully considering inter-scale correlation; secondly, when the transform domain coefficient is modeled, the selected distribution model is not deeply analyzed, and whether the established model is optimal to the selected transform domain is not fully proved; third, most digital watermarking methods based on statistical modeling are image watermarking directions, and there is not much attention and research in audio watermarking.
Disclosure of Invention
The invention provides a digital watermark detection method based on multivariate generalized Gaussian distribution, aiming at solving the technical problems in the prior art.
The technical solution of the invention is as follows: a digital watermark detection method based on multivariate generalized Gaussian distribution comprises watermark embedding and watermark extraction, and is characterized in that:
appointing:representing a host audio signal;which represents the length of the watermark or watermarks,;representing the length of each segment asThe audio frequency of (a) is selected,represents the high-frequency sub-band of the second scale,representing a third scale high frequency sub-band;represents the window length;represents the original position of the significant SWT coefficients;representsA segment of the importance coefficient of the image,representsA significant coefficient segment;represents a scattering matrix;representing a shape parameter;representing a scale parameter;a representative dimension;
the watermark embedding is carried out according to the following steps:
a. initial setting
b. watermark embedding
b.2 pairsCarrying out three-stage SWT conversion to obtainAndcalculatingThe short-time energy of each SWT coefficient is arranged in descending order according to the size:
b.3 filtering out coefficients near the end of the audio and coefficients that are too close to each other, before selectionAn important SWT coefficient is used as a watermark embedding initial position, and after recording and selectionA coefficient ofWill beThe location of the embedded watermark is also used asThe watermark embedding location of (a);
b.4, two multiplicative embedding strength functions are constructed for embedding watermark bits of 1 or 0:
b.6 willAndcombining other sub-bands to perform inverse SWT to obtain digital audio containing watermarks;
the watermark extraction is carried out according to the following steps:
c. coefficient multivariate generalized Gaussian distribution statistical modeling
c.1, determining an important SWT coefficient according to the record, and determining a SWT coefficient section of the watermark to be extracted;
c.4 respectively calculating the probability of the multivariate generalized Gaussian distribution under the two assumptions of embedding watermark '1' and embedding watermark '0' in each coefficient sectionAnd:
wherein the content of the first and second substances,,,to representThe coefficients of the noise are not included in the noise,to representA noise-free coefficient;
d. constructing a maximum likelihood detector for watermark extraction
d.1 pair of watermark-containing audiosPerforming segmentation processing onCarrying out three-stage SWT conversion to obtainAnd;
d.3 utilization ofAndas a parameter estimation sample, firstly aligning according to the maximum likelihood methodEstimating, and combining with Newton-Raphson methodEstimating, and obtaining the final result according to the estimationAndto pairEstimating:
d.4, constructing a maximum likelihood detector, and extracting specific watermark information of each watermark coefficient section to be extracted:
d.5, obtaining the optimal watermark sequence by using a voting principle.
Firstly, carrying out segmentation processing on original audio, constructing an SWT (single-tone transform) important coefficient section with higher local energy according to a short-time energy calculation method, modifying the coefficient of the SWT important coefficient section in a multiplicative embedding mode, and carrying out inverse SWT (single-tone transform) conversion on a sub-band of the modified coefficient and other sub-bands to obtain digital audio containing watermarks; then, carrying out statistical modeling on the SWT coefficient by using the correlation existing between the SWT high-frequency sub-bands and using multivariate generalized Gaussian distribution, and carrying out parameter estimation by combining an MLE (maximum likelihood estimation) method and a Newton-Raphson iteration method; and finally, constructing a maximum likelihood detector by using a log-likelihood ratio method to extract watermark information. The experimental result shows that the method of the invention utilizes the multivariate generalized Gaussian distribution to construct a more accurate model, thereby effectively improving the detection precision and simultaneously keeping the good balance between the imperceptibility and the robustness.
Compared with the prior art, the invention has the following beneficial effects:
firstly, modeling is carried out on the SWT coefficient by adopting multivariate generalized Gaussian distribution, so that the parent-child correlation between scales and the brother correlation in the scales are fully utilized, and the distribution characteristics of the coefficient can be more accurately fitted;
secondly, parameter estimation is carried out by combining an MLE method and a Newton-Raphson iteration method, and the detection precision of the model is further improved.
Drawings
Wav audio, fig. 1 is a waveform diagram of original audio, watermark-containing audio, and difference audio in wedding.
Fig. 2 is a diagram of a robustness test result of a conventional attack according to an embodiment of the present invention.
Fig. 3 is a diagram of a robustness test result of desynchronization attack according to an embodiment of the present invention.
Fig. 4 is a flowchart of watermark embedding according to an embodiment of the present invention.
Fig. 5 is a flowchart of watermark extraction according to an embodiment of the present invention.
Detailed Description
The method mainly comprises three stages: watermark embedding, coefficient multivariate generalized Gaussian distribution statistical modeling and constructing a maximum likelihood detector for watermark extraction.
Appointing:representing a host audio signal;which represents the length of the watermark or watermarks,;representing the length of each segment asThe audio frequency of (a) is selected,represents the high-frequency sub-band of the second scale,representing a third scale high frequency sub-band;represents the window length;represents the original position of the significant SWT coefficients;representsA segment of the importance coefficient of the image,representsA significant coefficient segment;represents a scattering matrix;representing a shape parameter;representing a scale parameter;a representative dimension;
watermark embedding is shown in fig. 4 and is performed as follows:
a. initial setting
b. watermark embedding
b.2 pairsCarrying out three-stage SWT conversion to obtainAndcalculatingThe short-time energy of each SWT coefficient is arranged in descending order according to the size:
b.3 filtering out coefficients near the end of the audio and coefficients that are too close to each other, before selectionAn important SWT coefficient is used as a watermark embedding initial position, and after recording and selectionA coefficient ofWill beThe location of the embedded watermark is also used asThe watermark embedding location of (a);
b.4, two multiplicative embedding strength functions are constructed for embedding watermark bits of 1 or 0:
b.6 willAndcombining other sub-bands to perform inverse SWT to obtain digital audio containing watermarks;
watermark extraction is performed as shown in fig. 5, according to the following steps:
c. coefficient multivariate generalized Gaussian distribution statistical modeling
c.1, determining an important SWT coefficient according to the record, and determining a SWT coefficient section of the watermark to be extracted;
c.4 respectively calculating the probability of the multivariate generalized Gaussian distribution under the two assumptions of embedding watermark '1' and embedding watermark '0' in each coefficient sectionAnd:
wherein the content of the first and second substances,,,to representThe coefficients of the noise are not included in the noise,to representA noise-free coefficient;
d. constructing a maximum likelihood detector for watermark extraction
d.1 pair of watermark-containing audiosPerforming segmentation processing onCarrying out three-stage SWT conversion to obtainAnd;
d.3 utilization ofAndas a parameter estimation sample, firstly aligning according to the maximum likelihood methodEstimating, and combining with Newton-Raphson methodEstimating, and obtaining the final result according to the estimationAndto pairEstimating:
d.4, constructing a maximum likelihood detector, and extracting specific watermark information of each watermark coefficient section to be extracted:
d.5, obtaining the optimal watermark sequence by using a voting principle.
Experimental testing and parameter setting:
the experiment was performed on a MATLAB R2011a platform, the audio carrier being a mono digital audio signal with a sampling frequency of 41kHz, a resolution of 16 bits, and a length of 20 seconds.
Wav audio, fig. 1 is a waveform diagram of original audio, watermark-containing audio, and difference audio in wedding.
Fig. 2 is a diagram of a robustness test result of a conventional attack according to an embodiment of the present invention.
Fig. 3 is a diagram of a robustness test result of desynchronization attack according to an embodiment of the present invention.
The experimental result shows that the method of the invention utilizes the multivariate generalized Gaussian distribution to construct a more accurate model, thereby effectively improving the detection precision and simultaneously keeping the good balance between the imperceptibility and the robustness.
Claims (1)
1. A digital watermark detection method based on multivariate generalized Gaussian distribution comprises watermark embedding and watermark extraction, and is characterized in that:
appointing: a represents a host audio signal; i is1*J1Represents the watermark length, k 1,21*J1(ii) a B represents audio of length L per segment, B2Representing the second-scale high-frequency sub-band, B3Representing a third scale high frequency sub-band; d represents the window length; p (i) represents the original position of the significant SWT coefficients; a. thek={x1,x2,...,xDRepresents B2Section of importance, Bk={x1,x2,...,xDRepresents B3A significant coefficient segment; m represents a scattering matrix; beta represents a shape parameter; m represents a scale parameter; p represents a dimension;
the watermark embedding is carried out according to the following steps:
a. initial setting
Acquiring a host audio signal A and initializing the host audio signal A;
b. watermark embedding
b.1, carrying out segmentation treatment on the A, wherein the length of each segment is as follows: l ═ I1*J1*100;
b.2 carrying out three-level SWT conversion on B to obtain B2And B3Calculating B2The short-time energy of each SWT coefficient is arranged in descending order according to the size:
b.3 filtering out coefficients near the end of the audio and adjacent coefficients too close, selecting the previous I1*J1An important SWT coefficient is used as a watermark embedding initial position, and D-1 coefficients after the recording and selection are AkA 1 to B2The location of the embedded watermark is also taken as B3The watermark embedding location of (a);
b.4, two multiplicative embedding strength functions are constructed for embedding watermark bits of 1 or 0:
f1(x)=arctan(x/10+4.5),f0(x)=arccot(x/30+2.5);
b.5 modification AkAnd BkThe middle corresponding coefficient obtains A'kAnd B'k:
b.6 mixing A'kAnd B'kCombining other sub-bands to perform inverse SWT to obtain digital audio containing watermarks;
the watermark extraction is carried out according to the following steps:
c. coefficient multivariate generalized Gaussian distribution statistical modeling
c.1, determining an important SWT coefficient according to the record, and determining a SWT coefficient section of the watermark to be extracted;
c.3 obtaining f by numerical method1、f0Inverse function g of1、g0;
c.4 respectively calculating the multivariate generalized Gaussian distribution probability H under the two assumptions of embedding watermark '1' and embedding watermark '0' in each coefficient section1And H0:
Wherein Q ═ g1(y1i),g1(y2i)],R=[g0(y1i),g0(y2i)],y1iIs represented by B'2Coefficient without noise, y2iIs represented by B'3A noise-free coefficient;
d. constructing a maximum likelihood detector for watermark extraction
d.1, carrying out segmentation processing on the watermark-containing audio A ', and carrying out three-stage SWT conversion on the B ' to obtain B '2And B'3;
d.2, carrying out initialization processing on M and beta:
M(i,j)=ρ|i-j| i,j∈[0,p-1];
d.3 utilize B'2And B'3As a parameter estimation sample, estimating M according to a maximum likelihood method, estimating beta by combining a Newton-Raphson method, and estimating M according to the estimated M and beta:
d.4, constructing a maximum likelihood detector, and extracting specific watermark information of each watermark coefficient section to be extracted:
d.5, obtaining the optimal watermark sequence by using a voting principle.
Priority Applications (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201711363551.4A CN108010532B (en) | 2017-12-18 | 2017-12-18 | Digital watermark detection method based on multivariate generalized Gaussian distribution |
Applications Claiming Priority (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201711363551.4A CN108010532B (en) | 2017-12-18 | 2017-12-18 | Digital watermark detection method based on multivariate generalized Gaussian distribution |
Publications (2)
Publication Number | Publication Date |
---|---|
CN108010532A CN108010532A (en) | 2018-05-08 |
CN108010532B true CN108010532B (en) | 2021-09-24 |
Family
ID=62059793
Family Applications (1)
Application Number | Title | Priority Date | Filing Date |
---|---|---|---|
CN201711363551.4A Active CN108010532B (en) | 2017-12-18 | 2017-12-18 | Digital watermark detection method based on multivariate generalized Gaussian distribution |
Country Status (1)
Country | Link |
---|---|
CN (1) | CN108010532B (en) |
Citations (4)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
WO2005006326A2 (en) * | 2003-07-11 | 2005-01-20 | Koninklijke Philips Electronics N.V. | Watermark embedding and detection |
CN101039371A (en) * | 2006-03-18 | 2007-09-19 | 辽宁师范大学 | Novel method of digital watermarking for protecting literary property of presswork |
CN102074237A (en) * | 2010-11-30 | 2011-05-25 | 辽宁师范大学 | Digital audio watermarking method based on invariant characteristic of histogram |
CN102147912A (en) * | 2011-03-30 | 2011-08-10 | 北京航空航天大学 | Adaptive difference expansion-based reversible image watermarking method |
-
2017
- 2017-12-18 CN CN201711363551.4A patent/CN108010532B/en active Active
Patent Citations (4)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
WO2005006326A2 (en) * | 2003-07-11 | 2005-01-20 | Koninklijke Philips Electronics N.V. | Watermark embedding and detection |
CN101039371A (en) * | 2006-03-18 | 2007-09-19 | 辽宁师范大学 | Novel method of digital watermarking for protecting literary property of presswork |
CN102074237A (en) * | 2010-11-30 | 2011-05-25 | 辽宁师范大学 | Digital audio watermarking method based on invariant characteristic of histogram |
CN102147912A (en) * | 2011-03-30 | 2011-08-10 | 北京航空航天大学 | Adaptive difference expansion-based reversible image watermarking method |
Also Published As
Publication number | Publication date |
---|---|
CN108010532A (en) | 2018-05-08 |
Similar Documents
Publication | Publication Date | Title |
---|---|---|
Fan et al. | Chaos-based discrete fractional Sine transform domain audio watermarking scheme | |
Mohanty et al. | Invisible watermarking based on creation and robust insertion-extraction of image adaptive watermarks | |
CN107945097B (en) | Lu-lolly image watermarking method based on joint statistical model correction | |
Su et al. | SNR-constrained heuristics for optimizing the scaling parameter of robust audio watermarking | |
CN106780281B (en) | Digital image watermarking method based on Cauchy's statistical modeling | |
CN106898358B (en) | Robust digital audio watermarking algorithm from time-frequency analysis angle | |
CN108596823A (en) | A kind of insertion of the digital blind watermark based on sparse transformation and extracting method | |
Zhao et al. | Tampered region detection of inpainting JPEG images | |
Wang et al. | Statistical image watermarking using local RHFMs magnitudes and beta exponential distribution | |
Jana et al. | A new DCT based robust image watermarking scheme using cellular automata | |
CN102867513B (en) | Pseudo-Zernike moment based voice content authentication method | |
CN108053360B (en) | Digital image watermark detection method based on multi-correlation HMT model | |
CN104766263A (en) | Color image watermark embedding and detecting method based on quaternion Legendre moment correction | |
CN108010532B (en) | Digital watermark detection method based on multivariate generalized Gaussian distribution | |
CN107910010B (en) | Digital watermark detection method based on multi-parameter Weibull statistical modeling | |
CN108877819B (en) | Voice content evidence obtaining method based on coefficient autocorrelation | |
Wang et al. | A new statistical image watermark detector in RHFMs domain using beta-exponential distribution | |
Yue et al. | Rights protection for trajectory streams | |
Kaur et al. | A secure and high payload digital audio watermarking using features from iris image | |
CN108062957B (en) | Strong robust digital watermark detection method based on steady local features | |
Lin et al. | DWT based on watermarking algorithm and its implementing with DSP | |
Chen et al. | Multipurpose audio watermarking algorithm | |
CN110047495A (en) | Large capacity Audio Watermarking Algorithm based on 2 grades of singular value decompositions | |
Raval et al. | Secure and robust watermarking technique | |
Wang et al. | Blind image watermark decoder in NSST-FPCET domain using Weibull Mixtures-HMT |
Legal Events
Date | Code | Title | Description |
---|---|---|---|
PB01 | Publication | ||
PB01 | Publication | ||
SE01 | Entry into force of request for substantive examination | ||
SE01 | Entry into force of request for substantive examination | ||
GR01 | Patent grant | ||
GR01 | Patent grant |