CN101447070B - Digital watermarking protection method of two-dimensional vector graph based on canonical correlation analysis - Google Patents
Digital watermarking protection method of two-dimensional vector graph based on canonical correlation analysis Download PDFInfo
- Publication number
- CN101447070B CN101447070B CN2008102039975A CN200810203997A CN101447070B CN 101447070 B CN101447070 B CN 101447070B CN 2008102039975 A CN2008102039975 A CN 2008102039975A CN 200810203997 A CN200810203997 A CN 200810203997A CN 101447070 B CN101447070 B CN 101447070B
- Authority
- CN
- China
- Prior art keywords
- canonical correlation
- watermarking
- digital watermarking
- watermark
- correlation analysis
- 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.)
- Expired - Fee Related
Links
Images
Abstract
The invention relates to a digital watermarking protection method of a two-dimensional vector graph based on canonical correlation analysis. The digital watermarking protection method is based on the analysis of the shortcomings of the existing vector graph watermarking algorithm on the anti-attack aspect and introduces the canonical correlation analysis into the digital watermarking technology according to the characteristics of the canonical correlation analysis. The digital watermarking protection method comprises two implementation steps: the proposal of utilizing the DFT domain algorithmfor embedding a watermarking on a canonical correlation variable obtained through the canonical correlation analysis is adopted during the embedding stage of the watermarking; when at the stage of detection, the correlation detection is carried out on the watermarking according to the affine invariant feature and the statistical feature of the canonical correlation variable obtained by the canonical correlation analysis, and the determination of whether the watermarking is contained or not is finally carried out. The method can resist various geometric attacks, noise-adding filtration, and starting point changing and the compressive attack thereof.
Description
Technical field
The present invention relates to a kind of method that CCA (canonical correlation analysis) is applied to the digital watermarking protection of 2 dimensional vectors, belong to the information safety protection field.
Background technology
Issue and the propagation of polar plot on network that be applied as of digital watermark technology provides safer advantage.At present, the digital watermarking of 2 dimensional vectors mainly is to come embed watermark by the method for revising the vector graphics profile.Concrete watermarking algorithm but respectively has quality.According to the difference of watermark embedded location, mainly be divided into spatial domain and transform domain water mark method.
1. spatial domain method: being that coordinate figure by direct modification profile summit comes embed watermark information, mainly is effectively to utilize the position relation on summit or the statistical property of coordinate etc.The representative watermarking project of revising the vertex position relation that is based on, in these schemes, polar plot is divided into the piece of specifying size, in these pieces, is initial point a bit with certain of piece, forms new coordinate figure in piece, is embedded in watermark bit at each piece.
2. transform domain method: be not to embed, but from polar plot, choose the coordinate on summit, and sort, carry out the territory conversion then, watermark is embedded in their conversion coefficient by certain rule by direct modification apex coordinate.Generally speaking, the transform-domain algorithm robustness is better, is the emphasis of studying at present, and main conversion has DFT (Fourier transform), DWT (wavelet transformation) and DCT (discrete cosine transform).
3. other transform-domain algorithm: profile parametersization, ICA (independent component analysis) handles, grid-spectrogram and Chinese remainder theorem etc.
The scheme in above-mentioned three fields respectively has quality.The morning that the method for spatial domain occurs, can only resist some slight attacks, algorithm does not have other media work spatial domain algorithm characteristic of simple yet.The transform domain method has good robustness relatively, but the algorithm that has implements more complicated, and the affine attacking ability of anti-polar plot is poor.As for the method for other transform domains, they belong to non-mainstream scheme, and the degree of depth of research is more shallow, has significant limitation.
Summary of the invention
The objective of the invention is to defective, a kind of digital watermarking protection method of 2 dimensional vectors based on canonical correlation analysis is provided, can resist various geometric attacks, add the filtering of making an uproar, change starting point and their comprehensive attack at prior art.
Main design of the present invention is: in the deficiency of present polar plot watermarking algorithm in anti-attack, according to the characteristics of canonical correlation analysis self, be introduced in the digital watermark technology, provided a kind of very effective polar plot digital watermarking protection method.
This method is to seek a kind of algorithm that can farthest resist various watermarks attacks, comprising:
(1) resist geometric attacks or claim affine attack
(2) anti-adding, make an uproar and filtering
(3) the anti-starting point that changes
(4) anti-summit backward
(5) anti-comprehensive the attack
For geometric attack, polar plot carries out obtaining one group of identical data or descriptor after this conversion after considering to find a kind of conversion can make former polar plot and geometric attack, we have introduced CCA for this reason, because find that by our analysis CCA has following affine property:
Two stochastic variables of CCA are restricted to 2 dimensions, two coordinates on just in time corresponding profile diagram summit then are reflected on the figure and can be expressed as at the corresponding relation before and after the affined transformation through the canonical correlation variable that CCA obtains: 1) overlap 2) through X-axis symmetry 3) through Y-axis symmetry 4) obtain through X-axis and twice symmetry of Y-axis.
Briefly, 2 n-dimensional random variable ns are after carrying out identical affined transformation 1) canonical correlation coefficient is constant, and 2) the canonical correlation variable exists above-mentioned four kinds of possible relations, 3) if regulation canonical correlation variable is arithmetic number, then affine front and back canonical correlation variable is constant.
Because CCA is the statistical property of utilizing data, for big distortion and insensitive for noise, so make an uproar and filtering has good robustness for adding based on the watermarking algorithm of CCA.
When starting point that changes the summit or summit backward, the starting point of the canonical correlation variable that CCA obtains also will change, this has just destroyed the synchronous of watermark, so, also need to find a kind of to synchronous insensitive algorithm, the watermarking algorithm in DFT territory can be resisted this attack just, so we consider to use this algorithm on the canonical correlation variable that obtains, so just can be good at dealing with problems, but at this moment also to consider the parity that starting point changes, because the influence that is subjected to of canonical correlation variable at this moment is different, when watermark detection, need calculate two kinds of correlations under the situation for this reason.
Top various attack their influence for the canonical correlation variable is separate, and the just simple stack of their comprehensive attack does not have big influence.
But, use for CCA still has problem, because CCA require two immediately the sample number of variable be identical, this just needs us that some restrictive conditions are arranged when choosing the summit, if pressing the target odd even with the summit, we are divided into two groups of sequences as carrying out the stochastic variable that CCA handles, our the profile vertex sequence length chosen just must be even number so, and this is that the vertex sequence of odd number is a fatal restrictive condition for length.So, at length be odd number vertex sequence we use a kind of new group technology---annular parity packet method.So-called annular method of grouping is exactly that summit with whole sequence is by ring shooting, begin to carry out parity packet from any by a certain direction (counterclockwise or clockwise), till dividing into groups all to begin to occur separately circulation to the odd even group, even former sequence length is 2n+1, and then the odd even group length after the grouping is 2n+1.After adopting this method grouping, the summit that taking out us when follow-up canonical variable merges needs gets final product.It should be noted that and when occurring changing starting point and summit backward, when detecting, will calculate one group of correlation more.
According to above-mentioned inventive concept, the present invention adopts following technical proposals:
A kind of digital watermarking protection method of 2 dimensional vectors based on canonical correlation analysis is characterized in that implementation step is: the embedding flow process of 1) implementing digital watermarking; 2) testing process of enforcement digital watermarking.
The operation steps of the embedding flow process of digital watermarking is:
(1) the profile diagram apex coordinate sequence P barycenter of original polar plot is made zero in elder generation, is divided into two groups of P according to following target parity then
oAnd P
e, i.e. odd number group and even number set; Being vertex sequence length is the group technology of even number above, is odd number for vertex sequence length, needs to use above-mentioned annular parity packet method, obtains odd number group P equally
oWith even number set P
e
(2) with P
oAnd P
eCarry out CCA, obtain P
oAnd P
eEach self-corresponding canonical correlation variable U and V get positive back by P with U and V then
oAnd P
eOdd even originally is combined into one group of sequence VP in proper order; Being odd number (if 2n+1) for vertex sequence length gets when just being combined into one group of sequence VP at U and V, does not need U and all data of V are merged, and it is just passable that the preceding n item that only needs to take out the preceding n+1 of U and V is merged into VP.
(3) utilization DFT territory watermarking algorithm is to the VP embed watermark, and the VP that contains watermark represents with WVP;
(4) WVP elder generation parity packet and revert back original positive negativity is carried out the inverse operation of CCA then, obtains to contain the sequence of parity WP of watermark
oAnd WP
eFor vertex sequence length is the situation of odd number (if 2n+1), with WVP elder generation parity packet and reduce after the positive negativity, and be the inverse operation of carrying out CCA behind the data polishing of giving up in (2) again.If be noted that the also data behind the embed watermark of these data that refill, can not be directly with the data of giving up originally.
(5) contain the WP of watermark
oAnd WP
eBe combined into unique sequence by odd even, original barycenter is returned in translation then.Promptly revert to the vector polygon sequence WP that contains watermark
oWhen vertex sequence length is odd number (if 2n+1), still to take out earlier separately preceding n+1 or the n item number according to after remerge to behind the unique sequence, finish the work of back again.What the testing process of digital watermarking adopted is the method for coherent detection, and concrete operation steps is:
(1) after the polar plot profile vertex sequence Q barycenter to be detected normalization, is divided into two groups of Q of odd even by odd even
oAnd Q
eProcessing mode when vertex sequence length is odd number here is identical with the step (1) of embedded system.
(2) to Q
oAnd Q
eCarry out CCA, obtain canonical correlation variable M and N,, be combined into unique sequence VQ after the N data are just got M;
Processing mode when vertex sequence length is odd number is identical with the step (2) of embedded system.
(3) calculate the Fourier descriptors Z of VQ according to DFT territory watermarking algorithm;
(4) calculate | the correlation C of Z/ and embed watermark, by comparing the size of C and threshold value T, judge whether the profile of polar plot to be detected contains watermark.
Above judgement be judgement under the simplest situation, in practice we and do not know that what kind of polar plot to be detected stood and attacked, so we should consider all information summaries when detecting.For this reason, also need to ask for following several correlations:
4.1 when profile vertex sequence length is even number, starting point is moved correlation behind the odd number.At this moment will with two correlations respectively with threshold ratio, if having a value greater than threshold value one value less than threshold value, just the judgement contain watermark, other is judged to and does not contain watermark.
When 4.2 profile vertex sequence length is odd number, with the Q in the detection (2) that obtains
oAnd Q
eDetected value after the exchange.Equally just two correlations respectively with threshold ratio, if having a value greater than threshold value one value less than threshold value, just judgement contains watermark, otherwise does not contain watermark.
The inventive method is compared with existing algorithm, has following conspicuous outstanding substantive distinguishing features and remarkable advantage:
The first, fully take into account the influence that geometric attack causes polar plot, so adopt the CCA method to be transformed into the canonical correlation variable state that affine attack is had no effect to the affine polar plot of attacking sensitivity originally.This just is equivalent to affined transformation and does not exist, and this watermark detection to the back is highly beneficial.
The comprehensive attack of the watermark that the second, takes into full account, i.e. the attack that will get in touch to some extent originally by CCA is mapped to the attack independent of each other on the canonical correlation variable.
Three, consider the various attacks that may exist greatly, promptly introduce DFT territory watermarking algorithm, well resisted the attack that changes starting point.
Four, this method has extraordinary robustness, has guaranteed to have judged high-accuracy.
Description of drawings
Fig. 1 is based on the process flow diagram of digital watermarking protection method of 2 dimensional vectors of canonical correlation analysis, and picture in picture (a) is that watermark embeds flow process, and figure (b) is the watermark detection flow process.
Fig. 2 is the vector outline figure before and after the embodiment embed watermark, and picture in picture (a) is original vector outline figure, and figure (b) is for containing the vector outline figure of watermark.
Fig. 3 be the present invention and DFT territory method among the embodiment ROC relatively.
Fig. 4 is the ROC curve of embodiment under comprehensive the attack.
Embodiment
Details are as follows in conjunction with the accompanying drawings for a preferred embodiment of the present invention: with reference to Fig. 1; europ_map.svg is specifically implemented; this digital watermarking protection method embodiment based on 2 dimensional vectors of canonical correlation analysis is divided into two big steps, i.e. the embedding flow process of digital watermarking and the testing process of digital watermarking.We adopt CCA to obtain the canonical correlation variable that has unchangeability to affine at first step, and this makes watermark have very strong robustness to geometric attack; Simultaneously, we also have been incorporated into the watermarking algorithm in DFT territory, have strengthened the ability of the anti-starting point change of watermark.And in second step, we have reduced the False Rate of watermark by calculating a plurality of correlations, have improved the judgement performance of system.Specifically can be with reference to figure 1.
Details are as follows in conjunction with the accompanying drawings for two big steps of present embodiment:
One, the embedding flow process of digital watermarking
In order to be implemented in the application on the europ_map.svg, we at first need to select the profile apex coordinate sequence P of a polar plot, and its summit length is even number, then it is implemented following detailed operation:
(1) earlier the P barycenter is made zero, be divided into odd number group P according to following target parity then
oWith even number set P
e
(2) with P
oAnd P
eCarry out CCA, obtain P
oAnd P
eEach self-corresponding canonical correlation variable U and V get positive back by P with U and V then
oAnd P
eOdd even originally is combined into one group of sequence VP in proper order;
(3) utilization DFT territory watermarking algorithm obtains containing the WVP of watermark to the VP embed watermark;
(4) WVP elder generation parity packet and revert back original positive negativity is carried out the inverse operation of CCA then, obtains to contain the sequence of parity VP of watermark
oAnd WP
e
(5) contain the WP of watermark
oAnd WP
eBe combined into unique sequence by odd even, original barycenter is returned in translation then.Promptly revert to the vector polygon sequence WP that contains watermark.Specifically can be with reference to embedding flow process and Fig. 2 of figure 1.
Two, the testing process of digital watermarking
After finishing the first step, the polar plot that contains watermark just can be issued or be used on the net.When the suspection polar plot is encroached right or run into the copyright ownership problem, just need detect watermark.Below we just are described in detail in conjunction with the operation of the europ_map.svg polar plot that contains watermark to the detection system of digital watermarking.Europ_map.svg profile vertex sequence to be detected is Q, and it is carried out following operation:
(1), is divided into odd number group idol Q by odd even then with the normalization of Q barycenter
oWith even number set Q
e
(2) to Q
oAnd Q
eCarry out CCA, obtain the canonical correlation variable M and the N of their correspondences,, be combined into unique sequence VQ after the N data are just got M;
(3) calculate the Fourier descriptors Z of VQ according to DFT territory watermarking algorithm;
(4) calculate | the correlation C of Z/ and embed watermark, by comparing the size of C and threshold value T, judge whether the profile of polar plot to be detected contains watermark.Specifically can be with reference to the testing process of figure 1.
Here, we are in order to verify advantage of the present invention, we by calculate watermark false dismissed rate and false declaration rate draw out ROC curve under the various situations, the ROC curve can be good at reflecting superiority of the present invention.Specifically can be with reference to figure 3 and Fig. 4.
Claims (1)
1. digital watermarking protection method based on 2 dimensional vectors of canonical correlation analysis is characterized in that implementation step is:
A. implement the embedding flow process of digital watermarking;
B. implement the testing process of digital watermarking;
The step that the enforcement digital watermarking of described steps A embeds flow process is:
A. the profile diagram apex coordinate sequence P barycenter of original polar plot is made zero in elder generation, is divided into two groups of P according to following target parity then
oAnd P
e, i.e. odd number group and even number set are odd number for vertex sequence length, need to use annular parity packet method;
B. with P
oAnd P
eCarry out canonical correlation analysis, obtain P
oAnd P
eEach self-corresponding canonical correlation variable U and V get positive back by P with U and V then
oAnd P
eOdd even originally is combined into one group of sequence VP in proper order, is odd number for vertex sequence length, only takes out required part;
C. use DFT territory watermarking algorithm to the VP embed watermark, the VP that contains watermark represents with WVP;
D.WVP elder generation's parity packet also reverts back original positive negativity, carries out the inverse operation of canonical correlation analysis then, obtains to contain the sequence of parity WP of watermark
oAnd WP
e, vertex sequence length is odd number, needs first polishing data to carry out the inverse operation of canonical correlation analysis again;
E. the WP that contains watermark
oAnd WP
eBe combined into unique sequence by odd even, original barycenter is returned in translation then, promptly reverts to the vector polygon sequence WP that contains watermark, and sequence length is an odd number, needs to take out the data that need and merges;
The step of the enforcement digital watermarking testing process of described step B is:
A. after the polar plot profile vertex sequence Q barycenter to be detected normalization, be divided into two groups of Q of odd even by odd even
oAnd Q
e, the processing mode when vertex sequence length is odd number is identical with the step a that implements digital watermarking embedding flow process;
B. to Q
oAnd Q
eCarry out canonical correlation analysis, obtain canonical correlation variable M and N, with M, be combined into unique sequence VQ after the N data are just got, the processing mode when vertex sequence length is odd number is identical with the step b that implements digital watermarking embedding flow process;
C. calculate the Fourier descriptors Z of VQ according to DFT territory watermarking algorithm;
D. calculate | the correlation C of Z| and embed watermark, by comparing the size of C and threshold value T, judge whether the profile of polar plot to be detected contains watermark; For the accuracy of judging also needs to ask for following several correlations:
When i. profile vertex sequence length is even number, starting point is moved correlation behind the odd number;
When ii. profile vertex sequence length is odd number, with the Q among the step b that obtains
oAnd Q
eDetected value after the exchange.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN2008102039975A CN101447070B (en) | 2008-12-04 | 2008-12-04 | Digital watermarking protection method of two-dimensional vector graph based on canonical correlation analysis |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN2008102039975A CN101447070B (en) | 2008-12-04 | 2008-12-04 | Digital watermarking protection method of two-dimensional vector graph based on canonical correlation analysis |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN101447070A CN101447070A (en) | 2009-06-03 |
| CN101447070B true CN101447070B (en) | 2011-03-30 |
Family
ID=40742737
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN2008102039975A Expired - Fee Related CN101447070B (en) | 2008-12-04 | 2008-12-04 | Digital watermarking protection method of two-dimensional vector graph based on canonical correlation analysis |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN101447070B (en) |
Families Citing this family (8)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN101923698B (en) * | 2009-06-11 | 2013-01-30 | 株式会社理光 | Method and device for embedding and detecting watermark information |
| CN102903075B (en) * | 2012-10-15 | 2015-04-08 | 西安电子科技大学 | Robust watermarking method based on image feature point global correction |
| CN103049881B (en) * | 2012-12-22 | 2016-02-17 | 浙江工业大学 | Based on the vector watermark attack method of vector geographic data structure |
| CN103281178B (en) * | 2013-06-08 | 2016-01-20 | 深圳大学 | A kind of anonymous communication method and system |
| CN105488360B (en) * | 2015-11-27 | 2018-06-01 | 北京印刷学院 | A kind of suffix is the anti-fake processing method and device of the threedimensional model file of fbx forms |
| CN105512522B (en) * | 2015-11-27 | 2018-08-03 | 北京印刷学院 | A kind of suffix is the anti-fake processing method and device of the threedimensional model file of 3ds formats |
| CN105488362B (en) * | 2015-11-27 | 2018-06-01 | 北京印刷学院 | A kind of suffix is the anti-fake processing method and device of the threedimensional model file of obj forms |
| CN112801844A (en) * | 2021-01-19 | 2021-05-14 | 广州数智网络科技有限公司 | Reversible watermark of high-fidelity semi-fragile two-dimensional vector diagram |
-
2008
- 2008-12-04 CN CN2008102039975A patent/CN101447070B/en not_active Expired - Fee Related
Also Published As
| Publication number | Publication date |
|---|---|
| CN101447070A (en) | 2009-06-03 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN101447070B (en) | Digital watermarking protection method of two-dimensional vector graph based on canonical correlation analysis | |
| Yu et al. | A robust watermarking scheme for 3D triangular mesh models | |
| Fadl et al. | Robust copy–move forgery revealing in digital images using polar coordinate system | |
| Al-Qershi et al. | Evaluation of copy-move forgery detection: datasets and evaluation metrics | |
| CN103049877A (en) | Method for embedding and detecting watermark based on vertex distribution feature digital three-dimensional model | |
| CN104091303A (en) | Robust image hashing method and device based on Radon transformation and invariant features | |
| CN108122225A (en) | Digital image tampering detection method based on self-adaptive features point | |
| CN102629325B (en) | Image characteristic extraction method, device thereof, image copy detection method and system thereof | |
| Yang et al. | A steganalytic algorithm for 3D polygonal meshes | |
| Soni et al. | multiCMFD: fast and efficient system for multiple copy-move forgeries detection in image | |
| CN102024244B (en) | Method and device for embedding and detecting watermarks based on image characteristic region | |
| CN104504645A (en) | Method for embedding and detecting robust image watermarks on basis of circular-harmonic-Fourier moments | |
| CN108876694B (en) | Three-dimensional model blind digital watermarking algorithm based on Schur decomposition | |
| Wang et al. | Blind detection of region duplication forgery by merging blur and affine moment invariants | |
| CN102314667B (en) | Vertex weight value-based OBJ (object)-format three-dimensional model digital-watermarking method | |
| Zhao et al. | A robust image hashing method based on Zernike moments | |
| CN103310134A (en) | Vector data watermark anti-counterfeiting method based on geographical semantics support | |
| Zhou et al. | Watermark embedded in polygonal line for copyright protection of contour map | |
| Zhao et al. | Perceptual image hash for tampering detection using Zernike moments | |
| Pham et al. | Robust object-based watermarking using feature matching | |
| Shah et al. | Copy-move forgery detection using hybrid transform and K-means clustering technique | |
| Yang et al. | A robust scheme for copy detection of 3D object point clouds | |
| Wang et al. | A feature-based robust digital image watermarking against desynchronization attacks | |
| CN105389767B (en) | It is a kind of to resist the digital watermarking generation and detection method and system that radial transformation is attacked | |
| Wang et al. | Copy move image forgery detection based on polar Fourier representation |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| C06 | Publication | ||
| PB01 | Publication | ||
| C10 | Entry into substantive examination | ||
| SE01 | Entry into force of request for substantive examination | ||
| C14 | Grant of patent or utility model | ||
| GR01 | Patent grant | ||
| C17 | Cessation of patent right | ||
| CF01 | Termination of patent right due to non-payment of annual fee |
Granted publication date: 20110330 Termination date: 20131204 |