CN103873871A - Geometric-attack-resistant robust video watermark method based on timeline spread spectrum - Google Patents

Abstract

The invention discloses a geometric-attack-resistant robust video watermark method based on timeline spread spectrum. In the watermark embedding procedure, a video frame image is shown through RGB color space, embedded two-value pictorial information is shown through a 0-1 array, a watermark information array is made to correspond to the B component color array of the video frame image in position, the watermark information in the corresponding position is embedded to a timeline in a blind spread spectrum watermark mode, and in order to resist attack on a video frame sequence, pseudorandom information used when spread spectrum embedding is performed is saved by changing the high-frequency coefficient comparison difference value generated after the R color component and the G color component of the video frame image are subjected to three-layer DWT. In the watermark extracting process, firstly, by comparing the high-frequency coefficient generated after the R color component and the G color component of the video frame image are subjected to three-layer DWT, the pseudorandom information is extracted, and then watermark information on the timeline is extracted. With the method, the robustness of the video watermark to geometric attack is improved, and certain robustness to compressing and the attack on the video frame sequence is also achieved.

Description

A kind of anti-geometry attack robust video watermark process based on time shaft spread spectrum

Technical field

The present invention relates to a kind of robust video digital watermark method, particularly a kind of water mark method of resisting geometric attack.

Background technology

As everyone knows, digital multimedia has been substituted traditional analog form, by the development of Internet communication technology, digital multimedia can conveniently copy and propagate, and the problem of piracy and the copyright dispute that cause thus also become day by day serious social concern.Estimate according to RIAA, the whole world is every year because the pirate economic loss causing is up to 5,000,000,000 dollars, and American film employer's organization estimates, piracy makes the annual income of American film industry reduce 2,500,000,000 dollars.

Traditional copyright protection means are cipher authentication techniques; cryptographic technique can ensure the information security in the communication process of copyright from sender to recipient; once but copyright is decrypted, cryptographic technique just no longer has protective effect, bootlegger is copy propagation arbitrarily.

In order to make up the deficiency of cryptographic technique in copyright protection field, a kind of new copyright protection means---digital watermark technology has obtained paying close attention to and development.Digital watermarking can be embedded in Digital Media (image, sound, document, video) non the relevant information of representative works copyright; it can prove the ownership of authorship to its works; and as qualification, the illegal evidence of encroaching right of prosecution; the integrality of copyright be can pass through the determination and analysis of watermark to ensure simultaneously, thereby intellectual property protection and the false proof effective means of digital multimedia become.

Video is as the widely used multimedia form of people; play an important role in daily life; the importance of its copyright protection is self-evident; robustness is the essential condition that video watermark need to be satisfied; and be the research emphasis of video watermark for the robustness of geometric attack always; this method is fully to study on the basis of the existing various video watermark process of resisting geometric attack; in conjunction with the advanced technology in the fields such as computer vision, signal processing, the novel robust video watermark process being highly resistant to taking geometric attack as main various attack proposing.

Summary of the invention

The technical problem to be solved in the present invention is: design a kind of robust video watermark process common attacks such as geometric attack of resisting of function admirable, to protect better the copyright of digital video.

The technical solution adopted for the present invention to solve the technical problems: a kind of robust video watermark process of the geometric attack resisted based on time shaft spread spectrum, comprise watermark embed process and watermark extraction process two parts, the watermark information embedding is two-value picture, the time shaft data of utilizing video frame images are preserved watermark information by blind spread-spectrum watermark technology, utilize difference between three layers of DWT conversion coefficient of video frame images different colours component to preserve and embed and extract the required pseudo-random information of watermark.

Geometric attack is namely as rotation, convergent-divergent, rule or the irregular geometric transformations such as shearing, can damage the spatial synchronization between video frame images and watermark, traditional watermark embedded technology is not generally highly resistant to the adverse effect that geometric attack causes watermark extracting, because adjacent frame of video can experience same conversion in the time suffering geometric attack, just can not be subject to like this impact of geometric attack at the synchronized relation of the time domain time shaft pixel sequence that namely the same position place of each frame forms, then utilize the pixel sequence of time shaft to adopt spread-spectrum watermark technology to preserve watermark information, just geometric attack can be passed to watermark information to the impact of frame of video, that is to say that watermark information can experience same geometric transformation with frame of video, if adopt two-value picture as watermark information, just can rely on human eye the information recognition capability of picture to be eliminated to the impact of geometric attack.

Due to features such as the required memory space of video is large, transmission difficulties, generally need video watermark to there is blind Detecting characteristic, preserve watermark information so need to utilize blind spread-spectrum watermark technology, although video is three dimensional signal, when original blind spread-spectrum watermark technical finesse, regard video as one-dimensional signal, telescopiny first will be by watermark information spread spectrum, and its conversion formula is as follows:

w' _i＝w _j,j×cr≤i＜(j+1)×cr,i∈N

Wherein w _jrepresent original watermark information, w _j{ N is natural number to ∈ for 1,1}, j ∈ N, and cr is spread spectrum coefficient, w' _ifor the watermark information after spread spectrum.After spread spectrum completes, watermark information is embedded on pixel value, its computing formula is as follows:

p′ _i＝p _i+α×r _i×w _i

Wherein p _ifor original pixel value, α is amplification coefficient, r _ifor pseudo random sequence, r _i∈ { 1,1}, w' _ifor the watermark information after spread spectrum, p' _ifor the pixel value after watermarked, i ∈ N, N is natural number.

In leaching process, first calculate relevant and, its computational process is as follows:

$s_j=\underset{i=j\×\mathrm{cr}}{\overset{(j+1)\×\mathrm{cr}-1}{\mathrm{\Σ}}}{p}_i^{\′}\×r_i=\underset{i=j\×\mathrm{cr}}{\overset{(j+1)\×\mathrm{cr}-1}{\mathrm{\Σ}}}{p}_i\×r_i+\underset{i=j\×\mathrm{cr}}{\overset{(j+1)\×\mathrm{cr}-1}{\mathrm{\Σ}}}\mathrm{\α}\×r_i\×w_i^{\′}\×r_i\≈\mathrm{cr}\×\mathrm{\α}\×w_j$

Wherein p' _ifor pixel value, r _ifor pseudo random sequence, r _i{ 1,1}, cr is spread spectrum coefficient to ∈, s _jsymbol just represent original j position watermark information, that is: sign (s _j)=sign (cr × α × w _j)=sign (w _j)=w _j.

Because blind spread-spectrum watermark Technology Need uses pseudo random sequence to embed and extracts watermark information, like this watermark extraction process be just easily subject to attack for sequence of frames of video (such as frame exchange, LOF, frame on average etc.) impact, in order to improve the robustness of this water mark method, by utilizing the contrast between video frame images different colours component preserve embedding and extract the required pseudo-random information of watermark, be so just highly resistant to the attack for sequence of frames of video.

The advantage that the present invention compared with prior art had is:

1, can resist well geometric attack, such as rotation, shearing, convergent-divergent, upset etc.;

2, can resist preferably the attack for sequence of frames of video, such as the exchange of LOF, frame, frame on average etc., the common attacks such as compression are also had to reasonable robustness;

3, algorithm is realized simply, and the process that embeds and extract is clear and definite, and computation complexity is little;

4, designed watermark is blind watermark, that is to say and in the time of watermark extracting, does not need original video and original embed watermark information.

Brief description of the drawings

Fig. 1 is overall framework figure of the present invention;

Fig. 2 is watermarking algorithm flow chart of the present invention;

Fig. 3 is watermark extracting algorithm flow chart of the present invention.

Embodiment

Video watermark process comprises watermark embedding and watermark extracting two parts, and its overall framework figure is Fig. 1.

Watermark embed process flow chart is shown in Fig. 2, and its step is as follows:

Step 1: the two-value picture size that represents watermark information is adjusted into 1/9 of two field picture size, then, centered by this two-value picture, fills neighboring area with white, build the binary bitmap onesize with two field picture, w represents with matrix.

Step 2: read i frame of video, i is sequence number, i ∈ N, N is natural number.

Step 3: to the Blue channel components of i two field picture, travel through this frame picture pixel, carry out watermark embedding by following formula:

p' _i(x,y)＝p _i(x,y)+α*r _i*(2*w(x,y)-1)

Wherein (x, y) is coordinate, p _ifor original pixel value, α is the amplification coefficient that value is greater than 0, r _ifor pseudo random sequence, r _i∈ 1,1}, and w is two-value picture matrix, wherein 0 represents black, 1 represents white, why selects Blue passage, is because human eye is to least sensitivity of blue component.

Step 4: the Red component to i two field picture and Green component, respectively it is carried out to 3 layer of 2 dimension DWT conversion, obtain the horizontal high fdrequency component chGreen of the horizontal high fdrequency component chRed of the 3rd layer of Red component after conversion, longitudinally high fdrequency component cvRed and Green component, longitudinal high fdrequency component cvGreen, each component can be regarded the two-dimensional matrix of formed objects as, embeds pseudo random sequence value r by following rule _i:

If r _ibeing 1 and D=|chRed (x, y)-chGreen (x, y) | < T, expands the difference of chRed and chGreen by two formulas below:

$\left\{\begin{array}{c}\mathrm{chGree}{n}^{\&prime;}(x,y)=\mathrm{chGreen}(x,y)-\frac{T-D}{2},\mathrm{chGreen}(x,y)<\mathrm{chRed}(x,y)\\ \mathrm{chGree}{n}^{\&prime;}(x,y)=\mathrm{chGreen}(x,y)+\frac{T-D}{2},\mathrm{chGreen}(x,y)>\mathrm{chRed}(x,y)\end{array}\right.$

$\left\{\begin{array}{c}\mathrm{chRe}{d}^{\&prime;}(x,y)=\mathrm{chRed}(x,y)-\frac{T-D}{2}\mathrm{chGreen}(x,y)>\mathrm{chRed}(x,y)\\ \mathrm{chRe}{d}^{\&prime;}(x,y)=\mathrm{chRed}(x,y)+\frac{T-D}{2},\mathrm{chGreen}(x,y)<\mathrm{chRed}(x,y)\end{array}\right.$

If r _ibeing 1 and D=|chRed (x, y)-chGreen (x, y) |>=T does not change.

If r _ifor-1 and D=|chRed (x, y)-chGreen (x, y) | > T, by two formulas are dwindled chRed and chGreen below difference:

$\left\{\begin{array}{c}\mathrm{chGree}{n}^{\&prime;}(x,y)=\mathrm{chGreen}(x,y)-\frac{T-D}{2},\mathrm{chGreen}(x,y)<\mathrm{chRed}(x,y)\\ \mathrm{chGree}{n}^{\&prime;}(x,y)=\mathrm{chGreen}(x,y)+\frac{T-D}{2},\mathrm{chGreen}(x,y)>\mathrm{chRed}(x,y)\end{array}\right.$

$\left\{\begin{array}{c}\mathrm{chRe}{d}^{\&prime;}(x,y)=\mathrm{chRed}(x,y)-\frac{T-D}{2}\mathrm{chGreen}(x,y)>\mathrm{chRed}(x,y)\\ \mathrm{chRe}{d}^{\&prime;}(x,y)=\mathrm{chRed}(x,y)+\frac{T-D}{2},\mathrm{chGreen}(x,y)<\mathrm{chRed}(x,y)\end{array}\right.$

If r _ifor-1 and D=|chRed (x, y)-chGreen (x, y) |≤T does not change.

CvRed is identical with the operation of chRed and chGreen with cvGreen, above-mentioned T value in various for value be positive threshold value.

Step 5: preserve frame of video.

Watermark extraction process flow chart is shown in Fig. 3, and its step is as follows:

Step 1: define a matrix m onesize with two field picture, its each value of initialization is 0.

Step 2: read i frame of video, Red component to i two field picture and Green component, respectively it is carried out to 3 layer of 2 dimension DWT conversion, obtain the 3rd layer of horizontal high fdrequency component chRed, chGreen and longitudinal high fdrequency component cvRed, the cvGreen after conversion, travel through this two component pixel values, by following Rule Extraction pseudo random sequence value r _i, wherein sum is that initial value is 0 count value, value can just can be born:

If D=|chRed (x, y)-chGreen (x, y) |≤T, sum value subtracts 1;

If D=|chRed (x, y)-chGreen (x, y) | > T, sum value adds 1;

If D=|cvRed (x, y)-cvGreen (x, y|≤T sum value subtracts 1;

If D=|cvRed (x, y)-cvGreen (x, y) | > T, sum value adds 1.

After traversal finishes, extract r according to following formula _ivalue:

$r_i=\left\{\begin{array}{c}1,\mathrm{sum}>0\\ -1,\mathrm{sum}<0\end{array}\right.$

Step 3: for Blue component, traversal picture pixel, operates by following formula:

m(x,y)＝m(x,y)+p' _i(x,y)*r _i

Wherein (x, y) is coordinate, p' _ifor pixel value, m is the set matrix of step 1, r _ifor pseudo random sequence, r _i∈ { 1,1}.

Step 4: read after all frame of video, Ergodic Matrices m, setting watermark information is matrix w, and w and m are onesize, and w presses following formula assignment:

$w(x,y)=\left\{\begin{array}{c}0,m(x,y)<0\\ 1,m(x,y)\&GreaterEqual;0\end{array}\right.$

Matrix w is the watermark information of extraction, the bianry image of namely original embedding.

Claims (4)

1. the anti-geometry attack robust video watermark process based on time shaft spread spectrum, it is characterized in that: comprise watermark embedding and watermark extracting two parts, the watermark information embedding is two-value picture, utilize the time shaft data of video frame images to preserve watermark information by spread-spectrum watermark technology, utilize the wavelet conversion coefficient value contrast between video frame images different colours component to preserve embedding and extract the required pseudo-random information of watermark;

Described watermark embed process is:

(1.1) two-value picture is adjusted into video frame size, with two-value picture matrix, w represents;

(1.2) the Blue channel components to i two field picture, travels through this frame picture pixel, carries out watermark embedding by following formula:

p' _i(x,y)＝p _i(x,y)+α*r _i*(2*w(x,y)-1)；

Wherein, (x, y) is coordinate, p _ifor original pixel value, α is the amplification coefficient that value is greater than 0, r _ifor pseudo random sequence, r _i{ N is natural number to ∈ for 1,1}, i ∈ N, and w is two-value picture matrix, wherein 0 represents black, and 1 represents white;

(1.3) the Red component to i two field picture and Green component, respectively it is carried out to 3 layer of 2 dimension DWT conversion, obtain the horizontal high fdrequency component chGreen of the horizontal high fdrequency component chRed of the 3rd layer of Red component after conversion, longitudinally high fdrequency component cvRed and Green component, longitudinal high fdrequency component cvGreen, each component can be regarded the two-dimensional matrix of formed objects as, set threshold T, T get on the occasion of, if r _ibe 1, the absolute difference of adjusting the parameter value of the same position of matrix chRed and matrix chGreen is greater than T, if r _ifor-1, the absolute difference of adjusting the parameter value of the same position of matrix chRed and matrix chGreen is less than T, and matrix cvRed and cvGreen process equally;

(1.4) preserve frame of video;

Described watermark extraction process is:

(2.1) a matrix m onesize with two field picture of definition, its each value of initialization is 0;

(2.2) the Red component to i two field picture and Green component, respectively it is carried out to 3 layer of 2 dimension DWT conversion, obtain the horizontal high fdrequency component chGreen of the horizontal high fdrequency component chRed of the 3rd layer of Red component after conversion, longitudinally high fdrequency component cvRed and Green component, longitudinal high fdrequency component cvGreen, each component can be regarded the two-dimensional matrix of formed objects as, whether the absolute value of the difference of the parameter value of the same position of judgment matrix chRed and matrix chGreen, matrix cvRed and matrix cvGreen is greater than T, if it is in the majority to be greater than the parameter of T, r _ibe 1, if it is in the majority to be less than the parameter of T, r _ifor-1;

(2.3), for Blue component, traversal picture pixel, operates by following formula:

m(x,y)＝m(x,y)+p' _i(x,y)*r _i

Wherein (x, y) is coordinate, p' _ifor pixel value, m is set matrix in step 2.1;

(2.4) read after all frame of video, Ergodic Matrices m, setting watermark information is matrix w, and w and m are onesize, and w presses following formula assignment:

$w(x,y)=\left\{\begin{array}{c}0,m(x,y)<0\\ 1,m(x,y)\&GreaterEqual;0\end{array}\right.$

W is the watermark information of extraction, the two-value picture matrix of namely original embedding.

2. the anti-geometry attack robust video watermark process based on time shaft spread spectrum according to claim 1, preserve horizontal high fdrequency component chRed, longitudinally high fdrequency component cvRed and the Green component of in pseudo-random information process, adjusting the 3rd layer of Red component after wavelet transformation horizontal high fdrequency component chGreen, longitudinally the concrete operations of high fdrequency component cvGreen are:

If r _ibeing 1 and D=|chRed (x, y)-chGreen (x, y) | < T, expands the difference of chRed and chGreen by two formulas below:

$\left\{\begin{array}{c}\mathrm{chGree}{n}^{\&prime;}(x,y)=\mathrm{chGreen}(x,y)-\frac{T-D}{2},\mathrm{chGreen}(x,y)<\mathrm{chRed}(x,y)\\ \mathrm{chGree}{n}^{\&prime;}(x,y)=\mathrm{chGreen}(x,y)+\frac{T-D}{2},\mathrm{chGreen}(x,y)>\mathrm{chRed}(x,y)\end{array}\right.$

$\left\{\begin{array}{c}\mathrm{chRe}{d}^{\&prime;}(x,y)=\mathrm{chRed}(x,y)-\frac{T-D}{2}\mathrm{chGreen}(x,y)>\mathrm{chRed}(x,y)\\ \mathrm{chRe}{d}^{\&prime;}(x,y)=\mathrm{chRed}(x,y)+\frac{T-D}{2},\mathrm{chGreen}(x,y)<\mathrm{chRed}(x,y)\end{array}\right.$

If r _ibeing 1 and D=|chRed (x, y)-chGreen (x, y) |>=T does not change;

If r _ifor-1 and D=|chRed (x, y)-chGreen (x, y) | > T, by two formulas are dwindled chRed and chGreen below difference:

$\left\{\begin{array}{c}\mathrm{chGree}{n}^{\&prime;}(x,y)=\mathrm{chGreen}(x,y)-\frac{T-D}{2},\mathrm{chGreen}(x,y)<\mathrm{chRed}(x,y)\\ \mathrm{chGree}{n}^{\&prime;}(x,y)=\mathrm{chGreen}(x,y)+\frac{T-D}{2},\mathrm{chGreen}(x,y)>\mathrm{chRed}(x,y)\end{array}\right.$

$\left\{\begin{array}{c}\mathrm{chRe}{d}^{\&prime;}(x,y)=\mathrm{chRed}(x,y)-\frac{T-D}{2}\mathrm{chGreen}(x,y)>\mathrm{chRed}(x,y)\\ \mathrm{chRe}{d}^{\&prime;}(x,y)=\mathrm{chRed}(x,y)+\frac{T-D}{2},\mathrm{chGreen}(x,y)<\mathrm{chRed}(x,y)\end{array}\right.$

If r _ifor-1 and D=|chRed (x, y)-chGreen (x, y) |≤T does not change;

CvRed is identical with the operation of chRed and chGreen with cvGreen, above-mentioned T value in various for value be positive threshold value, r _ifor value is 1 or-1 pseudo random sequence value.

3. the anti-geometry attack robust video watermark process based on time shaft spread spectrum according to claim 1, is characterized in that: the value of described α is 5≤α≤15.

4. the anti-geometry attack robust video watermark process based on time shaft spread spectrum according to claim 1, is characterized in that: the value of described T is 20≤T≤40.

Images