CN102510490A - Video watermark realizing method against geometric attack based on three-dimensional discrete cosine transform (DCT) - Google Patents

Abstract

The invention relates to a video watermark realizing method against geometric attack based on three-dimensional discrete cosine transform (DCT), which firstly embeds a watermark, and (1) global three-dimensional DCT conversion is carried out to an original video section, and a visual characteristic vector which resists geometric attack is extracted from a conversion efficient; (2) a binary logic sequence is worked out with the visual characteristic vector and the watermark to be embedded through a Hash function, and the binary logic sequence is stored in a third party, and the watermark is extracted; (3) the global three-dimensional DCT conversion is carried out to a tested video, and the characteristic vector of the video which resists geometric attack is extracted; and (4) the watermark is extracted with the characteristics of the Hash function and the binary logic sequence which is stored in the third party. The method,based on a three-dimensional DCT conversion video digital zero-watermark technology realizes the embedment of the watermark without affecting the video image, moreover the method has higher robustness and stronger attack resistant capability, and provides a powerful technical path for protecting the copyright of the video.

Description

A kind of video watermark implementation method based on three-dimensional DCT resist geometric attacks

Technical field

The invention belongs to field of multimedia signal processing, relate to a kind of video digital watermark technology based on three-dimensional DCT, specifically is a kind of video watermark implementation method based on three-dimensional DCT resist geometric attacks.

Background technology

Along with the fast development of multimedia and Internet technology, people on the net can very convenient download multimedia messages, but the Copyright Protection of incident multi-medium data is also serious day by day.When people transmitted through the network information of carrying out easily, also the piracy for Digital Media in the Internet provided convenience.Digital media content though learning, conventional cipher can solve safe transfer and access control, in case after password was deciphered, just can arbitrarily be copied, propagates and distort.Therefore, how both to make full use of the facility of internet, and can effectively protect the safe storage, transmission of copyright and the information of Digital Medias such as image, video etc. again, become a very urgent subject.Digital watermarking is a kind of Digital Media copyright protection means that proposed in recent years; Through detecting and extract digital watermarking; Can identify and verify out author, owner, the publisher of Digital Media works or the information of authorizing the consumer; Can also review the illegal propagation of copyright, be a kind of effective technical means of carrying out the digital copyright protection at present, remedied the deficiency of conventional encryption technique.But; Along with Development of Multimedia Technology, the Computer Processing video capability improves constantly, and video product is more and more (like DVD; VCD; HDTV, VOD multi-media material, video tape etc.), need solution badly for the copyright protection of a large amount of consumer video products, the research of video watermark becomes the field that Chinese scholars is paid close attention to gradually.

At present less for the research of the video digital watermark algorithm of resist geometric attacks.The vision mode that is adopted among the application JPEG such as Wolfgang has proposed the DCT territory watermarking algorithm of image adaptive; Propositions such as Hartung utilize the algorithm of thought embed watermark in the MPEG-2 compressed video of spread spectrum; Watermark signal is through expansion, amplification and modulation; Obtain a quasi-random sequence, then it is carried out 8 * 8 DCT, and the DCT coefficient is added on 8 * 8 the DCT coefficient of MPEG-2 code stream; Simitopoulos etc. have proposed a kind of algorithm at mpeg stream compression domain embed watermark, combine visual analysis and block sort technology, select the quantification in I frame brightness module DCT territory to exchange (AC) coefficient, embed watermark adaptively; Li Hua, Zhu Guangxi, Zhu Yaoting has proposed a kind of based on human eye vision sensor model territory digital watermarking hidden method according to the human eye vision frequency response function; Busch, Hsu, Dittmann etc. have proposed the video watermark algorithm based on the DCT coefficient, and they combine the still image watermarking algorithm of Koch and Zhao and human visual system's characteristic, make the watermark of embedding satisfy not sentience.

Obviously, common video watermark all is similar rest image digital watermark, and watermark information directly is inserted into compression stroke or changes its coefficient in transform domain, can't well resist geometric attack.This paper has proposed a kind of based on three-dimensional DCT robust video digital watermark algorithm; Its thinking is that the characteristic vector of video, Hash function, third party's notion are combined; Experimental result show this watermarking algorithm not only have stronger incompressible, Gauss disturbs and conventional attack ability such as medium filtering, also has geometric attack abilities such as anti-preferably rotation, convergent-divergent, translation and shearing.

In a word, in video, embed the method for the digital watermarking of geometric attacks such as anti-rotation, convergent-divergent, translation, shearing, and be zero digital watermark, such research still belongs to blank at present, does not see public reported.

Summary of the invention

The purpose of this invention is to provide a kind of video watermark implementation method based on three-dimensional DCT resist geometric attacks; Digital video zero digital watermark based on three-dimensional dct transform; Under the situation that does not influence video image, realize the embedding of watermark; Have higher robustness and stronger anti-attack ability, for protection video copyright provides strong technological path.

To achieve these goals; The present invention is performed such: carry out overall three-dimensional dct transform (here for video for video-frequency band; Be not divided into little 8x8x8 stereo block and carry out three-dimensional dct transform), in three-dimensional dct transform coefficient, extract the visual feature vector of a resist geometric attacks; And digital watermark and cryptography combined, realized the anti-geometry and the conventional attack of digital watermarking.The method that the present invention adopted comprises watermark embedding and watermark extracting two large divisions; First is that watermark embeds, and comprising: (1) obtains the characteristic vector V (j) of a resist geometric attacks through video being carried out overall three-dimensional dct transform; (2) according to the characteristic vector V (j) of watermark W (j) and the video that extracts through the Hash functional operation; Generate a two-valued function sequence Key (j), with two-valued function sequence Key (j), have the third party then; Second portion is a watermark extracting, comprising: the characteristic vector V ' that video to be measured is obtained in (3) (j), (4) utilize have third-party two-valued function sequence Key (j) and video to be measured characteristic vector V ' (j), extract watermark W ' (j).

Method of the present invention is elaborated as follows at present:

First: the embedding of watermark

At first with one group of binary pseudo-random W that can represent copyright information, W={w (j) | w (j)=0,1; 1≤j≤L} is as digital watermarking, original video be designated as F={f (i, j, k) | f (i, j, k) ∈ R; 1≤i≤M, 1≤j≤N, 1≤k≤P) }, wherein; (i, j k) represent voxel (Voxel) data value of watermark sequence and video respectively for w (j) and f; Grey scale pixel value in this similar two dimensional image is established N=M (length and width of establishing every frame picture are the same), and the embedding of watermark is following:

1) through original video being carried out overall three-dimensional dct transform, obtains a characteristic vector V (j) of this video;

(i, j k) carry out overall three-dimensional dct transform, obtain three-dimensional DCT coefficient matrix FD (i to original video F earlier; J, k), again from three-dimensional DCT coefficient matrix FD (i, j; K) in, L value before taking out, and through computing obtains the characteristic vector V (j) of this video to three-dimensional DCT coefficient symbols; Specific practice is to work as the DCT coefficient for we are with " 1 " expression just, and with " 0 " expression (reason is part as follows), program process was described below when coefficient was negative or zero:

FD(i，j，k)＝DCT3(F(i，j，k))

V(j)＝Sign(FD(i，j，k))

2) the characteristic vector V (j) according to watermark W (j) and video generates a two-valued function sequence Key (j)

$\mathrm{Key}\left(j\right)=V\left(j\right)\⊕W\left(j\right)$

Key (j) is by the characteristic vector V (j) of video and watermark W (j), and the HASH function commonly used through cryptography generates.Preserve Key (i), need use when extracting watermark afterwards.Through Key (j) is applied for to the third party as key,, reach the purpose of copyright protection to obtain the ownership of original works.

Second portion: the extraction of watermark

3) the visual feature vector V ' that obtains video to be measured (j)

If video to be measured is that (k), through obtaining the method that three-dimensional DCT coefficient matrix is FD ' (i, j is k), by above-mentioned steps 1) behind the overall three-dimensional dct transform to video, the characteristic vector V ' that tries to achieve video to be measured (j) for i, j for F ';

FD’(i，j，k)＝DCT3(F’(i，j，k))

V’(j)＝Sign(FD’(i，j，k))

4) in video to be measured, extract watermark W ' (j)

According to the characteristic vector V ' that has third-party Key (j) that generates when the embed watermark and video to be measured (j), utilize watermark W ' that Hash character can extract video to be measured (j)

$W^{,}\left(j\right)=\mathrm{Key}\left(j\right)\⊕V^{,}\left(j\right)$

Differentiate the owner of video to be measured again according to W (j) and W ' degree of correlation (j).

The present invention and existing video watermark technology relatively have following advantage:

At first because the present invention is based on the digital watermark technology of three-dimensional dct transform; The embedding of watermark and extraction are in frequency domain, to carry out; Experimental data through the back confirms that this watermark not only has stronger anti-conventional attack ability, and have stronger nonshrinkly put, geometric attack abilities such as translation and shearing; Secondly, the watermark of embedding is a kind of zero watermark, does not influence the content of original video.

Below we from the explanation of theoretical foundation and experimental data:

1) 3 d-dem cosine transform (three-dimensional DCT)

Three-dimensional dct transform formula is following:

Corresponding size is M * N * P video, and 3 d-dem cosine direct transform (DCT) formula is following:

$F(u,v,w)=c\left(u\right)c\left(v\right)c\left(w\right)[\underset{x=0}{\overset{M-1}{\mathrm{\Σ}}}\underset{y=0}{\overset{N-1}{\mathrm{\Σ}}}\underset{p=0}{\overset{P-1}{\mathrm{\Σ}}}f(x,y,z)\·\cos \frac{(2x+1)\mathrm{u\π}}{2M}\cos \frac{(2y+1)\mathrm{v\π}}{2N}\cos \frac{(2z+1)\mathrm{w\π}}{2P}]$

u＝0，1，...，M-1；v＝0，1，...，N-1；w＝0，1，...，P-1；

In the formula:

$c\left(u\right)=\left\{\begin{array}{cc}\sqrt{1/M}& u=0\\ \sqrt{2/M}& u=\mathrm{1,2},...,M-1\end{array}\right.,$ $c\left(v\right)=\left\{\begin{array}{cc}\sqrt{1/N}& v=0\\ \sqrt{2/N}& v=\mathrm{1,2},...,N-1\end{array}\right.$

$c\left(w\right)=\left\{\begin{array}{cc}\sqrt{1/P}& w=0\\ \sqrt{2/P}& w=\mathrm{1,2},...,P-1\end{array}\right.$

Here f (x, y, z) be video V (x, y, the voxel of z) locating (voxel) data value, (u, v w) are the corresponding 3D-DCT conversion coefficient of this voxel data to F.

2) the visual feature vector choosing method of video:

The main cause of present most of watermarking algorithm resist geometric attacks ability is: people are embedded in digital watermarking in voxel or the conversion coefficient, and the slight geometric transformation of video usually can cause the bigger suddenly variation of voxel data value or transform coefficient values.The watermark that is embedded in like this in the video is just attacked easily.If can find the characteristic vector of a reflecting video content; So when little geometric transformation takes place in video; Tangible sudden change can not take place in this characteristic vector value; We are associated the characteristic vector of the digital watermarking that will embed and this video then, and the digital watermarking that embeds so just has resist geometric attacks ability preferably.The DCT coefficient observation discovery that we obtain through a large amount of videos being carried out overall dct transform;, a video-frequency band (realizes) when being carried out common geometric transformations such as convergent-divergent through every frame picture is carried out geometric transformation; Some variations possibly take place in the size of three-dimensional DCT Low Medium Frequency coefficient value; But its coefficient symbols remains unchanged basically, and we explain through some experimental datas of table 1.The former figure that is used as test in the table 1 is Fig. 1 (a); It is the frame picture in the film Shirley Temple video-frequency band (Shirley Temple.mat); What " the 1st row " showed in the table 1 is video type under attack, and this two field picture that receives behind the conventional attack is seen Fig. 1 (b)-(d); This two field picture that receives behind the geometric attack is seen Fig. 1 (e)-(l)." the 2nd row " expression of table 1 be the Y-PSNR (PSNR) of video after under attack; " the 3rd row " of table 1 arrive " the 11st row ", and this is F (1,1,1)-9 Low Medium Frequency coefficients of F (3,3,1) of getting in the three-dimensional DCT coefficient matrix.For conventional attack, these Low Medium Frequency coefficient values F (1,1,1)-F (3,3,1) remains unchanged and the DCT coefficient value approximately equal of former video basically; For geometric attack, the part coefficient has bigger variation, but we can find that video is when receiving geometric attack, and the size of most of DCT Low Medium Frequency coefficient has taken place to change but its symbol does not change basically.With " 1 " expression (containing value is zero coefficient), negative coefficient is with " 0 " expression, so for original video figure with positive DCT coefficient for we; F (1,1,1)-F (3 in the three-dimensional DCT coefficient matrix; 3,1) coefficient, corresponding coefficient symbols sequence is: " 100010011 "; Specifically see the 12nd row of table 1, we observe these row and can find, no matter conventional attack still is that the maintenance of this symbol sebolic addressing of geometric attack and original video is similar; With original video normalizated correlation coefficient all big (seeing table 1 " the 13rd row "), all greater than 0.5 (having got 9 three-dimensional DCT coefficient symbols here for the purpose of convenient).

But in order to prove that further the characteristic vector of extracting as stated above is a key character of this video, we different video measurement objects (seeing Fig. 2 (a)-(h)), carry out overall three-dimensional dct transform again; Obtain corresponding DCT coefficient F (1,1,1)-F (4; 4; 4),, preceding 64 DCT coefficients have been got here from the statistics angle.And obtain coefficient correlation each other, result of calculation is as shown in table 2.Can find out from table 2: the coefficient correlation between the video feature vector of same video-frequency band is maximum, and its value is 1.0.Phase relation numerical value is less between the video feature vector of different video, and it is worth less than 0.5.And this is desired consistent with us.This explanation has reflected the characteristic of video by the video features value of the method extraction of this invention.

In sum, we pass through the analysis to the overall three-dimensional DCT coefficient of video, utilize the symbol sebolic addressing of three-dimensional DCT Low Medium Frequency coefficient to obtain a kind of method that obtains the visual feature vector of video.

The low frequency part coefficient of the overall 3D-DCT conversion of table 1 video and receive different the attack after changing value

^*The 1.0e+004 of dct transform coefficient unit

The coefficient correlation of table 2 different video characteristic vector (vector length 64bit)

	Va	Vb	Vc	Vd	Ve	Vf	Vg	Vh
									Va	1.000	0.065	0.250	0.281	0.125	0.187	0.030	0.125
Vb	0.065	1.000	0.065	0.280	0.125	0.180	0.090	0.250
									Vc	0.250	0.065	1.000	0.030	0.310	0.310	0.030	0.250
Vd	0.280	0.280	0.030	1.000	-0.03	-0.160	0.180	0.080
									Ve	0.125	0.125	0.310	-0.030	1.000	0.420	0.090	0.420
Vf	0.187	0.180	0.310	-0.160	0.420	1.000	0.220	0.220
									Vg	0.030	0.090	0.030	0.180	0.090	0.220	1.000	0.280
Vh	0.125	0.250	0.250	0.080	0.420	0.220	0.280	1.000

Description of drawings

Fig. 1 (a) is a two field picture (acquiescence is the 1st frame among the video data Shirley Temple.mat) of original video.

Fig. 1 (b) is that to receive intensity be the two field picture after 4% the Gaussian noise.

Fig. 1 (c) is the two field picture after JPEG compression (compression quality is 10%).

Fig. 1 (d) is the two field picture (filtering parameter is [3x3]) behind medium filtering).

Fig. 1 (e) is the two field picture through 15 ° of up time rotations.

Fig. 1 (f) is the two field picture of video Shirley Temple through 2 times of convergent-divergents.

Fig. 1 (g) is the two field picture of video Shirley Temple through 0.5 times of convergent-divergent.

Fig. 1 (h) is that vertical direction moves down a two field picture of 10%.

Fig. 1 (i) is a move to right two field picture of 8% of level.

Fig. 1 (j) is that Z-direction is sheared the two field picture after 15%.

Fig. 1 (k) is that Y direction is sheared the two field picture after 15%.

Fig. 1 (l) is that X-direction is sheared the two field picture after 15%.

Fig. 2 (a) is the two field picture of video Cat and Mouse.

Fig. 2 (b) is the two field picture of video Xi Yangyang.

Fig. 2 (c) is the two field picture of video Shaolin temple.

Fig. 2 (d) is a kind of two field picture of video Afanda.

Fig. 2 (e) is the two field picture of video The king of lion.

Fig. 2 (f) is the two field picture of video Li Xiaolong.

Fig. 2 (g) is the two field picture of video Roman Holiday.

Fig. 2 (h) is the two field picture of video Shirley Temple.

Fig. 3 (a) does not add watermark one two field picture when disturbing.

Fig. 3 (b) does not add the watermark detector output when disturbing.

The two field picture (Gaussian noise intensity 10%) that Fig. 4 (a) adds Gauss when disturbing.

The watermark detector output that Fig. 4 (b) adds Gauss when disturbing.

Two field picture (the compression quality parameter is 2%) after Fig. 5 (a) JPEG compression.

Watermark detector output after Fig. 5 (b) JPEG compression.

Frame picture behind Fig. 6 (a) medium filtering (filtering parameter is [5x5], filtering 10 times).

The output of the watermark detector behind Fig. 6 (b) medium filtering.

Two field picture behind Fig. 7 (a) up time rotation 10 degree.

The output of Fig. 7 (b) up time rotation 10 degree back watermark detectors.

Fig. 8 (a) zoom factor is a two field picture of 2.0.

Fig. 8 (b) zoom factor is 2.0 watermark detector output.

Fig. 9 (a) zoom factor is a two field picture of 0.5.

Fig. 9 (b) zoom factor is 0.5 watermark detector output.

Figure 10 (a) moves down a two field picture of 10%.

Figure 10 (b) moves down the output of the watermark detector after 10%.

A move to right two field picture of 8% of Figure 10 (c).

The move to right output of the watermark detector after 8% of Figure 10 (d).

Figure 11 (a) shears the two field picture after 10% in X-direction.

Figure 11 (b) shears the output of 10% back watermark detector in X-direction.

Embodiment

Below in conjunction with accompanying drawing the present invention is described further:

Use 1000 groups independently every group of sequence length of binary pseudo-random (value for+1 or 0) be=64bit, in these 1000 groups of data, we appoint and extract one group (we select the 500th group) here, as the watermark sequence that embeds.One two field picture of original video is seen Fig. 1 (a), is the 1st frame of taking from the video (Shirley Temple.mat), and in experiment, the size of video is 300x300x6), representation of video shot be F (i, j, k), 1≤i wherein, j≤300; The 3D-DCT coefficient matrix that 1≤k≤6 is corresponding be FD (i, j, k), 1≤i wherein, j≤300; 1≤k≤6, consider robustness and disposable embed watermark capacity we get 64 coefficients of Low Medium Frequency (8x8x1=64), watermark is designated as W (j), 1≤j≤64; After detecting W ' through watermarking algorithm, we have judged whether that through calculating normalizated correlation coefficient NC (Normalized Cross Correlation) watermark embeds.

A two field picture that does not add when disturbing (is given tacit consent to the 1st frame of selecting video here, is tested with video and have 6 two field pictures composition.)

Fig. 3 (a) is a frame watermarking images that does not add when disturbing;

Fig. 3 (b) does not add when disturbing, and the output of watermark detector can be seen NC=1.00, obviously detects the existence of watermark.

Below we judge the anti-conventional attack ability and the resist geometric attacks ability robustness of this video watermark through concrete test.

The ability of the anti-conventional attack of watermarking algorithm:

(1) video data adds the experiment of Gaussian noise processing attack

Use imnoise () function in video watermark, to add gaussian noise.

Fig. 4 (a) is the frame picture when Gaussian noise intensity is 10%, and the PSNR=12.073dB of video is visually very fuzzy;

Fig. 4 (b) is the output of watermark detector, can clearly detect the existence of watermark, NC=1.0.

Table 3 is the anti-Gauss of watermark detection data when disturbing.Can see from experimental data; When Gaussian noise intensity when being 20%, the PSNR of video watermark reduces to 9.783dB, at this moment detects watermark; Coefficient correlation NC=1.0 still can detect the existence of watermark. and this explanation adopts this invention that good anti-Gaussian noise ability is arranged.

The anti-Gaussian noise interfering data of table 3 watermark

Noise intensity (%)	2	4	6	8	10	20
							PSNR(dB)	18.199	15.481	13.928	12.861	12.073	9.783
NC	1.00	1.00	1.00	1.00	1.00	1.00

(2) video data JPEG processed compressed is attacked experiment

Adopt image compression quality percentage watermarking images to be carried out the JPEG compression as parameter;

Fig. 5 (a) is that compression quality is a frame picture of 2%, and blocking artifact has appearred in this figure;

Fig. 5 (b) is the response of watermark detector, NC=1.0, and it is obvious to detect effect.

Table 4 is the test data of the anti-JPEG compression of video watermark.When compression quality is merely 2%, at this moment compression quality is lower, still can record the existence of watermark, NC=1.0.

The experimental data of the anti-JPEG compression of table 4 watermark

Compression ratio (%)	2	6	10	20	40	60
							PSNR(dB)	24.672	28.509	30.717	33.751	37.042	39.424
NC	1.00	1.00	1.00	1.00	1.00	1.00

(3) video data medium filtering processing attack experiment

Fig. 6 (a) is that the medium filtering parameter is [5x5], and the filtering number of repetition is a frame picture of 10, and image outline is not too clearly demarcated;

Fig. 6 (b) is the response of watermark detector, NC=0.966, and it is obvious to detect effect.

Table 5 is the anti-medium filtering ability of video watermark, and it can be seen from the table, when the medium filtering parameter is [7x7], the filtering number of repetition is 10 o'clock, still can record the existence of video watermark, NC=0.966.

The anti-medium filtering experimental data of table 5 watermark

Watermarking algorithm resist geometric attacks ability:

(1) experiment is attacked in the rotation transformation of video

Fig. 7 (a) is that a two field picture of video turns clockwise 10 °, the PSNR=14.743dB of video watermark at this moment, and signal to noise ratio is very low;

Fig. 7 (b) is the watermarking images of detection, can obviously detect the NC=0.723 that exists of watermark.

Table 6 is the anti-rotation of watermark challenge trial data.Can see in the table that when watermarking images rotates 15 ° NC=0.619 still can detect watermark and exist.

Experimental data is attacked in the anti-rotation of table 6 watermark

The rotation number of degrees	-20°	-15°	-10°	10°	15°	20°
							PSNR(dB)	12.639	13.425	14.743	14.743	13.421	12.637
NC	0.412	0.521	0.723	0.717	0.619	0.403

(2) the video data scale transformation is attacked experiment

Fig. 8 (a) for video through the two field picture behind 2 times of the convergent-divergents;

Fig. 8 (b) for video through behind 2 times of the convergent-divergents, watermarking detecting results can obviously detect the existence of watermark, NC=1.0.

Fig. 9 (a) for video through the two field picture behind 0.5 times of the convergent-divergent;

Fig. 9 (b) for video through behind 0.5 times of the convergent-divergent, watermarking detecting results can obviously detect the existence of watermark, NC=1.0.

Table 7 is the convergent-divergent challenge trial data of video watermark, from table 7 can see when the video scaling factor little to 0.2 the time, coefficient correlation NC=1.0 can obviously record the existence of watermark.

Table 7 watermark convergent-divergent is attacked experimental data

The convergent-divergent multiple	0.2	0.5	1.5	2.0	3.0	4.0
							NC	1.00	1.00	1.00	1.00	1.00	1.00

(3) the video data translation transformation is attacked experiment

Figure 10 (a) is for vertically moving down a two field picture of 10%, PSNR=16.282dB at this moment, and signal to noise ratio is lower;

Figure 10 (b) can obviously detect the existence of watermark, NC=0.883 for watermark detector output.

Figure 10 (c) is a move to right two field picture of 8% of level, PSNR=16.481dB at this moment, and signal to noise ratio is lower;

Figure 10 (d) can obviously detect the existence of watermark, NC=0.840 for watermark detector output.

Table 8, table 9 are the anti-translation challenge trial of watermark data.From table, to learn or when moving horizontally 15% when vertical moving 10%, two NC values all are higher than >=and 0.5, can obviously detect the existence of watermark, so this digital watermarking has stronger anti-translation capability.

The anti-vertical translation of table 8 watermark is attacked experimental data

Vertical moving	Move down 2%	Move down 6%	Move down 10%	On move 2%	On move 6%	On move 10%
							PSNR	24.789	18.673	16.282	24.789	16.673	16.282
NC	1.00	0.909	0.883	0.966	0.813	0.75

The anti-horizontal translation of table 9 watermark is attacked experimental data

Move left and right	Move to right 4%	Move to right 8%	Move to right 15%	Move to left 4%	Move to left 8%	Move to left 15%
							PSNR	24.789	16.481	14.105	19.607	16.481	14.105
NC	0.966	0.841	0.559	0.973	0.909	0.595

(4) video data shearing attack experiment

Figure 11 (a) can find that for shearing the two field picture after 10% by X-direction original relatively two field picture has cut bigger one.

Figure 11 (b) watermarking detecting results can obviously detect the existence of watermark, NC=0.557.

Table 10 is the anti-shearing data of watermark, from table, can see, when shearing from Z-direction, shearing displacement is 10% o'clock, still can detect the existence of watermark, and NC=1.0 explains that this watermarking algorithm has stronger anti-shear ability.

The anti-shearing attack experimental data of table 10 watermark

Through above description of test, the embedding grammar of this watermark has stronger anti-conventional attack ability and geometric attack ability, and the embedding of watermark do not influence the value of video data, is a kind of zero watermark.

Claims (1)

1. video watermark implementation method based on three-dimensional DCT resist geometric attacks; It is characterized in that: based on the extraction of the visual feature vector of three-dimensional dct transform of the overall situation and resist geometric attacks; And digital watermark and cryptography and " third party " notion combined; Realized the anti-geometry and the conventional attack of video digital watermark, this video digital watermark implementation method is divided into two parts, amounts to four steps:

First is that video watermark embeds: through the embedding operation to watermark, obtain corresponding two-valued function sequence Key (j);

1) original video is carried out overall three-dimensional dct transform, from the DCT coefficient, obtain the vectorial V (j) of the resist geometric attacks of this figure according to the symbol sebolic addressing of Low Medium Frequency coefficient;

2) utilize Hash function and the watermark W (j) that will embed, obtain a two-valued function sequence Key (j), $\mathrm{Key}\left(j\right)=V\left(j\right)\⊕W\left(j\right);$

Preserve Key (j), will use when extracting watermark below, through applying for as key Key (j) to the third party, to obtain ownership former figure;

Second portion is a watermark extracting: the characteristic vector V ' of the resist geometric attacks through two-valued function sequence Key (j) and video to be measured (j) extracts watermark W ' (j);

3) video to be measured is carried out overall three-dimensional dct transform; In the DCT coefficient, the visual feature vector V ' that goes out video data to be measured according to the symbol extraction of Low Medium Frequency coefficient (j);

4) utilize Hash function character to extract watermark,

W (j) and W ' (j) are carried out normalizated correlation coefficient calculating, confirm the ownership of video data.

Images