CN101540898B - AVS video digital watermark method based on nonnegative matrix decomposition - Google Patents

Abstract

The invention discloses an AVS video digital watermark method based on nonnegative matrix decomposition, which mainly solves the problems of low transmittance and poor attacking repellence to frame deletion and frame recomposition. The invention generates watermark based on content by utilizing light characteristics of a video I frame and extracts kinetic characteristics of video signals by utilizing the nonnegative matrix decomposition during the watermark embedding, and can self-adaptively adjust the embedding strength and the embedding position of the watermark by combining texture and light characteristics of the video to embed the watermark into intermediate frequency DCT coefficients of prediction residual errors of an I frame light block. During the watermark verification, the I frame is only decoded, the video product is judged whether the watermark is embedded or not by comparison of the extracted watermark and the verification of the error rate of the watermark. The invention has high transmittance and robustness, can be used in the video product with AVS standard, and provides reliable copyright protection for the AVS video product.

Description

AVS video digital watermark method based on the nonnegative matrix decomposition

Technical field

The invention belongs to video, multimedia signal processing technique field, be specifically related to the method for video digital watermark, can be used for AVS normal video product, for the copyright of AVS video product provides protection.

Background technology

The video digital watermark technology is focus and the difficult point in the current digital watermark research direction.Focus is, the release of a large amount of consumer digital audio/video products makes with the digital watermarking to be that the market demand of digital product property right protection technology of important component part is more urgent.Difficult point is, because the characteristics that video is own, require people must seek new special technique and come video is operated, for example: the real-time requirement of data scale that video is huge and video watermark has determined video can't use the Image Watermarking Technique of many maturations; On time-domain, because of the more accurate human vision system model that comprises characteristics such as temporal masking effect is not set up as yet fully, technical know-how is studied the comparison difficulty, adds the restriction of concrete video coding compressed format, the easier limitation that causes technology to use; And because some at the new attack form of video watermark such as the appearance of frame deletion, interframe collusion etc., are different from unique requirement of rest image watermark for video watermark has proposed some

The video digital watermark method that has proposed at present has:

1.Lino E Coria, Mark Pickering, Panos Nasiopoulos, Rabab K.Ward, " A VideoWatermarking Scheme Based on the Dual-Tree Complex Wavelet Transform; " IEEETransactions on Information Forensics and Security, 2008,3 (3): 466-474, this method is carried out the decomposition of multistage antithesis tree Complex wavelet to each frame of original video, watermark is embedded in the low frequency sub-band after the decomposition, can be good at resisting geometric attacks such as rotation, convergent-divergent, shearing.This method belongs to the forward type video watermark process, can cause bulk information to lose behind the compressed encoding, can not resist video compression and handle.

2.Jing Zhang, A.T.S.Ho, G Qiu, et al.Robust Video Watermarking of H.264/AVC, IEEE Trans on Circuits and Systems-II:Express Briefs, 2007,54 (2): 205-209, this method is utilized spread spectrum, watermark is embedded in the DCT coefficient of video coding prediction residual, adjusts embedment strength in conjunction with the human-eye visual characteristic self adaptation simultaneously.Only utilize the static masking characteristics of human eye during this method construct human visual system model, do not considered the distinctive temporal masking effect of video.

3.Alper Koz, A.Aydin Alatan, Oblivious Spatio-Temporal Watermarking of DigitalVideo by Exploiting the Human Visual System.IEEE Trans.Circuits Syst.Video Techn.2008,18 (3): 326-337, this method is by calculating the human eye vision time contrast limit, self adaptation is determined watermark embed strength, though this method has been considered the temporal masking effect of video, but calculation of complex, real-time is relatively poor.

4. Hu Yang, Zhang Chuntian, Su Yuting is based on Video Information Hiding Algorithms H.264/AVC, electronic letters, vol, 2008,4 (4): 690-694, this method realizes that by the intra prediction mode of modulation I frame 4 * 4 luminance block watermark embeds, embedded location is determined in conjunction with key by each piece own characteristic, this method has the transparency and fail safe preferably, but poor robustness can not be resisted processing and attack at video.

In sum, existing video digital watermark has following deficiency: 1) video watermark process should combine with concrete coding standard, and existing video watermark process is applicable at present that mostly based on MPEG or H.26x serial compressed encoding standard the video watermark process of AVS standard almost is in blank.2) poor robustness can not be resisted the distinctive attack of some videos, for example: attacks such as frame deletion and frame reorganization; 3) only utilize the static masking characteristics of human eye mostly, do not considered the distinctive temporal masking effect of video.

Summary of the invention

The present invention seeks to deficiency at above-mentioned prior art; a kind of AVS video digital watermark method that decomposes based on nonnegative matrix is proposed; improving watermark opposing frame deletion and the distinctive attacking ability of frame reorganization video, for the copyright of AVS digital product provides reliable protection.

The key problem in technology of realizing the object of the invention is: in watermark embed process, utilize the brightness of video I frame to generate content-based watermark, utilize nonnegative matrix to decompose the motion feature that extracts vision signal, and in conjunction with the texture and the brightness of video, the embedment strength of adaptive adjustment watermark and position are embedded in watermark in the intermediate frequency DCT coefficient of I frame luminance block prediction residual.In the watermark verification process, only need decoding I frame, by the watermark of relatively extracting and the similitude of checking watermark, thereby judge whether video is embedded into watermark.Concrete scheme is as follows:

One, the AVS video digital watermark embedding grammar that decomposes based on nonnegative matrix comprises following process:

(1) to original video sequence divide into groups successively, dimension-reduction treatment and nonnegative matrix resolution process, and extract the motion feature of the sub-piece of each 8 * 8 brightness of I frame according to the result that nonnegative matrix is decomposed;

(2) extract the brightness Lum (k) of the sub-piece of I each 8 * 8 brightness of frame from original video sequence, and generate content-based watermark W (k) according to the brightness that extracts;

(3), extract the textural characteristics of the sub-piece of each 8 * 8 brightness of I frame according to the average coding cost of AVS infra-frame prediction;

(4), calculate the mixing visual masking coefficient of k sub-piece according to the motion feature, brightness and the textural characteristics that extract:

$M\left(k\right)=\mathrm{round}(\mathrm{\α}\·(1+\frac{\mathrm{Lum}\left(k\right)-90}{165})+\mathrm{\β}\·\frac{\mathrm{Mov}\left(k\right)}{\mathrm{Lum}\left(k\right)}+\mathrm{\η}\·\mathrm{TextMask}\left(k\right))$

In the formula, α, β and η are that visual signature is sheltered weighted factor, and Mov (k) is k the motion feature that sub-piece extracts, and Lum (k) is k the brightness that sub-piece extracts, TextMask (k) is k the textural characteristics that sub-piece extracts, and round represents bracket function;

(5) watermark W (k) is embedded in the intermediate frequency DCT coefficient of I frame k the sub-piece intra prediction of 8 * 8 brightness, and utilizes and mix the embedment strength that visual masking coefficient M (k) adjusts watermark.

(6) with reference to the AVS video encoding standard watermarked DCT coefficient is carried out entropy coding, output contains the AVS video code flow of watermark.

Two, the AVS video digital watermark extracting method of decomposing based on nonnegative matrix comprises following process

1) the AVS video compression coding stream that contains watermark is carried out partial decoding of h and handle the I frame of only decoding all;

2) extract watermark W ' in the residual error DCT coefficient of the sub-piece of each that from decode procedure, obtains 8 * 8 brightness;

3) mean flow rate of the sub-piece of calculating decoding back I each 8 * 8 brightness of frame, and by mean flow rate generation checking watermark W ";

4) watermark W ' that relatively extracts and checking watermark W ", calculate between them error rate e, setting threshold T is if e＜T proves that then this video has been embedded into watermark, otherwise then is not embedded into watermark.

This method has the following advantages:

1) nonnegative matrix is decomposed as a kind of low-rank approach method, can reduce the dimension of data characteristics greatly, has effectively saved storage and computational resource, thus the motion feature that from large-scale video data, extracts video that can be real-time;

2) according to the adaptive adjustment watermark embed strength of video motion characteristic, brightness and textural characteristics that extracts, improved the transparency of watermark;

3) the content-based watermark that generates by the luminance video feature, watermark changes with frame of video, satisfies the requirement of statistics invisibility, has improved the fail safe of watermark;

Therefore 4) this method has adopted I frame independently watermark embedding and testing mechanism, as long as there is an I frame not to be destroyed, just can detect watermark accurately, can improve the ability of watermark opposing frame deletion, frame reorganization video peculiar attack.

5) by analyzing the AVS encoding-decoding process, propose a kind of video watermark process, can provide reliable protection for the copyright of AVS digital product towards the AVS standard.

Description of drawings

Fig. 1 is a watermark embed process block diagram of the present invention;

Fig. 2 is a watermark extraction process block diagram of the present invention;

Fig. 3 is the sectional drawing of I frame before and after the test video watermark embeds.

Embodiment

One. the basic theory introduction

1. nonnegative matrix is decomposed

1999, Lee and Seung delivered the nonnegative matrix decomposition method first on " Nature ", were called for short NMF.This method is that all elements is under the non-negative condition a kind of matrix disassembling method to its realization in matrix, institute after decomposing simultaneously is important also to satisfy the nonnegativity restriction, from the viewpoint of algebraically, NMF is non-negative algebraically decomposed form in a kind of inherence or a technique of expression of finding data.

NMF has following characteristics: on the one hand, the low-rank that obtains former data matrix by matrix decomposition approaches, thereby can find its immanent structure feature, and can reduce the dimension of data characteristics greatly, improved operation efficiency, storage and computational resource have effectively been saved, for large-scale data is provided by the otherwise effective technique approach that provides; On the other hand, do not occur negative value among the result of decomposition, have interpretation and clear physical meaning, because can not have negative composition in many physical signallings; The nonnegativity restriction also makes the expression mode have certain sparse property in addition, and sparse relatively expression mode can suppress partial occlusion to a certain extent, light changes and the external worlds such as rotation of object change the adverse effect of bringing to feature extraction.The psychology of NMF and physiological structure are according to being human brain to the perception of things integral body by its part perception additivity is constituted, it provides a kind of thinking for the relation between object analysis local feature and the global feature, be that global feature is the non-negative wire combination of local feature, this non-negative resolution characteristic more conforms with the experience directly perceived of human visual perception, so NMF has intelligent.

Along with further investigation to NMF, NMF is constantly improving and is developing always, become one of multidimensional data handling implement most popular in the fields such as signal processing, biomedical engineering, pattern recognition, computer vision and Image Engineering gradually, be successfully applied at present in the researchs such as the detection of people's face and identification, music signal analysis separate with classification, blind source.

2.AVS standard

The AVS standard be based on China's innovative technology and part public technology from primary standard, code efficiency is than the high 2-3 of MPEG-2 doubly, and is suitable with MPEG-4; And technical scheme is succinct, and the chip implementation complexity is low, has reached the highest level of second generation standard; The AVS standard has solved the fast knot of MPEG-4 patent grant problem by succinct one-stop permission policy, is the international standard of open formulation, is easy to promote; In addition, the AVS standard is that a cover comprises system, video, audio frequency, medium copyright management in interior complete standards system, for the digital audio/video industry provides more comprehensive solution.In sum, the AVS standard is the choosing of going up of second generation information source standard.

Two, related symbol explanation

B: two-dimentional nonnegative matrix to be decomposed;

U: decompose the basic matrix that obtains by the B nonnegative matrix;

H: decompose the coefficient matrix that obtains by the B nonnegative matrix;

S (j): the component motion of j pixel of I frame;

Mov (k): the motion feature that the sub-piece of k brightness of I frame extracts;

Lum (k): the brightness that the sub-piece of k brightness of I frame extracts;

W (k): the content-based watermark that the sub-piece of k brightness of I frame generates;

TextMask (k): the textural characteristics that the sub-piece of k brightness of I frame extracts;

M (k): the mixing visual masking coefficient of k the sub-piece of brightness of I frame;

D (k): k not watermarked intermediate frequency coefficient of the sub-piece of brightness of I frame;

D ' is (k): the intermediate frequency coefficient after k sub-piece of brightness of I frame is watermarked;

W ' is (i): the watermark of extracting in k sub-piece of brightness of I frame;

Lum ' is (k): the mean flow rate of k the sub-piece of brightness of decoded I frame;

W " (k): the checking watermark that the sub-piece of k brightness of decoded I frame generates;

E: extract the error rate between watermark and the checking watermark.

Three, decompose the AVS video watermark process based on nonnegative matrix

With reference to Fig. 1, digital watermark embed process of the present invention is as follows:

Step 1 is extracted video motion characteristic

(1.1) original video sequence being carried out packet transaction, guarantee that each group has and have only an I frame, each frame of video group is carried out one dimension launch, be two-dimentional nonnegative matrix with the 3 D video data conversion;

(1.2) nonnegative matrix that obtains after grouping, the dimension-reduction treatment being carried out nonnegative matrix according to following iterative formula decomposes:

$U_{\mathrm{ia}}=U_{\mathrm{ia}}\frac{\underset{l}{\mathrm{\Σ}}\frac{H_{\mathrm{al}}{B}_{\mathrm{il}}}{{\left(\mathrm{UH}\right)}_{\mathrm{il}}}}{\underset{k}{\mathrm{\Σ}}{H}_{\mathrm{ak}}}$

$H_{\mathrm{aj}}=H_{\mathrm{aj}}\frac{\underset{l}{\mathrm{\Σ}}\frac{U_{\mathrm{la}}{B}_{\mathrm{lj}}}{{\left(\mathrm{UH}\right)}_{\mathrm{lj}}}}{\underset{k}{\mathrm{\Σ}}{U}_{\mathrm{ka}}}$

$U_{\mathrm{il}}=1,U_{\mathrm{ia}}=\frac{U_{\mathrm{ia}}}{\underset{l}{\mathrm{\Σ}}{U}_{\mathrm{la}}}$

In the formula, B represents two-dimentional nonnegative matrix to be decomposed, B _IlThe element of capable the 1st row of representing matrix B i, U are represented the basic matrix that the decomposition of B nonnegative matrix obtains, U _IaThe element of the capable a row of representing matrix U i, H represents the coefficient matrix that the decomposition of B nonnegative matrix obtains, H _AjThe element of the capable j row of representing matrix H a;

(1.3) decomposition obtains coefficient matrix H according to nonnegative matrix, is calculated as follows the component motion of each pixel of I frame:

$S\left(j\right)=\underset{a=2}{\overset{T}{\mathrm{\Σ}}}{H}_{\mathrm{aj}}$

In the formula, the component motion of j pixel of S (j) expression I frame, H _AjThe element of the capable j row of representing matrix H a, T represents the line number of coefficient matrix H,

Because first base vector to basic matrix in the nonnegative matrix decomposable process has added the static constraints of pressure, make each element of this vector all equate, therefore first row that decomposes the coefficient matrix H that obtains has been represented the stationary component of video, and other has then represented the component motion of video;

(1.4) component motion of all pixels of each 8 * 8 luminance block of I frame is sued for peace, extract the motion feature of luminance block.

Step 2 is extracted the luminance video feature, generates content-based watermark.

(2.1) be calculated as follows the mean flow rate of each 8 * 8 luminance block of I frame, as the brightness of the luminance block of extracting;

$\mathrm{Lum}\left(k\right)=\frac{1}{64}\underset{i=1}{\overset{8}{\mathrm{\Σ}}}\underset{j=1}{\overset{8}{\mathrm{\Σ}}}E(k,i,j)$

In the formula, Lum (k) is k the brightness that the sub-piece of brightness extracts, and (k, i j) are the pixel value of I frame k the capable j row of 8 * 8 luminance block i to E;

(2.2) press following formula according to the brightness that extracts and generate content-based watermark.

In the formula, W (k) is k the content-based watermark that the sub-piece of brightness generates, and Lum (k) is k the brightness that the sub-piece of brightness extracts.

Step 3 is extracted the video texture feature.

The rest image texture characteristic extracting method is more complicated all generally, can not satisfy the real-time requirement of video watermark, and the present invention proposes a kind of quick texture characteristic extracting method based on the AVS video encoding standard, and process is as follows:

(3.1) to all 8 * 8 luminance block of I frame, with reference to the AVS video encoding standard, extract its coding cost under 5 kinds of intra prediction modes, the coding cost under every kind of pattern has been reacted the textural characteristics on the sub-piece different directions;

(3.2) calculate the textural characteristics of the mean value of described 5 coding costs as luminance block:

$\mathrm{TextMask}\left(k\right)=\frac{1}{5}\underset{m=1}{\overset{5}{\mathrm{\Σ}}}\cos t(k,m)$

In the formula, (k m) is the coding cost of k luminance block of I frame under m kind predictive mode to cost.

Step 4 is calculated and is mixed the visual masking coefficient.

The human visual system has brightness masking effect, texture masking effect and temporal masking effect, therefore in conjunction with described this three-type-person's class vision system characteristic, according to the video motion characteristic, brightness and the textural characteristics that extract, be calculated as follows the mixing visual masking coefficient of k sub-piece:

$M\left(k\right)=\mathrm{round}(\mathrm{\α}\·(1+\frac{\mathrm{Lum}\left(k\right)-90}{165})+\mathrm{\β}\·\frac{\mathrm{Mov}\left(k\right)}{\mathrm{Lum}\left(k\right)}+\mathrm{\η}\·\mathrm{TextMask}\left(k\right))$

In the formula, α is that brightness is sheltered weighted factor, β is that motion feature is sheltered weighted factor, η is that textural characteristics is sheltered weighted factor, Mov (k) is k the motion feature that sub-piece extracts, Lum (k) is k the brightness that sub-piece extracts, and TextMask (k) is k the textural characteristics that sub-piece extracts, and round represents bracket function.

Step 5, watermarked.

The AVS video coding adopts intra-frame encoding mode to the I frame, data are more stable, if the destroyed then video quality of I frame can be subjected to very big influence, therefore the present invention selects the I frame watermarked, simultaneously for reducing the influence of watermark embedding to video quality, watermark is embedded on the intermediate frequency DCT coefficient of I frame 8 * 8 luminance block prediction residual, embedding formula is as follows:

In the formula, D (k) is k the not watermarked intermediate frequency coefficient of the sub-piece of brightness, and D ' is the intermediate frequency coefficient of k the sub-piece of brightness after watermarked (k).

Step 6, entropy coding.

With reference to the AVS video encoding standard watermarked DCT coefficient is carried out entropy coding, output contains the AVS video code flow of watermark.

With reference to Fig. 2, digital watermarking leaching process of the present invention is as follows:

Steps A, decoding I frame.

With reference to AVS video decode standard, the AVS code stream that contains watermark is carried out partial decoding of h, only decoding I frame wherein.

Step B extracts watermark.

Because watermark is to be embedded on the intermediate frequency DCT coefficient of I frame 8 * 8 luminance block prediction residual, presses following formula and extract watermark:

In the formula, W ' (i) represents the watermark of extracting from k the sub-piece of 8 * 8 brightness, and D ' (k) represents to contain the intermediate frequency DCT coefficient of the luminance block prediction residual of watermark.

Step C generates the checking watermark.

(1) be calculated as follows the mean flow rate of the sub-piece of each 8 * 8 brightness of decoded I frame:

${\mathrm{Lum}}^{\′}\left(k\right)=\frac{1}{64}\underset{i=1}{\overset{8}{\mathrm{\Σ}}}\underset{j=1}{\overset{8}{\mathrm{\Σ}}}{E}^{\′}(k,i,j)$

In the formula, Lum ' is the mean flow rate of k the sub-piece of brightness (k), and (k, i j) are the pixel value of decoded I frame k the capable j row of 8 * 8 luminance block i to E ';

(2) press following formula and generate the checking watermark:

In the formula, Lum ' is the mean flow rate of decoded I frame k the sub-piece of 8 * 8 brightness (k), W " (k) is decoded I frame k the checking watermark that the sub-piece of 8 * 8 brightness generates.

Step D, watermark verification.

Relatively the watermark of extracting in each I frame and the checking watermark of generation are calculated as follows the error rate e between them, and setting threshold T=0.4 if e＜T illustrates that then this frame of video has been embedded into watermark, otherwise then is not embedded into watermark,

$e=\underset{k=1}{\overset{L}{\mathrm{\Σ}}}[W^{\′}\left(i\right)W^{\′\′}\left(k\right)/L]\×100\%$

In the formula, L is a watermark length.

Effect of the present invention can further specify by following experiment simulation:

1. simulated conditions

The test video sequence of experimental selection is the akiyo video and the suzie video of QCIF form, and video length 88 frames, experiment software environment are visual c++ 6.0.For verifying the performance of the video watermark process that this paper proposes, several tests below having carried out respectively:

(1) transparency test: test the influence of watermark embedding to the video visual quality according to human eye subjective feeling and objective Y-PSNR PSNR;

(2) code check constancy test: video codes rate variations situation before and after the test watermark embeds;

(3) real-time test: the test watermark embeds the influence of extracting the coding and decoding video time;

(4) robustness test: designed a series of attack tests, comprise that weight contracts, the spiced salt add make an uproar, Gauss add make an uproar, attacks such as shearing in the Gauss's low-pass filtering, mean filter, Wiener filtering, frame, frame reorganization, frame deletion, with the robustness of method of testing.

2. simulation result

The simulation experiment result is respectively shown in Fig. 3, table 1, table 2, table 3 and table 4, wherein:

Fig. 3 is transparent subjective testing result.Fig. 3 a is the sectional drawing of test video akiyo I frame when not watermarked, Fig. 3 b is the sectional drawing of the watermarked back of test video akiyo I frame, Fig. 3 c is the sectional drawing of test video suzie I frame when not watermarked, Fig. 3 d is the sectional drawing of the watermarked back of test video suzie I frame, does not have obvious distortion by watermarked as can be seen rear video quality among Fig. 3.

Table 1 is transparent objective examination result, and the watermark Y-PSNR that embeds the rear video coding has not only descended 0.39 and 0.46dB respectively than watermarked as can be seen from Table 1, and the visually-perceptible quality of video is not had obvious influence.

The Y-PSNR result data of video coding before and after table 1 watermark embeds

Table 2 is a code check constancy test result, and the watermark bit rate that embeds latter two test video has only increased by 7.53% and 5.85% as can be seen from Table 2, and this method has code check constancy preferably.

The bit rate of video coding before and after table 2 watermark embeds

Table 3 is the real-time test result, and watermark scramble time of embedding latter two test video has increased by 2.63% and 1.47% as can be seen from Table 3, and decode time has only increased by 5.15% and 5.23%, and this method has good real-time performance.

The used time of coding and decoding video before and after table 3 watermark embeds

Table 4 is the robustness test result, the present invention is for other various attack except that mean filter is attacked as can be seen from Table 4, the error rate of extracting watermark all can correctly detect watermark all less than 0.4, therefore can prove that this is a kind of digital video watermarking method of strong robust.

Watermark extracting result data after table 4 various attack

Claims (2)

1. AVS video digital watermark embedding grammar that decomposes based on nonnegative matrix comprises following process:

(1) to original video sequence divide into groups successively, dimension-reduction treatment and nonnegative matrix resolution process, and extract the motion feature of the sub-piece of each 8 * 8 brightness of I frame, wherein carrying out according to the following procedure of step (1) according to the result that nonnegative matrix is decomposed:

(1a) original video sequence being carried out packet transaction, guarantee that each group has and have only an I frame, each frame of video group is carried out one dimension launch, be two-dimentional nonnegative matrix with the 3 D video data conversion;

(1b) nonnegative matrix that obtains after grouping, the dimension-reduction treatment being carried out nonnegative matrix according to following iterative formula decomposes:

Wherein, B represents two-dimentional nonnegative matrix to be decomposed, B _IlThe element of the capable l row of representing matrix B i, U represents the basic matrix that the decomposition of B nonnegative matrix obtains, U _IaThe element of the capable a row of representing matrix U i, H represents the coefficient matrix that the decomposition of B nonnegative matrix obtains, H _AjThe element of the capable j row of representing matrix H a;

(1c) decomposition obtains coefficient matrix H according to nonnegative matrix, is calculated as follows the component motion of each pixel of I frame:

Wherein, the component motion of j pixel of S (j) expression I frame, H _AjThe element of the capable j row of representing matrix H a, T represents the line number of coefficient matrix H;

(1d) component motion of all pixels of each 8 * 8 luminance block of I frame is sued for peace, extract the motion feature of luminance block;

(2) extract the brightness Lum (k) of the sub-piece of I each 8 * 8 brightness of frame from original video sequence, and generate content-based watermark W (k), wherein carrying out according to the following procedure of step (2) according to the brightness that extracts:

(2a) be calculated as follows the mean flow rate of each 8 * 8 luminance block of I frame, as the brightness of the luminance block of extracting;

Wherein, Lum (k) is k the brightness that the sub-piece of brightness extracts, and (k, i j) are the pixel value of I frame k the capable j row of 8 * 8 luminance block i to E;

(2b) press following formula and generate content-based watermark according to the brightness that extracts;

Wherein, W (k) is k the content-based watermark that the sub-piece of brightness generates, and Lum (k) is k the brightness that the sub-piece of brightness extracts;

(3) according to the average coding cost of AVS infra-frame prediction, extract the textural characteristics of the sub-piece of each 8 * 8 brightness of I frame, wherein step (3) is carried out according to the following procedure:

(3a) to all 8 * 8 luminance block of I frame, extract its coding cost under 5 kinds of intra prediction modes in the AVS video encoding standard;

(3b) calculate the textural characteristics of the mean value of described 5 coding costs as luminance block:

Wherein, (k m) is the coding cost of k luminance block of I frame under m kind predictive mode to cost;

(4), calculate the mixing visual masking coefficient of k sub-piece according to the motion feature, brightness and the textural characteristics that extract:

Wherein, α shelters weighted factor for the visual brightness feature, β shelters weighted factor for the visual movement feature, η shelters weighted factor for the visual texture feature, Mov (k) is k the motion feature that sub-piece extracts, Lum (k) is k the brightness that sub-piece extracts, and TextMask (k) is k the textural characteristics that sub-piece extracts, and round represents bracket function;

(5) watermark W (k) is embedded in the intermediate frequency DCT coefficient of I frame k the sub-piece intra prediction of 8 * 8 brightness, and utilizes and mix the embedment strength that visual masking coefficient M (k) adjusts watermark;

(6) with reference to the AVS video encoding standard watermarked DCT coefficient is carried out entropy coding, output contains the AVS video code flow of watermark.

2. AVS video digital watermark extracting method of decomposing based on nonnegative matrix comprises following process:

(1), the AVS video compression coding stream that contains watermark is carried out partial decoding of h handle, only decoding I frame wherein with reference to AVS video decode standard;

(2) extract watermark W ' by following formula in the residual error DCT coefficient of the sub-piece of each that from decode procedure, obtains 8 * 8 brightness:

Wherein, W ' (k) represents the watermark of extracting from k the sub-piece of 8 * 8 brightness, and D ' (k) represents to contain the intermediate frequency DCT coefficient of the luminance block prediction residual of watermark;

(3) generate checking watermark W ", carrying out according to the following procedure of step (3) wherein:

(3a) be calculated as follows the mean flow rate of the sub-piece of each 8 * 8 brightness of decoded I frame:

Wherein, Lum ' is the mean flow rate of k the sub-piece of brightness (k), and (k, i j) are the pixel value of decoded I frame k the capable j row of 8 * 8 luminance block i to E ';

(3b) press following formula and generate the checking watermark:

Wherein, Lum ' is the mean flow rate of decoded I frame k the sub-piece of 8 * 8 brightness (k), W " (k) is decoded I frame k the checking watermark that the sub-piece of 8 * 8 brightness generates;

(4) watermark W ' that relatively extracts and checking watermark W ", be calculated as follows between them error rate e, setting threshold T=0.4, if e＜T proves that then this video has been embedded into watermark, otherwise then is not embedded into watermark,

Wherein, L is a watermark length.

Images