CN106303696A - H.264 video content authenticating water-mark method based on pivot information distribution ratio stability - Google Patents

Abstract

The present invention discloses a kind of H.264 video content authenticating water-mark method based on pivot information distribution ratio stability, including: the Y-component of frame each to original video carries out 2 grades of DWT conversion of piecemeal, and set up data block characteristics code at transform domain based on key, feature watermark is embedded into based on key intersection the low frequency sub-band of other block；When watermark embeds, use the mode quantifying to adjust the first pivot information distribution ratio；The watermark of feature with corresponding recovery by comparing reconstruct during certification, and combine neighborhood characteristics to judge the verity of data block contents.The present invention is in safety, the transparency, meet functional need to aspects such as the impacts of code check, and under the normal signal treatment conditions of H.264/AVC second-compressed and period, can effectively distinguish video and keep the process of content and maliciously distort, and positioning tampering position.

Description

H.264 video content authenticating water-mark based on pivot information distribution ratio stability Method

Technical field

The present invention relates to information security field, especially relate to a kind of based on pivot information distribution ratio stability H.264 video content authenticating water-mark method.

Background technology

Along with developing rapidly and the appearance of various Video editing software of multi-media processing technology, video data can be by easily Distort and forge so that difficulty is distinguished between truth and false.Relating to the fields such as court's proof, Medical Authentication, military information, national security Time, distorting of video information is likely to result in inestimable consequence.The most effectively differentiate the authenticity questions of digital video content Obtain paying close attention to widely.

The technology being presently used for digital video content certification mainly has perception Hash and digital watermark technology.Due to video When content authentication, there are needs and individually transmit the overhead problem of summary info, and can not position in perception Hash Distorting the particular location of content, application has significant limitation.Use the authentication techniques of half fragile video watermark, for any guarantor Hold the operation of content, such as lossy compression method, form conversion, denoising etc., be then considered as the acceptable amendment that means no harm, and to content Malice distort sensitivity.

In the last few years, the document about video watermark was reported the most, and the overwhelming majority is towards H.264/AVC standard , but these algorithms are after only experiencing normal signal process or only weight contracts, watermark can be survived, but is experiencing conventional letter successively Number process and weight contracting after, then detect failure.Additionally, in actual applications, in order to reach video is usurped, the malice of content Distort, the signal processing that means no harm, reduce the purpose such as bandwidth, need video is carried out second-compressed, i.e. decoding, signal processing, again Encoding operation, and revise GOP parameter most probably when second time compression coding, and after experiencing the process of this process, video Content (semantic) does not changes.

Commonly referred to as video experiences the operation of twice change ginseng compression coding and the process of the normal signal of period is " second-compressed ". It is obvious that Video Watermarking Scheme based on encoding domain and code stream territory is not applied for the application scenarios of this second-compressed, reason It is that the frame type of I, B, P frame when video watermark embeds and detected video can not be set up corresponding relation, i.e. watermark and embed Asynchronous with detection.Therefore, video content level identifying algorithm to be designed for second-compressed can only consider to be placed on the former of video Beginning territory, this also robustness to algorithm propose very high request.Each frame in view of video original domain is still image, but because of Different at the aspect such as compression coding, performance requirement of image and video, about digital picture content-level authenticating water-mark algorithm not Video field can be introduced directly into.Up till now, the video watermark document that second-compressed attacks can be resisted and have no report, study towards H.264/AVC the video content level identifying algorithm of second-compressed, it has also become one, this field problem demanding prompt solution.

Summary of the invention

The invention discloses a kind of H.264 video content authenticating water-mark side based on pivot information distribution ratio stability Method, under H.264 secondary becomes ginseng compression and the normal signal treatment conditions of period, can effectively distinguish the operation keeping contents processing With malice Tampering attack, and realize tampering location.

Concrete technical scheme of the present invention is as follows:

A kind of H.264 video content authenticating water-mark method based on pivot information distribution ratio stability, including:

Frame Y-component each to original video carries out piecemeal, and each sub-block performs two grades of DWT conversion and low-frequency data block SVD converts, based on key key₁Two grades of high-frequency sub-band of the different sub-blocks controlled, set up each sub-block by energy relationship Condition code；

Low frequency sub-band in each each sub-block embeds the feature watermark from other sub-block based on key intersection, embedding in watermark During entering, if data block is stand growth model or minute surface data block, first do a little pretreatment, it is ensured that data block at least 2 unusual Value；When the first singular value contribution rate is modified, following condition should be met simultaneously:

1) the first singular value contribution rate have to be larger than the second singular value contribution rate；

2) adjust the contribution rate that remaining singular value is corresponding so that the relative scale between singular value contribution rate after adjustment simultaneously Relation with keep constant before.

Further, by after before adjusting according to two grades of DWT lower frequency regions of data block, the summation of each singular values is multiplied by adjustment each very Different value contribution rate, generates new each singular value, then by SVD inverse transformation, generates the two grades of DWT low frequencies of data block containing watermark Territory, finally performs two grades of IDWT and obtains the data block containing watermark.

Meanwhile, the inverse quantization Rule Extraction data block watermark corresponding with during watermark qualified insertion is used.

The present invention when carrying out video single frame detection, the watermark extracted according to each sub-block of Y-component and corresponding data block therewith The concordance of the feature rebuild, sets up and tentatively distorts matrix T₀, then according to domain features, build on the basis of tentatively distorting matrix The vertical matrix T that distorts:

If at T₀The least T of the quantity that 9 neighborhoods are 1 of middle corresponding point_s, then in T, corresponding point are 1, are ultimately determined to be tampered；

If at T₀The quantity that 9 neighborhoods are 1 of middle corresponding point is less than T_s, then in T, corresponding point are 0, are finally judged to true Block.

Further, in definition T, the occupation rate of 1 is bit error rate BER, the judgement the most maliciously distorted as video Criterion, concrete judgment mechanism is as follows:

1) as BER≤τ₁, video can suffer from normal signal and processes and re-encoding, and content frame is true；

2) τ is worked as₁＜ BER≤τ₂, video stands normal signal and processes and re-encoding, and content frame is the most credible；

3) as BER ＞ τ₂, video suffers content tampering and re-encoding, and content frame is unreliable；

Wherein, wherein τ₁、τ₂For detection threshold.

Further, if the condition 2 met in judgment mechanism) and condition 3) time, mark with " 1 " distorted in matrix T The tampered position that bright detection frame is corresponding, as visual auxiliary judgement information.

The present invention is in safety, the transparency, meet functional need to aspects such as the impacts of code check, and H.264/AVC Under the normal signal treatment conditions of second-compressed and period, can effectively distinguish video and keep the process of content to distort with malice, and Positioning tampering position.

Accompanying drawing explanation

Fig. 1 is the theory diagram of H.264 video content authenticating water-mark method in the present invention；

The feature that represents Fig. 2 stands the stability diagram after 13 kinds of individual events attacks of twice change ginseng coding and period, wherein, Fig. 2 A () distinguishes feature steadiness figure the most under fire for 5 video segments in test object；Fig. 2 (b) represents test population The stability diagram of feature；

Fig. 3 is the stability test procedure chart of singular value contribution rate；

Fig. 4 (a) is the result of testing scheme 1；

Fig. 4 (b) is the result of testing scheme 2；

Fig. 4 (c) is the SV under twice encoding and decoding with attack condition₁Absolute increment (Δ R₁) and cumulative probability (CP) Graph of a relation；

Fig. 5 is the first singular value contribution rate quantizing rule figure；

Fig. 6 is the first singular value contribution rate inverse quantization rule schema；

Fig. 7 represents and distorts ratio p_TRespectively with robustness (P_r|T), false alarm rate (P_fr|T), false dismissed rate (P_fa|T) curve linear relationship；

Fig. 8 is video each frame variation diagram in SSIM value before and after watermark；

Fig. 9 is no-watermark video no-watermark video comparison diagram after H.264-E1 on code check；

Figure 10 is process and the Detection results figure that video keeps content；

Figure 11 distorts and Detection results figure for malice.

Detailed description of the invention

Below in conjunction with the accompanying drawings technical scheme is described in further detail.

Theory diagram as shown in Figure 1, the Y-component of first each to original video frame carries out 2 grades of DWT conversion of piecemeal, and Set up data block characteristics code at transform domain based on key, then feature watermark is embedded into based on key intersection the low frequency of other block Subband；When watermark embeds, use the mode quantifying to adjust the first pivot information distribution ratio.By comparing the spy of reconstruct during certification Levy and the watermark of corresponding recovery, and combine neighborhood characteristics to judge the verity of data block contents.

H.264 video content authenticating water-mark method in the present embodiment, specifically comprises the following steps that

Step 1: video single frames feature generates.

1.1) feature generates method；

Frame Y-component each to original video carries out piecemeal, and performs two grades of DWT, based on key key₁The difference controlled Two grades of high-frequency sub-band of sub-block, set up the condition code of each sub-block by energy relationship.

1.2) feature stability data is obtained；

For observing the condition code of the above-mentioned generation practical stability after attacking, perform following a series of tests.From mark Quasi-video library from choosing 5 CIF forms (352 × 288) video (Flower, Hall, Akiyo, Stefan, Mobile), and from Each video chosen is randomly drawed 20 frames and is combined into a test object.Test is carried out under Matlab2011b environment, choosing With the coder/decoder of H.264/AVC JM10.2, design following 13 kinds of attacks: 1-product noise (0.002), 2-Gaussian noise (0.002), 3-salt-pepper noise (0.002), 4-filtering average (5 × 5), 5-filtering intermediate value (5 × 5), 6-filtering wiener (5 × 5), 7-equalization (164), 8-the most dimmed (-30), the absolute blast of 9-(+30), 10-the most dimmed ([0,0.7]), 11-increases relatively Bright ([0.4,1]), 12-scaling big (1.8 × 1.8) and 13-scale little (0.8 × 0.8).Additionally, encode major parameter for the first time For: QP₁=24, GOP₁=IPPPPPPPI...；Second time coding major parameter is: QP₂=28, GOP₂=IBBBPBBBI....

Shown in test result Fig. 2.Fig. 2 (a) gives 5 video segments in test object and distinguishes feature the most under fire stably Implementations, scope of data is between 0.85 to 0.97, and wherein Flower video segment adds attack (7-equalization at twice coding (164) value is minimized time).The stability data block not less than 0.94 and 0.90 accounts for 60% and 87% respectively.Fig. 2 (b) reflects The stability of test population feature, scope of data is between 0.89 to 0.96, and stability is not less than 0.94 and 0.90 Account for 69.2% and 92.3% respectively；Data block characteristics is 0.9356 by the average stability that twice coding and individual event are attacked.

Step 2: video single frame watermark embeds.

2.1) stability of pivot information distribution ratio is proved；

Owing to principal component analysis (PCA) and singular value decomposition (SVD) have equivalence, by proving singular value contribution rate Stability carrys out Proof Equivalence.

2.2) verify the stability of singular value contribution rate and obtain performance data；

For the Y-component of the test every frame of object, it is divided into the sub-block (16 × 16) of non-overlapping copies, then to every height Block performs the SVD conversion of two grades of DWT conversion and low-frequency data block.Herein, with singular value contribution rate rate of changeMeasure number Singular value λ according to block A_iContribution rate stability:

${\mathrm{\ΔR}}_i^A=|R_i^{A^{\′}}-R_i^A|---\left(1\right)$

${\mathrm{VR}}_i^A={\mathrm{\ΔR}}_i^A/R_i^A---\left(2\right)$

WhereinWithWith rear 2 grades of corresponding DWT low frequency sub-band (LL before expression data block A is hacked respectively₂) i-th Singular value SV_iContribution rate；For the knots modification of i-th singular value contribution rate before and after attacking.

Test process is as it is shown on figure 3, according to whether comprise H.264 secondary coding-decoding, divide two kinds of testing schemes.Data block is subject to Each singular value SV after attack_iContribution rate rate of change VR_iWith cumulative probability CP relation as shown in Figure 4, wherein Fig. 4 (a) is testing scheme The result of 1, i.e. without H.264 encoding-decoding process, Fig. 4 (b) is the result of testing scheme 2, comprises attacking of twice coding and period Hitting, Fig. 4 (c) is calculating SV on the basis of testing scheme 2₁Absolute increment Δ R₁Curve linear relationship with CP.According to test number According to, available following main result:

(1) singular value SV_iThe stability of contribution rate reduces along with the increase of i；

(2) when without H.264 encoding, CP (VR₁≤ 0.2)=0.9902；

(3) when H.264 encoding for twice of QP Yu GOP parameter value, CP (VR are revised at the same time₁≤ 0.2)=0.8053；

CP (Δ R after (4) twice coding₁≤ 0.2)=0.9465, CP (Δ R₁≤ 0.1)=0.801.

By above-mentioned checking, show that data block singular value contribution rate has certain stability, also illustrate simultaneously based on It is feasible that the stability of the first singular value contribution rate designs the watermarking algorithm of tolerable H.264 twice change ginseng compression coding.

2.3) quantization of pivot information distribution ratio and adjustment；

Carry out modulation of equal value first pivot information distribution ratio by the contribution rate of quantitative map the first singular value and embed water Official seal breath (quantization parameter q represents).If data block is stand growth model or minute surface data block, first do a little pretreatment, unusual by first It is worth or the second singular value, or the two is modified as a numerical value the least simultaneously, it is ensured that data block at least 2 singular values.When When one singular value contribution rate is modified, following 2 conditions to be met: 1) the first singular value contribution rate have to be larger than the second singular value Contribution rate；2) contribution rate that other singular value is corresponding also adjusts accordingly so that between other singular value contribution rate after adjustment Relative scale relation with keep constant before.While realizing the watermark transparency, so guarantee the same of watermark detection and embedding Step.

If matrix A={ a_ij}_m×n, its singular value is λ₁≥λ₂≥...≥λ_r＞ 0, wherein r is the order of A.When r≤1, A For null matrix data block or minute surface data block, pretreatment need to be carried out: (1), as r=0, makes λ₁=10^-5,λ₂=10^-9；(2) r is worked as When=1, make λ₂=λ₁×10^-5.According to formula (1), each singular value contribution rate and be 1, and haveWherein r > 1.In order to A embeds 1bit feature watermark f_w, first quantify to adjust the first singular value contribution rateRevise remaining unusual value tribute the most accordingly Offer rate, thus can realize the adjustment of each singular value, finally perform SVD inverse transformation.

If quantization step is q ∈ (0,1), rightQuantization method as shown in Figure 5.Number axis is divided into two class intervals, i.e. [(k-1) q, kq) with [(2k-1) q, 2kq), wherein k is the integer not less than 1.As the f embedded_wWhen being 0,1, respectively willAdjust The whole starting point to two class intervals, is designated asAnd meet Computational methods are as follows:

Wherein floor (.) is downward floor operation.WhenTime, it is adjusted to further

Next consider the adjustment of remaining singular value contribution rate, i.e. adjustArriveStrange for reducing Different value contribution rate adjust after impact on video visual quality, method of adjustment makees following constraint: (1) each singular value contribution rate is dull Successively decrease, and belong to interval [0,1)；(2) non-zero singular value contribution rate sum keeps self-consistentency；(3) except the first singular value contribution Outside rate, the proportionate relationship between remaining each singular value contribution rate keeps self-consistentency.So adjust and can ensure that watermark embeds and inspection Survey and synchronize, and also keep before and after making the pivot information (except the first pivot) of the data block A ratio on respective distribution arrangement Unanimously.To this end, setThen the computation rule of remaining singular value contribution rate is defined as:

$R_j^{A^{\′}}=R_j^A\×factor,(j=2,3,...,r),$

WhereinFor the contribution rate that jth singular value before amendment is corresponding, factor is contribution rate zoom factor.Thus may be used , as i ≠ k, and i, k=2,3 ..., during r: (1)(2) ifThen(3) if HaveFor guaranteeingI.e.It must is fulfilled for following condition:

$R_1^{A^{\′}}>R_2^A/(1-R_1^A+R_2^A)$

IfThen with 2q as step-length, raiseUntil meeting condition, then recalculate

Finally calculate each singular value after adjusting, i.e. adjust λ_jTo λ '_j(j=1,2 ..., r):

${\mathrm{\λ}}_j^{\′}=R_j^{A^{\′}}\×\underset{i=1}{\overset{r}{\Σ}}{\mathrm{\λ}}_i$

Wherein λ_j、λ′_jSingular value before and after the amendment of corresponding data block respectively；Then obtain embedding 1bit by SVD inverse transformation The data block A ' of information.

2.4) video single frame watermark embeds；

Two grades of DWT lower frequency regions in each sub-block embed feature watermark, and feature watermark, from other block, is controlled by key key2 System, the intersection i.e. setting up feature watermark embeds mechanism.

By each singular value contribution after the summation of each singular value is multiplied by adjustment before adjusting according to two grades of DWT lower frequency regions of data block Rate, generates new each singular value, then by SVD inverse transformation, generates the two grades of DWT lower frequency regions of data block containing watermark, finally holds Two grades of IDWT of row obtain the data block containing watermark.Each data block of the Y-component of frame each to video repeats above-mentioned sub-block watermark Telescopiny, it is achieved the single frames of video watermark embeds.

Step 3: use with the generation of video single frames feature that method is identical carries out video single frames feature reconstruction.

Step 4: video single frame watermark is extracted.

Extract framework consistent with video single frame watermark embedding framework, use and watermark qualified insertion during data block watermark extracting Time corresponding inverse quantization rule.

Watermark extracting is the inverse process that watermark embeds, and this method realizes the blind property to watermark and extracts, and i.e. need not original material The participation of material.Describe i-th data block B at this_i1bit watermark extraction process:

(1) A=B is made_i,LL2, A is performed SVD and decomposes；

(2) the first singular value value contribution rate is calculated

(3) with quantization step q pairCarry out inverse quantization

$f_{w^{\′}}=\mathrm{mod}\left(round\right(R_1^{A^{\′}}/q),2)$

Wherein round (.), mod (.) respectively round up and remainder operates, and its rule is as shown in Figure 6.

Step 5: video single frame detection.

5.1) distort matrix to generate；

The watermark extracted according to each sub-block of Y-component and the concordance of feature that corresponding data block is rebuild therewith, set up preliminary Distort matrix T₀(matrix dot is that 0 expression corresponding blocks is true, is that 1 expression is distorted)；Then according to domain features, square is tentatively being distorted Set up on the basis of Zhen and distort matrix T: if at T₀The least T of the quantity that 9 neighborhoods are 1 of middle corresponding point_s, then in T, corresponding point are 1, finally It is defined as being tampered；If at T₀The quantity that 9 neighborhoods are 1 of middle corresponding point is less than T_s, then in T, corresponding point are 0, are finally judged to True block.Wherein T_sIt is to apply, according to reality, the parameter that is manually set, the property indices that influences whether to judge (false alarm rate, False dismissed rate, decision threshold etc.), it is however generally that T_s=5.

5.2) decision mechanism；

Owing to video to be detected often experienced by the operation that means no harm of signal processing and re-encoding, frame data change and are Of overall importance, 1 distribution therefore distorted in matrix T is to be uniformly distributed based on sparse, and there may be the collection of minority minor radius Middle distribution；If content frame local is maliciously distorted, 1 in matrix T there is also the integrated distribution of relatively large radius.The most straight Connect to use and distort matrix T and can not the most effectively distinguish the reasonable employment of video and distort with malice, definition bit-errors for this Rate BER (Bit Error Rate), i.e. the occupation rate of 1 in T, the decision criteria the most maliciously distorted as video.

1) as BER≤τ₁, video can suffer from normal signal and processes and re-encoding, and content frame is true；

2) τ is worked as₁＜ BER≤τ₂, video stands normal signal and processes and re-encoding, and content frame is the most credible；

3) as BER ＞ τ₂, video suffers content tampering and re-encoding, and content frame is unreliable.

Wherein τ₁、τ₂For detection threshold, depending on occurrence is applied according to reality by user；When occur 2), 3) situation time use Distort the tampered position that " 1 " in matrix T indicates that detection frame is corresponding, as visual auxiliary judgement information.

5.3) decision threshold and detection performance；

Video content to carry out malice distort, assailant must first decode the H.264 video flowing containing watermark, then may Perform normal signal process and distort vestige to cover up, finally perform to change the re-encoding of GOP Yu QP parameter simultaneously.If portraying Individual data block A_i,jRobustness after distorting (is designated as P_r|T), false alarm rate (be designated as P_fr|T), false dismissed rate (be designated as P_fa|T).If p_T For the ratio of distorting, p_T∈[0,1]。

Separately set A_i,jBefore and after keeping contents processing, its watermark is respectively p with the stability of feature_w、p_f, and with w and w_t', F and f_t' represent A respectively_i,jBefore Tampering attack with after watermark and feature.

Carrying out theory analysis by above-mentioned malice being distorted the detection of process, P can be set up respectively_r|T、P_fr|T、P_fa|TWith p_w、 p_f、p_TAnalytical Expression.

Attack faced by this method mainly includes that twice change ginseng compression coding and the multi-signal of period process, wherein Compression coding relates to numerous parameter, and process is extremely complex, and different signal processing intensity is affected by each autoregressive parameter, so It is difficult to set up p_wWith p_fAnalytical Expression.To this end, step 1.2) and step 2.2) test data based on, set up the ginseng of the two Examine value.Can obtain according to formula (2) and inverse quantization rule:

As q=0.2,

$p_w=P\{0\≤{\mathrm{\ΔR}}_1^A\≤0.1\}+P\{0.4\≤{\mathrm{\ΔR}}_1^A\≤0.5\}$

Have according to test data:

$\begin{array}{c}{p}_w=CP\left(0.1\right)+CP\left(0.5\right)-CP\left(0.4\right)\\ =0.801+1-1\\ =0.801\end{array}---\left(4\right)$

Further according to step 1.2) test data can obtain,

p_f=0.9356 (5)

According to formula (4) and (5), and the P set up_r|T、P_fr|T、P_fa|TRespectively with p_w、p_f、p_TAnalytical Expression, programmed meter Calculation can be described to distort ratio p_TRespectively with P_r|T、P_fr|T、P_fa|TCurve linear relationship, as it is shown in fig. 7, work as p_TWhen=0, i.e. keep content Process and attack, P_r|T=0.96, P_fa|T=0, P_fr|T=0.04；Work as p_TWhen=0.3, P_r|T=0.7948, P_fa|T=0.1465, P_fr|T =0.0832.

To this end, can be at q=0.2, T_sOn the premise of=5, set τ₁=0.04.As q ≠ 0.2, T_sWhen ≠ 5, according to formula (3) similar with (4) method p is recalculated_w, further according to formula (5) and the P that set up_r|TWith p_w、p_f、p_TAnalytical Expression, so may be used Calculate τ₁Reference value.Owing to keeping the process of content by Video coding, signal processing, video sample (quantity, resolution and stricture of vagina Reason) etc. the impact of complicated factor, τ₂Reference value be difficult to derive calculate, it is proposed that be arranged to τ₂=0.1.When the blind property of end detects The BER value calculated during by detection and τ₁、τ₂Relatively, can automatically current test frame be judged to truly, the most credible, unreliable it One.

Due to when distorting ratio (p_T) the least time, or keep contents processing intensity the biggest time all there may be: τ₂≤BER< τ₁, i.e. it is judged as content frame the most credible, now indicates, with " 1 " distorted in matrix T, the tamper bit that detection frame is corresponding Put, enter manual intervention.

Method in the present embodiment in safety, the transparency, the aspects such as the impact of code check are met functional need, Er Qie H.264/AVC, under the normal signal treatment conditions of second-compressed and period, can effectively distinguish video and keep process and the evil of content Meaning is distorted, and positioning tampering position.

(1) safety

Method owing to proposing is watermark based on cipher controlled intersection embedded technology, can effectively resist cryptanalysis and attack Hit, piece attack, vector quantization attack and forgery attack targetedly together.

For the main performance of method in test the present embodiment, under Matlab2011b environment integrated H.264/AVC The coder/decoder of JM10.2 is tested.Choose 5 QCIF forms (176 × 144, YUV 4:2:0) standard video sequence (News, Mobile, Silent, Coastguard, Hall) is as test object；Coding uses Baseline configuration literary composition for the first time Part, major parameter is QP1=26, GOP1=IPPPPPPI..., is designated as H.264-E1；Second time coding uses Main configuration literary composition Part, major parameter is QP2=28, GOP2=IBBBPBBBI..., is designated as H.264-E2；Other parameter uses default setting.Calculate Quantization parameter q in method is set to 0.2.Additionally, the attacking of the holding contents processing used between H.264-E1 and H.264-E2 Hit with reference to step 1.2).Mainly storing and propagate in compression coding mode due to video, this experiment is after H.264-E1 Video with/without watermark is carried out correlated performance compare.

(2) the watermark transparency

For the objective evaluation watermark impact on visual quality, use structural similarity knots modification Δ SSIM, i.e. have no-watermark Video is the difference on SSIM metric after H.264-E1.SSIM metric, in the range of 0 to 1, shows to compare during equal to 0 Object entirely different, when being 1 illustrate comparison other completely the same.

Fig. 8 provides video each frame change in SSIM value before and after watermark, the fluctuation range of Δ SSIM from-0.024 to- 0.002.It is obvious that according to the tolerance of SSIM, watermark the video visual quality introduced declines can be ignored.

(3) code check impact

As it is shown in figure 9, be given with or without the comparison on code check after H.264-E1 of watermark video no-watermark video, watermark causes Code check averagely increases by 8.41%.

(4) differentiation that the process of content is distorted is kept with malice

In many applications, user more concerned be video content or whether semanteme is changed, rather than constitute video data Digital minor alteration.Protected video is being issued the most afterwards, and the application approach that content can be kept constant can be summarized as 3 kinds below: (1) directly uses, i.e. decode broadcasting, do not do any data and process；(2) decode for reducing code check to propagate, adjust Parameter re-encodes；(3) decode according to application-specific demand, image enhaucament, adjustment parameter re-encode.

For verifying this authentication method detection performance to holding contents processing, perform following experiment.First setting procedure 5.2) two detection threshold τ related to₁=0.04, τ₂=0.1；As shown in Figure 8, table 1 gives related description and inspection to Detection results Survey result.In Fig. 10, (a1)～(a4), (b1)～(b4), the detection effect of (c1)～(c4) the most corresponding above-mentioned 3 kinds of application Really.Processing procedure in table 1 " H.264-E1+13-scale little (0.8 × 0.8)+H.264-E2 " represents that original video loads watermark After perform H.264-E1 coding/decoding, individual event signal processing (step 1.2) of numbered 13 successively), H.264-E2 become ginseng compile again Code；Additionally, " frame-B19 " and " frame-I8 " represent corresponding the 9th frame and the 8th frame that design sketch is test video respectively in table 1, And the frame type of last coding is respectively B frame and I frame.By by BER experiment value and τ₁、τ₂Relatively, can be automatically currently Test frame is judged to truly, one of the most credible, unreliable.The video corresponding with Figure 10 keeps explanation and the detection of contents processing Result is as shown in table 1, wherein, and BER_Theoretical=P_{Fr | G is theoretical}=0.04；τ₁=0.04, τ₂=0.1；

Table 1 video keeps explanation and the testing result of contents processing

Additionally, continue to distort the malice of video spatial domain content to carry out experiment, Detection results and related description are respectively such as figure 11, shown in table 2, wherein maliciously distort on the basis of above-mentioned holding contents processing ((3rd) plants application approach) and carry out, bag Include locally to distort and replace with whole frame.Owing to losing at tampered region watermark information, and that the embedding of single authentication ' unit is 1bit Watermark, so being 0.5 relative to the false dismissed rate theoretical value of local, tampered region.Add as eliminating by keeping contents processing to introduce Noise and the filtration measure that uses, simultaneously also eliminate that tampered region is identified sparse piece, thus cause relative to usurping The experiment value of the false dismissed rate changing local, district is slightly above 0.5 (such as Figure 11-(a3, b3)).As the ratio of distorting (p_T) the least time, content frame May be judged as the most credible, now to enter manual intervention process according to detection masking-out figure.

Table 2 video malice distorts explanation and testing result (τ₁=0.04, τ₂=0.1)

Tested by above two classes, demonstrate this authentication method and can effectively filter video and keep in actual applications at content The situation of reason, also illustrates a certain proportion of malice has been distorted preferable discrimination simultaneously.By Video coding, signal processing, regard Frequently the impact of the complicated factor such as sample (quantity, resolution and texture), about P in table 1, table 2_r|T、P_fa|T、P_fr|T, BER respective There is deviation in theoretical value and experiment value, but because deviation is the least, is just illustrating that the step 1 in technical scheme tests data with step 2 The practicality of the relevant performance derivation model set up on basis.

The foregoing is only the preferable implementation example of the present invention, be not limited to the present invention, all in present invention spirit and Within principle, any modification, equivalent substitution and improvement etc. made, should be included within the scope of the present invention.

Claims (6)

1. a H.264 video content authenticating water-mark method based on pivot information distribution ratio stability, it is characterised in that bag Include:

Frame Y-component each to original video carries out piecemeal, and each sub-block performs two grades of DWT conversion and the SVD of low-frequency data block Conversion, based on key key₁Two grades of high-frequency sub-band of the different sub-blocks controlled, set up the spy of each sub-block by energy relationship Levy code；

Low frequency sub-band in each each sub-block embeds the feature watermark from other sub-block based on key intersection, embeds in watermark During, if data block is stand growth model or minute surface data block, first does a little pretreatment, it is ensured that data block at least 2 singular values；When When first singular value contribution rate is modified, following condition should be met simultaneously:

1) the first singular value contribution rate have to be larger than the second singular value contribution rate；

2) adjust the contribution rate that remaining singular value is corresponding so that the relative scale relation between singular value contribution rate after adjustment simultaneously With keep constant before.

2. H.264 video content authenticating water-mark method as claimed in claim 1, it is characterised in that by according to two grades of DWT of data block Each singular value contribution rate after the summation of each singular value is multiplied by adjustment before lower frequency region adjustment, generates new each singular value, then leads to Cross SVD inverse transformation, generate the two grades of DWT lower frequency regions of data block containing watermark, finally perform two grades of IDWT and obtain the data containing watermark Block.

3. H.264 video content authenticating water-mark method as claimed in claim 2, it is characterised in that use and quantify embedding with watermark The inverse quantization Rule Extraction data block watermark of fashionable correspondence.

4. H.264 video content authenticating water-mark method as claimed in claim 3, it is characterised in that carry according to each sub-block of Y-component The concordance of the feature that the watermark that takes and therewith corresponding data block are rebuild, sets up and tentatively distorts matrix T₀, then special according to field Levy, set up on the basis of tentatively distorting matrix and distort matrix T:

If at T₀The least T of the quantity that 9 neighborhoods are 1 of middle corresponding point_s, then in T, corresponding point are 1, are ultimately determined to be tampered；

If at T₀The quantity that 9 neighborhoods are 1 of middle corresponding point is less than T_s, then in T, corresponding point are 0, are finally judged to true block.

5. the most H.264 video content authenticating water-mark method as claimed in claim 4, it is characterised in that definition distorts in matrix T 1 Occupation rate be bit error rate BER, the decision criteria the most maliciously distorted as video, concrete judgment mechanism is as follows:

1) as BER≤τ₁, video can suffer from normal signal and processes and re-encoding, and content frame is true；

2) τ is worked as₁＜ BER≤τ₂, video stands normal signal and processes and re-encoding, and content frame is the most credible；

3) as BER ＞ τ₂, video suffers content tampering and re-encoding, and content frame is unreliable；

Wherein, wherein τ₁、τ₂For detection threshold.

6. H.264 video content authenticating water-mark method as claimed in claim 5, it is characterised in that if meeting in judgment mechanism Condition 2) and condition 3) time, indicate detection tampered position corresponding to frame with distorting " 1 " in matrix T, as visual auxiliary Help judgement information.