CN104853215A - Video steganography method based on motion vector local optimality preservation - Google Patents

Abstract

The present invention relates to a video steganography method based on motion vector local optimality preservation. Criteria for motion vector local optimality and a search algorithm for optimal candidate motion vectors are established. On this basis, by constructing a reasonable motion vector disturbance cost calculation model and selecting the most efficient steganographic codes, covert channels are established to complete steganographic insertion of secret information, and an extraction process of the secrete information is thus completed. Syndrome trellis code (STC) steganographic codes are preferably adopted. The method provided by the present invention is a motion vector field video steganography method which is based on motion vector local optimality preservation, has a high insertion efficiency and can effectively resist conventional steganalysis attacks, and is especially applicable to covert communication situations with high safety requirements.

Description

Based on the video steganography method that motion vector local optimality keeps

Technical field

The present invention relates to a kind of video steganography (Video Steganography) method, be specifically related to a kind of based on the highly concealed type video steganography method of motion vector local optimality maintenance and the application in covert communications thereof, the method belongs to the sub-field of Information hiding in field of information security technology.

Background technology

Present information concealing technology mainly comprises steganography (Steganography), steganalysis (Steganalysis) and digital watermarking (Digital Watermarking).Steganography mainly study how by digital multimedia documents such as secret information embedded images, video, audio frequency to reach the object of covert communications; Steganalysis mainly adopts the method such as machine learning, pattern recognition to carry out steganography classification judgement to file to be measured.

Generally, the design of steganographic algorithm needs to consider and weigh following several respects key element:

1) not sentience: the undistinguishable for human body of the carrier before and after steganography, namely cannot only rely on the audio visual of the mankind judge that whether file to be measured is through steganography;

2) robustness: steganography file, after the noise jamming of bearing to a certain degree and multi-media processing (as compression, cutting out, size change over etc.), still can recover the secret information be wherein embedded into exactly;

3) embedding capacity: under the prerequisite ensureing not sentience and to a certain degree robustness, embed secret information as much as possible in carrier;

4) Embedding efficiency: embedding under the certain prerequisite of amount of information, reduce the amendment number of times to carrier or degree as far as possible;

5) fail safe: even if having the priori about steganographic algorithm, existing steganalysis method also cannot distinguish general carrier and the file through this algorithm steganography with sufficiently high probability.

Along with making rapid progress of Internet technology, weeding out the old and bring forth the new of the flourish and efficient video coding standard of the tertiary industry such as video request program, video calling, video relies on its abundant visual performance and powerful information propagation capabilities, just progressively replace image, become current most popular communication media and more suitably steganography carrier, therefore, video steganography also result in the extensive concern of this area scholar.

Video steganography can be divided into spatial domain (Spatial Domain) video steganography and compression domain (Compressed Domain) video steganography generally: the former is by directly revising the original pixel value of frame of video to embed secret information; Latter introduces steganography disturbance in the cataloged procedure of video, makes compressed encoding and steganography embed and can carry out simultaneously.Compressed domain video steganography is according to the difference of embedded domain, can be divided into based on motion vector (Motion Vector, MV) video steganography (list of references: C.Xu, X.Ping, and T.Zhang.Steganography in compressed video stream.In Proc.1st Int.Conf.Innov.Comput., Inf.Control, vol.1:269 – 272,2006., H.Aly.Data hiding in motion vectors of compressed videobased on their associated prediction error.IEEE Transactions on Information Forensics and Security, vol.6 (1): 14 – 18,2006., Y.Cao, X.Zhao, D.Feng, and R.Sheng.Video steganography withperturbed motion estimation.In Proc.13th Int.Conf.IH, vol.6958:193 – 207,2011., Y.Yao, W.Zhang, N.Yu, and X.Zhao.Defining embedding distortion for motion vector-based videosteganography.Multimedia Tools and Applications, Article in Press, 2014, DOI:10.1007/s11042-014-2223-8.), based on the video steganography of conversion coefficient (Transform Coefficient), based on the video steganography of intra prediction mode (Intra Prediction Mode), video steganography based on inter-frame forecast mode (Inter Prediction Mode) and this few class of video steganography based on entropy code (Entropy Coding).

Through patent consulting, in field of the present invention, existing related application situation is as follows:

(1) number of patent application be 201410233838.5 Chinese patent " video steganography method based on the disturbance of macroblock partitions mode " disclose a kind of based on H.264/AVC video encoding standard, and by amendment macroblock partitions mode to embed the video steganography method of secret information.This patented method is preferred steganography code STC (Syndrome Trellis Code in telescopiny, verification grid code) and WPC (Wet Paper Code, l Water Paper is encoded) build double-deck private communication channel, thus effectively improve embedding capacity and Embedding efficiency.This patented method belongs to the video steganography based on inter-frame forecast mode, does not relate to and revising the disturbance of motion vector, and therefore, this patent application is obviously different from basic object of the present invention, mentality of designing and specific implementation.

(2) number of patent application be 201310275143.9 Chinese patent " a kind of video steganography method based on motion vector " disclose a kind of video steganography method based on motion vector.This patented method is when the steganography of carrying out secret information embeds, first to encode current video sequence, obtain the motion vector that wherein all P frames comprise, and calculate each horizontal motion vector of gained and the steganography embedding cost of vertical component respectively according to the distortion function preset, STC is adopted to carry out embedding disturbance to the carrier that all horizontal components form again, and embed cost according to the steganography of steganography result adjustment vertical component, STC is again adopted to carry out embedding disturbance to the carrier that all vertical components form subsequently, and then according to this video sequence of steganography result recompile of horizontal and vertical component, obtain final steganography video file.The distortion function that this patented method defines has considered the spatial domain of motion vector, the residual error that relativity of time domain and estimation produce, make the motion vector be modified can keep Neighborhood Statistics characteristic as far as possible, reduce steganography disturbance to video coding performance (code check simultaneously, fidelity) impact that produces, therefore, this patented method can effective resisted movement vector field steganalysis feature MVRB and ADB_MV (list of references: Y.Cao, X.Zhao, and D.Feng.Video steganalysis exploiting motion vector reversion-based features.IEEE Signal ProcessingLetters, vol.19 (1): 35-38, 2012., Y.Su, C.Zhang, and C.Zhang.A video steganalytic algorithmagainst motion-vector-based steganography.Signal Processing, vol.91 (8): 1901-1909,2011.).But, due to the alter mode of this patented method to motion vector remain traditional " to a certain component of motion vector add one or subtract one ", i.e. (mv' _x, mv' _y) ∈ { (mv _x+ 1, mv _y), (mv _x-1, mv _y), (mv _x, mv _y+ 1), (mv _x, mv _y-1) } (wherein (mv _x, mv _y) and (mv' _x, mv' _y) represent original respectively and through the amended motion vector of steganography), therefore the local optimality (Local Optimality) of motion vector is likely destroyed, therefore the attack (list of references: K.Wang of the most effective steganalysis feature AoSO in current motion vector territory cannot be resisted, H.Zhao, and H.Wang.Video Steganalysis Against MotionVector-Based Steganography by Adding or Subtracting One Motion Vector Value.IEEETransactions on Information Forensics and Security, vol.9 (5): 741-751, 2014.).As a comparison, the angle that the present invention keeps from motion vector local optimality, devise brand-new reasonably embedding cost function and motion vector perturbation scheme, effectively can resist the attack of AoSO, it is a kind of highly concealed type communication means, therefore, this patent application is obviously different from basic object of the present invention, mentality of designing and implementation.

Summary of the invention

The object of the invention is to, provide a kind of based on motion vector local optimality keep, have higher Embedding efficiency and can effectively resist existing steganalysis attack motion vector field video steganography method.

The present invention compares other motion vector field video steganography method, establish a kind of criterion of motion vector local optimality and the searching algorithm of optimal candidate motion vector (Best Candidate MV), on this basis, by building rational motion vector disturbance cost computation model and preferred steganography code efficiently, thus set up private communication channel to complete the steganography embedding of secret information.Visible, the method that the present invention proposes is different from any video steganography method in the past, is specially adapted to the covert communications scene higher to security requirement.

According to investigation, current most motion vector field video steganography method does not adopt steganography code to improve Embedding efficiency and fail safe, moreover, all motion vector field video steganography method all cannot resist the attack of AoSO (see background technology).For above not enough and shortcoming, improvement of the present invention is as follows: first, preferably and apply suitable steganography code, not only increases Embedding efficiency and fail safe, also effectively reduces the impact that steganography disturbance causes video coding performance; Secondly, abandon traditional motion vector alter mode, consider steganography operation to the impact of fail safe and video coding performance, filter out most suitable local optimum motion vector as finally revising result by reasonable searching algorithm efficiently, thus effectively can resist the attack of the motion vector field steganalysis method such as AoSO.

Specifically, the technical solution used in the present invention is as follows:

Based on the video steganography method that motion vector local optimality keeps, wherein the telescopiny of secret information comprises the following steps:

1) preliminary treatment: in current video frame, to each motion vector (mv _x, mv _y), calculate LSB (mv _x+ mv _y) obtain binary number 0 or 1, and write all result of calculation as vector x, wherein LSB () represents that getting lowest bit position (Least SignificantBit) operates, mv _xand mv _yrepresent the horizontal and vertical component of motion vector respectively;

2) build private communication channel: using step 1) in gained vector x as the carrier of private communication channel, according to the operation principle of preferred steganography code, the parameter appointed in advance according to communicating pair generates parity check matrix H, and the motion vector disturbance cost computation model that combination is preset finally constructs the x' meeting Hx'=m, wherein m represents secret information to be embedded in this private communication channel, and x' represents the carrier vector through amendment;

3) according to step 2) can determine in current video frame, have which motion vector to need amendment;

4) according to above acquired results coding current video frame, and respectively each motion vector to be modified is replaced to corresponding optimal candidate motion vector according to the optimal candidate motion vector sought algorithm preset, and then complete the secret information of private communication channel is embedded;

5) according to the frame coded sequence of video file, to remainder every frame repeated execution of steps 1) to 4) until secret information all embeds complete, obtain final steganography video file.

Further, said method also comprises the leaching process of secret information, comprises the following steps:

1) decoding current video frame, to each motion vector (mv _x, mv _y), calculate LSB (mv _x+ mv _y) obtain binary number 0 or 1, and write all result of calculation as vector x, wherein lowest bit bit manipulation is got in LSB () expression;

2) according to the operation principle of preferred steganography code, the parameter appointed in advance according to communicating pair generates parity check matrix H, and calculate Hx, acquired results is the secret information bit of extraction;

3) according to the frame decoding order of video file, to remainder every frame repeated execution of steps 1) to 2) until secret information all extracts complete.

On such scheme basis, the present invention improves further, namely to carry out preferably the steganography code adopted.The private communication channel set up in telescopiny (Fig. 1), adopt steganography code Syndrome Trellis Code (STC, verification grid code) carry out the embedding of secret information, in leaching process (Fig. 2), the operation principle according to STC extracts secret information from this private communication channel.Being summarized as follows about STC:

If x=is (x ₁, x ₂..., x _n) ^tfor original binary carrier vector; X'=(x' ₁, x' ₂..., x' _n) ^tfor the carrier vector after steganography embeds; for arithmetic number vector and represent the element x in x _ibe modified as 1-x _icost value; H is parity matrix; M is secret information vector.Then STC can construct the x' meeting Hx'=m, and makes steganography embed cost minimum, wherein, function δ meets $\mathrm{\δ}(x,y)=\left\{\begin{array}{cc}0& (x=y)\\ 1& (x\≠y)\end{array}\right..$

In the Inter prediction module (Fig. 3) of video compression coding, can obtain motion vector corresponding to original block (Original Block) by estimation (Motion Estimation), it points to the best prediction block (Prediction Block) in reference frame.The residual block (Residual Block) obtained subsequently is after conversion (Transformation) and quantification (Quantization) operation, its result not only needs to carry out entropy code (Entropy) and obtains binary code stream, also need to obtain rebuilding residual block (Reconstructed Residual Block) through inverse quantization (Rescale) and inverse transformation (Inverse Transformation) operation, and the reconstructed block (Reconstructed Block) that final acquisition original block is corresponding.Because quantization operation is that information detracts irreversible process, therefore reconstructed block and original block are not identical, and both similarity degrees are determined by motion-estimation precision and the acting in conjunction of quantization operation step-length.

When video encoder carries out estimation to original block to be encoded, usual utilization rate aberration optimizing (Rate DistortionOptimization, RDO) compromise between model choosing comprehensively code check and the distortion factor, selects to have minimum RDO functional value and J within the scope of motion vector sought _mvthe motion vector of=D+ λ R is as optimal motion vectors corresponding to this block to be encoded, wherein, D represents the distortion factor between the prediction block of motion vectors point and original block, bit number needed for R presentation code motion vector, λ represents the parameter regulated between D and R, λ is larger, more tends to the distortion of low bit-rate height, less, more tends to high code check low distortion.

Based on the above brief introduction to inter prediction encoding overall flow and estimation, the existing criterion of motion vector local optimality, motion vector disturbance cost computation model and the optimal candidate motion vector sought algorithm proposed the present invention does following detailed description.

[1] a kind of criterion of motion vector local optimality:

For motion vector MV _0,0=(mv _x, mv _y), if its prediction block pointed to is Pred (MV _0,0), the reconstructed block obtained through inter prediction encoding is Rec (MV _0,0).Compute matrix Pred (MV _0,0) and Rec (MV _0,0) between absolute error and (Sum of AbsoluteDifference, SAD), i.e. sad _0,0=SAD (Rec (MV _0,0), Pred (MV _0,0)).In like manner, for MV _0,08 adjacent around motion vector MV _i,j=(mv _x+ i, mv _y+ j) (-1≤i, j≤1) (Fig. 4), obtain the prediction block Pred (MV of their sensings respectively _i,j), and calculate sad _i,j=SAD (Rec (MV _0,0), Pred (MV _i,j)).If sad _i,jminimum value in (-1≤i, j≤1) is sad _0,0, then motion vector MV is claimed _0,0possesses local optimality.

Note: wherein, M _i,jand N _i,jbe respectively the element being positioned at (i, j) in matrix M and N;

[2] a kind of computation model of motion vector disturbance cost:

To motion vector MV _0,0=(mv _x, mv _y) horizontal and vertical direction on 4 adjacent motion vectors and MV _-1,0=(mv _x-1, mv _y), MV _1,0=(mv _x+ 1, mv _y), MV _{0 ,-1}=(mv _x, mv _y-1) and MV _0,1=(mv _x, mv _y+ 1) (Fig. 4) carries out local optimum sex determination, if there is N (0≤N≤4) individual motion vector to possess local optimality, then to motion vector MV in them _0,0the disturbance cost of carrying out steganography embedding is:

$\mathrm{Cost}\left({\mathrm{MV}}_{\mathrm{0,0}}\right)=\frac{1}{{(N+1)}^{N+1}}\·\underset{\mathrm{MV}\∈\mathrm{\Ω}}{\max }\left(\right|J_{{\mathrm{MV}}_{\mathrm{0,0}}}-J_{\mathrm{MV}}\left|\right),$

Wherein, Ω={ MV _-1,0, MV _1,0, MV _{0 ,-1}, MV _0,1, namely the disturbance cost of motion vector depends on the quantity of local optimum motion vector in its contiguous range, and the intensity of variation of RDO functional value;

[3] a kind of searching algorithm of optimal candidate motion vector:

If according to the result that steganography embeds, need motion vector MV _0,0=(mv _x, mv _y) modify, then perform following steps successively from all candidate motion vectors, select most suitable local optimum motion vector as finally revising result:

A) respectively to MV _0,0four motion vector MV in contiguous range _-1,0=(mv _x-1, mv _y), MV _1,0=(mv _x+ 1, mv _y), MV _{0 ,-1}=(mv _x, mv _y-1) and MV _0,1=(mv _x, mv _y+ 1) local optimum sex determination is carried out;

B) by MV _-1,0, MV _1,0, MV _{0 ,-1}and MV _0,1in local optimum motion vector composition candidate motion vector set close Ψ, if Ψ non-NULL, then jump to step e), otherwise continue perform step c);

C) region of search of a motion vector delimited, its size (the candidate motion vector quantity comprised) and being directly proportional final search time, should choose according to Practical Calculation resource or ability, in addition, for any candidate motion vector MV'=(mv in this region _x', mv _y'), LSB (mv must be met _x+ mv _y) ≠ LSB (mv _x'+mv _y');

D) for each candidate motion vector in this region of search, carry out local optimum sex determination respectively, determine the motion vector of wherein all local optimums, and by they composite set Ψ;

E) for each local optimum motion vector in Ψ, calculating its RDO functional value, selecting the local optimum motion vector having minimum RDO functional value as finally revising result.

Based on above description, the video steganography method kept based on motion vector local optimality of the employing steganography code STC that the present invention proposes, wherein the telescopiny of secret information comprises the following steps (if no special instructions, following steps perform by computer):

1) in current video frame, by each motion vector (mv _x, mv _y) horizontal and vertical component summation after get lowest bit position, i.e. LSB (mv _x+ mv _y), obtain binary number 0 or 1, and by all result of calculation composition binary system carrier vector x;

2) in current video frame, for each motion vector MV, the motion vector disturbance cost computation model according to presetting calculates its steganography disturbance cost Cost (MV), and by all result of calculation composition of vector Γ;

3) build private communication channel: according to the operation principle of steganography code STC, first according to the preassigned load factor α of communicating pair and parameter matrix Η, build parity matrix Η _sTC, then construct and meet Η _sTCx'=m and make steganography embed cost D (x, x', Γ) reach minimum x', wherein, m represents secret information to be embedded in private communication channel, x' represent by private communication channel steganography embed after be modified carrier vector; If the binary number on x with x' correspondence position is not identical, then represent that corresponding original motion vector needs amendment, if identical, represent and do not need amendment;

4) according to the amendment result of carrier vector in private communication channel, for each motion vector needing amendment, the optimal candidate motion vector sought algorithm according to presetting is replaced to the minimum local optimum motion vector of RDO functional value in candidate motion vector;

5) embed with the steganography completing secret information according to above acquired results coding current video frame;

6) according to the frame coded sequence of video file, to remainder every frame repeated execution of steps 1) to 5) until secret information all embeds complete, obtain final steganography video file.

Further, the leaching process of the secret information of steganography code STC is adopted to comprise the following steps (if no special instructions, following steps perform by computer):

1) decoding current video frame, to each motion vector (mv _x, mv _y) calculate LSB (mv _x+ mv _y) obtain binary number 0 or 1, and write all result of calculation as binary vector x;

2) secret information in private communication channel is extracted: according to the operation principle of STC, according to the preassigned load factor α of communicating pair and parameter matrix build parity matrix Η _sTC, calculate Η _sTCx, acquired results is the secret information bit of extraction;

3) according to the frame decoding order of video file, to remainder every frame repeated execution of steps 1) to 2) until secret information all extracts complete.

The beneficial effect of video steganography method of the present invention to correlative technology field is as follows:

1) Embedding efficiency is high, and steganography disturbance is little.Motion vector field video steganography method nearly all at present does not all adopt steganography code to improve Embedding efficiency in telescopiny, thus effectively cannot reduce steganography and operate the disturbance that the aspects such as video coding performance are caused, make the compression efficiency of video and fidelity be subject to negative effect; By contrast, the present invention is preferred steganography code STC, thus the steganography that can complete secret information with higher Embedding efficiency embeds, in other words, under the condition that load factor is identical, the present invention will far fewer than other motion vector field video steganography method to the amendment number of times of carrier video, thus can effectively reduce steganography disturbance and keep the coding efficiency of video.

2) effectively can resist steganalysis, possess higher-security.First, present invention employs steganography code STC and improve Embedding efficiency, decrease the amendment number of times to carrier video and degree, thus effectively reduce steganography and operate the disturbance that the aspects such as video coding performance are caused, make steganography video and original video undistinguishable in visually-perceptible; Secondly, the present invention has abandoned traditional motion vector alter mode, in telescopiny, adopt default optimal candidate motion vector sought algorithm to ensure to be modified the local optimality of motion vector, thus effectively can resist the attack of the most powerful steganalysis feature AoSO in current motion vector territory.Therefore, the present invention compares other motion vector field video steganography method, has higher steganography fail safe.

3) customizable motion vector disturbance cost computation model.The present invention is embedded with the steganography of carrying out secret information by amendment motion vector, therefore needs to carry out effective evaluation to the disturbance cost of motion vector.Different evaluation mechanism may cause different steganography embed effect and have an impact to the fail safe of steganographic algorithm.The present invention, by calculating in current motion vector contiguous range the quantity of local optimum motion vector and the intensity of variation of RDO functional value to evaluate the disturbance cost of this motion vector, preferentially revises in contiguous range when steganography embeds and there is more local optimum motion vectors and RDO functional value changes those less motion vectors.Visible, the present invention can ensure that in steganography telescopiny video coding performance does not produce compared with large disturbances and motion vector to be modified preferentially can be changed to the local optimum motion vector (if existence) in its contiguous range consuming time to shorten steganography as far as possible.In addition, for different application scenarioss, the present invention can customize different motion vector disturbance cost computation models, therefore has higher flexibility.

4) by regulate region of search size with steganographic algorithm performance and embed consuming time between compromise.In the optimal candidate motion vector sought algorithm that the present invention presets, when there is not local optimum motion vector in motion vector contiguous range to be modified, a region of search exhaustive search wherein can be delimited and go out the minimum local optimum motion vector of RDO functional value as finally revising result.Generally, the area of region of search is larger, and namely the quantity of candidate motion vector is more, and the best local optimum motion vector finally obtained is just unreasonable to be thought, but also needs the search more grown consuming time.Therefore, when carrying out steganography and embedding, the demand of current application scene to time cost and fail safe can be considered, flexible modulation is carried out to make the actual compromise of laminating embedding between consuming time and algorithm performance to region of search size.

Accompanying drawing explanation

Fig. 1 is that secret information of the present invention embeds schematic diagram;

Fig. 2 is that secret information of the present invention extracts schematic diagram;

Fig. 3 is video interprediction encoding framework schematic flow sheet;

Fig. 4 is motion vector (mv _x, mv _y) and its contiguous range in relative position schematic diagram between motion vector;

Fig. 5 is that secret information of the present invention embeds flow chart;

Fig. 6 is that secret information of the present invention extracts flow chart.

Embodiment

Below by specific embodiment and by reference to the accompanying drawings 5-6 the invention will be further described.

Before employing the present invention carries out the embedding of secret information steganography, first to needing the data embedded to be encrypted, random binary data stream can be obtained.The secret information of the video steganography method based on the maintenance of motion vector local optimality that the present invention proposes embeds flow process as shown in Figure 5, and concrete operations details is as follows:

1) acquisition needs the frame of video of carrying out steganography embedding, for each motion vector (mv in this frame _x, mv _y), obtain lowest bit position and the LSB (mv of its horizontal and vertical component sum _x+ mv _y), and by all result of calculation composition binary system carrier vector x;

2) for each motion vector MV in this frame of video _0,0=(mv _x, mv _y), obtain its steganography according to the motion vector disturbance cost computation model preset and embed cost: according to the motion vector local optimality criterion preset, statistics MV _0,0four motion vector MV in contiguous range _-1,0=(mv _x-1, mv _y), MV _1,0=(mv _x+ 1, mv _y), MV _{0 ,-1}=(mv _x, mv _y-1) and MV _0,1=(mv _x, mv _y+ 1) possess the quantity N (0≤N≤4) of the motion vector of local optimality in, and comprise MV based on the calculating of rate-distortion optimization model _0,0at the RDO functional value J of interior five motion vectors _mV(MV ∈ { MV _0,0, MV _-1,0, MV _1,0, MV _{0 ,-1}, MV _0,1), and then calculate MV _0,0steganography embed cost

$\mathrm{Cost}\left({\mathrm{MV}}_{\mathrm{0,0}}\right)=\frac{1}{{(N+1)}^{N+1}}\·\underset{\mathrm{MV}\∈\mathrm{\Ω}}{\max }\left(\right|J_{{\mathrm{MV}}_{\mathrm{0,0}}}-J_{\mathrm{MV}}\left|\right),$

Wherein, Ω={ MV _-1,0, MV _1,0, MV _{0 ,-1}, MV _0,1, and the steganography of each motion vector is embedded cost composition arithmetic number vector Γ;

3) apply steganography code Syndrome Trellis Code (STC) set up private communication channel and embed secret information wherein.First the load factor α appointed in advance according to communicating pair and parameter matrix set up the parity matrix Η needed for STC _sTC, then using step 1) gained vector x is as the carrier needed for STC, and integrating step 2) gained vector Γ, application STC constructs and meets Η _sTCx'=m and make overall steganography embed cost D (x, x', Γ) minimum x', wherein m represents the secret information embedded by private communication channel, x' represents the carrier vector after steganography embeds, if the binary number on x with x' correspondence position is different, then illustrates that corresponding original motion vector needs amendment, otherwise then do not need amendment;

4) according to step 3) result, in this frame of video each need amendment motion vector MV _0,0=(mv _x, mv _y), according to the optimal candidate motion vector sought algorithm preset, it is modified: by MV _0,0four motion vector MV in contiguous range _-1,0=(mv _x-1, mv _y), MV _1,0=(mv _x+ 1, mv _y), MV _{0 ,-1}=(mv _x, mv _y-1) and MV _0,1=(mv _x, mv _y+ 1) the local optimum motion vector composition set Ψ in, if Ψ non-NULL, then selects RDO functional value is minimum in Ψ local optimum motion vector as finally revising result, if Ψ be sky, i.e. and MV _-1,0, MV _1,0, MV _{0 ,-1}, MV _0,1all do not possess local optimality, then delimit the region of search (its scale needs to determine according to Practical Calculation resource and algorithm institute allowed time cost) of a candidate motion vector and all local optimal motion vector composition in this region is gathered Ψ, and then the local optimum motion vector that in selection Ψ, RDO functional value is minimum is as finally revising result.

5) according to step 4) result to encode this frame of video, namely complete and the secret information of private communication channel embedded;

6) according to the frame coded sequence of video file, to remainder every frame repeated execution of steps 1) to 5) until secret information all embeds complete, obtain final steganography video file.

The secret information of the video steganography method based on the maintenance of motion vector local optimality that the present invention proposes extracts flow process as shown in Figure 6, and concrete operations details is as follows:

1) frame of video needing to extract secret information is obtained, for each motion vector (mv in this frame _x, mv _y), obtain lowest bit position and the LSB (mv of its horizontal and vertical component sum _x+ mv _y), and by all result of calculation composition binary vector x;

2) secret information in private communication channel is extracted.First the load factor α appointed in advance according to communicating pair and parameter matrix set up the parity matrix Η needed for Syndrome Trellis Code (STC) _sTC, re-use step 1) and gained vector x calculating Η _sTCx, result of calculation is the secret information bit of extraction;

3) according to the frame decoding order of video file, to remainder every frame repeated execution of steps 1) to 2) until secret information all extracts complete.

As can be seen from above embodiment, in embedding flow process of the present invention, according to the needs of application scenarios, both different motion vector disturbance cost computation models can have been customized to affect the compressed encoding performance (code check, fidelity etc.) of the execution speed of steganographic algorithm, fail safe and steganography video, the region of search size that can also adjust optimal candidate motion vector with algorithm performance and embed consuming time between compromise, therefore the present invention has stronger adaptability and flexibility.

In order to outstanding explanation the present invention proposition is a kind of highly concealed type motion vector field video steganography method, adopts following configuration prepare video sample collection and carry out steganalysis experiment:

1) YUV sequence: collected by the Internet and obtain the YUV420 sequence that 50 resolution are CIF (352 × 288), length is 300 frames.

2) video encoder: adopt x264 video encoder of increasing income to prepare compressed video sample, in order to reduce time overhead, arranging coding class is basic class Baseline.

3) compressed video parameter: bit rate (bitrate) is 0.5mb/s, 1mb/s, 3mb/s or 10mb/s, and frame per second is 30fps.

4) steganography parameter: the load factor (payload) of carrier video is 0, the load factor of steganography video is 0.25 or 0.5.

5) carrier video sample collection preparation: according to above configuration instruction, 50 YUV420 files are prepared 4 carrier video sample collection, be designated as 0.5mb/s_0 respectively according to the form of " bitrate_payload ", 1mb/s_0,3mb/s_0,10mb/s_0.

6) steganography video sample collection preparation: according to above configuration instruction, 50 YUV420 files are prepared 8 steganography video sample collection, be designated as 0.5mb/s_0.25 respectively, 0.5mb/s_0.5,1mb/s_0.25,1mb/s_0.5,3mb/s_0.25,3mb/s_0.5,10mb/s_0.25,10mb/s_0.5.

7) steganalysis configuration: AoSO feature is extracted to carrier video sample collection and steganography video sample collection, is divided into 8 stack features set and carries out steganalysis respectively, i.e. { 0.5mb/s_0,0.5mb/s_0.25}, { 0.5mb/s_0,0.5mb/s_0.5}, { 1mb/s_0,1mb/s_0.25}, { 1mb/s_0,1mb/s_0.5}, { 3mb/s_0,3mb/s_0.25}, { 3mb/s_0,3mb/s_0.5}, { 10mb/s_0,10mb/s_0.25}, { 10mb/s_0,10mb/s_0.5}; For above each characteristic set, the feature choosing arbitrarily wherein 40 pairs of videos is trained for grader, obtain SVMs (the Support Vector Machine based on Gaussian kernel and polynomial kernel, SVM), the feature of residue 10 pairs of videos is used for steganography classification judgement, repeats this process 100 times and averages to the data obtained.

According to above experimental configuration, gained steganalysis result is as shown in table 1, can find out, final average detected accuracy all very low (only having one group of accuracy more than 60%), therefore, the inventive method can resist the attack of the most effective steganalysis feature AoSO in current motion vector territory completely, is a kind of highly concealed type video steganography method, is specially adapted to the covert communications scene higher to security requirement.

Table 1 adopts AoSO feature to carry out the average detected accuracy of steganalysis to institute of the present invention extracting method

Above embodiment is only in order to illustrate technical scheme of the present invention but not to be limited; those of ordinary skill in the art can modify to technical scheme of the present invention or equivalent replacement; and not departing from the spirit and scope of the present invention, protection scope of the present invention should be as the criterion with described in claim.

Claims (8)

1., based on the video steganography method that motion vector local optimality keeps, it is characterized in that, the telescopiny of secret information comprises the following steps:

1) in current video frame, to each motion vector (mv _x, mv _y), calculate LSB (mv _x+ mv _y) obtain binary number 0 or 1, and write all result of calculation as vector x, wherein LSB represents and gets lowest bit bit manipulation;

2) using the carrier of gained vector x as private communication channel, according to the operation principle of the steganography code selected, the parameter of arranging in advance according to communicating pair generates parity check matrix H, and the motion vector disturbance cost computation model that combination is preset finally constructs the x' meeting Hx'=m, wherein m represents secret information to be embedded in this private communication channel, and x' represents the carrier vector through amendment;

3) according to step 2) determine the motion vector needing amendment in current video frame;

4) according to above acquired results coding current video frame, and respectively each motion vector to be modified is replaced to corresponding optimal candidate motion vector according to the optimal candidate motion vector sought algorithm preset, and then complete the secret information of private communication channel is embedded;

5) according to the frame coded sequence of video file, to remainder every frame repeated execution of steps 1) to 4) until secret information all embeds complete, obtain final steganography video file.

2. the method for claim 1, is characterized in that, also comprises the leaching process of secret information, and its step is as follows:

1) decoding current video frame, to each motion vector (mv _x, mv _y), calculate LSB (mv _x+ mv _y) obtain binary number 0 or 1, and write all result of calculation as vector x, wherein LSB represents and gets lowest bit bit manipulation;

2) according to the operation principle of the steganography code selected, the parameter appointed in advance according to communicating pair generates parity check matrix H, and calculate Hx, acquired results is the secret information bit of extraction;

3) according to the frame decoding order of video file, to remainder every frame repeated execution of steps 1) to 2) until secret information all extracts complete.

3. method as claimed in claim 1 or 2, it is characterized in that, described steganography code is STC steganography code.

4. method as claimed in claim 1 or 2, it is characterized in that, described default motion vector disturbance cost computation model is:

To motion vector MV _0,0=(mv _x, mv _y) horizontal and vertical direction on 4 adjacent motion vectors and MV _-1,0=(mv _x-1, mv _y), MV _1,0=(mv _x+ 1, mv _y), MV _{0 ,-1}=(mv _x, mv _y-1) and MV _0,1=(mv _x, mv _y+ 1) carry out local optimum sex determination, if there is N number of motion vector to possess local optimality in them, wherein 0≤N≤4, then to motion vector MV _0,0the disturbance cost of carrying out steganography embedding is:

Wherein, Ω={ MV _-1,0, MV _1,0, MV _{0 ,-1}, MV _0,1, namely the disturbance cost of motion vector depends on the quantity of local optimum motion vector in its contiguous range, and the intensity of variation of RDO functional value.

5. method as claimed in claim 4, is characterized in that, if need the motion vector of modifying to be MV _0,0=(mv _x, mv _y), then described default optimal candidate motion vector sought algorithm is:

A) respectively to MV _0,0four motion vector MV in contiguous range _-1,0=(mv _x-1, mv _y), MV _1,0=(mv _x+ 1, mv _y), MV _{0 ,-1}=(mv _x, mv _y-1) and MV _0,1=(mv _x, mv _y+ 1) local optimum sex determination is carried out;

B) by MV _-1,0, MV _1,0, MV _{0 ,-1}and MV _0,1in local optimum motion vector composition candidate motion vector set close Ψ, if Ψ non-NULL, then jump to step e), otherwise continue perform step c);

C) delimit the region of search of a motion vector, its size and final search time are directly proportional, and choose according to Practical Calculation resource or ability; For any candidate motion vector MV'=(mv in this region _x', mv _y'), meet LSB (mv _x+ mv _y) ≠ LSB (mv _x'+mv _y');

D) for each candidate motion vector in this region of search, carry out local optimum sex determination respectively, determine the motion vector of wherein all local optimums, and by they composite set Ψ;

E) for each local optimum motion vector in Ψ, calculating its RDO functional value, selecting the local optimum motion vector having minimum RDO functional value as finally revising result.

6. method as claimed in claim 5, it is characterized in that, the method for described local optimum sex determination is:

For motion vector MV _0,0=(mv _x, mv _y), if its prediction block pointed to is Pred (MV _0,0), the reconstructed block obtained through inter prediction encoding is Rec (MV _0,0); Compute matrix Pred (MV _0,0) and Rec (MV _0,0) between absolute error and, i.e. sad _0,0=SAD (Rec (MV _0,0), Pred (MV _0,0)); In like manner, for MV _0,08 adjacent around motion vector MV _i,j=(mv _x+ i, mv _y+ j), wherein-1≤i, j≤1, obtains the prediction block Pred (MV of their sensings respectively _i,j), and calculate sad _i,j=SAD (Rec (MV _0,0), Pred (MV _i,j)); If sad _i,jin minimum value be sad _0,0, then motion vector MV is claimed _0,0possesses local optimality.

7. method as claimed in claim 6, is characterized in that, adopts the telescopiny of the secret information of STC steganography code to comprise the following steps:

1) in current video frame, by each motion vector (mv _x, mv _y) horizontal and vertical component summation after get lowest bit position, i.e. LSB (mv _x+ mv _y), obtain binary number 0 or 1, and by all result of calculation composition binary system carrier vector x;

2) in current video frame, for each motion vector MV, the motion vector disturbance cost computation model according to presetting calculates its steganography disturbance cost Cost (MV), and by all result of calculation composition of vector Γ;

3) according to the operation principle of steganography code STC, first according to the preassigned load factor α of communicating pair and parameter matrix build parity matrix Η _sTC, then construct and meet Η _sTCx'=m and make steganography embed cost D (x, x', Γ) reach minimum x', wherein, m represents secret information to be embedded in private communication channel, x' represent by private communication channel steganography embed after be modified carrier vector; If the binary number on x with x' correspondence position is not identical, then represent that corresponding original motion vector needs amendment, if identical, represent and do not need amendment;

4) according to the amendment result of carrier vector in private communication channel, for each motion vector needing amendment, the optimal candidate motion vector sought algorithm according to presetting is replaced to the minimum local optimum motion vector of RDO functional value in candidate motion vector;

5) embed with the steganography completing secret information according to above acquired results coding current video frame;

6) according to the frame coded sequence of video file, to remainder every frame repeated execution of steps 1) to 5) until secret information all embeds complete, obtain final steganography video file.

8. method as claimed in claim 7, is characterized in that, adopts the leaching process of the secret information of STC steganography code to comprise the following steps:

1) decoding current video frame, to each motion vector (mv _x, mv _y) calculate LSB (mv _x+ mv _y) obtain binary number 0 or 1, and write all result of calculation as binary vector x;

2) secret information in private communication channel is extracted: according to the operation principle of STC, according to the preassigned load factor α of communicating pair and parameter matrix build parity matrix Η _sTC, calculate Η _sTCx, acquired results is the secret information bit of extraction;

3) according to the frame decoding order of video file, to remainder every frame repeated execution of steps 1) to 2) until secret information all extracts complete.