CN105872555B - A kind of steganalysis algorithm for the insertion of H.264 video motion vector information - Google Patents
A kind of steganalysis algorithm for the insertion of H.264 video motion vector information Download PDFInfo
- Publication number
- CN105872555B CN105872555B CN201610177535.5A CN201610177535A CN105872555B CN 105872555 B CN105872555 B CN 105872555B CN 201610177535 A CN201610177535 A CN 201610177535A CN 105872555 B CN105872555 B CN 105872555B
- Authority
- CN
- China
- Prior art keywords
- motion vector
- order difference
- video
- differnce
- feature
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Expired - Fee Related
Links
Classifications
-
- H—ELECTRICITY
- H04—ELECTRIC COMMUNICATION TECHNIQUE
- H04N—PICTORIAL COMMUNICATION, e.g. TELEVISION
- H04N19/00—Methods or arrangements for coding, decoding, compressing or decompressing digital video signals
- H04N19/46—Embedding additional information in the video signal during the compression process
- H04N19/467—Embedding additional information in the video signal during the compression process characterised by the embedded information being invisible, e.g. watermarking
-
- H—ELECTRICITY
- H04—ELECTRIC COMMUNICATION TECHNIQUE
- H04N—PICTORIAL COMMUNICATION, e.g. TELEVISION
- H04N17/00—Diagnosis, testing or measuring for television systems or their details
Landscapes
- Engineering & Computer Science (AREA)
- Multimedia (AREA)
- Signal Processing (AREA)
- Health & Medical Sciences (AREA)
- Biomedical Technology (AREA)
- General Health & Medical Sciences (AREA)
- Compression Or Coding Systems Of Tv Signals (AREA)
Abstract
The invention discloses a kind of steganalysis algorithms for the insertion of H.264 video motion vector information, purpose is, steganalysis fast and effeciently can be carried out to H.264 video, especially for can preferably keep statistical property constant and the embedded mobile GIS verification and measurement ratio with higher of low insertion rate, used technical solution are as follows: motion vector is accurately extracted, higher difference feature is devised on the basis of first-order difference feature, the category feature preferably can carry out steganalysis to the steganographic algorithm for having vector statistics feature to compensate, utilize the thought of integrated study, multiple simple classification devices are trained respectively, design a kind of integrated steganography classifier based on FLD, integrated classifier is not only greatly improved on classification speed, and than there is higher classification accuracy using single same category device.
Description
Technical field
The invention belongs to Information Hiding Techniques fields, and in particular to a kind of to be embedded in for H.264 video motion vector information
Steganalysis algorithm.
Background technique
Steganography and steganalysis are the main contents of Information Hiding Techniques, and steganography is as the important of safe information transmission
Means, can be applied to the levels such as military affairs, information, national security, provide technical guarantee for communication security.Since video carrier has
There is the features such as larger data redundancy, human eye is relatively weak to moving image sensitivity, so that digital video is that large capacity is hidden
Communication provides necessary condition.Constantly heat up about the research of video concealing technology at present, on network using digital video as
The software that carrier carries out Information hiding also occurs in succession.Video information steganalysis must be important as one project and weighed
Depending on and concern.
Motion prediction and compensation technique utilize the correlation of video frame over time and space, dramatically save video counts
According to memory space, be widely used in all kinds of video compression standards.Motion vector is as in compression Video coding
Important component, stability with higher, a small amount of change can't cause big influence to image quality, very suitable
Cooperation is the position of Information hiding, quite a few video steganographic algorithm is exactly the movement arrow that selection is met certain condition at present
Amount, by information be embedded in motion vector amplitude, phase angle, the direction of motion, horizontal component and vertical component or predicted value vector it
In, this kind of information steganography algorithm safety with higher based on motion vector, not sentience and information are embedded in and extract
Real-time, the features such as being combined with video encoding standard.
H.264 as a kind of new, widely used video compression standard, have support smaller with previous standard compared with
Piecemeal, more accurate prediction, may include more motion vectors in compressed encoding, and H.264 this makes normal video is very
It is suitble to carry out hiding information using motion vector as embedded location.Such Information Hiding Techniques is light due to the modulation to motion vector
It is micro-, therefore very little is influenced on video frame quality, specific regularity is not present for video frame itself, therefore can not scheme from rebuilding
Presence as reflecting hiding information on frame, traditional statistical analysis algorithms based on video frame pixel can not be effectively detected point
Such video steganographic algorithm is precipitated.
Currently, the rational approach of a large amount of digital video steganography and steganalysis has been carried out in domestic and foreign scholars,
Practical, highly-safe, perfect in shape and function Steganalysis is also actively being studied by many countries.But it is generally speaking related
Video Steganalysis, especially to video compression standard of new generation is combined, for the steganalysis algorithm of motion vector
It studies still more limited.
Currently, the Steganalysis development for image is very fast.Since the 1990s mid-term, propose both at home and abroad
Image latent writing analytical technology has as many as tens kinds, and direction is had nothing in common with each other, and wherein the overwhelming majority belongs to statistics characteristic analysis method.
Such as: Fridrich proposes a kind of steganalysis algorithm based on single order and second order distributed feature for jpeg image steganography,
And by it as a comparison jpeg image steganographic algorithm insertion mechanism appraisal tool;Jackson etc. extracts feature in Farid algorithm
On the basis of, devise the Stego-detection system (CIS) based on artificial immunity;Avcibas selects similar between bit-planes
Then degree establishes classifier by multiple linear regression model, but this method is only applicable to LSB steganography to establish feature vector
The detection of algorithm;Westfeld is according to the value before and after LSB hiding information in carrier image to the system of (pairs of value)
Meter property difference devises chi-square criterion method.Domestic Wang Shuozhong, Zhang Xinpeng et al. is to regard to digital picture steganography and steganalysis
The analysis that it is careful that technology is contrasted published " Steganography system and steganalysis " book in 2005.It is relatively more detailed in book
Describe some common steganography methods, steganalysis method and anti-steganography method.
Compared with image latent writing analytical technology, the development of video Steganalysis is later and relatively slow.Existing known steganography point
Analysis algorithm mainly has a steganalysis algorithm for MSU video hide tools of the very equal propositions of Su Yu, and the algorithm is from embedding data
Caused blocking artifact variation is started with afterwards, and for the video of not embedding data, the blocking artifact distribution of 16x16 block and 8x8 are basic
Equally, when there is data insertion, the blocking artifact of 16x16 is apparently higher than 8x8 blocking artifact, on this basis, resets a thresholding
Value carries out Stego-detection.But the algorithm is the dedicated Stego-detection for MSU, and the video to other methods steganography is simultaneously not suitable for.
The steganalysis method based on collusion attack that Udit Budhiad etc. [8] is proposed utilizes the secret of collusion attack removal target frame
Confidential information implements collusion attack to given video, as long as video rate of change is very slow and implements the window L choosing of collusion attack
Taking appropriately can obtain the approximation of original video, to estimate the secret information of insertion.The algorithm is due to being to utilize phase
The time redundancy of adjacent interframe, when video variation is violent, when having more global motion, Detection accuracy is decreased obviously.
Julien.S.Jainsky etc. proposes asymptotic memoryless detection method.Using the motion vector of present frame and its before and after frames, use
Video motion interpolation reconstruction video, is then compared with original video, analyses whether to have carried out information finally by ARE arbiter
It hides.The time redundancy of video is equally only utilized in the algorithm, when video is not exclusively embedded in, especially current video frame
Hiding information, and its front and back consecutive frame not hide when, present frame can accurately be detected, but hidden for large capacity information
The whole detection accuracy rate of video sequence is hidden without too big raising.Su Yu very waits the straight using the first-order difference of motion vector of propositions
The characteristics of side's figure, its corresponding statistic is calculated, the algorithm for being directly based upon motion vector information insertion has preferable inspection
Efficiency is surveyed, but such as motion vector is compensated, so that its histogram is kept original characteristic, then can not effectively be detected.Yu
The steganalysis algorithm for motion vector insertion that Deng is proposed, the algorithm is using the corresponding statistic of second differnce as spy
Sign, carries out corresponding steganalysis.This has preferable analytical effect to the steganographic algorithm that first-order characteristics are kept, but for single order
The steganographic algorithm of the equal retention performance of second order can not obtain preferable verification and measurement ratio.
The detection of video steganalysis is at the early-stage, and related algorithm is less, carries out studying oneself to it being extremely urgent.It is existing
Detection algorithm need to be improved in applicability and Detection accuracy, therefore embedding specifically for H.264 normal video motion vector
The detection algorithm entered has certain practical value.
Summary of the invention
In order to solve the problems in the prior art, the present invention proposes a kind of for the insertion of H.264 video motion vector information
Steganalysis algorithm, steganalysis fast and effeciently can be carried out to H.264 video, especially for can preferably keep uniting
Count the embedded mobile GIS verification and measurement ratio with higher of the constant and low insertion rate of characteristic.
In order to achieve the goal above, the technical scheme adopted by the invention is as follows: the following steps are included:
1) motion vector extracts: being decoded to video, obtains the motion vector of each macro block, as motion vector difference
The data of feature extraction;
2) extraction of motion vector Differential Characteristics is carried out to the data that step 1) obtains:
2.1) first-order difference feature extraction:
Calculate the first-order difference distortion factor E of motion vector:
Wherein, kurtosis (▽ Si) be motion vector first-order difference kurtosis value, ▽ SiFor motion vector first-order difference,
▽Si=Si-Si+1, SiIt is the value of motion vector in i-th piece, h [- 2], h [- 1], h [0], h [1], h [2] respectively indicate ▽ Si
Probability mass function;
Calculate the mean value of single order vector difference
Wherein, q is the amplitude range of motion vector first-order difference, hS[n] is ▽ SiProbability mass function;
Calculate the variance of single order vector difference
Wherein, μsFor the mean value of motion vector, hS[n] is ▽ SiProbability mass function;
By E,As first-order difference feature;
2.2) second differnce feature and third order difference feature extraction:
Calculate the center mass function of motion vector second differnce:
Wherein,Respectively indicate ▽2Xi, second differnce ▽2S、▽2The probability mass function of η
Fourier's series, i are the numbers of block in a frame, and N is the number of block in video frame;
Calculate the variance of second differnce:
Wherein, μsFor the mean value of motion vector,It is second differnce ▽2The probability mass function of S, q are motion vector
The amplitude range of first-order difference;
Calculate the kurtosis value of second differnce and third order difference:
Wherein,WithThe mean value of motion vector second differnce and third order difference is respectively indicated,The variance square of motion vector second differnce and third order difference is respectively indicated, N is video frame
Middle piece of number;
By second differnce statisticAnd third order difference statisticAs higher difference feature;
3) video clip of input is carried out according to first-order difference feature and higher difference feature using FLD integrated classifier
Detection judgement, statistics are judged as the number of steganography and normal video frame wherein to determine entire video clip, i.e. completion steganography point
Analyse algorithm.
Video is decoded using FFmpeg decoder in the step 1), in decode_slice_header ()
The frame number that current decoded frame is obtained in function, obtains the type of each macro block, in mc_dir_ in h1_motion () function
The motion vector of each macro block is obtained in part () function, and is output in TXT text, is mentioned as motion vector Differential Characteristics
The data taken.
The calculation formula of the kurtosis value of motion vector first-order difference is as follows in the step 2.1):
Wherein, μ (▽ Si) be motion vector first-order difference mean value, σ4(▽Si) indicate motion vector first-order difference side
Difference square, N are the number of block in video frame.
The center mass feature of first-order difference function is introduced in the step 2.1) to reflect motion vector first-order difference
Energy distribution situation, pass through calculate the Fourier transform of first-order difference probability mass function after center mass feature C1(H
[m]), to reflect vector first-order difference in frequency domain energy variation situation:
Wherein, H [i] is indicated, T is indicated, i indicates the number of block in a frame.
The first-order difference of motion vector is defined as follows in the step 2.1):For carrier video
The first-order difference Distribution value of motion vector meets the super-Gaussian distribution that peak value is 0, and the horizontal component after embedding information is divided with vertical
The motion vector first-order difference of amountWithIt respectively indicates are as follows:
Wherein,WithThe respectively embedding information of the horizontal component of first-order difference and vertical component,WithPoint
Not Wei horizontal component and vertical component first-order difference.
The second differnce of adjacent motion vectors is defined in the step 2.2) are as follows:
▽2Si=▽ Si-▽Si+1
The horizontal and vertical component second differnce of motion vector indicates after information insertion are as follows:
Wherein,WithThe respectively embedding information of the horizontal component of second differnce and vertical component,WithThe respectively first-order difference of horizontal component and vertical component.
Third order difference is defined as follows in the step 2.2):
FLD integrated classifier includes several base learners, each base learner B in the step 3)l, l=
1 ..., L is RdThe mapping of → { 0,1 }, wherein 0 indicates carrier frame, 1 indicates steganography frame, RdFor all full dimensional characteristics, often
The decision-making value of one base learner be adjusted to etc. priori in the case where, minimize trained error rate are as follows:
Wherein PFA,PMDIt is the probability of false-alarm and missing inspection respectively.
Each described base learner is using Fisher linear discriminant as learning tool.
The lth base learner of the FLD integrated classifier is in training setUpper training
It obtains, whereinIt is randomly selected subset, simultaneouslyBe from set 1 ..., NtrnIn adopt
The bootstrapping sample that sample obtains,The feature vector of each base learner indicates are as follows:
WhereinIt is the mean value of every one kind: It is the Scatter Matrix in class, wherein λ is steady
Parameter is determined to guarantee SW+ λ I is positive definite.
Compared with prior art, the present invention accurately extracts motion vector, by being simplified and being modified to FFmpeg, only
Retain h.264 decoded portion therein, and increase the output function of motion vector information, while decoding quickly and accurately
Realize the extraction of motion vector, it is versatile.Higher difference feature is devised on the basis of first-order difference feature, the category feature
Steganalysis preferably can be carried out to the steganographic algorithm for having vector statistics feature to compensate, detection efficiency is high, versatility, stabilization
Property is strong.Video image characteristic quantity is big, with single complex classifier, at support vector machines either Method Using Relevance Vector Machine
Detection time needed for reason is all very long, and with simple classification device such as FLD, then detection effect is poor, in time efficiency and accuracy rate
Between be difficult to seek an equalization point, using the thought of integrated study, multiple simple classification devices are trained respectively, design one
Integrated steganography classifier of the kind based on FLD, integrated classifier are not only greatly improved on classification speed, but also than using single
One same category device has higher classification accuracy.The steganalysis algorithm of resisted motion Vector Message insertion of the present invention,
And realize its software systems, on the basis of feature extraction and classifier design, design can resist motion vector information insertion
Steganalysis algorithm, and using H.264 carrier as test object, realize steganalysis software systems.
Detailed description of the invention
Fig. 1 is that motion vector of the present invention extracts flow chart;
Fig. 2 is integrated classifier training and detection of the present invention;
Fig. 3 is working model figure of the present invention;
Fig. 4 a is the first-order difference distribution map of the motion vector of normal video;Fig. 4 b is the one of the motion vector of concealed video
Order difference distribution map.
Specific embodiment
Below with reference to specific embodiment and Figure of description the present invention will be further explained explanation.
Video steganalysis of the invention is based on H.264/AVC standard, specifically for motion vector in H.264 video
Embedding grammar carries out steganalysis, and key technology is how fast and effeciently to carry out steganalysis, especially needle to H.264 video
To can preferably keep statistical property constant and the embedded mobile GIS verification and measurement ratio with higher of low insertion rate, referring to Fig. 3, specifically
It is as follows:
1) motion vector extracts:
Steganalysis is carried out for motion vector, the accurate extraction of motion vector is the premise of steganalysis, for H.264
For code stream, motion vector is extracted, first has to be decoded it, the encoder for meeting H.264 coding standard at present has very
More, different encoders generally requires its corresponding decoder to decode, therefore selects a versatility good, fireballing
H.264 encoder is particularly significant, and FFmpeg is the cross-platform video and audio codec of a open source, it is provided to current
The decoding scheme of mainstream audio and video, in order to enable there is preferable versatility to the code stream of different coding device, referring to Fig. 1,
The present invention is decoded video using FFmpeg, and FFmpeg is powerful and versatile, but it is mainly in Linux platform
Lower exploitation, in order to make its stable operation under windows system, it is compiled again under windows environment, together
When code is simplified, only remain decoded portion wherein H.264, joined on this basis motion vector output mould
Block obtains the frame number of current decoded frame in decode_slice_header () function, obtains in h1_motion () function
The type of each macro block obtains each piece of motion vector in mc_dir_part () function, these information is output to
Initial data in TXT text, as the calculating of motion vector Differential Characteristics.
2) motion vector Differential Characteristics extract:
It is characterized in whether the key point of steganalysis, feature are effectively directly related to the height of verification and measurement ratio, based on movement
The information insertion of vector can be regarded as is added noise in the horizontal component and vertical component of motion vector, and difference can be with table
Show as follows:
WithIt is the value of horizontal and vertical movement vector component in i-th piece respectively, N is the number of block in a frame,
WithIt is the information of insertion, probability mass function PMF is expressed as follows,P is
The insertion rate of information in algorithm, k indicates the modified values in algorithm to carrier vector, as k=1 to the shadow of carrier after embedding information
Minimum is rung, the embedded mobile GIS based on motion vector is the S that selection meets certain condition under normal circumstancesiCome embedding information, such as Si
Amplitude, phase angle etc.,Motion vector after indicating embedding information.To the steganographic algorithm based on motion vector point
Analysis, essence are exactly to find out the difference of the motion vector of carrier and concealed video, therefore we will design corresponding statistic and come instead
S after answering information to be embedded iniDistortion situation, Difference Calculation is carried out to motion vector in the present invention, and with the system of these difference results
Metering is as detection feature;
(1) first-order difference feature:
It is had good correlation between the adjacent motion vector in space, uses one of neighboring block motion vector in same frame
Order difference embodies this correlation properties.The first-order difference of motion vector is defined as follows:Due to correlation,
The super-Gaussian distribution that peak value is 0 is met for the first-order difference Distribution value of carrier video motion vector.Shown in its distribution map 4a:
The motion vector first-order difference of horizontal and vertical component after embedding information can be expressed as:
According to ηiProbability mass function, ▽ ηiPMF be, It is hereby achieved thatDistribution, such as Fig. 4 b
It is shown, meet the Gaussian Profile that peak value is 0.The kurtosis of stochastic variable sufficiently reflects its distribution feelings compared with normal distribution
Steep under condition, therefore the motion vector that normal motion vector crosses information than steganography has bigger kurtosis value.Motion vector
First-order difference is defined as follows:
The distribution situation of front and back first-order difference is embedded according to information, the first-order difference distortion factor for defining vector is as follows:
▽ηiWith ▽ SiIndependently of each other, so ▽ XiProbability mass function be ▽ ηiWith ▽ SiThe volume of probability mass function
Product, uses hx[n],hs[n],hη[n] respectively indicates ▽ Xi, the probability mass function of ▽ S, ▽ η, then hx[n]=hs[n]*hη[n]。
It converts it to frequency domain and obtains Hx[m]=Hs[m]Hη[m], H are the discrete Fourier transform of h.HηThe derivation of [m] is as follows:
Available Hη[m]≤hη[0]+2hη[k]=hη[0]+hη[k]+hη[- k]=1
So Hx[n]≤Hs[n]。
Center mass feature is introduced, to reflect the transformation of frequency domain vector:
Mean value, the variance of single order vector difference are calculated simultaneously to reflect the situation of change of vector correlation
Wherein, ▽ Si=Si-Si+1, SiIt is the value of motion vector in i-th piece, μ (▽ Si) it is ▽ SiMean value, σ4(▽
Si) it is ▽ SiSquare of variance.H [n] is ▽ SiProbability mass function, H be h Discrete Fourier Transform, μsFor motion vector
Mean value, q be motion vector first-order difference amplitude range.
The E of motion vector horizontal and vertical component will be calculated,As first-order difference feature.
(2) second differnce feature and third order difference feature:
The insertion of information destroys the original correlativity of motion vector, and single order feature in general can be preferable
Reflect this distortion of vector, the motion vector of carrier video follows the super-Gaussian distribution figure of 0 peak value, such as 4a, and concealed video
Motion vector follow the Gaussian Profile of 0 peak value, such as Fig. 4 b, this be single order feature it is effective it is basic where.For being sweared in movement
It being directly embedded into amount for the algorithm of information, first-order difference feature has preferable verification and measurement ratio, but first-order difference calculating is simple,
It is comparatively easy that information remains unchanged the distribution of motion vector first-order difference.Different points such as are followed for two classes
Most important the difference is that wherein [- 2] h in cloth first-order difference, h [- 1], h [0], numerical value differs in h [1], h [2], concealed
The quantity of h [0] will be less than the quantity in normal video in video motion vector first-order difference, and h [- 2], h [- 1], h [1], h
[2] quantity in extra normal video is then wanted.Remaining hiI=± 3 ... then difference is little by ± n.If the energy in Information hiding
Enough keep h [- 2], h [- 1], h [0], h [1], h [2] numerical value it is almost the same, then first-order difference feature will fail.In order to overcome
This drawback, to the steganographic algorithm remained unchanged with single order feature verification and measurement ratio equally with higher, using high rank difference
One as feature of statistic.
Define the second differnce of adjacent motion vectors:
▽2Si=▽ Si-▽Si+1
Then the horizontal and vertical component second differnce of information insertion front and back vector can be expressed as follows:
WithRespectively indicate ▽2Xi,▽2S,▽2The probability mass function of η, then due to mutually indepedent
Property, it is similar with single order situation, it obtainsBeing converted to frequency domain hasRoot
According to the derivation of document [12], the center mass function of second differnce meets following relationship:
H2[n] is the Fourier's series of second differnce probability mass function.
The variance of second differnce is calculated simultaneously:
In order to enable detection algorithm has stronger versatility, have to the steganographic algorithm with feature retention performance more preferable
Detection effect, joined the third order difference statistical nature of motion vector again in detection feature, third order difference is defined as follows:
In view of such as to keep the statistical property of first-order difference constant, then the motion vector for needing to change is more, three scales
The distribution curve of score value will change, and kurtosis is the finger of the high and steep degree of point of the intensity or distribution curve for measuring distribution
The kurtosis definition of mark, therefore a part using second differnce and third order difference kurtosis value as feature, second order and third order difference
It is as follows:
WithThe mean value of motion vector second differnce and third order difference is respectively indicated,
WithThe variance square of motion vector second differnce and third order difference is respectively indicated, N is the number of block in video frame.
By the second differnce statistic of motion vector horizontal component and vertical component
And third order difference statisticAs higher difference feature;
It extracts motion vector first-order difference statistical nature and higher difference statistical nature collectively constitutes the detection of steganalysis
Feature.
3) integrated classifier designs:
The target of steganalysis is to detect the existence of the secret information in digital multimedia.Digital Media, such as image,
Video and audio is ideal carrier for steganography method, this is because they include a large amount of independent elements, these yuan
Element can be modified slightly to one classified information of insertion.Up to the present it is difficult using the method for statistics description to these
Carrier accurately models, and this considerably increases the difficulty of steganalysis.It is based particularly on the system extracted in carrier and steganography carrier
It is infeasible that characteristic, which is counted, come the detection method for estimating potential probability distribution.Therefore, detection is generally regarded as an engineering
Supervised classification problem in habit.
Despite the presence of a large amount of machine learning method, support vector machines is current most popular method.This mainly by
In SVMs has solid Fundamentals of Mathematics, it is to be overcome overfitting again simultaneously based on Statistical Learning Theory and work as intrinsic dimensionality
It remains to provide good result when big than number of samples.However, the complexity of SVM training reduces its use value, very
To not being on very big data set at one, training is also required to for a long time, this is because calculating the Gram matrix for indicating core
Complexity be proportional to square of intrinsic dimensionality Yu training set size product.Moreover, training process itself is at least training sample
A double optimization problem.These problems limit the scale solved the problems, such as when SVM application in practice, while requiring analysis again
Person has to the more accurate feature of building to meet the constraint that these are complicated as caused by computing resource.Integrated classifier provides
It is very big to carry out steganalysis from origin, it can not be limited by intrinsic dimensionality to design some features, while can not examine
The limitation of training sample number is considered to carry out the learning process of a more quick detector.
By multiple base learners, the stand-alone training in one group of carrier video frame and steganography video frame obtains integrated classifier, often
One base learner is exactly a simple classifier, this classifier establishes the son in the feature space of random (uniform) selection
Spatially.
Each base learner Bl, l=1 ..., L are a RdThe mapping of → { 0,1 }, wherein 0 expression carrier frame, 1
Indicate steganography frame.It should be noted that although learner is defined on all full dimensional characteristics RdOn, but all base learners
The dimension d of feature spacesubIt can choose the value more much smaller than full dimension d, this makes it possible to be effectively reduced computation complexity.
Although the classification performance of each independent base learner is very weak, when the value of L is sufficiently large, after carrying out tactful fusion, accurately
Degree will be greatly improved, and may finally restrain.The decision-making value of each base learner be adjusted to etc. priori
In the case of, minimize trained error rate are as follows:
Wherein PFA,PMDIt is the probability of false-alarm and missing inspection respectively.The present invention
Although intrinsic dimensionality is not high in, and the data volume detected for video is very big, and this integrated classifier can greatly
Shorten detection time and improves classification accuracy.
The present invention is using Fisher linear discriminant as the learning tool of each base learner, and this is mainly due to it
With lower trained complexity.The maximum situation of event consumption is to solve the inverse of covariance matrix in class and their sums
When.In addition, such weak and unstable classifier increases the degree of scatter of integrated study.
Referring to Fig. 3, since FLD is the classification tool of a standard, description part relevant to Ensemble classifier in the present invention.
The lth base learner is in training setWhat upper training obtained, wherein
It is randomly selected subset, simultaneouslyBe from set 1 ..., NtrnIn the obtained bootstrapping sample of sampling,Often
One base learner can be indicated with following feature vector:
WhereinIt is the mean value of every one kind:
It is the Scatter Matrix in class,
Wherein λ is steadiness parameter to guarantee SW+ λ I is positive definite, so as to avoid in practice, works as SWIt is counted when being unusual or ill
It is worth the unstability calculated.To a testing feature vector y ∈ Ytst, first of base learner is by calculating mappingAnd with
Threshold value (this threshold value needs to pre-adjust to meet expected performance standard) is compared to obtain this base learner
Classification results.After obtaining all L decisions, the output of final integrated classifier utilizes this L decision without the (more of weight
Number) temporal voting strategy is combined, i.e., it is carried out pair by the booking result summation of all single base learners and with decision-making value L/2
Than providing final decision result.Notice that this threshold value can be adjusted between [0, L], so that it is wrong to control different two classes
Significance level accidentally or acquisition one complete reception operating characteristic ROC curve.This project design classifier in we
It is L/2 by adjusting thresholds, false alarm rate and omission factor average value has been made to reach minimum.
FLD integrated classifier training when by frame as unit of, obtain integrated classifier parameter.Since video sequence has very
Strong continuity, it will be assumed that the smallest insertion unit is not less than 50 frames (about 2 seconds), is carried out as unit of 50 frame video clips
Detection, statistics is wherein judged as the number of steganography and normal video frame, and is corrected with priori error rate to judgement, final basis
It is determined as steganography video frame in segment and is determined as the quantity of normal video frame to determine whole fragment.
Motion vector of the invention accurately extracts, and is simplified and is modified to FFmpeg, only retains and therein h.264 decodes
Part, and increase the output function of motion vector information, the extraction of motion vector is quickly and accurately realized while decoding,
It is versatile.Vector characteristic extraction algorithm devises higher difference feature on the basis of first-order difference feature, the category feature energy
Steganalysis preferably is carried out to the steganographic algorithm for having vector statistics feature to compensate, detection efficiency is high, versatility, stability
By force.Video image characteristic quantity is big, is handled with single complex classifier, such as support vector machines either Method Using Relevance Vector Machine
Required detection time is all very long, and with simple classification device such as FLD, then detection effect is poor, time efficiency and accuracy rate it
Between be difficult to seek an equalization point, the present invention integrates the thought that integrated study is utilized in steganography classifier design, to multiple simple
Classifier is trained respectively, designs a kind of integrated steganography classifier based on FLD.Integrated classifier is not only on classification speed
It is greatly improved, and than there is higher classification accuracy using single same category device.Design resisted motion vector
The steganalysis algorithm of information insertion, and realize its software systems.On the basis of feature extraction and classifier design, design can
The steganalysis algorithm of motion vector information insertion is resisted, and using H.264 carrier as test object, realizes steganalysis software
System.The video steganalysis system that the present invention is developed, the function of realization are effectively analyzed and determined out to based on H.264/
The video of AVC standard motion vector information insertion, verification and measurement ratio reach 70% or more.
Claims (10)
1. a kind of steganalysis algorithm for the insertion of H.264 video motion vector information, which is characterized in that including following step
It is rapid:
1) motion vector extracts: being decoded to video, obtains the motion vector of each macro block, as motion vector Differential Characteristics
The data of extraction;
2) extraction of motion vector Differential Characteristics is carried out to the data that step 1) obtains:
2.1) first-order difference feature extraction:
Calculate the first-order difference distortion factor E of motion vector:
Wherein, kurtosis (▽ Si) be motion vector first-order difference kurtosis value, ▽ SiFor motion vector first-order difference, ▽ Si
=Si-Si+1, SiIt is the value of motion vector in i-th piece, h [- 2], h [- 1], h [0], h [1], h [2] respectively indicate ▽ SiIt is general
Rate mass function;
Calculate the mean value of single order vector difference
Wherein, q is the amplitude range of motion vector first-order difference, hS[n] is ▽ SiProbability mass function;
Calculate the variance of single order vector difference
Wherein, μsFor the mean value of motion vector, hS[n] is ▽ SiProbability mass function;
By E,As first-order difference feature;
2.2) second differnce feature and third order difference feature extraction:
Calculate the center mass function of motion vector second differnce:
Wherein,Respectively indicate ▽2Xi, second differnce ▽2S、▽2The Fourier of the probability mass function of η
Variation, i are the numbers of block in a frame, and N is the number of block in video frame;
Calculate the variance of second differnce:
Wherein, μsFor the mean value of motion vector,It is second differnce ▽2The probability mass function of S, q are one scale of motion vector
The amplitude range divided;
Calculate the kurtosis value of second differnce and third order difference:
Wherein,WithThe mean value of motion vector second differnce and third order difference is respectively indicated,The variance square of motion vector second differnce and third order difference is respectively indicated, N is video frame
Middle piece of number;
By second differnce statisticAnd third order difference statisticMake
For higher difference feature;
3) video clip of input is detected according to first-order difference feature and higher difference feature using FLD integrated classifier
Judgement, statistics are judged as the number of steganography and normal video frame wherein to determine entire video clip, i.e. completion steganalysis is calculated
Method.
2. a kind of steganalysis algorithm for the insertion of H.264 video motion vector information according to claim 1, special
Sign is, is decoded using FFmpeg decoder to video in the step 1), in decode_slice_header ()
The frame number that current decoded frame is obtained in function, obtains the type of each macro block, in mc_dir_ in h1_motion () function
The motion vector of each macro block is obtained in part () function, and is output in TXT text, is mentioned as motion vector Differential Characteristics
The data taken.
3. a kind of steganalysis algorithm for the insertion of H.264 video motion vector information according to claim 1, special
Sign is that the calculation formula of the kurtosis value of motion vector first-order difference is as follows in the step 2.1):
Wherein, μ (▽ Si) be motion vector first-order difference mean value, σ4(▽Si) indicate that the variance of motion vector first-order difference is flat
Side, N are the number of block in video frame.
4. a kind of steganalysis algorithm for the insertion of H.264 video motion vector information according to claim 3, special
Sign is, the center mass feature of first-order difference function is introduced in the step 2.1) to reflect motion vector first-order difference
Energy distribution situation passes through the center mass feature C after calculating first-order difference probability mass function Fourier transform1(H [m]),
To reflect vector first-order difference in frequency domain energy variation situation:
Wherein, H [i] indicates first-order difference ▽ SiThe Fourier transformation of probability mass function h [i], wherein ▽ Si=Si-Si+1, SiIt is
The value of motion vector in i-th piece, T indicate that the sum of block in a frame, i indicate the number of block in a frame.
5. a kind of steganalysis algorithm for the insertion of H.264 video motion vector information according to claim 4, special
Sign is that the first-order difference of motion vector is defined as follows in the step 2.1):Carrier video is transported
The first-order difference Distribution value of dynamic vector meets the super-Gaussian distribution that peak value is 0, horizontal component and vertical component after embedding information
Motion vector first-order differenceWithIt respectively indicates are as follows:
Wherein,WithThe respectively horizontal component of first-order differenceAnd vertical componentEmbedding information,WithPoint
It is notWithFirst-order difference,WithRespectively horizontal componentAnd vertical componentFirst-order difference;With
Indicate the horizontal and vertical component in i-th piece before motion vector information insertion.
6. a kind of steganalysis algorithm for the insertion of H.264 video motion vector information according to claim 1, special
Sign is, the second differnce of adjacent motion vectors is defined in the step 2.2) are as follows:
▽2Si=▽ Si-▽Si+1
The horizontal and vertical component second differnce of motion vector is expressed as after information insertionWith
Wherein,WithIn respectively i-th piece in motion vector horizontal and vertical component embedding information first-order difference,WithRespectivelyWithSecond differnce,WithRespectivelyWithSecond differnce.
7. a kind of steganalysis algorithm for the insertion of H.264 video motion vector information according to claim 1, special
Sign is that third order difference is defined as follows in the step 2.2):
8. a kind of steganalysis algorithm for the insertion of H.264 video motion vector information according to claim 1, special
Sign is that FLD integrated classifier includes several base learners, each base learner B in the step 3)l, l=
1 ..., L is RdThe mapping of → { 0,1 }, wherein 0 indicates carrier frame, 1 indicates steganography frame, RdFor all full dimensional characteristics, often
The decision-making value of one base learner be adjusted to etc. priori in the case where, minimize trained error rate are as follows:
Wherein PFA,PMDIt is the probability of false-alarm and missing inspection respectively.
9. a kind of steganalysis algorithm for the insertion of H.264 video motion vector information according to claim 8, special
Sign is that each described base learner is using Fisher linear discriminant as learning tool.
10. a kind of steganalysis algorithm for the insertion of H.264 video motion vector information according to claim 9,
It is characterized in that, first of base learner of the FLD integrated classifier is in training setIt is upper trained
It arrives, whereinIt is randomly selected subset, simultaneouslyBe from set 1 ..., NtrnIn sampling
Obtained bootstrapping sample,The feature vector of each base learner indicates are as follows:
Wherein μ,It is the mean value of every one kind: It is the Scatter Matrix in class, wherein λ is steady
Parameter is determined to guarantee SW+ λ I is positive definite.
Priority Applications (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201610177535.5A CN105872555B (en) | 2016-03-25 | 2016-03-25 | A kind of steganalysis algorithm for the insertion of H.264 video motion vector information |
Applications Claiming Priority (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201610177535.5A CN105872555B (en) | 2016-03-25 | 2016-03-25 | A kind of steganalysis algorithm for the insertion of H.264 video motion vector information |
Publications (2)
Publication Number | Publication Date |
---|---|
CN105872555A CN105872555A (en) | 2016-08-17 |
CN105872555B true CN105872555B (en) | 2019-01-15 |
Family
ID=56625225
Family Applications (1)
Application Number | Title | Priority Date | Filing Date |
---|---|---|---|
CN201610177535.5A Expired - Fee Related CN105872555B (en) | 2016-03-25 | 2016-03-25 | A kind of steganalysis algorithm for the insertion of H.264 video motion vector information |
Country Status (1)
Country | Link |
---|---|
CN (1) | CN105872555B (en) |
Families Citing this family (9)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN107172434B (en) * | 2017-06-27 | 2019-06-18 | 郑州师范学院 | A method of built-in steganography is carried out to H.265 video |
CN107197298B (en) * | 2017-06-27 | 2019-06-25 | 郑州师范学院 | A kind of reversible steganography method of HEVC video of the undistorted drift based on multi-component system |
CN107333139B (en) * | 2017-06-27 | 2019-12-13 | 郑州师范学院 | H.265 semi-reversible video steganography information method based on intra-frame prediction mode |
CN107947898B (en) * | 2017-11-15 | 2020-08-07 | 深圳大学 | Information detection method and device based on optimized grouping variance and receiving equipment |
CN108460405A (en) * | 2018-02-02 | 2018-08-28 | 上海大学 | A kind of image latent writing analysis Ensemble classifier optimization method based on deeply study |
CN113902921B (en) | 2018-11-30 | 2022-11-25 | 腾讯科技(深圳)有限公司 | Image processing method, device, equipment and storage medium |
CN110062242B (en) * | 2019-03-04 | 2021-05-04 | 中山大学 | H.264 video steganography algorithm based on UED |
CN110149457B (en) * | 2019-05-08 | 2021-05-04 | 中山大学 | Multi-objective optimization H.264 video steganography method with constraint conditions |
CN110324634B (en) * | 2019-07-05 | 2021-10-01 | 中国科学技术大学 | Video steganography method based on motion vector embedding distortion decomposition |
Citations (4)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
KR100734845B1 (en) * | 2005-11-18 | 2007-07-03 | 한국전자통신연구원 | Method for discriminating malicious video using image quality classification and apparatus thereof |
CN103338376A (en) * | 2013-07-02 | 2013-10-02 | 中国科学技术大学 | Video steganography method based on motion vector |
CN103888773A (en) * | 2014-02-19 | 2014-06-25 | 南京邮电大学 | Video steganography analysis method based on mutual information and motion vectors |
CN104837011A (en) * | 2015-05-04 | 2015-08-12 | 中国科学院信息工程研究所 | Content self-adaptive video steganalysis method |
-
2016
- 2016-03-25 CN CN201610177535.5A patent/CN105872555B/en not_active Expired - Fee Related
Patent Citations (4)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
KR100734845B1 (en) * | 2005-11-18 | 2007-07-03 | 한국전자통신연구원 | Method for discriminating malicious video using image quality classification and apparatus thereof |
CN103338376A (en) * | 2013-07-02 | 2013-10-02 | 中国科学技术大学 | Video steganography method based on motion vector |
CN103888773A (en) * | 2014-02-19 | 2014-06-25 | 南京邮电大学 | Video steganography analysis method based on mutual information and motion vectors |
CN104837011A (en) * | 2015-05-04 | 2015-08-12 | 中国科学院信息工程研究所 | Content self-adaptive video steganalysis method |
Non-Patent Citations (2)
Title |
---|
基于灰色关联分析的H_264_AVC视频隐写算法;张英男 等;《武汉大学学报(理学版)》;20141231;第60卷(第6期);全文 |
基于运动矢量分量差的视频隐写算法;王珏 等;《计算机工程》;20111231;第37卷(第23期);全文 |
Also Published As
Publication number | Publication date |
---|---|
CN105872555A (en) | 2016-08-17 |
Similar Documents
Publication | Publication Date | Title |
---|---|---|
CN105872555B (en) | A kind of steganalysis algorithm for the insertion of H.264 video motion vector information | |
CN102917227B (en) | Compressive sensing-based adaptive video information hiding method | |
CN105844238A (en) | Method and system for discriminating videos | |
CN103544498A (en) | Video content detection method and video content detection system based on self-adaption sampling | |
CN104519361A (en) | Video steganography analysis method based on space-time domain local binary pattern | |
CN104837011B (en) | Content self-adaptive video steganalysis method | |
CN103747271A (en) | Video tamper detection method and device based on mixed perceptual hashing | |
CN106127667A (en) | A kind of jpeg image self adaptation steganography method | |
CN103747254A (en) | Video tamper detection method and device based on time-domain perceptual hashing | |
CN103037212A (en) | Adaptive block compressing sensing image coding method based on visual perception | |
CN107027051B (en) | A kind of video key frame extracting method based on linear dynamic system | |
CN104301733A (en) | Video steganalysis method based on feature fusions | |
Miller et al. | Detection theory for graphs | |
CN103747255A (en) | Video tamper detection method and device based on airspace perceptual hashing | |
CN104636764A (en) | Image steganography analysis method and device | |
CN103067713A (en) | Method and system of bitmap joint photographic experts group (JPEG) compression detection | |
CN103903271B (en) | Image forensics method for natural image and compressed and tampered image based on DWT | |
Consolini et al. | Permutation entropy analysis of complex magnetospheric dynamics | |
CN105913008B (en) | Based on the assumption that the crowd's accident detection method examined | |
Wang et al. | A posterior evaluation algorithm of steganalysis accuracy inspired by residual co-occurrence probability | |
CN103888773A (en) | Video steganography analysis method based on mutual information and motion vectors | |
CN106303696B (en) | H.264 video content authenticating water-mark method based on pivot information distribution proportion stability | |
Ren et al. | Student behavior detection based on YOLOv4-Bi | |
Huang et al. | An experimental study on the security performance of YASS | |
CN106791871B (en) | A kind of hiding detection method of motion vector modulation intelligence |
Legal Events
Date | Code | Title | Description |
---|---|---|---|
C06 | Publication | ||
PB01 | Publication | ||
C10 | Entry into substantive examination | ||
SE01 | Entry into force of request for substantive examination | ||
GR01 | Patent grant | ||
GR01 | Patent grant | ||
CF01 | Termination of patent right due to non-payment of annual fee |
Granted publication date: 20190115 |
|
CF01 | Termination of patent right due to non-payment of annual fee |