CN105933711B - Neighborhood optimum probability video steganalysis method and system based on segmentation - Google Patents
Neighborhood optimum probability video steganalysis method and system based on segmentation Download PDFInfo
- Publication number
- CN105933711B CN105933711B CN201610464312.7A CN201610464312A CN105933711B CN 105933711 B CN105933711 B CN 105933711B CN 201610464312 A CN201610464312 A CN 201610464312A CN 105933711 B CN105933711 B CN 105933711B
- Authority
- CN
- China
- Prior art keywords
- video
- feature
- steganalysis
- neighborhood
- fractionation regimen
- 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.)
- Active
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
- H04N19/00—Methods or arrangements for coding, decoding, compressing or decompressing digital video signals
- H04N19/10—Methods or arrangements for coding, decoding, compressing or decompressing digital video signals using adaptive coding
- H04N19/102—Methods or arrangements for coding, decoding, compressing or decompressing digital video signals using adaptive coding characterised by the element, parameter or selection affected or controlled by the adaptive coding
- H04N19/103—Selection of coding mode or of prediction mode
-
- 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
- H04N19/10—Methods or arrangements for coding, decoding, compressing or decompressing digital video signals using adaptive coding
- H04N19/102—Methods or arrangements for coding, decoding, compressing or decompressing digital video signals using adaptive coding characterised by the element, parameter or selection affected or controlled by the adaptive coding
- H04N19/12—Selection from among a plurality of transforms or standards, e.g. selection between discrete cosine transform [DCT] and sub-band transform or selection between H.263 and H.264
- H04N19/122—Selection of transform size, e.g. 8x8 or 2x4x8 DCT; Selection of sub-band transforms of varying structure or type
-
- 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
- H04N19/10—Methods or arrangements for coding, decoding, compressing or decompressing digital video signals using adaptive coding
- H04N19/169—Methods or arrangements for coding, decoding, compressing or decompressing digital video signals using adaptive coding characterised by the coding unit, i.e. the structural portion or semantic portion of the video signal being the object or the subject of the adaptive coding
- H04N19/17—Methods or arrangements for coding, decoding, compressing or decompressing digital video signals using adaptive coding characterised by the coding unit, i.e. the structural portion or semantic portion of the video signal being the object or the subject of the adaptive coding the unit being an image region, e.g. an object
- H04N19/176—Methods or arrangements for coding, decoding, compressing or decompressing digital video signals using adaptive coding characterised by the coding unit, i.e. the structural portion or semantic portion of the video signal being the object or the subject of the adaptive coding the unit being an image region, e.g. an object the region being a block, e.g. a macroblock
-
- 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
- H04N19/90—Methods or arrangements for coding, decoding, compressing or decompressing digital video signals using coding techniques not provided for in groups H04N19/10-H04N19/85, e.g. fractals
- H04N19/94—Vector quantisation
Abstract
The present invention discloses a kind of neighborhood optimum probability video steganalysis method and system based on segmentation, selects one kind from Fractionation regimen set according to video feature;The set that the optimal SAD of neighborhood of all macroblock partition modes is counted to a video frame, calculates the steganalysis feature an of video frame, merges to feature, obtains the final steganalysis feature of a video frame;Accordingly, successively extracting the steganalysis feature of all frames in a video.Detection of the present invention towards video motion vector steganography method, aiming at the problem that traditional steganalysis feature based on prediction residual is interfered vulnerable to quantizing distortion, by macroblock partition to the quantization of quantizing distortion and the combination of neighborhood optimum probability feature, realize to the general of motion vector steganography method and accurately steganalysis.
Description
Technical field
The present invention relates to multi-media safety and digital media processing techniques field, in particular to a kind of discriminating digit video is
The no Steganalysis scheme being embedded in by secret information.
Background technique
Modern Steganography is the technology that confidential corespondence is carried out using Digital Media, and steganalysis is the anti-of Steganography
To detection technique, target is to judge whether be concealed with secret information in the Digital Medias such as image, audio, video.With video
The prevalence of the universal and internet video application of equipment is acquired, digital video becomes the hiding carrier easily obtained;Video carrier
Volume is larger, and enough concealed spaces can be provided for secret information.In recent years the steganography based on digital video and tool by
Cumulative more, this proposes stern challenge to the steganalysis of digital video.
Due to being H.264/AVC current the most widely used video encoding standard, in practical applications very likely at
For video information hiding carrier, therefore main herein consider based on video steganalysis H.264/AVC.
Motion vector (motion vector, MV) and prediction residual are the important parameters in video compression coding.Movement arrow
Amount is the relative coordinate distance in interframe encode between prediction block and present encoding block, and prediction residual is present encoding block and prediction
The difference of block.At Video coding end, motion vector becomes compressed bit stream together with prediction residual, is used to transmit or store;?
Video decoding end can rebuild original video by extracting motion vector and prediction residual from code stream.
The main purpose of Video coding is to keep video code flow few as far as possible under the premise of guaranteeing video quality.In video
Interframe encode in, encoder finds optimal motion vector using estimation (motion estimation, ME), makes
Bit number needed for obtaining encoding motion vector and prediction residual reaches minimum [1].This process usually by rate-distortion optimization model Lai
It realizes, i.e., so that certain Lagrange cost function reaches minimum, the most common Lagrange cost function such as following formula:
J=SAD+ λ BITS (MVD) (formula 1)
Wherein SAD be prediction residual PE absolute error and, MVD is motion vector residual error (motion vector
Difference, MVD), bit number needed for BITS (MVD) represents coding MVD, λ is Lagrange multiplier, and value is usually such as
Under [2]:
Wherein QP is quantization parameter (quantization parameter, QP).
H.264/AVC the full-size of Video coding block is fixed as 16 × 16, referred to as macro block.In order to preferably approach view
Body form in frequency scene and reach more accurate matching effect, macro block generally requires further to be divided during ME
At multiple sub-blocks, such case is referred to as variable block length (variable block size, VBS).H.264/AVC one is provided
16 × 16 macro block can be divided into one 16 × 16 segmentation, two 16 × 8 segmentations, two 8 × 16 segmentations or four 8 × 8
Segmentation, these segmentations are referred to as macroblock partition (MB partitions).8 × 8 segmentation also whistle macro blocks, it can also continue to by
One 8 × 8 segmentation, two 8 × 4 segmentations, two 4 × 8 segmentations or four 4 × 4 segmentations are divided into, it is macro that these segmentations are referred to as son
Block divides (subMB partitions).Above-mentioned macroblock partition mode is as shown in Figure 1.
Video steganography method based on motion vector is the hot spot of video Steganography.Such method passes through directly modification movement
The numerical value of vector simultaneously adjusts corresponding prediction residual simultaneously, to achieve the purpose that be embedded in secret information.
Due to the optimality principle of Video coding, in the coding side of video, for the video (cover) without steganography, most
The SAD of prediction residual corresponding to excellent motion vector usually has lesser numerical value, that is, there is local optimality;And for steganography
Video (stego) after insertion, motion vector become suboptimum by optimal, the SAD of corresponding prediction residual may have compared with
Big numerical value, and its local optimality also changes correspondingly [3] [4].
Current most effective steganalysis is characterized in the local optimality based on prediction residual and constructs [3].Due to view
Quantization step in frequency cataloged procedure is lossy compression, the local optimality of prediction residual and its SAD video coding side and
Decoding end will be different.And the steganalysis feature based on prediction residual can only be constructed from the decoding end of video, therefore such
Feature inevitably will be by the interference of quantizing distortion.In addition, SAD is only utilized in the steganalysis feature based on prediction residual
Local optimality [3] [4], and steganography insertion to the modification of motion vector also change SAD neighborhood optimality statistics rule
Rule, there has been no research and propose related art scheme at present.
Therefore, effective steganalysis feature how is constructed using prediction residual, and reduces quantizing distortion to feature
Interference, to improve the verification and measurement ratio of steganalysis, has great importance for steganalysis.
Bibliography:
[1]I.Richardson,The H.264Advanced Video Compression Standard.New
York,NY,USA:Wiley,2010.
[2]Wiegand T,Schwarz H,Joch A,et al.Rate-constrained coder control
and comparison of video coding standards[J].Circuits and Systems for Video
Technology,IEEE Transactions on,2003,13(7):688-703.
[3]Wang K,Zhao H,Wang H.Video steganalysis against motion vector-
based steganography by adding or subtracting one motion vector value[J]
.Information Forensics and Security,IEEE Transactions on,2014,9(5):741-751..
[4]Ren Y,Zhai L,Wang L,et al.Video steganalysis based on subtractive
probability of optimal matching feature[C]//Proceedings of the 2nd ACM
workshop on Information hiding and multimedia security.ACM,2014:83-90.
Summary of the invention
The present invention is directed to existing the problems of the steganalysis feature based on prediction residual, realizes a kind of robustness
By force, the high general steganalysis feature of accuracy rate provides the neighborhood optimum probability video Steganalysis scheme based on segmentation.
Technical solution of the present invention provides a kind of neighborhood optimum probability video steganalysis method based on segmentation, and feature exists
In: following steps are executed when in steganalysis latent structure,
Step 1, according to video feature, one kind is selected from following Fractionation regimen set,
Step 2, for a video frame, the set of the optimal SAD of neighborhood of all macroblock partition modes is countedIt realizes
It is as follows,
If p is the index of Fractionation regimen, p=1 ..., P, P is the Fractionation regimen number in set of modes selected by step 1, if
First of motion vector is V in certain video framel=(Vl h,Vl v), Δ h and Δ v are V respectivelyl hAnd Vl vKnots modification, local eight neighborhood
It is denoted as
Wherein, l is the index being segmented in pth kind Fractionation regimen;To divide mould in set of modes selected by step 1
Under formula p
Step 3, the steganalysis feature f an of video frame is calculatedp,iIt is as follows,
Wherein, LpIt is the sum of the segmentation in pth kind Fractionation regimen,Indicate setIn element number;IfThen parameterOtherwise
Step 4, according to preset threshold value T, to feature fp,iIt merges, obtains the final steganalysis of video frame
Feature Fp,i,
Step 5, return step 2 handles next video frame, successively extracts the steganography point of all frames in a video
Analyse feature.
A kind of neighborhood optimum probability video steganalysis system based on segmentation, sets in steganalysis latent structure subsystem
It sets and comprises the following modules,
First module, for selecting one kind from following Fractionation regimen set according to video feature,
Second module, for counting the set of the optimal SAD of neighborhood of all macroblock partition modes for a video frameRealization is as follows,
If p is the index of Fractionation regimen, p=1 ..., P, P is the Fractionation regimen number in set of modes selected by the first module
Mesh, if first of motion vector is V in certain video framel=(Vl h,Vl v), Δ h and Δ v are V respectivelyl hAnd Vl vKnots modification, part
Eight neighborhood is denoted as
Wherein, l is the index being segmented in pth kind Fractionation regimen;To divide in set of modes selected by the first module
Under mode p
Third module, for calculating the steganalysis feature f an of video framep,iIt is as follows,
Wherein, LpIt is the sum of the segmentation in pth kind Fractionation regimen,Indicate setIn element number;IfThen parameterOtherwise
4th module is used for according to preset threshold value T, to feature fp,iIt merges, obtains final hidden of a video frame
Write analysis feature Fp,i,
5th module is successively extracted in a video and is owned for ordering the second module to handle next video frame
The steganalysis feature of frame.
The present invention discloses a kind of neighborhood optimum probability video steganalysis latent structure technical solution based on segmentation, towards
The detection of video motion vector steganography method, for traditional steganography based on prediction residual (prediction error, PE)
The problem of feature is interfered vulnerable to quantizing distortion is analyzed, steganalysis feature is constructed in terms of two.Firstly, proposing that a kind of quantization is lost
Genuine quantization method measures the quantizing distortion size of PE using the macroblock partition mode of video;Secondly, according to steganography insertion pair
The change of the optimality of absolute error and (sum of absolute difference, SAD) in motion vector contiguous range,
It is proposed a kind of neighborhood optimum probability feature.By macroblock partition to the quantization of quantizing distortion and the knot of neighborhood optimum probability feature
It closes, realizes to the general of motion vector steganography method and accurately steganalysis.It is an advantage of the present invention that not occurring also at present
Effective method reduces interference of the quantizing distortion to the steganalysis feature based on prediction residual, and the present invention make feature according to
The degree of quantizing distortion is finely divided, and influence of the quantizing distortion to the sensibility of feature can be effectively reduced, which has
Stronger versatility is conducive to the detection effect for improving steganalysis.
Detailed description of the invention
Fig. 1 is macroblock partition and the schematic diagram that macro block is divided.
Fig. 2 is the relation schematic diagram of macroblock partition mode and QP, α.
Fig. 3 is 3 × 3 window schematic diagrames of part of motion vector and prediction residual, and wherein Fig. 3 a is with original motion vector
Centered on 3 × 3 motion vector windows, Fig. 3 b be corresponding 3 × 3SAD window.
Fig. 4 is the relation schematic diagram of macroblock partition mode and neighborhood optimum probability, and wherein Fig. 4 a is the relationship of cover video
Schematic diagram, Fig. 4 b are the relation schematic diagram of stego video.
Fig. 5 is the steganalysis latent structure flow chart of the embodiment of the present invention.
Specific embodiment
Below in conjunction with drawings and examples the present invention will be described in detail technical solution.
The present invention constructs steganography using the local optimality of the SAD of macroblock partition mode and contiguous range interior prediction residual error
Feature is analyzed, and trains steganalysis model and test video sample in conjunction with SVM classifier.The technical side of the embodiment of the present invention
The principle of case is as follows:
1 latent structure method
The quantization method of 1.1 quantizing distortions
1.1.1 the relationship of quantizing distortion and macroblock partition mode
Document [3] assumes that the 2D-DCT coefficient of prediction residual PE obeys following laplacian distribution:
Wherein, y is the 2D-DCT coefficient of prediction residual PE, | y | indicate the absolute value of y, α is the ginseng of laplacian distribution
Number.And further conclude that quantizing distortion and quantization parameter QP correlation, and with the negatively correlated pass of profile parameter
System, wherein α is related with the accuracy of estimation ME.
The compression degree of QP expression video;α represents the distribution shape of 2D-DCT coefficient, and α is smaller, be distributed it is gentler, it is on the contrary
It is then distributed more precipitous.However in the video using fixed compression ratio, the QP of each encoding block is variation;α then with video field
Whether rationally related [3] scape moves speed, Texture complication and searching algorithm, and the two is all not easy to determine and measure.This meaning
Quantizing distortion (SAD local optimality) degree of each encoding block of video variability and uncertainty is presented, if to this
Situation is not distinguish the detection performance that will reduce the feature of the steganalysis based on PE;If can to the range of quantizing distortion into
Row quantization, different degrees of quantizing distortion respectively correspond respective PE feature, are beneficial to improve the effect of steganalysis.Under
How face quantifies research to quantizing distortion.
For formula 3, the present invention it can be concluded that the distribution expectation E (y)=0, the α of variance Var (y)=2/2, can be obtained simultaneously
Second geometric moment E (the y of y2) such as following formula:
As unit of segmentation, if its prediction residual PE is pi,j, the 2D-DCT coefficient symbols y of PE is accordingly denoted as yi,j, i=0,
1 ..., M-1, j=0,1 ..., N-1, wherein M and N distribution is the height and width of segmentation.It is protected according to the energy of orthogonal transformation or norm
Invariance is held, discrete cosine transform is the distribution for changing energy, and converts front and back energy and remain unchanged, therefore can
:
According to formula 4 and formula 5, the present invention can obtain following formulas:
If the mean square error (mean squared error, MSE) of segmentation is denoted as MSE, can obtain:
According to formula 6 and formula 7, the present invention can obtain following formulas:
From formula 8 as can be seen that MSE is equal to pi,jVariance.In addition document [1] point out SAD with MSE monotonic increase, therefore
The present invention it can be concluded that
Wherein, SAD indicate prediction residual PE absolute error and.
QP and MSE and SAD also have certain relationship.For one and same coding block, according to formula 1 and formula 2, QP is bigger, and λ is just
Bigger, MVD proportion just will increase in Lagrange cost function, and which limits the search ranges of MV so that find compared with
Probability for matched reference block becomes smaller, and causes to obtain bigger MSE and SAD.Therefore QP and α may be by MSE and SAD
Roughly measure.
Although MSE can be used and SAD quantifies quantizing distortion, the value range of MSE and SAD itself are equally very
Extensively, MSE or floating number cause operability lower.In addition, present invention discover that between MSE and SAD and the Fractionation regimen of macro block
There is also simple monotonicities, for example the size divided is bigger, and MSE and MAD are (for the SAD of more various sizes of segmentation, with difference
It is worth quadratic sum (mean absolute difference, MAD) to replace, it is the average value of SAD) it is just smaller.To prove to be somebody's turn to do
Saying compares the difference point of akiyo (static state, texture are simple), soccer (motion intense, texture-rich) the two videos below
The ratio of corresponding average MSE, average SAD, segmentation proportion and its local optimality is cut, as shown in Fig. 2, different macro blocks
Segmentation and average MSE, the average MAD, macroblock partition ratio (partition proportion, PP) drawn game in the case of different Q P
Portion's optimum probability (local optimal probability, LOP), wherein (a) is partially akiyo, QP25;(b) partially it is
Akiyo, QP35;It (c) is partially soccer, QP25;It (d) is partially soccer, QP35.
1.1.2 the principle analysis of macroblock partition mode
Relationship between QP and α and macroblock partition mode can also intuitively show that the principle is as follows: since QP is bigger,
ME can tend to reduce the size of MVD code stream, it is clear that select biggish macroblock partition to reduce the quantity of MV and can more save MVD code
Stream.Such as in Fig. 2, having a size of 16 × 16 be segmented in QP value it is larger when, shared by ratio PP it is also bigger.For α,
Other than the relationship of searching algorithm and Fractionation regimen is unobvious, movement is more violent, and texture is more complicated, and lesser segmentation is more suitable
It closes reflection motion state and is more conducive to describe grain details.It is equally the segmentation of small size, soccer view such as in Fig. 2
PP in frequency can be larger.For these reasons, it along with Fractionation regimen operates simple and easy, can be replaced with it
MSE and SAD quantifies the quantizing distortion of encoding block.
The 1.2 neighborhood optimum probability features based on segmentation
1.2.1 the principle of neighborhood optimum probability
If first of motion vector is V in one frame of cover videol=(Vl h,Vl v), wherein Vl hAnd Vl vIt is V respectivelylLevel
Component and vertical component.Since the steganography method based on motion vector is usually one or two component for modifying motion vector
Minimum important position (least significant bit, LSB), then in cover video, motion vector VlIn telescopiny
In the variation of possible value will form 3 × 3 windows, be denoted as
Wherein Δ h and Δ v is V respectivelyl hAnd Vl vKnots modification.If only focusing on the part of motion vector modification, formed
One local eight neighborhood
Correspond to the motion vector set of a possible value
WhereinIndicate not modified original motion vector,Expression was modified
Motion vector.Although MV is only modified in the steganography insertion based on motion vector, it is also required to re-start motion compensation simultaneously
It obtains new PE and brings error to avoid to reconstructed block, thereforeAlso the SAD set of a possible value is corresponded to With?In distribution form it is as shown in Figure 3a and Figure 3b shows, whereinIt indicates not having in cover video modified original
Absolute error and,Indicate the absolute error modified and.
In cover video, due to the local optimality of SAD, often haveThis
KindReferred to as local optimum SAD (LO-SAD);And in stego video, for modifying the MV of LSB, the MV before modification is still
So exist in its eight neighborhood, generally hasThisThe referred to as optimal SAD of neighborhood
(NO-SAD).This is the situation at Video coding end, and in video decoding end, above-mentioned rule is still maintained to a certain extent
[3]。
Accordingly, it can be said that in cover video, the probability of the optimal SAD appearance of neighborhoodWould generally compare
It is smaller;In stego video, probabilityWould generally be bigger, wherein
SAD optimal for the neighborhood of cover video and stego video also has quantity in addition to the difference in existing probability
On difference.In the eight neighborhood of encoding block, the quantity value range of the optimal SAD of neighborhood is 0-8, wherein 0 means the block not
There are the optimal SAD of neighborhood, and there are local optimum SAD.The optimal SAD of neighborhood in cover video is mainly caused by quantizing distortion;
And in stego video the optimal SAD of neighborhood generation, other than the factor for having quantizing distortion, there are also steganography insertion influence.Cause
This, the optimal SAD of neighborhood quantitative distribution in cover video and stego video would also vary from.This conclusion can lead to
Cross it is following experiments have shown that.
For different Fractionation regimens, the optimal SAD of the neighborhood of cover video and stego video is flat on different number
Equal probability is shown in Fig. 4, and experiment uses 36 pairs of cover videos and stego video, and video comes from standard testing video sequence, stego view
The embedded mode of frequency is to modify biggish MV component using LSBM, and the QP of two class videos is 25.It can be with from Fig. 4 a and Fig. 4 b
, it is evident that in stego video, the ratio that the optimal SAD quantity of neighborhood is 0 is less than cover video, and the optimal SAD of neighborhood
The ratio of segmentation of the quantity greater than 0 is significantly larger than cover (neighborhood optimal SAD quantity is especially apparent when being 1-4).In addition it can
See, either cover video or stego video, the optimal SAD quantity of neighborhood be 0 segmentation be all it is most, with segmentation
The reduction of size, the optimal SAD quantity of neighborhood are not that 0 ratio gradually increases, this also demonstrates quantizing distortion in different segmentation moulds
Influence size in formula.
Below by the probability by the number of the optimal SAD of neighborhood in different values, it is based in conjunction with Fractionation regimen to construct
The neighborhood optimum probability (partition based neighborhood optimal probability, PB-NOP) of segmentation is special
Sign.
1.2.2 the extraction of neighborhood optimum probability feature
According to Fig. 1, if P16×16,P16×8,P8×16,P8×8,P8×4,P4×8,P4×4Represent original 7 kinds of segmentations mould of macro block
Formula defines following 4 kinds of new macroblock partition modes, can be described as set of modes:
Wherein,All Fractionation regimens are considered as one kind, are equivalently employed without the influence for considering division size;With two
4 kinds of Fractionation regimens in macroblock partition are synthesized one mode by kind macroblock partition mode, and 3 kinds points in bundle macroblock partition
It cuts mode and synthesizes another mode;According to the size of original Fractionation regimen, 5 kinds of new macroblock partition modes are formed;
Then using 7 kinds of original macroblock partition modes, do not merge.
If the set of the optimal SAD of the local neighborhood individually dividedAre as follows:
Wherein, p is the index of Fractionation regimen, p=1 ..., P, set of modesIn Fractionation regimen numberIt is real
Apply { 1,2,5,7 } i ∈ in example;L is the index being segmented in pth kind Fractionation regimen;For set of modesMiddle segmentation mould
Under formula pIts Fractionation regimen is p;It is then set of modesUnder middle Fractionation regimen p
The then feature f of pth kind Fractionation regimenp,iAre as follows:
Wherein LpIt is the sum of the segmentation in pth kind Fractionation regimen;Indicate setIn element number;IfThen parameterOtherwiseI refers toValue, that is, local eight neighborhood
In the optimal SAD of neighborhood (NO-SAD) number.
There is similar part or neighborhood optimum probability point by the segmentation that Fig. 3 and Fig. 4 can be seen that identical dimensioned area
Cloth, therefore PB-NOP uses Fractionation regimen(additional experiments also turn out that PB-NOP is usedWhen detection effect it is best);In addition lead to
It crosses Fig. 4 and can further be seen that and work asWhen value larger (being greater than 5), fp,iValue it is all minimum, in order to keep feature more compact and mention
The robustness of high feature, merges Partial Feature using characteristic threshold value, the feature F after being mergedp,iExpression side
Formula:
Wherein T is characteristic threshold value, in embodiment, takes T=5, and those skilled in the art can preset value when specific implementation.
Therefore the characteristic dimension of PB-NOP is P (T+1)=30.
The detection of 2 steganalysis
The basic process of steganalysis detection is as follows:
Step 2.1, the video sample of yuv format is inputted, if video is H.264 compressed format, needs to be first converted into YUV
Format.Utilize H.264/AVC video encoder and the steganography tool identical cover sample of difference generation quantity and corresponding
Stego sample.
Step 2.2, the pairs of video sample that 2.1 obtain is randomly divided into the identical two parts of quantity, a part is as instruction
Practice collection, another part verifies the effect of disaggregated model as test set.
Step 2.3, the steganalysis feature of training set and test set sample is extracted according to the method in 1 part.
Step 2.4, using the cover sample characteristics and corresponding stego sample characteristics in training set, and LibSVM is combined
Classifier trains general steganalysis model.
Step 2.5, the detection performance of steganalysis model is verified with the feature of test set sample.
Referring to Fig. 5, the embodiment of the present invention provides a kind of steganalysis latent structure towards video motion vector Steganography
Method, the specific configuration process of the steganalysis feature in step 2.3 the following steps are included:
Step 2.2.1 the characteristics of according to video itself, selects suitable new Fractionation regimen, i.e. set of patterns by formula 13
It closesFor example, should make if video belongs to motion intense and the content with texture complexity comprising moreIt takes biggish
Numerical value, such asWithIt is on the contrary then should makeLesser numerical value is taken, such asWithIt selects in embodiments of the present invention
Fractionation regimenWhen it is implemented, can voluntarily select to set by those skilled in the art.
Step 2.2.2 counts the optimal SAD's of neighborhood of all macroblock partition modes according to formula 14 for a video frame
SetIn embodimentThat is p=1,2 ..., 5, i=0,1 ..., 8.
Step 2.2.3 calculates the steganalysis feature f of a video frame according to formula 15p,i, p=1,2 ..., 5, i=0,
1,…,8。
Step 2.2.4, according to formula 16 and threshold value T, to feature fp,iIt merges, obtains the final steganography of video frame
Analyze feature Fp,i。
Step 2.2.5, return step 2.2.2 handle next video frame, repeat step 2.2.2-2.2.4, successively
Extract the steganalysis feature of all frames in a video.
When it is implemented, computer software mode, which can be used, supports automatic running process.It generally can be first in step 2.2.2
Next video frame is judged whether there is, is, continues process, otherwise terminates.
When it is implemented, modular mode, which can also be used, realizes corresponding system.Embodiment provides a kind of neighbour based on segmentation
Domain optimum probability video steganalysis system is comprised the following modules in the setting of steganalysis latent structure subsystem,
First module, for selecting one kind from following Fractionation regimen set according to video feature,
Second module, for counting the set of the optimal SAD of neighborhood of all macroblock partition modes for a video frameRealization is as follows,
If p is the index of Fractionation regimen, p=1 ..., P, P is the Fractionation regimen number in set of modes selected by the first module
Mesh, if first of motion vector is in certain video frameΔ h and Δ v are respectivelyWithKnots modification, part eight
Neighborhood is denoted as
Wherein, l is the index being segmented in pth kind Fractionation regimen;To divide in set of modes selected by the first module
Under mode p
Third module, for calculating the steganalysis feature f an of video framep,iIt is as follows,
Wherein, LpIt is the sum of the segmentation in pth kind Fractionation regimen,Indicate setIn element number;IfThen parameterOtherwise
4th module is used for according to preset threshold value T, to feature fp,iIt merges, obtains final hidden of a video frame
Write analysis feature Fp,i,
5th module is successively extracted in a video and is owned for ordering the second module to handle next video frame
The steganalysis feature of frame.
Each module specific implementation can be found in corresponding steps, and it will not go into details by the present invention.
Specific embodiment described herein is only an example for the spirit of the invention.The neck of technology belonging to the present invention
The technical staff in domain can make various modifications or additions to the described embodiments or replace by a similar method
In generation, however, it does not deviate from the spirit of the invention or beyond the scope of the appended claims.
Claims (2)
1. a kind of neighborhood optimum probability video steganalysis method based on segmentation, it is characterised in that: in steganalysis feature structure
Following steps are executed when making,
Step 1, if P16×16,P16×8,P8×16,P8×8,P8×4,P4×8,P4×4The original 7 kinds of Fractionation regimens for representing macro block, according to
Video feature selects one kind from following Fractionation regimen set,
Wherein ∪ indicates the union operation of set;
Step 2, for a video frame, the set of the optimal SAD of neighborhood of all macroblock partition modes is countedRealization is as follows,
If p is the index of Fractionation regimen, p=1 ..., P, P is the Fractionation regimen number in set of modes selected by step 1, if certain is regarded
First of motion vector is V in frequency framel=(Vl h,Vl v), Vl hAnd Vl vIt is V respectivelylHorizontal component and vertical component, Δ h and Δ v
It is V respectivelyl hAnd Vl vKnots modification, local eight neighborhood is denoted as
Wherein, l is the index being segmented in pth kind Fractionation regimen;For under Fractionation regimen p in set of modes selected by step 1
's
Step 3, the steganalysis feature f an of video frame is calculatedp,iIt is as follows,
Wherein, LpIt is the sum of the segmentation in pth kind Fractionation regimen,Indicate setIn element number;IfThen parameterOtherwise
Step 4, according to preset threshold value T, to feature fp,iIt merges, obtains the final steganalysis feature of a video frame
Fp,i,
Step 5, return step 2 handles next video frame, and the steganalysis for successively extracting all frames in a video is special
Sign.
2. a kind of neighborhood optimum probability video steganalysis system based on segmentation, it is characterised in that: in steganalysis feature structure
Subsystem setting is made to comprise the following modules,
First module, for setting P16×16,P16×8,P8×16,P8×8,P8×4,P4×8,P4×4Represent original 7 kinds of segmentations mould of macro block
Formula selects one kind from following Fractionation regimen set according to video feature,
Second module, for counting the set of the optimal SAD of neighborhood of all macroblock partition modes for a video frameIt is real
It is now as follows,
If p is the index of Fractionation regimen, p=1 ..., P, P is the Fractionation regimen number in set of modes selected by the first module, if
First of motion vector is V in certain video framel=(Vl h,Vl v), Vl hAnd Vl vIt is V respectivelylHorizontal component and vertical component, Δ h
It is V respectively with Δ vl hAnd Vl vKnots modification, local eight neighborhood is denoted as
Wherein, l is the index being segmented in pth kind Fractionation regimen;For Fractionation regimen in set of modes selected by the first module
Under p
Third module, for calculating the steganalysis feature f an of video framep,iIt is as follows,
Wherein, LpIt is the sum of the segmentation in pth kind Fractionation regimen,Indicate setIn element number;IfThen parameterOtherwise
4th module is used for according to preset threshold value T, to feature fp,iIt merges, obtains final steganography of video frame point
Analyse feature Fp,i,
5th module successively extracts all frames in a video for ordering the second module to handle next video frame
Steganalysis feature.
Priority Applications (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201610464312.7A CN105933711B (en) | 2016-06-23 | 2016-06-23 | Neighborhood optimum probability video steganalysis method and system based on segmentation |
Applications Claiming Priority (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201610464312.7A CN105933711B (en) | 2016-06-23 | 2016-06-23 | Neighborhood optimum probability video steganalysis method and system based on segmentation |
Publications (2)
Publication Number | Publication Date |
---|---|
CN105933711A CN105933711A (en) | 2016-09-07 |
CN105933711B true CN105933711B (en) | 2019-02-26 |
Family
ID=56832050
Family Applications (1)
Application Number | Title | Priority Date | Filing Date |
---|---|---|---|
CN201610464312.7A Active CN105933711B (en) | 2016-06-23 | 2016-06-23 | Neighborhood optimum probability video steganalysis method and system based on segmentation |
Country Status (1)
Country | Link |
---|---|
CN (1) | CN105933711B (en) |
Families Citing this family (6)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN107808100B (en) * | 2017-10-25 | 2020-03-31 | 中国科学技术大学 | Steganalysis method for specific test sample |
CN108965887B (en) * | 2018-06-29 | 2020-05-22 | 中国科学院信息工程研究所 | Video information hiding method and device based on inter-block decoupling |
CN111263157A (en) * | 2020-02-27 | 2020-06-09 | 武汉大学 | Video multi-domain steganalysis method based on motion vector consistency |
CN112312141B (en) * | 2020-08-17 | 2021-08-13 | 中国科学技术大学 | HEVC video steganography method based on pixel adaptive compensation |
CN112738355B (en) * | 2020-12-24 | 2021-10-26 | 华南理工大学 | Non-degraded HEVC video steganography method capable of resisting deep learning network detection |
CN112804533B (en) * | 2021-02-08 | 2022-04-26 | 上海交通大学 | HEVC video steganalysis network method and system based on block information mask |
Citations (2)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN103108188A (en) * | 2013-03-01 | 2013-05-15 | 武汉大学 | Video steganalysis method based on partial cost non-optimal statistics |
CN103338376A (en) * | 2013-07-02 | 2013-10-02 | 中国科学技术大学 | Video steganography method based on motion vector |
Family Cites Families (1)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
US8355165B2 (en) * | 2006-12-05 | 2013-01-15 | Canon Kabushiki Kaisha | Image processing apparatus and control method for restoring color image data |
-
2016
- 2016-06-23 CN CN201610464312.7A patent/CN105933711B/en active Active
Patent Citations (2)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN103108188A (en) * | 2013-03-01 | 2013-05-15 | 武汉大学 | Video steganalysis method based on partial cost non-optimal statistics |
CN103338376A (en) * | 2013-07-02 | 2013-10-02 | 中国科学技术大学 | Video steganography method based on motion vector |
Non-Patent Citations (2)
Title |
---|
基于不确定性推理的JPEG图像通用隐藏信息检测技术;朱婷婷等;《电子学报》;20130215;第41卷(第2期);第233-238页 |
基于相关性异常的H.264/AVC视频运动矢量隐写分析算法;王丽娜等;《电子学报》;20140815;第42卷(第8期);第1457-1464页 |
Also Published As
Publication number | Publication date |
---|---|
CN105933711A (en) | 2016-09-07 |
Similar Documents
Publication | Publication Date | Title |
---|---|---|
CN105933711B (en) | Neighborhood optimum probability video steganalysis method and system based on segmentation | |
CN103002289B (en) | Video constant quality coding device for monitoring application and coding method thereof | |
CN111355956B (en) | Deep learning-based rate distortion optimization rapid decision system and method in HEVC intra-frame coding | |
CN104378643B (en) | A kind of 3D video depths image method for choosing frame inner forecast mode and system | |
CN102801976B (en) | Inter-frame module selecting method based on three-dimensional wavelet video code | |
CN103546758B (en) | A kind of fast deep graphic sequence inter mode decision fractal coding | |
CN108028945A (en) | The apparatus and method of conversion are performed by using singleton coefficient update | |
CN111263157A (en) | Video multi-domain steganalysis method based on motion vector consistency | |
CN109819260A (en) | Video steganography method and device based on the fusion of multi-embedding domain | |
CN105120290A (en) | Fast coding method for depth video | |
CN104853215B (en) | The video steganography method kept based on motion vector local optimality | |
CN105979269B (en) | Motion vector field video steganography method based on novel insertion cost | |
CN109286812A (en) | A kind of HEVC video quality estimation method | |
Fu et al. | Efficient depth intra frame coding in 3D-HEVC by corner points | |
CN105915916B (en) | Video steganalysis method based on the estimation of motion vector distortion performance | |
CN104853186A (en) | Improved video steganalysis method based on motion vector reply | |
CN110246093B (en) | Method for enhancing decoded image | |
CN106791828A (en) | High performance video code-transferring method and its transcoder based on machine learning | |
CN108769696A (en) | A kind of DVC-HEVC video transcoding methods based on Fisher discriminates | |
CN107040786B (en) | A kind of H.265/HEVC video steganalysis method adaptively selected based on space-time characteristic of field | |
CN106131553B (en) | A kind of video steganalysis method based on motion vector residual error correlation | |
CN103546747B (en) | A kind of depth map sequence fractal coding based on color video encoding pattern | |
Wang et al. | Segmentation based video steganalysis to detect motion vector modification | |
CN104093034A (en) | H.264 video streaming self-adaptive error concealing method of similarity face constraining region | |
CN106101713B (en) | A kind of video steganalysis method based on the optimal calibration of window |
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 |