CN107040786B - A kind of H.265/HEVC video steganalysis method adaptively selected based on space-time characteristic of field - Google Patents
A kind of H.265/HEVC video steganalysis method adaptively selected based on space-time characteristic of field Download PDFInfo
- Publication number
- CN107040786B CN107040786B CN201710145997.3A CN201710145997A CN107040786B CN 107040786 B CN107040786 B CN 107040786B CN 201710145997 A CN201710145997 A CN 201710145997A CN 107040786 B CN107040786 B CN 107040786B
- Authority
- CN
- China
- Prior art keywords
- motion vector
- time
- domain
- frame
- airspace
- 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
- 238000000034 method Methods 0.000 title claims abstract description 40
- 239000013598 vector Substances 0.000 claims abstract description 231
- 239000000284 extract Substances 0.000 claims abstract description 18
- 230000031068 symbiosis, encompassing mutualism through parasitism Effects 0.000 claims abstract description 8
- 238000000605 extraction Methods 0.000 claims description 12
- 239000008186 active pharmaceutical agent Substances 0.000 claims description 7
- 239000000203 mixture Substances 0.000 claims description 4
- 230000006835 compression Effects 0.000 claims description 3
- 238000007906 compression Methods 0.000 claims description 3
- 241000208340 Araliaceae Species 0.000 claims 1
- 235000005035 Panax pseudoginseng ssp. pseudoginseng Nutrition 0.000 claims 1
- 235000003140 Panax quinquefolius Nutrition 0.000 claims 1
- 235000008434 ginseng Nutrition 0.000 claims 1
- XLYOFNOQVPJJNP-UHFFFAOYSA-N water Substances O XLYOFNOQVPJJNP-UHFFFAOYSA-N 0.000 claims 1
- 238000005259 measurement Methods 0.000 abstract description 4
- 238000012795 verification Methods 0.000 abstract description 4
- 238000005516 engineering process Methods 0.000 abstract description 2
- 238000010586 diagram Methods 0.000 description 7
- 238000010276 construction Methods 0.000 description 4
- 238000003780 insertion Methods 0.000 description 4
- 230000037431 insertion Effects 0.000 description 4
- 230000007812 deficiency Effects 0.000 description 3
- 230000005856 abnormality Effects 0.000 description 2
- 230000003044 adaptive effect Effects 0.000 description 2
- 230000008901 benefit Effects 0.000 description 2
- 230000008859 change Effects 0.000 description 2
- 238000001514 detection method Methods 0.000 description 2
- 238000012986 modification Methods 0.000 description 2
- 230000004048 modification Effects 0.000 description 2
- 230000001568 sexual effect Effects 0.000 description 2
- 238000012706 support-vector machine Methods 0.000 description 2
- 238000012360 testing method Methods 0.000 description 2
- 238000012549 training Methods 0.000 description 2
- 238000004458 analytical method Methods 0.000 description 1
- 230000009286 beneficial effect Effects 0.000 description 1
- 238000004364 calculation method Methods 0.000 description 1
- 230000000295 complement effect Effects 0.000 description 1
- 238000013461 design Methods 0.000 description 1
- 230000008569 process Effects 0.000 description 1
- 238000012545 processing Methods 0.000 description 1
- 238000013139 quantization Methods 0.000 description 1
- 239000004576 sand Substances 0.000 description 1
- 238000006467 substitution reaction Methods 0.000 description 1
- 230000009466 transformation Effects 0.000 description 1
- 230000000007 visual effect Effects 0.000 description 1
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
Abstract
The invention discloses a kind of H.265/HEVC video steganalysis method adaptively selected based on space-time characteristic of field, steps are as follows: decoding video extracts the compressed domains such as coding unit division, motion vector to P frame；Motion vector scanning chain is generated, symbiosis frequency abstraction airspace motion vector correlative character is utilized；According to the motion-vector prediction technology in H.265/HEVC, information calculate the time domain prediction motion vector of current prediction unit by same position prediction unit and its at a distance from reference frame etc.；It is poor that time domain prediction motion vector and current motion vector are made, and extracts relativity of time domain feature；Candidate video frame is chosen according to the relationship of motion complexity and time-space domain correlation integrated value；Adaptively selected airspace or time-domain motion vector correlative character are as final classification feature；Finally trained and Classification and Identification.The present invention utilizes the motion-vector prediction feature in H.265/HEVC, creatively adaptively selected to the progress of spatially and temporally motion vector correlative character, effectively improves steganalysis verification and measurement ratio.
Description
Technical field
The present invention relates to the Steganalysis fields in video compress domain, and in particular to one kind is adaptive based on space-time characteristic of field
The H.265/HEVC video motion vector steganalysis method that should be selected.
Background technique
Steganography and steganalysis (or Stego-detection) are the important branch of information security field.Digital video is because of it
Data volume is big, can accommodate the feature more than secret letter quantity, becomes ideal steganography carrier.It H.265/HEVC is newest Video coding
Standard, from prior-generation H.264/AVC in using the macro block of fixed size as coding basic unit it is different, H.265/HEVC allow to adopt
With coding unit (Coding Unit, CU) not of uniform size, including 64 × 64,32 × 32,16 × 16 and 8 × 8.To texture compared with
Complex region generallys use lesser encoding block, and generallys use biggish encoding block to flat site, this makes Video coding
Mode is more flexible, can combine accuracy and code rate.Compared with H.264/AVC standard, under identical visual quality, code stream
Length can be reduced half.It has H.265/HEVC gradually applied in the various products of internet, therefore has studied with it as load at present
The steganography and steganalysis algorithm of body have important theory significance and practical application value.
Video information steganography is to be inserted into secret letter or be hidden into normal video, therefore can from the angle of information theory
At being to add noise to normal video, video distortion is caused to increase, while reducing phase of the video content in time domain and airspace
Guan Xing.Specifically, the correlation of adjacent pixel values in same frame is made on airspace to be reduced；And in the time domain, due to time domain
The presence of prediction, noise is but also the correlation of adjacent interframe pixel value reduces.Common steganography can substantially be transported from modification
These three aspects of transformation coefficient after dynamic vector, prediction mode and quantization are set about, and the present invention is mainly discussed to swear for modification movement
The Stego-detection technology of amount.Modifying influence of the motion vector to video can be from prediction residual, weight compressed encoding and correlation three
The variation of aspect embodies: secret letter insertion may make motion vector be directed toward other reference blocks, and prediction residual is caused to increase；And
The motion vector the modified trend that oriented original motion vector restores after weight compressed encoding；The noise meeting that secret letter insertion introduces
Reduce airspace motion vector correlation and time-domain motion vector correlation.Therefore, based on the steganalysis method of motion vector
Also mainly related in prediction residual, weight compressed encoding and/or motion vector to embedding close video by comparing not embedding close video
Property these three aspects on statistical nature difference, identify that whether there is or not embedding close using the method for pattern classification.In steganalysis most often
Pattern classifier includes support vector machines (Support Vector Machine, SVM).
There is Su et al. in 2011 in SIGNAL currently based on the typical steganalysis method of related sexual abnormality
Paper " the A Video Steganalytic Algorithm Against Motion- delivered on PROCESSING periodical
Vector-Based Steganography ", they think that the insertion of secret letter can change the statistical property of motion vector, and utilize
Motion vector histograms central moment extracts feature from time domain and airspace respectively, and finally time domain and spatial feature are connected, and it is final to be formed
Characteristic of division vector carry out Classification and Identification as the input of SVM.Lina WANG in 2014 et al. delivers in " electronic letters, vol "
Paper " the H.264/AVC video motion vector steganalysis algorithm based on related sexual abnormality " proposes that steganography operation can destroy airspace
The correlation of upper adjacent motion vectors, and four-way Motion vector scanning chain is designed according to the characteristics of H.264 predicting piecemeal, it extracts
Its symbiosis frequecy characteristic constructs svm classifier feature vector.The deficiency of above two method is not accounting for spatially and temporally to transport
Relationship between dynamic vector correlation, the former is only simply connected these two types of features, and airspace fortune is only utilized in the latter
Dynamic vector correlative character.Studies have shown that in moving more violent video frame, when airspace motion vector correlation is often better than
Domain motion vector correlation；And in moving more slow video frame, time-domain motion vector correlation is typically much stronger than airspace movement
Vector correlation.The method of Su et al. spatially and temporally feature will simply be connected, and obvious deficiency is feature vector dimension
Increase, computational complexity increases；And the method for Lina WANG et al. has only selected spatial feature, does not make full use of and moves in time domain
The exception information of vector.Since H.264/AVC and H.265/HEVC video steganography has certain similitude, the present invention is for upper
Deficiency is stated, a kind of steganalysis method adaptively selected towards the space-time characteristic of field of H.265/HEVC video is proposed, it can root
According to the content and coding characteristic of video, the feature that reflection motion vector changes in adaptively selected time domain and airspace, construction SVM spy
Levy vector.So far, it not yet appears in the newspapers for the adaptively selected steganalysis method of space-time characteristic of field of H.265/HEVC video
Road.
Summary of the invention
The purpose of the present invention is in view of the above shortcomings of the prior art, provide one kind adaptively to select based on space-time characteristic of field
The H.265/HEVC video motion vector steganalysis method selected.This method being capable of adaptively selected airspace or time-domain motion vector
Correlative character composition characteristic vector effectively improves steganalysis verification and measurement ratio in the case where not increasing intrinsic dimensionality, reduces
The time of classifier training and classification.
The purpose of the present invention can be achieved through the following technical solutions:
A kind of H.265/HEVC video motion vector steganalysis method adaptively selected based on space-time characteristic of field, it is described
Method the following steps are included:
Step 1, by video code flow entropy decoding to compression domain, its coding unit is read for all P frames and predicting unit is drawn
Point mode, with position prediction unit, motion vector, motion vector residual error these compressed domains；
Step 2 extracts airspace motion vector correlative character: for each P frame, according to its predicting unit division side
Each motion vector is added in the four-way Motion vector scanning chain SS of airspace by formula, in airspace four-way Motion vector scanning chain SS
In, the component being utilized respectively on both horizontally and vertically forms two chains, the Pearson correlation coefficient between two chains is calculated, and
The direction for selecting Pearson correlation coefficient absolute value small carries out feature extraction, by the party in the four-way Motion vector scanning chain SS of airspace
In upward motion vector components deposit feature extraction object MS, and airspace motion vector four-way is obtained as difference to its adjacent position
Difference chain DS extracts symbiosis frequecy characteristic to DS to get airspace motion vector correlative character vector CS is arrived；
Step 3 obtains time domain prediction motion vector: it is pre- to obtain its same position in encoded frame to each predicting unit
Unit is surveyed, according to current prediction unit at a distance from its reference frame, with position prediction unit at a distance from its reference frame and same position
The motion vector computation of predicting unit goes out the time domain prediction motion vector of current prediction unit；
Step 4 extracts time-domain motion vector correlative character: time domain prediction motion vector that step 3 is obtained and current
It is poor that motion vector is made, and is stored in time domain four-way Motion vector scanning chain ST according to the division mode of its predicting unit, respectively benefit
Two chains are formed with the component on both horizontally and vertically, calculate Pearson correlation coefficient, and select Pearson correlation coefficient exhausted
To object of the small direction as feature extraction is worth, component in this direction is stored in time-domain motion vector four-way difference chain DT
In, symbiosis frequecy characteristic is extracted to get time-domain motion vector correlative character vector CT to DT；
Step 5, selection candidate video frame: using the motion complexity G of the calculated for pixel values present frame of consecutive frame, for every
One G value has a corresponding threshold value λ, using motion vector residual computations spatial correlation integrated value P at that time, if P is greater than threshold
Value λ, it is determined that the frame is candidate video frame, and enters step 6, if P is less than or equal to threshold value λ, not as candidate video frame, directly
It is its final feature that airspace motion vector correlative character is selected in selecting, and enters step 7；
Step 6 adaptively chooses spatially and temporally motion vector correlative character: for candidate video frame, utilizing movement
0 ratio calculates the size of airspace motion vector correlation and time-domain motion vector correlation in vector four-way difference chain, respectively
Ratio shared by calculating 0 in airspace motion vector four-way difference chain and time-domain motion vector four-way difference chain, if airspace movement arrow
Amount correlation is stronger, i.e., 0 proportion is higher in airspace motion vector four-way difference chain, then selects airspace motion vector correlation
Feature be final feature, if time-domain motion vector correlation is stronger, i.e., in time-domain motion vector four-way difference chain 0 proportion compared with
Height then selects time-domain motion vector correlative character for final feature；
The feature input classifier of extraction is trained and is classified by step 7.
Further, in the step 2, construct airspace four-way Motion vector scanning chain, include top edge, lower edge,
Four scan chains of left edge and right hand edge, building method are will to be in coding tree unit respectively on the basis of coding tree unit
Top edge, lower edge, left edge and right hand edge motion vector be added in order top edge scan chain, lower edge scan chain,
In left edge scan chain and right hand edge scan chain, and the scan chain for forming that length is L that joined end to end, due to H.265/HEVC
In, the size of each coding tree unit is constant, therefore has one-to-one close between the adjacent motion vectors in same edge
System, is not influenced by prediction block sizes, two scannings that the horizontal and vertical component composition length using motion vector is L-1
Chain calculates it Pearson correlation coefficient respectively, and feature is extracted in the direction for selecting Pearson correlation coefficient absolute value small.
Further, in the step 3, H.265/HEVC middle time domain prediction motion vector is not directly by same position prediction list
The motion vector of member obtains, but passes through what the flexible adjustment of corresponding ratio obtained, if current prediction unit and reference frame away from
It is tb at a distance from its reference frame with position prediction unit, the motion vector with position prediction unit is Col_MV, then currently from for td
The time domain prediction motion vector of predicting unit is (Col_MV × td)/tb.
Further, in the step 4, using the method construct time domain four-way Motion vector scanning chain similar with step 2
With calculating Pearson correlation coefficient.
Further, in the step 5, select not embedding close frame for candidate video frame, because in never embedding close video frame
The stronger feature of correlation, which is extracted, as final feature can improve classification performance, improve classification accuracy rate, whether judge present frame
There are a kind of corresponding relationships: motion complexity for the motion complexity and time-space domain correlation for being the frame for the foundation of candidate video frame
Spatial correlation is weaker at that time for higher video frame, and spatial correlation is stronger at that time for the low frame of motion complexity, and embedding close operation
It will lead to that time-space domain correlation is obviously reduced but motion complexity is basically unchanged, the method for the invention is to each motion complexity
A threshold value is set, the frame that time-space domain correlation integrated value is greater than threshold value is determined as candidate video frame.
Further, it in the step 6, is measured spatially and temporally using in motion vector four-way difference chain 0 ratio
Motion vector correlation, since motion vector difference chain is to be obtained by adjacent motion vectors as difference, adjacent motion vectors are equal
More, 0 number is more in difference chain, conversely, 0 number is fewer in difference chain.
Compared with the prior art, the invention has the following advantages and beneficial effects:
1, the present invention innovatively mutually ties airspace motion vector correlative character with time-domain motion vector correlative character
It closes, can according to video content adaptive select the stronger feature of correlation as final classification feature, two kinds can be made full use of
The complementary nature of correlation improves steganalysis verification and measurement ratio.
2, the present invention selects the biggish feature of correlation as final point candidate video frame in time domain and airspace
Category feature, the dimension of feature vector and be used only airspace motion vector correlative character or time-domain motion vector correlative character when
Unanimously, can be reduced half compared to the concatenated method characteristic dimension of feature, thus this method have lower complicated classification degree and
Faster classification speed；
3, H.265/HEVC new motion-vector prediction feature is innovatively used in steganalysis by the present invention, by obtaining
The time domain prediction motion vector for taking current prediction unit extracts its time-domain motion vector correlation, when utilization is H.265/HEVC middle
Domain motion vector correlation improves the performance of airspace motion vector correlative character than the stronger feature in H.264/AVC；
4, the present invention need to only be carried out in the decoding end of H.265/HEVC standard, without high encoded of complexity
Journey has computation complexity low, the fast feature of the speed of service.
Detailed description of the invention
Fig. 1 is a kind of H.265/HEVC video motion vector adaptively selected based on space-time characteristic of field of the embodiment of the present invention
The flow diagram of steganalysis method.
Fig. 2 (a) is structurally boundary scan chain schematic diagram of the embodiment of the present invention, and Fig. 2 (b) is that construction lower edge scan chain shows
It is intended to, Fig. 2 (c) is construction left edge scan chain schematic diagram, and Fig. 2 (d) is construction right hand edge scan chain schematic diagram.
Fig. 3 is H.265/HEVC time domain prediction motion vector schematic diagram of the embodiment of the present invention.
Fig. 4 is the ROC curve figure of classification results in the embodiment of the present invention.
Specific embodiment
Present invention will now be described in further detail with reference to the embodiments and the accompanying drawings, but embodiments of the present invention are unlimited
In this.
Embodiment:
It is hidden to present embodiments provide a kind of H.265/HEVC video motion vector adaptively selected based on space-time characteristic of field
Analysis method is write, flow diagram is as shown in Figure 1, be broadly divided into seven steps, including decoding video extracts compressed domain, extracts
Airspace motion vector correlative character obtains time domain prediction motion vector, extracts relativity of time domain feature, chooses candidate video
Frame, adaptively selected airspace or time-domain motion vector correlative character are as final classification feature, trained and Classification and Identification.Below
Implementation process of the invention is discussed in detail as embodiment using the video library that 10 CIF videos form.It is used in embodiment
H.265/HEVC as decoder, carrying close video can be passed through with certain insertion official's test model HM-13.0 by original video
The typical steganography of amount operates to obtain, and the present embodiment is using Aly et al. in IEEE Transactions on Information
Classic paper " the Data Hiding in Motion Vectors of delivered on Forensics and Security periodical
Method in Compressed Video Based On Their Associated Prediction Error " is full to make
The close sample of embedding load.
The first step, decoding video extract compressed domain.
The main purpose of decoding video code stream is that necessary parameter, such as prediction block division side are provided for subsequent steganalysis
The information such as formula, motion vector, motion vector residual error.By taking above-mentioned CIF video as an example, by video binary code stream entropy decoding to compression
Domain obtains the division mode of each coding unit in each P frame, predicting unit, obtains four-way Motion vector scanning for second step
Chain provides judgment basis.In H.265/HEVC, each coding tree unit has a motion-vector field m_acCUMvField,
Motion vector all in the coding tree unit is contained, we can be inclined in coding tree unit according to each predicting unit
Shifting amount obtains motion vector.
Second step extracts airspace motion vector correlative character.
Due to predicting unit division mode multiplicity in H.265/HEVC and not of uniform size, motion vector may be
There are multiple adjacent motion vectors in same direction, causes correlation extraction difficult.The present embodiment is swept using four-way motion vector
The method of chain is retouched to describe the neighborhood relationships of motion vector.H.264/AVC the basic unit encoded in is macro block, and H.265/
It is coding unit that basic unit is encoded in HEVC.Different from H.264/AVC middle macroblock size fixation, H.265/HEVC middle coding is single
Member size be it is variable, can divide to obtain by the coding tree unit of fixed size, thus this method with coding tree unit and
Non-coding unit is benchmark tectonic movement vector scan chain.Since each coding tree unit size is constant, it is in same one side
There is one-to-one relationship between the adjacent motion vectors of edge, do not influenced by prediction block sizes.Such as Fig. 2 (a), Fig. 2 (b), Fig. 2
(c) and shown in Fig. 2 (d), the square that heavy line surrounds is a coding tree unit, and the square that fine line surrounds is a volume
Code unit, the rectangle that dotted line surrounds are a predicting unit, and this method will be in coding tree unit top edge and lower edge respectively
Prediction block, i.e. Fig. 2 (a) are added to top edge scan chain by sequence from left to right with the motion vector in shaded block in Fig. 2 (b)
In lower edge scan chain, coding tree unit left edge and right hand edge prediction block, i.e. shade in Fig. 2 (c) and Fig. 2 (d) will be in
Motion vector in block is added in left edge scan chain and right hand edge scan chain by sequence from top to bottom, to guarantee each
Motion vector simplifies the extraction of correlation with a uniqueness for direction neighborhood.
By the scan chain of four direction by top edge scan chain, lower edge scan chain, left edge scan chain, right hand edge scanning
The sequence of chain joins end to end, and obtaining the Motion vector scanning chain SS that length is L, (all motion vector horizontal components are constituted in note SS
Collection be combined into SSx, the collection that all motion vector vertical components are constituted is combined into SSy), and it is divided into i=1 to L-1 and i=2 is arrived
Two groups of motion vector set of L (i is the position in SS), are utilized respectively motion vector vertical direction and horizontal direction component calculate this
The Pearson correlation coefficient of two groups of motion vector set:
Wherein, rxAnd ryThe Pearson correlation coefficient of motion vector horizontal and vertical direction is respectively indicated, value range is
[-1,1]。SSx,iAnd SSx,i+1Respectively indicate i-th and i+1 motion vector horizontal durection component, SS in SSy,iAnd SSy,i+1
I-th and i+1 motion vector vertical durection component in SS are respectively indicated,WithRespectively such as following formula
Definition:
The present embodiment carries out feature extraction to the lesser direction of Pearson correlation coefficient absolute value, because of fortune in this direction
Dynamic vector component correlations are poor, and steganography is more serious to its related damage.It chooses motion vector direction and generates subsequent characteristics
The rule for extracting object MS is as follows:
It is poor to make to each adjacent motion vectors component in MS, obtains airspace motion vector four-way difference chain DS, it may be assumed that
DSi=MSi-MSi+1 (8)
Symbiosis frequecy characteristic is extracted to motion vector four-way difference chain, i.e., to two in DS at a distance of the motion vector point for l
Amount calculates its joint probability for being respectively equal to n and m:
Wherein n and m takes the integer in [- 2,2], and l takes 1 or 2.Combined according to each exploitation of the above formula to n, m and l general
Rate, and be successively added in the motion vector correlative character vector CS of airspace.
Third step obtains time domain prediction motion vector.
For each predicting unit in present frame, time domain predicted motion vector can be sweared by the movement of same position prediction unit
Amount, with position prediction unit at a distance from its reference frame and the distance between present frame and reference frame are scaled to obtain.Because
H.265/HEVC the thought of object uniform motion is mainly utilized in the time domain prediction of middle motion vector, if it is determined that is separated by certain
The motion vector of the two field pictures of distance, then between this two frame any one frame predicted motion vector can by the frame to this two
A distance with reference to interframe is calculated.As shown in H.265/HEVC time domain prediction motion vector schematic diagram in Fig. 3, this method is logical
The same position prediction unit Col_PU for obtaining current prediction unit is crossed, and according to Col_PU and its reference frame distance tb, Col_PU
Corresponding motion vector Col_MV and current prediction unit and reference frame distance td calculates time domain prediction fortune by following formula
Dynamic vector:
4th step extracts relativity of time domain feature.
It is for the time domain prediction motion vector of acquisition, it is poor with current motion vector work, and according to method shown in Fig. 2
It is deposited into time domain four-way Motion vector scanning chain ST and (remembers that the collection of all motion vector horizontal components compositions in ST is combined into STx,
The collection that all motion vector vertical components are constituted is combined into STy), then it is utilized respectively the horizontal and vertical component institute group of motion vector
At two chains, according to formula (1) (2) calculate Pearson correlation coefficient, be denoted as rtxAnd rty.Select Pearson correlation coefficient exhausted
To object of the lesser direction as feature extraction is worth, motion vector components in this direction are stored in time-domain motion vector four-way
In difference chain DT, it may be assumed that
To time-domain motion vector four-way difference chain DT, time domain symbiosis frequecy characteristic is extracted according to formula (9), obtains time domain phase
Closing property feature vector CT.
5th step chooses candidate video frame.
The purpose that the present embodiment chooses candidate video frame is and to make its correlation high for the not embedding close video frame of determination
Feature is as final classification feature.Some not embedding close video frames airspace motion vector correlation itself is not strong, may be with the close view of load
Frequency frame is very nearly the same, and this frame easilys lead to classification error, to reduce overall verification and measurement ratio.And time-domain motion vector is related
Property and airspace motion vector correlation often have complementarity, and airspace motion vector correlation is better than in the video of motion intense
Time-domain motion vector correlation, in moving slow video, it is related that time-domain motion vector correlation is better than airspace motion vector
Property, therefore overall classification performance can be improved by selecting the feature of high correlation for not embedding close video frame, but answer
Avoid carrying the influence of close video frame as far as possible, because identification can be reduced instead to the feature for carrying close video frame selection high correlation
Rate.
For each P frame, its motion complexity is sought by the pixel value on consecutive frame:
Wherein, k is k-th of P frame, and M and N are the width and height of video frame, in the present embodiment respectively 352 and 288.
fk(i, j) and fk-1(i, j) respectively indicates the pixel value of kth frame and -1 frame of kth at the position (i, j).The G (k) the big, represents view
The movement of frequency frame is more violent, on the contrary then to represent video frame motion gentler.
For each P frame, spatial correlation integrated value at that time is measured by calculating the reciprocal of motion vector residual error mean value:
Wherein avgMVD (i) is the mean value of motion vector residual values in i-th of coding tree unit.CTUNum is institute in P frame
There is the quantity of coding tree unit, is in the present embodiment 30.The average MVD length of each coding tree unit is sought in formula (13)
It is expressed from the next:
Wherein, CTUWidth and CTUHeight respectively indicates the width and height of coding tree unit, in the present embodiment all
The width of j-th of predicting unit in i-th of coding tree unit is respectively indicated for 64, PUWidth (i, j) and PUHeight (i, j)
And height, MVD (i, j) indicate j-th of motion vector residual values in i-th of coding tree unit, including xijAnd yijTwo components,
Its length calculation method is as follows:
After the motion complexity G and time-space domain correlation integrated value P that calculate present frame, if complicated movement angle value be greater than etc.
In 4, then it is assumed that movement is more violent, and relativity of time domain is very poor, does not have referential substantially, directly takes airspace motion vector related
Property feature be final classification feature, otherwise P is compared with the threshold value λ in formula (16), if P, greater than threshold value λ, which is
Candidate video frame simultaneously enters the 6th step, if P is less than or equal to threshold value λ, which directly takes airspace to move not as candidate video frame
Vector correlation feature is final classification feature.
6th step, adaptively selected airspace or time-domain motion vector correlative character are as final classification feature.
For selected candidate video frame, the frame is sought by calculating in its motion vector four-way difference chain 0 ratio
Spatially and temporally motion vector correlation:
Wherein zeronumSAnd zeronumTRespectively spatially and temporally in motion vector four-way difference chain 0 quantity,
Length (DS) and Length (DT) is respectively the total length of spatially and temporally motion vector four-way difference chain.
If RS>=RT, then select airspace motion vector correlative character as final classification feature, if RS<RT, then select
Time-domain motion vector correlative character is as final classification feature.
7th step, trained and Classification and Identification.
The final feature of acquisition input classifier is trained and Classification and Identification.The present embodiment makees 50% video frame
For training sample, in addition 50% video frame is trained and is classified using LIB-SVM as test sample, obtains true positive rate
(carrying all P frames in close video to be detected as carrying the ratio of close video) is 91.165%, and true negative rate (does not carry in close video and owns
P frame is detected as not carrying the ratio of close video) it is 92.272%.In the present embodiment, the ROC curves of classification results as shown in figure 4,
It can be seen that ROC curve is close to the upper left corner, there is preferable classification performance, it was demonstrated that effectiveness of the invention.
The above, only the invention patent preferred embodiment, but the scope of protection of the patent of the present invention is not limited to
This, anyone skilled in the art is in the range disclosed in the invention patent, according to the present invention the skill of patent
Art scheme and its patent of invention design are subject to equivalent substitution or change, belong to the scope of protection of the patent of the present invention.
Claims (6)
1. a kind of H.265/HEVC video motion vector steganalysis method adaptively selected based on space-time characteristic of field, feature
It is, the described method comprises the following steps:
Step 1, by video code flow entropy decoding to compression domain, its coding unit and predicting unit division side are read for all P frames
Formula, with position prediction unit, motion vector, motion vector residual error these compressed domains；
Step 2 extracts airspace motion vector correlative character: will according to its predicting unit division mode for each P frame
Each motion vector is added in the four-way Motion vector scanning chain SS of airspace, in the four-way Motion vector scanning chain SS of airspace, point
Not Li Yong both horizontally and vertically on component form two chains, calculate the Pearson correlation coefficient between two chains, and select
The small direction of Pearson correlation coefficient absolute value carries out feature extraction, by the four-way Motion vector scanning chain SS of airspace in this direction
Motion vector components deposit feature extraction object MS in, and airspace motion vector four-way difference is obtained as difference to its adjacent position
Chain DS extracts symbiosis frequecy characteristic to DS to get airspace motion vector correlative character vector CS is arrived；
Step 3 obtains time domain prediction motion vector: obtaining its same position prediction list in encoded frame to each predicting unit
Member, according to current prediction unit at a distance from its reference frame, with position prediction unit at a distance from its reference frame and same position prediction
The motion vector computation of unit goes out the time domain prediction motion vector of current prediction unit；
Step 4 extracts time-domain motion vector correlative character: the time domain prediction motion vector and current kinetic that step 3 is obtained
It is poor that vector is made, and is stored in time domain four-way Motion vector scanning chain ST according to the division mode of its predicting unit, is utilized respectively water
Component in gentle vertical direction forms two chains, calculates Pearson correlation coefficient, and select Pearson correlation coefficient absolute value
Component in this direction is stored in time-domain motion vector four-way difference chain DT by object of the small direction as feature extraction, right
DT extracts symbiosis frequecy characteristic to get time-domain motion vector correlative character vector CT；
Step 5, selection candidate video frame: using the motion complexity G of the calculated for pixel values present frame of consecutive frame, for each
G value has a corresponding threshold value λ, using motion vector residual computations spatial correlation integrated value P at that time, if P is greater than threshold value λ,
It then determines that the frame is candidate video frame, and enters step 6, if P is less than or equal to threshold value λ, not as candidate video frame, directly select
Selecting airspace motion vector correlative character is its final feature, and enters step 7；
Step 6 adaptively chooses spatially and temporally motion vector correlative character: for candidate video frame, utilizing motion vector
0 ratio calculates the size of airspace motion vector correlation and time-domain motion vector correlation in four-way difference chain, calculates separately
Ratio shared by 0 in airspace motion vector four-way difference chain and time-domain motion vector four-way difference chain, if airspace motion vector phase
Closing property is stronger, i.e., 0 proportion is higher in airspace motion vector four-way difference chain, then selects airspace motion vector correlative character
For final feature, if time-domain motion vector correlation is stronger, i.e., 0 proportion is higher in time-domain motion vector four-way difference chain,
Then select time-domain motion vector correlative character for final feature；
The feature input classifier of extraction is trained and is classified by step 7.
2. a kind of H.265/HEVC video motion vector adaptively selected based on space-time characteristic of field according to claim 1
Steganalysis method, it is characterised in that: in the step 2, construct airspace four-way Motion vector scanning chain, include top edge,
Four lower edge, left edge and right hand edge scan chains, building method are will to be in coding respectively on the basis of coding tree unit
It sets unit top edge, lower edge, left edge and the motion vector of right hand edge is added to top edge scan chain in order, lower edge is swept
It retouches in chain, left edge scan chain and right hand edge scan chain, and the scan chain for forming that length is L that joined end to end, due to
H.265/HEVC in, the size of each coding tree unit is constant, therefore has one between the adjacent motion vectors in same edge
One-to-one correspondence is not influenced by prediction block sizes, is L-1's using the horizontal and vertical component composition length of motion vector
Two scan chains calculate it Pearson correlation coefficient respectively, and the direction for selecting Pearson correlation coefficient absolute value small is extracted special
Sign.
3. a kind of H.265/HEVC video motion vector adaptively selected based on space-time characteristic of field according to claim 1
Steganalysis method, it is characterised in that: in the step 3, H.265/HEVC middle time domain prediction motion vector is not directly by same
The motion vector of position prediction unit obtains, but passes through what the flexible adjustment of corresponding ratio obtained, if current prediction unit and ginseng
The distance for examining frame is td, is tb at a distance from its reference frame with position prediction unit, and the motion vector with position prediction unit is Col_
MV, then the time domain prediction motion vector of current prediction unit is (Col_MV × td)/tb.
4. a kind of H.265/HEVC video motion vector adaptively selected based on space-time characteristic of field according to claim 2
Steganalysis method, it is characterised in that: in the step 4, using the method construct time domain four-way motion vector similar with step 2
Scan chain and calculating Pearson correlation coefficient.
5. a kind of H.265/HEVC video motion vector adaptively selected based on space-time characteristic of field according to claim 1
Steganalysis method, it is characterised in that: in the step 5, select not embedding close frame for candidate video frame, because of never embedding close view
The stronger feature of correlation is extracted in frequency frame as final feature can improve classification performance, improve classification accuracy rate, and judgement is current
Frame whether be candidate video frame foundation be the frame motion complexity and time-space domain correlation there are a kind of corresponding relationships: movement
Spatial correlation is weaker at that time for the higher video frame of complexity, and spatial correlation is stronger at that time for the low frame of motion complexity, and embedding
Close operation will lead to that time-space domain correlation is obviously reduced but motion complexity is basically unchanged, and the method for the invention is to each movement
Complexity sets a threshold value, and the frame that time-space domain correlation integrated value is greater than threshold value is determined as candidate video frame.
6. a kind of H.265/HEVC video motion vector adaptively selected based on space-time characteristic of field according to claim 1
Steganalysis method, it is characterised in that: in the step 6, measure airspace using in motion vector four-way difference chain 0 ratio
With time-domain motion vector correlation, since motion vector difference chain is to be obtained by adjacent motion vectors as difference, adjacent motion vectors
Equal is more, and 0 number is more in difference chain, conversely, 0 number is fewer in difference chain.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201710145997.3A CN107040786B (en) | 2017-03-13 | 2017-03-13 | A kind of H.265/HEVC video steganalysis method adaptively selected based on space-time characteristic of field |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201710145997.3A CN107040786B (en) | 2017-03-13 | 2017-03-13 | A kind of H.265/HEVC video steganalysis method adaptively selected based on space-time characteristic of field |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN107040786A CN107040786A (en) | 2017-08-11 |
| CN107040786B true CN107040786B (en) | 2019-06-18 |
Family
ID=59534447
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201710145997.3A Expired - Fee Related CN107040786B (en) | 2017-03-13 | 2017-03-13 | A kind of H.265/HEVC video steganalysis method adaptively selected based on space-time characteristic of field |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN107040786B (en) |
Families Citing this family (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN109982071B (en) * | 2019-03-16 | 2020-08-11 | 四川大学 | HEVC (high efficiency video coding) dual-compression video detection method based on space-time complexity measurement and local prediction residual distribution |
| CN112135137B (en) * | 2019-06-25 | 2024-04-09 | 华为技术有限公司 | Video encoder, video decoder and corresponding methods |
| CN111462765B (en) * | 2020-04-02 | 2023-08-01 | 宁波大学 | Adaptive audio complexity characterization method based on one-dimensional convolution kernel |
| WO2023130285A1 (en) * | 2022-01-05 | 2023-07-13 | Oppo广东移动通信有限公司 | Method, apparatus and system for predicting temporal motion information, and method, apparatus and system for constructing candidate list |
Citations (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN104519361A (en) * | 2014-12-12 | 2015-04-15 | 天津大学 | Video steganography analysis method based on space-time domain local binary pattern |
| CN106131553A (en) * | 2016-07-04 | 2016-11-16 | 武汉大学 | A kind of video steganalysis method based on motion vector residual error dependency |
Family Cites Families (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| WO2010108024A1 (en) * | 2009-03-20 | 2010-09-23 | Digimarc Coporation | Improvements to 3d data representation, conveyance, and use |
-
2017
- 2017-03-13 CN CN201710145997.3A patent/CN107040786B/en not_active Expired - Fee Related
Patent Citations (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN104519361A (en) * | 2014-12-12 | 2015-04-15 | 天津大学 | Video steganography analysis method based on space-time domain local binary pattern |
| CN106131553A (en) * | 2016-07-04 | 2016-11-16 | 武汉大学 | A kind of video steganalysis method based on motion vector residual error dependency |
Also Published As
| Publication number | Publication date |
|---|---|
| CN107040786A (en) | 2017-08-11 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN107040786B (en) | A kind of H.265/HEVC video steganalysis method adaptively selected based on space-time characteristic of field | |
| Cao et al. | Video steganalysis exploiting motion vector reversion-based features | |
| CN105933711B (en) | Neighborhood optimum probability video steganalysis method and system based on segmentation | |
| CN101478691B (en) | Non-reference evaluation method for Motion Jpeg2000 video objective quality | |
| CN104199627B (en) | Gradable video encoding system based on multiple dimensioned online dictionary learning | |
| CN107079165A (en) | Use the method for video coding and device of prediction residual | |
| CN107197297A (en) | A kind of video steganalysis method of the detection based on DCT coefficient steganography | |
| CN103152578A (en) | H.264 video watermark embedding and extraction method based on mixed coding/decoding | |
| CN102496165A (en) | Method for comprehensively processing video based on motion detection and feature extraction | |
| CN109819260A (en) | Video steganography method and device based on the fusion of multi-embedding domain | |
| CN105657431A (en) | Watermarking algorithm based on DCT domain of video frame | |
| CN104853215B (en) | The video steganography method kept based on motion vector local optimality | |
| CN105979269A (en) | Motion vector domain video steganography method based on novel embedding cost | |
| CN104853186A (en) | Improved video steganalysis method based on motion vector reply | |
| CN101765011B (en) | Method and device for scaling motion estimation | |
| Hu et al. | Optimized spatial recurrent network for intra prediction in video coding | |
| CN105915916B (en) | Video steganalysis method based on the estimation of motion vector distortion performance | |
| CN108171325A (en) | Sequential integrated network, code device and the decoding apparatus that a kind of multiple dimensioned face restores | |
| CN101389032A (en) | Intra-frame predictive encoding method based on image value interposing | |
| CN105721875B (en) | A kind of video motion vector Stego-detection method based on entropy | |
| CN107895355B (en) | Motion detection and image contrast self-adaptive enhancement system and method | |
| CN110324634A (en) | It is a kind of to be embedded in the video steganography method that distortion is decomposed based on motion vector | |
| Qin et al. | An improved method of image denoising based on wavelet transform | |
| CN106101713B (en) | A kind of video steganalysis method based on the optimal calibration of window | |
| CN107888931A (en) | A kind of method using video statistics feature prediction error susceptibility |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| PB01 | Publication | ||
| PB01 | Publication | ||
| SE01 | Entry into force of request for 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 | ||
| CF01 | Termination of patent right due to non-payment of annual fee |
Granted publication date: 20190618 |