CN103108188A - Video steganalysis method based on partial cost non-optimal statistics - Google Patents
Video steganalysis method based on partial cost non-optimal statistics Download PDFInfo
- Publication number
- CN103108188A CN103108188A CN2013100660098A CN201310066009A CN103108188A CN 103108188 A CN103108188 A CN 103108188A CN 2013100660098 A CN2013100660098 A CN 2013100660098A CN 201310066009 A CN201310066009 A CN 201310066009A CN 103108188 A CN103108188 A CN 103108188A
- Authority
- CN
- China
- Prior art keywords
- video
- cost
- motion vector
- piece
- local
- 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.)
- Granted
Links
Images
Abstract
The invention provides a video steganalysis method based on partial cost non-optimal statistics. A fundamental principle of producing a motion vector is deeply excavated in the steganalysis method. The video steganalysis method based on the partial cost non-optimal statistics starts from the local optimum characteristics which should be kept and produced by the motion vector, and thereby the steganalysis method is enabled not to be limited to a particular video coding standard and a coding tool, is applicable to video encoding formats adopting interframe prediction technology such as moving picture expert group (MPEG)-2, MPEG-4 and H.264 and has good and wide generality and a high detection accurate rate. The video steganalysis method based on the partial cost non-optimal statistics can be applied to national security, army, government and enterprise sectors, achieves effective detection and monitoring of video content security, prevents illegal personnel from transferring intelligence information with a large data amount through video data and has important research significance and application values for ensuring the national intelligence safety, preventing enterprise confidential information from being revealed and the like.
Description
Technical field
The present invention relates to digital video information and hide the field, relate in particular to a kind of hidden general steganalysis method of writing of digital video motion vector.
Background technology
Along with the development of video compression technology, the network communications technology and network flow-medium business, video has become the mainstream media that the network information is transmitted gradually.Household video camera, smart mobile phone day by day universal, and the application of various video editing softwares easy and simple to handle, make people can carry out easily the recording of video, montage, and by the Internet communication instrument, as YouTube, the excellent cruel video shared platform that waits is transmitted video data quickly and easily.And meanwhile, because video has definitely large Information hiding redundant space, be secret communication carrier best after image.Information Hiding Techniques is as a kind of information security technology; both can be for the protection of country, the important information of enterprise and copyright owner's interests; simultaneously also may by the lawless person even terroristic organization utilize, reach the purpose of escaping monitoring and propagating illegal information.Realization is carried out detection and the early warning of secret intelligence information transfer behavior to illegal use Information hiding instrument, and is significant for the information safety that guarantees country, army, enterprise.
Steganalysis (Steganalysis) is the countermeasure techniques of Information hiding, by to the intrinsic statistical property of carrier with hiddenly write the analysis that causes the trickle change of carrier characteristics, detect in the carrier informations such as digital picture, audio frequency, video whether have the behavior of secret information and secret communication, thereby reach destruction, the detection of carrying confidential information, the purpose of even extracting secret information.Although yet relatively ripe for the Steganalysis of image-carrier at present, less about the open source literature of video steganalysis.Due to the difference between video and image, make the video Steganalysis can't directly use existing image latent writing analytical technology and method, need to and need the characteristics of the steganographic algorithm of antagonism according to self carrier information characteristic, carry out statistical analysis targetedly.
Utilize motion vector to carry out a kind of principal mode that Information hiding is the current video hidden algorithm.From the data volume angle that information embeds, in compressed video stream, among various data elements, the movable information proportion is only second to the DCT coefficient information; Embed angle analysis to the carrier data quality influence from information, the slight modulation of movable information can be made up by the adjustment of corresponding reference block, this makes video quality after embedding information without any mass loss; Feasibility angle from the information embedding, due to present various video encoding standards, comprise MPEG-2, MPEG-4, H.264 etc., its algorithm is on the determining of motion vector, all taked more open portion, make each system can adopt the algorithm of designed, designed to be optimized, therefore have larger uncertainty on this link, have the primary condition that embeds modulation.Therefore, motion vector becomes the present video hidden important embedding territory of writing, and has extremely important and urgent demand for the Steganalysis research of video motion vector.
Referring to document [1] Y.T.Su, C.Q.Zhang, and C.T.Zhang, " A video steganalytic algorithm against motion-vector-based steganography; " Signal Processing91,1901 – 1909 (2011). wherein take the lead in the motion vector steganalysis algorithm is studied, for the MPEG-2 video, based on there being stronger correlation between adjacent macroblocks motion vector in video local motion zone, the steganalysis algorithm of based on motion vector correlation is proposed.But the method is bad for the hidden detection effect of writing of low embedding rate, and its based on adjacent macroblocks motion vector correlation have certain limitation.
referring to document [2] Y.Cao, X.F.Zhao, and D.G.Feng, " Video steganalysis exploiting motion vector reversion-based features, " IEEE Signal Process.Lett.19, 35 – 38 (2012). a kind of video steganalysis method based on calibration is wherein proposed, think that the hidden motion vector of writing video after calibration has the hidden trend of writing front respective motion vectors size that returns to, utilize this characteristics, by coming the structural feature collection with the unregulated hidden otherness of writing video after comparison calibration, reach and distinguish hidden video and the non-hidden purpose of writing video write.It is effective that the method is write video for hidden under low embedding rate, but because it relies on the cause of calibration, and the correctness of calibration is had higher requirements.
Referring to document [3] Yu Deng, Yunjie Wu, and Linna Zhou, " Digital video steganalysis using motion vector recovery-based features, " Applied Optics, Vol.51, Issue20, pp.4667-4677 (2012). wherein also propose a kind of video steganalysis method based on calibration, realize calibration by Local Polynomial core regression model, and come structural feature by the seesaw difference of vector of comparison calibration.
Above-mentioned video steganalysis method is all hidden the writing of video for MPEG-2 or MPEG-4 standard, because the macroblock partitions mode of MPEG-2 and MPEG-4 video is unified, so above analytical method has a common defective, can't be applicable to exactly the video encoding standard that this class H.264 has multiple macroblock partition pattern.
Steganalysis of the present invention is based on video motion vector generating principle in present various video coding algorithm, and therefore, before the summary of the invention statement, the basic principle that at first motion vector in video code flow is generated is introduced.
The basic contraction principle of digital video is to realize by time redundancy and the spatial redundancy of eliminating interframe at present.Wherein time redundancy refers to exist between adjacent video frames similitude and the correlation of height.In existing video compression standard, adopting inter prediction encoding is a kind of universal method of eliminating the video time redundancy, specific practice is present frame to be divided into a plurality of macro blocks of formed objects, carry out the piece match search in reference video frame, search for the macro block the highest with the macroblock to be encoded similarity, define this macro block and be the prediction piece, the prediction piece is similar to is used as current macroblock to be encoded, the relative displacement of the position of this prediction piece and current macro is called motion vector (motion vector, MV).But the prediction piece also not exclusively is equal to current macroblock to be encoded, has a difference between current macro and prediction piece, and defining this difference is the piece residual error, and the prediction piece will add that the piece residual error just can revert to current macro, and this piece residual error is exactly the spatial redundancy of interframe.Therefore, the coding present frame only needs the MV of each macro block in frame and residual error are encoded and gets final product.In order to reach good compression effectiveness, should lack as much as possible the code bit of MV and residual coding, due to how much weighing with the coding cost of coding code bit, namely the coding cost of each piece should be minimum.The predicted macroblock of search and current macro Optimum Matching in reference frame is so that the process of coding Least-cost is called as estimation (motion estimate, ME), the direct purpose of estimation is to locate optimum motion vector, obtains optimum coding cost by motion vector.Therefore all encoders are all take the minimum code code as target as can be known, and the coding cost of the current macro that produces by estimation should have optimal properties, and this is also the basic principle of steganalysis feature extraction in the present invention.
Above-mentioned Video coding principle belongs to the general character of video encoding standard.According to statistics, only in 2011, adopt on network H.264 that the video of coding standard accounts for 80% of Internet video total amount, present Video coding is to take as the leading factor with standard H.264, previous MPEG-2, MPEG-4 standard and the situation of depositing.And current based on the hidden steganalysis method of writing of video motion vector all only for the video of MPEG-2, MPEG-4 standard, write helplessly to the video that adopts standard H.264 is hidden, this has brought great hidden danger to information security.Therefore, the steganalysis of video not only will have the versatility on steganography method, and the versatility on video encoding standard and coding tools also will be arranged.The present invention mainly solve take H.264, MPEG-4, Moving Picture Experts Group-2 video write as the hidden of carrier, the below tells about H.264 some exclusive encoding characteristics of standard.
Tell about the motion estimation process in standard H.264 as an example of the X264 encoder example.In order to make estimation more accurate, H.264 standard is taked more careful partitioning scheme to the 16x16 macro block, and namely a 16x16 macro block can be divided by 4 kinds of sizes: 1 16x16,2 16x8,2 8x16,4 8x8.These sub-macro blocks that are divided into are called macroblock partition.Wherein, all right further following division of each 8x8 macro block: 1 8x8,2 8x4,2 4x8,4 4x4.The division that 8x8 macro block is carried out is called sub-macroblock partition.The dividing mode of these macroblock partition and sub-macroblock partition is referred to as the macroblock partition pattern, so, in standard H.264, each macroblock partition pattern can become the base unit of motion compensation, any that the size of present encoding piece may be 16x16 in the 4x4, and be no longer only this a kind of size of 16x16 piece.For a P macro block or B macro block, encoder travels through according to a certain method all macroblock partition patterns of this macro block and carries out estimation, tries to achieve corresponding MV and residual error and coding cost thereof, selects most suitable macroblock partition pattern according to the coding cost.the a certain concrete pattern of cutting apart for P macro block or B macro block, the idiographic flow of estimation is at first according to the MV of the adjacent block of present encoding piece, utilize the median prediction method to obtain motion vectors (MVP), then the motion estimation algorithm that sets according to encoder is done the motion-vector search of integer-pel precision to MVP prediction piece pointed in reference frame, the MV of the last integer-pel precision that search obtains to process in reference frame does the motion-vector search of 1/2 pixel precision and the motion-vector search of 1/4 pixel precision, obtain the MV prediction piece pointed of final fraction pixel precision.
Relational language is explained
1) coding cost: what of required bit number when a piece is encoded, the coding cost comprises residual error cost and the Motion Vector Cost of piece, and hereinafter referred is cost, and the computing formula of the cost of a piece is COST
blk=COST (D)+COST (mv), wherein, COST
blkCoding cost for this piece, D is the motion compensated residual value, mv is motion vector, and COST (D) is the coding cost of motion compensated residual value, and COST (mv) is the coding cost of motion vector (referring to the coding cost of motion vector residual values in H264).
2) eight neighborhood motion vectors: establish the motion vector that mv=(h, v) is the Video coding piece, wherein, h and v are respectively horizontal component and the vertical components of mv.If function f (x)=x+k, k ∈ { 1,0,1}, and f (mv)=(f (h), f (v)) arranged, and all results of f (mv) consist of a set C, we call set C '=C-{mv} the eight neighborhood motion vector set of mv, and the motion vector in eight neighborhood motion vector set is called the eight neighborhood motion vectors of mv.
3) N neighborhood motion vector: enlarge the radius of neighbourhood of mv, as becoming d by 1, obtain f (x)=x+k, k ∈ d ... ,-1,0,1 ..., d} can obtain the N neighborhood motion vector of mv, wherein N=(2d+1)
2-1.
4) the non-optimum of local cost (Local Cost Not-Perfect, LCNP): representing that the current coding cost of this piece compares with the coding of its N neighborhood reference position, is not minimum, and namely within subrange, the coding of current block is not optimum.
5) the non-optimum probability of local cost (Rate of Local Cost Not-Perfect, RLCNP): in a video-frequency band to be detected, the number with encoding block of the non-optimal properties of local cost accounts for the ratio of all encoding block numbers in this video-frequency band.When RLCNP is higher, the non-optimum of code efficiency in this video-frequency band is described, RLCNP is lower illustrates that the code efficiency in this video-frequency band is relatively excellent.
6) Cover: original video, not through the hidden video of writing processing
7) Stego: the hidden video of writing, by the hidden video of writing processing
8) calibration: after compressed video bit stream is decoded, obtain decoded YUV raw video image sequence, then, adopt the compression coding parameter identical with former compressed video to compress, generate the calibration rear video.Because motion vector is the temporary variable that video compression generates, therefore, after calibration process, the non-optimum probability of local cost of video is closer to its cover video.
Summary of the invention
The present invention is directed to the hidden characteristics of writing of digital video, start with from the basic principle of estimation, propose the video motion vector steganalysis algorithm based on the non-optimal statistical of local cost.
Technical scheme of the present invention is a kind of video steganalysis method based on the non-optimal statistical of local cost, comprises training process and forecasting process,
The flow process of training process comprises the following steps,
Step 1.1, the input training sample set, the training video that training sample is concentrated comprises hidden video and the non-hidden video of writing write; All training videos that training sample is concentrated are calculated characteristics Δ P all, and arbitrary training video calculated characteristics Δ P is comprised following substep,
Step 1.1.1 as original video X, calculates the non-optimum probability P1 of local cost of original video X with training video;
Step 1.1.2, X calibrates to original video, obtains calibrating video
Step 1.1.3 calculates the calibration video
The non-optimum probability P2 of local cost;
Step 1.1.4, calculated characteristics Δ P=P1-P2;
Step 1.2, the feature Δ P of all training videos of training sample being concentrated by grader trains and draws threshold value T; The flow process of forecasting process comprises the following steps,
Step 2.1 as original video X, is calculated the non-optimum probability P1 of local cost of original video X with video to be measured;
Step 2.2, X calibrates to original video, obtains calibrating video
Step 2.3 is calculated the calibration video
The non-optimum probability P2 of local cost;
Step 2.4, calculated characteristics Δ P=P1-P2, if Δ P＞T thinks that video to be measured is the hidden video of writing, otherwise the hidden video of writing of video right and wrong to be measured.
And, to original video X or calibration video
When calculating the non-optimum probability of local cost as pending video, specific implementation is as follows,
(1) in the pending video of calculating, each has the encoding block B of motion vector
iThe non-optimal value NP of local cost
i, account form is as follows,
A) by piece B
iAnd piece B
iMotion vector mv
i, at piece B
iReference frame in determine piece B
iReference block, be designated as piece
B) computing block B
iWith piece
Between luminance component residual error D
i, and computing block B
iCoding cost Cost
iIt is as follows,
Cost
i＝COST(D
i)+COST(mv
i)
Wherein, COST (D
i) be piece B
iWith piece
Between luminance component residual error D
iThe coding cost, COST (mv
i) be piece B
iThe coding cost of motion vector;
C) respectively to piece B
iN neighborhood motion vector mv '
jCalculate the coding cost Cost of each neighborhood motion vector
ij, 1≤j≤N, wherein minimum value is designated as MIN (Cost
ij); Account form is as follows,
Cost
ij＝CODT(D′
j)+COST(mv′
j)
Wherein, CODT (D '
j) be piece
With neighborhood motion vector mv '
jBetween luminance component residual error D '
jThe coding cost, COST (mv '
j) be neighborhood motion vector mv '
jThe coding cost;
D) computing block B
iThe non-optimal value NP of local cost
iIt is as follows,
(2) calculate the LCNP probability of pending video
M is the encoding block sum that has motion vector in pending video.
The present invention has deeply excavated the basic principle that motion vector produces in the steganalysis method, producing the local optimum feature that should keep from motion vector starts with, thereby make this steganalysis method no longer be confined to a certain specific video encoding standard and coding tools, applicable to MPEG-2, MPEG-4, H.264 wait the video code model that adopts the inter prediction technology, have well and widely versatility and high detection accuracy.This invention can be applicable to national security, army, government and business enterprice sector, the effective examination and controlling of realization to video content safety, prevent that unauthorized person from transmitting the information of big data quantity by video data, for ensureing national intelligence safety and preventing that enterprise's confidential information leakage aspect etc. has important Research Significance and using value.
Description of drawings
Fig. 1 is the training process figure of general steganalysis model.
Fig. 2 is the forecasting process figure of general steganalysis model.
Fig. 3 is the feature extraction flow chart of the embodiment of the present invention.
Fig. 4 is the flow chart with the threshold test feature of the embodiment of the present invention.
Fig. 5 is eight neighborhood motion vector set of the embodiment of the present invention and the schematic diagram that motion vector embeds.
Fig. 6 is the steganalysis testing process schematic diagram of the embodiment of the present invention.
Fig. 7 is the LCNP probability of video of the embodiment of the present invention and the graph of a relation of embedding rate.
Fig. 8 is that the LCNP probability calibrating residual error of stego and the cover of different embedding rates before and after the calibration of the embodiment of the present invention contrasts schematic diagram.
Fig. 9 is the hidden feature schematic diagram of writing of stego under the cover of the embodiment of the present invention and different embedding rate thereof.
Embodiment
A kind of digital video general steganalysis method provided by the invention comprises training process and forecasting process.Training process comprises concentrates all videos to calibrate, extract feature and calculated threshold to training sample; Forecasting process comprises calibrates and extracts feature with the same manner to video to be measured, whether the size of comparative feature value and the threshold value that trains is the hidden video of writing according to comparing to determine this video.It is as follows that steganalysis of the present invention detects thought:
In Video coding, in order to improve compression efficiency, must make the encoding code stream of video as far as possible little.Therefore, in the interframe encode process, be used for should lacking the number of coded bits of motion vector MV and residual block, namely the coding cost of current block should be minimum as far as possible, and this just need to select optimum motion vector in motion estimation process, makes the value of following formula minimum
COST
blk=COST(D)+COST(mv)
Wherein, COST
blkBe the total coding cost of present encoding piece, D and COST (D) are respectively the residual sum residual coding cost between present encoding piece and reference block, mv and COST (mv) are respectively the coding cost (in standard H.264, COST (mv) should be the cost of motion vector residual error) of motion vector and the motion vector of macroblock to be encoded.
The target of video code between frames is to guarantee that the COST of selected MV is minimum.Theoretically, if use the global search algorithm, the MV of estimation acquisition should remain on Least-cost in whole reference frame.But in order to guarantee the computational efficiency of encryption algorithm, often use fast search algorithm to replace the global search algorithm improving search speed in the actual coding process, this just makes the MV that finally obtains local optimum but not global optimum often.
The algorithm that motion vector information is hidden can be summarized as two classes: 1) revise the range value of one or two component of MV according to certain selection rule after estimation/coding, as the LSB method; 2) select another MV to replace original MV according to selection rule in motion estimation process.No matter be which kind of method, all modify and adjust on the optimum MV that Video coding has chosen, all the numerical values recited of the MV of a part of encoding block caused change, the MV that makes local optimum is no longer local optimum, and the local optimum cost that then makes current block is no longer also local optimum.Simultaneously, the coding cost of a piece is in coding side local optimum whether, and this characteristic still keeps in decoding end.
Therefore, in not through the hidden video of writing of motion vector (cover), ratio with encoding block of local optimum MV is larger, namely has the non-optimum of local cost (Local Cost Not-Perfect, the ratio of encoding block LCNP) is lower, and its value maintains in certain scope usually.And in through the hidden video of writing of motion vector (stego), destroy possibly this characteristic of local optimum of motion vector due to the embedding of information, the ratio of encoding block with the non-optimum of local cost is usually larger than cover, these characteristics can be enlightened our design and be distinguished cover and stego based on the video steganalysis method of the non-optimal statistical of local cost, and this is also the source of the thought of video steganalysis method in this patent.Therefore, the present invention proposes by detecting in video to be measured the non-optimum probability P of local cost of each motion vector institute corresponding blocks
NPThereby, analyze this video and whether have that motion vector is hidden to be write.
The latent structure method of steganalysis is as follows: the non-optimum probability of local cost of calculation training video or video to be measured at first; Then this video is decoded and the recompile process, obtain the calibration video; Calculate the non-optimum probability of local cost of calibration video; The non-optimum probability of local cost of video before and after calibration is done difference operation, and its difference is as judging that whether the video motion vector is by the hidden feature of writing.
Below in conjunction with embodiment and accompanying drawing, technical solution of the present invention is described in detail.
Steganalysis method of the present invention adopts the classification mechanism of pattern recognition, is divided into training and prediction two large divisions.The training part as shown in Figure 1, comprise that the training sample set (namely all sample videos, comprise cover and stego) that training video is consisted of carries out feature extraction, trains with grader selected feature afterwards, obtain classification thresholds, grader can adopt existing techniques in realizing; Forecasting process carries out same feature extraction to sample to be tested (being video to be measured) as shown in Figure 2, with classification thresholds, feature is carried out forecast test and obtains final result.Calibration and feature extraction have all been used in the training of embodiment and prediction, and the feature extraction part for original video, is calculated its LCNP probability P 1 as shown in Figure 3; Original video is calibrated, obtained calibrating video, calculate the LCNP probability P 2 of calibration video; Calculated characteristics Δ P=P1-P2; Calibration refers to compressed video bit stream is unziped to the YUV sequence in spatial domain, and obtain the relevant parameter of video compression coding from compressed video bit stream, YUV sequence to video weighs compressed encoding with same compressed encoding parameter, if the motion vector of video was once write by hidden, the motion vector of the video after the calibration and piece residual error will return to the state before embedding so.The process that predicted portions makes a decision the feature of extracting with threshold value is as shown in Figure 4: if Δ P＞T thinks the hidden video of writing, otherwise the hidden video of writing of right and wrong.The below is respectively to the training part of steganalysis and being described in detail of predicted portions.
The training process of steganalysis is in order to obtain the threshold value of LCNP probability characteristics, and this threshold value is used for judging whether hidden writing of video to be measured in forecasting process.During concrete enforcement, threshold value can also rule of thumb directly provide except can obtaining by training.The flow process of training process is as follows:
Step 1.1, all training videos that training sample is concentrated are calculated characteristics Δ P all, and arbitrary training video calculated characteristics Δ P is comprised following substep,
Step 1.1.1, the training video for training sample is concentrated as original video X, calculates the non-optimum of the local cost of its LCNP() probability P 1.
Step 1.1.2, X calibrates to original video, obtains calibrating video
Step 1.1.3 calculates the calibration video
The non-optimum of the local cost of LCNP() probability P 2.
Step 1.1.4, calculated characteristics Δ P=P1-P2.
Step 1.2 is trained all feature Δ P by grader and is drawn threshold value T.
The forecasting process of steganalysis is to video extraction feature to be measured, utilizes the threshold value that training process obtains that feature is judged, draws to predict the outcome.The flow process of forecasting process is as follows:
Step 2.1 with video to be measured, as original video X, is calculated the non-optimum of the local cost of its LCNP() probability P 1.
Step 2.2, X calibrates to original video, obtains calibrating video
Step 2.3 is calculated the calibration video
The non-optimum of the local cost of LCNP() probability P 2.
Step 2.4, calculated characteristics Δ P=P1-P2 is if Δ P＞T think that video to be measured is stego, otherwise video to be measured is cover.
The core novelty of inventive method of the present invention is mainly reflected in latent structure extraction aspect, and the below does a description to principle and the method for latent structure.
If mv=(h, v) is a motion vector, h and v are respectively horizontal component and the vertical components of mv.If f (x)=x+k, k ∈ d ... ,-1,0,1 ... d}, and f (mv)=(f (h) is arranged, f (v)), all results of f (mv) consist of a set C, and the present invention is set C '=C-{mv}(C '=C/{mv}) call the N neighborhood motion vector set of mv, N=(2d+1)
2-1.In the concrete process of implementing, those skilled in the art can select the scope of d value to determine the hunting zone of local optimum, when d=1, are eight neighborhood motion vector set.
referring to document [4] C.Xu, X.Ping, and T.Zhang, " Steganography in compressed video stream, " in Proceedings of IEEE First International Conference on Innovative Computing, Information and Control (IEEE, 2006), pp.269 – 272(2006). referring to document [5] H.A.Aly, " Data hiding in motion vectors of compressed video based on their associated prediction error, " IEEE Trans.Inform.Forensics Security6, 14 – 18 (2011). in the steganography method of document [4] and document [5], the embedding of secret information is to adopt LSB coupling or LSB to replace, actual be to one or two component of original motion vector do add 1 or subtract 1 the operation, this result that motion vector is revised makes unit distance of MV off-target MV after embedding, and (unit distance minimum in standard H.264 is 1/4 pixel distance, minimum is 1/2 pixel distance in MEPG-2 and MEPG-4 standard), MV after namely embedding belongs to eight neighborhood motion vector set of optimum movement vector.
The eight neighborhood motion vector set of MV and example thereof be as shown in Figure 5: in the accompanying drawing left side, center (h, v) is optimum MV, and the zone at (h, v) place is the top left corner pixel position of this motion vector institute corresponding blocks.Suppose that local optimum MV is mv=(h, v), the MV after embedding information is mv ', mv ' ∈ C '.Separately h or v are modified, mv ' may be in (h, v-1), (h, v+1), (h-1, v) or (h+1, v); Revise simultaneously h and v, mv ' may be in (h-1, v-1), (h+1, v-1), (h-1, v+1) or (h+1, v+1) these 4 diagonal positions.In the accompanying drawing right side, expression is when the horizontal component of mv and vertical component embed respectively 1 and-1 simultaneously, and mv has changed over mv ', and result makes the prediction piece P of motion vector points become another and predicts piece P ', and the piece residual error has also become D ' by D.
If piece and mv and mv ' the residual error cost between the prediction piece that points to separately be respectively COST (D) and COST (D '), so obtain
COST＝COST(D)+COST(mv)
COST′＝COST(D′)+COST(mv′)
Because mv is local optimum, mv ' is in again the adjacent locations of mv, so COST＜COST ' is arranged.
Referring to document [6] Y.Cao, X.Zhao, D.Feng, and R.Sheng, " Video steganography with perturbed motion estimation, " Lect.Notes Comput.Sci.6958,193 – 207 (2011). the mv ' that steganography method wherein obtains not necessarily belongs to the eight neighborhood motion vector set of mv, but can solve this problem by the neighborhood scope that enlarges mv, for example make the radius of neighbourhood of mv become d by 1, obtain f (x)=x+k, k ∈ { d, ...,-1,0,1, ..., d}.
In steganalysis, obtain the MV of a piece from compressing video frequency flow to be measured, decoding obtains reconstructed block corresponding to this MV, calculates the residual error cost of reconstructed block and this motion vector prediction piece pointed and the cost of MV, and cost COST obtains encoding; Calculate again the coding cost of reconstructed block and all neighborhood motion vectors prediction piece pointed, obtain other M different COST '
1～M, wherein minimum value be designated as MIN (COST '
1～M).If this MV is not modified, have under high probability COST＜MIN (COST '
1～M); If this MV once was modified, occur COST＞MIN (COST '
1～M) Probability maximum.
Certainly, even MV is not modified, also have COST＞MIN (COST '
1～M) situation; Although MV has been modified, also still may occur COST＜MIN (COST '
1～M) this situation.Reason is as follows: what adopt when 1) choosing the optimal motion vector in motion estimation process is fast search algorithm, and which kind of fast search algorithm all can exist the search blind area, even that is to say according to searching algorithm and found optimum motion vector, also may there be more excellent motion vector and be taken into account because of the limitation of algorithm around it; 2) fast search algorithm is to carry out in certain region of search, if the optimal motion vector that obtains is positioned at the border, region of search, this motion vector can only be optimum in the region of search, also may there be more excellent motion vector in search outside the border, and may obtains searching for more excellent motion vector outside the border to the embedding of optimal motion vector.Through experimental results demonstrate, the probability that these exceptions occur keeps within the specific limits usually.
For example, as shown in accompanying drawing 6 left sides, optimum movement vector (h, v) becomes (h-1, v+1) after revising; As shown in accompanying drawing 6 right sides, try to achieve (h-1 when analyzing, v+1) eight neighborhood motion vectors, calculate current reconstructed block and (h-1, v+1) the coding cost COST between prediction piece pointed and the coding cost COST ' corresponding with eight neighborhood motion vectors, it is not minimum can obtaining (h-1, v+1) corresponding coding cost COST, can judge that (h-1, v+1) is the motion vector of revising.
So when COST＞MIN (COST '
1～M) time, the present invention just thinks that the coding cost of this piece is not local optimum, and all pieces are done same operation, just can count LCNP probability in a video.By experiment, the embedding rate that can draw stego is higher, and the ratio that motion vector is revised is higher, and the ratio of LCNP in whole video is also just larger.As shown in Figure 7,8 hidden videos of writing are arranged, be designated as hidden video 1, hidden video 2, hidden video 3, hidden video 4, hidden video 5, hidden video 6, the hidden video 7 of writing write write write write write write.Each line segment represents a stego video, totally 7 stego videos; Abscissa is the difference embedding rate of same stego, and ordinate is the LCNP of each stego under difference embedding rate shared ratio in video, can see, when the embedding rate of stego increased, its LCNP probability was linear increment trend basically.
But often can not obtain a corresponding cover of stego, also just can not obtain the LCNP probability in cover.The modification of motion vector only affects compression effectiveness and does not affect the content quality of reconstruction video, use with the same parameter of encoding for the first time reconstruction video is re-started compressed encoding, the present invention is called calibration to this process, the motion vector of calibration gained will have the trend that reverts to the original motion vector, i.e. the motion vector that calibration generates is similar with cover to a great extent.So in this steganalysis method, the present invention is similar to the method for calibrating and obtains cover, the stego after namely stego being calibrated through decoding, recodification.Prove by experiment, stego after calibration and cover have very large similitude, as shown in Figure 8, to be cover comprise from stego(under different embedding rates that cover does not calibrate to abscissa, the stego under cover calibration, 25%, 50%, 75%, 100% embedding rate), ordinate is LCNP probability separately, can see, stego under each embedding rate is roughly the same with the shared ratio of the LCNP probability of cover after through calibration.
Experiment shows, different types of cover video is before and after calibration, the difference of LCNP probability is usually in a minimum scope, and the LCNP probability after the LCNP probability of stego and calibration subtracts each other, and its difference is compared with the difference of the LCNP probability of cover calibration front and back and more easily presented linear relationship.As shown in Figure 9, be 7 cover videos and this cover formed stego under difference embedding rate, the difference tendency of they LCNP probability before and after calibration has been described in figure.Abscissa is the difference embedding rate of cover and stego, ordinate is the difference of LCNP probability before and after calibration, can find out, the difference of the LCNP probability of cover before and after calibration is very little, along with the increase of embedding rate, the difference of the LCNP probability of stego before and after calibration is linear increment also.Therefore, as long as within first by training, the difference of the LCNP probability of cover before and after calibrating being limited to threshold value T, obtain again the difference of the LCNP probability of video to be measured before and after calibration, stego or cover by the video to be measured that relatively just can judge to it and T, so the present invention deducts the difference of the LCNP probability of calibrating video to the LCNP probability of video as feature.
Be more than the description to LCNP probability characteristics aufbauprinciple, the below is the method for embodiment latent structure.
It is consistent that all training videos that the step 1 of embodiment pair training sample is concentrated extract respectively the realization of feature Δ P, step 2 couple video extraction feature Δ P to be measured.Continuous a plurality of frame of video are called a video-frequency band in time, and the video-frequency band of LCNP probability to be asked for is designated as pending video, and pending video comprises P frame and B frame, for arbitrary piece B with motion vector in pending video
i(piece B
iSize may be any one in the Video coding macroblock mode, as 4x4,1≤i≤M, M is the encoding block sum that has motion vector in pending video, be the number sum of piece in all P frames and B frame in pending video, lower with), establishing its motion vector is mv
i, piece B
iReference block be
, C
iGet 0 or 1 expression mv
iWhether write by hidden, the computational process of the LCNP probability of this frame of video is as follows:
One, calculate the LCNP probability P 1 of original video X, be about to original video X and be designated as pending video, carry out following steps:
2) calculate each in pending video and have the encoding block B of motion vector
iThe non-optimal value NP of local cost
i, 1＜i＜M wherein, NP
i{ 0,1}, M have the number of the encoding block of motion vector to ∈ in pending video, M is the number sum of piece in all P frames and B frame in original video X herein; When the current motion vector of this piece is the local optimum cost, NP
i=0, otherwise, NP
i=1.Piece B
iThe non-optimal value NP of local cost
iComputational methods as follows:
A) by piece B
iAnd mv
i, determine its reference block in the reference frame of this piece
B) calculate B
iWith
Between Y component (being luminance component) residual error D
i, and calculate B
iThe coding cost:
Cost
i＝COST(D
i)+COST(mv
i)
Wherein, COST (D
i) be piece B
iWith piece
Between luminance component residual error D
iThe coding cost, COST (mv
i) be piece B
iThe coding cost of motion vector;
C) respectively to B
iN neighborhood motion vector mv '
j(1≤j≤N) calculate the coding cost of each neighborhood motion vector wherein, wherein minimum value is designated as MIN (Cost
ij); Account form is as follows,
Cost
ij＝CODT(D′
j)+COST(mv′
j)
Wherein, CODT (D '
j) be piece
With neighborhood motion vector mv '
jBetween luminance component residual error D '
jThe coding cost, COST (mv '
j) be neighborhood motion vector mv '
jThe coding cost;
D) calculate the non-optimal value NP of local cost of this piece
i:
3) calculate the LCNP probability P 1 of pending video
Two, original video X is realized calibration, generate the calibration video
X decodes to original video, and it is decoded as the YUV sequence, then, adopts identical encryption algorithm (as H.264, MEPG-4, MPEG-2 etc.) again to compress, and becomes compressed bit stream, namely calibrates video
Three, calculate the calibration video
LCNP probability P 2
To the calibration video
Decode, according to the method for step 1, calculate the calibration video
LCNP probability P 2.Namely at the calibration video
Middle execution 1) calculate the encoding block B that each has motion vector
iThe non-optimal value NP of local cost
i, then according to NP
iObtain calibrating video
The LCNP probability
, M namely calibrates video herein
In the number sum of piece in all P frames and B frame.
Four, calculate Δ P=P1-P2, the LCNP feature of Δ P as original video X.
Contrast to existing steganalysis algorithm compare, the present invention has following advantage:
1) highly versatile.Because the steganalysis method of inventing combines the process of the basic principle of estimation and motion vector embedding, fundamentally grasped the hidden essence of writing of based on motion vector, so no matter be MPEG-2, MPEG-4 or video H.264, because the basic thought of its estimation is consistent, method in the present invention is all applicable, therefore has versatility widely.Although be to adopt H.264 that the video of standard performs an analysis in this manual, also can be applied on other video standards.
2) intrinsic dimensionality is low.Because the LCNP probability of video has reflected the hidden essence of writing of motion vector deeply, only just can effectively distinguish stego and cover with one-dimensional characteristic, greatly reduce complexity.
3) verification and measurement ratio is high.Choose the motion vector steganographic algorithm of three kinds of videos in list of references [4], [5], [6] in experimentation, selected different steganalysis methods to detect to the video of its different embedding rates.Contrast by result, can find out obviously that the video steganalysis method based on the non-optimal statistical of local cost is having obvious advantage than other several steganalysis methods aspect the verification and measurement ratio of low embedding rate video.Therefore, the LCNP probability can effectively reflect the hidden characteristics of writing of the motion vector of digital video, and the present invention is significant at the party's mask.
Specific embodiment described herein is only to the explanation for example of the present invention's spirit.Those skilled in the art can make various modifications or replenish or adopt similar mode to substitute described specific embodiment, but can't depart from spirit of the present invention or surmount the defined scope of appended claims.
Claims (2)
1. video steganalysis method based on the non-optimal statistical of local cost is characterized in that: comprise training process and forecasting process,
The flow process of training process comprises the following steps,
Step 1.1, the input training sample set, the training video that training sample is concentrated comprises hidden video and the non-hidden video of writing write; All training videos that training sample is concentrated are calculated characteristics Δ P all, and arbitrary training video calculated characteristics Δ P is comprised following substep,
Step 1.1.1 as original video X, calculates the non-optimum probability P1 of local cost of original video X with training video;
Step 1.1.2, X calibrates to original video, obtains calibrating video
Step 1.1.3 calculates the calibration video
The non-optimum probability P2 of local cost;
Step 1.1.4, calculated characteristics Δ P=P1-P2;
Step 1.2, the feature Δ P of all training videos of training sample being concentrated by grader trains and draws threshold value T;
The flow process of forecasting process comprises the following steps,
Step 2.1 as original video X, is calculated the non-optimum probability P1 of local cost of original video X with video to be measured;
Step 2.2, X calibrates to original video, obtains calibrating video
Step 2.3 is calculated the calibration video
The non-optimum probability P2 of local cost;
Step 2.4, calculated characteristics Δ P=P1-P2, if Δ P＞T thinks that video to be measured is the hidden video of writing, otherwise the hidden video of writing of video right and wrong to be measured.
2. according to claim 1 based on the video steganalysis method of the non-optimal statistical of local cost, it is characterized in that: to original video X or calibration video
When calculating the non-optimum probability of local cost as pending video, specific implementation is as follows,
(1) in the pending video of calculating, each has the encoding block B of motion vector
iThe non-optimal value NP of local cost
i, account form is as follows,
A) by piece B
iAnd piece B
iMotion vector mv
i, at piece B
iReference frame in determine piece B
iReference block, be designated as piece
B) computing block B
iWith piece
Between luminance component residual error D
i, and computing block B
iCoding cost Cost
iIt is as follows,
Cost
i＝COST(D
i)+COST(mv
i)
Wherein, COST (D
i) be piece B
iWith piece
Between luminance component residual error D
iThe coding cost, COST (mv
i) be piece B
iThe coding cost of motion vector;
C) respectively to piece B
iN neighborhood motion vector mv '
jCalculate the coding cost Cost of each neighborhood motion vector
ij, 1≤j≤N, wherein minimum value is designated as MIN (Cost
ij); Account form is as follows,
Cost
ij＝CODT(D′
j)+COST(mv′
j)
Wherein, CODT (D '
j) be piece
With neighborhood motion vector mv '
jBetween luminance component residual error D '
jThe coding cost, COST (mv '
j) be neighborhood motion vector mv '
jThe coding cost;
D) computing block B
iThe non-optimal value NP of local cost
iIt is as follows,
(2) calculate the LCNP probability of pending video
M is the encoding block sum that has motion vector in pending video.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201310066009.8A CN103108188B (en) | 2013-03-01 | 2013-03-01 | Based on the video steganalysis method of local cost non-optimal statistics |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201310066009.8A CN103108188B (en) | 2013-03-01 | 2013-03-01 | Based on the video steganalysis method of local cost non-optimal statistics |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN103108188A true CN103108188A (en) | 2013-05-15 |
| CN103108188B CN103108188B (en) | 2015-09-02 |
Family
ID=48315716
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201310066009.8A Expired - Fee Related CN103108188B (en) | 2013-03-01 | 2013-03-01 | Based on the video steganalysis method of local cost non-optimal statistics |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN103108188B (en) |
Cited By (13)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN103281473A (en) * | 2013-06-09 | 2013-09-04 | 中国科学院自动化研究所 | General video steganalysis method based on video pixel space-time relevance |
| CN103888773A (en) * | 2014-02-19 | 2014-06-25 | 南京邮电大学 | Video steganography analysis method based on mutual information and motion vectors |
| CN104853186A (en) * | 2015-06-08 | 2015-08-19 | 中国科学院信息工程研究所 | Improved video steganalysis method based on motion vector reply |
| CN104853215A (en) * | 2015-04-17 | 2015-08-19 | 中国科学院信息工程研究所 | Video steganography method based on motion vector local optimality preservation |
| CN105915916A (en) * | 2016-05-12 | 2016-08-31 | 中国科学院信息工程研究所 | Video steganalysis method based on motion vector rate-distortion performance estimation |
| CN105933711A (en) * | 2016-06-23 | 2016-09-07 | 武汉大学 | Partition-based video steganography analysis method and system having neighbourhood optimal probability |
| CN106101713A (en) * | 2016-07-06 | 2016-11-09 | 武汉大学 | A kind of video steganalysis method based on the calibration of window optimum |
| 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 |
| CN106845242A (en) * | 2016-08-26 | 2017-06-13 | 中国科学院信息工程研究所 | A kind of steganographic detection and extracting method based on IS4 software features |
| CN107197297A (en) * | 2017-06-14 | 2017-09-22 | 中国科学院信息工程研究所 | A kind of video steganalysis method of the detection based on DCT coefficient steganography |
| CN107396112A (en) * | 2017-08-01 | 2017-11-24 | 深信服科技股份有限公司 | A kind of coding method and device, computer installation, readable storage medium storing program for executing |
| CN107682703A (en) * | 2017-10-27 | 2018-02-09 | 中国科学院信息工程研究所 | Video steganalysis method based on the detection of inter-frame forecast mode recovery characteristic |
| CN107808100A (en) * | 2017-10-25 | 2018-03-16 | 中国科学技术大学 | For the steganalysis method of fc-specific test FC sample |
Citations (5)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| WO2008089377A2 (en) * | 2007-01-19 | 2008-07-24 | New Jersey Institute Of Technology | A method and apparatus for steganalysis for texture images |
| CN101246722A (en) * | 2008-03-14 | 2008-08-20 | 天津大学 | AVS optical disk duplication control method based on digital watermarking |
| US20100091981A1 (en) * | 2008-04-14 | 2010-04-15 | Yun-Qing Shi | Steganalysis of Suspect Media |
| CN102147913A (en) * | 2011-04-11 | 2011-08-10 | 北京航空航天大学 | Steganalysis method based on image smoothness variation characteristics |
| CN102843576A (en) * | 2012-07-25 | 2012-12-26 | 武汉大学 | Steganography analyzing method aiming at modem-sharing unit (MSU) |
-
2013
- 2013-03-01 CN CN201310066009.8A patent/CN103108188B/en not_active Expired - Fee Related
Patent Citations (5)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| WO2008089377A2 (en) * | 2007-01-19 | 2008-07-24 | New Jersey Institute Of Technology | A method and apparatus for steganalysis for texture images |
| CN101246722A (en) * | 2008-03-14 | 2008-08-20 | 天津大学 | AVS optical disk duplication control method based on digital watermarking |
| US20100091981A1 (en) * | 2008-04-14 | 2010-04-15 | Yun-Qing Shi | Steganalysis of Suspect Media |
| CN102147913A (en) * | 2011-04-11 | 2011-08-10 | 北京航空航天大学 | Steganalysis method based on image smoothness variation characteristics |
| CN102843576A (en) * | 2012-07-25 | 2012-12-26 | 武汉大学 | Steganography analyzing method aiming at modem-sharing unit (MSU) |
Cited By (23)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN103281473B (en) * | 2013-06-09 | 2015-04-15 | 中国科学院自动化研究所 | General video steganalysis method based on video pixel space-time relevance |
| CN103281473A (en) * | 2013-06-09 | 2013-09-04 | 中国科学院自动化研究所 | General video steganalysis method based on video pixel space-time relevance |
| CN103888773A (en) * | 2014-02-19 | 2014-06-25 | 南京邮电大学 | Video steganography analysis method based on mutual information and motion vectors |
| CN104853215B (en) * | 2015-04-17 | 2018-12-28 | 中国科学院信息工程研究所 | The video steganography method kept based on motion vector local optimality |
| CN104853215A (en) * | 2015-04-17 | 2015-08-19 | 中国科学院信息工程研究所 | Video steganography method based on motion vector local optimality preservation |
| CN104853186A (en) * | 2015-06-08 | 2015-08-19 | 中国科学院信息工程研究所 | Improved video steganalysis method based on motion vector reply |
| CN105915916A (en) * | 2016-05-12 | 2016-08-31 | 中国科学院信息工程研究所 | Video steganalysis method based on motion vector rate-distortion performance estimation |
| CN105915916B (en) * | 2016-05-12 | 2019-02-22 | 中国科学院信息工程研究所 | Video steganalysis method based on the estimation of motion vector distortion performance |
| CN105933711A (en) * | 2016-06-23 | 2016-09-07 | 武汉大学 | Partition-based video steganography analysis method and system having neighbourhood optimal probability |
| CN105933711B (en) * | 2016-06-23 | 2019-02-26 | 武汉大学 | Neighborhood optimum probability video steganalysis method and system based on segmentation |
| 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 |
| CN106101713A (en) * | 2016-07-06 | 2016-11-09 | 武汉大学 | A kind of video steganalysis method based on the calibration of window optimum |
| CN106101713B (en) * | 2016-07-06 | 2018-10-09 | 武汉大学 | A kind of video steganalysis method based on the optimal calibration of window |
| CN106845242A (en) * | 2016-08-26 | 2017-06-13 | 中国科学院信息工程研究所 | A kind of steganographic detection and extracting method based on IS4 software features |
| CN106845242B (en) * | 2016-08-26 | 2020-04-14 | 中国科学院信息工程研究所 | Hidden information detection and extraction method based on IS4 software features |
| CN107197297B (en) * | 2017-06-14 | 2019-12-10 | 中国科学院信息工程研究所 | Video steganalysis method for detecting steganalysis based on DCT coefficient steganalysis |
| CN107197297A (en) * | 2017-06-14 | 2017-09-22 | 中国科学院信息工程研究所 | A kind of video steganalysis method of the detection based on DCT coefficient steganography |
| CN107396112A (en) * | 2017-08-01 | 2017-11-24 | 深信服科技股份有限公司 | A kind of coding method and device, computer installation, readable storage medium storing program for executing |
| CN107396112B (en) * | 2017-08-01 | 2021-03-09 | 深信服科技股份有限公司 | Encoding method and device, computer device and readable storage medium |
| CN107808100A (en) * | 2017-10-25 | 2018-03-16 | 中国科学技术大学 | For the steganalysis method of fc-specific test FC sample |
| CN107808100B (en) * | 2017-10-25 | 2020-03-31 | 中国科学技术大学 | Steganalysis method for specific test sample |
| CN107682703A (en) * | 2017-10-27 | 2018-02-09 | 中国科学院信息工程研究所 | Video steganalysis method based on the detection of inter-frame forecast mode recovery characteristic |
| CN107682703B (en) * | 2017-10-27 | 2019-11-26 | 中国科学院信息工程研究所 | Video steganalysis method, device, equipment and computer readable storage medium based on the detection of inter-frame forecast mode recovery characteristic |
Also Published As
| Publication number | Publication date |
|---|---|
| CN103108188B (en) | 2015-09-02 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN103108188B (en) | Based on the video steganalysis method of local cost non-optimal statistics | |
| CN104754355B (en) | A kind of image decoding apparatus | |
| CN110521205A (en) | Sub- predicting unit temporal motion vector prediction for coding and decoding video | |
| CN102783149B (en) | For the adaptive motion resolution of video coding | |
| CN105681795B (en) | For carrying out the method and encoder of Video coding to frame sequence | |
| CN109587479A (en) | Inter-frame prediction method, device and the codec of video image | |
| CN106105215A (en) | Photo current is used as the reference of video coding | |
| TWI621351B (en) | Image prediction decoding device, image prediction decoding method and image prediction decoding program | |
| TW201440502A (en) | Device and method for scalable coding of video information based on high efficiency video coding | |
| Wu et al. | A flexible H. 264/AVC compressed video watermarking scheme using particle swarm optimization based dither modulation | |
| CN112866720B (en) | Motion vector prediction method and device and coder-decoder | |
| EP1997318A1 (en) | Method and apparatus for encoding and decoding the compensated illumination change | |
| He et al. | Detection of double compression in MPEG-4 videos based on block artifact measurement | |
| CN110166771A (en) | Method for video coding, device, computer equipment and storage medium | |
| CN107948646B (en) | A kind of video abstraction generating method and video re-encoding method | |
| CN104853186B (en) | A kind of improved video steganalysis method that is replied based on motion vector | |
| Gorur et al. | Skip decision and reference frame selection for low-complexity H. 264/AVC surveillance video coding | |
| CN105933711A (en) | Partition-based video steganography analysis method and system having neighbourhood optimal probability | |
| Fu et al. | Efficient depth intra frame coding in 3D-HEVC by corner points | |
| CN106101714A (en) | One and the tightly coupled H.264 video information hiding method of compression encoding process | |
| Zhai et al. | Combined and calibrated features for steganalysis of motion vector-based steganography in H. 264/AVC | |
| CN105915916B (en) | Video steganalysis method based on the estimation of motion vector distortion performance | |
| Jin et al. | Plenoptic 2.0 intra coding using imaging principle | |
| Wang et al. | Segmentation based video steganalysis to detect motion vector modification | |
| CN109565600A (en) | Method and apparatus for carrying out image watermarking in Prediction Parameters |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| C06 | Publication | ||
| PB01 | Publication | ||
| C10 | Entry into substantive examination | ||
| SE01 | Entry into force of request for substantive examination | ||
| C14 | Grant of patent or utility model | ||
| GR01 | Patent grant | ||
| CF01 | Termination of patent right due to non-payment of annual fee |
Granted publication date: 20150902 Termination date: 20180301 |
|
| CF01 | Termination of patent right due to non-payment of annual fee |