CN107197297A - A kind of video steganalysis method of the detection based on DCT coefficient steganography - Google Patents
A kind of video steganalysis method of the detection based on DCT coefficient steganography Download PDFInfo
- Publication number
- CN107197297A CN107197297A CN201710447336.6A CN201710447336A CN107197297A CN 107197297 A CN107197297 A CN 107197297A CN 201710447336 A CN201710447336 A CN 201710447336A CN 107197297 A CN107197297 A CN 107197297A
- Authority
- CN
- China
- Prior art keywords
- video
- frames
- spatial domain
- steganalysis
- steganography
- 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
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 present invention provides a kind of video steganalysis method of the detection based on DCT coefficient steganography, this method operates the influence to the timely domain correlation in video spatial domain according to steganography of the analysis based on DCT coefficient steganography, and calculating spatial feature collection using DCT cores and embedded cost obtains spatial domain histogram.And the similar macro blocks of interframe are connected by motion vector, construct spatial domain burst and calculate temporal signatures collection and obtain time domain histogram.Above-mentioned spatial domain histogram and time domain histogram are merged into final steganalysis feature set, and are entered into grader and is trained generation steganalysis grader.Video to be measured is extracted into its steganalysis feature as stated above afterwards, and the steganalysis feature is inputted into the steganalysis grader and carries out analysis and distinguishing.This method can steganography video of the effective detection based on DCT coefficient steganography, improve the accuracy rate that steganalysis is especially detected to high-quality video.
Description
Technical field
The present invention relates to the steganalysis (Steganalysis) of the Information hiding subdomains in field of information security technology
Method, more particularly to DCT cores are used in a kind of video steganography of detection based on DCT coefficient steganography of the timely domain construction feature in spatial domain
Analysis method, and application of this method in video steganography of the detection based on DCT coefficient steganography.
Background technology
The purpose of steganalysis is detected in multimedia file with the presence or absence of embedded secret information, effectively containment steganography skill
Abuse of the art in stolen close communication, hiding illegal contact etc..The high speed of current era, video coding technique and high performance network
Development, promotes compression video to become one of maximum multimedia file of disturbance degree.Meanwhile, compression video can be steganography and hidden
Write analysis and sufficient redundant information available is provided, therefore, current grind is become using video as the steganography of carrier and steganalysis
Study carefully focus.
Current video steganography method is generally combined with video compression, by change the motion in compression process to
Amount, inter-frame forecast mode, quantization DCT parameters etc. are embedded in message to realize video steganography.Wherein, the steganography tool based on DCT coefficient
There are high capacity and low complex degree, it is adaptable to the real-time application of steganography.Such as Lin (T.Lin, K.Chung, P.Chang,
Y.Huang,H.Liao,and C.Fang.2013.An improved DCT-based perturbation scheme for
high capacity data hiding in H.264/AVC intra frames.Journal of Systems and
Software 86,3(2013),604–614.),Ma(X.Ma,Z.Li,J.Lv,and W.Wang.2009.Data Hiding
in H.264/AVC Streams with Limited Intra-Frame Distortion Drift.In
2009International Symposium on Computer Network and Multimedia Technology.1–
5；X.Ma,Z.Li,H.Tu,and B.Zhang.2010.A Data Hiding Algorithm for H.264/AVC Video
Streams Without Intra-Frame Distortion Drift.IEEE Transactions on Circuits
And Systems for Video Technology 20, Oct 2010,1320-1330.), Nakajima (K.Nakajima,
K.Tanaka,T.Matsuoka,and Y.Nakajima.2005.Rewritable Data Embedding on MPEG
Coded Data Domain.In 2005IEEE International Conference on Multimedia and
Expo.682-685.), Wong (K.Wong, K.Tanaka, K.Takagi, and Y.Nakajima.2009.Complete
Video Quality-Preserving Data Hiding.IEEE Transactions on Circuits and
Systems for Video Technology 19,10 (Oct 2009), 1499-1512.) etc. proposition video steganography method,
Calculating consumption during video full decoder can be prevented effectively from and recoded, makes the video steganography based on DCT coefficient turn into quick
Easy-to-use steganography method.
For the influence that is caused to the spatial domain of video flowing and relativity of time domain of analysis steganography, a series of steganalysis methods by
Propose, such as Da (Ting Da, ZhiTang Li, and Bing Feng.2015.A Video Steganalysis
Algorithm for H.264/AVC Based on the Markov Features.Springer International
Publishing, Cham, 47-59.), Pankajakshan (V.Pankajakshan and A.Ho.2007.Improving
Video Steganalysis Using Temporal Correlation.In Proceedings of the Third
International Conference on International Information Hiding and Multimedia
Signal Processing,Volume 01.IEEE Computer Society,Washington,DC,USA,287–
290.), Zarmehi (N.Zarmehi and M.Akhaee.2016.Digital video steganalysis toward
Spread spectrum data hiding.IET Image Processing 10,1 (2016), 1-8.) propose method.
Above-mentioned steganalysis method is not to be exclusively used in the video steganalysis method based on DCT coefficient, but it is such hidden to direct detection
The research direction write.In addition, the steganography research based on quantization DCT coefficient is more ripe in image, in jpeg image
Efficient steganalysis algorithm, such as Holub and Fridrich (V.Holub and J.Fridrich.2015.Low-Complexity
Features for JPEG Steganalysis Using Undecimated DCT.IEEE Transactions on
Information Forensics and Security10,2Feb 2015,219–228；V.Holub and
J.Fridrich.2015.Phase-aware projection model for steganalysis of JPEG
Images.2015.) the algorithm of design, the also design for the present invention provides Research Thinking.
The content of the invention
It is an object of the invention to provide a kind of video steganalysis method of the detection based on DCT coefficient steganography, this method is led to
Influence of the analysis steganography operation to the timely domain correlation in video spatial domain is crossed, spatial domain and time domain steganalysis feature set are constructed respectively,
Realize the effective detection to the video steganography method based on DCT coefficient steganography.
To reach above-mentioned purpose, the technical solution adopted in the present invention is：
A kind of video steganalysis method of the detection based on DCT coefficient steganography, its step includes：
1) original video collection is prepared, and it is corresponding using steganographic algorithm generation to be analyzed based on a part of original video collection
Steganography video set；
2) above-mentioned each original video and each steganography video are decoded to K GOP (Group of Pictures) respectively
Unit, wherein K are determined according to the length of video and GOP units；
3) the I frames of each GOP units are decoded and extract spatial feature collection to spatial domain, and to each spatial domain I frames, spatial domain Nogata is obtained
Figure；
4) the P frames/P frames and B frames of each GOP units are decoded, and temporal signatures collection is constructed in decoding process, time domain is obtained straight
Fang Tu；
5) steganalysis feature set is obtained according to above-mentioned spatial domain histogram and time domain histogram；
6) above-mentioned steganalysis feature set input grader is trained, obtains steganalysis grader；
7) according to step 2) to step 5) methods described extracts the steganalysis feature of video to be measured, and be inputted described
Steganalysis grader carries out analysis and distinguishing.
Further, step 1) described in the preparation method of original video collection refer to：User is according to unified compression parameters
Yuv video stream is compressed, one or more original video collection are obtained.
Further, the compression parameters include size, length, resolution ratio.
Further, step 3) described in each spatial domain I frames extract spatial feature collection obtain the histogrammic step in spatial domain
Including：
The convolution of spatial domain I frames and DCT cores 3-1) is calculated, the noise residual error of spatial domain I frames is obtained, and obtain by quantization operation
The quantizing noise residual error of spatial domain I frames；
3-2) according to the existing embedding grammar based on DCT steganographic algorithms, different change coefficients are obtained to macro block spatial domain pixel
The influence of value, and calculate the insertion cost for changing current macro；
3-3) according to the above-mentioned embedded cost of quantizing noise residual sum of above-mentioned spatial domain I frames, calculate spatial feature collection and obtain sky
Domain histogram.
Further, step 4) described in construction temporal signatures collection obtain the histogrammic step of time domain and include：
Similar macro blocks 4-1) are connected by motion vector and reconstruct spatial domain burst；
The convolution of above-mentioned spatial domain burst and DCT cores 4-2) is calculated, the noise residual error of spatial domain burst is obtained, and by quantifying to grasp
Obtain the quantizing noise residual error of P frames/P frames and B frames；
4-3) according to above-mentioned P frames/P frames and the quantizing noise residual sum step 3-2 of B frames) in insertion cost, calculate time domain
Feature set obtains time domain histogram.
Further, step 5) specifically include：Above-mentioned spatial domain histogram is merged according to the symmetry principle of DCT cores and time domain is straight
Fang Tu, obtains spatial feature and temporal signatures after the spatial feature after dimensionality reduction and temporal signatures, connection dimensionality reduction and obtains steganography point
Analyse feature set.
Further, step 6) described in grader be integrated classifier or Gaussian kernel SVMs.
Further, step 7 is repeated several times), finally differentiated average result as foundation.
The present invention is to the beneficial effect in video steganalysis field：The present invention provides a kind of detection and is based on DCT coefficient
The video steganalysis method of steganography, and this method be it is first be exclusively used in detect the video steganalysis based on DCT coefficient steganography
Method.This method operates the influence to the timely domain correlation in video spatial domain according to steganography of the analysis based on DCT coefficient steganography, uses
DCT cores and embedded cost calculate spatial feature collection and obtain spatial domain histogram.And by the similar macro blocks of motion vector connection interframe,
Construct spatial domain burst and calculate temporal signatures collection and obtain time domain histogram.Above-mentioned spatial domain histogram and time domain histogram are merged into
Final steganalysis feature set, and be entered into grader be trained generation steganalysis grader.Afterwards will be to be measured
Video extracts its steganalysis feature as stated above, and the steganalysis feature is inputted into the steganalysis grader progress
Analysis and distinguishing.This method can steganography video of the effective detection based on DCT coefficient steganography, improve steganalysis especially to high-quality
The accuracy rate that video is detected.
Brief description of the drawings
Fig. 1 is video steganalysis method flow chart of a kind of detection based on DCT coefficient steganography that the present invention is provided；
Fig. 2 is spatial feature collection calculation flow chart in the present invention；
Fig. 3 is time domain layered structure schematic diagram in the present invention；
Fig. 4 is sequential movements vector calculating schematic diagram in the present invention；
Fig. 5 is that the present invention connects macro block schematic diagram using motion vector；
Fig. 6 be the inventive method from DCTR analyses feature, Da analysis method under different quantization parameters and embedded rate it is right
The Detection accuracy comparison diagram of Ma steganography method；
Fig. 7 be the inventive method from DCTR analyses feature, Da analysis method under different quantization parameters and embedded rate it is right
The Detection accuracy comparison diagram of Lin steganography method.
Embodiment
To enable the features described above and advantage of the present invention to become apparent, special embodiment below, and coordinate institute's accompanying drawing work
Describe in detail as follows.
A kind of implementation process such as Fig. 1 of video steganalysis method of the detection that the present invention is provided based on DCT coefficient steganography
It is shown.The technical solution adopted in the present invention mainly include the following steps that (unless otherwise specified, following steps by computer and
The software and hardware of electronic equipment is performed)：
1st, original video and steganography video set are prepared.
User compresses yuv video stream according to unified compression parameters (such as size, length, resolution ratio), obtain one or
Multiple original video collection.Based on one group of original video collection, corresponding one group of steganography video is generated using steganographic algorithm to be analyzed
Collection.
2nd, steganalysis feature set is extracted.
Above-mentioned original video and steganography video are decoded into K GOP unit respectively, and (wherein K is according to video and GOP units
Length is determined), the start frame of each GOP units is I frames, and operations described below is performed to k-th of GOP unit (1≤k≤K)：
1) each I frames are decoded to spatial domain, to each spatial domain I frames FkExtract spatial feature collection, its characteristic extraction procedure such as Fig. 2 institutes
Show, comprise the following steps：
A) by seeking FkWith DCT cores G convolution, the noise residual error U (F of spatial domain I frames are calculatedk, G) and={ Fk*G}.And pass through
Quantization operation obtains the quantizing noise residual error U (F of spatial domain I framesk, G, Q) and=Q (U (Fk, G) and/q), wherein q is fixed quantization step
Long, Q is with { 0,1,2 ..., TrBe barycenter quantizer, wherein TrIt is interceptive value.(a) convolution and amount in the step such as Fig. 2
Shown in changing；
B) according to the existing embedding grammar based on DCT steganographic algorithms, different change coefficients are obtained to macro block spatial domain pixel value
Influence ρi,j, and calculate the insertion cost δ (ρ for changing current macroi,j), wherein ρi,jSubscript i and j be respectively macro block it is horizontal seat
The index of mark and ordinate.Shown in (d) macro block classification and (e) cost are calculated in the step such as Fig. 2；
C) for the I frames that resolution ratio is M × N, according to the above-mentioned embedded cost of the quantizing noise residual sum of above-mentioned spatial domain I frames,
Calculate spatial feature collection and obtain spatial domain histogramThe step is as shown in (b) histogram calculation in Fig. 2.
Wherein, χ is the length and width pixel value of macro block；Residual error after u, v are dct transform and quantified indexes coefficient, and 0≤u, v
≤χ-1；Q is quantizer, and its barycenter is { 0,1,2 ..., Tr, TrFor interceptive value；τ is the histogrammic value in spatial domain, and-Tτ
≤τ≤Tτ, wherein Tτ≥0；WithRespectively macro block number of the I frames in spatial domain in vertical and horizontal.
2) the P frames/P frames and B frames of each GOP units are decoded, temporal signatures collection is constructed in decoding process；Wherein GOP units
P frames/P frames and B frames refer to that the GOP units that have include P frames, and some GOP units include P frames and B frames.It includes following step
Suddenly：
A) similar macro blocks are connected by motion vector and reconstructs spatial domain burst Pk.Motion vector for connection needs to meetWherein SAD is the residual absolute value sum between predicted macroblock and original macro, and μ is
SAD standard deviation, Nei defines the range size of neighboring macroblocks.The figure layer of reconstruct is as shown in figure 3, each row is by T and spatial domain I frames
Corresponding similar macro blocks are constituted, and wherein T is the frame number in the GOP units.If there is L connection in the GOP units, reconstruct is empty
Domain burst is made up of L rows altogether；
B) above-mentioned spatial domain burst P is calculatedkWith DCT cores G convolution, spatial domain burst P is obtainedkNoise residual error U (Pk, G), and
Quantified to obtain the quantizing noise residual error U (P of P frames/P frames and B framesk,G,Q)；
C) to the spatial domain burst P by T × L macro block splicingk, calculate temporal signatures collection and obtain time domain histogram
Wherein, the residual error after u, v are dct transform and quantified indexes coefficient, and 0≤u, v≤χ -1；Q is quantizer, its matter
The heart for 0,1,2 ..., Tr, TrFor interceptive value；τ is the histogrammic value of time domain, and-Tτ≤τ≤Tτ, wherein Tτ≥0；δ
(ρi) represent PkInsertion cost with the similar macro blocks connected in a line is identical, is equal to FkThe insertion generation of middle respective macroblock
Valency.
3) spatial domain histogram is merged according to the symmetry principle of DCT cores(in Fig. 2 (c) dimension-reduction treatment institute
Show) and time domain histogramObtain the spatial feature after dimensionality reduction and temporal signatures.Connect the spatial domain after dimensionality reduction special
Temporal signatures of seeking peace can obtain final steganalysis feature set.
To each GOP units, steganalysis feature set is extracted according to the operation of step 2, until currently original regard that be disposed
Frequency collection and steganography video set.
3rd, the training and configuration of steganalysis grader.It will be extracted in step 2 from original video collection and steganography video set
Steganalysis feature set input grader, grader is trained, generate steganalysis grader.
4th, video to be measured is analyzed.Video to be measured is received, the video to be measured is extracted in the operation for being first according to step 2
Steganalysis feature, then will acquisition steganalysis feature input steganalysis grader in analyzed, repeatedly,
Finally differentiated average result as foundation.
A specific embodiment is named to illustrate the present invention.The embodiment is in H.264/AVC video encoding standard
Under to based on DCT coefficient steganography video steganography carry out analysis, it is only analysis method of the present invention in H.264/AVC standard
In application, the effect of this method can be absolutely proved.But proposed by the present invention is a general framework, except the embodiment it
Outside, the steganalysis that this method can be applied under other video compression standards.Therefore other realities that the framework based on the present invention is proposed
Example is applied, protection scope of the present invention is belonged to.
In H.264/AVC video encoding standard, inter prediction provides two kinds of brightness block sizes：4 × 4 and 16 × 16.
Due to 16 × 16 pieces of brightness, generally change is gentle and change of the human eye to brightness value is more sensitive, therefore existing steganography method is only
Consider 4 × 4 pieces of brightness in I frames.In order to reduce computation complexity and realize quick video compress, H.264/AVC standard is adopted
With Integer DCT Transform and quantization operation, the process can formulate and be expressed as RQDCT=ARAT.Wherein, RQDCTFor Integer DCT Transform
And the residual error after quantifying, R is the residual error of macro block, and A is unit orthogonal matrix.And unit orthogonal matrix is expressed as：
In the video frame, the residual block after the dct transform of the i-th row jth row is represented by：
In H.264/AVC standard, 0≤m, n≤3 represent the ranks index of 4 × 4 residual errors；0≤u, v≤3, expression 4 ×
Residual error after 4DCT is converted and quantified indexes coefficient；wuAnd wvDct transform coefficient is represented, works as u, during v=0And
If u ＞ 0, v ＞ 0, then wu=1, wv=1.Based on above-mentioned theory, the present invention propose H.264/AVC in 4 × 4DCT cores, be used for
Spatial domain and the extraction of temporal signatures.
A kind of video steganalysis method for detection DCT coefficient steganography that the present invention is provided, it is in implementation H.264/AVC
Method is mainly included the following steps that：
(1) original video collection and steganography video set are prepared.100 4:2:The standard CIF sequences of 0YUV forms are used for this
Embodiment.The frame per second of video sequence is 30fps, and video length is 100 frames to 2000 frames.Ma steganography method and Lin steganography
Method will be used to be embedded in secret information, and use data bit (the bits embedded per non-being embedded in per nonzero coefficient
Zero coefficient, bpnc) Metric Embedding rate.For the Detection results of test the inventive method in varied situations, the reality
Example is applied respectively using conventional quantization parameter (i.e. QP=28,32,36) compression video, and using different steganography methods different embedding
Enter and steganography (Ma steganography method is carried out to compression video under rate：0.05bpnc, 0.1bpnc；Lin steganography method：
0.05bpnc, 0.1bpnc and 0.15bpnc).
(2) steganography video is decoded, for each GOP units, steganalysis feature set is extracted using the inventive method.
A) spatial feature collection is calculated.Calculate the convolution of spatial domain I frames and DCT cores, the quantizing noise of spatial domain I frames after being quantified
Residual error.Its formula is as follows：
U(Fk, G) and={ U(u,v)|0≤u,v≤3}
U(u,v)=Fk*G(u,v)
U(Fk, G, Q) and=Q (U (Fk,G)/q)
Wherein, G(u,v)For DCT cores, and it is defined as follows：
Different insertion cost set of the change coefficients under are obtained based on DCT coefficient steganographic algorithm according to existing.
δ(ρi,j)={ δ (ρi,j)1st,δ(ρi,j)2st,δ(ρi,j)3st,δ(ρi,j)4st,δ(ρi,j)5st}={ 8Q2,8Q2,8Q2,
4Q2,16Q2}
With reference to convolution operation, calculating obtains spatial domain histogram.
B) temporal signatures collection is calculated.The macro block similar to spatial domain I frames is connected by motion vector, Fig. 4 is refer to, is based on
The triangulo operation relation existed in GOP units between motion vector, the motion vector (Sequential being linked in sequence
Motion Vector, SMV).In frameIn, for macro blockIt can obtainIn frameIn, for macro blockIt can obtainSimilarly in frame FPIn, for macro block MBPIt can obtainPlease
With reference to Fig. 5, during motion vector is connected, if predicted macroblock exceedes the border in reference frame, connection Maximum overlap face
Long-pending macro block.If the size for connecting macro block is not 4 × 4, current macro is divided into the sub-macroblock that size is 4 × 4.Pass through
4 × 4 macro blocks of connection are spliced, spatial domain burst is can obtain.Based on the spatial domain burst, time domain histogram can obtain.
C) symmetry principle based on DCT cores merges spatial domain histogram and time domain histogram, and spatial domain histogram and time domain are straight
The number of square figure is down to 9 from 16, and the dimension of each feature set is 16 × 9 × (Tr+ 1)=144 × (Tr+1)。
(3) train and configure steganalysis grader.In embodiments of the present invention, using integrated classifier and Gaussian kernel
SVMs is classified, and the steganography and non-steganography video of any selection 50% extract steganography point using the method for step (2)
Feature is analysed, and steganalysis feature input grader is trained, steganalysis grader is obtained.
(4) video to be measured is detected.The steganalysis grader completed using training is to remaining 50% video sequence
Row are analyzed, and carrying out feature extraction to the video to be measured first by step (2) (extracts the steganalysis of the video to be measured
Feature), then the steganalysis feature of acquisition is inputted and tested in steganalysis grader.Detection accuracy is true positives
The average value of rate and true negative rate, is repeated 20 times experiment, and the foundation finally differentiated is used as using Average Accuracy.
For the Detection results of relatively more different steganalysis methods, the DCTR during the present embodiment is analyzed using image latent writing divides
The video steganalysis method that analysis method and Da are proposed as the inventive method control group.The Detection results of different analysis methods
Listed in table 1, Fig. 6 is that the inventive method analyzes feature, Da analysis method in different quantization parameters and embedded rate from DCTR
Under to the Detection accuracy comparison diagram of Ma steganography method；Fig. 7 is the inventive method and DCTR analyses feature, Da analysis method
To the Detection accuracy comparison diagram of Lin steganography method under different quantization parameters and embedded rate.
Table 1：Detection results of the different analysis methods under different quantization parameters and embedded rate to Da and Lin steganography method
Learnt by observation table 1, Fig. 6 and Fig. 7, Detection results of the inventive method in three kinds of steganalysis are best, and
In most cases, DCTR analyzes analysis method of the Detection results better than Da of feature.With the reduction of quantization parameter, three kinds
The Detection results of steganalysis largely all increase.This be due to the video quality that is compressed using less quantization parameter compared with
Height, the feature therefrom extracted is also more effective.Further, since the dimension of feature is relatively low, it is confidential than using integrated using supporting vector
The good classification effect of grader.Although for Ma and Lin steganography method, Da and DCTR have certain Detection results, this
Inventive method greatly improves Detection accuracy.In the case where quantization parameter is 28 and embedded rate is 0.15bpnc, detection
Lin steganography method can reach 99.13% accuracy rate.For Ma steganography method, in quantization parameter be 28 and embedded rate is
During 0.1bpnc, accuracy rate is up to 95.81%.
Embodiment in above embodiment, video steganalysis method proposed by the present invention can be to base
Efficiently detected in the video steganalysis method of DCT coefficient steganography, and to the Detection results of high-quality steganography video more
It is good.
Implement to be merely illustrative of the technical solution of the present invention rather than be limited above, the ordinary skill people of this area
Member can modify or equivalent substitution to technical scheme, without departing from the spirit and scope of the present invention, this hair
Bright protection domain should be to be defined described in claims.
Claims (8)
1. a kind of video steganalysis method of the detection based on DCT coefficient steganography, its step includes：
1) original video collection is prepared, and corresponding steganography is generated using steganographic algorithm to be analyzed based on a part of original video collection
Video set；
2) above-mentioned each original video and each steganography video are decoded to K GOP unit respectively, wherein K is according to video and GOP
The length of unit is determined；
3) the I frames of each GOP units are decoded and extract spatial feature collection to spatial domain, and to each spatial domain I frames, spatial domain histogram is obtained；
4) the P frames/P frames and B frames of each GOP units are decoded, and temporal signatures collection is constructed in decoding process, time domain Nogata is obtained
Figure；
5) steganalysis feature set is obtained according to above-mentioned spatial domain histogram and time domain histogram；
6) above-mentioned steganalysis feature set input grader is trained, obtains steganalysis grader；
7) according to step 2) to step 5) methods described extracts the steganalysis feature of video to be measured, and it is inputted the steganography
Analyze grader and carry out analysis and distinguishing.
2. the method as described in claim 1, it is characterised in that step 1) described in the preparation method of original video collection refer to：
User compresses yuv video stream according to unified compression parameters, obtains one or more original video collection.
3. method as claimed in claim 2, it is characterised in that the compression parameters include size, length, resolution ratio.
4. the method as described in claim 1, it is characterised in that step 3) described in spatial feature is extracted to each spatial domain I frames
Collection, which obtains the histogrammic step in spatial domain, to be included：
The convolution of spatial domain I frames and DCT cores 3-1) is calculated, the noise residual error of spatial domain I frames is obtained, and spatial domain is obtained by quantization operation
The quantizing noise residual error of I frames；
3-2) according to the existing embedding grammar based on DCT steganographic algorithms, different change coefficients are obtained to macro block spatial domain pixel value
Influence, and calculate the insertion cost for changing current macro；
3-3) according to the above-mentioned embedded cost of quantizing noise residual sum of above-mentioned spatial domain I frames, it is straight that calculating spatial feature collection obtains spatial domain
Fang Tu.
5. method as claimed in claim 4, it is characterised in that step 4) described in construction temporal signatures collection obtain time domain Nogata
The step of figure, includes：
Similar macro blocks 4-1) are connected by motion vector and reconstruct spatial domain burst；
The convolution of above-mentioned spatial domain burst and DCT cores 4-2) is calculated, the noise residual error of spatial domain burst is obtained, and obtain by quantization operation
To the quantizing noise residual error of P frames/P frames and B frames；
4-3) according to above-mentioned P frames/P frames and the quantizing noise residual sum step 3-2 of B frames) in insertion cost, calculate temporal signatures
Collection obtains time domain histogram.
6. the method as described in claim 1, it is characterised in that step 5) specifically include：Merged according to the symmetry principle of DCT cores
Above-mentioned spatial domain histogram and time domain histogram, the spatial domain obtained after the spatial feature after dimensionality reduction and temporal signatures, connection dimensionality reduction are special
Temporal signatures of seeking peace obtain steganalysis feature set.
7. the method as described in claim 1, it is characterised in that step 6) described in grader be integrated classifier or Gaussian kernel
SVMs.
8. the method as described in claim 1, it is characterised in that step 7 is repeated several times), using average result as according to progress
It is final to differentiate.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201710447336.6A CN107197297B (en) | 2017-06-14 | 2017-06-14 | Video steganalysis method for detecting steganalysis based on DCT coefficient steganalysis |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201710447336.6A CN107197297B (en) | 2017-06-14 | 2017-06-14 | Video steganalysis method for detecting steganalysis based on DCT coefficient steganalysis |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN107197297A true CN107197297A (en) | 2017-09-22 |
| CN107197297B CN107197297B (en) | 2019-12-10 |
Family
ID=59879712
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201710447336.6A Expired - Fee Related CN107197297B (en) | 2017-06-14 | 2017-06-14 | Video steganalysis method for detecting steganalysis based on DCT coefficient steganalysis |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN107197297B (en) |
Cited By (8)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| 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 |
| CN109862395A (en) * | 2019-03-29 | 2019-06-07 | 中国人民解放军战略支援部队信息工程大学 | A kind of video flowing steganographic detection method and apparatus |
| CN110390941A (en) * | 2019-07-01 | 2019-10-29 | 清华大学 | MP3 audio hidden information analysis method and device based on coefficient correlation model |
| WO2021037082A1 (en) * | 2019-08-28 | 2021-03-04 | 上海寒武纪信息科技有限公司 | Method and apparatus for processing data, and related product |
| CN112637605A (en) * | 2020-11-11 | 2021-04-09 | 中国科学院信息工程研究所 | Video steganalysis method and device based on analysis of CAVLC code words and number of nonzero DCT coefficients |
| WO2021097771A1 (en) * | 2019-11-21 | 2021-05-27 | Suzhou Aqueti Technology Co., Ltd. | Ics-frame transformation method and apparatus for cv analysis |
| CN112949352A (en) * | 2019-12-10 | 2021-06-11 | 北京地平线机器人技术研发有限公司 | Training method and device of video detection model, storage medium and electronic equipment |
| CN113556439A (en) * | 2021-06-08 | 2021-10-26 | 中国人民解放军战略支援部队信息工程大学 | Rich Model steganography detection feature selection method based on feature component correlation |
Citations (6)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US20070030996A1 (en) * | 2005-08-02 | 2007-02-08 | Lsi Logic Corporation | Method and/or apparatus for video watermarking and steganography using simulated film grain |
| CN102905134A (en) * | 2012-10-22 | 2013-01-30 | 山东省计算中心 | Adaptive video digital steganography method |
| CN103108188A (en) * | 2013-03-01 | 2013-05-15 | 武汉大学 | Video steganalysis method based on partial cost non-optimal statistics |
| 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 |
| 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 |
| CN106713917A (en) * | 2016-12-05 | 2017-05-24 | 南京航空航天大学 | Video steganography algorithm based on motion vector difference |
-
2017
- 2017-06-14 CN CN201710447336.6A patent/CN107197297B/en not_active Expired - Fee Related
Patent Citations (6)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US20070030996A1 (en) * | 2005-08-02 | 2007-02-08 | Lsi Logic Corporation | Method and/or apparatus for video watermarking and steganography using simulated film grain |
| CN102905134A (en) * | 2012-10-22 | 2013-01-30 | 山东省计算中心 | Adaptive video digital steganography method |
| CN103108188A (en) * | 2013-03-01 | 2013-05-15 | 武汉大学 | Video steganalysis method based on partial cost non-optimal statistics |
| 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 |
| 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 |
| CN106713917A (en) * | 2016-12-05 | 2017-05-24 | 南京航空航天大学 | Video steganography algorithm based on motion vector difference |
Non-Patent Citations (1)
| Title |
|---|
| 邓永，赵险峰等: "一种基于图像特征选择分类器的隐写分析方法", 《第十二届全国信息隐藏暨多媒体信息安全学术大会论文集》 * |
Cited By (11)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| 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 |
| CN109862395A (en) * | 2019-03-29 | 2019-06-07 | 中国人民解放军战略支援部队信息工程大学 | A kind of video flowing steganographic detection method and apparatus |
| CN109862395B (en) * | 2019-03-29 | 2021-05-04 | 中国人民解放军战略支援部队信息工程大学 | Video stream hidden information detection method and device |
| CN110390941A (en) * | 2019-07-01 | 2019-10-29 | 清华大学 | MP3 audio hidden information analysis method and device based on coefficient correlation model |
| WO2021037082A1 (en) * | 2019-08-28 | 2021-03-04 | 上海寒武纪信息科技有限公司 | Method and apparatus for processing data, and related product |
| WO2021097771A1 (en) * | 2019-11-21 | 2021-05-27 | Suzhou Aqueti Technology Co., Ltd. | Ics-frame transformation method and apparatus for cv analysis |
| CN113170160A (en) * | 2019-11-21 | 2021-07-23 | 无锡安科迪智能技术有限公司 | ICS frame transformation method and device for computer vision analysis |
| CN112949352A (en) * | 2019-12-10 | 2021-06-11 | 北京地平线机器人技术研发有限公司 | Training method and device of video detection model, storage medium and electronic equipment |
| CN112637605A (en) * | 2020-11-11 | 2021-04-09 | 中国科学院信息工程研究所 | Video steganalysis method and device based on analysis of CAVLC code words and number of nonzero DCT coefficients |
| CN113556439A (en) * | 2021-06-08 | 2021-10-26 | 中国人民解放军战略支援部队信息工程大学 | Rich Model steganography detection feature selection method based on feature component correlation |
Also Published As
| Publication number | Publication date |
|---|---|
| CN107197297B (en) | 2019-12-10 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN107197297A (en) | A kind of video steganalysis method of the detection based on DCT coefficient steganography | |
| Jin et al. | Statistical study on perceived JPEG image quality via MCL-JCI dataset construction and analysis | |
| CN103002289B (en) | Video constant quality coding device for monitoring application and coding method thereof | |
| Wu et al. | Learned block-based hybrid image compression | |
| CN108347612B (en) | Monitoring video compression and reconstruction method based on visual attention mechanism | |
| CN107925763A (en) | The transcoding, coding transform method and apparatus of the selection of block level transforming and implicit signaling in Multi-level segmentation | |
| CN109120937A (en) | A kind of method for video coding, coding/decoding method, device and electronic equipment | |
| CN104378636B (en) | A kind of video encoding method and device | |
| CN110944200B (en) | Method for evaluating immersive video transcoding scheme | |
| CN109151475A (en) | A kind of method for video coding, coding/decoding method, device and electronic equipment | |
| CN107396102A (en) | A kind of inter-frame mode fast selecting method and device based on Merge technological movement vectors | |
| CN108965887A (en) | A kind of video information hiding method and device based on uncoupling between block | |
| CN104853186B (en) | A kind of improved video steganalysis method that is replied based on motion vector | |
| CN104853215B (en) | The video steganography method kept based on motion vector local optimality | |
| CN107155112A (en) | A kind of compressed sensing method for processing video frequency for assuming prediction more | |
| CN114363623A (en) | Image processing method, image processing apparatus, image processing medium, and electronic device | |
| CN109819260A (en) | Video steganography method and device based on the fusion of multi-embedding domain | |
| CN107454413A (en) | A kind of method for video coding of keeping characteristics | |
| dos Santos et al. | The good, the bad, and the ugly: Neural networks straight from jpeg | |
| CN105681803B (en) | A kind of HEVC video information hiding methods of large capacity | |
| CN111263157A (en) | Video multi-domain steganalysis method based on motion vector consistency | |
| CN101426148A (en) | Video objective quality evaluation method | |
| WO2020227911A1 (en) | Method for accelerating coding/decoding of hevc video sequence | |
| CN106331730A (en) | Double-compression detection method by using quantification factor same as H.264 video | |
| KR20200119372A (en) | Artificial Neural Network Based Object Region Detection Method, Device and Computer Program Thereof |
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: 20191210 Termination date: 20200614 |