CN103945228B - Video frame in copy move altering detecting methods based on space-time relationship - Google Patents
Video frame in copy move altering detecting methods based on space-time relationship Download PDFInfo
- Publication number
- CN103945228B CN103945228B CN201410130688.5A CN201410130688A CN103945228B CN 103945228 B CN103945228 B CN 103945228B CN 201410130688 A CN201410130688 A CN 201410130688A CN 103945228 B CN103945228 B CN 103945228B
- Authority
- CN
- China
- Prior art keywords
- video
- frame
- threshold
- sequence
- copy
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Active
Links
Landscapes
- Image Analysis (AREA)
Abstract
The invention provides a kind of video frame in copy move altering detecting methods based on space-time relationship, which carries out NO emissions reduction process according to the method for building up of image gaussian pyramid to frame of video first, the temporal associativity and spatial correlation of monitor video content is recycled to distinguish adaptive setting time threshold value and peak threshold, filter out the suspicious frame pair that there may be that frame in copy move is distorted afterwards using phase place correlation and based on the detection method of Block- matching, and calculate the binary map of description tampered region position.Conclusive judgement is carried out to the packet of suspicious frame using time threshold finally.The present invention has larger minimizing in operation time than analogous algorithms, and enhances algorithm for the robustness of Gaussian noise.When tampered region being described using binary map, add the operation of morphologic filtering eliminate the interference that erroneous judgement point is brought with noise spot to a certain extent, improve the accuracy rate of detection.
Description
Technical field
The invention belongs to video tampering detection field, and in particular to a kind of video frame in copy- based on space-time relationship
Move altering detecting methods.
Background technology
With developing rapidly for information technology, digital video is obtained in various fields and is extensively applied.That day by day popularized is high-quality
Amount digital camera equipment, powerful Video editing software and a large amount of video sharing web sites for rising regard acquisition numeral
Frequently, and which is carried out maliciously distorting and becomes more and more easier.In judicial evidence collection, the verity so that digital video is maliciously distorted
It is difficult to be guaranteed with integrity.How to judge that a piece of digital video is distorted with the presence or absence of malice exactly, have become information
The important research topic in security fields one.
At present, typical video to be distorted be divided into frame according to operation object and distort and distorted with interframe.Video frame in copy-
It is to replicate the sport foreground in certain section of video sequence that move is distorted, and pastes other certain not overlapping of same video
In section video sequence, to reach the purpose for changing video original meaning, belong to one kind that frame in is distorted.Digital video altering detecting method
Two kinds of active detecting and passive detection can be divided into.The video tampering detection of early stage mostly is the mode of active detecting, this kind of method
Depend on the pretreatment to video.Video distorts the statistical nature that passive detection algorithm relies solely on digital video itself, such as
Quantizing noise, textural characteristics, motion vector dependency etc. are judging whether video is maliciously tampered.Due to without the need in advance to video
Processed, passive detection method practicality is higher by contrast.
Find through the retrieval to prior art, Chinese patent literature CN101707711A, publication date is
2010.05.12, a kind of " the video sequence copy-move altering detecting methods based on compression domain " is described, is this process employs
The content similarity of video GOP and kinematic similitude degree are usurped with the presence or absence of copy-move with detecting video sequence as judgement feature
Change.The one of the method has the disadvantage greatly to distort for interframe and is detected, and frame in copy-move is distorted and does not detect energy
Power.This causes a kind of robust and effectively frame in copy-move altering detecting method is urgently suggested.Existing usurp for frame in
The algorithm for changing, is predominantly detected to be covered by same background and is distorted as foreground area, and be directed to what frame in copy-move was distorted
Through the retrieval to existing document, detection algorithm, finds that only having W.Wang proposed, its method is using phase place coherent detection two
Two interframe whether there is suspicious similar area, after primary election recycle the similarity of spatial domain to determine phase two suspicious frames
Like area, the size of final reference area determines whether that there is frame in copy-move distorts, but the method amount of calculation is high,
Fixed threshold is poor for applicability and has certain limitation to cause false alarm rate higher due to decision pattern.
Content of the invention
For defect of the prior art, the present invention proposes a kind of video frame in copy- based on space-time relationship
Move altering detecting methods.The method carries out NO emissions reduction process first with the method for building up of gaussian pyramid to frame of video, then
Using the adaptive given threshold of the temporal associativity and spatial correlation of monitor video content, finally using phase place correlation and base
Judge that with the presence or absence of frame in copy-move distort two-by-two by interframe for video in the detection method of Block- matching, finally using time threshold
Value carries out conclusive judgement again to the packet of suspicious frame.
According to a kind of video frame in copy-move altering detecting method based on space-time relationship that the present invention is provided, tool
Body step includes:
The first step:Video to be detected and reference video is read in, video to be detected and reference video are solved respectively
Each frame of video for frame of video and is converted to gray-scale maps by code, to form sequence of frames of video S to be measured respectivelyd(i, j, t) and
Reference video frame sequence Sr(i, j, t), 1≤i≤M, 1≤j≤N, 1≤t≤L, wherein, i is pixel row coordinate, and j is pixel
Point range coordinate, time indexs of the t for frame of video, M are the image pixel number that often row includes, and N is the picture that image each column includes
Vegetarian refreshments number, L are video totalframes;
Second step:To sequence of frames of video S to be measuredd(i, j, t) and reference video frame sequence Sr(i, j, t) is respectively according to building
The mode of vertical gaussian pyramid carries out a NO emissions reduction computing, obtains the new sequence of frames of video S to be measured after NO emissions reductiond_new(i,
J, t) and reference video frame sequence Sr_new(i, j, t), then by sequence Sd_new(i, j, t) and Sr_new(i, j, t) is utilized respectively
Background subtraction method obtains the video sequence S after removing time redundancyd_re(i, j, t) and Sr_re(i,j,t);And remove the time
Reference video frame sequence after redundancy is made up of the multiple video sequences comprising sport foreground, and wherein, the time is superfluous
Remaining, refer to the background frames not comprising sport foreground;
3rd step:To video sequence Sr_re(i, j, t), by adaptive threshold set algorithm, obtains time threshold
TthresholdWith phase place correlation peak threshold value Pthreshold;
4th step:Using phase place correlation peak threshold value Pthreshold, first to video sequence Sd_re(i, j, t) is calculated
The phase place correlation peak of interframe, makes decisions to peak value two-by-two, and requires the absolute value of the difference of selected its time index of two frames
It is more than time threshold Tthreshold;If peak value exceedes phase place correlation peak threshold value Pthreshold, then relative displacement is extracted, if relative
Displacement is more than given threshold, then carry out the block level coupling of two suspicious interframe using the relative displacement, and otherwise two frame does not exist multiple
Relation processed;Wherein, given threshold is referred to, m/12, and m is frame of video often row pixel sum;
5th step:The binary map for being marked with replication region position between two frames is obtained by block level coupling, to the binary map
First expanded the morphologic filtering for corroding again, calculated the area in largest connected region in binary map after filtering, by the area with
The area threshold of setting is compared, and determines that the two interframe domain of the existence is distorted, otherwise do not exist if more than threshold value;Wherein,
The area threshold of the setting is proportional with original image area;
6th step:Repeat the 4th step, the 5th step until the satisfactory frame of video of all time index relations is through inspection
Survey, that is, detected all frames of video to be measured;
7th step:Using time threshold Tthreshold, it is grouped to there is the frame of video for replicating stickup relation, if point
The video frame number at least having a certain group after group exceedes given threshold, then finally determine that the video sequence has frame in copy-
Move is distorted, and is not otherwise existed.
Preferably, the second step includes procedure below:
Step A1:To every frame S (i, j, t in sequence of frames of videok),tk∈[1,lnew] carried out using Gassian low-pass filter template
Filter in spatial domain smoothed after sequence of frames of video S_lowpass(i,j,tk), wherein, tkFor the time index of video, lnewFor
Video length;
Step A2:Sequence of frames of video S after to smoothing_lowpass(i,j,tk) interlacing is carried out every the down-sampled of row, adopt drop
Video frame size after sample is reduced to a quarter of former frame of video, so just completes a NO emissions reduction computing;
Step A3:The sequence of frames of video after NO emissions reduction is read, and sequence of frames of video and background frames is obtained using background subtraction method
Difference image sequence;
Step A4:To being corroded again and expansive working after difference image sequence binaryzation, by bianry image area of non-zero regions
The area in domain judges that frame of video whether there is sport foreground;
Step A5:For sequence of frames of video S to be measuredd_new(i, j, t) directly chooses the composition of the frame of video containing sport foreground
New sequence, and to reference video frame sequence Sr_new(i, j, t) also needs to divide the frame of video containing sport foreground according to labelling
Section forms subsequence.
Preferably, the 3rd step includes procedure below:
Step B1:According to the reference video frame sequence S that segmentation is obtainedr_new(i, j, t), counts Sr_new(i, j, t) each cross-talk
Length l of video sequence1,l2....lN;
Step B2:, used as model is estimated, the value using following maximum likelihood moments estimation is as the time for selected quantum condition entropy
Threshold value:
Wherein,For quantum condition entropy estimates of parameters, N is reference video total length;
Step C1:Reference video frame sequence S is obtained to segmentationr_new(i, j, t), is partitioned into front scenery using background subtraction method
Whether body, judge foreground object occupied area less than threshold value mn/48, and wherein, m is frame of video often row pixel sum, and n is video
Frame each column pixel sum;The prospect for being only partitioned into negligible amounts is thought if less than threshold value, next execution step C2;
Step C2:Obtain reference video frame sequence Sr_newAfter sport foreground in (i, j, t) per cross-talk sequence, calculate respectively
The average brightness in the sport foreground region of N number of subsequence;
Step C3:The sport foreground regional luminance meansigma methodss of all of subsequence are obtained μ1,μ2...μN, average in brightness
Minimum and maximum mean flow rate μ is found in value sequencemax=max (μ1,μ2...μN),μmin=min (μ1,μ2...μN);
Step C4:After completing step C3, two groups and μ is obtainedmax,μminThe corresponding subsequence of value, selectes two cross-talks respectively
Certain frame in sequence is used as start frame, it is desirable to which the start frame includes the complete prospect for occurring first;
Step C5:Start frame is entered with subsequent frame the calculating of line phase correlation function successively, so, starting is utilized respectively
Frame is calculated a series of corresponding respectively μ with subsequent framemax,μminPhase place correlation function peak amplitudeAskMaximumAndMinima
Threshold value for ensureing to set has higher recall ratio, does following judgement:
IfThen
IfThen
λ1, λ2For the parameter for setting.
Preferably, the 4th step includes procedure below:
Step D1:The video sequence S of readingd_reTwo frame frame of video in (i, j, t) are designated as S respectivelyd_re(i,j,tk) and
Sd_re(i,j,tl), and | tk-tl| > Tthreshold, respectively to Sd_re(i,j,tk) and Sd_re(i,j,tl) FFT is carried out for frequency
Rate figure Sd_re(u,v,tk) and Sd_re(u,v,tl);
Step D2:Calculated by equation below and obtain frequency diagram Sd_re(u,v,tk) and Sd_re(u,v,tl) phase place related
Scheme D (u, v), concrete formula is:
Wherein, * is represented and is taken conjugate operation to each element in matrix;
Step D3:Related for the phase place for obtaining figure D (u, v) is carried out image d (i, j) that IFFT inverse transformations obtain spatial domain, is tied
Close correlation peak threshold value PthresholdFind in spatial domain picture d (i, j) with the presence or absence of more than PthresholdPeak point, if exist,
Coordinate (the Δ that then record peak-peak is locatedx,Δy), with upper left point as origin, if | Δx| > m/12 and | Δy| > m/12,
Then proceed block level coupling, otherwise it is assumed that two frames for reading in do not have similar region;
Step E1:For Sd_re(i,j,tk), by (Δx,Δy) as displacement, to Sd_re(i,j,tk) do circulation shifting
Dynamic process obtains new images
Step E2:WillAnd Sd_re(i,j,tl) it is divided into 8 × 8 overlapping pixels block;Left with block of pixels
The position that the coordinate representation block of pixels of upper angle pixel is located, calculatesAnd Sd_re(i,j,tl) correspondence position pixel
The similarity of block, if the luminance matrix of two block of pixels is respectively M1(i,j),M2(i, j), then calculate similarity using equation below
ρ:
WhereinWithFor the meansigma methodss of two pixel Block Brightness, calculateAnd Sd_re(i,j,tl) all right
Answer the similarity of position block of pixels, if ρ more than set similarity threshold, then it is assumed that two block of pixels be identical, and
The pixel value of corresponding position in binary map B (i, j) is set to 1;Otherwise, the pixel value of binary map corresponding position is set to 0;
Wherein 1≤i≤M/2,1≤j≤N/2.Wherein, similarity threshold takes 0.99.
Preferably, the 7th step includes procedure below:
Step F1:If tampering detection algorithm detects m to there is the frame that copy-move is distorted, first according to frame when
Between subscript sequential search is carried out to there is the frame that copy-move is distorted to m;When initial, searched with the value of time index successively
Rope, the time index of first pair of tampering detection is stored in two arrays respectively;
Step F2:There is the frame that copy-move is distorted in sequential search second pair:
If second pair has two frames that copy-move is distorted, its time index is stored in array with above two respectively
The absolute value of the difference of subscript time is respectively less than or is equal to α1Tthreshold, then by second pair exist frame that copy-move is distorted when
Between subscript be stored in respectively before array in, wherein, α1For adjusting thresholds coefficient;
If second pair of two time index that there is the frame that copy-move is distorted is stored under array with previous respectively
The absolute value of the difference between timestamp is more than α1Tthreshold, then newly set up two arrays to deposit the time index of this pair of frames;
Next exist the time index of the frame pair that copy-move is distorted by with all of array that sets up before in deposit
The time index of frame be compared, if be respectively less than to be equal to the time index absolute difference of element in all arrays before set
Determine threshold value, be then deposited in the closest array of time domain;If absolute with the element time subscript in all arrays before
Difference is all higher than given threshold, then newly set up two arrays to deposit the time index of this frame pair, until detecting to all
M be grouped to there is the frame that copy-move is distorted;The time index number included in different arrays is designated as n respectivelyi, i=1,
2,3,...;
Step F3:To the packet subscript number n for obtainingiThe threshold alpha obtained with step F21TthresholdCompared as follows:
Work as ni≥α1TthresholdWhen, then it is assumed that there is frame in copy-move and distort in the video;
Work as ni< α1TthresholdWhen, then it is assumed that there is no frame in copy-move and distort in the video.
A kind of video frame in copy-move altering detecting method and existing skill based on space-time relationship that the present invention is provided
Art compares, with following obvious substance outstanding feature and remarkable advantage:
1. the method has carried out the pretreatment of a NO emissions reduction using the method for setting up Gauss pyramid to video sequence, this
The process of sample on the premise of guarantee accuracy rate does not affect is greatly reduced operand and improves algorithm for additive Gaussian
The robustness of noise.
2. the method using adaptive threshold value set mechanism, can according to different monitor video content characteristics to arrange when
Between threshold value and phase place correlation peak threshold value, the two threshold values are the key parameters for being related to verification and measurement ratio and accuracy rate, compared to original
The strategy of preset parameter in algorithm.The method has higher motility and more excellent Detection results.
3. the method performed morphologic filtering to the binary map of manipulation of record position before tampered region area is judged
Operation, the operation made in binary map as noise and the erroneous point that causes of erroneous judgement are eliminated the accuracy that improve detection.
Description of the drawings
The detailed description that non-limiting example is made with reference to the following drawings by reading, the further feature of the present invention,
Objects and advantages will become more apparent upon:
Fig. 1 is the overall flow schematic diagram that video frame in distorts passive detection algorithm;
Fig. 2 is phase place related operation and block level coupling idiographic flow schematic diagram;
Fig. 3 is overlapping block dividing method schematic diagram.
Specific embodiment
With reference to specific embodiment, the present invention is described in detail.Following examples will be helpful to the technology of this area
Personnel further understand the present invention, but the invention is not limited in any way.It should be pointed out that the ordinary skill to this area
For personnel, without departing from the inventive concept of the premise, some deformations and improvement can also be made.These belong to the present invention
Protection domain.
A kind of video frame in copy-move altering detecting method based on space-time relationship of the present invention, the method are main
The method that is set up using gaussian pyramid is to video sequence NO emissions reduction, and the means using background subtraction remove the redundancy in video
Content, is hereafter calculated using phase place correlation function and there may be the frame of video that region is distorted, and to suspicious frame of video to carrying out
Block level is mated, and final packet judgement video whether there is to be distorted.Compared with prior art, the method have in operation time compared with
Big minimizing, and algorithm is enhanced for the robustness of Gaussian noise.Compared with the algorithm of traditional fixed threshold, using self adaptation
The time threshold and peak threshold that algorithm is obtained has more motility, for different types of video can show preferably performance.
Moreover, the step of adding morphologic filtering before binary map calculates tampered region area can remove to a certain extent
The interference brought with noise spot by erroneous judgement point, improves the accuracy rate of detection.
It is that sport foreground in just certain section video sequence is replicated that video frame in copy-move is distorted, and pastes
In other certain section of video sequences not overlapped of same video, means are distorted to reach the one kind for the purpose for changing video original meaning,
Belong to one kind that frame in is distorted.
The present invention is achieved by the following technical solutions, first to decompression after reference video and video to be detected carry out
The NO emissions reduction of gaussian pyramid is processed.Reference video after using process obtains adaptive time threshold and phase place phase
Close peak threshold.Again to process after video to be detected in frame calculate phase place correlation function two-by-two, if exist peak value and its
Amplitude then records peak more than peak threshold and carries out block level coupling, then carries out shape to the binary map obtained after block level coupling
State is filtered, the binary map maximum non-zero area after statistical filtering, is recognized if maximum non-zero area is more than the area threshold for setting
There is distorting for frame in copy-move for two interframe, and keep a record, otherwise there is no the relation of distorting.There is frame in all
The frame that copy-move is distorted is grouped, per group of frame number for including after statistical packet, time threshold after frame number exceedes adjustment
When finally judge that the copy-move that the video has frame in is distorted.
Video frame in distorts the overall flow of passive detection algorithm as shown in Figure 1.Below with resolution as 360 × 288,
Compressed format is wmv, and frame per second is 25fps, and totalframes is the reference of the video sequence to be measured and totalframes of 520 frames more than 520
Video is tested.
The first step:Read in video to be detected and reference video, using matlab in mmreader decode it for
The frame of video of bmp forms is simultaneously converted into gray-scale maps using rgb2gray, respectively formation sequence Sd(i, j, t) and Sr(i,j,
T) to carry out subsequent treatment, wherein 1≤i≤360,1≤j≤288,1≤t≤520, i are pixel row coordinate, and j is pixel
Point range coordinate, time indexs of the t for frame of video;
Second step:To decoded sequence of frames of video S to be measuredd(i, j, t) and reference video frame sequence Sr(i, j, t) is pressed
A NO emissions reduction computing is carried out according to the mode for setting up gaussian pyramid.New video sequences S after NO emissions reductiond_new(i, j, t) and
Sr_new(i, j, t), wherein 1≤i≤180,1≤j≤144,1≤t≤520 are obtained using background subtraction method and remove time redundancy
Video sequence S after (i.e. totally stationary frame)d_re(i, j, t) and Sr_re(i, j, t), wherein 1≤i≤180,1≤j≤
144,1≤t≤l_new, it is clear that l_new≤520.And Sr_re(i, j, t) is by multiple video sequences S comprising sport foreground1
(x,y,t)S2(x,y,t)...SN(x, y, t) is constituted.
Described video sequence is carried out a NO emissions reduction computing according to the mode for setting up gaussian pyramid and is subtracted using background
Division obtains the video sequence concrete steps of removing time redundancy and refers to:
Step A1:To every frame S (i, j, t in video sequencek),tk∈[1,lnew] using size be 3 × 3, standard deviation is 1
Gassian low-pass filter template h (i, j), carries out frame of video S after filter in spatial domain is smoothed_lowpass(i,j,tk)=S (i, j,
tk) * h (i, j), wherein, (*) it is two-dimensional convolution computing.
Step A2:To smooth after video sequence carry out interlacing every row down-sampled, make down-sampled after frame of video chi
The very little a quarter for being reduced to former frame of video, so just completes a NO emissions reduction computing.
Step A3:Read smooth rear pending frame of video, it is assumed that with t1Moment frame as initial background frames f (i, j,
t1) (it is the frame of video only containing background that can manually select the first frame), by itself and t1Moment frame of video S (i, j, t1) make the difference.Differed from
Partial image D (i, j, t1)=| S (i, j, t1)-f(i,j,t1)|.
Step A4:To being corroded again and expansive working after difference image binaryzation, the disk mould that width is 5 is selected respectively
Plate and width are that 8 disc template is carried out, and the non-zero pixels number of the bianry image after statistical disposition, and number is more than mn/
48 (m is frame of video often row pixel sum, and n is frame of video each column pixel sum) think that video has sport foreground and to which
Making marks, otherwise being judged as background frames, and background frames renewal is carried out using equation below, more new formula is:
F (i, j, t after renewal2)=(1- α) f (i, j, t2)+αf(i,j,t1)
Wherein α be context update speed, f (i, j, t2) it is t2Moment frame.
Step A5:For video S to be measuredd_new(i, j, t) directly chooses the frame of video containing sport foreground and constitutes new sequence
, and to reference video Sr_new(i, j, t) also needs to do following process according to labelling:Judging that present frame is the frame comprising prospect
Afterwards, whether the former frame for checking present frame is the frame comprising prospect, if the frame comprising prospect, then adds present frame previous
In sequence belonging to frame.If it is not, then restarting to set up one section of sequence, and continue the situation of calculated for subsequent frame.To whole section
Video carries out above-mentioned process, can be by reference video Sr_new(i, j, t) inner video signal process comprising sport foreground out, is obtained
Multiple video sequence S comprising sport foregroundr_new1(i,j,t)Sr_new2(i,j,t)...Sr_newN(i,j,t).
3rd step:Reference video frame sequence S after to processr_re(i, j, t), by adaptive threshold set algorithm, obtains
Arrive time threshold TthresholdWith phase place correlation peak threshold value Pthreshold.
Described auto-adaptive time threshold setting algorithm is mainly included the following steps that:
Step B1:Video sequence S of the reference video obtained according to step 2 segmentationr_new1(i,j,t)Sr_new2(i,j,
t)...Sr_newN(i, j, t), counts length l of each cross-talk video sequence1,l2....lN.
Step B2:A kind of probability Distribution Model is selected for different video types, using probability Distribution Model and before
The time samples of acquisition carry out parameter estimation, and try to achieve the average of sample, select quantum condition entropy here as estimating model, profit
Value with maximum likelihood moments estimation the following is used as time threshold:
Wherein,For quantum condition entropy estimates of parameters, N is reference video total length;
The estimated value is just as time threshold Tthreshold.When tampering detection is carried out, when time interval is more than above-mentioned threshold value
Frame just enter the calculating of line phase correlation function.This is because, logically identical prospect will not be placed by interpolater
In the video sequence in the same time period.
Described self adaptation phase place correlation peak threshold setting algorithm is mainly included the following steps that:
Step C1:In step 2, segmentation obtains video sequence S of reference videor_new1(i,j,t)Sr_new2(i,j,
t)...Sr_newN(i,j,t).This algorithm only considers the situation for having minority prospect simple motion.Using the background subtraction in step 2
Whether method is partitioned into foreground object, judge foreground object occupied area less than threshold value mn/48.Then think only to be partitioned into less than threshold value
The prospect of negligible amounts, meeting the condition then carries out below step.
Step C2:After obtaining the sport foreground per cross-talk sequence, the sport foreground region of N number of video sequence is calculated respectively
Average brightness, such as:The sport foreground regional luminance average value mu that k-th subsequence is calculated using equation belowk:
DtFor the region that kth cross-talk sequence t frames have prospect, lkFor the frame number that k-th sub-series of packets contains.
Step C3:The sport foreground regional luminance meansigma methodss of all of subsequence are obtained μ1,μ2...μN.Average in brightness
Minimum and maximum mean flow rate μ is found in value sequencemax=max (μ1,μ2...μN),μmin=min (μ1,μ2...μN).
Step C4:After completing step C3, two groups and μ is obtainedmax,μminThe corresponding video sequence of value, is designated as S respectivelymax
(x, y, t), Smin(x, y, t), the length of video-frequency band are calculated as l respectivelymax, lmin.This two sections of video sequences contain whole section respectively
There is in reference video higher average brightness value and relatively low average brightness value sport foreground.To Smax(x, y, t), Smin(x,y,t)
S in artificial selected video sequencemax(x,y,tk), tk∈[1,lmax] and Smin(x,y,tl), tl∈[1,lmin] as starting
Frame, it is desirable to which the start frame includes the complete prospect for occurring first.
Step C5:By Smax(x,y,tk) and Smin(x,y,tl) enter the calculating of line phase correlation function with subsequent frame successively.
So, sequence Smax(x, y, t), Smin(x, y, t) is utilized respectively start frame, and to be calculated a series of phase place to subsequent frame related
The peak amplitude of functionTwo group phase place correlation functions are sought respectively
Maximum and minima:
Threshold value for ensureing to set has higher recall ratio, does following judgement:
IfThen
IfThenParameter lambda is typically taken1=1.4, λ2=1.1.
4th step:Phase place correlation peak threshold value P obtained using step 3threshold, first to new video sequence to be measured
Sd_re(i, j, k) calculates the phase place correlation peak of interframe two-by-two, and peak value is made decisions, it is desirable to selected its time index of two frames
Difference absolute value be more than time threshold Tthreshold.If peak value exceedes given threshold, relative displacement (Δ can be extractedx,
Δy), if | Δx| > m/12 and | Δy| > m/12, then proceed block level coupling, otherwise there is no replication relation in two frame.
This is because peak value occurs in the phase place correlation function meeting at the origin of identical image, the object that some original places are hovered needs to disappear with this
Except which disturbs.
Described phase place related algorithm concrete steps refer to (accompanying drawing 2):
Step D1:Assume that the two frame frame of video that reads in are designated as S respectivelyd_re(i,j,tk) and Sd_re(i,j,tl), and time difference
|tk-tl| > Tthreshold, FFT is carried out for frequency diagram S to which respectivelyd_re(u,v,tk) and Sd_re(u,v,tl).
Step D2:The phase place correlation figure that can obtain two frames is calculated by equation below, and concrete formula is:
Wherein, * is to take conjugation to each element in matrix.
Step D3:Related for the phase place for obtaining figure D (u, v) is carried out image d (i, j) that IFFT inverse transformations obtain spatial domain, is tied
Close correlation peak threshold value P that step 3 is obtainedthresholdFind in spatial domain picture d (i, j) with the presence or absence of more than PthresholdPeak value
Point, if existing, records the coordinate (Δ that peak-peak is locatedx,Δy) (with upper left point as origin), if | Δx| > m/12 and |
Δy| > m/12, then proceed block level coupling, otherwise it is assumed that two frames for reading in do not have similar region.Described block level coupling
Concrete steps refer to:
Step E1:One of two frames optionally to be matched, it is assumed herein that select Sd_re(i,j,tk).By (Δx,Δy) make
For displacement, using the dip-image function pairs in matlab Digital Image Processing tool kits, which does loopy moving process and obtains
Obtain new images
Step E2:WillAnd Sd_re(i,j,tl) it is divided into 8 × 8 overlapping pixels block (accompanying drawing 3).With picture
The position that the coordinate representation block of pixels of plain block top left corner pixel point is located, calculatesAnd Sd_re(i,j,tl) corresponding position
The similarity of block of pixels is put, if two block of pixels are respectively M1(i,j),M2(i, j), then calculate similarity using equation below:
WhereinWithFor the meansigma methodss of two pixel Block Brightness, calculateAnd Sd_re(i,j,tl) all right
The similarity of position block of pixels is answered, if ρ is more than the similarity threshold (being traditionally arranged to be 0.99) for setting, then it is assumed that two block of pixels
It is identical, and the pixel value by correspondence position in binary map B (i, j) (coordinate position of block of pixels top left corner pixel) place
It is set to 1.Otherwise, the pixel value of binary map corresponding position is set to 0.Wherein 1≤i≤180,1≤j≤144.
5th step:Binary map B (i, j) for being marked with tampered region position between two frames is obtained through block level coupling.To this
Binary map is first expanded the morphologic filtering for corroding again, selects the disc template that width is 5 and the disk that width is 8 respectively
Template is carried out.The maximum area of connected region in binary map is calculated after the completion of filtering, by the area and the area threshold for setting
Sthreshold=mn/48 is compared, and determines that the two interframe domain of the existence is distorted, otherwise do not exist if more than threshold value.
6th step:Repeat the 4th step, the 5th step until the satisfactory frame of video of all time index relations is through inspection
Survey, that is, detected all frames of video to be measured.
7th step:Using time threshold Tthreshold, it is grouped to there is the frame of video for replicating stickup relation, if point
The video frame number at least having a certain group after group exceedes given threshold α1Tthreshold, then finally determine that the video sequence has frame
Interior copy-move is distorted, and is not otherwise existed.
Described utilization time threshold carries out packet to the frame of video that there is duplication stickup relation and comprises the concrete steps that:
Step F1:If tampering detection algorithm detects m to there is the frame that copy-move is distorted.First according to frame when
Between subscript sequential search is carried out to there is the frame that copy-move is distorted to m.When initial, with the value of time index, (this is to examine successively
Consider the time continuity of motion) scan for, the time index t that there is the frame that copy-move is distorted by first couple1And t1+Δ
t1Array x is stored in respectively1[n],x2In [n].
Step F2:Next, there is the frame that copy-move is distorted in sequential search second pair again:
If second pair has two frames that copy-move is distorted, its time index t2And t2+Δt2Respectively with above two
It is stored in the absolute value of the difference of the subscript time of array | t1-t2| and | t1-t2+Δt1-Δt2| respectively less than or equal to α1Tthreshold,
Array x before then be stored in the time index that second pair has the frame that copy-move is distorted respectively1[n],x2In [n].
If second pair of two time index that there is the frame that copy-move is distorted is stored under array with previous respectively
The absolute value of the difference between timestamp | t1-t2| and | t1-t2+Δt1-Δt2| more than α1Tthreshold, then two arrays x are newly set up3[n],
x4[n] is depositing the time index of this pair of frames.
Next exist the time index of the frame pair that copy-move is distorted by with all of array that sets up before in deposit
The time index of frame be compared, and carry out and identical operation be described above.Until to all m for detecting to exist
The frame that copy-move is distorted is grouped.Multipair array can be obtained.Different arrays xi[n], i=1,2,3 ... in include
Time index number is designated as n respectivelyi, i=1,2,3 ....
Step F3:To the packet subscript number n for obtainingiThe threshold value obtained with step F2 is compared as follows:
Work as ni≥α1TthresholdWhen, we claim the video to there is frame in copy-move to distort.
Work as ni< α1TthresholdWhen, we claim the video to there is no frame in copy-move to distort.If algorithm has n time by mistake
Inspection, and n > α2N ought wherein λ ∈ [0,1], we claim tampering detection algorithm to there is flase drop to the video.
Generally α1=0.7, α2=0.2.α2Referred to as flase drop is spaced regulation coefficient.
Above the specific embodiment of the present invention is described.It is to be appreciated that the invention is not limited in above-mentioned
Particular implementation, those skilled in the art can make various modifications or modification within the scope of the claims, this not shadow
Ring the flesh and blood of the present invention.
Claims (4)
1. a kind of video frame in copy-move altering detecting method based on space-time relationship, it is characterised in that concrete steps bag
Include:
The first step:Video to be detected and reference video is read in, video to be detected and reference video are decoded as respectively
Frame of video and each frame of video is converted to gray-scale maps, to form sequence of frames of video S to be measured respectivelyd(i, j, t) and reference
Sequence of frames of video Sr(i, j, t), 1≤i≤M, 1≤j≤N, 1≤t≤L, wherein, i is pixel row coordinate, and j is pixel point range
Coordinate, time indexs of the t for frame of video, M are the image pixel number that often row includes, and N is the pixel that image each column includes
Number, L are video totalframes;
Second step:To sequence of frames of video S to be measuredd(i, j, t) and reference video frame sequence Sr(i, j, t) is high according to setting up respectively
This pyramidal mode carries out a NO emissions reduction computing, obtains the new sequence of frames of video S to be measured after NO emissions reductiond_new(i,j,t)
And reference video frame sequence Sr_new(i, j, t), then by sequence Sd_new(i, j, t) and Sr_new(i, j, t) is utilized respectively background
Relief method obtains the video sequence S after removing time redundancyd_re(i, j, t) and Sr_re(i,j,t);And remove time redundancy
Reference video frame sequence afterwards is made up of the multiple video sequences comprising sport foreground, and wherein, the time redundancy is
Refer to the background frames not comprising sport foreground;
3rd step:To video sequence Sr_re(i, j, t), by adaptive threshold set algorithm, obtains time threshold TthresholdWith
Phase place correlation peak threshold value Pthreshold;
4th step:Using phase place correlation peak threshold value Pthreshold, first to video sequence Sd_re(i, j, t) calculates frame two-by-two
Between phase place correlation peak, when making decisions to peak value, and requiring that the absolute value of the difference of selected its time index of two frames is more than
Between threshold value Tthreshold;If peak value exceedes phase place correlation peak threshold value Pthreshold, then relative displacement is extracted, if relative displacement is big
In given threshold, then the block level coupling of two suspicious interframe is carried out using the relative displacement, otherwise two frame does not have replication relation;
Wherein, given threshold is referred to, m/12, and m is frame of video often row pixel sum;
5th step:The binary map for being marked with replication region position between two frames is obtained by block level coupling, the binary map is carried out
The morphologic filtering for corroding again is first expanded, the area in largest connected region in binary map after filtering is calculated, by the area and setting
Area threshold be compared, if more than determining that the two interframe domain of the existence is distorted if area threshold, otherwise do not exist;
6th step:Repeat the 4th step, the 5th step until the satisfactory frame of video of all time index relations is through detecting, i.e.,
All frames of video to be measured are detected;
7th step:Using time threshold Tthreshold, it is grouped to there is the frame of video for replicating stickup relation, if after packet
The video frame number at least having a certain group exceedes frame number threshold value, then finally determine that the video sequence has frame in copy-move and usurps
Change, otherwise do not exist;
The second step includes procedure below:
Step A1:To every frame S (i, j, t in sequence of frames of videok),tk∈[1,lnew] space is carried out using Gassian low-pass filter template
Domain filters the sequence of frames of video S after being smoothed_lowpass(i,j,tk), wherein, tkFor the time index of video, lnewFor video
Length;
Step A2:Sequence of frames of video S after to smoothing_lowpass(i,j,tk) carry out interlacing every row down-sampled, make down-sampled it
Video frame size afterwards is reduced to a quarter of former frame of video, so just completes a NO emissions reduction computing;
Step A3:The sequence of frames of video after NO emissions reduction is read, the difference for sequence of frames of video and background frames being obtained using background subtraction method
Partial image sequence;
Step A4:To being corroded again and expansive working after difference image sequence binaryzation, by bianry image non-zero region
Area judges that frame of video whether there is sport foreground;
Step A5:For sequence of frames of video S to be measuredd_new(i, j, t) directly chooses the frame of video containing sport foreground and constitutes new sequence
Row, and to reference video frame sequence Sr_new(i, j, t) is also needed the frame of video segmentation shape containing sport foreground according to labelling
Into subsequence.
2. the video frame in copy-move altering detecting method based on space-time relationship according to claim 1, its feature
It is, the 3rd step includes procedure below:
Step B1:According to the reference video frame sequence S that segmentation is obtainedr_new(i, j, t), counts Sr_new(i, j, t) each cross-talk video
Length l of sequence1,l2....lN;
Step B2:, used as model is estimated, the value using following maximum likelihood moments estimation is as time threshold for selected quantum condition entropy:
Wherein,For quantum condition entropy estimates of parameters, N is reference video total length;
Step C1:Reference video frame sequence S is obtained to segmentationr_new(i, j, t), is partitioned into foreground object using background subtraction method,
Whether foreground object occupied area is judged less than threshold value mn/48, wherein, m is frame of video often row pixel sum, and n is frame of video
Each column pixel sum;The prospect for being only partitioned into negligible amounts is thought if less than threshold value, next execution step C2;
Step C2:Obtain reference video frame sequence Sr_newAfter sport foreground in (i, j, t) per cross-talk sequence, calculate respectively N number of
The average brightness in the sport foreground region of subsequence;
Step C3:The sport foreground regional luminance meansigma methodss of all of subsequence are obtained μ1,μ2...μN, in average brightness sequence
Minimum and maximum mean flow rate μ is found in rowmax=max (μ1,μ2...μN),μmin=min (μ1,μ2...μN);
Step C4:After completing step C3, two groups and μ is obtainedmax,μminThe corresponding subsequence of value, selectes two cross-talk sequences respectively
In certain frame as start frame, it is desirable to the start frame includes the complete first prospect for occurring;
Step C5:Start frame is entered successively the calculating of line phase correlation function with subsequent frame, so, be utilized respectively start frame with
Subsequent frame be calculated a series of respectively correspond to μmax,μminPhase place correlation function peak amplitudeAskMaximumAndMinima
Threshold value for ensureing to set has higher recall ratio, does following judgement:
IfThen
IfThen
λ1, λ2For the parameter for setting.
3. the video frame in copy-move altering detecting method based on space-time relationship according to claim 1, its feature
It is, the 4th step includes procedure below:
Step D1:The video sequence S of readingd_reTwo frame frame of video in (i, j, t) are designated as S respectivelyd_re(i,j,tk) and Sd_re(i,
j,tl), and | tk-tl| > Tthreshold, respectively to Sd_re(i,j,tk) and Sd_re(i,j,tl) FFT is carried out for frequency diagram Sd_re
(u,v,tk) and Sd_re(u,v,tl);
Step D2:Calculated by equation below and obtain frequency diagram Sd_re(u,v,tk) and Sd_re(u,v,tl) the related figure D of phase place (u,
V), concrete formula is:
Wherein, * is represented and is taken conjugate operation to each element in matrix;
Step D3:Related for the phase place for obtaining figure D (u, v) is carried out image d (i, j) that IFFT inverse transformations obtain spatial domain, in conjunction with phase
Close peak threshold PthresholdFind in spatial domain picture d (i, j) with the presence or absence of more than PthresholdPeak point, if exist, remember
Coordinate (the Δ that record peak-peak is locatedx,Δy), with upper left point as origin, if | Δx| > m/12 and | Δy| > m/12, then after
Continuing carries out block level coupling, otherwise it is assumed that two frames for reading in do not have similar region;
Step E1:For Sd_re(i,j,tk), by (Δx,Δy) as displacement, to Sd_re(i,j,tk) do at loopy moving
Reason obtains new images
Step E2:WillAnd Sd_re(i,j,tl) it is divided into 8 × 8 overlapping pixels block;With block of pixels upper left corner picture
The position that the coordinate representation block of pixels of vegetarian refreshments is located, calculatesAnd Sd_re(i,j,tl) correspondence position block of pixels phase
Like spending, if the luminance matrix of two block of pixels is respectively M1(i,j),M2(i, j), then calculate similarity ρ using equation below:
WhereinWithFor the meansigma methodss of two pixel Block Brightness, calculateAnd Sd_re(i,j,tl) all correspondence positions
The similarity of block of pixels, if ρ is more than the similarity threshold for setting, then it is assumed that two block of pixels are identical, and by two-value
In figure B (i, j), the pixel value of corresponding position is set to 1;Otherwise, the pixel value of binary map corresponding position is set to 0;Wherein 1
≤i≤M/2,1≤j≤N/2.
4. the video frame in copy-move altering detecting method based on space-time relationship according to claim 1, its feature
It is, the 7th step includes procedure below:
Step F1:If tampering detection algorithm detects m to there is the frame that copy-move is distorted, first according to the time of frame under
Mark carries out sequential search to m to there is the frame that copy-move is distorted;When initial, scanned for the value of time index successively, will
First pair of time index that there is the frame that copy-move is distorted is stored in two arrays respectively;
Step F2:There is the frame that copy-move is distorted in sequential search second pair:
If second pair has two frames that copy-move is distorted, its time index respectively with above two subscripts for being stored in array
The absolute value of the difference of time is respectively less than or is equal to α1Tthreshold, then under the time that there is the frame that copy-move is distorted by second pair
Mark in the array before being stored in respectively, wherein, α1For adjusting thresholds coefficient;
If second pair of two time index that there is the frame that copy-move is distorted respectively with the previous lower timestamp for being stored in array
Between difference absolute value be more than α1Tthreshold, then newly set up two arrays to deposit the time index of this pair of frames;
Next exist the time index of the frame pair that copy-move is distorted by with the frame deposited in all of array that sets up before
Time index be compared, if be respectively less than to be equal to the time index absolute difference of element in all arrays before setting threshold
Value, then be deposited in the closest array of time domain;If with the element time subscript absolute difference in all arrays before
Given threshold is all higher than, then newly sets up two arrays to deposit the time index of this frame pair, until to all m pairs for detecting
There is the frame that copy-move distorts to be grouped;The time index number included in different arrays is designated as n respectivelyi, i=1,2,
3,...;
Step F3:To the packet subscript number n for obtainingiThe threshold alpha obtained with step F21TthresholdCompared as follows:
Work as ni≥α1TthresholdWhen, then it is assumed that there is frame in copy-move and distort in the video;
Work as ni< α1TthresholdWhen, then it is assumed that there is no frame in copy-move and distort in the video.
Priority Applications (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201410130688.5A CN103945228B (en) | 2014-03-28 | 2014-03-28 | Video frame in copy move altering detecting methods based on space-time relationship |
Applications Claiming Priority (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201410130688.5A CN103945228B (en) | 2014-03-28 | 2014-03-28 | Video frame in copy move altering detecting methods based on space-time relationship |
Publications (2)
Publication Number | Publication Date |
---|---|
CN103945228A CN103945228A (en) | 2014-07-23 |
CN103945228B true CN103945228B (en) | 2017-03-15 |
Family
ID=51192672
Family Applications (1)
Application Number | Title | Priority Date | Filing Date |
---|---|---|---|
CN201410130688.5A Active CN103945228B (en) | 2014-03-28 | 2014-03-28 | Video frame in copy move altering detecting methods based on space-time relationship |
Country Status (1)
Country | Link |
---|---|
CN (1) | CN103945228B (en) |
Families Citing this family (10)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN104270644A (en) * | 2014-09-28 | 2015-01-07 | 上海交通大学 | Video inter-frame tampering detecting method based on velocity field coincidence |
CN104954807B (en) * | 2015-06-25 | 2018-02-23 | 西安理工大学 | The video dubbing altering detecting method of resist geometric attackses |
CN105578198B (en) * | 2015-12-14 | 2019-01-11 | 上海交通大学 | Based on when inclined feature the homologous Copy-Move detection method of video |
CN105657435B (en) * | 2016-01-25 | 2019-04-05 | 福建师范大学 | Altering detecting method is pasted in video single frames duplication based on quantization DCT coefficient |
CN106375773B (en) * | 2016-09-08 | 2019-11-26 | 福建师范大学 | Altering detecting method is pasted in frame duplication based on dynamic threshold |
CN106447656B (en) * | 2016-09-22 | 2019-02-15 | 江苏赞奇科技股份有限公司 | Rendering flaw image detecting method based on image recognition |
CN109218752B (en) * | 2017-06-30 | 2022-04-19 | 中国电信股份有限公司 | Video playing, providing and detecting method, set top box, media server and system |
CN110162665B (en) * | 2018-12-28 | 2023-06-16 | 腾讯科技(深圳)有限公司 | Video searching method, computer device and storage medium |
CN110084781B (en) * | 2019-03-22 | 2021-11-09 | 西安电子科技大学 | Passive evidence obtaining method and system for monitoring video tampering detection based on feature points |
JP7314683B2 (en) * | 2019-07-23 | 2023-07-26 | 株式会社Jvcケンウッド | MOVING IMAGE ENCODER, MOVING IMAGE ENCODED METHOD, MOVING IMAGE ALTERATION DETERMINATION METHOD, AND MOVING IMAGE ALTERATION DETERMINATION PROGRAM |
Citations (3)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
EP2164261A2 (en) * | 2002-07-11 | 2010-03-17 | Panasonic Corporation | Filtering strength determination method, moving picture coding and decoding method |
CN101707711A (en) * | 2009-11-03 | 2010-05-12 | 上海大学 | Method for detecting tampering of video sequence by Copy-Move based on compressed domain |
CN103384331A (en) * | 2013-07-19 | 2013-11-06 | 上海交通大学 | Video inter-frame forgery detection method based on light stream consistency |
-
2014
- 2014-03-28 CN CN201410130688.5A patent/CN103945228B/en active Active
Patent Citations (3)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
EP2164261A2 (en) * | 2002-07-11 | 2010-03-17 | Panasonic Corporation | Filtering strength determination method, moving picture coding and decoding method |
CN101707711A (en) * | 2009-11-03 | 2010-05-12 | 上海大学 | Method for detecting tampering of video sequence by Copy-Move based on compressed domain |
CN103384331A (en) * | 2013-07-19 | 2013-11-06 | 上海交通大学 | Video inter-frame forgery detection method based on light stream consistency |
Non-Patent Citations (2)
Title |
---|
图像内容纂改真实性检测算法分析与研究;金梦;《中国优秀硕士学位论文全文数据库 信息科技辑》;20100815(第8期);正文第29-36页 * |
图像区域复制粘贴检测技术研究;冯占磊;《中国优秀硕士学位论文全文数据库 信息科技辑》;20090715(第7期);正文第15-16页 * |
Also Published As
Publication number | Publication date |
---|---|
CN103945228A (en) | 2014-07-23 |
Similar Documents
Publication | Publication Date | Title |
---|---|---|
CN103945228B (en) | Video frame in copy move altering detecting methods based on space-time relationship | |
CN106845408B (en) | Street garbage identification method under complex environment | |
Cozzolino et al. | Camera-based image forgery localization using convolutional neural networks | |
CN103345758B (en) | Jpeg image region duplication based on DCT statistical nature distorts blind checking method | |
Chang et al. | A forgery detection algorithm for exemplar-based inpainting images using multi-region relation | |
Kumar et al. | A fast DCT based method for copy move forgery detection | |
CN103561271B (en) | The moving target of static camera shooting is removed video spatial domain altering detecting method | |
US20200250840A1 (en) | Shadow detection method and system for surveillance video image, and shadow removing method | |
CN103544703B (en) | Digital picture splicing detection method | |
CN101493937B (en) | Method for detecting content reliability of digital picture by utilizing gradient local entropy | |
CN103093458B (en) | The detection method of key frame and device | |
CN104933721A (en) | Spliced image-tamper detection method based on color filter array characteristic | |
CN104837011B (en) | Content self-adaptive video steganalysis method | |
CN111652875B (en) | Video forgery detection method, system, storage medium and video monitoring terminal | |
CN106815583A (en) | A kind of vehicle at night license plate locating method being combined based on MSER and SWT | |
CN105488787A (en) | Geometric active contour model-based coastline detection method for remote sensing image | |
CN101299274B (en) | Detecting method and system for moving fixed target | |
Muhammad et al. | A non-intrusive method for copy-move forgery detection | |
CN105657435A (en) | Single video frame copy and paste tamper detection method based on quantized DCT coefficient | |
CN102521821A (en) | Automatic identification and tampered area positioning method in digital image | |
CN103561274B (en) | Video time domain tamper detection method for removing moving object shot by static camera lens | |
CN109660814B (en) | Method for detecting deletion tampering of video foreground | |
Chang et al. | A passive multi-purpose scheme based on periodicity analysis of CFA artifacts for image forensics | |
CN102024244A (en) | Method and device for embedding and detecting watermarks based on image characteristic region | |
CN103136517A (en) | Rolling fingerprint image sequence real-time mosaic method based on key column selection |
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 |