CN109726733A - A kind of video tamper detection method based on frame-to-frame correlation - Google Patents
A kind of video tamper detection method based on frame-to-frame correlation Download PDFInfo
- Publication number
- CN109726733A CN109726733A CN201811373982.3A CN201811373982A CN109726733A CN 109726733 A CN109726733 A CN 109726733A CN 201811373982 A CN201811373982 A CN 201811373982A CN 109726733 A CN109726733 A CN 109726733A
- Authority
- CN
- China
- Prior art keywords
- frame
- video
- gist feature
- correlation
- gist
- 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.)
- Pending
Links
- 238000001514 detection method Methods 0.000 title claims abstract description 87
- 201000011243 gastrointestinal stromal tumor Diseases 0.000 claims abstract description 99
- 238000012217 deletion Methods 0.000 claims abstract description 35
- 230000037430 deletion Effects 0.000 claims abstract description 35
- 238000003780 insertion Methods 0.000 claims abstract description 35
- 230000037431 insertion Effects 0.000 claims abstract description 35
- 238000005259 measurement Methods 0.000 claims abstract description 15
- 238000004364 calculation method Methods 0.000 claims abstract description 12
- 239000000284 extract Substances 0.000 claims abstract description 7
- 230000001174 ascending effect Effects 0.000 claims description 3
- 238000001914 filtration Methods 0.000 claims description 2
- 238000012545 processing Methods 0.000 claims description 2
- 230000009466 transformation Effects 0.000 claims description 2
- 238000000034 method Methods 0.000 abstract description 22
- 230000000694 effects Effects 0.000 abstract description 5
- 238000012360 testing method Methods 0.000 description 20
- 238000002474 experimental method Methods 0.000 description 8
- 238000012795 verification Methods 0.000 description 8
- 238000010586 diagram Methods 0.000 description 6
- 230000002159 abnormal effect Effects 0.000 description 3
- 230000005856 abnormality Effects 0.000 description 3
- 238000005516 engineering process Methods 0.000 description 3
- 230000000007 visual effect Effects 0.000 description 3
- 230000009286 beneficial effect Effects 0.000 description 2
- 230000008859 change Effects 0.000 description 2
- 238000012937 correction Methods 0.000 description 2
- 230000000875 corresponding effect Effects 0.000 description 2
- 230000002708 enhancing effect Effects 0.000 description 2
- 238000007689 inspection Methods 0.000 description 2
- 238000013432 robust analysis Methods 0.000 description 2
- 230000002776 aggregation Effects 0.000 description 1
- 238000004220 aggregation Methods 0.000 description 1
- 230000015572 biosynthetic process Effects 0.000 description 1
- 238000006243 chemical reaction Methods 0.000 description 1
- 230000002596 correlated effect Effects 0.000 description 1
- 230000007423 decrease Effects 0.000 description 1
- 238000011161 development Methods 0.000 description 1
- 238000011156 evaluation Methods 0.000 description 1
- 238000009472 formulation Methods 0.000 description 1
- 239000000203 mixture Substances 0.000 description 1
- 230000008569 process Effects 0.000 description 1
- 230000003362 replicative effect Effects 0.000 description 1
- 238000011160 research Methods 0.000 description 1
- 238000011524 similarity measure Methods 0.000 description 1
- 238000003786 synthesis reaction Methods 0.000 description 1
- 238000005303 weighing Methods 0.000 description 1
Landscapes
- Image Analysis (AREA)
Abstract
The invention discloses a kind of video tamper detection methods based on frame-to-frame correlation, comprising the following steps: video is decoded into independent sequence of frames of video first, extracts the GIST feature of each frame；Then the Euclidean distance of GIST feature is calculated；Obtained video frame is handled using filter group, the Euclidean distance of the GIST feature for each frame that then comparing calculation obtains is obtained with the presence or absence of duplicated frame；Then the Spearman related coefficient for calculating adjacent interframe GIST feature carrys out and measures the correlation of consecutive frame, obtains with the presence or absence of insertion frame；Then it using the intensity of anomaly of the related coefficient of LO algorithm measurement video frame GIST feature, is then obtained according to intensity of anomaly with the presence or absence of frame deletion；Method provided by the present invention can use a kind of three kinds of different interframe tampering detection tasks of feature realization, the video frame number for all having preferable performance in terms of detection effect and robustness, and capable of accurately determining the video frame number being replicated and being inserted into.
Description
Technical field
The invention belongs to video tampering detection technical fields, and in particular to a kind of video based on frame-to-frame correlation distorts inspection
Survey method.
Background technique
In recent years, with the digital equipments such as the continuous development of internet and digital technology and digital camera, video camera
It is widely used, people are more easily accessible to all kinds of images, video than ever, these digitized products make people's
It lives more rich and varied.But with various powerful images, Video editing software (such as Adobe Photoshop and
Adobe Premiere) use, digital picture, video be easy to be tampered and attack, this makes the true of digital contents
Reality is destroyed with integrality, thus greatly reduces the confidence level of source video sequence and content.
Such synthesis and practical joke to video, allows people to gradually lose the trust to video content.If distorted
Content is related to national security, business information, individual privacy etc., can bring immeasurable damage to country, enterprises and individuals surely
It loses, be unfavorable for the harmony of society and stablize.Therefore, the integrality of the authenticity and content that detect digital source video sequence has become
One urgent problem to be solved.The research to video tamper detection method is unfolded, has very important significance.
Currently, the tampering detection technology for video content is extremely limited, existing method mainly utilizes content not
Consistency is detected, it may be assumed that directly extracts the various features of media object content, comprising: global characteristics, local feature, fortune
Dynamic feature, space-time characteristic etc. are used as evidence, are compared by characteristic matching or Similarity measures with the threshold value of setting, detection view
Whether frequency content is by distorting.These detection methods mainly include in video dubbing detection, frame insertion, frame deletion detection and frame
Tampering detection.
Summary of the invention
The object of the present invention is to provide a kind of video tamper detection methods based on frame-to-frame correlation, solve existing
Method detection accuracy present in technology is not high, the weak problem of robustness.
The technical scheme adopted by the invention is that a kind of video tamper detection method based on frame-to-frame correlation, including with
Lower step:
Step 1, video is decoded into independent sequence of frames of video, extracts the GIST feature of each frame；
Step 2, the Euclidean distance of the GIST feature obtained through step 1 is calculated；
Step 3, the video frame obtained through step 1 is handled using filter group, then compares and is calculated through step 2
The Euclidean distance of the GIST feature of each frame arrived, obtains with the presence or absence of duplicated frame；
Step 4, video frame is inserted into tampering detection: calculating the Spearman of the adjacent interframe GIST feature obtained through step 1
Related coefficient carrys out and measures the correlation of consecutive frame, obtains with the presence or absence of insertion frame；
Step 5, the GIST feature of adjacent two frame video frame deletion tampering detection: is calculated using the calculation method of step 4
Between Spearman related coefficient, using LO algorithm measurement video frame GIST feature related coefficient intensity of anomaly, then
It is obtained according to intensity of anomaly with the presence or absence of frame deletion.
The features of the present invention also characterized in that:
Wherein specific steps in the step 1 are as follows: video to be detected is decoded into and is independent sequence of frames of video, it is right
Frame sequence carries out one-dimensional Haar wavelet transformation, low frequency component is extracted, as new frame sequence；It include the video of M frame, note to one
New frame sequence is { 1,2 ..., M }, extracts the GIST feature of each frame, the GIST feature for obtaining M frame is denoted as: F={ F1,
F2,...,FM, wherein Fi={ fi,1,fi,2,...,fi,qBe the i-th frame GIST feature, q be GIST feature dimension；
Wherein the specific steps are be by the independent sequence of frames of video obtained through step 1 selection first frame in the step 2
Reference frame calculates the Euclidean distance c between the GIST feature of each frame and the GIST feature of reference framei(i=1 ..., M):
Wherein, Fi={ fi,1,fi,2,...,fi,qBe the i-th frame GIST feature, q be GIST feature dimension；
Wherein the specific steps are filtering processings in the step 3: using the Gabor filter group in 4 scales, 8 directions
The video frame obtained through step 1 is filtered；Compare the GIST feature of each frame being calculated through step 2 it is European away from
From if ci=cj, then have duplication situation between the i-th frame and jth frame；
Wherein the step 4 specifically includes: calculating the Spearman phase relation between the GIST feature of adjacent two frame first
Number zi, the correlation distance for then defining the GIST feature of the i-th frame and the GIST feature of i+1 frame is CiIf note insertion boundary
The maximum value and second largest value of two correlation distance are respectively Cmax、Csec, and the ratio between the second largest value of correlation distance and average value are greater than
Some threshold value Sin, whenWhen, it is considered as in this frame sequence that there are frame insertions.
Wherein the correlation distance of the GIST feature of the GIST feature and i+1 frame of i-th frame is C in the step 4iAre as follows:
The Spearman related coefficient between the GIST feature of adjacent two frame is wherein calculated in the step 4 are as follows:
For the video of a M frame, F is enabledi={ fi,1,fi,2,...,fi,qBe the i-th frame GIST feature, q is that GIST is special
The dimension of sign takes q=512 using GIST feature calculation method.To Fi、Fi+1It is ranked up, while being ascending or descending order, respectively
Obtain the ordered set F' of two groups of elementsi={ f'i,1,f'i,2,...,f'i,q}、 F'i+1={ f'i+1,1,f'i+1,2,...,
f'i+1,q}.It will set F'i、F'i+1In corresponding element subtract each other, to obtain a difference set:
kj=f'i,j-f'i+1,j, { k1,k2,...,kq} (3)
Spearman coefficient between the GIST feature of i-th frame and the GIST feature of i+1 frame are as follows:
Wherein in the step 5 intensity of anomaly calculation method are as follows: meter the i-th frame GIST feature and i+1 frame GIST
Spearman coefficient between feature is denoted as zi, for the video of a M frame, obtain M-1 related coefficient z={ z1,
z2,...,zM-1, it is related including calculating using the intensity of anomaly of the related coefficient of LO algorithm measurement video frame GIST feature
Euclidean distance di between coefficientl,iWith local density lrl,i, then introducing indicates in ziL small distance in all videos
Frame Ml,i, finally according to formula
Therefore, LOl,iSize mean that the Spearman between the GIST feature of the i-th frame and the GIST feature of i+1 frame
Coefficient ziIntensity of anomaly；Introduce threshold value ScdIf LOl,I>Scd, then have frame deletion situation between the i-th frame and i+1 frame.
Wherein in the step 5 Euclidean distance calculation method are as follows: calculate each ziWith the Euclidean distance of components other in z
dil,i, l is given positive integer, dil,iFor the l small distance of i-th of related coefficient, as shown in formula (4):
dil,i=r (| | zj-z||j≠i},l) (6)
Wherein, function r indicates first of minimum value that data sort from large to small in set.
Wherein in the step 5 local density calculation method are as follows:
Ml,i={ zj|||zj-zi| | < dil,i,j≠i} (8)
Wherein, Ml,iIt indicates in ziL small distance in all video frames, | | indicate set in element number.
The beneficial effects of the invention are as follows
The present invention provides a kind of video tamper detection methods based on frame-to-frame correlation, and it is multiple effectively to detect video frame
System, frame insertion and frame deletion are distorted, and three kinds of different interframe tampering detection tasks are realized using a kind of feature, for different
Database has very high verification and measurement ratio and is not influenced by video capture camera, and the video of different-format also can be examined accurately
It surveys, all has for various the map functions such as enhancing of gamma correction, contrast, fuzzy, horizontal mirror image, vertical mirror, rotation etc.
Relatively good robustness, has certain resist geometric attacks performance, and precision ratio with higher also has when deletion frame number is less
There is preferable detection effect.
Detailed description of the invention
Fig. 1 is that original video frame sequence is illustrated in a kind of video tamper detection method based on frame-to-frame correlation of the invention
Figure；
Fig. 2 is that video sequence is distorted in single frames duplication in a kind of video tamper detection method based on frame-to-frame correlation of the invention
List intention；
Fig. 3 is actual scene original video sequence in a kind of video tamper detection method based on frame-to-frame correlation of the invention
Column example；
Fig. 4 is that the duplication of original scene single frames is usurped in a kind of video tamper detection method based on frame-to-frame correlation of the invention
Change video sequence example；
Fig. 5 is continuous multiple frames duplication signal in a kind of video tamper detection method based on frame-to-frame correlation of the invention
Figure；
Fig. 6 is actual scene video sequence figure in a kind of video tamper detection method based on frame-to-frame correlation of the invention
3 continuous multiple frames replicate the example figure distorted；
Fig. 7 be in a kind of video tamper detection method based on frame-to-frame correlation of the invention original video frame sequence A and
B；
Fig. 8 is that schematic diagram is distorted in frame insertion in a kind of video tamper detection method based on frame-to-frame correlation of the invention；
Fig. 9 is that frame is inserted into instance of video in a kind of video tamper detection method based on frame-to-frame correlation of the invention；
Figure 10 is video frame deletion schematic diagram in a kind of video tamper detection method based on frame-to-frame correlation of the invention；
Figure 11 is that video frame deletion distorts reality in a kind of video tamper detection method based on frame-to-frame correlation of the invention
Example；
Figure 12 is the phase of the video frame in a kind of video tamper detection method based on frame-to-frame correlation of the invention in Fig. 3
To Euclidean distance；
Figure 13 is the phase of the video frame in a kind of video tamper detection method based on frame-to-frame correlation of the invention in Fig. 4
To Euclidean distance；
Figure 14 is the phase of the video frame in a kind of video tamper detection method based on frame-to-frame correlation of the invention in Fig. 6
To Euclidean distance；
Figure 15 is shown in Fig. 9 in a kind of video tamper detection method based on frame-to-frame correlation of the invention to distort video
Frame correlation distance；
The abnormality degree of video sequence frame in Figure 16 Figure 11.
Specific embodiment
The following describes the present invention in detail with reference to the accompanying drawings and specific embodiments.
The present invention provides a kind of video tamper detection methods based on frame-to-frame correlation, comprising the following steps:
Step 1, video is decoded into independent sequence of frames of video
Video to be detected is decoded into and is independent sequence of frames of video, one-dimensional Haar small echo is carried out to frame sequence and is become
It changes, low frequency component is extracted, as new frame sequence；Include the video of M frame to one, remember that new frame sequence is { 1,2 ..., M },
The GIST feature for extracting each frame, the GIST feature for obtaining M frame are denoted as: F={ F1,F2,...,FM, wherein Fi={ fi,1,
fi,2,...,fi,qBe the i-th frame GIST feature, q be GIST feature dimension；
Step 2, the Euclidean distance of GIST feature is calculated
The first frame of selecting video sequence is reference frame, calculates the GIST feature of each frame and the GIST feature of reference frame
Between Euclidean distance ci(i=1 ..., M) (we are referred to as the opposite Euclidean distance of the i-th frame):
Wherein, Fi={ fi,1,fi,2,...,fi,qBe the i-th frame GIST feature, q be GIST feature dimension.
Step 3, it is filtered
In order to improve efficiency of algorithm, we filter video frame using the Gabor filter group in 4 scales, 8 directions
Wave, therefore q=4 × 8=32.
In general, each frame in sequence of frames of video is all different, so that the GIST feature of each frame is also different, because
This, if there is l identical Euclidean distance values, then explanation have l frame be it is identical, that is, having l-1 frame is by certain in this video
What one frame replicated, it may be assumed that if ci=cj, then have duplication situation between the i-th frame and jth frame.
Step 4, video frame replicates tampering detection
Single frames duplication is distorted, and is referred to and is replicated to obtain new video sequence by a certain frame continuous several times, as Fig. 1 indicates original
Video sequence schematic diagram, Fig. 2 indicates that video sequence schematic diagram is distorted in single frames duplication, wherein the 4th frame has been replicated 3 times.Fig. 3 table
Show that the original video sequence example an of actual scene, Fig. 4 indicate that video sequence example is distorted in the single frames duplication of actual scene,
In, the 1st, 2,7,8 frames be in original video Fig. 3 the 1st, 2,3,4 frames, the 3rd to 6 frame is by 4 replicating original videos
What the 2nd frame obtained, the video distorted in this way does not see difference in visual effect；
Continuous multiple frames duplication is distorted, and is referred to and is replicated the continuous multiple frames in original video sequence to obtain new frame sequence
Column, as shown in figure 5, be in Fig. 1 the 4th, 5,6 frames replicate once.Fig. 6 is that actual scene video sequence Fig. 3 is continuously more
Frame replicates the example distorted, the 1st in Fig. 6,2,3,4,5,11,12 frames be 1-6 frame in Fig. 3,6-10 frame is logical in Fig. 6
Cross what 1-5 frame in 1 Fig. 3 of duplication obtained, observer does not see the notable difference in visual effect yet；
Step 4.1, single frames duplication is distorted, if ci+1It is secondary that l (l >=2) occurs in value, i.e. ci+1=ci+2=...=
ci+l, then this video sequence is replicated l-1 times there are single frames and is distorted.Figure 12 is the opposite Euclidean distance of the video frame in Fig. 3, figure
13 be the opposite Euclidean distance of the video frame in Fig. 4.It can be seen that visually very indistinguishable frame, phase from Figure 12 and Figure 13
There is apparent difference to Euclidean distance.As can be seen from Figure 13, the opposite Euclidean distance of 2-6 frame is identical, indicates the 2nd frame extremely
6th frame is that single frames duplication is distorted, this is consistent with actually distorting for Fig. 4.
Step 4.2, successive frame duplication is distorted, if ci+1、ci+2、…、ci+m(ci+1≠ci+2≠...≠ci+m) value
It is secondary to there are l (l >=2), it may be assumed that
Then this sequence of frames of video is replicated there are continuous m frame and is distorted for l times, is divided into n-m frame.Figure 14 is the video frame of Fig. 6
Opposite Euclidean distance value.Can intuitively it find out from Figure 14, the 1st frame and the 6th frame, the 2nd frame and the 7th frame, the 3rd frame and the 8th frame,
The opposite Euclidean distance of 4 frames and the 9th frame, the 5th frame and the 10th frame be it is the same, indicate exist between 1-5 frame and 6-10 frame
Successive frame duplication is distorted, and number of copy times is 1 time；
Step 5, video frame is inserted into tampering detection
Refer to and original video is decoded into sequence of frames of video, is inserted into the video sequence continuous in another sequence of frames of video
Multiframe generates new video sequence, is illustrated in figure 7 original video frame sequence A and B, and Fig. 8 is that schematic diagram is distorted in frame insertion, is
In the video B in Fig. 7 the 6th, 7,8 frames are inserted between the 6th frame of video A and the 7th frame.Fig. 9 is that frame is inserted into video reality
Example, inserts the frame a, b, c, d, e in other video sequences in video frame 1,2,3.
Step 5.1, it is assumed that handled video is that frame shown in Fig. 9 is inserted into instance of video, and original video only has 3 frames,
The frame a in other video sequences, b, c, d, e are inserted between 2nd frame and the 3rd frame.
Step 5.2, the Spearman related coefficient of GIST feature is calculated
Since the insertion of heterologous video frame can destroy the consistency of original video sequence inter texture feature, and Spearman
Related coefficient can describe the degree " being monotonically correlated " between data, i.e., with the increase of one of numerical value, can cause another
The degree that a numerical value increases or decreases.Therefore, we by calculate the Spearman related coefficient of adjacent interframe GIST feature come
And the correlation of measurement consecutive frame, thus the video frame that detection is inserted into；
Calculate the Spearman related coefficient between the GIST feature of adjacent two frame:
For the video of a M frame, F is enabledi={ fi,1,fi,2,...q,,fiBe the i-th frame GIST feature, q is that GIST is special
The dimension of sign takes q=512 using GIST feature calculation method same as the prior art.To Fi、 Fi+1It is ranked up (simultaneously
For ascending or descending order), respectively obtain the ordered set F' of two groups of elementsi={ f'i,1,f'i,2,...,f'i,q}、F'i+1=
{f'i+1,1,f'i+1,2,...,f'i+1,q}.It will set F'i、F'i+1In corresponding element subtract each other, to obtain a difference set:
kj=f'i,j-f'i+1,j, { k1,k2,...,kq} (3)
Spearman coefficient between the GIST feature of i-th frame and the GIST feature of i+1 frame are as follows:
Define the correlation distance C of the GIST feature of the i-th frame and the GIST feature of i+1 frameiAre as follows:
When there are video frame insertion, the consistency of original sequence of frames of video is destroyed step 5.3, is inserted into frame and primitive frame
The correlation of boundary will reduce.Therefore, when being inserted into continuous multiple frames in a video, insertion frame and adjacent primitive frame
The related coefficient z of GIST feature can become smaller between (insertion boundary), and correlation distance C can become larger.If two of note insertion boundary
The maximum value and second largest value of correlation distance are respectively Cmax、Csec, and the ratio between the second largest value of correlation distance and average value are greater than threshold value
(in an experiment, we measure S by many experiments to Sinin=4.7), it may be assumed thatWe are considered as in this frame sequence
There are frame insertions.It might as well assume Cmax=Ci, Csec=Cj, then illustrate i+1 frame to jth frame for insertion frame, insertion frame number be j-i
Frame.
For example, Figure 15 indicates the correlation distance of the frame sequence GIST feature shown in Fig. 9 for distorting video, it can from Figure 15
Out, the 2nd is several much larger than other with the 7th correlation distance value, this is because successive frame insertion reduces its correlation, so that its phase
It closes distance to increase, i.e., there is successive frame insertion to distort between the 2nd frame and the 7th frame, insertion frame number is 5 frames.
Step 6, video frame deletion tampering detection
Video frame deletion, which is distorted, to be referred to original video is decoded into sequence of frames of video, and the continuous multiple frames in frame sequence are deleted
Obtain new sequence of frames of video, in order to delete the activity trajectory of certain object or persons in video.Although theoretically may be used
Only to delete 1 frame, but since the frame rate of usual video is that 25~30 frames are per second, only deletes 1 frame and can't reach and distort
Effect, common frame deletion are all to delete continuous multiple frames.As shown in Figure 10, it is video frame deletion schematic diagram, is in Fig. 1
Video sequence after 6-8 frame deletion.Figure 11 is that a video frame deletion distorts example, indicate to delete in original video the 5th,
6, the sequence of frames of video after 7,8 frames, visually sees, the movement not too big variation of people's running in frame deletion rear video,
Still it runs to the left, observer will not feel lofty visual effect, so being only intuitively not easy to judge whether there is frame
Deletion is distorted.
Step 6.1, it calculates Spearman related coefficient: utilizing the Spearman phase between the GIST feature of adjacent two frame
Relationship number.Spearman coefficient between the GIST feature of i-th frame and the GIST feature of i+1 frame is denoted as zi, for a M
The video of frame obtains M-1 related coefficient z={ z1,z2,...,zM-1}。
Using the phase of LOF (local outlier factor, local outlier factor) algorithm measurement video frame GIST feature
The intensity of anomaly of relationship number.Specific calculating process is as follows:
Step 6.2 calculates the Euclidean distance between related coefficient: calculating each ziWith the Euclidean distance of components other in z
dil,i, l is given positive integer, dil,iFor the l small distance of i-th of related coefficient, as shown by the equation:
dil,i=r (| | zj-z||j≠i},l) (6)
Wherein, function r indicates first of minimum value that data sort from large to small in set.In view of continuously there are 24 frames
When picture, human eye can not individually differentiate every frame picture, therefore l=24；
Step 6.3, it calculates local density: considering dil,iVideo frame GIST characteristic correlation coefficient is not can accurately reflect
Therefore intensity of anomaly introduces local density lrl,i:
Ml,i={ zj|||zj-zi| | < dil,i,j≠i} (8)
Wherein, Ml,iIt indicates in ziL small distance in all video frames, | | indicate set in element number, with
The increase of distance, a possibility that data exception also become larger therewith, therefore, local density is due to containing the aggregation between data
With dispersing characteristic, so it can reflect a degree of data exception；
The intensity of anomaly of step 6.4 calculating Spearman coefficient: under normal conditions, the intensity of anomaly of data simultaneously also with its
The data of surrounding are related, therefore our definition data ziAbnormality degree LOl,iAre as follows: Ml,iIn data be averaged local density and number
According to ziLocal density ratio, as shown by the equation:
Therefore, LOl,iSize mean that the Spearman between the GIST feature of the i-th frame and the GIST feature of i+1 frame
Coefficient ziIntensity of anomaly.
Step 6.5 video frame deletion tampering detection
The given video comprising M frame, we are using abnormal degree series { LOl,1,LOl,2,...,LOl,M-1Video frame described
The abnormal conditions of GIST feature Spearman related coefficient, if there is frame deletion situation in video, sequence of frames of video is adjacent
The abnormality degree of interframe GIST feature Spearman related coefficient just will increase, even LOl,I>Scd(wherein ScdFor threshold value, testing
In, we measure S by many experimentscd=7.8), then there is frame deletion situation between the i-th frame and i+1 frame.Figure 16 gives
To the abnormality degree test result of the video sequence in Figure 11.As can be seen from Figure 16, the 4th value in abnormal degree series is much larger than
Other values, this is because caused by sequence variation degree increases, this result meets in Figure 11 after deleting frame between the 4th frame and the 5th frame
Frame deletion situation.
Illustrate beneficial effects of the present invention from experimental result:
This experiment is mainly used to test the verification and measurement ratio of video frame copy detection method of the invention.For quantitatively evaluating algorithm
Detection performance, the concept that we define verification and measurement ratio is as follows:
Then, our methods described in step 3 detect, and the testing result obtained under different situations is as follows:
Verification and measurement ratio of the video frame copy detection method of 1. step 3 of table description to the video frame duplication of disparate databases
The video frame copy detection method of 2. step 3 of table description is to video frame copy detection rate captured by different brands camera
Verification and measurement ratio of the video frame copy detection method of 3. step 3 of table description to the frame duplication of different-format video
The video frame copy detection method of 4. step 3 of table description is compared with the verification and measurement ratio result of Existing methods
From table 4, it can be seen that the verification and measurement ratio that method of the invention replicates frame is better than Existing methods, and Existing methods
Middle part cannot detect single frames duplication and distort.From table 1 to table 3 as can be seen that method of the invention is in disparate databases
Video has very high frame copy detection rate, and is not influenced by video capture camera, also has for the video of different-format
Higher frame copy detection rate.
Video frame replicates tampering detection robust analysis in method of the invention:
Robustness refers to that after video experienced the operation of content holding, detection method is still effective.In order to detect this
The robustness of the method for invention, we used 95 videos of SULFA database as test video.These videos are distinguished
Carried out following operation: then step 4 is used in fuzzy, contrast enhancing, mirror image (horizontal, vertical), rotation, scaling, gamma correction
Described method is detected, and the results are shown in Table 5.As can be seen from Table 5, it is kept when test video have passed through above content
Operation after, the method for the present invention still maintains higher verification and measurement ratio, this is because these map functions will not change relatively
The uniqueness of Euclidean distance value.
The robustness test result of the video frame copy detection method of 5. step 3 of table description
The detection performance of video frame insertion altering detecting method of the invention, two evaluation indexes that we use are respectively
Recall ratio (recall) and precision ratio (precision)；Recall ratio refers to the ratio correctly detected in all videos being tampered
Example.Precision ratio refers to the percentage that video is correctly detected in the video of all detections, we provide recall ratio ReWith precision ratio Pr
Formulation definition:
Wherein, NcIndicate true positives frame number, that is, what is be correctly detected out distorts video frame number；NmIndicate false positive frame number, i.e.,
It is erroneously detected as distorting the video frame number of frame；NfIndicate false negative frame number, that is, be missed survey distorts video frame number.
Video frame described in 6. step 4 of table is inserted into detection method and is inserted into the test result distorted to video frame
| Method | Nc | Nm | Nf | Re | Pr |
| Proposed method | 128 | 8 | 2 | 94.11% | 98.46% |
| Chao | 2863 | 137 | 140 | 95.43% | 95.34% |
| Zheng | 223 | 3 | 14 | 98.67% | 94.09% |
Video frame described in 7. step 4 of table is inserted into detection method to the result of different insertion frame number detections
Frame is carried out to test video using the insertion detection method of video frame described in step 4 and is inserted into tampering detection.?
In experiment, we are threshold value S firstin, measured by many experiments, work as SinWhen=4.7, for video frame insertion operation, originally
The recall ratio of inventive method is 94.11%, and precision ratio (has reached opposite between recall ratio and precision ratio to put down for 98.46%
Weighing apparatus).Table 6 lists testing result.From testing result as can be seen that compared with prior art, method of the invention is with higher
Precision ratio, but recall ratio is reduced to a certain extent, this is because method of the invention is more concerned about detection accuracy.
Table 7 lists the insertion detection method of video frame described in step 4 to the testing result of different insertion frame numbers, surveys
The sequence of frames of video of examination is divided into two groups, wherein 34 sequence of frames of video are less than the insertion of 25 frames, 102 sequence of frames of video are
The insertion of 100 frames.From table 7 it will be seen that when being inserted into frame number less than 25 frame, video frame of the invention is inserted into detection method
More effectively, when being inserted into frame number greater than 25 frame, video frame insertion detection method precision of the invention is higher.
Frame deletion inspection is carried out to test video using video frame deletion detection method described in step 5 in the present invention
It surveys.In an experiment, our threshold value S firstcd, measured by many experiments, work as ScdWhen=7.8, reaches recall ratio and look into
Relative equilibrium between quasi- rate.Table 8 lists the testing result for deleting different frame numbers, as the result of the prior art, deletes
Recall ratio and precision ratio when the recall ratio and precision ratio of 25 frames are lower than 100 frame of deletion.This is because the texture between consecutive frame
Feature is more similar, and when the frame number of deletion is less, detection accuracy can be reduced.But our scheme is 25 deleting frame number
When frame, recall ratio and precision ratio are superior in the prior art as a result, illustrating that our scheme also has when deletion frame number is less
There is preferable effect.
Video frame deletion detection method described in 8 step 5 of table is to the testing result for deleting different frame numbers
The robust analysis of video frame insertion detection method and frame deletion detection method of the invention:
Frame insertion is distorted with frame deletion tests 50 videos respectively, wherein video 45 are distorted, original video 5.It is former
Beginning video is all from KTH database.Table 9 is the robustness test result that video frame described in step 4 is inserted into detection method, table
10 be the robustness test result of video frame deletion detection method described in step 5.Due to the detection of frame insertion and frame deletion
The related coefficient of consecutive frame GIST feature, the operation that the key reaction of robustness result is kept in content have all been used in scheme
In influence to content frame texture.From table 9 and table 10 as can be seen that our scheme is to geometric operations such as scaling, mirror image, rotations
All it is robust, illustrates it with certain resist geometric attacks performance.
The robustness test result of the insertion detection method of video frame described in 9. step 4 of table
| 90 degree of selection | Select 180 degree | 0.5 times of scaling | 2 times of scaling | Horizontal mirror image | Vertical mirror | |
| Re | 93.75% | 93.75% | 95.83% | 91.67% | 93.75% | 93.75% |
| Pr | 95.74% | 95.74% | 95.83% | 95.65% | 95.74% | 95.74% |
The robustness test result of video frame deletion detection method described in 10. step 5 of table
Claims (10)
1. a kind of video tamper detection method based on frame-to-frame correlation, which comprises the following steps:
Step 1, video is decoded into independent sequence of frames of video, extracts the GIST feature of each frame；
Step 2, the Euclidean distance for the GIST feature extracted through step 1 is calculated；
Step 3, the video frame obtained through step 1 is handled using filter group, then compares and is calculated through step 2
The Euclidean distance of the GIST feature of each frame, obtains according to comparing result with the presence or absence of duplicated frame；
Step 4, video frame is inserted into tampering detection: the Spearman for calculating the adjacent interframe GIST feature obtained through step 1 is related
Coefficient measures the correlation of consecutive frame, then judges whether there is insertion frame according to correlation；
Step 5, video frame deletion tampering detection: between the GIST feature for calculating adjacent two frame using the calculation method of step 4
Spearman related coefficient, using the intensity of anomaly of the related coefficient of LO algorithm measurement video frame GIST feature, then basis
Intensity of anomaly is obtained with the presence or absence of frame deletion.
2. a kind of video tamper detection method based on frame-to-frame correlation according to claim 1, which is characterized in that described
Specific steps in step 1 are as follows: video to be detected is decoded into and is independent sequence of frames of video, frame sequence is carried out one-dimensional
Haar wavelet transformation extracts low frequency component, as new frame sequence；Include the video of M frame to one, remembers that new frame sequence is
{ 1,2 ..., M } extracts the GIST feature of each frame, and the GIST feature for obtaining M frame is denoted as: F={ F1,F2,...,FM, wherein
Fi={ fi,1,fi,2,...,fi,qBe the i-th frame GIST feature, q be GIST feature dimension.
3. a kind of video tamper detection method based on frame-to-frame correlation according to claim 2, which is characterized in that described
Specific steps in step 2 are as follows: it is reference frame that the independent sequence of frames of video obtained through step 1, which is chosen first frame, is calculated each
Euclidean distance c between the GIST feature of frame and the GIST feature of reference framei(i=1 ..., M):
Wherein, Fi={ fi,1,fi,2,...,fi,qBe the i-th frame GIST feature, q be GIST feature dimension.
4. a kind of video tamper detection method based on frame-to-frame correlation according to claim 3, which is characterized in that described
The specific steps are filtering processings in step 3: using the Gabor filter group in 4 scales, 8 directions to the view obtained through step 1
Frequency frame is filtered；The Euclidean distance for comparing the GIST feature for each frame being calculated through step 2, if ci=cj, then i-th
There is duplication situation between frame and jth frame.
5. a kind of video tamper detection method based on frame-to-frame correlation according to claim 1, which is characterized in that described
Step 4 specifically includes: calculating the Spearman related coefficient z between the GIST feature of adjacent two frame firsti, then define i-th
The correlation distance of the GIST feature of the GIST feature and i+1 frame of frame is CiIf two correlation distances of note insertion boundary
Maximum value and second largest value are respectively Cmax、Csec, and the ratio between the second largest value of correlation distance and average value are greater than some threshold value Sin, whenWhen, it is considered as in this frame sequence that there are frame insertions.
6. a kind of video tamper detection method based on frame-to-frame correlation according to claim 5, which is characterized in that described
The correlation distance of the GIST feature of the GIST feature and i+1 frame of i-th frame is C in step 4iAre as follows:
7. a kind of video tamper detection method based on frame-to-frame correlation according to claim 5, which is characterized in that described
The Spearman related coefficient between the GIST feature of adjacent two frame is calculated in step 4 are as follows:
For the video of a M frame, F is enabledi={ fi,1,fi,2,...,fi,qBe the i-th frame GIST feature, q be GIST feature dimension
Number, using GIST feature calculation method, takes q=512；To Fi、Fi+1It is ranked up, while being ascending or descending order, respectively obtain two
The ordered set F' of group elementi={ f'i,1,f'i,2,...,f'i,q}、F'i+1={ f'i+1,1,f'i+1,2,...,f'i+1,q, it will collect
Close F'i、F'i+1In corresponding element subtract each other, to obtain a difference set:
kj=f'i,j-f'i+1,j, { k1,k2,...,kq} (3)
Spearman related coefficient between the GIST feature of i-th frame and the GIST feature of i+1 frame are as follows:
8. a kind of video tamper detection method based on frame-to-frame correlation according to claim 7, which is characterized in that described
The calculation method of intensity of anomaly in step 5 are as follows: between the GIST feature of the i-th frame of meter and the GIST feature of i+1 frame
Spearman related coefficient is denoted as zi, for the video of a M frame, obtain M-1 related coefficient z={ z1,z2,...,zM-1,
Using the intensity of anomaly of the related coefficient of LO algorithm measurement video frame GIST feature, including the Europe calculated between related coefficient
Formula distance dil,iWith local density lrl,i, then introducing indicates in ziL small distance in all video frame Ml,i, obtain public affairs
Formula
Wherein, LOl,iSize mean that the Spearman between the GIST feature of the i-th frame and the GIST feature of i+1 frame is related
Coefficient ziIntensity of anomaly；Introduce threshold value ScdIf LOl,I>Scd, then have frame deletion situation between the i-th frame and i+1 frame.
9. a kind of video tamper detection method based on frame-to-frame correlation according to claim 8, which is characterized in that described
The calculation method of Euclidean distance in step 5 are as follows: calculate each ziWith the Euclidean distance di of components other in zl,i, l is to give just
Integer, dil,iFor the l small distance of i-th of related coefficient, as shown in formula (4):
dil,i=r (| | zj-z||j≠i},l) (6)
Wherein, function r indicates first of minimum value that data sort from large to small in set.
10. a kind of video tamper detection method based on frame-to-frame correlation according to claim 8, which is characterized in that institute
State the calculation method of local density in step 5 are as follows:
Ml,i={ zj|||zj-zi| | < dil,i,j≠i} (8)
Wherein, Ml,iIt indicates in ziL small distance in all video frames, | | indicate set in element number.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201811373982.3A CN109726733A (en) | 2018-11-19 | 2018-11-19 | A kind of video tamper detection method based on frame-to-frame correlation |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201811373982.3A CN109726733A (en) | 2018-11-19 | 2018-11-19 | A kind of video tamper detection method based on frame-to-frame correlation |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN109726733A true CN109726733A (en) | 2019-05-07 |
Family
ID=66294619
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201811373982.3A Pending CN109726733A (en) | 2018-11-19 | 2018-11-19 | A kind of video tamper detection method based on frame-to-frame correlation |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN109726733A (en) |
Cited By (7)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN110418129A (en) * | 2019-07-19 | 2019-11-05 | 长沙理工大学 | Digital video interframe altering detecting method and system |
| CN110796053A (en) * | 2019-10-21 | 2020-02-14 | 北京奇艺世纪科技有限公司 | Video detection method and device, electronic equipment and computer readable storage medium |
| CN110853033A (en) * | 2019-11-22 | 2020-02-28 | 腾讯科技（深圳）有限公司 | Video detection method and device based on inter-frame similarity |
| CN111144314A (en) * | 2019-12-27 | 2020-05-12 | 北京中科研究院 | Method for detecting tampered face video |
| CN111860414A (en) * | 2020-07-29 | 2020-10-30 | 中国科学院深圳先进技术研究院 | Method for detecting Deepfake video based on multi-feature fusion |
| CN114022928A (en) * | 2021-10-22 | 2022-02-08 | 长沙理工大学 | Dual-stream-based method and system for detecting Deepfakes video |
| CN117315678A (en) * | 2023-08-24 | 2023-12-29 | 安徽省征信股份有限公司 | Digital image tampering identification method, system, equipment and storage medium |
Citations (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| EP1276328A2 (en) * | 2001-07-13 | 2003-01-15 | Motorola, Inc. | Watermarking in video transmissions |
| CN105657435A (en) * | 2016-01-25 | 2016-06-08 | 福建师范大学 | Single video frame copy and paste tamper detection method based on quantized DCT coefficient |
-
2018
- 2018-11-19 CN CN201811373982.3A patent/CN109726733A/en active Pending
Patent Citations (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| EP1276328A2 (en) * | 2001-07-13 | 2003-01-15 | Motorola, Inc. | Watermarking in video transmissions |
| CN105657435A (en) * | 2016-01-25 | 2016-06-08 | 福建师范大学 | Single video frame copy and paste tamper detection method based on quantized DCT coefficient |
Non-Patent Citations (1)
| Title |
|---|
| 汤超: "基于帧间相关性的视频篡改检测方法研究", 《中国优秀硕士学位论文全文数据库信息科技辑》 * |
Cited By (14)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN110418129B (en) * | 2019-07-19 | 2021-03-02 | 长沙理工大学 | Digital video interframe tampering detection method and system |
| CN110418129A (en) * | 2019-07-19 | 2019-11-05 | 长沙理工大学 | Digital video interframe altering detecting method and system |
| CN110796053B (en) * | 2019-10-21 | 2022-07-29 | 北京奇艺世纪科技有限公司 | Video detection method and device, electronic equipment and computer readable storage medium |
| CN110796053A (en) * | 2019-10-21 | 2020-02-14 | 北京奇艺世纪科技有限公司 | Video detection method and device, electronic equipment and computer readable storage medium |
| CN110853033A (en) * | 2019-11-22 | 2020-02-28 | 腾讯科技（深圳）有限公司 | Video detection method and device based on inter-frame similarity |
| CN110853033B (en) * | 2019-11-22 | 2022-02-22 | 腾讯科技（深圳）有限公司 | Video detection method and device based on inter-frame similarity |
| CN111144314A (en) * | 2019-12-27 | 2020-05-12 | 北京中科研究院 | Method for detecting tampered face video |
| CN111144314B (en) * | 2019-12-27 | 2020-09-18 | 北京中科研究院 | Method for detecting tampered face video |
| CN111860414A (en) * | 2020-07-29 | 2020-10-30 | 中国科学院深圳先进技术研究院 | Method for detecting Deepfake video based on multi-feature fusion |
| CN111860414B (en) * | 2020-07-29 | 2023-10-24 | 中国科学院深圳先进技术研究院 | Method for detecting deep video based on multi-feature fusion |
| CN114022928A (en) * | 2021-10-22 | 2022-02-08 | 长沙理工大学 | Dual-stream-based method and system for detecting Deepfakes video |
| CN114022928B (en) * | 2021-10-22 | 2024-07-26 | 长沙理工大学 | DEEPFAKES video detection method and system based on double streams |
| CN117315678A (en) * | 2023-08-24 | 2023-12-29 | 安徽省征信股份有限公司 | Digital image tampering identification method, system, equipment and storage medium |
| CN117315678B (en) * | 2023-08-24 | 2024-04-26 | 安徽省征信股份有限公司 | Digital image tampering identification method, system, equipment and storage medium |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN109726733A (en) | A kind of video tamper detection method based on frame-to-frame correlation | |
| Kang et al. | Robust median filtering forensics using an autoregressive model | |
| Wang et al. | Detection of image region duplication forgery using model with circle block | |
| Zeng et al. | Image splicing localization using PCA-based noise level estimation | |
| Ng et al. | Blind detection of photomontage using higher order statistics | |
| Al-Qershi et al. | Passive detection of copy-move forgery in digital images: State-of-the-art | |
| Muhammad et al. | Copy-move forgery detection using dyadic wavelet transform | |
| CN102226920B (en) | Cutting-resistant JPEG image compression history and synthetic tamper detection method | |
| Zhu et al. | Blind image splicing detection via noise level function | |
| CN103747271B (en) | Video tamper detection method and device based on mixed perceptual hashing | |
| CN104244016B (en) | A kind of H264 video contents altering detecting method | |
| CN109903302A (en) | A kind of altering detecting method for stitching image | |
| CN107808100B (en) | Steganalysis method for specific test sample | |
| CN110457996A (en) | Moving Objects in Video Sequences based on VGG-11 convolutional neural networks distorts evidence collecting method | |
| CN102957915A (en) | Double JPEG (Joint Photographic Experts Group) compressed image-targeted tempertamper detection and tempertamper locating method | |
| Siddiqi et al. | Image Splicing‐Based Forgery Detection Using Discrete Wavelet Transform and Edge Weighted Local Binary Patterns | |
| CN102609947B (en) | Forgery detection method for spliced and distorted digital photos | |
| Yang et al. | Image tamper detection based on noise estimation and lacunarity texture | |
| Chu et al. | Detectability of the order of operations: An information theoretic approach | |
| CN112801037A (en) | Face tampering detection method based on continuous inter-frame difference | |
| Li et al. | Robust median filtering detection based on the difference of frequency residuals | |
| CN106709915B (en) | Image resampling operation detection method | |
| Panchal et al. | Multiple forgery detection in digital video based on inconsistency in video quality assessment attributes | |
| CN105740858B (en) | A kind of image copy detection method based on region of interesting extraction | |
| Yohannan et al. | Detection of copy-move forgery based on Gabor filter |
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 | ||
| RJ01 | Rejection of invention patent application after publication | ||
| RJ01 | Rejection of invention patent application after publication |
Application publication date: 20190507 |