An Intelligent Technique for Video Authentication
-
Upload
journal-of-computing -
Category
Documents
-
view
217 -
download
0
Transcript of An Intelligent Technique for Video Authentication
8/2/2019 An Intelligent Technique for Video Authentication
http://slidepdf.com/reader/full/an-intelligent-technique-for-video-authentication 1/7
An Intelligent Technique for Video Authentication
Saurabh Upadhyay, Sanjay Kumar Singh
Abstract—Video authentication has gained much attention in recent years. However many existed authentication techniques have their own
advantages and obvious drawbacks; we propose a novel authentication technique which uses an intelligent approach for video
authentication. Our methodology is a learning based methodology which uses SVM (support vector machine) for learning and classificationpurpose and a video database as sample data. The proposed algorithm does not require the computation and storage of any digital
signature or embedding of any watermark. Therefore it works for raw videos (videos captured in any situation), and useful for real life
application of authentication. It covers all kinds of tampering attacks of spatial and temporal tampering. It uses a database of more than 1200
tampered and non-tampered videos and gives excellent results with 94.57% classification accuracy.
Index Terms —Authentication, Fragile watermarking, Digital signature, Intelligent techniques.
—————————— ——————————
1 INTRODUCTION
IGITAL video authentication has been a topic of immense
interest to researchers in the past few years. Authentica‐
tion of a digital video refers to the process of determining
that the
video
taken
is
original
and
has
not
been
tampered
with. In some applications the authenticity of video data is of paramount interest such as in video surveillance, forensic in‐
vestigations, law enforcement and content ownership [3]. For example, in court of law, it is important to establish the trust‐worthiness of any video that is used as evidence.
As in another scenario, for example, suppose a stationaryvideo recorder for surveillance purpose, is positioned on thepillar of a railway platform to survey every activity on thatplatform along a side. It would be fairly simple to remove acertain activity, people or even an event by simply removing ahandful of frames from this type of video sequences. On theother hand it would also be feasible to insert, into this video,
certain objects and people, taken from different cameras and indifferent time. A video clip can be doctored in a specific way todefame an individual. On the other hand criminals get freefrom being punished because the video (used as evidence),showing their crime cannot be proved conclusively in the courtof law. In the case of surveillance systems, it is difficult to as-sure that the digital video produced as evidence, is the same asit was actually shot by camera. In another scenario, a newsmaker cannot prove that the video played by a news channel istrustworthy; while a video viewer who receives the videothrough a communication channel cannot ensure that videobeing viewed is really the one that was transmitted [6]. Theseare the instances where modifications cannot be tolerated.Therefore there is a compelling need for video authentication.So video authentication is a process which ascertains that thecontent in a given video is authentic and exactly same as whencaptured. For verifying the originality of received video con-tent, and to detect malicious tampering and preventing varioustypes of forgeries, performed on video data, video authentica-tion techniques are used. These techniques also detect the types
and locations of malicious tampering. In fact a wide range powerful digital video processing tools are available in thmarket that allow extensive access, manipulations and reuse
visual materials[2]. Since different video recording devices anclose circuit television camera system become more convenieand affordable option in the private and public sectors, there a corresponding increase in the frequency in which they arencountered in criminal investigations [4]. The video evidenchave significant role in criminal investigations due to their abity to obtain detailed information from their own. And thehave tremendous potential to assist in investigations [4]. Therfore it would be necessary to take utmost care to make suthat the given video evidence, presented in the court, is authetic.
2. VIDEO TAMPERING
When the content of information, being produced by a givevideo sequence, is maliciously altered, then it is called tampeing of video data. It can be done for several purposes, for instance to manipulate the integrity of an individual. Since wide range of sophisticated and low cost video editing sofware are available in the market that makes it easy to manipulate the video content information maliciously, it projects serous challenges to researchers to be solved
2.1 Video Tampering Attacks
There are several possible attacks that can be applied to altthe contents of a video data. Formally a wide range of authentication techniques have been proposed in the literature b
most of them have been primarily focused on still imageHowever the basic task of video authentication system is prove whether the given video is tampered or not. But in seeral applications, due to large availability of information video sequences, it may be more significant if the authentiction system can tell where the modifications happened (It indcates the locality property of authentication) and how the videis tampered [1]. On considering these where and how, the video tampering attacks can have different classifications. A lot works have been done that briefly address the classificatiobased on where [3], [1]. And some papers address the classication based on how [5]. A video sequence can be viewed ascollection of consecutive frames with temporal dependency,
a three dimensional plane. This is called the regional propert
D
————————————————
Saurabh Upadhyay is with the Department of Computer Science & Engi-neering, Saffrony Institute of Technology, Gujarat, India.
San jay Kumar Singh is with the Department of Computer Engineering,I.T., Banaras Hindu University, Varanasi, India.
JOURNAL OF COMPUTING, VOLUME 4, ISSUE 2, FEBRUARY 2012, ISSN 2151-9617
https://sites.google.com/site/journalofcomputing
WWW.JOURNALOFCOMPUTING.ORG 67
8/2/2019 An Intelligent Technique for Video Authentication
http://slidepdf.com/reader/full/an-intelligent-technique-for-video-authentication 2/7
of the video sequences. When a malicious alteration is per-formed on a video sequence, it either attacks on the contents ofthe video (i.e. visual information presented by the frames of thevideo), or attacks on the temporal dependency between theframes. Based on the regional property of the video sequences,we can broadly classify the video tampering attacks into threecategories: spatial tampering attacks, temporal tampering at-tacks and the combination of these two, spatio-temporal tam-pering attacks [1]. In [13], authors have presented a wide classi-fication of video tampering attacks including the sub classifica-tions of spatial and temporal tampering.
3 PREVIOUS WORK
In last two decades watermark and digital signature basedtechniques have been widely used for the purpose of videoauthentication. Basically fragile watermarking and digital sig-natures are the two commonly used schemes for video authen-tication [1]. The authentication data is embedded in to the pri-mary multimedia sources in fragile watermarking schemes.While in digital signature based schemes, the authentication
data is stored separately either in user defined field, as like, inheader of MPEG sequence or in a separate file. In addition ofthese two techniques, intelligent techniques have also beenintroduced for video authentication [3, 14]. Intelligent videoauthentication techniques are basically learning based tech-niques which use video databases as sample data for the pur-pose of learning (training) [3, 14]. Apart from these, digital sig-nature, watermarking and intelligent techniques, some otherauthentication techniques are also introduced by researchers,which are specifically designed for various cases of maliciousattacks. Genuinely video authentication techniques are broadlyclassified in to four categories: Digital signature based tech-niques, watermark based techniques, intelligent techniques
and other authentication techniques. During the authenticationprocess, digital signatures can be saved in two different ways.Either they can be saved in the header of the compressedsource data or it can be saved as an independent file. Furtherthey can be produced for verification. Since the digital signa-ture remains unchanged when the pixel values of the frames ofthe video are changed, they provide better results in the con-sideration of robustness. In the digital signature basedschemes, the digital signature of the signer to the data dependson the content of data on some secret information which is onlyknown to signer [15]. Hence the digital signature cannot beforged, and the end user can verify the received video data byexamining whether the contents of video data match the in-formation conveyed in the digital signature. In fact, in videoauthentication, the digital signature can be used to verify theintegrity of video data which is endorsed by the signer [15].The Johns Hopkins University Applied Physics Laboratory(APL) has developed a system for digital video authentication[16]. The video authentication system computes secure com-puter generated digital signatures for information recorded bya standard digital video camcorder. While recording, com-pressed digital video is simultaneously written to digital tapein the camcorder and transferred from the camcorder in to thedigital video authenticator. This video authentication systemsplits the video in to individual frames and generates threeunique digital signatures per frame-one each for video, audioand (camcorder) control data-at the camcorder frame rate.
Here the key cryptography is used. One key called a “private”
key is used to generate the signatures and is destroyed whenthe recording is completed. The second key is a “public” keywhich is used for verification. The signatures that are generated make it easy to recognize tampering. If a frame has beenadded, it would not have the signature and will be instantlydetected and if an original frame is tampered the signaturewould not match the new data and it will be detected as tam-pering in verification process.In last two decades, a wide variety of watermark based authentication techniques have been presented by various researchersin literature. Based on the application areas, watermarking canbe classified in different categories [5].In addition of ensuring the integrity of the digital data and rec-ognizing the malicious manipulations, watermarking can beused for the authentication of the author or producer of thecontent. In watermark based video authentication techniquesgenerally, watermarks are embedded in digital videos withoutchanging the meaning of the content of the video data. Furtherthey can be retrieved from the video to verify the integrity ofvideo data. Since the watermarks are embedded in the contentof video data, once the data is manipulated, these watermarks
will also be altered such that the authentication system canexamine them to verify the integrity of video data.Fabrizio et al. use the video authentication template, whichuses bubble random sampling approach for synchronizationand content verification in the context of video watermarking[17]. The authentication template is introduced in order to en-sure temporal synchronization and to prevent content tamper-ing in video data [17]. The owners or producers of informationresources are being worried of releasing proprietary information to an environment which appears to be lacking in security [18]. On the other hand with the help of powerful videoediting tools one can challenge the trustworthiness of digitavideos. Chang-yin Liang et al introduced a video authentica
tion system which is robust enough to separate the maliciousattacks from natural video processing operations with thecloud watermark [19].Intelligent video authentication techniques use video databasesfor learning purpose. The database comprises tampered andnon tampered video clips. An intelligent technique for videoauthentication, proposed by M.Vatsa et al, uses inherent videoinformation for authentication [3], thus making it useful forreal world applications.Apart from digital signature, watermarking and intelligenauthentication techniques, some other techniques are proposedby various researchers in the literature for the purpose of au-thentication of digital videos.
Mohan Kankanhalli et al. proposed a video authenticationtechnique which is based on motion trajectory and crypto-graphic secret sharing [9]. In this technique, the given video isfirstly segmented into shots then all the frames of the videoshots are mapped to a trajectory in the feature space by whichthe key frames of the video shot are computed. Once the keyframes are obtained, a secret frame is computed from the keyframes information of the video shot. These secret frames areused to construct a hierarchical structure and after that finalmaster key is obtained. The authentication technique uses thismaster key to verify the authenticity of the video. Any modification in a shot or in the important content of a shot would bereflected as changes in the computed master cap.
3.1 Limitations of existing video authentication tech-
JOURNAL OF COMPUTING, VOLUME 4, ISSUE 2, FEBRUARY 2012, ISSN 2151-9617
https://sites.google.com/site/journalofcomputing
WWW.JOURNALOFCOMPUTING.ORG 68
8/2/2019 An Intelligent Technique for Video Authentication
http://slidepdf.com/reader/full/an-intelligent-technique-for-video-authentication 3/7
niques
Different challenges are there with the existing video authenti-cation techniques. There is no issue related with the size of au-thentication code in digital signature based authenticationtechniques. However they provide better results regardingrobustness, since the digital signature remains unchangedwhen there is a change in pixel values of the video frames. But
if the location where digital signature is stored is compromisedthen it is easy to deceive the authentication system, which inturn may give wrong decision. On the other hand fragile wa-termark based authentication algorithms perform better thanalgorithms based on conventional cryptography [2]. Fragileand semi fragile watermark based algorithms show good re-sults for detecting and locating any malicious manipulationsbut often they are too fragile to resist incidental manipulations,and robustness is also challenged in watermark based videoauthentication systems. Moreover embedding the watermarkmay change the content of video which is not permissible incourt of law [3].
Most of the other video authentication techniques are estab-
lished for specific tampering attacks. Moreover existing au-thentication techniques are also affected by compression andscaling operations. On considering all these limitations of exist-ing video authentication techniques, we have proposed an in-telligent technique for video authentication which does notrequire computation and storage of any key or embedding ofany secret information in the video data. Instead of our algo-rithm uses a video database of 20 non-tampered originally rec-orded videos and their more than 1200 tampered copies. Thedetails of video database have been given in experimental re-sults and discussion section.
4 PROPOSED METHODOLOGY
To address these challenges we have proposed an effectivevideo authentication algorithm which computes the inherentlocal features information from digital video frames statistical-ly and establishes a relationship among the frames. A SupportVector Machine (SVM) [7] based learning algorithm is thenused to classify the video as tampered or non-tampered. Thealgorithm uses inherent video information for authentication,thus making it useful for real world applications.
4.1 Support Vector Machine Support Vector Machine, pioneered by Vapnik [7], is a power-ful methodology for solving problems in nonlinear classifica-tion, function estimation and density estimation [8]. The main
idea of a support vector machine is to construct a hyper planeas the decision surface in such a way that the margin of separa-tion between two classes of examples is maximized. It per-forms the classification task by constructing hyper planes in amultidimensional space and separates the data points into dif-ferent classes. SVM uses an iterative training algorithm to max-imize the margin between two classes [3, 8]. The mathematicalformulation of SVM is as follows:Let ,
be the training sample of data vectors,where is the input pattern for the example and is thecorresponding desired response. It is assumed that the pattern(class) represented by the subset 1 and the pattern rep-resented by the subset 1, are linearly separable. The
equation of generalized decision function can be written as:
∑ ̄
Where is a nonlinear function representing hidden nodand̄ , , … . , and b is a bias. To otain a non linear decision boundary which enhances the dicrimination power, the above equation can be rewritten as:
∑ ̄
Here ,
is the nonlinear kernel that enhances the discrim
ination power and ∝ is the Lagrangian multiplier [3]. Basicala nonlinear SVM uses a kernel function , to map thinput space to the feature space so that the mapped data bcomes linearly separable. One example of such kernel is thRBF kernel.
, ‖ ‖, 0
Where and represent the input vectors and is the RBparameter [3]. Additional details of SVM can be found in [7].
4.2 Proposed Video Authentication Algorithm
The common attacks on a video data, for tampering, are: framremoval, frame addition, and frame alteration [3, 13]. In [13we have presented an algorithm which takes the labeled videframes as input to SVM and trains a hyper plain that is capabof classifying the unlabelled video frames. We have used thaverage object area and entropy of the video frames as sttistical local information (SL) for training purpose anshowed the results for two common tampering attackframe addition and frame removal. In this paper we havproposed an algorithm which uses the average object area anentropy as statistical local information for training purpose buit takes the labeled videos as input to SVM and output of thtraining is a trained hyper plain which is capable of classifyinunlabelled videos. Here we have shown the results of our alg
rithm for all the three common tampering attacks: frame rmoval, frame addition and frame alteration (Spatial tampeing). However the proposed algorithm can handle all types malicious attacks. Since we are using SVM based learning anclassification technique, it can also differentiate between attacand acceptable operations. The concept of the proposed algrithm is shown in Fig. 1. The proposed video authenticatioalgorithm computes the statistical local features informatiobetween two consecutive video frames. Here we take the abslute difference of every two consecutive video frames. The average object area and entropy of difference frames are used astatistical local features information. They are worked here the basis for SVM learning. This information is computed loca
ly using statistical tools and then classification is performeusing support vector machine [7]. Based on the functionalitthe proposed algorithm is divided into two stages: (1) SVLearning and (2) tamper detection and classification usinSVM.
4.2.1 SVM Learning
SVM learning is the first step of the proposed algorithm, sthat it can classify the tampered and non-tampered video datFor this purpose a database of tampered and non-tamperevideos is used. In SVM learning a kernel is trained. Training performed using a manually labeled training video database. the video in the training data is tampered, then it is assigne
the label 0 otherwise (if it is not tampered) the label is 1.
JOURNAL OF COMPUTING, VOLUME 4, ISSUE 2, FEBRUARY 2012, ISSN 2151-9617
https://sites.google.com/site/journalofcomputing
WWW.JOURNALOFCOMPUTING.ORG 69
8/2/2019 An Intelligent Technique for Video Authentication
http://slidepdf.com/reader/full/an-intelligent-technique-for-video-authentication 4/7
From the training videos, statistical local information (Averageobject area and Entropy) are extracted. This labeled infor-mation is then used as input to the SVM which performs learn-ing and generates a non-linear hyper plane that can classify thevideo as tampered and non-tampered. All these steps involvedin the training of the kernel are explained in the Learning Al-gorithm.
Learning Algorithm Input: Labelled training videos. Output: Trained SVM with a non‐linear hyper plane to classify
tampered and non‐tampered video data. Algorithm:
1. Individual frames are obtained from the different
tampered and non tampered videos.
2. The difference frames of all the videos are obtained by
taking absolute difference between every two consecu‐
tive frames. In absolute difference, we subtract each
pixel value in second frame from the corresponding
pixel value in the first frame.
3. All these difference frames of every video are convert‐
ed into binary frames.
4. The total number of objects in first binary difference
frame and
their
area
are
calculated.
Then
the
average
object area and the entropy of the first binary differ‐
ence frame are computed statistically.
5. If the area of an object in a binary frame isa then the
average object areaof the binary frame would be
1 ∑
Where N is the total number of ob‐
jects
6. If the average object area and entropy of a binary
frame are and then the statistical local infor‐
mation of
that
video
of
the
video
database
would
be
defined as ∑ ,
This statistical local information is a column vector of
size 1 2 where m is the total number of bina‐
ry frames extracted from that video of the database.
7. Steps 1‐6 are performed on all the labeled training
videos and the statistical local information is com‐
puted for each video of the video database.
8. Statistical local information and labels of all the train‐
ing videos are given as input to the Support Vector
Machine.
9. In the learning process the SVM kernel [7] is trained to
classify the tampered and non tampered video data
Output of this training is a trained hyper plain with
classified tampered and non tampered video data.
4.2.2 Tamper detection and classification
We now explain the proposed tamper detection and classifica-
tion algorithm. Input to this classification algorithm is a videodata whose authenticity needs to be established. As performedin SVM learning algorithm, statistical local information of allthe binary frames of the given video is computed and thetrained SVM is used to classify the video. If the SVM classifiesthe input video as tampered then the location of tampering iscomputed. The tamper detection and classification algorithm isdescribed below.
Tamper Detection and Classification Input: Unlabelled video data
Output: Classification result as tampered and non‐tamperedvideo data.
Algorithm:
1. Using steps 1‐6 of the SVM learning algorithm, the sta
tistical local information for the input video is
computed.
2. This statistical local information of the input video da
ta is projected into the SVM hyper plane to classify the
video as tampered or non‐tampered. If the output o
SVM is label 1 for the given input video then the given
video is authentic otherwise it is tampered.
3. Plot the statistical local information (average objec
area and entropy) of all the tampered and non
tampered videos.
4. Local
values
showing
the
maximum
deviation
in
theplot are the values corresponding to the tampered
videos.
5. Plot the trained SVM classifier which shows the sup
port vectors for the training video data
Fig. 2 shows the video frames from a tampered video that hasbeen subjected to frame addition attack. Here a new frame hasbeen inserted at frame position 7. Similarly Fig. 3 shows thevideo frames of a temporally tampered video that has beensubjected to frame removal attack. Here twenty frames aredropped in a video sequence (from frame 21 to frame 40). Infigure 4, a kind of frame alteration attack has been shown in
which a small device is removed from the original frame in thetampered frame.
5 EXPERIMENTAL RESULTS AND DISCUSSION
The proposed algorithm shows excellent results for temporatampering attacks. Fig. 5.1 shows the plot of average objecarea values as statistical local information for the 105 probevideos of the video database in frame addition attack. The plotshows that the average object area values of the 6th, 11th, 14th
and 17th videos are significantly higher as compared to the average object area values of other videos.
JOURNAL OF COMPUTING, VOLUME 4, ISSUE 2, FEBRUARY 2012, ISSN 2151-9617
https://sites.google.com/site/journalofcomputing
WWW.JOURNALOFCOMPUTING.ORG 70
8/2/2019 An Intelligent Technique for Video Authentication
http://slidepdf.com/reader/full/an-intelligent-technique-for-video-authentication 5/7
Since videos 6th, 11th, 14th and 17th are common videos and leadto higher average object area values, these are detected as tam-pered videos subjected to frame addition attack.
In the similar way Fig. 5.2 shows the plot of entropy values asstatistical local information for the 105 probe videos of the vid-eo database in frame addition attack. Henceforth from this plotalso videos 6th, 11th, 14th and 17th are detected as tampered vid-eos. In the similar manner Fig. 6 shows the plot of average ob-
ject area values as statistical local information, for the 105 tam-pered and non-tampered videos from the video database inframe removal attack. The plot shows that the average objectarea values of the binary difference frames 8th, 19th, 27th and 67th videos are significantly higher compared to average object area
values of other probe videos. We dropped here 20 frames for
frame removal attack.
Here the videos 8th, 19th, 27th and 67th are detected as tamperevideos, since these videos are having higher average objearea values as compared to average object area values of othvideos.For spatial tampering, we have modified the spatial content the frames of the video with the help of professional softwaand created the tampered videos for our video database. Thestampered videos include almost all kinds of spatial tamperin
attack. Fig. 7 shows the plot of average object area values astatistical local information for the 66 probe videos of the videdatabase in spatial tampering attack.
In this figure the average object area values of 2nd, 10th, 21st, an32nd videos are comparatively higher than the average objearea values of other videos. Therefore the videos regarding 2n
10th, 21st, and 32nd values in x-axis are declared here as tampered videos.The validation process of proposed tamper detection algorithis performed using a video database which contains twenvideos. Experimental protocols for validation process are follows:
1. Video database contains 20 originally recorded no
tampered videos with 352 frames each captured at 23 fpThe frame size of each video clip is720 576. This vid
data is used as the ground truth. For each of the 20 vide
different copies are created by subjecting them to differe
video tampering attacks. In the database, for frame remov
attack 11 copies are created for each video in which 2
frames have been dropped at different positions and 11 co
ies are created for each video in which 50 frames have bee
dropped at different position. For the frame addition attac
we first select a video within the category. Frames of th
video are inserted at random positions in the remaining vi
eos of the video database to generate 22 tampered copies
each ground truth videos. We thus have 20 ground truth vi
JOURNAL OF COMPUTING, VOLUME 4, ISSUE 2, FEBRUARY 2012, ISSN 2151-9617
https://sites.google.com/site/journalofcomputing
WWW.JOURNALOFCOMPUTING.ORG 71
8/2/2019 An Intelligent Technique for Video Authentication
http://slidepdf.com/reader/full/an-intelligent-technique-for-video-authentication 6/7
eos, more than 400 videos with the frame removal attack
and more than 400 videos with the frame addition attack.
2. For frame alteration (spatial tampering) attacks we used pro-
fessional software. With the help of this software we altered
the contents of the frames of each ground truth video. This
alteration is performed in various aspects, such as, object
addition and object removal from the frames. 22 copies of
each video of the video database are created, subjected tospatial tampering attacks.
3. 15 ground truth videos together with 900 tampered videos
are used to train the support vector machine, for frame re-
moval, frame addition and frame alteration attack. This
SVM training is performed for all the three kinds of attack,
separately with different tampered videos.
4. 10 different non-tampered copies of the remaining 5 ground
truth videos are created and these 50 non-tampered videos
together with more than 300 tampered videos are used as the
probe database to determine the performance of the pro-
posed algorithm.
The performance of the proposed video authentication algo-rithm is evaluated with this experimental protocol.All of the computations are performed using the hardware con-figuration of Pentium ® Dual-Core CPU 2.20 GHz computerwith 2 GB RAM under MATLAB programming environment.The RBF parameter used in the proposed algorithm is comput-ed empirically using the training video frames. The best suitedvalue of RBF parameter (γ ) among 1 to 5 is 3. The value of 3 gives the maximum classification accuracy. We thereforeused 3 for classification on the probe data.The results given in table 1 summarize the performance of theproposed video authentication algorithm. For authentic videosand videos subjected to the frame addition attack, our algo-
rithm gives the result with maximum accuracy and yields100% correct classification. For frame removal attack, a classifi-cation accuracy of 96 is obtained. For frame removal attack ouralgorithm misclassified four tampered videos, because the ob-
jects movements in difference frames were very small. Forframe alteration (spatial tampering) attacks, the proposed algo-rithm gives the result with 85% accuracy and misclassifies 15tampered frames out of 100 tampered frames. Thus the overallclassification accuracy of the proposed algorithm is 94.57%.
Thus the overall classification accuracy of the proposed algo-
rithm is 94.57%. These results show the efficacy of our pro-
posed video authentication algorithm for all the three commontampering attacks, namely frame addition, frame removal at-tack and spatial tampering attacks. We also compared the per-formance of the proposed video authentication algorithm withthe motion trajectory based video authentication algorithm [9]Table 2 depicts a theoretical comparison of both algorithmsMotion trajectory based algorithm [9] is fast and simple butunable to detect some of the tampering attacks (as spatial tam-pering attacks). On the other hand our proposed algorithmuses an intelligent technique, namely SVM classification whichis able to detect both kinds of attack, spatial as well as tem-poral.
Thus our proposed algorithm covers a wide range of tamper-ing attacks with good classification accuracy and a minor in-crease in computational time.
6 CONCLUSION
Video authentication is a very challenging problem and of high
importance in several applications such as in forensic investi
gations of digital video for law enforcement agencies, video
surveillance and presenting video evidence in court of law
Existing video authentication algorithms use watermarking or
digital signature based algorithms. Digital signature based al
gorithm can be deceived, if the digital signature is compro
mised and watermarking based algorithms are not acceptable
in court of law because they have been altered during water
mark
embedding
and
extraction.
To
address
these
issues
wehave proposed an efficient video authentication algorithm
which can detect multiple video tampering attacks. Our pro
posed algorithm computes the statistical local information o
all of the binary difference frames of the given video and pro
jects them into a non‐linear SVM hyper plane to determine i
the video is tampered or not. The algorithm is validated on an
extensive video database containing more than 1200 tampered
and 20 ground truth videos. The results show that the pro
posed algorithm yields a classification accuracy of 94.57%. In
future we would like to extend the proposed algorithm for
rapid camera movement and night vision shot video tamper
ing.
JOURNAL OF COMPUTING, VOLUME 4, ISSUE 2, FEBRUARY 2012, ISSN 2151-9617
https://sites.google.com/site/journalofcomputing
WWW.JOURNALOFCOMPUTING.ORG 72
8/2/2019 An Intelligent Technique for Video Authentication
http://slidepdf.com/reader/full/an-intelligent-technique-for-video-authentication 7/7
. References [1]. Peng Yin, Hong heather Yu, “Classification of Video Tampering
Methods and Countermeasures using Digital Watermarking” Proc. SPIE
Vol. 4518, p. 239-246, Multimedia Systems and Applications IV
[2]. Adil Hauzia, Rita Noumeir (2007) “Methods for image authentication:
a survey.” In: Proceedings of the Multimedia Tools Appl (2008) 39:1-46,
DOI 10.1007/s11042-007-0154-3.
[3]. S. Upadhyay, S.K. Singh, M. Vatsa, and R. Singh, “Video authentication
using relative correlation information and SVM”, In Computational Intelli-
gence in Multimedia Processing: Recent Advances (Springer Verlag) Edited
by A.E. Hassanien, J. Kacprzyk, and A. Abraham, 2007.
[4]. Law Enforcement/Emergence Services Video Association (LEWA).
[5]. Jana Dittman, Anirban Mukharjee and Martin Steinbach, “Media inde-
pendent watermarking classification and the need for combining digital
video and audio watermarking for media authentication”. International
conference on Information Technology: Coding and Computing, 2000.
[6]. P.K. Atrey, W. Yan, and M.S. Kankanhalli, “A scalable signature
scheme for video authentication”, presented at Multimedia Tools Appl.,
2007, pp.107-135.
[7]. Vapnik VN (1995). “The Nature of Statistical Learning Theory”,
Springer Verlag
[8]. Singh R, Vatsa M, Noore A (2006) “Intelligent Biometric infor-
mation fusion using support vector Machine”. In Soft Computing in
Image Processing: Recent Advances, Springer Verlag, 327-350.
[9]. Wei-Qi Yan an Mohan S Kankanhalli, “Motion Trajectory Based
Video Authentication” ISCA(3) 2003: 810-813
[10]. Dajun He, Qibin Sun, Qi Tian, “A semi fragile Object based video
authentication system” IEEE ISCAS 2003, Bangkok.
[11]. R. Gennaro and P. Rohatgi,” How to sign digital Stream”, Crypto’
97, pp. 180-197, 1997.
[12]. J. M. Park, E. K. P. Chong and H. J. Siegel, “Efficient multicast
packet authentication using Signature amortization”, IEEE symposium
on security and privacy, pp. 227-240, 2002.
[13]. Upadhyay, Saurabh; Singh, Sanjay K;, "Learning based video authen-
tication using statistical local information," Image Information Processing
(ICIIP), 2011 International Conference on , vol., no., pp.1-6, 3-5 Nov. 2011 doi:
10.1109/ICIIP.2011.6108953.
[14]. R. Singh, M. Vatsa, S.K. Singh, and S. Upadhyay, “ Integrating SVM
Classification with SVD Watermarking for Intelligent Video Authentication”, In
Telecommunication Systems Journal – Special Issue on Computational
Intelligence in Multimedia Computing, Springer, 2008 IV.
[15]. P. Wohlmacher, “Requirements and Mechanism of IT-Security Including
Aspects of Multimedia Security”, Multimedia and Security Workshop at
ACM Multimedia 98, Bristol, U. K., Sep. 1998.
[16]. Johns Hopkins APL creates system to detect Digital Video Tampering.
http://www.jhu.edu/ .
[17]. Fabrizio Guerrini, Reccardo Leonardi and Pierangelo Migliorati, “ A
new video authentication template based on bubble random sampling”, Proc. of
the European Signal Processing Conference 2004.
[18]. M.P. Queluz, “ Authentication of Digital Images and Video: Generic Models
and a New Contribution”, Signal Processing: Image Communication, Vol.16,
pp. 461-475, January 2001.
[19]. Chang-yin Liang, Ang Li, Xia-mu Niu, “Video authentication and tamp
detection based on cloud model”, Proceedings of the Third International Co
ference on International Information Hiding and Multimedia Signal Pr
cessing (IIH-MSP 2007), p.225-228, November 26-28, 2007.
Saurabh Upadhyay received the B. Tech. degree in computer science aengineering in 2001 and is currently working toward the Ph.D. degree computer science at U.P. Technical University, India. He is an Associa
Professor in the Department of Computer Science and EngineerinSaffrony Institute of Technology Gujarat, India. He is actively involved the development of a robust video authentication system which can identtampering to determine the authenticity of the video. His current areas interest include pattern recognition, video and image processing, watemarking, and artificial intelligence.
Sanjay K. Singh is Associate Professor in Department of Computer Engneering at Institute of Technology, BHU, India. He is a certified Novel Engneer and Novel administrator. His research has been funded by UGC anAICTE. He has over 50 publications in refereed journals, book chapterand conferences. His research interests include computational intelligencbiometrics, video authentication and machine learning. Dr. Singh is member of IEEE, ISTE and CSI.
JOURNAL OF COMPUTING, VOLUME 4, ISSUE 2, FEBRUARY 2012, ISSN 2151-9617
https://sites.google.com/site/journalofcomputing
WWW.JOURNALOFCOMPUTING.ORG 73