Ci31560566

Anjanadevi B,Nagesh Vadaparthi, Jyothi V,Satyanarayana Reddy G/ International Journal of
Engineering Research and Applications (IJERA) ISSN: 2248-9622 www.ijera.com
Vol. 3, Issue 1, January -February 2013, pp.560-566
A Novel Approach For Privacy Preserving Videosharing And
Merging
Anjanadevi B,Nagesh Vadaparthi, Jyothi V,Satyanarayana Reddy G
Department of IT, MVGR College of Engineering, Vizianagaram, AP, India

Abstract
In present days rapid growth of internet preserving data mining. A privacy-preserving data
has paved a path for increased utilization of mining technique must ensure that
distributed applications. The number of any information which is disclosed ―cannot be
applications aredrastically increased for a traced to an individual‖ or ―does not constitute an
distribution of video information to various intrusion ―.
places. In recent years, videos are also An improvement in our knowledge about
playingmajor role at various surveillance an individual could be considered an intrusion. The
applications. Hence, propose a novel approach to latter is particularly likely to cause a problem for
share the videos to various places while providing data mining, as the goal is to improve our
privacy.In this paper we used an efficient knowledge. Even though the target is often groups
algorithm to merge the given video. This method of individuals, knowing more about a group does
provides various parameters to preserve the increase our knowledge about individuals in the
privacy and accuracy. Our proposed framework group. This means we need to measure both the
is highly efficient than various existing knowledge gained and our ability to relate it to a
approaches like Smart Cameras, Homomorphic particular individual, and determine if these exceed
Encryption and Secure Multi-Party computation the thresholds.Privacy, therefore, happens to be
to carry out privacy preserving video serious concern in the age of video surveillance.
surveillance. This work opens up a new avenue Widespread usage of surveillance cameras [3], in
for practical and provably secure offices and other business establishments, pose a
implementations of vision algorithms. The significant threat to the privacy of the employees
proposed system along with motion segmentation and visitors. It raises the specter of an invasive `Big
results will be used to detect and track peoples or Brother' society. In this regards, certain privacy laws
objects. have been introduced to guard an individual‗s
privacy/rights. Despite these, video surveillance
Index Terms-Privacy, Video Surveillance, remains vulnerable to abuse by unscrupulous
Sharing algorithm, Merging Algorithm, Vision operators with criminal or voyeuristic aims and to
algorithms institutional abuse for incriminatory purposes.
These legitimate concerns frequently slow the
I. INTRODUCTION deployment of surveillance systems. The challenge
The Growth of internet has increased the of introducing privacy and security in such a
usage and distribution of multimedia content among practical surveillance system has been stifled by
remote locations.In present internet form, lack of enormous computational and communication
security is observed while distributing the overhead required by the solutions.
multimedia information.But security of sensitive The privacy of the system is based on
information [1] is of primary concern in the field of splitting the information present in an image into
commercial,medical, military systems and even at multiple shares. The parameters used for shattering
work places.Privacy plays a major role in internet are primes p1,p2,p3……pi and scale factor ‗sc‘ are
applications. Privacy is pertains to data is "freedom constant for each shattering operation and are in
from unauthorized intrusion". With respect to general assumed to be public. The only possible
privacy-preserving data mining [2]. If users have information leakage of the secret is that retained by
given the authorization to use data for the particular each share. We now analytically show that with an
data mining task, then there is no privacy issue. And optimal parameter selection, the information
also the user is not authorized, what user constitutes retained by a share is negligible.Information privacy
―intrusion‖ A common standard among most is concerned with preserving the confidentiality of
privacy laws (Ex-European Community privacy information and is therefore the most relevant kind
guidelines or the U.S. healthcare laws) is that of privacy with respect to the internet and email
privacy only applies to ―individually identifiable monitoring or electronic monitoring [3]. Therefore,
data‖. By combining intrusion and individually it is essential to develop an efficient method which
identifiable leads to a standard to judge privacy can ensure that the data is not tampered. Though
encryption techniques are popular and assures the

560 | P a g e

integrity and secrecy of information, single point security co-exists in the domain of computer vision.
failure is the major vulnerability [3] for large We exploit the following facts, which are valid for
information (like satellite photos,medical images or most of the computer vision tasks. Meaningful
even video information) contents. Privacy in the images from real-world have following properties
workplace is a very important issue that can cause that are of interest to us.
some controversy. There are certain privacy laws
that are designed to protect employees and when Limited and fixed Range
using video surveillance, privacy of employees must The values that an algorithm can take are
be considered [4]. Privacy issues on video finite and are from a limited fixed range. And more
surveillance cameras are most likely to occur when importantly the range is known aprior [5]. Thus, the
the employees do not realize that they are being algorithms that have multiple possible answers, but
filmed (covert video surveillance) or when cameras only one of them within this possible range, are as
are located in places where people do not want to be valid as solutions that have only one answer. Such
filmed. There are also concerns that, video algorithms need not be useful (or correct) for a
surveillance (particularly covert video surveillance) general purpose (non-vision) tasks. In this approach,
may unfairly target certain minority groups. we have exploited this by designing efficient, secure
Video surveillance is a critical tool for a surveillance solutions that has infinite answers, but
variety of tasks such as law enforcement, personal only one of them in the valid range. Interestingly, in
safety, traffic control, resource planning, and this process, we I also circumvent Oblivious
security of assets, to name a few. Rapid Transfer, thereby gainingin efficiency.
development/deployment of closed circuit television
(CCTV) technology plays a key role in observing Scale Invariance
suspicious behavior. However, the proliferation in The information in the image remains
the use of cameras for surveillance has introduced practically unchanged even if we change the units of
severe concerns of privacy. Everyone is constantly measurement or scale the whole data. This is not
being watched on the roads, offices, supermarkets, true for most non-image information. We exploit
parking lots, airports, or any other commercial this to design a wrapper algorithm which converts a
establishment. This raises concerns such as, partially secure algorithm to a completely secure
watching you in your private moments, locating you one. For example, consider a partially secure
at a specific place and time or with a person, spying algorithm that may reveal the LSB of all the pixels.
on your everyday activities, or even implicitly Suppose this algorithm is run on an input which is
controlling some of your actions. Advantage of scaled (at least by a factor of two) with all the LSBs
video surveillance is it keeps track of video randomized. Then note that with practically no
information for future use and is helpful in change in the output, the original input (before
identifying people in the crime scenes etc. scaling) is completely secure. Keeping in mind the
Disadvantage of the present system is that it is above properties, this design is a vision specific,
difficult to maintain heavy amount of raw video distributed framework for surveillance tasks that
data and Human interaction. This requires higher preserves privacy. The emphasis is on performing
bandwidth for transmitting the visual data. this efficiently, thus facilitating proactive
Privacy preserving video surveillance surveillance. In this framework, each video is
addresses these contrasting requirements of shattered into shares and each one is sent to a
confidentiality and utility. The objective is to allow different site in such a way that each site has no
the general surveillance to continue, without information about the scene (data-specific secret
disrupting the privacy of an individual. This novel sharing). The complete algorithm runs in this
technology addresses the critical issue of ―privacy distributed setup, such that at the end of the
invasionin an efficient and cost-effective way. protocol, all that the observer recovers is the final
Privacy concerns often prevent multiple competitive output from the results obtained at each of the
organizations from sharing and integrating data site.The extent to which employers can use video
taken from various videos. In traditional approaches, surveillance to monitor employees may depend on
various algorithms are developed to prevent privacy the state they are living in and the type of business
in some extent. But in privacy preserving video they are conducting. In most cases, employees
surveillance uses the secret sharing technique to should be notified of the video surveillance that is
achieve complete privacy and efficient computation taking place and the video surveillance should not
of surveillance algorithms. include areas where employees have a right to
expect some privacy. In some cases, covert video
Role of Visual Data surveillance is allowed (for instance, when an
While general purpose secure computation employee is suspected of criminal activity), but an
appears to be inherently complex and oftentimes employer may need to get permission from the
impractical, we show that due to certain "suitable" relevant authorities before implementing this.
properties of visual information, efficiency and General video surveillance for security purposes

561 | P a g e

will, in most cases, not be considered to breach any decrypted by the corresponding private key. These
privacy rights in the workplace.Video surveillance is keys are related mathematically, but the private key
not the only type of monitoring that may be cannot be practically derived from the public key.
conducted in the workplace. In practice, PKC can be used to ensure
Thus, to resolve these problems, secret confidentiality of the data. The messages encrypted
sharing schemes have been proposed based on with a recipient‗s public key can only be decrypted
threshold in 1979. Secret Sharing refers to a method using the corresponding private key. The private key
for distributing a secret among a group of servers of which is known only to the intended receiver.
each of which is allocated a share of the secret [6]. Asymmetric key algorithms are generally found to
The secret can be reconstructed only when the be computationally expensive. Some ofthe popular
shares are combined together; on their own, the PKC algorithms include Pailliers and El-Gamal.
individual shares give no information of the secret.
Several types of secret sharing schemes have been Secure Multi-Party Computation(SMC)
proposed in literature. Shamir's secret sharing In cryptography, secure multi-party
scheme [6] represents the secret as the y-intercept of computation or secure computation or multi-party
an n-degree polynomial, and shares correspond to computation (MPC) is a problem that was initially
points on the polynomial. In contrast, Blakley's suggested by Andrew C. Yao [5] in 1982 paper. In
scheme specifies the secret as a point in n- that publication, the millionaire problem was
dimensional space [3], and gives out shares that introduced: Alice and Bob two millionaires who
correspond to hyper planes that intersect the secret want to find out who is richer without revealing the
point. The primary motivation behind Secret precise amount of their wealth. Yao proposed a
Sharing is of securing a secret over multiple servers. solution allowing Alice and Bob to satisfy their
However, computing functions on the input secretly curiosity while respecting the constraints. The
shared among n-servers requires highly concept is important in the field of cryptography and
communication intensive protocols (which relies on is closely related to the idea of zero-knowledgeness.
some sort of SMC). Furthermore, such schemes In general, it refers to computational
result in huge data expansion, which becomes in- systems in which multiple parties wish to jointly
efficient for large secrets, such as live-videos (as in compute some value based on individually held
our case). For example, Shamir's shares are each as secret bits of information, but do not wish to reveal
large as the original secret, where as Blakley's their secrets to one another in the process. For
scheme is even less space-efficient than Shamir's. example, two individuals who each possess some
Each secret share is a plane, and the secret is the secret information—x and y, respectively—may
point at which three shares intersect. Two shares wish to jointly compute some function f(x,y)
yield only a line intersection. without revealing any information about x and y
The primary motivation is proposing a paradigm other than can be reasonably deduced by knowing
shift for our real-timetasks was to reduce the actual value of f(x,y), where "reasonably
communication. We next show that visual-data has deduced" is often interpreted as equivalent to
certain characteristic properties, which can be computation within polynomial time.
exploited to define a tailor made Secret Sharing The primary motivation for studying
scheme exclusively for visual data. Compared to the methods of secure computation is to design systems
standard CRT based Secret Sharing schemes, our that allow for maximum utility of information
scheme significantly reduces the data expansion by without compromising user privacy. SMC uses
at least a factor of,where is the number of interactions between multiple parties to achieve a
servers/shares. Such reduction is prominent for huge specific task, while keeping everyone oblivious of
data such as live-video feeds, thus making privacy others data. Introducing privacy and security in
preserving video surveillance practical. visual data processing was attempted with
considerable success in different domains. Blind
Video surveillance techniques Public Key vision [7] allows someone to run their classifier on
Encryption (PKC) another person‗s data without revealing the
The process of converting the plaintext (P) algorithm or gaining knowledge of the data.
to cipher text (C) using an algorithm is called Shashank exploited the clustered nature of image
encryption (E). On the other hand, restoring the databases to improve the efficiency of SMC for
plaintext from the cipher text is called decryption example based image retrieval by processing
(D). Public key Encryption (PKC), also known as multiple queries together.
asymmetric cryptography, is a form of cryptography
in which key used to encrypt a message differs from Smart Cameras
the key used to decrypt it. Private Key is kept secret, Smart cameras do surveillance in the
while the public key can be widely distributed. The camera itself or try to mask sensitive information in
message that needs to be conveyed to the recipient is the videos. The former requires expensive
encrypted using his public key. It can only be programmable cameras and are restricted to single

562 | P a g e

camera algorithms. Changing the algorithms is organized as follows, Section-I describes the
tedious and costly. The second approach addresses procedure to preserve the privacy before secret
problem specific concerns in surveillance videos. distribution andSection-II explains the Privacy
Face swapping and face de-identification [8] try to preserving technique using sharing algorithm and
modify face images such that they can be Merging algorithm section-III describesthe
automatically detected, but cannot be correctly ExperimentalResults and finally, Section-IV
recognized. Face detection techniques uses concludes the paper.
programmable cameras to carry out detection and Privacy is a major concern in video
masking of the regions of interest. transmission. Before transmission of video from one
All the above approaches rely on the place to another,it is essential to change the form of
success of detection of interest regions and do not the information to provide security to the
provide any guarantee of privacy. Moreover, the information which is to be transmitted. Therefore, to
original video is lost in all. We note that most of the attain this process, consider a video file (.avi or
current privacy preserving algorithms are based on .mpg) and split into number of frames. Select one of
the generic framework of SMC requires heavy the frames F from video V. For enabling the
communication to achieve secure computation. For distributed secured processing, frame F is
instance, a single multiplication is carried out via distributed to N number of parties.If frame F is
complex distributed protocol involving oblivious directly distributed,there is a chance of information
transfer [9], which is a highly communication leakage.Hence, to avoid information leakage, we
intensive subroutine in SMC. need to maintain privacy to the input frame.In this
Factually, the round-trip time in a LAN is process, scaling (scale the positive integer with each
of the order of a few milliseconds, whereas several pixel) and randomization(generation of random
floating operations take no more than few number and summation with each pixel) are applied
nanoseconds. Clearly, these delays are too high, to the frame. After scaling and randomization, the
while dealing with voluminous data like frame is processed for secret sharing.
surveillance videos. Hence, solutions based on SMC
are impractical for our application. In this work, the II.PRIVACYPRESERVINGTECHNIQUE
paradigm of secret sharing is to achieve private and USING SHARING ALGORITHM AND
efficient computation of surveillance algorithms MERGING ALGORITHM
[10]. Secret sharing (SS) methods [11,12] try to split Input: Capturing a video from camera and
any data into multiple shares such that no share by spilt into frames.
itself has any useful information, but together, Step1: Initially, choose N number of relatively
they retain all the information of the original data. primes( gcd(I,J)=1 the I and J are relatively prime)
However, the standard SS methods, which were where N equal to number of shares.
invented to address secure storage of data, results in Step2: Apply the Scaling (Scale the pixels with
significant data expansion (each share is at least the fixed integer i.e.,sc) and Randomization (adding
size of the data). randomvalues to each pixel) for ensuring accuracy
Computing on the shares is inefficient as it before sharing algorithm.
would require some sort of SMC. In this work, we To obtain the shares from the secret S as
exploit certain desirable properties of visual data share1= S mod 𝑝1 ,
such as fixed range and insensitivity to data scale, to share2= S mod 𝑝2 ,
achieve distributed efficient and secure computation share3=S mod 𝑝3
of surveillance algorithms in the above framework.
This approach also addresses the concerns related to shareN=S mod 𝑝 𝑁 ,
video surveillance, presented below. To achieve where, 𝑃1 , 𝑃2 , … . 𝑃 𝑁 𝑎𝑟𝑒𝑟𝑒𝑙𝑎𝑡𝑖𝑣𝑒𝑙𝑦𝑝𝑟𝑖𝑚𝑒.
distributed secure processing and storage and also Step3: Each share is sent to individual
address these issues more effectively we have computational severs and apply the affine
developed a new system to supporting privacy transformation on each individual computational
preserving video surveillance. server.
But all these techniques have various Step4: Affine Transformation on share1 as 𝑝. 𝑑 𝑖 +
disadvantages such as increase in share size,poor 𝑞.𝑠𝑐𝑚𝑜𝑑𝑝1, where p,q are fixed values which are
contrast ratio in the reconstructed image or other used for privacy and di is pixel values of share1 and
issues related to security before computation of the
sc is scaling factor.
image. Step5: Affine Transformation is applied to each
Hence, to overcome these drawbacks, a independent server and keeps results for obtaining
new approach is proposed using sharing and original result.
Merging algorithms, which are more efficient than
the other techniques. In this approach, we have
MERGING ALGORITHM
utilized Chinese remainder theorem for merging
various shares into a secret. The rest of the paper is

563 | P a g e

We need to show that a solution exists and We have
that is unique modulo M. 880*N1 1 mod 3,
1. To construct a simultaneous solution, Let 520*N2 1 mod 5,
Mi= M / mi,i=1,2,3,…………………,n. 240*N3 1 mod 11,
Where, Mi is the product of moduli except
for ithterm. 165*N4 1 mod 16.
2. In this, GCD(M/mi,mi)=1. Using Extended Solving using the Extended Euclidean algorithm,
Euclidean algorithm we have N1 = 1, N2 = 2, N3 = 5, N4 = -3.
3. We can find Ni such that M/mi*Ni 1 Therefore, x = 2*880*N1 + 3*528*N2 +
mod mi Then, 4*240*N3 + 5*165*N4
4. X a1*(M/m1)*N1+ a2*(M/m2)*N2+ = 2*880*1 + 3*528*2 + 4*240*5 + 5*165*(-3)
………+ ar*(M/mr)*Nr. = 7253 (mod 2640)
= 1973.
5. Therefore, X ai*{M/mi}*Ni
We have x = 1973 as a common solution to
Rest of the terms yield the result to a value
the above system of congruence. All other solutions
zero, Since M/mj 0 mod mi, when i≠j. X satisfies
are of the form 1973+ M*i,i=1, 2, 3 . . . and so on.
all the congruencies in the system. X is the unique
solution for modulo M.
Using Merging Algorithm to solve III. EXPERIMENTAL RESULTS
X 2 mod 3, In this process, consider any video file(eg:
walk.avi), split into number of frames. Choose any
X 3 mod 5, input frame and apply scaling & randomization to
X 4 mod 11, preserve the privacy. Now, Scaled and Randomized
X 5 mod 16. image doesn‘t reveal any useful information. Hence,
Clearly, the moduli are relatively prime in pairs. using the secret sharing technique, we can achieve
M= 3*5*11*16 = 2640. the efficient privacy preserving process. Figure 1.c,
M1 = 2640/3 = 880,M2 = 2640/5 = 528, M3 = Figure 1.d and Figure 1.e are the individual shares
2640/11 = 240, M4 = 2640/16 = 165. which are sent to the computational servers.

Figure 1.:(a)Input Frame taken from video. (b) After Scaling and Randomization (c)Share1 of the input frame
(d)Share2 of the input frame (e)Share3 of the input Frame (f)Transformed Share1 of the input frame

564 | P a g e


Figure 2: (a)Transformed Share2 (b)Transformed Share3 (c) Reconstructed Frame

In this example, we used three computational server. The corresponding
computational servers to perform the Transformed shares are shown in Figure
transformation of shares. There, we applied 2.Finally,Observer will merge these three
transformation formula (p.di+q.SC) mod pi, here transformed shares using efficient sharing and
p,q are positive integers that are used in merging algorithms.Our Transformation will give
reconstruction phase (for retrieving original loss-less result in less time.
image). Here, di is the pixel values of the shares In this section, we also presented few other
and SC is the positive scale factor and pi is the experimental results where sharing and Merging
prime number of the corresponding shares. Affine algorithms applied.
transformation is independent of each

Figure 3.:(a)Input Frame taken from video. (b)Frame After Scaling.(c)After Scaling and Randomization
(d)Share1 of the input frame (e)Share2 of the input Frame (f)Share3 of the input frame

Figure 4: (a)Transformed Share1 (b)Transformed Share2 (c)Transformed Share3(d) Reconstructed Frame

565 | P a g e


The merging algorithm is applied on S. Yu. Online available at
3different types of videos and the PSNR values are http://www.charuaggarwal.net/toc.pdf.
calculated with various scaling factors. The table-1 [3] ManeeshUpmanyu, Anoop M.
below shows the PSNR values for different scaling Namboodiri, KannanSrinathan and C.V.
factors. Jawahar, ―Efficient Privacy Preserving
TABLE 1:PEAK SIGNAL – TO – NOISE – RATIO Video Surveillance‖, In Twelfth
Scaling Factor International Conference on Computer
Image resolution Vision (ICCV), 2009.
33 80 120
[4] Hazel Oliner, ―Email and internal
DFS.AVI(320 x 240) 56.624 99 99
monitoring in the Workplace: Information
DELTA.MPEG (320x200) 53.8 99 99 Privacy and Contracting out‖, Industrial
Law Journal 321 – 322, (2002) 31 (4).
WALK.AVI (320 x 240) 55.025 99 99
[5] A. C. Yao. ―Protocols for secure
computations‖.InProc. 23rd IEEE
Symp.on Foundations of Comp. Science,
pages 160–164, Chicago, 1982. IEEE.
120 [6] C.C. Thien and J.C. Lin, ―Secret image
100 sharing," Computers & Graphics, vol. 26,
no. 5, pp.765 - 770,2002.
80 [7] S. Avidan and M. Butman. ―Blind vision‖.
60 InProc. of European Conference on
40 Computer Vision, 2006.
Series1 [8] Model-based Face de-idenfication
20
Series2 technique is online available at
0 ieeexplore.ieee.org/xpls/abs_all.jsp?arnum
Series3 ber=1640608
[9] C Narasimha Raju, GanugulaUmadevi,
KannamSrinathan and C V Jawahar, ―A
Novel Video Encryption Technique Based
on Secret Sharing‖, ICIP 2008.
[10] B.Anjanadevi, P Sitharama raj,V
Jyothi,V,Valli Kumari. A Novel approach
for Privacy Preserving in Video using
Extended Euclidean algorithm Based on
Chinese remainder theorem. International
IV. CONCLUSION Journal Communication & Network
In this approach, we obtained the results in Security (IJCNS), Volume-I, Issue-II,
much effective manner and also found that the 2011.
computational time is less when compared to the [11] J. C. Benaloh. Secret sharing
other techniques. It is also observed that the size of homomorphisms: keeping shares of a
the output image file is almost equal to the original secret secret. CRYPTO, 283:251–260,
size where in other techniques, it is found that the 1986.
output file size is larger than the original one. [12] Craig Gentry. Fully homomorphic
Hence, our approach is more efficient than encryption using ideal lattices. STOC,
others.This approach opens up a new avenue for pages 169–178, 2009.
practical and provably secure implementations of
various vision algorithms where distribution of data
is over multiple computers. These results as
guidelines along with motion segmentation can
detect and track peoples.

REFERENCES
[1] G. Blakley, ―Safeguarding cryptographic
keys," presented at the Proceedings of the
AFIPS 1979National Computer
Conference, vol. 48, Arlington, VA, June
1997, pp. 313 - 317.
[2] An Introduction to Privacy-Preserving
Data Mining Charu C. Aggarwal, Philip

566 | P a g e

Ci31560566

Recomendados

Recomendados

Más contenido relacionado

La actualidad más candente

La actualidad más candente (20)

Destacado

Destacado (20)

Similar a Ci31560566

Similar a Ci31560566 (20)

Ci31560566