Вы находитесь на странице: 1из 18

PERCEPTUAL VIDEO HASHING BASED ON THE

ACHLIOPTASS RANDOM PROJECTIONS


Sandeep R.
Supervisor: Prof. P. K. Bora
Department of Electronics and Electrical Engineering,
Indian Institute of Technology Guwahati.
Alliance University Interview
19-March-2014
Alliance University Interview 19-March-2014 1/ 18
OUTLINE
1
Perceptual Hashing.
2
Prior Work and Motivation.
3
Theoretical Framework.
4
Proposed Algorithm.
5
Simulation Results.
6
Conclusions and Future Works.
Alliance University Interview 19-March-2014 2/ 18
PERCEPTUAL HASHING
Content-identication problem.
Challenge of nding the perceptually similar videos in a large
database.
Content-authentication problem.
Authenticating the video even though its digital representation is
different from the original.
Cryptographic hashing technique.
Perceptual hashing technique.
Alliance University Interview 19-March-2014 3/ 18
PERCEPTUAL HASHING (CONTD)
Takes two inputs, a video V V and a secret key K K , to
produce a k-bit binary hash value h = H(V, K).
Properties of H(V, K) [Venkatesan 2000, Monga 2005]
1
One-way Function:
V H(V, K).
2
Compactness:
Size(H(V, K)) Size(V) .
3
Perceptual Robustness:
Pr (d
H
(H(V, K), H(V
sim
, K))
1
) 1.
4
Visual Fragility:
Pr

d
H

H(V, K), H

V
di f f
, K

>
2

1.
5
Unpredictability:
Pr (d
H
(H(V, K
1
), H(V, K
2
)) >
3
) 1.
Alliance University Interview 19-March-2014 4/ 18
PRIOR WORK AND MOTIVATION
Prior Work.
Radial projection of key frames. [DeRoover 2005,Roover 2005]
3D-DCT [Coskun 2004,Coskun 2006]
Centroid of gradient orientations. [Lee 2006,Lee 2008]
Multilinear sub-space projections. [Li 2011,Li 2012]
Fast Johnson-Lindenstrauss transform (FJLT). [Lv 2008, Lv 2009]
Motivation.
Distance preserving property.
Computationally efcient.
Secured.
Essential features captured with minor distortion.
Alliance University Interview 19-March-2014 5/ 18
THEORETICAL FRAMEWORK
LEMMA (JOHNSON-LINDENSTRAUSS LEMMA [JOHNSON 1982])
A set of n points in ddimensional Euclidean space can be mapped
into kdimensional Euclidean space without distorting the distances
between any pair of points by more than a factor of (1 ).
k is logarithmic in n and independent of d.
k d.
(A) Points in ddimension. (B) Points in kdimension
FIGURE: Illustration of JL Lemma.
Alliance University Interview 19-March-2014 6/ 18
ACHLIOPTASS RANDOM MATRIX (ARM)
Projection satises JL lemma.
R
ARM
(i, j) =

+1 with probability 0.5


1 with probability 0.5
[Achlioptas 2001,Achlioptas 2003]
Projection involves only additions and subtractions.
Alliance University Interview 19-March-2014 7/ 18
PROPOSED ALGORITHM
FIGURE: Block diagram of the proposed video hashing algorithm
Input
RGB video V
in
, hash size k and algorithm parameters.
Preprocessing and normalization
Convert V
in
into a sequence of gray-level frames, V
gray
.
Normalize V
gray
via temporal sub-sampling and spatial re-sizing to
obtain V of size 646464.
Randomization
Depending on secret key K, randomly select N(32) overlapping 3D
sub-blocks V
i
, i = 1, 2, . . . , N, each of size U V W.
Concatenate the columns of each sub-block V
i
to form vector
f
i
R
d
such that d =U V W.
Construct the feature matrix F R
dN
from f
i
s to obtain
F = [f
1
f
2
. . . f
N
].
Alliance University Interview 19-March-2014 8/ 18
PROPOSED ALGORITHM (CONTD)
Random projections
Apply the ARM projection to F to get the hash matrix H =
1

k
RF.
Hash computation
Generate the intermediate hash vector h

=
N

i=1
h
i
N
, where h
i
is the i
th
column of the matrix H.
Obtain the median of h

.
Binarize the elements of h

to obtain h = [h
1
h
2
. . . h
k
]
T
where
h
j
=

0; h

j
<
1; h

, j = 1, 2, . . . , k.
Output
A hash vector h
Alliance University Interview 19-March-2014 9/ 18
SIMULATION RESULTS
Parameters Details
Number of videos 224 [xiph.org, reefvid.org]
Spatial resolution 192x144, 352x288, 384X288 and 512x288
Frame rates (in fps) 15, 25 and 30
Minimum number of frame 66
Maximum number of frame 1079
Alliance University Interview 19-March-2014 10/ 18
SIMULATION RESULTS (CONTD)
(A) Frame rotation by 10

. (B) Regular frame drop


FIGURE: ROC curves for single attack
Alliance University Interview 19-March-2014 11/ 18
SIMULATION RESULTS (CONTD)
(A) MPEG4 Compression with average CR
30:1.
(B) Gaussian Blurring using a 55 mask
size.
FIGURE: ROC curves for single attack
Alliance University Interview 19-March-2014 12/ 18
SIMULATION RESULTS (CONTD)
(A) Frame rate decreased to 15fps. (B) Contrast decreased by 10%.
FIGURE: ROC curves for single attack.
Alliance University Interview 19-March-2014 13/ 18
SIMULATION RESULTS (CONTD)
(A) AWGN with mean 0 and standard
deviation 60 and brightness increased by
10%. (B) Regular frame dropping and 5

rotation.
FIGURE: ROC curves for multiple attacks.
Alliance University Interview 19-March-2014 14/ 18
CONCLUSIONS AND FUTURE WORKS
Conclusions
Proposed a video hashing algorithm based on random projections
using Achlioptass random matrix.
Proposed algorithm showed excellent robustness to most of the
single and multiple image processing attacks.
Achlioptass random projection being computationally efcient
make the hash suitable for content-based video retrieval
application.
Future Works
Studying the performance of the proposed algorithm under
malicious attacks.
Comparing the performance of the proposed algorithm with that of
the other random projection based techniques like FJLT.
Alliance University Interview 19-March-2014 15/ 18
BIBLIOGRAPHY I
[1] R. Venkatesan, S.-M. Koon, M. Jakubowski, and P. Moulin, Robust image hashing, in Image
Processing, 2000. Proceedings. 2000 International Conference on, vol. 3, 2000, pp. 664 666
vol.3.
[2] V. Monga, Perceptually based methods for robust image hashing, PHDTHESIS, The
University of Texas at Austin, August 2005.
[3] C. De Roover, C. De Vleeschouwer, F. Lefebvre, and B. Macq, Robust video hashing based
on radial projections of key frames, Signal Processing, IEEE Transactions on, vol. 53, no. 10,
pp. 4020 4037, oct. 2005.
[4] C. D. Roover, C. D. Vleeschouwer, F. Lefbvre, and B. M. Macq, Robust image hashing based
on radial variance of pixels. in ICIP (3), 2005, pp. 7780.
[5] B. Coskun and B. Sankur, Robust video hash extraction, in Signal Processing and
Communications Applications Conference, 2004. Proceedings of the IEEE 12th, april 2004,
pp. 292 295.
[6] B. Coskun, B. Sankur, and N. Memon, Spatio-temporal transform based video hashing,
Multimedia, IEEE Transactions on, vol. 8, no. 6, pp. 1190 1208, dec. 2006.
[7] S. Lee and C. Yoo, Video ngerprinting based on centroids of gradient orientations, in
Acoustics, Speech and Signal Processing, 2006. ICASSP 2006 Proceedings. 2006 IEEE
International Conference on, vol. 2, may 2006, p. II.
[8] , Robust video ngerprinting for content-based video identication, Circuits and Systems
for Video Technology, IEEE Transactions on, vol. 18, no. 7, pp. 983 988, july 2008.
Alliance University Interview 19-March-2014 16/ 18
BIBLIOGRAPHY II
[9] M. Li and V. Monga, Desynchronization resilient video ngerprinting via randomized, low-rank
tensor approximations, in Multimedia Signal Processing (MMSP), 2011 IEEE 13th
International Workshop on, oct. 2011, pp. 1 6.
[10] , Robust video hashing via multilinear subspace projections, Image Processing, IEEE
Transactions on, vol. 21, no. 10, pp. 4397 4409, oct. 2012.
[11] X. Lv and Z. Wang, Fast johnson-lindenstrauss transform for robust and secure image
hashing, in Multimedia Signal Processing, 2008 IEEE 10th Workshop on, oct. 2008, pp. 725
729.
[12] X. Lv and Z. J. Wang, An extended image hashing concept: content-based ngerprinting
using fjlt, EURASIP J. Inf. Secur., vol. 2009, pp. 2:12:16, Jan. 2009.
[13] W. B. Johnson and J. Lindenstrauss, Extensions of lipschitz mappings into a hilbert space, in
Contemporary Mathematics, Proceeddings of the conference on Modern Analysis and
Probability, A. B. Richard Beals, Anatole Beck and A. Hajian, Eds., vol. 26, 1984, pp. 189206.
[14] D. Achlioptas, Database-friendly random projections. ACM Press, 2001, pp. 274281.
[15] , Database-friendly random projections: Johnson-lindenstrauss with binary coins, J.
Comput. Syst. Sci., vol. 66, no. 4, pp. 671687, Jun. 2003.
[16] http://media.xiph.org/video/derf/, test video sequences.
[17] http://www.reefvid.org/, test video sequences.
Alliance University Interview 19-March-2014 17/ 18
THANKS TO ONE AND ALL.
Alliance University Interview 19-March-2014 18/ 18

Вам также может понравиться