Professional Documents
Culture Documents
Abstract: An innovative MPEG-2 based robust, invisible and blind watermarking scheme for video is
presented. The proposed algorithm uses all the DC coefficients from all 8 x 8 block of discrete coefficient
transforms matrices generated from each IDR frame in order to embed the binary watermark. The watermark
can only be extracted by using the key, which also enhance the security of watermark itself. Therefore the
extraction will never possible without knowing the actual key. The robustness is evaluated by testing against
image processing attacks and video processing intentional and non-intentional attacks by evaluating two
parameters as Normalized Correlation and Bit error Rate in order to find the degree of similarity and degree of
dissimilarity respectively between the original and extracted watermark. The superiority of the proposed video
watermarking algorithm is that the excellent robustness and good perceptibility achieved without changing the
motion vectors during the DPCM process of MPEG-2 encoding scheme.
Keyword: Discrete cosine transforms, information retrieval, video compression, MPEG-2 Structure.
If y==0? If wi=0?
Yes Yes Yes Yes
No No
No change to DC value of & it is Change DC value to greatest even Change DC value to greatest
considered that the selected bit (wi=0 or integer lower than the DC value & it odd integer lower than the DC
1) is inserted into the currently selected is considered that the selected value for the currently selected
DCT block watermark bit ‘0’ is inserted into the DCT block & it is considered that
currently selected DCT block . the selected bit ‘1’ is inserted .
No No
Select first
Embedding of watermark bits Embedding process
watermark bit (wj) is started from the scratch completed.
Fig. 3: Watermark Embedding System
selected DCT coefficient then it is considered watermark. The extracted watermark is shown
that the selected one watermark bit ‘0’ is in Fig 4b.
inserted virtually. Now choose DC value from
the next adjacent DCT block. At the same time
also pick the watermark bit (0/1). Repeat the
same procedure to insert the other watermark
bit. This process is continued to embed the
other watermark bits from all the DCT block
from the currently selected I frame. Move to
another GOP to get another I frame for
repeating the same process for inserting the
rest watermark bits. This process is continued
till all watermark bits have been processed. If Fig 4a: Original Fig 4b: Extracted
still some DCT blocks are untouched then Watermark Watermark
watermark bits will be inserted again from the
scratch in order to cover all untouched DCT The simulation results have been evaluated for
block. In this way all DCT blocks from all I three major parameters as defined below
frame has been covered. It can be best
understand by the following example. Suppose 5.1 Perceptibility
5 watermark bits to be insert are { 0,0, 1,1,1}
and the actual DC values are { 183, 94, 72,
The perceptibility factor measures the degree
187, 190} then the updated value of DC
of similarity between the original video with
components are { 182, 94, 71, 187, 189} in
the watermarked video. It is defined by the
order to be considered that the watermark bits
following formula.
have been inserted.
NoofFrames
5 Experimental setup And Result PSNR 1/ NoofFrames
K 1
PSNR (i)
K
Analysis
Where PSNRK = 10 log10 255/ MSEK(mean
Proposed video watermarking algorithm is square error) is the peak signal to noise ration
simulated under MATLAB tool version measures between Kth original video frame and
R2009. Experimental results have been Kth Watermarked Video frame and MSE it is
evaluated by considering the standard videos defined as mean square error between these
namely ‘Akiyo.avi’ . The video is static since two frames, defined by the following formula
a newsreader is having only lips movements –
and rest part of the video including
N 1 M 1 2
Although in all the listed attacks, the best Where W and W ' are the original watermark
robustness results are obtained against frame and the extracted watermark, respectively. M x
replacement attack, frame deletion attack, N is the size of watermark image. NC and
frame swapping, frame averaging attack BER, both the parameters evaluate the
because of NC value in all these case is above robustness but former find the similarities
90%. These attacks must be checked for between the original and extracted watermark
robustness evaluation of any digital video and later described the degree of
watermarking scheme. As compare to previous dissimilarities. The numerical relationship
researchers work, no work have been between these two is that NC is inversely
demonstrate for frame replacement attack for propotional to BER i.e. higher the NC value
robustness evaluation till now & it is as and lower the BER indicates the higher
important as other frame specific attacks. It robustness is obtained for some constant ‘k’.
may be needed for those situation where the In a simple way, when NC=1 and BER=0
commercial video clip as advertisement have reflects that there is no viable difference
to be insert but the entire video length must between the original and the extracted
not be increased. The proposed algorithm also watermark. The proposed algorithm
tested the robustness against various types of calculated the BER between original and
inserting the noise. The results are good for the extracted watermark with and without attacks
small value of noise addition into the as shown in the Table 1.
watermarked video. As far as geometric attack
Table 1: NC, PSNR and BER along with extracted image after applying Attacks on Watermarked Video
No Attack Rotation at 0.1 : 30 frames Crop Speckle Noise at 0.0003 Salt & Pepper Noise at
NC: 0.95343 NC : 0.83419 NC : 0.89403 NC : 0.9404 0.0003:
PSNR: 38.3198 PSNR: 37.6641 PSNR: 34.3369 PSNR: 37.6865 NC : 0.93446
BER: 4.5400 BER: 15.8419 BER: 12.8849 BER:5.7837 PSNR: 38.3036
BER:6.3429
Gaussian Noise Median Filter: Frame Replacement Attack Frame Insertion Attack Frame Deletion Attack
at 0.0003 NC : 0.7354 (12 to 24 Frame Replaced (12 to 24 frame inserted (2 frames Deleted)
NC : 0.8185 PSNR: 37.7444 from ‘cat_video.avi’) : from cat_video.avi) NC : 0.90228
PSNR: 35.4524 BER:28.3974 NC : 0.94187 NC : 0.69888 PSNR: 38.2234
BER:17.2974 PSNR: 38.0504 PSNR: 36.2480 BER: 9.3918
BER:5.6534 BER:29.8682
----
6 State of Art Vs Simulation Results can be compared with those researchers who
have evaluated either NC or BER as shown in
The theme of the proposed technique is that the Table 2. Another important obsession is that no
robustness results can be compared with the existing schemes shown results for evaluating
results of previous research by covering same PSNR after various attacks. Since this factor is
and different variety of existing techniques as directly associated with the quality of
frequency domain based approach, DCT based watermarked video and therefore it cannot be
and motion vector based using MPEG structure overlooked and it is almost as important as
as shown in Table 2. Other consideration is that normalized correlation value. That is why; the
- some of the author evaluated the NC or other proposed scheme demonstrates PSNR along
calculated the BER for robustness point of view. with NC and BER before and after various types
In view of this, the proposed method evaluated of attacks.
both NC and BER so that the simulation results
Table 2: Comparison of the robustness between the proposed algorithm and other previously suggested algorithms
1 Rotation --- 0.60 --- 0.71 --- 49.87 --- 0.9140 38.1441 0.95343 15.8419
2 Cropping --- 0.68 --- 0.73 --- 6.52 --- --- 34.3369 0.89403 12.8849
3 Speckle Noise --- 0.90 0.91 35.04 11.34 --- --- 37.6865 0.9404 5.7837
4 Salt & Pepper --- 0.63 --- --- 34.67 --- --- 0.6684 38.3036 0.93446 6.3429
5 Gaussian Noise --- --- --- --- --- --- --- 0.6847 35..4524 0.8185 17.2974
6 Median Filter --- 0.54 --- 0.63 35.04 6.54 --- --- 37.7444 0.7354 28.3974
7 Wiener Filter --- --- --- --- 35.04 --- --- --- 38.4757 0.69956 28.3974
8 Frame Replacement --- --- --- --- --- --- --- --- 38.0504 0.94187 5.634
9 Frame Insertion --- --- --- --- --- --- --- --- 36.248 0.69888 29.8682
10 Frame Deletion --- 0.90 --- --- --- --- --- 0.8801 38.2234 0.90228 9.3918
11 Frame Swapping --- 0.90 --- --- --- --- --- 0.9011 38.1402 0.94197 5.8143
12 Frame Averaging --- --- --- --- --- --- --- 0.9030 38.0026 0.95343 4.5427
13 Compression --- --- --- --- --- 28.13 --- --- 37.4866 0.7127 27.0109
According to the experimental results, the 3. Another observation is that the existing
following observation made. schemes have shown their results for covering
few attacks. The proposed approach covered
1. It is observed from the above results that the almost all geometric attack, image processing
proposed method shows better results than other attacks and video specific attacks. Till now, no
existing schemes for most of the image and video researcher has evaluated the robustness against
specific attacks. frame replacement attack.
2. The quality of watermarked video is compared
with the original video tested through 4. By analyzing the table, it is concluded that the
perceptibility (PSNR) as shown in the Table 1. Its robustness results are almost better than previous
value lies between 38.0026 (minimum value with work except with [13], the results are tarnished
any type of attack applied) and 38.3198 only for cropping and median filter attack shown
(maximum value without any attack), shows that by bold case letter.
perceptibility is degraded to maximum of 1.23%, 5. Security of extraction of watermark is
which is very nominal. Hence it is concluded that depending upon the key & the key is ‘size of
the proposed algorithm successfully achieved the watermark’ itself. The watermark can never be
perceptibility requirements, most essential feature extracted by the attacker until he/she knows the
of any video watermarking, under any type of key.
intentional or non-intentional attack.
7 Conclusion Algorithm based on MPEG-2 Structure”, Paper in
press in IEEE proceedings, conference held on
In this paper, DCT based robust and 15-16 May’2015.
imperceptible digital video watermarking is [7] Ueno, Y., "A digital video watermarking
proposed by using the MPEG structure. The method by associating with the motion
results have been evaluated for covering the three estimation," Signal Processing, 2004. IEEE
main issues as robustness, imperceptibility and Proceedings. ICSP '04. 2004 7th International
payload capacity. Robustness is being Conference on , vol.3, no., pp.2576,2579 vol.3,
successfully simulated by covering fourteen 31 Aug.-4 Sept. 2004
intentional and non-intentional attacks as shown [8] Yuk Ying Chung, Fan Fei Xu, Faith Choy, “
in the table 1 and the beauty of this proposed Development of Digital Video Watermarking for
algorithm is that the watermark is recovered MPEG-2 Video, Proc of IEEE, 2006, pp. 1-4
successfully without degrading the perceptibility. [9] Danial Cross, Bijan G. Mobasseri, “
The only limitation with this approach is that the Watermarking for self authentication of
payload capacity is depend upon the number of compressed video”, IEEE proceeding,
DCT coefficients available in the I frames. The International Conference on , vol.2, pp.913-916
future work can be extended to test the robustness vol.2, 2002
against applying more attacks like ambiguity [10] Dr. Anil Kumar Sharma, Mr. Yunus
attacks, collusion attacks and joint attacks. Mohammed Pervej, “Simulation and analysis of
digital video watermarking using MPEG-2”
,International Journal on Computer Science and
References Engineering,3(7)2700-2706, 2011
[11] Mahesh R. Saghavi, Dr. Mrs Archana M.,
[1] Chiou-Ting Hsu; Ja-Ling Wu,, "DCT-based Prof. Dr. Rajeev, Kainjan S. Kotecha, “ A Robust
watermarking for video," Consumer Electronics, Scheme for Digital Video Watermarking based
IEEE Transactions on , vol.44, no.1, pp.206,216, on Scrambling of Watermark,” International
Feb 1998 Journal of Computer Applications (0975-8887),
[2] Mobasseri, B.G., "A spatial digital video Vol 35- No. 2, Dec 2011
watermark that survives MPEG," in Information [12] Mirza, H.H.; Thai, H.D.; Nagata, Y.; Nakao,
Technology: Coding and Computing, 2000. Zensho, "Digital Video Watermarking Based on
Proceedings. International Conference on, vol., Principal Component Analysis," Innovative
no., pp.68-73, 2000 Computing, Information and Control, 2007.
[3] You-Ru Lin; Hui-Yu Huang; Wen-Hsing Hsu, ICICIC '07.IEEE Second International
"An embedded watermark technique in video for Conference on , vol., no., pp.290,290, 5-7 Sept.
copyright protection," in Pattern Recognition, 2007
2006. ICPR 2006. 18th International Conference [13] Yuan-Gen Wang, Zhe Ming Lu, Liang Fan,
on , vol.4, no., pp.795-798, 20-24 Aug. 2006 Yun Zheng, “Robust dual watermarking
[4] Satyen Biswas, Sunil R. Das, “An adaptive algorithm for AVS video”, ELSEVIER, Signal
compressed MPEG-2 video watermarking processing image communication, vol 24, pp.
scheme”, IEEE transaction on instrumentation 333-344, 2009,
and measurement”, vol. 54, No.5, pp. 1853- 1861, [14] Lama Rajab, Tahani Al-Khatib, Ali Al-Haj,
Oct 2005. “ Hybrid DWT-SVD Video Watermarking” IEEE
[5] Lu Jianfeng; Yang Zhenhua; Yang Fan; Li Li, proceedings, international conference on, pp.
"A MPEG2 Video Watermarking Algorithm 588-592 2008
Based on DCT Domain," Digital Media and
Digital Content Management (DMDCM), 2011
Workshop on , vol., no., pp.194,197, 15-16 May
2011
[6] Rakesh Ahuja, S S bedi, “ Copyright
Protection using Blind Video Watermarking