2015 Online International Conference on Green Engineering and Technologies (IC-GET)
Index Mapping based Hybrid DWT-DCT Watermarking Technique for Copyright Protection of Videos Files Alavi Kunhu, Nisi K, Sadeena Sabnam, Majida A Department of Electronics and Communication Engineering Cochin College of Engineering and Technology Valanchery, India
[email protected] Abstract— In this paper, we propose a new blind colour video watermarking technique for the copyright protection of multimedia colour videos using index mapping concept. The novelty in the presented approach consists in designing a hybrid Discrete Wavelet Transform (DWT) and Discrete Cosine Transform (DCT) based digital video watermarking of colour watermark logo using index mapping technique. The distortion caused by watermarking is assessed by using peak signal to noise ratio (PSNR) and similarity structure index measure (SSIM) and robustness against different types of attacks have been assessed using StirMark. The proposed video watermarking algorithm provides better imperceptibility in harmony with the human visual system and offers higher robustness against signal processing attacks. Keywords— Video watermarking; Copyright protectio; discrete wavelet transform; discrete cosine transform; peak signal to noise ratio; similarity structure index measure.
Saeed AL-Mansoori Applications Development and Analysis Center (ADAC) Mohammed Bin Rashid Space Center (MBRSC) Dubai, UAE
[email protected] required to validate the copyright, the watermark is extracted from the digital interface to prove their legal owner. The watermark must be robust against several attacks either intentional or non-intentional [3]. To be effective, a watermark needs to have certain features such as being imperceptible and undeletable. Any watermarking technique consists of an encoding algorithm which embeds the watermark information into the host cover and decoding algorithm to extract the watermark information from watermarked cover [4]. The objective of this project is to develop a hybrid discrete wavelet transform and discrete cosine transform based video watermarking algorithm which embeds robust ownership watermark information into video frames to protect the copyright ownership II. REVIEW OF DWT AND DCT
I.
INTRODUCTION
As a result of the rapid growth in the digital computer technology, for the last few years there has been a large demand for video watermarking products due to the fact that there are so many videos available at no cost on the World Wide Web, which need to be, copyright protected and ownership authenticated [1]. The protection of intellectual property rights and digital media has become in great demand. Thus, the best way to protect multimedia products against illegal transactions is through hiding information within the cover called digital watermarking. The use of digital watermark technology, which has become more convenient, depends on hiding watermark in a digital medium. The watermark is the information regarding the owner of a copyright for this digital media. This could be a brand image, a serial number or any other digital information that describes the copyright [2]. For this digital interface, a watermark and a digital interface can be merged to make it difficult to be separated from each other. Whenever it is
The proposed digital video watermarking algorithm is based on hybrid Discrete Wavelet Transform (DWT) and Discrete Cosine Transform (DCT). A brief review of DWT and DCT are described below. A. Best Hiding Location Selection Technique
Fig. 1. Best hiding location of video frames
2015 Online International Conference on Green Engineering and Technologies (IC-GET) This proposed algorithm will select the best three watermark logo hiding locations for each video frames using DCT. Here first the 2level DWT decomposed LL2 block of selected video frame layer is divided into non-overlapping 8x8 sub-blocks and find out the 2D-DCT of each sub-block. For each sub-blocks the 16 lower frequencies DCT coefficients are screened to find the best three coefficients with highest magnitude and register their locations [4]. The best three locations which are repeated more are selected and the selected best three locations will vary from one video frame to another according to the spatial frequency contents of the video frame. Fig. 1 graphically shows the best 3 watermark hiding locations of various video frames.
band (FL), middle frequencies sub-band (FM) and high frequencies sub-band (FH). Fig. 3 shows the regions of frequency sub-bands in video frames blocks, while Fig. 4 shows the DCT energy distribution. As it can be seen the highest energy coefficients is concentrated on the left-top corner where the highest energy coefficient is called a DC component while other coefficients stand for AC components. Most of the signal energy lies at low-frequencies sub-band which contains the most important visual parts of the video frames. In addition, the Human Visual System (HVS) is more sensitive to this region.
B. Discrete Wavelet Transform (DWT)
300 200
Discrete wavelet analysis, use the concept of approximations and details. The approximations are the highscale, low-frequency components, which gives the signal its identity, while the details are the low-scale, high-frequency components, which imparts flavor. In the proposed hybrid watermarking technique, the discrete wavelet transform (DWT) decompose a video frame into a lower resolution approximation (LL) as well as horizontal (HL), vertical (LH) and diagonal (HH) detail components. This decomposition process can be repeated to compute multiple "scale" wavelet decomposition and 2 levels DWT decomposition model is shown in Fig. 2 [5, 6, 7]. Accurate aspects of the HVS is the one of main advantages of the wavelet transform and this allows us to use higher energy watermarks in regions that the HVS is known to be less sensitive to, such as the high resolution detail bands {LH, HL, HH}. Embedding watermarks in these regions allow us to increase the robustness of our watermark, at little to no additional impact on video quality. LL2
HL2 HL
LH2
LHl
HH2
100 0 -100 0 2 4
4 6
6 8
Fig. 3. DCT frequency sub bands
8
Fig. 4. DCT Energy Spectrum
Mathematically, the DCT transform can be expressed as follows: y(u, v)
M 1 N 1 2 2 (2m 1)u (2n 1)v u v x(m, n) cos cos M N 2M 2N u 0 v 0
(1)
where αu and αv are given by: 1 / u 1/ v
2,u0 , 1 , u 1,2,...., N 1 2 ,v0 1 , v 1,2,...., N 1
The Inverse DCT transform can be expressed as follows: x(m, n)
HHl
0 2
2 M
2 M 1 N 1 (2m 1)u (2n 1)v u v y(u, v) cos 2M cos 2 N (2) N u 0 v 0
where x(m, n) is the pixel intensity of selected channel of video frame block in row and column and y(u,v) is the DCT transform coefficients in row and column of the DCT block.
Fig. 2. 2 levels DWT decomposition model
III. PROPOSED METHODOLAGY C. Discrete Cosine Transform (DCT) DCT is a process of converting a signal from spatial domain to frequency domain. In the case of multimedia videos, by applying 2D-DCT each video frames are divided into three regions of frequencies called low frequencies sub-
A. Watermark Encoding Algorithm The proposed robust encoding algorithm can embeds up to 8 colour index map colour ownership watermark logo into the colour video file in hybrid frequency and wavelet domain
2015 Online International Conference on Green Engineering and Technologies (IC-GET) using Index Mapping, DWT and DCT respectively. Each colour pixel of the logo will require the insertion of 3 bits in the hybrid DWT-DCT domain. First we need to convert the colour watermark logo into 3bit logo index using index mapping table shown in Table I. Then the colour video file is converted into frames and transformed into RGB color space and selects a suitable channel for watermarking. Then 2-level DWT „haar‟ wavelet decomposition is applied to the selected channel to obtain the LL2 sub-band. Subsequently, the LL2 sub-band is divided into a non-overlapping (8×8) blocks and converted into a frequency domain using DCT. Then, the index mapped ownership watermark information is embedded into a selected best three coefficients location selected using best hiding location selection algorithm using scaled odd/even embedding technique. The complete encoding algorithm is summarized in Algorithm 1, where fk indicates video frame, l(x,y) represents the index mapped watermark information, Qe represents even quantization and Qo represents odd quantization to the nearest integer number. The symbol “Δ” indicates the watermark strength scaling factor. TABLE I. INDEX MAPPING TABLE FOR WATERMARK COLOUR LOGO Logo Colour Combinations Blue Green Yellow
Index Key 1 2 3
Binary Value 001 010 011
Red
4
100
White
5
101
Algorithm 1: Embedding copyright watermark Initialize: DCT block size, Wavelet type, Scaling factor (∆) Input: Selected channel of video frame (fk), Watermark information (lo) Output: Robust watermarked video frame (fwk)
Finding 2-level DWT [LL1, LH1, HL1, HH1] = DWT[fk] [LL2, LH2, HL2, HH2] = DWT[LL1]
Finding DCT of NHB 8×8 sub-blocks of LL2 for k=1 to NHB do for i= 1 to 8 do for j= 1 to 8 do find LL2k sub-block of LL2 end for end for DLL2k(u,v) = DCT {LL2k(i,j)}
end for Finding the best three hiding location
Encoding Index mapped watermark information
for k= 1 to NHB do if lo(x,y) = 0 then
DLL 2 k ( x, y ) Q DLL2k(x,y) = e DLL 2 k ( x, y )
else
(3)
DLL2 k ( x, y ) Qo DLL2 k ( x, y )
DLL2k(x,y) = end if end for
B. Watermark Decoding Algorithm The decoding algorithm extracts the watermarked ownership colour logo information from the DCT and Hash multi watermarked colour satellite image using the discrete cosine transform (DCT). Here first extract the logo index key using Odd/Even extraction technique and convert into watermark colour logo using index mapping table.
DLL 2k (u, v) Odd lr (i, j ) 0 if Q Even
(4)
lr (i, j ) 1
IV. RESULT ANALYSIS The proposed DCT and Hash multi watermarking algorithm performance is tested on various colour video files. Examples are shown in Fig. 5 where each video frame is 2048×2048×3 pixels and Fig. 6 shows 64×64×3 pixels Khalifa University color logos used as ownership watermark information. The different watermark strength scaling factors used are 8, 16 and 24. The distortion caused to the video images was assessed by using the peak signal to noise ratio (PSNR) and the structural similarity index measurement (SSIM), while the recovered watermark logo performance under the various attacks were analyzed using the normalized correlation (NC). The Peak Signal-to-Noise Ratio (PSNR) and the Structural Similarity Index Measurement (SSIM) are used as metric to measure quality of video watermarked frames [9, 11], Meanwhile the Bit Error Rate (BER) and the Normalized Correlation (NC) are used as a metric to measure the quality of extracted watermark colour logo information from the video frames.
2015 Online International Conference on Green Engineering and Technologies (IC-GET) L(o, w)
2 X Y A
A 2 X
2 Y
,
C (o, w)
2 X Y B
X2 Y2 B
,
XY C XY C where fo represents the original video frame and fw represents a watermarked video frame. L, C and S represent the luminance, contrast and structure components respectively, while α, β, γ are parameters used to adjust the relative importance of the luminance, contrast and structure components. S (o, w)
Video1 frame
Video2 frame
Video2 frame
The Normalized Correlation (NC) is a metric used to analyze the performance of the extracted watermark information. If both the original and the extracted watermark are the same, then the value of NC will be 1. NC is given by.
Video4 frame
NC (lox, y lex , y ) / lo 2 x, y
Fig. 5. Colour Video frames
(7)
x, y
where lo, le represents the original watermark and the extracted watermark information. TABLE II. PSNR AND SSIM ANALYSIS OF WATERMARKED COLOUR VIDEO FRAME Δ=8 Δ = 16 Δ = 24 PSNR SSIM PSNR SSIM PSNR SSIM 59.70 0.9997 53.40 0.9984 50.60 0.9971 Video1 59.26 0.9997 53.16 0.9986 50.14 0.9971 Video2 59.28 0.9997 53.23 0.9985 50.09 0.9968 Video3 59.00 0.9998 52.95 0.9990 49.90 0.9980 Video4
Original video frame
TABLE III. NC AND BER ANALYSIS OF EXTRACTED COLOUR LOGO Δ Video1 Video2 Video3 Video4 8 1st level DWT of video frame
2nd level DWT of video frame
Fig. 6. DWT decomposition of video1 frame 16
The Peak Signal to Noise Ratio (PSNR) penalizes the visibility of noise in the video frames. Thus, two video frames that are exactly the same will produce an infinite PSNR value. PSNR is given by: 2552 (5) PSNR 10 log10 MSE The Structural Similarity Index Measurement (SSIM) is a measure that compares local pattern of pixel intensities that have been normalized for luminance and contrast. The higher SSIM is, the larger the similarity between the compared video frames. SSIM is given by:
SSIM ( f , f w ) ( L( f , f w )) .(C ( f , f w )) .( S ( f , f w ))
(6)
24
Extracted logo NC
0.8217
0.8349
0.8457
0.8487
BER
0.3830
0.3582
0.3424
0.3350
NC
0.9908
0.9962
0.9964
0.9988
BER
0.0151
0.0066
0.0061
0.0029
NC
1.0
1.0
1.0
1.0
BER
0.0
0.0
0.0
0.0
Extracted logo
Extracted logo
TABLE IV. NC AND BER ANALYSIS OF EXTRACTED WATERMARK UNDER ROTATION ATTACK Δ 1 3 5 7 9 Degree degree degree Degree Degree NC 8 0.8455 0.8502 0.8536 0.8588 0.8627 BER 0.3368 0.3289 0.3267 0.3242 0.3218 NC 16 0.9948 0.9951 0.9943 0.9935 0.9950 BER 0.0083 0.0081 0.0095 0.0112 0.0090 NC 24 0.9952 0.9963 0.9974 0.9957 0.9982 BER 0.0078 0.0076 0.0085 0.0144 0.0159
2015 Online International Conference on Green Engineering and Technologies (IC-GET) 0.6
TABLE VI. NC AND BER ANALYSIS OF EXTRACTED WATERMARK UNDER SCALING ATTACK Δ 80% 70% 60% 50% 40% 30% NC 8 0.8246 0.8234 0.8198 0.8357 0.8070 0.7858 0.3841 0.9980
0.3975 0.9978
0.4045 0.9971
0.3622 0.9972
0.4325 0.9958
0.4943 0.9907
24
BER NC
0.0029 1.0000
0.0039 1.0000
0.0044 1.0000
0.0056 1.0000
0.0086 1.0000
0.0245 0.9999
BER
0
0
0
0
0
0.0005
As the watermark strength scaling factor increases, both PSNR and SSIM start to decrease. For example when the watermark strength scaling factor Δ = 12 for different images the PSNR was in the range 45 dB to 47 dB, but when Δ = 24 the PSNR was in the range 40dB to 41 dB. The robustness of the DCT and Hash multi watermarking algorithm was tested by applying the various attacks such as JPEG compression, image resize, image cropping, image rotation, image filtering, noise and synchronization attack.
0.4
0.3
0.2
0.1
3
5 7 Video Frame Filter [size]
9
Fig. 8. BER analysis of extracted watermark logo under filter attack
1
Scaling factor = 8 Scaling factor = 16 Scaling factor = 24
0.95 Normalized Correlation (NC)
16
BER NC
Scaling factor = 8 Scaling factor = 16 Scaling factor = 24
0.5
Bit Error Rate (BER)
TABLE V. NC AND BER ANALYSIS OF EXTRACTED WATERMARK UNDER FILTER Δ Filter 3x3 Filter 5x5 Filter 7x7 Filter 9x9 NC 8 0.8147 0.7784 0.7697 0.7854 BER 0.4179 0.5164 0.5893 0.5927 NC 16 0.9740 0.9275 0.8559 0.8013 BER 0.0473 0.1699 0.3618 0.5131 NC 24 0.9860 0.9661 0.9400 0.8675 BER 0.0208 0.0591 0.1347 0.3175
0.9
0.85
0.8
0.75
0.7 0.2
Scaling factor = 8 Scaling factor = 16 Scaling factor = 24
0.4
0.5 0.6 Video Frame Scaling [%]
0.7
0.8
Fig. 9. NC analysis of extracted watermark logo under scaling attack
0.5
0.9
0.45
0.85
0.35
Scaling factor = 8 Scaling factor = 16 Scaling factor = 24
0.4
Bit Error Rate (BER)
Normalized Correlation (NC)
0.95
0.3
0.8
0.3 0.25 0.2 0.15
3
5 7 Video Frame Filter [size]
9
Fig. 7. NC analysis of extracted watermark logo under filter attack
0.1 0.05 0 0.2
0.3
0.4
0.5 0.6 Video Frame Scaling [%]
0.7
0.8
Fig. 10. BER analysis of extracted watermark under scaling attack
2015 Online International Conference on Green Engineering and Technologies (IC-GET)
Scaling factor = 8 Scaling factor = 16 Scaling factor = 24
1.05
References Saeed AL-Mansoori and Alavi Kunhu, “Multi-Watermarking Scheme for Copyright Protection and Content Authentication of DubaiSat-1 Satellite Imagery”, SPIE of Optics and Photonics Conference 2013, California, USA, August 2013. [2] M.Kim, D.Li and S. Hong, “a Robust and Invisible Digital Watermarking Algorithm based on Multiple Transform Method for Image Contents”, Proceedings of the World Congress on Engineering and Computer Science 2013 Vol I, WCECS 2013,San Francisco, USA, October 2013. [3] M. Swanson, M. Kobayashi, A. Tewfik, Multimedia data embedding and watermarking techniques, Proc. IEEE 86 (1998) 1064–1087. [4] A. Al-Gindy, H. Al-Ahmad, R. Qahwaj, and A. Tawfik, "A frequency domain adaptive watermarking algorithm for still colour images," in International Conference on Advances in Computational Tools for Engineering Applications, ACTEA '09. , Beirut, Lebanon, 2009, pp. 186-191. [5] D. Kundur and D. Hatzinakos, “Digital Watermarking using Multiresolution Wavelet Decomposition”, Proceedings, IEEE International Conference Acoustic, Speech, Signal Processing, 1998. [6] Alavi Kunhu and Hussain Al-Ahmad, “A New Multi Watermarking Algorithm Based on DWT and Hash Functions for Color Satellite Images”, IEEE International Conference on Electronics, Circuits, and Systems- ICECS2013, Abu Dhabi, UAE, December 2013. [7] Huang Daren, Liu Jiufen, Huang Jiwu and Liu Hongmei “A DWT based image watermarking algorithm”, ICME IEEE International Conference on Multimedia and Expo, August 2001,pp. 313-316. [8] W. Lie, G. Lin, C. Wu, and T. Wang, “ Robust Image Watermarking on DCT Domain,” ISCAS 2000, IEEE International Symposium on Circuits and Systems, Geneva, Switzerland, May 2000, pp. 228-231. [9] Saeed AL-Mansoori and Alavi Kunhu, “Robust Watermarking Technique based on DCT to Protect the Ownership of DubaiSat-1 Images against Attacks”, IJCSNS International Journal of Computer Science and Network Security, VOL.12 No.6, June 2012. [10] Alavi Kunhu and Hussain Al-Ahmad, “A New Watermarking Algorithm for Color Satellite Images Using Color Logos and Hash Functions”, 5th IEEE International Conference on Computational Intelligence, Communication System and Networks - CICSYN2013, Madrid, Spain, June, 2013. [11] Z. Wang, A. Bovic, H. Sheikh, and E. Simoncelli, "Image Quality Assessment: From Error Visibility to Structural Similarity," IEEE Transactions on Image Processing, 2004, pp. 600-612.
Normalized Correlation (NC)
[1] 1
0.95
0.9
0.85
0.8
0.75 1
3 5 7 Video Frame Rotation [degree]
9
Fig. 11. NC analysis of extracted watermark under rotation attack
Scaling factor = 8 Scaling factor = 16 Scaling factor = 24
0.35
Bit Error Rate (BER)
0.3 0.25 0.2 0.15 0.1 0.05 0 1
3 5 7 Video Frame Rotation [degree]
9
Fig. 12. BER analysis of extracted watermark under rotation attack
V. CONCLUSION This paper proposed a novel digital video watermarking technique using a hybrid DWT-DCT function. In this study, a colour logo is utilized as a watermark and colour video frames are utilized as a cover. Based on our experimental results analysis we have come to the conclusion that the proposed video water marking is highly efficient and is capable of withstanding almost all sorts of signal processing attacks. The distortion caused by watermarking is assessed by using peak signal to noise ratio (PSNR) and similarity structure index measure (SSIM) and robustness against different types of attacks have been assessed using StirMark. Therefore, the proposed method is a very good candidate for copyright protection for video files.
ACKNOWLEDGMENT The authors acknowledge all the reviewers and editor in charge for their valuable comments on improving this paper.
