Top Banner
Adaptive Video Watermarking Using Motion Information Choong-Hoon Leea and Hwarig-Seok Ohb and Heung-Kyu Leer' a Department of Computer Science & Advanced Information Technology Research Center (AITrc) Korea Advanced Institute of Science and Technology (KAIST) 373-1 Gusong-Dong Yusong-Gu, Taejon, Korea bVisual Information Research Team Virtual Reality Research and Development Center Electronics and Telecommunications Research Institute (ETRI) 161 Kajeong-Dong Yusong-Gu, Taejon, 305-350, Korea ABSTRACT This paper presents an adaptive video watermarking using motion information. Because video data have one more dimension than image data, simple adoptation of image watermarking method to video data would reveal some types of visual artifact such as flickering. In the proposed scheme, same watermark information is embedded for same region in each frame to decrease such visual artifact. For higher robustness and invisibility, watermarking strength is adjusted by motion information and region complexity. Keywords: Video Watermarking, Wavelet Transform, Motion Estimation 1. INTRODUCTION Advance of computing power and network technology made multimedia data processing and distribution easier. Accordingly, copy or forgery of multimedia data is easier than before. In recent, multimedia data are high value- added products and illegal copy and forgery are serious problems in multimedia industry. Digital watermarking is newly emerging technique for solving such problems.1 Digital watermarking is a sort of information hiding technique that can be used for ownership assertion, fin- gerprinting, authentication, content labeling, usage control or content protection.2 In watermarking, user specific signal, watermark, is embedded into host data such as image, video, audio or any other types of multimedia data in invisible form. Embedded watermark is retrieved later for above purpose. Digital watermarking must satisfy some requirements such as imperceptibility, robustness and reliability to be used for copyright protection.3 Imperceptibility means that embedded watermark cannot be seen or must not degrade quality of host data. Although visible type of watermarking exists, it is not desirable because visible watermark degrades quality of original data and is more easily removable. Second requirement, robustness, means that embedded watermark must remains after several attacks for removing it. Several types of attack are possible such as data compression or signal processing. The embedded watermark information can be damaged by lossy video compression technique such as MPEG, H.261 or H.263. In addition to, several types of signal processing such as noise addition, AD/DA conversion, blurring, histogram modification can change embedded signal. Watermark must endure such attacks. Reliability means that embedded watermark must not be detected by unauthorized person and must be retrieved correctly by authorized one. Because video data have one more dimension - time dimension - than image data and successive similar frames are presented with very small time gap, visibility problem in video watermarking is more important than in image watermarking. Bad watermarking scheme for video would reveal visual artifact that is not seen in single image.4 To prevent such quality degradation in video caused by watermark embedding, more sophisticate watermarking scheme is needed. Further author information: (Send correspondence to Choong-Hoon Lee) Choong-Hoon Lee: E-mail: chlee©casaturn.kaist.ac.kr Hwang-Seok Oh: E-mail: hsoh©etri.re.kr Heung-Kyu Lee: E-mail: hklee©casaturn.kaist.ac.kr In Security and Watermarking of Multimedia Contents II, Ping Wah Wong, Edward J. DeIp, Editors, Proceedings of SPIE Vol. 3971 (2000) • 0277-786X/O0/$15.00 209
8

Adaptive Video Watermarking Using Motion Information Adaptive Vide… · Digital watermarking is newly emerging technique for solving such problems.1 Digital watermarking is a sort

Jul 17, 2020

Download

Documents

dariahiddleston
Welcome message from author
This document is posted to help you gain knowledge. Please leave a comment to let me know what you think about it! Share it to your friends and learn new things together.
Transcript
Page 1: Adaptive Video Watermarking Using Motion Information Adaptive Vide… · Digital watermarking is newly emerging technique for solving such problems.1 Digital watermarking is a sort

Adaptive Video Watermarking Using Motion InformationChoong-Hoon Leea and Hwarig-Seok Ohb and Heung-Kyu Leer'

a Department of Computer Science & Advanced Information Technology Research Center (AITrc)Korea Advanced Institute of Science and Technology (KAIST)

373-1 Gusong-Dong Yusong-Gu, Taejon, Korea

bVisual Information Research TeamVirtual Reality Research and Development Center

Electronics and Telecommunications Research Institute (ETRI)161 Kajeong-Dong Yusong-Gu, Taejon, 305-350, Korea

ABSTRACTThis paper presents an adaptive video watermarking using motion information. Because video data have one moredimension than image data, simple adoptation of image watermarking method to video data would reveal some typesof visual artifact such as flickering. In the proposed scheme, same watermark information is embedded for sameregion in each frame to decrease such visual artifact. For higher robustness and invisibility, watermarking strengthis adjusted by motion information and region complexity.

Keywords: Video Watermarking, Wavelet Transform, Motion Estimation

1. INTRODUCTIONAdvance of computing power and network technology made multimedia data processing and distribution easier.Accordingly, copy or forgery of multimedia data is easier than before. In recent, multimedia data are high value-added products and illegal copy and forgery are serious problems in multimedia industry. Digital watermarking isnewly emerging technique for solving such problems.1

Digital watermarking is a sort of information hiding technique that can be used for ownership assertion, fin-gerprinting, authentication, content labeling, usage control or content protection.2 In watermarking, user specificsignal, watermark, is embedded into host data such as image, video, audio or any other types of multimedia data ininvisible form. Embedded watermark is retrieved later for above purpose.

Digital watermarking must satisfy some requirements such as imperceptibility, robustness and reliability to beused for copyright protection.3 Imperceptibility means that embedded watermark cannot be seen or must notdegrade quality of host data. Although visible type of watermarking exists, it is not desirable because visiblewatermark degrades quality of original data and is more easily removable. Second requirement, robustness, meansthat embedded watermark must remains after several attacks for removing it. Several types of attack are possiblesuch as data compression or signal processing. The embedded watermark information can be damaged by lossy videocompression technique such as MPEG, H.261 or H.263. In addition to, several types of signal processing such asnoise addition, AD/DA conversion, blurring, histogram modification can change embedded signal. Watermark mustendure such attacks. Reliability means that embedded watermark must not be detected by unauthorized person andmust be retrieved correctly by authorized one.

Because video data have one more dimension - time dimension - than image data and successive similar framesare presented with very small time gap, visibility problem in video watermarking is more important than in imagewatermarking. Bad watermarking scheme for video would reveal visual artifact that is not seen in single image.4 Toprevent such quality degradation in video caused by watermark embedding, more sophisticate watermarking schemeis needed.

Further author information: (Send correspondence to Choong-Hoon Lee)Choong-Hoon Lee: E-mail: chlee©casaturn.kaist.ac.krHwang-Seok Oh: E-mail: hsoh©etri.re.krHeung-Kyu Lee: E-mail: hklee©casaturn.kaist.ac.kr

In Security and Watermarking of Multimedia Contents II, Ping Wah Wong, Edward J. DeIp,Editors, Proceedings of SPIE Vol. 3971 (2000) • 0277-786X/O0/$15.00 209

Page 2: Adaptive Video Watermarking Using Motion Information Adaptive Vide… · Digital watermarking is newly emerging technique for solving such problems.1 Digital watermarking is a sort

Figure 1. Relationship between successive frames and object moving

In video data, successive frames share same context very much. Same object and same background are presentedin two successive frame with a little location changes as in the Fig. 1. Therefore, the inconsistant modification ofsimilar regions such as the same object regions in successive frames can cause the visual artifact when the videosequence is presented in real-time fashion. To solve this problem, we propose a new video watermarking scheme thatembeds the same watermark information into the same regions in successive video frames. In proposed scheme, theproperty of human eyes considered also. Human eyes have different sensitivity according to the circumstance. Forexample, human cannot see the details of fast moving object and human eyes are not sensitive to the distortion inthe complex or highly textured region. In the proposed scheme, watermark is embedded with higher strength wherehuman eyes are less sensitive and with lower strength in the contrary case.

This paper is organized as follows. Section 2 describes our watermarking scheme in details. In section 3, exper-imental results of our scheme are shown. Quality of marked video and robustness against to some attacks such asMPEG compression and noise addition are presented. Section 4 describes concluding remarks and further research.

2. PROPOSED WATERMARKING METHODThis section describes our proposing algorithm in detail. In the proposed scheme, block based watermarking is usedand watermark is embedded only into selected block. Overall watermarking structure is as Fig. 2. Firstly, in the firstframe, embedding blocks are selected using some criteria. In the following frames, selected blocks in the first frameare tracked using block matching algorithm and same watermark signals with previous frames are embedded. In eachframe, selected blocks are locally transformed using wavelet transform and coefficients in the middle band frequencyare changed according to the watermark signal. Watermark information is embedded using masking function whichis calculated using complexity of the block and motion information of the blocks. After watermark embedding intransform domain, blocks are inversely transformed. Then, original blocks are replaced with watermarked blocks.

2.1. Selection of Watermark Embedding RegionIn our scheme, blocks that are appropriate for watermarking are selected in the first frame and watermark is embeddedinto the selected blocks. Selected blocks are tracked in the successive frames using motion estimation and samewatermark signal are embedded into same block. Blocks are selected by following procedure. Firstly, moving blocksare selected because the moving regions are parts of object and watermark embedding in object region is more usefulfor correct motion tracking. Secondly, the blocks that can be tracked exactly are selected among the selected blocksin the first selection. For example, homogeneous block has many similar blocks in following frame but in truthonly one block is the same block in context. Thus, in our scheme, the blocks that have much difference betweenDBD (Displaced Block Difference) with best matching block and secondly matching block are selected for watermarkembedding.

210

Frame n Frame n+1 Frame n+2

Page 3: Adaptive Video Watermarking Using Motion Information Adaptive Vide… · Digital watermarking is newly emerging technique for solving such problems.1 Digital watermarking is a sort

Original Video

The detail of block selection is as follows. Firstly, first frame is segmented into 16 x 16 blocks without overlappingand only moving blocks are selected as in Equation (1). In Equation (1), MV(B)I refers to length of motion vectorfor block B. Among the moving blocks, blocks where difference between DBDsecond and DBDfirst is larger thanthreshold (DBD) are selected as in Equation (2). DBD first and DBDsecond refer to DBD with best matchingblocks and secondly matching blocks respectively.

MV(B)I > 0DBDsecond — DBDfirst > DBD

(1)

(2)

In this paper, DBD is determined dynamically using DBDfirst . DBD j5 determined as DBD = ciDBD first.The a was set to 0.5 for experiment.

During the watermarking process, if the average DBD of blocks is higher than certain threshold then all blocksare reselected as in the first frame. In this case, we can assume that scene breaking or great camera moving occurs.

2.2. Watermark EmbeddingWatermark is embedded into wavelet transform domain because embedded signal into wavelet domain is robust tostandard video compressions such as MPEG or H.263 that are DCT based compression. Wavelet transform is appliedto each selected block as in the Fig. 3. As a result, two AC bands and one DC band are constructed. Human eyes areless sensitive to higher frequency elements. Thus, watermark embedding in higher frequency domain shows betterresults in visibility aspect. However, higher frequency coefficients are vulnerable to image processing or compression.This is not good in robustness aspect. Therefore, the watermark signal is embedded into middle band which are grayarea in Fig. 3.

Gaussian random number sequence according to N(0,1) is used for watermark signal. The same watermark signalthat is embedded into the matching block in the previous frame is used. The coefficients in (1,1), (1,2), (1,3) sub-bands in Fig. 3 are modified according to the watermark signal. To minimize visual artifact by watermarking, blockcomplexity, property of wavelet coefficient and motion information of block are considered.

Each random number is embedded into each coefficient using Equation (3),

= Xb,d, + tb,d X W, (3)

211

WatermarkSignal

WatermarkedVideo

Figure 2. Overall structure of watermark embedding scheme

Page 4: Adaptive Video Watermarking Using Motion Information Adaptive Vide… · Digital watermarking is newly emerging technique for solving such problems.1 Digital watermarking is a sort

Figure 3. \Vatermark embedding region

where 1hd refers to the ith wavelet coefficient in the direction d (horizontal, vertical or diagonal) in the block b and1b,d,i refers to the modified coefficient. The w refers to watermark signal which is Gaussian random value. The tb.drefers to watermark embedding strength according to each block b and coefficient direction d. The tb,d is composed ofmasking value according to coefficient direction, motion information and complexity of the block as in the Equation(4).

thd = )< x v6, (4)

where d refers to the watermark embedding strength according to coefficient direction d. The 0d has higher value inthe diagonal area than in horizontal or vertical area. This is because diagonal area contains less important coefficientsthan horizontal or vertical region and the change in diagonal area makes less visual artifact in the real frame.5 The

refers to the watermark embedding strength according to motion of the block. Because human cannot see thedetails of fast moving object. fast moving blocks have larger value for rnb.d. and slow moving blocks have smallervalue. That is, fast moving blocks are modified more than slowly moving blocks. Motion direction is consideredalso. \Vhen an object moves in horizontal direction then human eyes is less sensitive to the distortion in horizontaledge than in vertical edge. The contrary case is same. The horizontal area in the wavelet domain contains verticaldirection edges information and vertical area contains horizontal direction edge information. Diagonal area containsdiagonal edge information. Therefore, mb.d is adjusted according to motion direction. For horizontal area, mb,d hasthe value that is proportional to vertical element (dy) of motion vector as in Equation (5). For vertical area, mb.dhas the value that is proportional to horizontal element (dx) of motion vector. For diagonal area, mb,d has the valuethat is proportional to \/dx2 + dy2. The rnbd has minimum value mmjn, which prevents watermarking with tooweak embedding strength or zero strength. If mb,d is smaller than mmjn then is used for mb,d.

dx d = (1.2) region in Fig. 37nbd cx dy_________ d = (1.1) region in Fig. 3 (5)

+ dy2 d = (1.3) region in Fig. 3

In equation (4). Vb is adjusted by the complexity of the block b. Human eyes are less sensitive to the change or(listortion in complex region than smooth region. Thus. complexity of the block is considered for watermarking. Forcomplex region, the watermark is embedded with higher strength. In our scheme, the variance of the pixels in blockis used for complexity measure of the block. In this paper, Lb is proportional to the logarithm of pixel variance inthe block b as in Equation (6). The l'ar(Pb) refers to the variance of pixels in the block b.

212

Frame

local wavelet transform

Page 5: Adaptive Video Watermarking Using Motion Information Adaptive Vide… · Digital watermarking is newly emerging technique for solving such problems.1 Digital watermarking is a sort

Vb 0 log (Var(Pb)) (6)

2.3. Watermark DetectionEmbedded watermarks are detected using similarity function as in the Cox's scheme.6 Original video data andwatermarked video data are compared and embedded - possibly corrupted - watermark is extracted. Correlationbetween embedded watermark and extracted watermark is calculated and decision is made.

For watermark detection, blocks in the original video and corresponding blocks in the marked video are selectedas in the embedding process. Each block is locally transformed using wavelet transform. Then coefficients differencebetween two corresponding blocks is the embedded watermark.

w*=x*_x (7)

sIM(Ww*)=__ (8)

In the Equation (7) and (8) , X refers to the marked coefficient sequence in the watermarked video and X refersto the corresponding coefficient in the original video. Thus, W refers to the extracted mark. W refers to theembedded watermark. SIM(W, W*) shows higher value if W and W* are highly correlated. If SIM(W, W*) > Tthen we can decide that watermark W is embedded in the test image.

3. EXPERIMENTAL RESULTS3.1. Visibility TestOur method is concentrated on the visibility problem of watermarking. As a result, watermarked video and originalvideo are hardly distinguishable. Fig. 4 and 5 show the original video, Miss America, and watemarked version of itusing our scheme. Our scheme changes only small fraction of each frame. Moreover, changing of frames are donevery invisible form. Thus, difference between two videos can be hardly found. Each original and marked frame pairhas no difference visually and no visual artifact is shown in playing. Fig. 6 shows the watermark embedding blocksfor testing video.

3.2. Robustness to AttacksIn this experiment, we tested our scheme for robustness against MPEG compression and noise addition. Firstly,we tested the marked video without any processing. Fig. 7 shows the detecting result. The left figure shows thedetecting response to 1000 random watermarks. As shown in the figure, only for embedded watermark strongestresponse occurs. The right figure shows the response to each frame for embedded watermark. Strong responses areshown for all frames. Fig. 8 shows the detecting result after MPEG compression of watermarked video. As in theFig. 7, left figure shows the responses to 1000 random watermarks. Although the response to embedded watermarkis weaker than unprocessed video, it is much stronger than responses to all other watermark. Thus, exact detectionis possible. Right figure shows the responses for every frame. These responses are also weaker than the responses tounprocessed video data and response fluctuation is present. However, the responses are much stronger than responsesto other watermark also. Fig. 9 shows the detecting result after white noise addition. Test results are similar to theresults for MPEG compressed data. Response fluctuation in right figures is weaker than MPEG compressed data.

4. CONCLUSIONIn this paper, we proposed an adaptive video watermarking scheme that used motion information for watermarkembedding. Appropriate blocks for watermark embedding are selected using some criteria such as motion vectors andDBD. Selected blocks are the target of watermark embedding. For watermark embedding, blocks are transformedusing wavelet transform and wavelet coefficients are changed using random signal. All selected blocks are trackedframe-by-frame and same watermark is embedded into same block. This scheme enables more invisible watermarking.Experimental results show that visual artifacts are rarely shown. We also showed proposed watermarking schemeis robust against video compression and noise addition. However, some problems remain. Our algorithm is notrobust against geometrical attack or frame level attack such as frame skipping, frame averaging, resampling. Furtherresearch is needed for such problems. Moreover, watermark detection scheme without original video is needed.

213

Page 6: Adaptive Video Watermarking Using Motion Information Adaptive Vide… · Digital watermarking is newly emerging technique for solving such problems.1 Digital watermarking is a sort

ACKNOWLEDGMENTSThis work was supported by the Korea Science and Engineering Foundation (KOSEF) through the Advanced Infor-mation Technology Research Center(AITrc).

REFERENCES1. G. Voyatzis and I. Pitas, "The use of watermarks in the protection of digital multimedia products," Proc. of the

IEEE 87, pp. 1197—1207, July 1999.2. N.Memon and P. Wong, "Protecting digital media content," Communications of the ACM 41, pp. 34—43, July

1998.3. W. D. J.J.K. O Ruanaidh and F. Boland, "Watermarking digital images for copyright protection," Proc. mt.

Elect. Eng. Vis. Image Signal Processing 143, pp. 1087—1091, Aug. 1996.4. W. F. C. Busch and S. Wolthusen, "Digital watermarking: From concepts to real-time video application," IEEE

Computer Graphics and Applications , pp. 25—35, Jan./Feb. 1999.5. J. S. A.B. Watson, G.Y. Yang and J. Villasenor, "Visual thresholds for wavelet quantization error," Proc. SPIE

Human Vision and Electronic Imaging 2657, pp. 382—392, 1996.6. F. L. I.J. Cox, J. Kilian and T. Shamoon, "Secure spread spectrum watermarking for multimedia," IEEE Trans.

on Image Processing 6, pp. 1673—1687, Dec. 1997.

214

Page 7: Adaptive Video Watermarking Using Motion Information Adaptive Vide… · Digital watermarking is newly emerging technique for solving such problems.1 Digital watermarking is a sort

Figure 6. Watermark embedding block (framel. frame2, frame3)

Figure 4. Original video (framel, frame2, frame3)

Figure 5. Watermarked video (framel, frame2, frame3)

Page 8: Adaptive Video Watermarking Using Motion Information Adaptive Vide… · Digital watermarking is newly emerging technique for solving such problems.1 Digital watermarking is a sort

40 35

30 25 >

20

E

15

C')

5 0 -5

Figure

7. W

atermark

detector

response

to 1000

random

waterm

arks

(left)

and

waterm

ark

detector

response

in each

frame

(right)

16 14

12 10

E C

l) 2 0 -2 -4

20 18

16 14 12 10 E

U)

8 6 4 2 0

Figure

8. W

atermark

detector

response

to 1000

random

waterm

arks

(left)

and

waterm

ark

detector

response

in each

frame

(right)

after

MPE

G

compression

of m

arked

video.

10 8 6 >

•E

2 Cl)

o -2

-4

Figure

9.

Waterm

ark

detector

response

to 1000

random

waterm

arks

(left)

and

waterm

ark

detector

response

in each

frame

(right)

after

Gaussian

noise

addition

to w

atermarked

video.

216

40 35

30

>

25 : 20 E ö5 15

10 5

0 ¶-

0 0 N-

(0

U)

C)

C'J

- 0 0)

N- N

-

cv

0 N-

- '-

Lt)

C'J

0)

C

o o - C%

J

CJ

c)

Lt)

(0

(0

N-

U

0)

Random

Waterm

arks

'-

c'J

c)

to

co

N-

0)

0 . cJ

c) ,-

c'J

C)

U)

(0

N-

0 - CJ

C)

Fram

e

Num

ber

[mN

W

¶-

0 0)

N-

(0

L()

C!)

C'J

0 0)

N- N

-

c)

0 N-

- a

Lt)

C'J

0)

(0

C

C)

(0 '-

c %J

()

'- Lf)

CD

(0

N-

0)

Random

Waterm

arks

12

- c'J

C!)

(0

(0

N-

0)

0 '—

cJ

c) ,-

c'J

c)

(0

(0

N-

0 '-

OJ

C)

Fram

e

Num

ber

-6

16 14

12

>.

10

E (I)

6 4

2

,-

0 0)

cO

N-

(0

LO

C)

CJ

'-

0 0)

N- N

-

c)

0 N-

'-

(0

c 0)

(0

CJ

0)

(0 -

c'J

04

C')

U)

(0

(0

N-

co

cc

0)

Random

Waterm

arks

0 ,—

c'J

c)

(0

(0

N-

(0

0)

0 '-

c'J

c) - c c)

(0

N-

0(

Q

J

C)

Fram

e

Num

ber