ROBUST VIDEO WATERMARKING BASED ON 3D-DWT

USING PATCHWORK METHOD

Yadoallah Zamanidoost, Satar Mirza kuchaki

Department of Electrical Engineering, Iran University Science & Technology, Tehran, Iran

Zhinoos Razavi Hesabi, Antonio Navarro

Department of Electrical Engineering, Aveiro University, Aveiro, Portugal

Institute of Telecommunications, University Campuse, Aveiro, Portugal

Keywords: Video watermarking, Human vision system, Patchwork algorithm, 3D-DWT.

Abstract: The digital watermarks have recently been recognized as a solution for protecting the copyright of the

digital multimedia. In this paper, a new method for video watermarking with high transparency based on

3D-DWT is proposed. This algorithm is implemented on the basis of Human Vision System (HVS). By

using the patchwork methods in Discrete Wavelet Transform (DWT) domain, this algorithm is robust

against different attacks such as frame dropping, frame swapping, frame averaging, median filtering and

MPEG-2 video encoding. The experimental results show that the embedded watermark is robust and

invisible. The watermark was successfully extracted from the video after various attacks.

1 INTRODUCTION

In recent decade, information watermarking in

digital environment has attracted much interest

because of its ability in covering different aims.

Considering the high volume of video products in

recent years, special attention is focused on this type

of technology. Information watermarking is

embedment of a hidden message within another

signal. This signals named cover signal can be text,

digital image, audio or video file. Watermarking

follows different aims such as authentication,

reserving right of author, copy right and control of

data spreading, amongst others.

Three challenges exist in the field of

watermarking. In this process, data embedding must

be done in such a manner that watermarked signal

keeps its transparency. On the other hand, the signal

including hidden information may be exposed under

different processes such as filtering, geometric

transformation, adding noise, etc and after these

transformations, hidden message can be extractable.

This significant feature is called watermarking

system robustness.

With regard to application of watermarking in

video signals, high robustness in related algorithms

is a primary necessity. Another important challenge

in these systems is capacity. Capacity is by

definition the amount of information that can be put

on host signal while preserving its transparency and

robustness.

Many algorithms have been proposed for video

watermarking. First and foremost algorithm is

Hartung method (Harung, Girod, 1998). This

method is based on spread spectrum algorithms and

is executable on uncompressed and compressed

video signals. In addition to this method, various

algorithms have been considered to do watermarking

on uncompressed and compressed video signals like

data embedding in 3D-DCT domain (park, Lee, and

Moon, 2006) and 3D-DWT (Angiang, Jing, 2007)

for uncompressed video. There are other patterns in

which watermarking algorithms related to still

images are used and motion characteristics of video

have been used as parameters to modify presented

designs.

Patchwork algorithm was used for the first time

in image watermarking implemented on the basis of

comparison of two groups average, variance or other

signal properties (Yeo, Kim, 2003). This method has

been used in different transformation fields such as

DCT and DWT. (Kii, Onishi, Ozawa, 1999),

109

Zamanidoost Y., Mirza Kuchaki S., Razavi Hesabi Z. and Navarro A..

ROBUST VIDEO WATERMARKING BASED ON 3D-DWT USING PATCHWORK METHOD .

DOI: 10.5220/0003312001090113

In Proceedings of the International Conference on Imaging Theory and Applications and International Conference on Information Visualization Theory

and Applications (IMAGAPP-2011), pages 109-113

ISBN: 978-989-8425-46-1

 2011 SCITEPRESS (Science and Technology Publications, Lda.)

(2 1) (2 1) (2 1)

( , , ) ( ) ( ) ( )[ ( , , )*cos( )cos( )cos( )]

LLL

xU yV zW

FUVW CUCV CW f xyz

LLL













(1)

(OSluciak, Vargic, 2004). Patchwork algorithm in

audio watermarking shows a relatively high

performance leading to the high attack resistance

(Khademi, Akhaee, Ahadi, Amindavar, 2009). In

this paper, the patchwork algorithm is used for video

watermarking based on the 3D-DWT domain

(Anqiang, Jing, 2007).

2 WATERMARK EMBEDDING

PROCESS

Fig. 1 shows a block diagram of the watermark

embedding process in an original video signal.

Figure 1: Embedding watermark process.

2.1 Selecting of N Cube based on HVS

First, the input video signal is divided into cubes of

L*L*L. Then, according to HVS, N cubes are

chosen (Fig. 2).

Figure 2: Dividing video signal to cubes of L*L*L.

This choosing criterion is used as a secret key to

increase algorithm security. The best position for

hiding watermark in video signal is where human

eye is less sensitive. Human eye has two

fundamental weak points:

1) Human eyes can not see details on the fast

moving object

2) Human eyes are not sensitive to the distortion in

the complex or high connectivity texture region

These two features can be extracted from the video

signal by 3D-DCT transformation according to

equation (1):

where C(U),C(V),C(W) are constants. After

applying equation 1 on cubes of L*L*L, the amount

of texture and objects movement are obtained by

equation 2.































LLLKLLWVUFEE

LLKWVUFEE

KNM

KNT

***,),,(

)1*(1,),,(

(2)

N is the cube number and k is pixel number inside

that cube. According to equation (2), E

is texture

amount and E

is amount of object movement inside

the cube. A change in E

and E

results in a change

in the Secret key and also an increase in these

parameters causes the transparency to improve and

the capacity to decrease and vice versa.

2.2 3D-DWT Transformation

of Selected Cubes

There are two main methods for 3D-DWT

Transformation:

1) 2D-DWT transform other each frame of cube,

then 1D-DWT transform from each pixel row

(

Huang-yu, Ying, cheng-ke, 2004).

2) 1D-DWT transform from each pixel row of cube,

then 2D-DWT transform other each frame of cube.

In the present study, the first method has been

applied. Each frame was transformed into a 2D-

DWT with 3 levels and Haar filter. Then each pixel

row was transformed to a 1D-DWT with 3 levels

using Haar transform (Fig 3).

Figure 3: 3D-DWT with 3 levels and Haar filter.

IMAGAPP 2011 - International Conference on Imaging Theory and Applications

110

2.3 Watermark Embedding

by Patchwork Method

In this step, watermark is embedded to LLLH and

LLHL windows forming median frequency. By

embedding watermark in this window, watermarked

signal becomes robust to averaging and median

filtering attacks. First, adjacent row pairs are chosen

and variance of each of them is determined (Fig. 4).

Figure 4: Median frequency window.

As seen in Fig. 5, for watermarking LLL1 and

LLL2 are used.

Figure 5: LLL1 and LLL2 from LLL window.

Watermarking is explained by the following

algorithm:

If m == 0

LLL1=(LLL1+LLL2)/2+TH;

LLL2=(LLL1+LLL2)/2-TH;

Else

LLL1=(LLL1+LLL2)/2-TH;

LLL2=(LLL1+LLL2)/2+TH;

End

M is the watermark bit and TH is the threshold limit

of algorithm, by increasing TH, transparency of

video signal decreases while resistance against

frame dropping and frame swapping attack

increases. On the other hand decreasing TH

increases the PSNR of the video signal and

decreases the signal resistance against

aforementioned attacks.

2.4 Formation of Video Watermarked

After embedding watermark, video signal is

reconstructed. To do this, first, each pixel row was

transformed to a 1D-IDWT with 3 levels and Haar

filter, and then each frame was transformed to a 2D-

IDWT with 3 levels and Haar filter. At the end, all

cubes are combined together to form the

watermarked video.

3 WATERMARK EXTRACTING

PROCESS

Fig. 6 shows block diagram of watermark extracting

from watermarked video.

Figure 6: Watermark extracting process.

As seen in Fig. 6, first the watermarked video

signal is divided into cubes of L*L*L. Then, N

cubes are chosen according to secret key formed in

watermark embedding step. Afterwards, each cube is

transformed by a 3D-DWT with 3 levels and Haar

filter as mentioned in section 2.2. Adjacent pairs of

pixel rows of two LLLH and LLHL windows are

compared from variance point of view. The row with

higher variance is chosen. According to the

following algorithm, watermark is extracted by

comparison of LLL1 and LLL2.

If LLL1 >= LLL2

m = 0;

Else

m = 1;

End

4 EXPERIMENTAL RESULTS

In this experience 16*16*16 cubes are used and

amount of TH (embedding watermark threshold) is

40. The video sequences used in the experiments

include foreman and Stefan. The size of each frame

is 352x288. It is necessary to mention that to stand

up against MPEG-2 encoding attack; each class

includes I-Frames (Simitopoulos, Tsaftaris and

ROBUST VIDEO WATERMARKING BASED ON 3D-DWT USING PATCHWORK METHOD

111

Boulgouris, 2002). The embedded watermark

signature is a binary image (size 60x20 pixels),

shown in Fig.8 (a).

(a) Original video frame

PSNR=41.65 PSNR=40.87

(b) Watermarked video frame

Figure 7: Original and watermarked video frame.

An original frame from each video is shown in

Fig.7 (a). The corresponding watermarked frame for

each is shown in Fig.7 (b). Figure 8.(b) and (c)

shows the watermark image after extracting from

foreman and Stefan sequences.

(a) Original image (b) NC=0.9965 (c) NC=0.9945

Figure 8: Original watermark and extracted watermark.

PSNR (peak signal to noise ration) is one

commonly objective in perceptual quality measure.

PSNR is defined as:

)/255(10log10

MSEPSNR 

(3)

In this equation, mean squared error (MSE) is

formulated as:

















),(),(

jiXjiX

MSE

(4)

Where, X's are the coefficients of the original video

and

's are the coefficients of the watermarked

video. M and N stand for the height and width of the

image, respectively.

In order to evaluate the performance of the

watermarking algorithm objectively, BER (Bit Error

Rate) and NC (Normalized Cross-Correlation

Function) are introduced.

























BER

(5)





























1212

),(),(

),(

jiwjiw

WWNC

(6)

In equation (5),



w is the extracted watermark, and

w is the original watermark. In equation (6),

),( jiw denotes the original watermark image, and

),( jiw

denotes the extracted watermark image.

Without any attack, the experimental results are

shown in Table.1. The extracted watermark

signature is obviously similar to the original

watermark.

In order to test the robustness of this

watermarking scheme, a set of attack experiments

were performed.

4.1 Median Filtering Attack

The system performance against 3*3 median

filtering is shown in Table.1.

4.2 Frame Averaging Attack

Frame averaging is another significant attack to the

video watermark. In this experiment, we use the

average of current frame and its too nearest

neighbours to replace the current frame which are

formulated as:

1,...,3,2

),(),(),(

),(









yxfyxfyxf

yxf

kkk

(7)

Experimental results are shown in Table 1.

4.3 Lossy Compression

The MPEG lossy compression is one of the most

basic attacks to video watermark. Table 1 shows the

results after MPEG-2 Lossy compression.

Table 1: Performance in term of the Normalized

Correlation (NC) of the detect watermark.

Foreman Stefan

No Attack 0.9965 0.9945

Median Filtering 0.9632 0.9576

Averaging 0.9557 0.9527

MPEG2 0.9288 0.9189

IMAGAPP 2011 - International Conference on Imaging Theory and Applications

112

4.4 Frame Dropping

and Frame Swapping Attack

Frame dropping and swapping are very simple and

common attacks because of data redundancy. To

investigate the robustness of the proposed method

against these kinds of attacks, extraction of the

watermark after dropping and swapping different

rates of frames in the video clips was performed.

Experimental results are shown in Table.2.

Table 2: Results after frame dropping and frame Swapping

attack.

1/32 1/16 1/8 1/4

forman

Dropping 0.9878 0.9745 0.9588 0.9398

Swapping 0.9912 0.9813 0.9732 0.9532

Stefan

Dropping 0.9897 0.9867 0.9689 0.9478

Swapping 0.9932 0.9889 0.9773 0.9529

According to the above experimental results, the

watermarking scheme is robust against a variety of

common video processing attacks such as median

filtering; frame averaging, frame dropping, frame

swapping and MPEG2 lossy compression. The

extracted watermark is highly similar to the original

watermark.

This method is compared with non-blind method

(X. Niu, shenghe Sun, 2000) and blind method (A.

Essaouabi, E. Ibnelhaj, 2009) for Stefan sequences.

Figure 9.(a) and (b)

(a) The effect of the frame dropping

(b) The effect of the frame swapping

Figure 9: comparison of Blind, Non Blind and proposal

watermarking scheme for Stefan scene.

5 CONCLUSIONS

This paper presented semi-blind video watermarking

based on wavelet transform. By using the secret key

in embedding algorithm, this algorithm has high

security. Image that is used as a watermark is a

binary image. Procedure of this method includes

video processing, video embedding and video

detection that are described in details. This method

is a powerful method against attacks such as frame

averaging, frame swapping, frame dropping, median

filtering and MPEG2 lossy compression.

REFERENCES

F. Harung, B. Girod, 1998. watermarking of

uncompressed and compressed video signal

processing, 66 (3) pp. 283-301.

H. Park, S. H. Lee, and Y. S. Moon, 2006. Adaptive video

watermarking utilizing video characteristics in 3D-

DCT domain. lecture notes in computer science, pp.

397-406.

L. Angiang, L. Jing, 2007. A novel scheme for robust

video watermark in the 3D-DWT domain. first

international symposium on data, pp. 514-516.

Z. Huang-yu, L. Ying, W. Cheng-ke, 2004. A Blind

spatial-temporal algorithm based on 3D wavelet for

video watermarking. international conference on

multimedia and Expo (ICME) ,pp 1727-1730.

K. Yeo, H. J Kim, 2003. Generalized Patchwork

algorithm for image watermarking. international

conference on multimedia system, pp 261-265.

H. Kii, J. Onishi, S. Ozawa, 1999. The Digital

Watermarking method by using both patchwork and

DCT. IEEE international conference on multimedia

system ,pp 895-899.

OSluciak, R. Vargic, 2004. An Audio Watermarking

method based on wavelet patchwork algorithm. IEEE

international conference on multimedia.

N. Khademi, M. A. Akhaee, S. M. Ahadi, H.

Amindavar,August 2009. Robust Multiplicative

Patchwork method for audio watermarking," IEEE

Transaction on audio, speech and language processing,

Vol 17, NO.6,pp 1133-1141.

L. Anqiang, L. Jing, 2007. A Novel Scheme for robust

video watermarking in 3D-DWT Domain. First

International symposium on data, privacy and E-

Commerce IEEE ,pp 514-516.

A. Essaouabi, E. Ibnelhaj, 2009. A 3D wavelet-Based

method for digital video watermarking. IEEE

international conference on multimedia, pp 429-434.

D. Simitopoulos, S. Tsaftaris and N. Boulgouris, Aug.

2002.compressed-domain video watermarking of

MPEG stream. in proc. IEEE int. conf. on multimedia

and Expo, pp. 569-572.

ROBUST VIDEO WATERMARKING BASED ON 3D-DWT USING PATCHWORK METHOD

113