A Visual Cryptography Based Watermark Technology for Individual and Group

Images
Azzam SLEIT (Previously, Azzam IBRAHIM)
King Abdullah II School for Information Technology, University of Jordan,
Amman, Jordan

and

Adel ABUSITTA
King Abdullah II School for Information Technology, University of Jordan,
Amman, Jordan

ABSTRACT Copyright Office to obtain proof that he is the rightful owner. If

he did not register the image, then he should at least be able to
The ease by which digital information can be duplicated and show the film negative. However, with the rapid acceptance of
distributed has led to the need for effective copyright protection digital photography, there might never have been a negative. In
tools. Various techniques including watermarking have been theory, it is possible for the owner to use a watermark
introduced in attempt to address these growing concerns. Most embedded in the image to prove that he/she owns it. Digital
watermarking algorithms call for a piece of information to be watermarking describes methods and technologies that allow
hidden directly in media content, in such a way that it is hiding information, for example a number or text, in digital
imperceptible to a human observer, but detectable by a media, such as images, video and audio. The embedding takes
computer. This paper presents an improved cryptographic place by manipulating the contents of the digital data. The
watermark method based on Hwang and Naor-Shamir [1, 2] modifications of the pixel values have to be invisible.
approaches. The technique does not require that the watermark Furthermore, the watermark has to be robust or fragile,
pattern to be embedded in to the original digital image. depending on the application. With robustness we refer to the
Verification information is generated and used to validate the capability of the watermark to resist to manipulations of the
ownership of the image or a group of images. The watermark media, such as loss compression, scaling, and cropping, just to
pattern can be any bitmap image. Experimental results show enumerate some. Figure 1 shows the standard digital watermark
that the proposed method can recover the watermark pattern schema. An image watermark method must satisfy the following
from the marked image (or group of images) even if major two properties:
changes are reflected on the original digital image or any
member of the image group such as rotation, scaling and • Transparency: the embedded watermark pattern does not
distortion. visually spoil the original image fidelity and should be
perceptually invisible.
Keywords: Image Watermark, Group of Images, Pattern, Visual
Cryptography, Rotation, Scaling, Copyright. • Robustness: the watermark pattern is hard to detect and
remove illegally.

1. INTRODUCTION This paper proposes a solution for watermarking digital images.

Marking images will be conducted without embedding patterns
Digital watermarking is the practice of hiding a message in an into images. This leaves marked images unchanged with sizes
image, audio, video or other digital media element. Since the exactly equal to the original ones. The new solution will retrieve
late 1990s, there has been a massive production for digital the watermark pattern from rotated and / or resized image as is
watermarking algorithms [3–17]. The sudden increase is mostly without any noise in the watermark pattern. As will be
attributed to the increase in concern over copyright protection of demonstrated in the coming sections, retrieving the watermark
content. Because new devices store content in digital form, there pattern from the marked image will be reasonably fast. The
is no degradation in quality of data after a copy is made. next section of this paper briefly outlines existing watermarking
methods. Section 3 explains the proposed watermark method.
Traditionally, the owner of an image registers the image with Then, we discuss experimental results to demonstrate the merits
the Copyright Office by sending a copy to them. The Copyright of the proposed method.
Office archives the image, together with information about the
rightful owner. When dispute occurs, the real owner contacts the

24 SYSTEMICS, CYBERNETICS AND INFORMATICS VOLUME 5 - NUMBER 2

 Watermark

Watermarked
Data Digital Data
Watermarking

Secret / public key

.

Figure 1: Digital watermark schema

2. WATERMARKING AND INVARIANT IMAGE SIZE

Block1
Pixel Block 1 Block 2 superimposes on
This section focuses on the works the watermarking methods
Block 2
that do not directly embed watermarks into the original digital
images. Instead, verification information is generated which is Black (1, 0) (0, 1) (1, 1)
used to verify the ownership of the image. The visual
cryptography is a new concept defined by Naor-Shamir [2]. It is Black (0, 1) (1, 0) (1, 1)
an extended type of (t, n)-threshold scheme which is also named
the (t, n)-visual threshold scheme. In [2], the shadow of each White (1, 0) (1, 0) (1, 0)
participant is a transparency showing random dots. The shared
secret is an image composed of black and white pixels. Any t White (0, 1) (0, 1) (0, 1)
out of these n shadows can make the shared secret recognized
through the human visual system when they are stacked
together. Any t-1 (or less) shadows stacked together can Table 1: A (2, 2)- visual threshold scheme; Note: bit "1" denotes
generate no knowledge about the shared secret. The image black and bit "0" denotes white.
stored in the computer system can be considered a composition
d
of pixels. Let each pixel be stored in d bits. Then, a 2 gray-
leveled image can be shown by using a set of pixels. It only uses Hwang's method [1] is based on the simple (2, 2) visual
one bit to express each pixel. Table 1 illustrates a simple (2, 2)- threshold scheme presented by Naor-Shamir [2]. According to
threshold scheme based on Naor-Shamir’s idea [2]. It also Hwang, the owner should select h*n black/white image as
specifies the algorithm to encode each pixel in the shared image. his/her watermark pattern P and a key S which must be kept
This algorithm is applied to each pixel in the shared image in securely. Then, verification information V is generated from the
order to generate the corresponding subpixels in its original k*1 image M and the watermark pattern P using the key
corresponding two shadows. Each pixel P in the shared image is S; as follows:
divided into two subpixels in each of these two shadows. If P is
black, then the dealer randomly selects one of the first two rows 1. Use the secret key S as the seed to generate h*n different
in Table 1. If P is white, then the dealer randomly selects one of random numbers over the interval [0, k*1]. (Ri represents the i-
the last two rows in Table 1. Then, the dealer puts two two- th random number).
subpixel blocks from Columns 2 and 3 to the corresponding
positions in shadows 1 and 2, respectively. Let’s consider the 2. Assign the i-th pair (Vi1, Vi2) of the verification information
result when these two shadows are stacked together. For each V based on the following table:
pixel P in the shared image, if P is black, then it generates a
block with two black subpixels when these two shadows are
The color of The left most Assign the i-th
stacked together. If P is white, then it generates a block with
the with pixel bit of the Ri-th pair (Vi1 ,Vi2), of
one black subpixel and one white subpixel when these two
in watermark pixel of image verification
shadows are stacked together. The result is a collection of two
pattern is M is information V to
black/white subpixels, which are printed in close proximity to
be
each other so that the human visual system averages their
individual black/white contributions. Through the human visual Black "1" (0,1)
system, the block with two black subpixels will be recognized Black "0" (1,0)
as a black dot while the block with one black subpixel and one
white subpixel will be recognized as a white dot. Obviously, we White "1" (1,0)
can readily recognize if an image is the shared image with our White "0" (0,1)
visual system when these two shadows are stacked together.

Table 2: Rules to assign values of verification information.

SYSTEMICS, CYBERNETICS AND INFORMATICS VOLUME 5 - NUMBER 2 25

 3. Assemble all the (Vi1, Vi2) pairs to construct the verification 2. Sort in-place all the pixel values of the original image
information V. This verification information must be kept by Mi in ascending order. For example if the original
neutral organization. image Mi has the following gray values (12 1 6 1 2 7
18 12 6 9), then Mi will receive the following values
When the owner want to claim the ownership of an image F as a (1 1 2 6 6 7 9 12 12 18). The size of Mi is equal to that
copy of the original image M, he/she provides the secret key S, of MKi.
and the watermark pattern is restored using the image F and 3. Generate Mi' from Mi. Each pixel in Mi' represents 8
verification information V as follows: pixels of Mi. Bit 0 in pixel 0 of Mi' represents the most
significant bit from pixel 0 of Mi, bit 1 in pixel 0 of
1. Use S as the seed to generate h*n different random numbers Mi' represents the most significant bit from pixel 1 of
over the interval [0, k*1]. (Ri represents the i-th random Mi, and so on. For example, if the first 8 pixels of Mi
number). are: 11111101 11110111 11101000 11101000
10100100 01100001 01001101 00010010, then bit
2. Assign the color of the i-th pixel of the watermark pattern P' 0 of Mi' will receive1, and bit 1 of Mi' will be 1 as
based on the image F as follows: well and so on. The first pixel in Mi' will be
11111000.
a. Get the left-most bit, b, of the Ri-th pixel of 4. Divide Mi' into g groups of pixels, in which g =
image F, and if b is 1 then, assign fi = (1,0); MKi'/PK rounded to the largest integer. MKi'
otherwise assign fi = (0,1). represents the length of the image Mi'. The grouping is
b. If fi is equal to i-th pair of V then assigns the performed without any manipulation.
color of the i-th pixel of P' to be white; 5. Create array Xi which has the same length as P (i.e.
otherwise, assign it to be black. PK). Xi is assigned the exclusive-or of all the pixels in
each group of the groups generated from Mi'. If there
3. If P' can be recognized as P through the human, the neutral is still space in Xi while Mi' pixels are finished (this
organization shall adjudge that the image F is a copy of M. happened when PK > MKi') then Xi will repeat itself
until all PK pixels in Xi have been filled.
The previous method has deficiencies in that it does not give 6. Steps from 1 to 5 are repeated n times with different
consideration to the rotation, and scaling of images. Moreover, value of i where i = {1, 2,…n}
the watermark pattern must be a black and white image which 7. Finally, After generating Xi for each Mi, where i = 1,
does not offer flexibility to the owner [17]. The following 2, ….n, verification information V is generated
section provides solutions for the above mentioned deficiencies. according to the following segment:
For i = 1 to PK
V[i] = (P[i] XOR Se[i] XOR X1[i] XOR
X2[i] XOR ….. XOR Xn[i])
3. PROPOSED WATERMARK METHOD
The verification information V is given to a neutral
This paper offers a solution for the previously mentioned organization. In case the owner needs to claim ownership of
shortcomings. Let Mg represent a group of images to be marked. some data F, and F seems to be one of the group's images Mi
This group contain The following images {M1, M2, … Mn }. ,where i = {1, 2, …., n}. A watermark pattern P' is generated
The proposed method generates verification information V from from the verification information V and key Se according to the
a group image Mg as shown bellow. following steps:

First, the owner must select a secret key S of length 8...128 byte. 1. Replace F with its similar image in Mg and Repeat
The key must be a multiple of 8 bytes and will be expanded to steps from 1 to 5 of the embedding process (use VK
PK, where PK is the length of the watermark pattern as we will which is the length of the verification information
see later. The expanding algorithm is as follows: instead of PK and generation information of the key
as VK instead of PK). Then create array X from F to
1. Let S[i] be the i-th byte of the original key, Se[i] the i- be used along with the extended key Se.
th byte of the expanded key, and K the length of S. 2. Generate the watermark pattern P' according to the
2. Load the key S without change into the first K bytes following:
of Se. For i = 1 to VK
3. For i = K+1 to PK P'[i] = (V[i] XOR Se[i] XOR X1[i] XOR X2[i]
Se[i] = (Se[i-8] XOR Se[i-4] XOR XOR ….. XOR Xn[i])
Se[i-3] XOR Se[i-1]) <<< 3 3. If P' is equal to the original watermark pattern P, then
F is a copy of Mi (in case F is a copy of Mi with minor
Then, the owner should select a watermark pattern P which can changes then the changes will appear on P' as some
be any significant bitmap image. Consequently, the owner can distortions).
typify a group image Mg using the watermark pattern and the
expanded key Se according to the following steps for each Mi In summary, the proposed method calls for sorting the input
where i = 1, 2, ….n. group images. The algorithm takes the sorted image as input to
generate verification information instead of the original image.
1. Let PK be the length of the watermark pattern P, MKi Also, it is obvious that the watermark pattern is restored
the length of the image Mi, and MKi' the length of the successfully if no change occurs to any member in the group.
image Mi' which will be equal to MKi/8 as we will see This is because the embedding and verification processes are the
below.

26 SYSTEMICS, CYBERNETICS AND INFORMATICS VOLUME 5 - NUMBER 2

same except for the final step (i.e. the exclusive-or operation).
Mathematically,
4. EXPERIMENTAL RESULTS
(X XOR Y) XOR Y = X
The proposed algorithms are studied on 24-bitmap bean and
Then, it can be concluded that P'[i] = (V[i] XOR X[i]) XOR moon groups of images in Figure 2. Two watermark patterns are
S[i]) = (( P[i] XOR X[i]) XOR Se[i] XOR X'[i] ) XOR Se[i] = used; namely: Cheng and Tiger. Several types of distortions are
P[i] only if X[i] is the same as X'[i]. Also it can be seen that if experimentally investigated including rotation and scaling.
there is change in X[i] which represents a change in the image, Table 3 demonstrates that the watermark pattern can be
then this change will reflect on the restored watermark pattern recognized after rotating and/or scaling the images. Also, the
P'. Therefore, if the image Mi undergoes minor changes to proposed method works for significant patterns on groups of
become image F, the watermark pattern will still be recognized images unlike most of the existing techniques which work only
but it will have some distortions. on black and white patterns for individual images.

Figure 2: Lina and moon image groups

SYSTEMICS, CYBERNETICS AND INFORMATICS VOLUME 5 - NUMBER 2 27

 Pattern Used (Cheng) Problem

Image Scaling and Rotation

Black and white pattern

Original Image Scaled and Rotated Hwang Result Our Result

Image

28 SYSTEMICS, CYBERNETICS AND INFORMATICS VOLUME 5 - NUMBER 2

 Pattern Used (Tiger) Problem

Image Scaling and Rotation

Colored pattern

Original Image Scaled and Rotated Hwang Result Our Result

Image

Hwang method works

only on black and white
Pattern

Hwang method works

only on black and white
Pattern

Hwang method works

only on black and white
Pattern

Hwang method works

only on black and white
Pattern

SYSTEMICS, CYBERNETICS AND INFORMATICS VOLUME 5 - NUMBER 2 29

 Pattern Used (Cheng) Problem

Image Scaling and Distortion

Black and white pattern

Original Image Scaled and Hwang Result Our Result

Distorted Image

30 SYSTEMICS, CYBERNETICS AND INFORMATICS VOLUME 5 - NUMBER 2

 Pattern Used (Tiger) Problem

Image Scaling and Distortion

Colored Pattern

Original Image Scaled and distort Hwang Result Our Result

image

Hwang method works only

on black and white Patterns

Hwang method works only

on black and white Patterns

Hwang method works only

on black and white Pattern

Hwang method works only

on black and white Patterns

Table 3: Experimental results outlining the superiority of the proposed algorithms.

SYSTEMICS, CYBERNETICS AND INFORMATICS VOLUME 5 - NUMBER 2 31

 5. CONCLUTION SPIE/IS&T Conference on Security and Watermarking
of Multimedia Contents, Vol. 3657, San Jose, CA, Jan.
This paper presented a digital image copyright protection 1999, pp. 214–225.
method which does not require that the watermark pattern to be [7] C. I. Podilchuk and E. J. Delp, “Digital watermarking:
embedded in to the original image which leaves the marked Algorithms and applications”, IEEE Signal Processing
image equal to the original image. The watermark pattern is Magazine, Vol. 18, No. 4, Jul. 2001, pp. 33–46.
retrieved as is for rotated and resized images or group of [8] I. Cox, M. Miller, and J. Bloom, Digital Watermarking,
images. The proposed technique is capable of protecting a group Morgan Kaufmann, 2001.
of images and demonstrated superiority when compared to other [9] W. Stallings, Cryptography and Network Security, Third
techniques. The watermark pattern cannot be retrieved from the Edition, Prentice Hall, 2003.
marked image unless the key is given. Also the key cannot be [10] C. Rafael and E. Richard, Digital Image Processing,
retrieved even if all the algorithm components are known. Second Edition, Prentice Hall, 2002.
Security of the method can be controlled by the length of the [11] M. Swanson, B. Zhu and A. H. Tewfik, “Transparent
given key. For very long keys, the method is secure while for Robust Image Watermarking”, The Proceedings of IEEE
shorter keys the method obviously becomes less secure. International Conference on Image Processing, Vol. 3,
1996, pp. 211-214.
[12] G. Voyatzis and I. Pitas, “Applications of Total
6. REFERENCES Automorphisms in Image Watermarking”, The
Proceedings of IEEE International Conference on
[1] R. J. Hwang, "A Digital Copyright Protection Scheme Image Processing, Vol. 2, 1996, pp. 237-240.
Based on Visual Cryptography", Tamkang Journal of [13] Xia X. G., Boncelet C. G. and Arce G. R., “A
science and Engineering, Vol. 3, No. 3, 2000, pp. 97-106. Multiresolution Watermark for Digital Images”, The
[2] N. Naor and A. Shamir, "Visual Cryptography", Advances Proceedings of IEEE International Conference on
in Cryptology: Eurocrypt'94, Springer-Verlag, Berlin, Image Processing, Vol. 1, 1997, pp. 548-551.
1995, pp.1-12. [14] W. Bender, D. Gruhl, N. Morimotoand, “Techniques for
[3] R. B. Wolfgang, C. I. Podilchuk, and E. J. Delp, “Perceptual Data Hiding”, IBM System Journal, Vol. 35, No. 3, 1996,
watermarks for digital images and video”, Proceedings of pp. 313-336.
the IEEE, Vol. 87, No. 7, Jul. 1999, pp. 1108–1126. [15] K. Matsui, J. Ohnishi, and Y. Nakamura, “Embedding a
[4] E. T. Lin, A. M. Eskicioglu, R. L. Lagendijk, and E. J. Signature to Pictures under Wavelet Transform”, IEICE
Delp, “Advances in digital video content protection”, Transactions, Vol. J79-D-II, No. 6, 1996, pp. 1017-1024.
Proceedings of the IEEE, Vol. 93, No. 1, January 2005, [16] Hwang M. S., Chang C. C., and Hwang K. F., “A
pp. 171–183. Watermarking Technique Based on One-way Hash
[5] I. Cox, J. Kilian, T. Leighton, and T. Shamoon, “Secure Functions”, IEEE Transactions on Consumer
spread spectrum watermarking for multimedia”, IEEE Electronics, Vol. 45, No. 2, 1999, pp. 286-294.
Transactions on Image Processing, Vol. 6, No. 12, 1997, [17] A. Sleit, and A. Abusitta, “A Watermark Technology
pp. 1673–1687. Based on Visual Cryptography”, in the Proceedings of the
[6] J. Fridrich and M. Goljan, “Comparing robustness of 10th World Multi-Conference on Systemics,
watermarking techniques”, Proceedings of the Cybernetics and Informatics, Vol. 1, 2006, pp. 227-238.

32 SYSTEMICS, CYBERNETICS AND INFORMATICS VOLUME 5 - NUMBER 2