Professional Documents
Culture Documents
Abstract
Good image quality and high hiding capacity are two basic requirements of information
hiding systems. Technically, it is very challenging to achieve these two factors
simultaneously. The purpose of obtaining either high hiding capacity or good image quality
are various from application to application. Inspired from the wet paper codes proposed by
Fridrich et al., we propose an information hiding scheme for grayscale images. The proposed
scheme first uses a secret key to randomly select a subset of pixels from a cover image as dry
pixels. Next, the toral automorphism is applied to the cover image to maximize the number of
dry pixel pairs. Then, each secret digit in the base-9 numeral system is embedded into one dry
pixel pair. The experimental results show that the proposed scheme can achieve good image
quality (i.e., PSNR > 46 dB) and flexible hiding capacity. In addition, unauthorized users
without knowing the secret key and the secret parameters used for the toral automorphism
can not extract the embedded message.
1. Introduction
The advances in network technologies and digital devices facilitate digital data
distributions. However, distributing digital data over public networks such as the Internet is
not really safe due to copy violation, counterfeiting, forgery, and fraud. Therefore, the
protection of digital data, especially for confidential data, is in high demand. Traditionally,
secret data can be protected by cryptographic methods such as DES [1] or RSA [2]. The
drawback of cryptography is that cryptography can protect the secret data in transit, but once
they have been decrypted, the content of the secret data has no further protection [3].
Alternatively, confidential data can be protected by using information hiding techniques. An
information hiding system embeds secret data into a cover object (e.g., an image, audio,
video, or written text) to obtain an embedded object (also called a watermarked object in
1
International Journal of Smart Home
Vol.3, No.2, April, 2009
2
International Journal of Smart Home
Vol.3, No.2, April, 2009
embedded into one group of n cover pixels by increasing or decreasing only one cover pixel
in the group by 1. Let us denote the grayscale values of pixels in an n-pixel group as an n-
dimensional vector (p1, p2, ..., pn) and calculate the value y of the extraction function G as a
weighted sum modulo (2n + 1). That is, y = G(p1, p2, ..., pn) = 1≤i≤n (pi × i) mod (2n + 1).
According to the secret digit d and the value y of the extraction function G, the embedding
rules can be divided into three cases as follows. Case 1: if d = y, then d is embedded by no
modification on the pixel group. Case 2: if d ≠ y, calculate s = (d – y) mod (2n + 1). Next, if s
≤ n, then increase the value of ps by 1. Case 3: if d ≠ y, calculate s = (d – y) mod (2n + 1).
Then, if s > n, then decrease the value of p2n+1–s by 1. To improve the hiding capacity of the
EMD method, Chang et al. [7] proposed an information hiding scheme by using a Sudoku
solution. The review of Chang et al.’s scheme is presented in Subsection 2.1.
The aforementioned schemes conceal secret data in raster scan manner (i.e., from left to
right and from top to bottom). As a result, these schemes may be vulnerable to steganalytic
attacks because the embedding positions are publicly known. To enhance the steganographic
security of information hiding systems, Fridrich et al. proposed a novel steganographic
scheme called writing on wet paper [8]. Fridrich et al.’s scheme is reviewed in Subsection
2.2.
Thus, from the aforementioned considerations, aimed at improving the steganographic
security of Chang et al.’s scheme, we propose a steganographic scheme with good visual
quality (i.e., PSNR > 46 dB) and high hiding capacity by using the idea from the wet paper
codes. First, the proposed scheme utilizes a secret key to randomly select a subset of pixels
from a cover image as dry pixels. Next, the toral automorphism is performed on the cover
image so that the number of dry pixel pairs is maximized. Then, each secret digit in the base-
9 numeral system (also called a secret digit for short) is hidden into one dry pixel pair at a
time. The details of the proposed method are described in Section 3.
The rest of the paper is organized as follows. The reviews of Chang et al.’s scheme, wet
paper codes, and the toral automorphism are presented in Subsections 2.1, 2.2, and 2.3,
respectively. Section 3 details the proposed scheme. The experimental results and discussions
are presented in Section 4. Finally, some conclusions are made in Section 5.
2. Related works
2.1. Chang et al.’s scheme
Inspired from Zhang and Wang’s scheme, recently, Chang et al. [7] proposed an
information hiding scheme by using a Sudoku solution to enhance the hiding capacity of the
EMD method. Sudoku is a logic-based number placement puzzle [9]. Sudoku is a 99 matrix
which contains nine 33 sub-blocks, each contains different digits from 1 to 9. In addition,
each row and each column of a Sudoku grid also contain different digits from 1 to 9. An
example of a Sudoku solution is shown in Figure 1.
According to Sudoku properties, Chang et al.’s scheme converts a binary secret message
into secret digits in the base-9 numeral system and then modifies the values of cover pixel
pairs for concealing the secret digits. Thus, every digit in the Sudoku grid is decreased by 1 so
that the Sudoku grid contains digits from 0 to 8.
Next, the modified Sudoku solution is used to generate the reference matrix R, as shown in
Figure 2, which is used for the embedding and extracting processes.
3
International Journal of Smart Home
Vol.3, No.2, April, 2009
Each secret digit d is embedded into a cover pixel pair (pi, pi+1) as follows. Firstly, the
cover pixel pair (pi, pi+1) is located onto the reference matrix R at the row pi and the column
pi+1. Secondly, three sets of candidate elements are identified by C1 = {R(pi, pi+1 – 4), R(pi, pi+1
– 3), R(pi, pi+1 – 2), R(pi, pi+1 – 1), R(pi, pi+1), R(pi, pi+1 + 1), R(pi, pi+1 + 2), R(pi, pi+1 + 3), R(pi,
4
International Journal of Smart Home
Vol.3, No.2, April, 2009
pi+1 + 4)}, C2 = {R(pi – 4, pi+1), R(pi – 3, pi+1), R(pi – 2, pi+1), R(pi – 1, pi+1), R(pi, pi+1), R(pi +
1, pi+1), R(pi + 2, pi+1), R(pi + 3, pi+1), R(pi + 4, pi+1)}, and C3 = {R(xb, yb), R(xb, yb + 1), R(xb, yb
+ 2), R(xb + 1, yb), R(xb + 1, yb + 1), R(xb + 1, yb + 2), R(xb + 2, yb), R(xb + 2, yb + 1), R(xb + 2,
yb + 2)}, where xb = pi / 3 × 3 and yb = pi+1 / 3 × 3. Next, search from C1, C2, and C3 to
find out three candidate elements R(xh, yh), R(xv, yv), and R(xw, vw), respectively, such that
R(xh, yh) = R(xv, yv) = R(xw, vw) = d. Then, among the found candidate elements R(xh, yh), R(xv,
yv), and R(xw, vw), select the candidate element R(xf, yf) with a minimum distortion. That is,
R(xf, yf) = minj{h, v, w}{|pi – xj| + |pi+1 – yj|}. Finally, the stego pixel pair is obtained by (pi’,
pi+1’) = (xf, yf).
At the receiving end, the embedded secret digits can be exactly extracted from the
received stego image V sized H×W by d = R(pi , pi+1 ). The other details of Chang et al.’s
method can be found in the original article [7].
5
International Journal of Smart Home
Vol.3, No.2, April, 2009
Inspired from the idea of the WPC scheme, we propose a steganographic scheme with
good image quality (i.e., PSNR > 46 dB), flexible hiding capacity, and improved
steganographic security. The details of the proposed scheme are described below.
The proposed method works as follows. Based on the concept of the wet paper codes
(WPC) proposed by Fridrich et al. [8], firstly, we randomly select K dry pixels (i.e.,
changeable pixels) of the cover image X using a pseudo-random number generator (PRNG)
with a secret seed s, where 0 ≤ K ≤ H×W. The H×W – K remaining pixels are regarded as wet
pixels (i.e., unchangeable pixels). The secret seed s and K are considered as secret keys and
pre-shared with expected receivers. A temporary matrix T sized H×W is used to record the dry
pixels in the cover image X. Actually, T is a bit map in which T[i][j] = 1 if X[i][j] is a dry
pixel; otherwise, T[i][j] = 0, where 0 ≤ i ≤ H – 1 and 0 ≤ j ≤ W – 1. Secondly, pixels in the
cover image X are paired using some paring rule (e.g., in raster scan order). A pixel pair (pi,
pi+1) is defined as a restricted pair if pi and pi+1 are wet pixels. Otherwise, the pixel pair (pi,
pi+1) is defined as a non-restricted pair. The proposed method embeds secret digits into non-
restricted pairs only. That is, the restricted pairs are left unchanged during the embedding
process. Thirdly, the toral automorphism [10] is used to relocate pixels in the cover image X
such that the number of non-restricted pixel pairs in the cover image X is maximized. That is,
the number of restricted pixel pairs in X is minimized. Let the number of iterations that the
number of non-restricted pixel pairs in the cover image X is maximized and the recurrence
6
International Journal of Smart Home
Vol.3, No.2, April, 2009
time be u and t, respectively. The resulting image is denoted as Y. It is noted that when we
apply the toral automorphism u iterations to the cover image X, we also apply the toral
automorphism u iterations to the temporary matrix T to maintain the information of the
restricted and non-restricted pixel pairs. Fourthly, analogous to Chang et al.’s scheme, the
binary secret message S is converted into secret digits in the base-9 numeral system (also
called secret digits for short). These secret digits are then embedded into the image Y by
adopting Chang et al.’s method [7] to obtain the embedded image U. That is, Chang et al.’s
method is adopted to embed the secret digits into the non-restricted pixel pairs only. Finally,
the toral automorphism is applied t – u times to the embedded image U to get the final stego
image V. When the embedding process has been done, the sender sends the stego image V and
secret parameters K, s, and k to expected recipients, where k is the parameter of the toral
automorphism.
The proposed extracting process works as follows. First, the recipient uses the same PRNG
with seed s to randomly choose K dry pixels in the received stego image V. Second, the toral
automorphism is executed u iterations to the stego image V and the temporary matrix T to get
the embedded image U. Then, based on the temporary matrix T, the extracting process of
Chang et al.’s scheme is adopted to extract the embedded secret digits from the non-restricted
pixel pairs of the embedded image U. Finally, the extracted secret digits are concatenated and
converted back to the binary form to obtain the original secret message.
7
International Journal of Smart Home
Vol.3, No.2, April, 2009
the used toral automorphism and the secret seed of the used PRNG were set to be k = 2 and s
= 15, respectively.
Table 1. Hiding capacities and PSNR values of the proposed method for test
images with various values of K
Images Lena Baboon F16 Barbara
K PSNR C PSNR C PSNR C PSNR C
10% 53.83 0.29 53.84 0.29 53.89 0.29 53.89 0.29
20% 51.05 0.54 51.10 0.54 51.06 0.54 51.08 0.54
30% 49.57 0.77 49.56 0.77 49.55 0.77 49.58 0.77
40% 48.57 0.96 48.58 0.96 48.61 0.96 48.58 0.96
50% 47.89 1.13 47.91 1.13 47.91 1.13 47.90 1.13
60% 47.45 1.26 47.42 1.26 47.42 1.26 47.41 1.26
70% 47.09 1.37 47.08 1.37 47.08 1.37 47.05 1.37
80% 46.82 1.44 46.83 1.44 46.84 1.44 46.83 1.44
90% 46.70 1.49 46.68 1.49 46.71 1.49 46.70 1.49
100% 46.65 1.50 46.67 1.50 46.61 1.50 46.65 1.50
To compare the performance of two related works and the proposed method, the
performance results of Zhang and Wang’s and Chang et al.’s methods in terms of visual
quality and hiding capacity are shown in Table 2.
Tables 1 and 2 show that the proposed method becomes Chang et al.’s method when K =
100%. In other words, Chang et al.’s scheme is a special case of the proposed scheme.
In addition, the hiding capacity of the proposed method is less than that of Zhang and
Wang’s method when K ≤ 40%. However, the hiding capacity of the proposed method is
greater than that of Zhang and Wang’s method when 50% ≤ K ≤ 100%. It is very natural that
the PSNR value of the EMD method is higher than that of the proposed method. This is
8
International Journal of Smart Home
Vol.3, No.2, April, 2009
because Zhang and Wang’s method embeds one secret digit in the base-5 numeral system into
one cover pixel pair whereas the proposed method conceals one secret digit in the base-9
numeral system into one cover pixel pair.
Table 2. The performance results of Zhang and Wang’s and Chang et al.’s
methods
Methods Zhang and Wang Chang et al.
Images PSNR C PSNR C
Lena 52.09 1 46.65 1.5
Baboon 52.10 1 46.67 1.5
F16 52.12 1 46.61 1.5
Barbara 52.12 1 46.65 1.5
The purpose of using the toral automorphism in our proposed scheme is to maximize the
number of non-restricted pixel pairs in the cover image X. The gains in terms of dry pixel
pairs thanks to the use of the toral automorphism for test images with various values of K are
shown in Table 3. It is clear that the gain in secret bits, which can be embedded more in the
cover image X, via the use of the toral automorphism is treble the number of gained dry pixel
pairs.
Table 3. The gained dry pixel pairs using toral automorphism for test images with
various values of K
Factors Initial dry Maximized dry Gained dry Gained secret
K pixel pairs pixels pairs pixel pairs bits
10% 24907 24996 89 267
20% 47203 47432 229 687
30% 66847 67038 191 573
40% 83934 84117 183 549
50% 98304 98559 255 765
60% 110068 110379 311 933
70% 119207 119475 268 804
80% 125751 125978 227 681
90% 129729 129880 151 453
With the uses of the random selection of dry pixels from the cover image X and the toral
automorphism, the proposed method can improve the steganographic security of Zhang and
Wang’s and Chang et al.’s methods. That is, the unintended recipients can not extract the
embedded secret message unless they know the secret keys K, s, and k which are used for
selecting dry pixel pairs, PRNG, and the toral automorphism.
5. Conclusions
In this paper, we propose a steganographic scheme for grayscale images in spatial domain.
The proposed scheme uses the idea from the wet paper codes to randomly select a subset of
pixels from a cover image as dry pixels for the embedding process. In addition, the toral
automorphism is applied to the cover image to maximize the number of dry pixel pairs. The
proposed scheme can achieve good image quality and high hiding capacity. That is, the PSNR
value is greater than 46 dB for all test images and the hiding capacity can be achieved up to
1.5 bpp. Furthermore, the embedded message can not be extracted if the secret parameters K,
s, and k are unknown.
9
International Journal of Smart Home
Vol.3, No.2, April, 2009
References
[1] R. M. Davis, “The Data Encryption Standard in Perspective,” IEEE Communications Magazine, Vol. 16, No.
6, 1978, pp. 5-9.
[2] R. Rivest, A. Shamir, and L. Adleman, “A Method for Obtaining Digital Signatures and Public-key
Cryptosystems,” Communications of the ACM, Vol. 21, No. 2, 1978, pp. 120-126.
[3] I. J. Cox, M. L. Miller, J. A. Bloom, J. Fridrich, and T. Kalker, “Digital Watermarking and Steganography,”
Morgan Kauffman, 2007, ISBN: 978-0-12-372585-1.
[4] F. A. P. Petitcolas, R. J. Anderson, and M. G. Kuhn, “Information Hiding-A Survey,” Proceedings of the
IEEE, Vol. 87, No. 7, 1999, pp. 1062-1078.
[5] J. Mielikainen, “LSB Matching Revisited,” IEEE Signal Processing Letters, Vol. 13, No. 5, May 2006, pp.
285-287.
[6] X. Zhang and S. Wang, “Efficient Steganographic Embedding by Exploiting Modification Direction,” IEEE
Communications Letters, Vo. 10, No. 11, Nov. 2006, pp. 1-3.
[7] C. C. Chang, Y. C. Chou, and T. D. Kieu, “An Information Hiding Scheme Using Sudoku,” Proceedings of
The Third International Conference on Innovative Computing, Information and Control (ICICIC2008), Dalian,
China, Jun. 2008. pp. 17-21.
[8] J. Fridrich, M. Goljan, P. Lisonek, and D. Soukal, “Writing on Wet Paper,” IEEE Transactions on Signal
Processing, Vol. 53, No. 10, Oct. 2005, pp. 3923-3935.
[9] E. Russell and F. Jarvis, “Mathematics of Sudoku ii,” Mathematical Spectrum, Vol. 39, No. 2, Jan. 2007, pp.
54-58.
[10] G. Voyatzis and I. Pitas, “Applications of Toral Automorphisms in Image Watermarking,” Proceedings of the
IEEE International Conference on Image Processing, Lausanne, Switzerland, Vol. 2, Sep. 1996, pp. 237-240.
[11] X. Wu and Z. H. Guan, “A Novel Digital Watermark Algorithm Based on Chaotic Maps,” Physics Letters A,
Vol. 365, No. 5-6, Jun. 2007, pp. 403-406.
[12] Y. T. Wu, Y. S. Frank, “Digital Watermarking Based on Chaotic Map and Reference Register,” Pattern
Recognition, Vol. 40, No. 12, Dec. 2007, pp. 3753-3763.
Authors
The Duc Kieu received the B.S. degree in mathematics from the University
of Pedagogy, Vietnam, in 1995, the B.S. degree in information technology
from the University of Natural Sciences, Vietnam, in 1999, and the M.S.
degree in computer science from Latrobe University, Australia, in 2005.
Currently, he is a Ph.D. candidate under Professor Chen-Chen Chang’s
supervision at Feng Chia University, Taiwan. His current research interests
include information hiding, image processing, and data compression.
10
International Journal of Smart Home
Vol.3, No.2, April, 2009
11
International Journal of Smart Home
Vol.3, No.2, April, 2009
12