include the second file containing hidden information. The hidden message can ... key to hide and recover the information, respectively. There are several .... This estimated image has been corrupted by noise during its transmission through ...
2004 1st International Conference on Electrical and Electronics Engineering
Robust Steganography using Bit Plane Complexity Segmentation Silvia Torres Maya, Mariko Nakano Miyatake and Ruben Vazquez Medina Seccion de Estudios de Posgrado e lnvestigacion Escuela Superior de Ingenieria Mecanica y Electrica Unidad Culhuacan Instituto Politecnico Nacional AV. Santa Ana No. 1000 Col. San Francisco Culhuacan, Mexico, D.F., Mexico
Abstract - This paper presents a steganographical algorithm based on bit plane complexity segmentation (BPCS), which permits to implement hiding information into images for its secure transmission through no secure channeh Simulation results show that the proposed algorithm is robust against noisy channels. This algorithm offers higher hiding capacity due to that it exploits the variance of complex regions in each bit plane.
I. mTRODUCTION Digital representation of media facilitates the access and potentially improves the portability, efficiency, and accuracy of the information presented. However, undesirable effects of easy data access include an increased opportunity for violation of information security with or without modification of content. Cryptography has provided a useful approach to information security, and encryption programs are widely available. However, the encryption clearly marks a message as containing “interesting” information, and the encrypted message becomes subject to attack. Furthermore, in many cases it is desirable to send information without noticing anyone that information has been sent. The steganography, as its Greek root implies, is covered wriring. Although it was applied from ancient times to the hiding information, steganography is an emerging discipline for hiding information to achieve better security levels. Informally, steganography refers to the practice of hiding secret messages in communications over a public channel so that attackers will not be aware that the secret message exists. Unlike cryptography, which seeks to hide the content of the message, with steganography we seek to hide the existence of the message. Several constraints must be taken in account in a steganografhical process which affect on the whole system, among them, the quantity of data to be hidden, the need for invariance of these data under different channel conditions, distortions and lossy compression, and degree to which the data must be immune to interception, modification, or removal by non authorized person or attacker. Generally two files and a secret key are required for steganography to work. The first tile is cover data that will include the second file containing hidden information. The hidden message can be anything that is embeddable into a bit stream such as plain text or cipher text. In the basic form of steganography, transmitter and receiver share the same secret
0-7803-8531 -4/04/$20.00 02004 IEEE
key to hide and recover the information, respectively. There are several methods to hide information in digital images, from taking advantage of noisy areas where can be drown less attention in an image, to scatter the secret message randomly throughout the image. In this paper we present a steganographical algorithm which uses an image as the cover data, and we embed secret information in the bit-planes of the cover image. This technique makes use of the characteristics of the human visual system whereby a human cannot perceive any shape information in a very complicated binary pattern. We can replace all of the “noise-like” regions in the bit-planes of the cover image with the secret data without deteriorating the image quality. Unlike the similar methods based on BPCS, we use the variance of complexity in each region of the image to determine optimal regions to insert the secret message. The rest of the paper is organized as follows. In Section 11, we introduce the concept of bit plane complexity segmentation and image complexity, their relation and application to hiding information. Section 111, describe the proposed algorithm and the system model used in its implementation. T h e performance of the proposed steganographical method is shown via simulation and the results are presented in Section 1V. Finally, Section V presents some conclusions and future work about this paper. 11. BIT PLANE COMPLEXITY SEGMENTATION (BPCS)
A. Bit Plane Decomposition.
In this section we introduce the concept of bit plane decomposition, which is the basic idea behind our proposed steganographicaf algorithm. Bit plane segmentation decomposes an image into a set of binary images, one for each bit position of the pixel values. For %bits image (there are 8 bits per pixel), 8 bit planes can be decomposed. Figure 1 illustrates bit plane decomposition of a sample image. The least significant bit plane would be a binary image consisting of the least significant bit of each pixel in the image. The next higher bit plane would consist of the next higher bit of each pixel, and so on. Such decomposition provides one efficient way to perform hiding information in digital images.
40
Fig 1. Bit plane decomposiiion.
After the image is decomposed into its bit planes, the bits of one or more of the lowest bit planes will be replaced with secret information, and then recompose the bit planes back into an image. This decomposition process is central idea of BPCS, which are discussed next. B) Cutnpiexity To implement BPCS we need a binary complexity measure that can be applied to each region of each bit plane. There is no standard definition of image complexity. Most implementations of BPCS adopt a black-and-white border length to measure the image complexity. If the border is long, the image is complex, and if it is short, the image is simple. The total length of the black-and-white border is equal to the summation of the number of blacklwhite-changes along the rows and along the columns in an image. In this paper we propose a measure of complexity based on the variance on each region of each bit plane. The embedding capacity without obvious image degradation can be improved if we limit our embedding in the higher planes to regions of the image that are fairly “complex”; i.e., where there is already a natural variation in pixel values in the local region, Noise added to such regions through the embedding process is much less perceptible than noise added to flat regions. Using such a complexity measure, we can determine if each individual region of each bit plane is complex enough for embedding. Unlike the similar existent algorithms, our. method uses the variance as a complexity measure, which permits to improve the hiding capacity and Bit Error Rate (BER) of the system.
information and cover image. Due to that this algorithm is based on the image complexity on each bit plane, the secret key must include data about the exact mapping of secret information over the image. To determine if the complexity of determined region is enough for embedding, we use as reference a threshold value which is a function of the mean value of that region, and if the complexity of the region is greater or equat to this value, we deem that it is embeddable, and if not, we leave the region alone. Based on this idea, we can build a map with regions complex enough to embed information for each bit plane. Figure 3 shows insertion maps for 5,6,7 and 8 bit planes of a sample image using the variance of each region as complexity measure. The embedding algorithm uses these maps to insert information bits into each bit plane of the cover image. The channel models, those transformations suffered by the image during transmission, distribution and authorized usage by the intended recipient. The delivered image may also be intercepted and manipulated by an unauthorized agent (or even by the intended recipient) in order to delete or corrupt the stego-image to avoid recovering the original secret infomiation. Unlike the watermarking systems in which the existence of the watermark can be known by third party, even though it must be difficult to detect the original watermark from the watermarked image when the secret key is not known, in steganography, the insertion process must guarante the invisibility of the message such as nobody knows about the existence of some secret message.
T Host Image
n4 +
Stego Image
-
source
,j
111, PROPOSED STEGANOGRAPHJCAL SYSTEM.
H I
Complexity Computation
I
The proposed system model is shown in Fig. 2. The host image (cover image) is decomposed to the bit planes, while hidden source (secret information) is encoded and embedded to the cover image using bit planes variance criterion. In hidden inforniation encoder, each symbol of the message is mapped to 8x8 binary block before it be added to the image. One of the goals of the steganographical scheme is to make it difficult to guess the exact mapping between secret
I
t-H-t+ -Extraction
Decoder
14r:g+d I I
Bit Plane Decomposition
I
‘
U
Recovered message
Fig 2. Proposed steganographical system.
41
1
In the receiver:
1, 2.
Decompose the host image in its bit planes. Divide each bit plane in 8x8-bit blocks and compute the complexity for each one. If the complexity is higher than a specific threshold value, consider the block as noisy. 3. Design a map with the noisy blocks. 4. Extract the noisy blocks fiom the image. 5. Using the same procedure as the transmitter, decode these noisy blocks to get the originaI secret message. Obviously, the original host image is not necessary to get secret message. This is because the unique useful information is contained in its noisy regions of the stego-image. In the steganography, the goal is to recover the inserted secret message, no to recover the host image in its original form. Fig 3. Generated maps with noisy regions for bit planes 5,6,7 and 8 of the sample image.
In the receiver, an estimation of the stego-image is received. This estimated image has been corrupted by noise during its transmission through the channel. In this paper, the noise is assumed as additive and white Gaussian noise. Once the image is delivered, an inverse process must be realized to recover the original transmitted secret message. First, the received stego-image is decomposed to bit pIanes. Again, we have to compute the complexity for each region on each bit plane and obtain a map with noisy regions. Of course, only those modified blocks of the image will hold the threshold used in the complexity computation process in the receiver. The obtained maps will contain the specific regions of the stego-image where the secret message was inserted. Once the blocks have been extracted, the decoder will make an inverse process to recover the message in its original format. Next, a summary of the proposed algorithm is presented.
IV. SIMULATION RESULTS.
This section presents the obtained results via simulation. For this, a text of X-Kb was used as the secret information to transmit. The threshold value used to determine the noisy blocks was the mean value of all blocks to achieve a better distribution of the message into the image. Figure 4 shows the original host image and Figure 5 shows the same image but after the secret message has been inserted. These figures permit us to verify the characteristic of invisibility offered by the proposed steganographic algorithm. Figures 6 and 7 show, respectively, the symbol error rate (SER) and the bit error
rate (BER) after the message has been recovered in the receptor. Finally, Figure 8 shows the MSE achieved by the algorithm in simulation.
In the transmitter: 1.
2. 3. 4.
5. 6
Transform the secret information to equivalent form using blocks. Each block must have an W i t size. Transform every block to a binary 8x8-bit image. Decompose the host image in its bit planes. Divide each bit plane in 8x8-bit blocks and compute the complexity for each one. If the complexity is higher than a specific threshold vaIue, consider the block as noisy (high complexity level). Design a map with the noisy blocks. Replace the noisy blocks by the information blocks generated in Step 2 based on the insertion map.
Fig 4.Original host image.
42
V. CONCLUSIONS In this paper we presented an algorithm for steganography based on bit plane complexity segmentation. The uses of the variance as parameter to determine the optima1 regions on the image to insert the information improve the hiding capacity of the system and simulation results show that this algorithm is robust against additive and white Gaussian noise. Bit plane complexity segmentation has demonstrated be useful in steganographical methods and in future work, BPCS will be implemented in lossy compression schemes such as
JPEG2000. ACKNOWLEGMENTS
Fig 5 . Host image with inserted message. Id
...... ......... ............................. ..........:........................
....................... ;..........
.
.
..,........... L ....... .j
In-’ 1..._._.. __. i ..................
3 in8 ”
. .
~~
through the SUSST program during the realization of this research, and to professor Yamaguchi Kazuhiko of University of Electro-Communications of Japan for his support during the realization of this research.
... ..._... L.. ....... .J
....
tl..~-- ~ i
Thanks to the National Science and Technology Council (CONACyT) for its support in the realization of this work. We also are grateful with the University of Electro-Communications of Japan for the support provided
REFERENCES
-?\I
........ :................................... ~ ........... ~ . > ........... 1 ........... ,...
........
%---.
W. Bender, D. Gruhl, and N. Morimoto, “Techniques for data hiding,” in Proc. of the SPIE, (San Jose, CA), pp. 2420{2440, February 1995 M.D. Swanson, B. Zhu, A.H. Tewfik: “Robust Data Hiding for Images”, IEEE DSP Workshop, 1996. M. Goljan, J. Fridrich, and R. Du, “Distortion-free data embedding,” Proceedings of 4th Information Hiding Workshop, pp. 27-41, Pittsburgh, PA, April,
io‘
........... ,....................... ............, ........... ,.......... ................................ ...,..... .................. ,...........
2001. R. J. Anderson and A. P. Petitcolas, “On the
1
5
limits of steganography,” IEEE Journal on Selected Areas in Conzmunications, vol. 16, no. 4, pp. 474-481, May 1998. Stefan Katzenbeisser and Fabien A. P. Petitcolas. Information hiding techniques for stegunography and digital waremarking. Artech House Books, 1999. G . J. Simmons, “Results Concerning the Bandwidth of Subliminal Channels,” lEEE J. on Selected Areas in Communications, vol. 16, no. 4, pp, 463473, May 1998. F. Petitcolas, R. Anderson, and M. Kuhn, “Information Hiding - A Survey,” Proceedings of the IEEE, vol. 87, no. 7, pp. 1062-78, July 1999.
6
S N W n UBI
Fig 6 . Symbol error rate achieved.
i L2+=F7do
OMt
011
sm LIB)
Fig 7. Mean square error between the original transmitted message and the recovered message.
43