2016 6th International Conference on Information Communication and Management

Wavelet Based Image Steganographic System Using Chaotic Signals

Hikmat N. Abdullah Sura F. Yousif

College of Information Engineering College of Engineering
Al-Nahrain University Diyala University
Baghdad, Iraq Diyala, Iraq
[email protected] [email protected]

Alejandro A. Valenzuela
EMT department
BRS university of Applied Sciences
Sankt Augustin, Germany
[email protected]

Abstract—The main requirements of efficient steganographic

systems are transparency, robustness and security. In this I. INTRODUCTION
work, the proposed system is designed to achieve these Steganography is the art and science of hiding secret data
requirements, as follows: to increase the transparency and by embedding them within other information. The secret data
robustness, the proposed system uses the transform domain
itself could be graphics or sounds. Computer files [1], texts
technique, in this case the wavelet domain technique to embed
the color secret image into the color cover image. To achieve
[2], audio files [3], videos and digital images [4], all of
high security, chaotic systems are used to accomplish the which contain redundant information, can hide the secret
following two purposes: firstly to scramble the secret color data as carriers or covers. The basic requirements of any
image before hiding it into the color cover image, secondly to steganographic system are capacity, invisibility, security and
randomly select the embedding positions in the cover image. robustness [1]. The two most popular techniques used today
Three chaotic systems with many combinations are used for to hide secret information inside digital images are spatial
the scrambling and the embedding process. Experimental domain and transform domain. The spatial domain technique
results including Mean Square Error (MSE), Peak Signal to embeds the bits of the secret message straight to the least
Noise Ratio (PSNR), Correlation, Signal to Noise Ratio (SNR)
significant bits of the cover image. This results in high
and embedding capacity demonstrate that the proposed system
has good performance as compared with other existing
vulnerability to attacks such as formatting and image
methods without attack and under different types of attacks changes [2]. The transform domain technique on the other
such as filtering and noise attacks. Simulation results show that hand hides information in the significant parts of the cover
the combination of chaotic systems providing the best image in order to be more robust against attacks such as
performance include the use of Logistic and Hénon maps for filtering, cropping and lossy compression but still staying
the embedding process and Arnold´s cat map for the imperceptible to the human sensitive system. Discrete
encryption process. The MSE, PSNR, correlation, SNR and Wavelet Transform (DWT), Discrete Cosine Transform
embedding capacity obtained in this case reach up to 0.0010, (DCT) and Discrete Fourier Transform (DFT) are the most
78.1dB, 0.9999, 67.2dB and 0.39% respectively when the secret
popular examples of transform domain techniques. The
image is Airplane with 32x32 pixels and the cover image is Girl
with 512x512 pixels. For the extracted secret image, the chaotic
signal or image is decomposed using DWT into high
system set that provides the best performance includes the use frequency and low frequency parts. The signal's edge
of Arnold´s cat map for the encryption process. The MSE, components are defined in the high frequency parts (details)
PSNR, correlation and SNR obtained in this case reach up to while the low frequency parts (approximations) represent the
0.0846, 58.9dB, 1 and 56.1dB respectively. As compared with identity of the signal. Due to its time-frequency
other methods, the proposed method achieves gains without decomposition properties, DWT has been used in
any attack of 8.2dB for the stego image and 32.4dB for the steganography to give it more robustness [3] and is also used
extracted secret image in PSNR when the secret image is as a transform domain technique in our approach.
Airplane with 256x256 pixels and the cover image is Girl with Chaotic signals refer to systems that have nonlinear
512x512 pixels.
behavior. In contrast to other signals, sequences of chaotic
Keywords-component; steganography; secret image; chaos;
systems have their private characteristics, such as sensitivity
encryption to initial conditions. However, the most powerful
characteristics of chaotic signals are their unpredictability
due to the fact, that there are no repeated values within their
sequence, even for lengthy ones. Due to these excellent
statistical properties, they have been used recently in

978-1-5090-3495-6/16/$31.00 ©2016 IEEE 130

steganograhic systems in order to spread the information A. Logistic Map
over a wide range as compared to conventional periodic PN The logistic map is one of the popular and simplest
sequences [4]. chaotic maps used for generating random sequences, it is a
II. RELATED WORK one dimensional chaotic map. It has the following difference
equation:
Wherever Times is specified, Times Roman or Times
New Roman may be used. If neither is available on your
word processor, please use the font closest in appearance to
Times. Avoid using bit-mapped fonts if possible. True-Type where n is the state number, is the initial condition and r
1 or Open Type fonts are preferred. Please embed symbol is the control parameter. The value of should be in the
fonts, as well, for math, etc. Lin and Shin present a data
interval [0,1]. According to reference [4] the maximum
hiding scheme for gray scale image in the spatial domain [5]
randomness is achieved when the control parameter is in
were inverse embedding techniques are used to decrease the
the interval [3.57,4].
distortion of the stego images. Atawneh and Sumari [3]
propose a steganographic technique based on chaotic map B. Hénnon Map
and DWT, where a coding map is generated and DWT is The Hénon map is a two dimensional chaotic map. It
applied to the cover image. The secret image is then depends on two variables: and . It has the following
scrambled by using Arnold´s cat map before it is embedded difference equations:
into the cover image. Boora and Gambhir [6] present a
steganographic method to hide a binary image in a gray
(2)
scale image using wavelet domain and Arnold´s transform.
The secret image is scrambled using Arnold´s cat map, then
DWT is applied to both cover and scrambled images. (3)
Abdulhameed and Mahmood [7] propose a gray image
steganography based on contourlet transform and chaotic where a and b are the control parameters. In order to
map. Contourlet transform is used to embed a secret generate iterations of chaotic sequence, the value of a and b
message into a cover image and a modified Arnold cat map must be 1.4 and 0.3 respectively [10].
is used for increasing the key space. Bandyopadhyay et.al [4] C. Arnold's Cat Map
present a steganographic approach to hide a secret message
inside a color image based on spatial domain and chaotic Arnold´s cat map is a two dimensional chaotic map. The
map. Before embedding the secret image into the cover pixels of a digital image will be scrambled and the image
image, the secret image is encrypted using Logistic map to looks noisy when this chaotic map is applied to it. Arnold`s
increase the security. Parul et.al [8] propose an image cat map or as it is sometimes called Arnold´s transform has
steganography for color image using chaotic map and DWT. the following equation:
The secret and the cover images are splitted to three layers.
Then, DWT is applied to each layer of the cover image. (4)
Each layer of the secret image is transformed using Arnold´s
cat map then it is embedded into the cover image. where is the pixel location in the original image and
An efficient steganographic approach based on DWT is the new pixel location in the scrambled
and chaotic systems is reported in this paper. The proposed
image after Arnold´s transformation, is the size of
approach uses DWT technique to embed the secret image
the image. The iteration of pixel location in the image will
into the cover image in order to increase the transparency
result in scrambling the pixel location, which means that the
and robustness and uses chaotic systems to increase the
image is encrypted using this chaotic map. Only the pixels
security. The proposed steganographic system is a blind
locations are changed not their values in this scrambling
system because the original cover image is not needed to
process [6].
extract the secret image at the recipient side.
IV. THE PROPOSED SYSTEM
III. CHAOTIC SYSTEMS
The proposed system hides a colored secret image of
The equations of chaotic systems change over time
different sizes into a color cover image of size (512x512).
because they are dynamical systems. These equations are
The proposed system consists of two parts: the transmitter
either differential equations, which are called flows, or
and receiver part.
difference equations, which are called maps. For this reason,
there are two types of chaotic systems: chaotic flows and A. The Transmitter Part
chaotic maps. Some of the well-known chaotic flows are the The transmission process in the proposed system
Lorenz system and the Rössler system and some of the well- contains many stages and algorithms in order to achieve the
known chaotic maps are the Logistic map, the Hénon map main aim of steganography. The flow chart of the proposed
and the Arnold´s cat map [9,10]. system at the transmitter side is shown in Fig. 1. Since we
are dealing with colored images, the cover and secret

131
images are separated to three layers at the transmitter part: B. The Receiver Part
red, green and blue. Then a two level DWT is applied to The receiver cannot extract the secret information
each layer of the cover image and three chaotic systems without knowing the secret keys that have been used in the
generated: ch1, ch2 and ch3. Each layer of the secret image embedding process (ch1, ch2 and ch3). By applying a two
is scrambled or encrypted using ch3 and a one level DWT is level DWT to the stego image, the secret image is extracted
applied to each scrambled layer after the scrambling process. from the coefficients of the stego image using replicas of
The embedding process is done using ch1 and ch2. Channel ch1 and ch2 that were used at the transmitter side. The
ch1 will select the suitable coefficients in the cover image resulting scrambled or encrypted image will be decrypted
for the embedding process and ch2 will select the suitable using the replica of ch3. The secret image is then
bits in the selected coefficients and replace them with the successfully extracted by the inverse processes that have
coefficients of the secret image. Each layer of the secret been used for embedding the secret image. The flow chart of
image is embedded in the corresponding layer of the cover the proposed system at the receiver side is shown in Fig. 5.
image. The preparation of the cover and the secret image are
shown in Figs. 2 and 3 respectively. After the embedding
process is successfully completed and all coefficients of the
secret image are embedded in the coefficients of the cover
image, an Inverse Discrete Wavelet Transform (IDWT) is
applied to each layer of the cover image: red, green and blue.
The final step is concatenating the three layers to obtain the
stego image as shown in Fig. 4. The stego image is very
similar to the cover image.
(a) (b)
Figure 2. (a) Cover image, (b) 2-level DWT of the cover image.

Figure 3. (a) Secret image, (b) Scrambled image, (c) 1-level DWT of the
scrambled image.

(a) (b)
Figure 1. The flow chart of the proposed system at the transmitter side. Figure 4. (a) Cover image, (b) Stego image.

132
 for the two secret images Baboon and Airplane with different
sizes. The results are given in Tables I and II for the cover
image Girl and the secret image Airplane. The results are
given in terms of embedding capacity and the objective tests:
MSE, PSNR, Correlation (Cor.) and SNR for the stego
image and for the extracted secret image. Tables III and IV
show the summary of the results obtained in the four study
cases considered for the stego images and for the extracted
secret images respectively.
TABLE I. CH1, CH2 AND CH3 ARE LOGISTIC MAPS

Dimension of the
32*32 64*64 128*128 256*256
secret image
Capacity
0.39 1.56 6.25 25
(%)

Stego image
MSE 0.0010 0.0023 0.0047 0.0126
PSNR (dB) 78.0102 74.5157 71.4048 67.1275
Cor. 0.9999 0.9999 0.9999 0.9998
SNR (dB) 67.0920 63.5976 60.4866 56.2093
MSE 0.2148 0.2135 0.2167 0.2171

secret image
Extracted PSNR (dB) 54.8096 54.8360 54.7727 54.7632
Cor. 1 1 1 1
SNR (dB) 52.0331 52.0988 52.0622 52.0673

TABLE II. CH1 IS LOGISTIC MAP, CH2 IS HENNON MAP AND CH3 IS
ARNOD CAT MAP.

Dimension of the
32*32 64*64 128*128 256*256
secret image
Capacity
(%) 0.39 1.56 6.25 25
Stego image

MSE 0.0010 0.0023 0.0047 0.0126

Figure 5. The flow chart of the proposed system at the receiver side.
PSNR (dB) 78.1216 74.5489 71.4598 67.1338
Cor. 0.9999 0.9999 0.9999 0.9998
V. SIMULATION RESULTS
SNR (dB) 67.2034 63.6307 60.5417 56.2156
To evaluate the quality of the proposed system, many
tests have been carried out. First, the Mean Square Error MSE 0.0846 0.0851 0.0872 0.0865
Extracted secret

(MSE), which represents the difference between the cover PSNR (dB) 58.8553 58.8303 58.7267 58.7598
image

image and the stego image, is calculated. Then, the Peak

Signal to Noise Ratio (PSNR), which is used for measuring Cor. 1 1 1 1
the quality of the stego image as compared to the original SNR (dB) 56.0788 56.0932 56.0161 56.0639
cover image, was determined. If the value of PSNR is larger
than 30dB, then it is difficult for the human eye to
distinguish between the stego image and the cover image [5]. VI. ROBUSTNESS OF THE PROPOSED SYSTEM
Another test is the correlation test which is used to compare
To evaluate the performance of the proposed system,
the similarity between the cover image and the stego image.
different types of attacks were performed to show the
The cover image and the stego image are very similar if the
robustness of the stego image and the extracted secret image.
value of correlation is close to one [7]. The final test is the
These attacks were applied to the cover image Girl with the
Signal to Noise Ratio (SNR) test, which is used as a measure
secret image Airplane of size 256x256. The performance
to determine the distortion by noise [11]. The calculations of
without attacks is given in Table II, using a Logistic map for
the proposed system were obtained using Matlab (R2013a)
channel ch1, a Hénon map for ch2 and Arnold´s cat map for
on a personal MS Windows based computer running at 2.40
ch3. Table V gives the results of these attacks for both stego
GHz (Core i3 CPU). The calculations were done for the two
images and the extracted secret image.
cover images Lena and Girl with each 512x512 pixels and

133
 TABLE III. SUMMARY OF RESULTS OBTAINED IN THE FOUR STUDY method, as compared with ref. [4], which has the maximum
CASES CONSIDERED FOR THE STEGO IMAGES: LENA AND GIRL PSNR, is 8.2dB. Furthermore, the extracted secret image of
the proposed system is compared with the extracted secret
image of the references [7] and [8]. The results of the
comparison are given in Table VII. As from this table, the
extracted secret image using the proposed system provides
better results than other existing methods. The gains
obtained in PSNR by the proposed method, as compared
with ref. [7], which has the maximum PSNR, is 32.4dB.
TABLE VI. COMPARISION BETWEEN THE PROPOSED METHOD AND
OTHER EXISTING METHODS WITHOUT ATTACK
Ref. [3] Ref. Ref. [6] Ref. [7] Ref. [8] Proposed
[4] method
PSNR 47.3158 58.94 30.8231 42.7789 49.5629 67.1338
(dB)

TABLE IV. SUMMARY OF RESULTS OBTAINED IN THE FOUR STUDY TABLE VII. COMPARISION BETWEEN THE EXTRACTED SECRET IMAGE
CASES CONSIDERED FOR THE STEGO IMAGES: BABOON AND AIRPLANE AIRPLANE IN THE PROPOSED METHOD AND OTHER METHODS WITHOUT
ATTACK
Ref. [7] Ref. [8] Proposed method
PSNR(dB) 26.3730 12.0818 58.7598
Ref. [7] Ref. [8] Proposed method
PSNR(dB) 26.3730 12.0818 58.7598

B. With Attack
Table VIII compares the stego image Girl of the
proposed method with the methods in reference [12] with
attacks. Table IX compares the extracted secret image
Airplane of the proposed method with the method in
TABLE V. THE RESULTS OF ATTACKS ON THE STEGO IMAGE GIRL reference [13] with attacks. As shown from Table VIII, the
AND THE EXTRACTED SECRET IMAGE AIRPLANE
proposed system is robust against filtering attacks, salt and
pepper noise attack, when compared with other existing
methods. The gains obtained in PSNR and correlation by the
proposed method, as compared with ref. [12], for salt and
pepper noise attack are 12.3dB and 0.0653 respectively. For
median filter attack, the gains obtained and compared with
the same reference are 11.3dB and 0.0447 respectively. For
low pass filter attacks, the gains obtained and compared with
the same reference are 9.9dB and 0.0635 respectively.
TABLE VIII. COMPARISION BETWEEN THE STEGO IMAGE GIRL IN THE
PROPOSED METHOD AND OTHER METHODS WITH ATTACKS
Ref. [12] Proposed method
Attack type PSNR PSNR
Cor. Cor.
(dB) (dB)
Salt and pepper
noise 22.1052 0.9284 34.3944 0.9937
Strength (0.1)
Median filter (3*3) 31.0038 0.9543 42.3437 0.9990
VII. PERFORMANCE EVALUATION Low pass filter (3*3) 28.1968 0.9338 38.0473 0.9973
In order to evaluate the performance of the proposed Wiener filter (3*3) 34.2973 0.9716 36.3780 0.9960
system, its results are compared with the some existing
methods in the literature as follows:
TABLE IX. COMPARISION BETWEEN THE EXTRACTED SECRET IMAGE
AIRPLANE IN THE PROPOSED METHOD AND OTHER METHODS WITH
A. Without Attack ATTACK
In the without attack case, the performance of the
Attack type Ref. [13] Proposed method
proposed system is compared with the methods in
references [3, 4, 6-8]. In this comparison, the secret image PSNR(dB) PSNR(dB)
Salt and pepper noise 11.325 12.2045
Airplane of size 256x256 pixels from Table II is used and Strength (0.1)
the results depicted in Table VI. This table shows that the
proposed method gives better results than other existing Speckle noise 11.517 12.4209
Strength (0.01)
methods. The gain obtained in PSNR by the proposed

134
 Finally, for Wiener filter attacks, the gains obtained and Science and Technology, Vol.2, No.10, November 2010, pp.5985-
compared with the same reference are 2.1dB and 0.0244 5992.
respectively. [3] S. Atawneh, P. Sumari, "Hybrid and blind steganographic method for
digital images based on DWT and chaotic map," Journal of
Communications, Vol.8, No.11, November 2013, pp.690-699.
VIII. CONCLUSIONS
[4] D. Bandyopadhyay, K. Dasgupta, J. Mandal and P. Dutta, "A Novel
In this paper, an efficient, robust and secure approach for secure image steganography method based on chaos theory in spatial
image steganography using DWT and chaotic systems has domain," International Journal of Security, Privacy and Trust
been proposed. Both cover and secret images are Management, Vol.3, No.1, February 2014, pp.11-22.
decomposed using DWT to increase the invisibility of the [5] L. Jia, S. Shin and K. Yoo, "A reversible data hiding scheme using
proposed system. High security is achieved by using chaotic inverse embedding methods in double-embedding strategies," IEEE
International Conference on Informatics and Systems (INFOS 2010).
systems such that without knowing the three keys: chaotic
[6] M. Boora and M. Gambhir, "Binary text-image steganography,"
system1, chaotic system2 and chaotic system3, it is difficult International Journal on Recent Technology and Engineering, Vol.2,
to the attacker to break the proposed system. Even if he No.5, November 2013, pp.126-131.
knows that there is a hidden message, and he manages to [7] Z. Abdulhameed and M. Mahmood, "Steganography based on chaos
extract it, he will get the encrypted or scrambled image not and contourlet transform for hiding multimedia data," International
the original secret image. Tables I and II show that the Journal of Electronics and Communication Engineering &
proposed system has good performance for both stego and Technology, Vol.5, No.1, January 2014, pp.26-42.
extracted secret images in terms of MSE, PSNR, correlation, [8] P. Manju and H. Rohil, "Optimized image steganography using
discrete wavelet transform," International Journal of Recent
SNR and high embedding capacity which reaches up to one Development in Engineering and Technology, Vol.2, No.2, February
quarter of the size of the cover image (256x256) without 2014, pp.75-81.
any distortion to the stego image. As compared with other [9] A. Mummert, S. Sara and J. Silver, "Numerical calculation of
methods, the proposed method achieves gains without any Lyapunov exponents for three-dimensional systems of ordinary
attack of 8.2dB for the stego image and 32.4dB for the differential equations," M.Sc. Thesis, The Graduate College of
extracted secret image in PSNR when the secret image is Marshall University, 2011.
Airplane with 256x256 pixels and the cover image is Girl [10] P. Addison, Fractals and chaos. Institute of Physics Publishing, Dirac
House, Temple Back, Bristol, UK, 1997.
with 512x512 pixels. By comparing the proposed approach
with other existing approaches, the analysis is accomplished [11] M. Alwa, T. El-Tobely, M. Fahmy, M. Nasr and M. Abd El-Aziz,
"Fidelity and robust digital watermarking adaptivity pixel based on
and it can be concluded in terms of PSNR, correlation for medial pyramid of embedding error gray scale images," International
both stego image and extracted secret image with and Journal of Computer Science and Network Security, Vol.10, No.6,
without attacks, that the proposed approach is superior to June 2010, pp.284-314.
other existing approaches. [12] Z. Ruimei, W. Mei, H. Boning and L. Hua, "Digital image
watermarking based on wavelet transform," Proceedings of third
REFERENCES IEEE International Symposium on Intelligent Information
Technology Applications, 2009, pp.437-440.
[1] S. Thakar and M. Aggarwal, "A review –steganography,
"International Journal of Advanced Research in Computer Science [13] A. Verma, A. Nolkha, A. Singh and G. Jaiswal, "Implementation of
and Software Engineering, Vol.3, No.12, December 2013, pp.528-532. image steganography using 2-level DWT technique," International
Journal of Computer Science and Business Informatics, Vol.1, No.1,
[2] J. Ashok, Y. Raju, S. Munishankaralah and K. Srinivas, May 2013, pp.1-14.
"Steganography: An overview," International Journal of Engineering

135