Upload
razali
View
217
Download
5
Embed Size (px)
Citation preview
High Capacity and Inaudibility Audio Steganography Scheme
Haider Ismael Shahadi
Electronic and Communication Engineering Dept.
Tenaga National University (UNITEN)
Kajang, Malaysia
Razali Jidin
Electronic and Communication Engineering Dept.
Tenaga National University (UNITEN)
Kajang, Malaysia
Abstract—Steganography is an information hiding technique
where secret message is embedded into unsuspicious cover
signal. Measurement of good steganography algorithm
includes security, capacity, robustness and imperceptibility.
These measures are contradicted, therefore improving one,
affects the others. In this paper, we propose a new high
capacity audio steganography algorithm based on the wavelet
packet transform with adaptive hiding in least significant bits.
The adaptive hiding is determined depend on the cover
samples strength and bits block matching between message
and cover signals. The results show that message can be
embedded up to 42 % of the total size of the cover audio signal
with at least of 50 dB signal to noise ratio.
Keywords-: Audio Steganography, Wavelet Packet Transform,
Bits Block Matching, Inaudible Steganography, High capacity
Steganography.
I. INTRODUCTION
In the present era of information technology, eavesdropping can be reduced by employing cryptography or/and steganography. Steganography is a process of embedding secret messages in a cover signal to avoid illegal detection [1]. Steganography differs from cryptography in term of message visibility. It hides secret messages totally compared to cryptography where the secret message is still visible [2].
Steganography is frequently used in covert communication such as military and government communications. Often it requires relatively high payloads when compared to watermarking. The major requirements that should be satisfied for good steganography algorithms include perceptual transparency, payload or capacity and robustness [4]. High capacity is considered as an important aspect for steganography when compared to watermarking. For watermarking, robustness should be a dominant factor. Improvement for one of the mentioned requirements will tend to degrade the other performances as they are contradictory according to the magic triangle [5].
In recently years many techniques have been developed for information hiding [4, 7, 8], and most of these techniques used either image and video media but rarely use audio signal as a cover signal especially in high rate of data embedding, most likely due to Human Auditory System (HAS) which is more sensitive compared to the Human
Visual System (HVS) [8]. Although adopting audio signals as a cover signals may yield inferior inaudible performance, there are still suitable features such as transitory and unpredictability that makes sound signal as a suitable secure cover signal.
In this paper we proposed a new steganography algorithm that has high capacity and high output quality. The proposed algorithm is based on discrete wavelet packet transform (DWPT) with adaptive hiding in least significant bits. The adaptive hiding is determined depend on the cover samples strength and bits block matching between message and cover signals. The output stego-signal has Signal to Noise Ratio (SNR) above of 50 dB for the input capacity 300 kb/sec (this forms more than 42 % from the input file size). The message recovery process does not require original audio cover signal, also this algorithm has good security due to the arbitrary distribution of the secret message blocks in embedding process instead of a known embedded message distribution in many stego-algorithms.
II. RELATED WORK
Generally audio steganography and watermarking can be classified according to the embedding domain either in time or transform domain. The simplest hiding technique in time domain with acceptable capacity is the Least Significant Bits (LSB), but it is vulnerable due to changes in LSB that can possibly destroy the embedded message [5]. In the transform domain, there are many transform methods that can be employed in information hiding such as Fourier domain [5, 9], discrete cosine domain [5, 10], and wavelet domain [5, 8, 11, 12]. Each domain has its features in signal processing and information hiding [13], however, the wavelet domain has a major advantage over the others because it divides a signal into different frequency components with different resolutions, and then each component can be used in embedded process according to its power. DWT behaves similar to the time-frequency characteristics of human ears that has high time and low frequency resolution for the high frequencies, and high frequency and low time resolutions for the low frequencies components [14].
The Discrete Wavelet Transform (DWT) decomposes a signal at a level of decomposition into two components, high and low frequency components. Most power of the input signal is concentrated in low frequency component
104978-1-4577-2155-7/11/$26.00 c©2011 IEEE
called approximation signal, while little power exists in the high frequency component or known as a detail signal. The decomposition process starts by decompose a signal into two components, then will further decompose each the low frequency component into another two components, and the process repeats for further levels of decomposition [14]. The reconstruction of original signal is performed by the Inverse Discrete Wavelet Transform (IDWT).
The modification in the details signals has little effect on the reconstructed signal, rather depends on the number of levels of decomposition and which selected detail level was modified. However modification in the approximation signal or low frequency component may affect significantly the reconstructed signal. Therefore using details signals as a cover for information embedded process enable high payload and an acceptable quality, when it is used in the steganography [8]. However, information embedding in details signal can affect its robustness as it is possible to remove a secret message by signal processing for example an attacker may reset the details coefficients.
The LSB is the most common method employed in embedding process in DWT domain [8, 12, and 16]. This method has superior stego-signal quality and capacity.
In our previous work [17] we used DWPT and samples block matching that based on minimum distance between message and cover samples blocks. In [17] we did not use the LSB in the embedding process, instead we had chosen the replacing process between blocks of samples to maintain the message power in the high level, and therefore has improved robustness against the additive noise. The maximum capacity that we had achieved in [17] was 250 kb/sec with an acceptable quality.
In this work a high capacity and high quality audio steganography algorithm will be described. The purpose of this algorithm is to achieve a high embedding capacity and high output quality to gain advantages of three things, DWPT, bits block matching between message and cover, and adaptive hiding depend on the strength of the cover samples. The proposed algorithm has high embedding capacity reaches up to 300 kb/sec and high quality for output stego-signal (SNR above 50 dB). Another advantage for this algorithm over most algorithms is hardly to detect the positions of embedded secret message especially in low and medium capacity because the message bits blocks are distributed arbitrary according to the result of bits block matching process and this feature make the secret message more protected against detection and tampering. Furthermore the secret message recovery algorithm does not need the original audio cover signal.
The proposed algorithm starts by segmenting the input audio cover signal and then decomposing each segment by using DWPT. The DWPT is similar to DWT except decomposition is performed for both high and low frequency components. The decomposed signal by DWPT for L levels, yields (2
L) components with equal lengths, one
represents the approximation signal that has the highest power and lowest frequency, while the others are details signals with decreasing power, starting from the lowest to the highest frequencies details components. Subsequently
after several steps, the Inverse DWPT (IDWPT) is used to reconstruct the output stego signal.
The proposed scheme however does not use the approximation signal in embedding process to maintain the quality of output of stego-signal. The lowest frequency component of the details signal which has highest power with respect to other details signal is used to embed main key that is generated in bits block matching process. The remaining (2
L – 2) details signals are used in message
embedding process after several processing. The bits block matching compare between bits blocks of
secret message and cover after extracting the contents of bits positions that are dedicated for embedding process. These dedicated bits positions are computed depend on the strength of each sample as shown in section (3.A). The block matching output is used to generate embedding key that is used in both embedding and extraction processes.
III. PROPOSED HIDING SCHEME
Fig. 1shows the main structure of the proposed hiding scheme. The structure includes four main stages come after the stages of input audio segmenting and secret message preprocessing. The input audio signal is segmenting to Gsegments; each segment has length of Z samples. In message preprocessing, the secret message is also segmented to G segments with equal size; each one is arranged as a two dimension (2D) binary matrix with size MxN. The following four main stages are repeated G times to hide each secret message segment in one cover segment:
A. Cover Signal Decomposition and Preparing Stage
Each segment of the input audio cover signal is decomposed using L-levels of Haar DWPT to obtain 2
L
signals one represents the approximation coefficients signal and the others represent details coefficients signals. Each one of the produced signal has length of Z/2L
samples. We select 2
L-2 from the details signal starting from the highest
frequency component for embedding of the secret message. First of all, each details signal is scaled according to its maximum value and number of bits per sample of the input audio signal as follows:
mfDDs
mf
iDMax
ii
r
i
L
ii
×=
−=
−==
max
12
22,....,2,1),(max
(1)
Where: Max is a function that is used to compute maximum value (maxi) in each selected details signal Di, mfi is the multiplication factor that is used for scale Di to obtain the scaled details signal Dsi, and r represents number of bits per sample of the input audio signal.
The second step is using of the round function to approximate Dsi to the nearest integer and then convert each produced scaled details signal to binary matrix (Dsbi) each row in this matrix represents one sample. Then for each sample algorithm has determined the number of bits that
2011 7th International Conference on Information Assurance and Security (IAS) 105
will be dedicated for message embedding in the selected details signals as shown in (2).
)2/(,...2,1,22,...2,1,,,
LL
jiji ZjiiPNB =−=−= (2)
Where NBi,j is the number of bits that are dedicated for embedding in sample j from the details signal i . Pi,j is the position number of the first bit that has value of logic 1 from the most significant bit direction. For example if the sample Dsbi (j) has binary value 001011110001101 then Pi,(j) =13. The value of NB is determined adaptively depending on the strength of each sample and the degree of details signal which belongs to it. In (2) i represent number of bits which do not used in embedding process from P position toward least significant bits direction. These safety bits are increasing from the highest toward the lowest frequency components; in other word they are increasing from the lowest toward the highest power components to maintain the quality of the output signal. If NBi,j less than or equal zero then the sample j of details signal i will be neglected and will not be used in the embedding process.
After completing the above process for all selected scaled details signals samples, the embedded positions contents vector will be constructed as given in (3)
}...,,.........,{
}2/,.....2,1),:1,({
2221
,
−=
==
LEPCEPCEPCEPC
ZjNBjDsbEPC L
jiii
(3)
Where: EPC is the embedding positions contents vector. The final step is converted the EPC vector to W blocks, each block has length of N bits. The last reminder of bits is neglected if they are less than N.
B. Key Generating and Secret Message Embedding Stage
The preprocessed message segment (MP) that has size of MxN and the matrix of embedding positions contents that has size WxM are fed to the bits block matching process. In the bits block matching process, the bits blocks of MP and EPC matrix are compared to compute matching between each bits block (row) of MP and whole blocks (rows) of EPC to obtain the blocks matching matrix BM which has size of MxW. Each element in BM matrix represents number of matched bits in both i preprocessed message block and jblock of EPC matrix. Then, based on maximum matching in each row of BM, some of EPC blocks will be replaced with all MP blocks, also the stego-key will be constructed, then reconstruct the modified selected details signals as follows:
• For each row i in BM matrix algorithm computes the index of maximum value that represents maximum matching between message block i and cover blocks as follows:
MiBMMaxIndexind ii ,....2,1)),(( == (4)
• Modifying each element that has position ind in Match matrix rows (i+1 to M) to negative value (for example -1) to avoid embedding overwrite if there is any similarity in other blocks matching.
• Constructing the Stego-key vector (Key)
MiindKey ii ,......,2,1, == (5)
• Embedding the MP blocks in maximum matching positions of EPC based on key that is generated in the previous step by replacing EPC blocks by MPblocks according to (6) to build modified EPC.
MiMPEPC iindi,......,2,1, ==
(6)
• Arranging the modified EPC matrix in one dimension vector called MEPCV.
• Reconstruct the modified binary scaled matrix of each details signal as given in (7)
}...,,.........,{
22,..2,1,2/,.....2,1
):():1,(
2221
,,
−=
−==
=
LDsbDsbDsbDsb
iZj
NBkMEPCVNBjDsb
LL
jijijii
(7)
Where kj = kj-1+NBi,j if j > 1 and k=1 if j = 1 . Finally convert all the binary matrices of modified
selected scaled details signals to decimal and the final vectors are named MDSi, i=1,2,….., 2L-2.
C. Stego-Key Embedding Stage
Because of the arbitrary distribution of the message
blocks in embedding process in the previous section, the
recovery algorithm of the proposed scheme will need stego-
key and message size to extract the message blocks from the
stego-signal. Therefore, the stego-key with the message size
will embedded in the lowest frequency details signal (D2L
-1).
We choose this signal to embed the stego-key because it has
maximum power between all other details signals and that
make the stego-key more resistance against distortion or
lost. Before the embedding, the stego-key and message size
are converted to binary and arranged in one vector which is called final key (FKey). Then the Fkey vector is encrypted using simple stream cipher. The pseudo random number generator (PRN) is used to produce a random binary vector that has length similar to Fkey. The changing of PRN depends on the changing of the cipher key that is entered to the algorithm as a secret key. The final step in the encryption process is the XORing between the two binary vectors to obtain the encrypted Key vector (EKey).
PRNFKeyEKey ⊕= (8)
The Embedding process of encrypted key starts by scaling and converting to binary of the lowest frequency details signal and then compute the number of bits that dedicated for embedding process for each sample; exactly by the same steps that are used in section (3.A). Finally, the encrypted key vector bits are embedded on the positions according to NB2
L-1 which is computed using (2).
D. Stego-Signal Reconstruction Stage
In this stage all the modified details signals, including the lowest frequency details signal, are converted to decimal
106 2011 7th International Conference on Information Assurance and Security (IAS)
to obtain on MDSi, i=1,2,….2L-1 and then descaling the results as follows:
(9)
The IDWPT is used to reconstruct the segments of
stego- signal based on modified descaled details signals
MDi, i=1, 2… , 2L-1 and unmodified approximation signal.
The reconstructed segments will fed to segment collecting
step to reconstruct the final stegonagraphy algorithm output.
IV. MESSAGE RECOVERY ALGORITHM
In the message recovery algorithm that is shown in Fig. 2, the input stego-signal is segmented to G segments then each segment feed to algorithm stages. First stage is the decomposition and preparing of the segment. The function of this stage is exactly similar to its function in hiding algorithm, it decompose the input segment using L-levels Haar DWPT and then do the scaling and binary converting for all details signals. After that it computes the NBi using (2) in section (3.A) for whole details signals where i=1,2,…2L-1.
The second stage is the key recovery, in this stage the encrypted key extracted from the lowest details signal based on NB2
L-1. Then using same method which is used in
encryption and similar ciphering key to decipher the encrypted key and obtain the final stego-key.
The final stage is the message recovery blocks. In this stage, the embedding positions contents extracting from details signals starting from highest frequency details signal (Dsbi) until to (Dsb2
L-2) based on the number of bits vectors
NBi, where i=1,2,…,2L-2. Then the message blocks are reconstructed from embedding positions contents (EPC)depend on the final stego-key. Finally the message blocks are arranged and collecting all the message segments in one vector then it is converted to its original data type.
V. RESULTS AND DISCUSSIONS
The proposed algorithm was implemented by using Matlab (2010a) programming. The proposed algorithm was tested using three audio cover signals: male speaker, female speaker and music. Each signal has resolution of 16 bits per sample and sampling frequency 44100 samples/sec. Three data types: audio, image (as given in Fig. 3), and text are used in tests as a secret messages. The quality of output signal in each test was computed using SNR in (10).
2
1
1
2
10
)]()([
)(
log10
kCkC
kC
SNRZG
k
ZG
k
=
=
′−
= (10)
Where C and C�are input cover signal and output stego-
signal respectively, ZG represents the number of samples in each one of them.
The proposed scheme was tested for different hiding capacity and the results showed that it has excellent output quality. From the tests we find the proposed algorithm support high capacity rate reach up to 300 kb/sec and that is
form above 42% from the size of the input audio cover file at SNR above 50 dB for the output signal. Fig. 4 shows the relationship between SNR and embedding capacity for three different message type and three different cover signals. In Figs. 4-5 experiments, the following attributes were used: length of block N=128 bits, Z = 10000 samples and L=3.
Fig. 5 shows comparison between proposed scheme and conventional DWT that employed fixed LSB for embedding scheme. In this comparison we used mail audio as a cover signal and three types of data include audio, image and text as messages. The comparison showed the clearly superiority of the proposed scheme over the conventional DWT scheme in high embedded capacity, for example at 300 kb/sec the SNR is above 50 dB in our algorithm while it is in range of 21 dB in conventional DWT scheme for different data type messages.
Table 1 shows a comparison between different values of N and L with respect to SNR and processing time for fixed capacity ( about 100 kb/sec) and Z = 10000 samples. In these tests we use male speaker as a cover signal with length of 230000 samples and camera man image as a secret message with size of 256*256 pixels. The results in table show the reducing in the bits block size will increase the SNR because the increasing in matching rate also this will increase the processing time. The increasing in levels depth will increase in SNR and processing time but the process of increase L does not always allow because this depends on the entered capacity and bits block length (N).
The arbitrary result of bits block matching make the distribution of secret message blocks over the cover signals arbitrary and that increase the security of secret message, moreover the block length and message size encrypted with the stego-key. That entire make the detection of steganographic information that is based on statistical changes in the signal is complex especially in low and medium capacity.
VI. CONCLUSIONS
We have presented a high capacity and high stego-signal quality audio steganography scheme based on DWPT and bits block matching. The DWPT, bits block matching, and adaptive embedding in LSB of cover samples depend on their strength are enabled the algorithm to achieve very high embedding capacity for different data type that can reach up to 42 % from the input audio file size with lest of 50 dB SNR for the output stego signal.
The arbitrary results of the block matching generate an arbitrary key for embedding process, and that help in increasing the security of steganographic information in the proposed algorithm. Another advantage for the proposed algorithm is the reconstruction of the actual secret messages does not require the original cover audio signal and therefore, the cover signal can be any recorded audio by the hiding side. In future work we will use lifting wavelet transform instead of DWPT to eliminate the error that may occur in the construction of DWPT because the scaling and descaling processes and that may increase efficiency of the stego-algorithm.
12,.....2,1, −==Li
i imf
MDsMD
2011 7th International Conference on Information Assurance and Security (IAS) 107
0 0.5 1 1.5 2 2.5
x 105
-6000
-4000
-2000
0
2000
4000
6000
8000
Sample
Am
plit
ud
e
(a) Speech Message Signal (b) Image Message
20 40 60 80 100 120 140 160 180 200 220 240 260 280 30030
35
40
45
50
55
60
65
70
Embedded Capacity (kb/sec)
SN
R d
b
Audio Massage
Image Message
Text Message
(a) Mail Speaker Cover Audio Signal
20 40 60 80 100 120 140 160 180 200 220 240 260 280 30030
35
40
45
50
55
60
65
70
Embedded Capacity (kb/sec)
SN
R (d
b)
Audio MessageImage MessageText Message
(b) Female Speaker Cover Audio Signal
20 40 60 80 100 120 140 160 180 200 220 240 260 280 30030
35
40
45
50
55
60
6566
Embedded Capacity (kb/sec)
SN
R (d
b)
Audio MessageImage MessageText Message
(c) Music Cover Audio Signal
Message Embedding Stage Dsb
BM
Modified Details
Signals
DWPT
L-Level
Scaling and
Converting to
Binary
Compute NB
Extract Embedded
Positions Contents
Key Generating and
Embedding
Preprocessing
Input Audio Cover
Signal
Cover Signal Decomposition and Preparing Stage
Descaling and
Converting to
Decimal
IDWPT
L-Level
Stego
Key
Stego-Key Embedding
Ciphering
Scaling, Binary
Converting, NB
Computing
Key Embedding
Cipher
Key
Approx.
Signal
Non-Modified
Approx. Signal
Output
Signal
Secret Message
Lowest Frequency
Detail Signal
PM
EPC
Reconstruction Stage
Bits Blocks
Matching and
Replacing
Details
Signals
Segmentation
Segment
Collecting
Message Blocks
Recovery
NBi,
i=1,2,,…..2L-2
Encrypted Key
Recovery
Message Recovery Stage
Decipher Key
NB2L
-1
MDsbi,
i=1,2,,…..2L-2
Key Recovery
Stage
Input Audio
Stego-Signal
Message
Segments
Arranging
and
Collecting
Deciphering
Computing of
NB
Scaling, Binary
Converting DWPT
L-Level
MDsb2L
-1
Decomposition and Preparing Stage
Output Secret
Message
Segmentation
Figure 1 the General Structure of the Proposed Hiding Scheme
Figure 2 the Message Recovery Algorithm
Figure 3 Input Secret Messages
Figure 4 the Relationship between SNR and Embedding Capacity for
Different Cover Signals and different Data Type
108 2011 7th International Conference on Information Assurance and Security (IAS)
Table I SNR and Processing Time for Different N and L with capacity of 100 kb/sec
REFERENCES
[1] N. Provos and P. Honeyman, “Hide and seek: An introduction to steganography,” IEEE Security and Privacy Magazine, Vol. 1, No. 3, June 2003, pp. 32–44.
[2] H. Wang, and S. Wang, “Cyber warfare: Steganography vs. Steganalysis,” Communications of the ACM magazine, Vol. 47, No.10, October 2004, pp. 76-82.
[3] A. Cheddad, J. Condell, K. Curran, and P. M. Kevitt, “ Digital image steganography: Survey and analysis of current methods,” Journal of Signal Processing, Vol. 90, No. 3, March 2010, pp.727–752.
[4] Y. Wang; P.Moulin, “Perfectly Secure Steganography: Capacity, Error Exponents, and Code Constructions,” Information Theory IEEE Transactions, Vol. 54, No. 6, Jun 2008, pp. 2706 – 2722.
[5] N. Cvejic,“Algorithms for Audio Watermarking and Steganography,” MSc. thesis, Department of Electrical and Information Engineering, Information Processing Laboratory, University of Oulu, Finland Oulu, Finland, 2004.
[6] K. Bailey, and K. Curran, “An Evaluation of image based Steganography methods,” Journal of multimedia Tools and Applications, Vol. 30, No. 1, July, 2006, pp. 55-88.
[7] N. Meghanathan, and L. Nayak. “Steganalysis algorithms for detecting the Hidden information in image, audio and Video cover media,” International Journal of Network Security & Its Application (IJNSA), Vol.2, No.1, January 2010, pp. 43-55.
[8] S. Shahreza and M. Shalmani, “High capacity error free wavelet Domain Speech Steganography,” IEEE International conference on acoustics, speech, and signal processing, March 31 -April 4, 2008, pp. 1729 - 1732.
[9] A. Khashandarag, A. Oskuei, H. Mohammadi and M. Mirnia, “A Hybrid Method for Color Image Steganography in Spatial and Frequency Domain,” IJCSI International Journal of Computer Science Issues, Vol. 8, Issue 3, No. 2, May 2011, pp. 113-120.
[10] Z. Zhou, and L. Zhou “A Novel Algorithm for Robust Audio Watermarking Based on Quantification DCT Domain,” Third International Conference on International Information Hiding and Multimedia Signal Processing, vol. 1, 26-28 November, 2007, pp. 441-444.
[11] S. Wu, J. Huang, D. Huang and Y. Q. Shi, “A Self-Synchronized
Audio Watermarking in DWT Domain, “Circuits, Vol. 5, 23-26 May 2004, pp. 712-715.
[12] N. Cvejic, and T. Seppanen, A Wavelet Domain LSB Insertion Algorithm For High Capacity Audio Steganography,” Digital Signal Processing Workshop, 2002 and the 2nd Signal Processing Education Workshop. Proceedings of 2002 IEEE 10th, 13-16 October 2002. , pp. 53-55.
[13] M. Hasija1 , A. Jindal2,” Contrast of Watermarking Techniques in different domains,” IJCSI International Journal of Computer Science Issues, Vol. 8, Issue 3, No. 2, May 2011, pp. 559-563.
[14] G. Tzanetakis, G. Essl, and P. Cook "Audio Analysis using the Discrete Wavelet Transform, " In Proceeding WSES International Conference, Acoustics and Music: Theory and Applications (AMTA 2001) , kiathos, Greece, 2001. pp. 318-323.
[15] X. yang,W. P. Niu, M. Lu, , “A Robust Digital Audio Watermarking Scheme using Wavelet Moment Invariance,” The Journal of Systems and Software, Vol. 84, No. 8 , 2011, pp. 1408-1421
[16] R. Santosa, and P. Bao, “Audio to image wavelet transform based audio steganography,” Proceeding of 47th International Symposium, ELMAR, June 2005, pp. 209- 212.
[17] I. Haider, and J. Razali, “High Capacity and Resistance to Additive Noise Audio Steganography Algorithm,” IJCSI International Journal of Computer Science Issues, Vol. 8, Issue 5, No 2, September 2011, pp. 176-184.
Number of
DWPT Levels
(L)
Length of Block
(N bits)
Output
SNR (dB)
Processing
Time (sec)
3 128 56.7 28.993
3 265 55.8 21.773
3 512 55.4 18.876
3 1024 55.2 17.752
4 128 59.2 31.231
4 256 58.8 24.618
4 512 58.7 20.719
4 1024 58.1 19.258
20 40 60 80 100 120 140 160 180 200 220 240 260 280 30020
30
40
50
60
70
80
90
100
Capacity (kb/sec)
SN
R (d
b)
Proposed SchemeConventional DWT Scheme
20 40 60 80 100 120 140 160 180 200 220 240 260 280 30020
30
40
50
60
70
80
90
Capacity (kb/sec)
SN
R (d
b)
Proposed SchemeConventional DWT Scheme
(a) Audio Message Signal
(b) Image Message
20 40 60 80 100 120 140 160 180 200 220 240 260 280 30020
30
40
50
60
70
80
90
Capacity (kb/sec)
SN
R (d
b)
Proposed SchemeConventional DWT Scheme
(c) Text Message
Figure 5 Comparisons between Proposed Scheme and Conventional DWT Scheme for Embedding Capacity and SNR
2011 7th International Conference on Information Assurance and Security (IAS) 109