A secure image steganography based on burrows wheeler transform and dynamic bit embedding

International Journal of Electrical and Computer Engineering (IJECE)
Vol. 9, No. 1, February 2019, pp. 460~467
ISSN: 2088-8708, DOI: 10.11591/ijece.v9i1.pp460-467  460
Journal homepage: http://iaescore.com/journals/index.php/IJECE
A secure image steganography based on burrows wheeler
transform and dynamic bit embedding
Ahmed Toman Thahab
Department of Electrical and Electronic Engineering, University of Kerbala, Iraq
Article Info ABSTRACT
Article history:
Received Jan 11, 2018
Revised Oct 10, 2018
Accepted Oct 27, 2018
In modern public communication networks, digital data is massively
transmitted through the internet with a high risk of data piracy.
Steganography is a technique used to transmit data without arousing
suspicion of secret data existence. In this paper, a color image steganography
technique is proposed in spatial domain. The cover image is segmented into
non-overlapping blocks which are scattered among image size window using
Burrows Wheeler transform before embedding. Secret data is embedded in
each block according to its sequence in the Burrows Wheeler transform
output. The hiding method is an operation of an exclusive-or between a
virtual bit which is generated from the most significant bit and the least
significant bits of the cover pixel. Results of the algorithm are analyzed
according to its degradation of the output image and embedding capacity.
The results are also compared with other existing methods.
Keywords:
Burrows wheeler transform
dynamic embedding
Data security
Payload
XOR
Copyright © 2019 Institute of Advanced Engineering and Science.
All rights reserved.
Corresponding Author:
Ahmed T.Thahab,
Department of Electrical and Electronic Engineering,
University of Kerbala.
104 Al-Mwathafeen Road, Kerbala Province, Iraq.
Email: toeahmed@gmail.com
1. INTRODUCTION
A technique that conceals secret information inside another variety of information data is termed as
steganography [1]. It is broadly used nowadays in information security strategies over the last decade since
this kind of invisible communication prevents imperrmissible access to secret information [1]. Although
other data security strategies such as cryptography and encryption techniques exist for secure data
transmission, steganography uses files such as audio, image, video formats to transmit data oblivious to the
fact that data exits in the message [1]. Watermarking is also considered as a hiding technique, but it is used to
protect copyright of content such as small tags or company logos. Steganography conceals greater payload in
cover media [2].
Essentially, any stenographic system consists of three major components: The cover file where
information is hidden; the secret file and the stego-file which is considered as the output resultant from the
system. When embedding secret message data in a cover media, it ought to be visually undetectable and the
output stego-file should not possess any noticeable adaptation compared to the original cover image [3].
Imperceptibility is crucial factor when it comes to steganography techniques; it reflects the level of
correlation between the stego and the cover file. Moreover; other factors such as security and capacity are
also considered decisive factors for steganography algorithm.
Steganography has been under the scope of researcher for some time since; it is a vital field for
covert communication. Secret information can be embedded in the spatial domain specifically in the pixel
intensity value [4]. A pioneering embedding method used in steganography is the least significant bit
technique (LSB). This technique utilizes the least significant bits of a code-word and replaces them with the
most significant bits of the secret binary data. In [5], a new LSB technique is proposed using differencing.

Int J Elec & Comp Eng ISSN: 2088-8708 
A secure image steganography based on burrows wheeler transform… (Ahmed Toman Thahab)
461
The method takes the difference between bit No.5 and bit No.6 to hide secret message. If the result of
difference is not equal to the secret information bit the bit no. 5 is transverse. Results of this method show a
peak signal to noise ratio (PSNR) of 51.1803 with payload of 262144 bits. The main disadvantage of method
in [5] and generally LSB substitution is the techniques’ tolerance in terms of security and capacity [6], [7],
therefore; it has been combined with other techniques such as encryption and source coding etc.
Authors in [8] encoded the secret data using Huffman encoding method and embed the encoded
secret data in the cover image using LSB method. Although the technique achieved a PSNR of +57.43dB, the
embedding capacity of the technique used is 25% of the cover image size. Huffman encoding requires a
Huffman table to be known at the receiver side to decode the secret data. Processing time is also a critical
factor in the performance of this algorithm. Other authors used encryption to decrease the chances of data
security compromise. In [9], an improved version of least significant bit (LSB) method is proposed using
encryption. The secret data embedding positions are encrypted using a secret encryption key. The maximum
stego file quality (PSNR) is 53.7869dB without mentioning the embedding capacity of the algorithm.
For LSB to have more security features, authors in [10] utilized a Ron Code (RC4) to randomize the
concealment of secret data bits in the cover image instead of LSB conventional sequent embedding manner.
The complexity of the algorithm and its time processing consumption are disadvantageous to the proposed
method.
Signal transform tools such as discrete wavelet transform (DWT) is utilized in steganography to
increase embedding capacity without degenerating visual quality. Lately, numerous researches have been
founded on discrete wavelet transform (DWT), since the transform provides a vital frequency band
localization suitable for data embedding. In [11], authors proposed a steganography method using encryption
technique and DWT. The DWT is applied on the cover image; Redundancy in the cover image is determined
using threshold calculation based on the image statistics. The secret message is partitioned and encrypted
using RC4 algorithm. A DWT is also applied to the encrypted secret message. The embedding process
implies a simple replacement of DWT coefficients of the encrypted secret message in the previously
specified DWT of the cover image. Although this method is secure, it requires hard ware resources such as
processor speed and memory. Although frequency domain embedding is considered a secure embedding
domain, the extracted secret data will suffer some degradation from the forward and reverse transform
operation.
LSB spatial embeding is considered as fragile hiding technique in terms of security, therefore it
requires a combination with encryption techniques as in [10] but, these techniques require additional
processing and computational expensive. Manipulating the embedding procedure will provide a massive
security feature to the steganography algorithm. In this paper a new embedding procedure is developed for
hiding secret data inside a cover media utilizing the Burrows Wheeler transform (BWT) for manipulating the
sequence of data embedding inside the cover media according to the output of BWT. The embedding
procedure is based on an exclusive–or (XOR) operation between a virtual bit and the LSB of the cover pixel
coefficient.
The rest of the paper organized as follows: Section two explains the conventional LSB method.
Section three presents the steps of Burrows wheeler transform. Section four illustrates the embedding and
extraction method Section five gives the results of some tests for the proposed steganography algorithm and
comparison with other algorithms. Section six states the conclusions and future developments.
2. CONVENTIONAL LEAST SIGNIFICANT BIT
The least significant bit technique is considered as a simple method used in the steganography field.
If a secret message is considered as a binary bit stream it can be embedded into a cover media, (i.e.….Audio,
image, and video) by replacing the LSB bits of the cover image with the secret bits [12]. The number of
replaced bits depends on the method that is utilized in the algorithm. Suppose an image of an 8 bit grayscale
have the following two pixel binary values p1=[10111000] and p2=[10110011] and the secret message that is
required to be embedded is S=[10] in the pixels, the stego output pixels are S1=[10111010] and
S2=[10110010]. The error between [p1, p2] and the [S1, S2] is imperceptible to the human visual system
(HVS). Increasing embedding capacity requires embedding more bits into the right side of the cover stream
pixels. However, increasing the embedding capacity results in a quality degrade of stego image. Therefore; a
trade-off between embedding payload and quality of stego image ought to be considerable. Although this
method is considered simple and fast, its main drawbacks are limited capacity and vulnerable security.

 ISSN: 2088-8708
Int J Elec & Comp Eng, Vol. 9, No. 1, February 2019 : 460 - 467
462
3. BURROWS WHEELER TRANSFORM
Lossless data compression is a technique used to compress data by manipulating data in a form that
can be reversed to retrieve its original value without losing any vital data [13]. The Burrows wheeler
transform (BWT) is a technique based on block sorting which blocks of data are manipulated to be
transformed to an easier form for processing. Since the technique is used in many practical applications
especially in lossless compression techniques, there have been many extensions of the original BWT [14].
Data can be transformed in the BWT domain and vice versa.
3.1. The forward burrows wheeler transform
The BWT is reversible transformation, therefore; original data is reconstructed to its original form.
In order to transform the data using BWT, the output of the transformation is limited to two factors; the
output (B), and the index (inx) where the location of the original input is located [13], [14]. Consider a one
dimension pixel sequence P=[3 2 5 3 1]. The sequence (p) is copied to a two column table (Index and
sequence) as shown in Table 1. The index (inx) exceeds (0……n-1) where n: is number of values in pixel.
The sequence is then left-cyclic permutated into each consequent index row. The rows are then sorted
lexicographically as in Table 2.
Table 1. Shows the Left Cyclic Permutation
Inx No. Left Cyclic Permutation Sequences
0 3 2 5 3 1
1 2 5 3 1 3
2 5 3 1 3 2
3 3 1 3 2 5
4 1 3 2 5 3
Table 2. Shows the Lexicographically Sorted Rows
Inx No. Left Cyclic Permutation Sequences
0 1 3 2 5 3
1 2 5 3 1 3
2 3 1 3 2 5
3 3 2 5 3 1
4 5 3 1 3 2
The search for the input one dimension row is executed in Table 2 and which the original pixel is
indexed. In this example the original row is indexed at 3. The BWT (B) for the input pixel p=[3 2 5 3 1] is
the last column B=[3 3 5 1 2] and inx=3. The proposed algorithm utilizes the transform to randomize the
blocks which will be embedded. Since the lexicographical sorting depends on the input vector to the
transform, unique outputs will be produced. This technique will blossom the security grade of the method.
4. PROPOSED IMAGE STEGANOGRAPHY
In this section, we describe a novel color image steganography. As mentioned in previous section,
the main drawbacks of the LSB technique are security compromise and limited embedding capacity. In order
to overcome these drawbacks, a new image steganography approach is proposed using generated virtual bit.
The method increases the embedding capacity of the conventional LSB without introducing noticeable
distortion to the stego-image. The method also possesses security features protecting secret data inside the
cover image using Burrows wheeler transform to enhance the embedded data security. Figure 2 shows the
block diagram of the method.
Figure 2. Block diagram of the proposed hiding algorithm

463
A RGB image is input to the algorithm, and separated into three layers R, G, and B. Layers of the
color image are divided into k*k non-overlapping blocks. Utilizing the Burrows wheeler permits the assign of
alphabat codes or decimal numbers to the blocks. In the proposed work, each block is assigned a one decimal
code. For example using a 64*64 block for 128*128 image will have a four (64*64) blocks per layer,
therefore; four decimal codes exist. The decimal numbers representing blocks are converted to a one
dimension vector using either Hilbert scanning, raster scanning, zig-zag scanning….etc. In this paper, the
Hillbert scanning path is used since this type of scanner scans every block of any square image size. The one
dimension vector is input to the Burrows wheeler transform for randomizing the block’s sequences to be
embedded with secret data, this will increase the security of the secret data since intruders are unaware of the
sequence of the blocks, size of blocks, type of the scanner used and type of block asigning. This prior
embedding process is considered as first line of defense for the embedded secret data. Blocks of one layer are
input to BWT and other two layers are embedded due to the same sequence of the first layer. This will reduce
time and complex processing required by BWT; in addition the algorithm does not require the inverse BWT
for extracting the data.
4.1. Hiding secret data
The method hides secret data in the spatial domain of a color image without utilizing any signal
transform. The embedding method assumes a virtual bit at the least significant bits of each cover image pixel
which is generated from a two MSB bit according to (1):
(Vir) = MSB - ⊕ MSB - (1)
where : Vir: virtual bit generated.
MSB: the ith
most significant bit of cover bit & i=8, 7…n.
The XOR between the LSB bit and the virtual bit according to (2) and the result (Res) is compared
to the secret data bit. If the result is equal to the data bit, the LSB bit is unchanged. Otherwise change the
LSB bit to the difference bit. In this operation one bit is embedded in the ith LSB bit.
(𝑅𝑒𝑠) = 𝐿𝑆𝐵 ⊕ 𝑉𝑖𝑟 (2)
Where: LSB: the ith
least significant bit of cover bit.
Embedding operation is continued for n bits. If n=8, then only one bit is embedded in the LSB of the
cover pixel. If n=7, then two bits are embedded in the cover pixel etc. For an illustration of the procedure for
the embedding process. Table 3. shows the embedding procedure for the secret data in the sixth bit of the
cover data. The table assumes that the virtual bit is 0. Table 4. shows the illustration for the same embedding
process of Table 3. for virtual bit of 1. It is clear from Table 3. and Table 4. that each embedded codeword
yields to a unique code word, therefore; a complete extraction of secret data is applicable.
Table 3. Shows the Embedding Scheme for Virtual Bit 0
No.
Embedding Bit 0
Cover bits
B5 B6 B7 B8
Sec. data Stego bits
1 0000
0 0000
1 0100
2 0001
0 0001
1 0101
3 0010
0 0010
1 0110
4 0011
0 0011
1 0111
5 0100
0 0000
1 0100
6 0101
0 0001
1 0101
7 0110
0 0010
1 0110
8 0111
0 0011
1 0111
Table 4. Shows the Embedding Scheme for Virtual Bit 1
No.
Embedding Bit 0
Cover bits
B5 B6 B7 B8
Sec. data Stego bits
1 0000
0 0100
1 0000
2 0001
0 0101
1 0001
3 0010
0 0110
1 0010
4 0011
0 0111
1 0011
5 0100
0 0100
1 0000
6 0101
0 0101
1 0001
7 0110
0 0110
1 0010
8 0111
0 0111
1 0011

 ISSN: 2088-8708
464
4.2. Extracting secret data
Most of the steganography algorithms require a lot of keys in order to extract the secret data from
the stego image. These keys are either embedded or sent as header information. In addition to keys, other
algorithms require knowledge of the cover image to reconstructed secret data. The proposed embedding
algorithm does not require the cover image to extract secret data making the algorithm as a blind
steganography algorithm and keys required are block sizes of the segmentation (k) of image and the number
of bits embedded per pixel (n). Figure 3 shows the secret extraction block diagram.
Figure 3. Block diagram secret data extraction
The stego image is separated into R, G and B layers. Each layer is divided into (k*k) blocks and
have the same assignment and type of scanner. The burrows wheeler transform is applied to find the
embedding sequence of the blocks. The extraction of the ith bit from the stego pixel is according to (3).
(𝑅𝑒𝑠) = 𝑆𝑡𝑒𝑔 ⊕ 𝑉𝑖𝑟 (3)
Steg: the ith
least significant bit of steg bit.
If the Res is equal to the ith
bit of the stego, the secret bit is 𝑅𝑒𝑠.
Otherwise: secret bit is the inversion of the ith
bit.
The blindness proposed method can extract the secret data without creating errors.
5. RESULT AND DISCUSSION
The basic performance of any stenographic algorithm is the stego-image quality and the embedding
capacity that the algorithm provides to carry secret data. The stego quality is gauged using the image metric
peak signal to noise ratio (PSNR) which is given by (4) [15] & (5) [15] for a square image. Some authors
determine the PSNR for each layer of the color image; therefore, the quality metric (PSNR) will be averaged
over the three layers:
𝑃𝑆𝑁𝑅 = 10 𝑙𝑜𝑔
( )
(4)
𝑀𝑆𝐸 = ∑ ∑ (𝚤𝑚 − 𝑖𝑚) (5)
where: ım and im are the stego and cover pixels respectively.
N: is the number of pixels in row or column of a square image.
A superior image steganography algorithm performance is remarked when the algorithm achieves a
high PSNR and embedding capacity. Moreover, the vital issue on message recovery side is the errorless
message retrieval from the cover media. The proposed algorithm is assessed with four images with various
number of embedding bits. The size of the image is 256*256*3 and 512*512*3 color images. Figure 4 shows
the original images of Baboon and Lena. Other image metric such as normalized cross correlation (NCC) is

465
utilized to illustrate the high correlation presented between the stego-image. The calculation of NCC for a
square image is given in (6).
 





N
I
N
J
imim
N
NCC
1 1
2^
2
1
(6)
As stated in section 4, the image is segmented into k*k blocks since the input images are a size of
512*512 then k=64, therefore; there will be 64 blocks of 64*64. All blocks will be assigned decimal codes
and converted to a one dimension block using Hilbert scanning. The one dimension code vector is
randomized according to Burrows wheeler transform. The algorithm will embed one bit (n=8) and two bits
(n=7) in the cover image, Table 5 illustrates the PSNR for various bit embedding payloads. The secret data
stream is generated by a psedu-random generator.
Table 5. Stego Image Quality versus Payload Covers Images
Image Name and size Number of bits Payload in bits PSNR in dB NCC
Babbon 512*512*3
n=8 786432 51.1412 0.999
n=7 1572864 44.1509 0.989
Lena 512*512*3
n=8 786432 51.1462 0.999
n=7 1572864 44.1400 0.988
Viptraffic 128*128*3
n=8 491520 51.1120 0.999
n=7 983040 44.1383 0.979
Tajmahal 256*256*3
n=8 196608 51.1418 0.999
n=7 393216 44.1490 0.989
Kola 256*256*3
n=8 196608 51.1487 0.999
n=7 393216 44.1322 0.979
a. original Baboon and lena b. Stego image Baboon and lena
Figure 4. Shows the original and stego color 512*512 Baboon and Lena images
With many experiments made on other various images the range of PSNR for present technique is
51.5-51.12 dB. When three bits are embedded in a cover pixel, the PSNR is in the range of 40-35dB with
payload 2359296 bits. Embedding more bits in the cover image will degrade the quality of the image with
respect to the original image. Stego images shown in Figure 4(b) are highly correlated to the original images
compared with the originals. Therefore, the algorithm preserves image quality for high payloads and is highly
secure for data embedding. The proposed algorithm compared with other proposed algorithms in [15] [16] is
less complex, in addition to the required processing time. However, the method presents better PSNR and high
capacity embedding. Table 6 and Table 7 show the comparison between these techniques respectively.
Table 6. Comparison of Stego Image Quality and Payload for 256*256 Image
Image
Reference [15] Proposed Method n=8
APSNR in dB Payload in bits PSNR in dB Payload in bits
Lena 50.6208 147456 51.1353 196608
Baboon 50.6167 147456 51.1412 196608
Tagmahal 50.5972 147456 51.1418 196608
Kola 50.6279 147456 51.1487 196608

 ISSN: 2088-8708
466
Table 7. Comparison of Stego Image Quality and Payload for 512*512 Image
Image
Reference [16] Proposed Method n=8
PSNR in dB Payload in bits PSNR in dB Payloadin bits
Lena 44.0886 786436 51.1462 786432
Baboon 44.0307 785110 51.1412 786432
Sailboat 43.9728 786832 51.1372 786432
Paper 44.0092 785392 51.1419 786432
6. CONCLUSION AND FUTURE WORK
The essential objective of a steganography algorithm is to address the security of information which
is achieved by concealing secret communication data inside the cover media. It is crucial for the sender to
conceal high amount of secret information inside a cover media and send the information over a public
communication channel. However, intruders can pirate the information inside the cover media unless the stego
image is highly correlated to the original media. A novel image steganography method is proposed in this
paper utilizing the XOR function with generated virtual bits and a pre-embedding method based on Burrows
wheeler transform. The proposed steganography method highly complies with steganography objectives.
Embedding the secret data using the proposed embedding strategy produces a stego image highly correlated to
the original cover, since the strategy saves the deference of the result of the XOR function. The virtual bit
generated from the MSB binary cover pixel is neither affected by the noise added on the stego-image nor other
compression process, therefore; the method permits a complete reconstruction of the secret data. Burrows
Wheeler transform sorts either decimal number coded blocks or alphabat chericteristecs codes that are sorted
lexicographically. Therefore; the assigned coded blocks of the cover image is redistributed in a random
manner before embedding. The procedure enhances security of embedding data. The proposed algorithm can
be enhanced using the output of digital scramblers for assigning the codes of cover image blocks since the
output of the digital scramblers alters with the number of shift registers, the initial digital bit value memorized
inside the shift registers and their arrangments. Nevertheless, the algorithm can be utilized in hiding images
inside a cover video.
REFERENCES
[1] G. S. Swain et al, “A Novel Approach to RGB Channel Based Image Steganography Technique,” Int. Arab J. e-
Technology, vol. 2, pp. 181–186, 2012.
[2] M.Hussain et al, “Recursive Information Hiding Scheme through LSB, PVD Shifit and MPE”, IETE Technical
Review, vol. 35, pp.53-63, 2017.
[3] Kumar R et al, “A New Image Steganography Technique Based on Similarity in Secret Message”, The Next
Generation Information Technology Summit (4th International Conference), pp. 376-37, 2013.
[4] M. S Arya et al, “Improved Capacity Image Steganography Algorithm using 16- Pixel Differencing with n-bit LSB
Substitution for RGB Images:, International Journal of Electrical and Computer Engineering (IJECE), Vol. 6,
2016, pp 2735-2741.
[5] Ammad UI Islam et al, “Improved Image Steganography Technique based on MSB using Bit Differencing” 6th
International conference on Innovative Computing Technology (INTECH), pp. 265-269, 2016.
[6] Weiqi Luo et al, “ Edge Adaptive Image Steganography Based on LSB Matching Revisited”, IEEE Transactions on
Information Forensics and Security, pp. 201-214, 2010.
[7] Liu J et al, “Least-Significant-Digit Steganography in Low Bitrate Speech”, IEEE International Conference on
Communications (ICC), pp. 1133–1137, 2012.
[8] Das. R et al, “A Novel Steganography Method for Image Based on Huffman Encoding”, Proceedings of 3rd
National Conference on Emerging Trends and Applications in Computer Science, (NCETACS), pp.14- 18, 2012.
[9] Masud Karim S. M et al, “A new approach for LSB based image steganography using secret Key”, 14th
International Conference on Computer Information Technology (ICCIT) IEEE; 2011 p. 286–291, 2011.
[10]Akhtar N et al, ”Enhancing the security and Quality of LSB based Image Steganography. 5th International
Conference on Computational Intelligence and Communication Networks, IEEE; pp. 385–90, 2013.
[11]A. Al-ataby et al, “Modified High Capacity Image Steganography Technique Based on Wavelet Transform”,
International Arab Journal Inf. Technol., pp. 358–64, 2010.
[12] C. K. Chan et al, “Hiding data in image by simple LSB substitution”, Pattern Recognition Journal, Vol. 37, pp. 469-
474, 2004.
[13]M. Burrows et al, “A Block-sorting Lossless Data Compression Algorithm”, Research Report 124, Digital System
Research Center, 1994.
[14]Ziya. Arnavut and Spyros S. Megluveras, “Block Sorting and Compression. Data Compression Conference”,
pp.181-190, 1997.
[15]Thanikaiselvan. V et al, “High Security Image Steganograohy Using IWT and Graph Theory”, IEEE International
Conference on Signal and Image Processing Applications (ICSIPA), pp. 337-342, 2013.

467
[16]Bhowmik. S et al, “A New Approach in Color Image Steganography with high level of Perceptibility and Security”,
International Conference on Intelligent Control Power and Instrumentation (ICICPI), pp. 283-286, 2016.
BIOGRAPHY OF AUTHOR
Mr. Ahmed Toman Thahab received his B.Sc in Electrical Enginnering from the University of
Babylon in 2006. He was enrolled in the Graduate school of the same universty. He received his
M.sc degree in the Communication and Electronic Engineering in 2011. His main study was in
video compression based on wavelet Transform.

A secure image steganography based on burrows wheeler transform and dynamic bit embedding

More Related Content

What's hot (16)

Similar to A secure image steganography based on burrows wheeler transform and dynamic bit embedding (20)

More from IJECEIAES (20)

Recently uploaded (20)

A secure image steganography based on burrows wheeler transform and dynamic bit embedding