Available online at www.sciencedirect.

com

ScienceDirect
ICT Express 4 (2018) 6–13
www.elsevier.com/locate/icte

Improved chaos-based video steganography using DNA alphabets

Nirmalya Kar a , ∗, Kaushik Mandal a , Baby Bhattacharya b
a CSE Department, NIT Agartala, Jirania, India
b Mathematics Department, NIT Agartala, Jirania, India

Received 1 December 2017; accepted 7 January 2018

Available online 1 February 2018

Abstract

DNA based steganography plays a vital role in the field of privacy and secure communication. Here, we propose a DNA properties-based
mechanism to send data hidden inside a video file. Initially, the video file is converted into image frames. Random frames are then selected and
data is hidden in these at random locations by using the Least Significant Bit substitution method. We analyze the proposed architecture in terms
of peak signal-to-noise ratio as well as mean squared error measured between the original and steganographic files averaged over all video frames.
The results show minimal degradation of the steganographic video file.
⃝c 2018 The Korean Institute of Communications Information Sciences. Publishing Services by Elsevier B.V. This is an open access article under
the CC BY-NC-ND license (http://creativecommons.org/licenses/by-nc-nd/4.0/).

Keywords: Chaotic map; DNA; Linear congruential generator; Video steganography; Least significant bit

1. Introduction where they increased the secret message size which would be
embedded into video frames. A 3 − 3 − 2 approach has been
Relaying data in a secure manner between two or more
used for embedding data into images which means they take
communication points has captured the imagination of mankind
the Least Significant Bit of all RGB color components (3-bits
for ages. Steganography is concerned with concealing the fact
of Red, 3-bits of Green, and 2-bits of Blue). As compared to this
that a secret message is being sent, as well as concealing
technique, a method proposed by Ramalingam et al. [8] using a
the contents of the message [1]. Any steganography method
modified LSB algorithm provides better efficiency. Cao, Zhang,
must satisfy two properties such as perceptual transparency
Xianfeng and Yu [9] proposed a novel video steganography
and high data rate. DNA steganography analyzes and utilizes
scheme based on motion vectors as carriers to embed the
DNA molecular techniques to conceal information [2]. It is
secret message through compression processing using H.264
a new territory that emerged after the discovery of the com-
video compression processing. The algorithm also uses the
putational prowess of DNA [3]. Moreover, the randomness of
principle of linear block codes to reduce motion vectors’ modi-
DNA strands makes them more suitable to encrypt data and
fication rates. A video steganography method proposed by Bin,
hide sensitive information [4]. The chaotic map is a collec-
Li-Yi and Wei-Dong [10] was based on motion vectors by using
tion of a noisy sequence. So, utilizing the chaotic map, we
matrix encoding. Kelash, Wahab and El-sayed [11] proposed
identify random pixel locations in an image [5,6]. Most video-
a steganography algorithm to embed data into video clips
based steganography algorithms are developed based on the
directly, where each pixel in each video frame is divided into
relationships between frames. Eltahir et al. [7] presented a
two parts, where the number of bits which will be embedded in
video steganography based on the Least Significant Bit (LSB)
the right part are counted in the left part of the pixel based on a
∗ Corresponding author. color histogram.
E-mail addresses: nirmalya@ieee.org (N. Kar), baby.math@nita.ac.in (K. In our work, we use an .avi video file as the input video. This
Mandal).
Peer review under responsibility of The Korean Institute of Communica- video will be considered as the cover medium. We will collect
tions Information Sciences. all frames from the video file and store these in a folder. After
https://doi.org/10.1016/j.icte.2018.01.003
2405-9595/⃝ c 2018 The Korean Institute of Communications Information Sciences. Publishing Services by Elsevier B.V. This is an open access article under the
CC BY-NC-ND license (http://creativecommons.org/licenses/by-nc-nd/4.0/).
 N. Kar et al. / ICT Express 4 (2018) 6–13
Fig. 1. Overview of proposed architecture.
extraction, all frames will be converted into a set of .png frames.
Fig. 1 shows the entire procedure of our proposed approach.
2. Background description
The methodologies mentioned here are required to establish
our proposed approach. This paper does not make use of
biological properties; instead, it uses DNA sequences properties
to achieve its goal.
2.1. DNA alphabets
A DNA sequence consists of four alphabets: A, C, G, and
T , which can be represented by two bits. To demonstrate this,
assume, A = 00, T = 01, C = 10 and G = 11. Then, each
symbol of the language is encoded by a triplet of the DNA
base elements in such a manner that the natural structure of
the amino acids remains the same. There are three standard
methods of steganography that have been extensively practiced,
namely the insertion method, the complementary pair method,
and the substitution method. The broad intent of such methods
is to choose a reference DNA sequence S covertly, where only
the transmitter and the receiver are aware of and incorporate the
secret message P into S to obtain a new DNA sequence S ′ . The
second party (receiver) will receive S ′ , identify it, extract the
hidden message P, and ignore all other parts of the sequence
using the appropriate inverse algorithm. The randomness of
DNA strands generation makes it more suitable to encrypt data
and hide sensitive information [12,13].
2.2. DNA properties
PCR Technology, also known as “polymerase chain reac-
tion”, is normally used to open up the DNA, as it is extremely
7
difficult to control small measures of DNA. In practice, DNA
intensification methods incorporate cloning. The enhancement
productivity of PCR is high and can enhance an extensive
number of selected DNA in a brief time frame. In addition,
PCR will accomplish the enhancement by utilizing regular nu-
cleotide atoms [14]. DNA also has cutting properties. Restric-
tion endonucleases cut DNA in the manner all endonucleases
do, by cleaving the phosphodiester bond between an adjacent
phosphate and deoxyribose group in the phosphate backbone of
DNA. A specific region of the endonuclease called the active
site does this by causing the side chain of a certain amino acid
in the enzyme to form a temporary chemical bond with the
phosphate group. This breaks the existing bond between the
deoxyribose sugar and the phosphate, thereby cutting the DNA
chain at a location within a DNA molecule [15].
2.3. Linear congruential generator
A linear congruential generator (LCG) is an algorithm that
yields a sequence of pseudorandomized numbers calculated
with a discontinuous piecewise linear equation. The method
represents one of the oldest and best known pseudorandom
number generator algorithm [16]. The generator is designed by
the recurrence relation:
X n+1 = (a X n + c) mod m. (1)
Above, X n is the sequence of pseudorandom values and m is
the modulus; 0 ≤ a < m, a is the multiplier; 0 ≤ c < m, and c
is the incremental; 0 ≤ X 0 < m, X 0 is the seed value.
2.4. Chaotic map
In the proposed architecture, we use the Burger chaotic
map which is a deterministic nonlinear dynamical system.
8 N. Kar et al. / ICT Express 4 (2018) 6–13
The Burger 2-D chaotic map has been picked in view of
its disordered sequence [17]. It is a discrete-time dynamical
system given by:
X n+1 = X n2 − Yn2 + a X n + bYn (2)
Yn+1 = 2X n Yn − Yn2 + cX n + dYn . (3)
In our approach, the chaotic map performs an important role.
After selecting the individual frame from the frame pool, we
need to embed data in the selected frame. Here we introduce
randomness in selecting the pixel location for data hiding. We
place the Burger chaotic map in the selected frame and in
return, it will plot many pixel points in the frame. Out of those
points we choose only a few points to exactly embed data. In
this manner, we introduce a few extra layers of security in our
proposed architecture.
3. Proposed approach
3.1. Frame selection mechanism
The frame selection mechanism is an important part of our
approach (Fig. 2). Here, we have taken a sample DNA strand
(e.g. ACTGGTCACGTAGTAG) as input. Initially we perform
three different operations on the selected DNA strand.
1. Convert into binary and check its length.
2. Check number of codons in the DNA strand.
3. Convert it into its corresponding decimal values.
Taking decimal values as initial values, number of codons
as the multiplier, number of digits into decimal values as a
constant, and a user given key value which is at least 10
digits, we have generated a pseudorandom number by placing
all those values in the linear congruential generator. Now, we
select a random number and check its length in binary with the
length of the input DNA strand. If the length of DNA strand is
greater than the length of the random number, we apply DNA
cutting properties. Otherwise, we use polymerization properties
i.e., make a replica of the key DNA strand and compare it
with the random number until both lengths are the same. Our
intention is to select the same length codon from the key DNA
strand and compare it to the length of the random number.
Next, we perform an XOR operation and begin key generation.
The block diagram in Fig. 2 explains the entire procedure for
selection of frames and generation of the key.
3.2. Data embedding mechanism
After selecting the frame, the next major task is to embed
the data into it. But before embedding, we again introduce
randomization for better security. Instead of using the entire
frame for data hiding, we choose selected pixel points where
data will be embedded. We use the Burger chaotic map for pixel
plotting over the cover image. The initial value of the chaotic
map will always change for each frame. So, pixel plotting
over every frame will also always change so that an attacker
cannot identify a pattern. Again, we are not embedding the
data in all the pixel points received from the chaotic map.
We introduce random numbers from the linear congruential
generator for randomizing pixel selection as a more advanced
form of security. We know that each pixel is an RGB picture
i.e., size of every pixel is 3 ∗ 8 = 24 bits (8 bits—Red, 8 bits—
Green, 8 bits—Blue). Here we embed data into eight minimum
critical bits (3 bits from red, 3 bits from green, 2 bits from
blue) out of 24 bits for concealment reasons. The reason for
taking the blue color as 2-bits is because the human visual
system (HVS) is more sensitive to the blue color than red and
green. When installing an 8-bit message, we use a permutation
cipher depending on the key value for the arrangement of
RGBRGBGR ((b1;r1) (b2;g1) (b3;b1) (b4;r2) (b5;g2) (b6;b2)
(b7;g3) (b8;r3)), where (bi;rj) = (bit;pixel). Fig. 3 shows the
entire procedure of data embedding.
3.3. Extraction module
In the data extraction phase, the original message will be
retrieved from the stego video. To achieve this, we must isolate
the frames from the stego video. In the extraction procedure, the
same parameters and key value are used, as that information
is kept secret in our method. The data extraction steps are as
follows:
1. Convert the stego video into frames.
2. Find the stego-frame based on the frame selection algo-
rithm.
3. Generate the chaotic map based on the initial value.
4. Find pixel points based on the chaotic map in the stego-
frame.
5. Retrieve the bits in the order of 3, 3, 2 respectively, from
each pixel.
6. Reconstruct the secret information.
4. Results and discussion
We use the sky .avi as cover video with duration of 7 s and
file size of 12 Mb. Total number of frames in the input video
file = 114. The security key of the proposed algorithm is a
sequence of a DNA strand which is user given. Here, we are not
considering the length of the DNA strand as we will use DNA
cutting and polymerization properties. A sample of the input
DNA strand is ‘ACTGGTCACGTAGTAG’. We also introduce
the linear congruential generator for generating the random
number. Below we illustrate our approach with an example.
The initial values of LCG are taken as X n = 6685113491,
a = 16, c = 10, and m = 6421329685. The linear congruential
generator will generate a series of random numbers. In our
example, the selected random number = 4220540902.2, P =
32, Q = 34. So, 17 codons are selected starting from fourth
codon. The selected codon = GGTCACGTAGTAGACTG, and
the resultant value = 13474497729. The initial value of chaotic
map = 0.97729, frame number = 57, and the key for the Caesar
cipher operation = 13474497729, mod 26 = 17.
Plain Text: The concept of using DNA computing in the
fields of cryptography and Steganography has been identified
 N. Kar et al. / ICT Express 4 (2018) 6–13 9

Fig. 2. Frame selection procedure.

Fig. 3. Data embedding procedure. (For interpretation of the references to color in this figure legend, the reader is referred to the web version of this article.).

as a possible technology that may bring forward a new poten-
tial for unbreakable algorithms. In this paper, we proposed a
mechanism by which we can send data hidden in a video file.
Encrypted Text: Max vhgvxim hy nlbgz WGT vhfinmbgz
bg max ybxewl hy vkrimhzktiar tgw Lmxztghzktiar atl uxxg
bwxgmbybxw tl t ihllbuex mxvaghehzr matm ftr ukbgz yhkp-
tkw t gxp ahix yhk ngukxtdtuex tezhkbmafi. Bg mabl itixk, px
ikhihlxw t fxvatgblf ur pabva px vtg lxgw wtmt abwwxg bgmh
t obwxh ybex.
After obtaining the initial value for the chaotic map, we plot
Burger’s chaotic map over the selected frame. X b = 0.97729;
Yb = 0.97729. As the initial value of chaotic map changes for
each frame, data will embed in different locations of the frame.
Fig. 4(a), (b), (c) and (d) show plots of the chaotic map in
a particular frame and the selected points for data embedding.
After collecting the 8-bit RGB pixels, we use a permutation
cipher for the arrangement of RGBRGBGR ((b1;r1) (b2;g1)
(b3;b1) (b4;r2) (b5;g2) (b6;b2) (b7;g3) (b8;r3) where (bi;rj) =
10 N. Kar et al. / ICT Express 4 (2018) 6–13
Fig. 4. (a) All points plotted over selected frame. (b) All points of Burger Chaotic Map. (c) Selected points plotted over selected frame. (d) Selected points of Burger
Chaotic Map.
(bit;pixel)), Key value = 5, and embedded channel order =
GRBGRGBR.
The perceptual quality of the cover video frame and the stego
video frame has been evaluated using two quality evaluation
metrics, Mean Squared Error and Peak Signal to Noise Ratio.
There are several attacks possible based on statistical properties
on LSB embedding [18]. An initial approach was proposed by
Westfield and Pfitzmann named the Chi-square attack. The
Chi-square attack was basically developed for sequential at-
tack, but not for random embedding unless approximately all
samples have been used for carrying message bits. So, the
Chi-square attack is unable to reveal any information about
secret data in our proposed mechanism. We have also analyzed
our algorithm based on histogram analysis.
4.1. Pixel substitution
An encryption algorithm with a good diffusion property has
the ability of resisting a chosen plaintext attack. The diffusion
property indicates that encrypting two plain texts with a slight
difference using the same security key will yield totally dif-
ferent encryption results. In this proposed algorithm, selection
of pixels and RGB channel has also been performed randomly.
After channel selection, we embed cipher text into it. From
all points, we selected 30% pixels where data was embedded.
Each pixel was selected randomly based on random numbers
generated by a linear congruential generator and the RGB
channel was selected by a permutation cipher. After embedding
text into frames, we observed no obvious visual changes.
4.2. Peak Signal to Noise Ratio (PSNR)
PSNR and MSE are standard parameters used to determine
the superiority of two images in a steganographic system [19].
The PSNR determines the similarity between two images and
is reciprocal to MSE. PSNR is used to decide the similarity
between the cover frame and the resulting stego frame. Because
PSNR calculates the level of similarity between the cover frame
 N. Kar et al. / ICT Express 4 (2018) 6–13 11

Table 2
Results obtained from proposed algorithm and LSB techniques [8] for PSNR.
Name of video file Results of the proposed Results obtained
mechanism using [8]
PSNR Payload PSNR Payload
Sky.avi 51.51 1.56 48.56 1
Rhinos.avi 52.24 1.23 52.34 1
Count.avi 46.21 1.78 44.56 1

Fig. 6. MSE comparison between original and stego frames.

Table 3
Results obtained from proposed algorithm and LSB techniques [8] for MSE.
Name of video file Results of the proposed Results obtained
mechanism using [8]
MSE Payload MSE Payload
Sky.avi 0.49 1.56 0.42 1
Fig. 5. PSNR comparison between original and stego frames. Rhinos.avi 0.53 1.23 0.52 1
Count.avi 0.41 1.78 0.38 1

and the stego frame, higher values of PSNR indicate better

results. PSNR is calculated by:
M AX 2
P S N R = 10 ∗ log10 . (4)
MSE
Above, MAX = maximum pixel value (in the RGB image, one
pixel value is 24 bits (Red: 8 bits, Green: 8 bits, Blue: 8 bits)).
Here PSNR is calculated for all three channels independently,
so the maximum value is (28 − 1) = 255.
We compared the results of the proposed algorithm with
Video Steganography using the Modified LSB Algorithm pro-
posed by Ramalingam [8]. Here, for comparison we took three
videos with the same resolution and frame rate as the author
had taken. Details of the video files are given in Table 1.
In Table 2, we compare the PSNR and payload between
the proposed algorithm and modified LSB techniques [8]. As
higher PSNR indicates that the difference between the original
and stego videos is lower, we can state that the proposed
algorithm generated better results overall (see Fig. 5).
4.3. Mean Squared Error (MSE)
MSE is used to show the differences between the cover
video frame and the stego video frame. This is accomplished
by taking the sum of the differences between the corresponding
pixel values of both frames and then dividing the sum by
the size of the frame. Because MSE calculates the degree of
dissimilarities between cover and stego frames, lower the value
of MSE the better the result. Mean squared error is computed
by the following:
h ∑
∑ h
M S Ee = (Cicj − Sicj ). (5)
n=1 n=1
Above, c = color component, w = width of image, h =
height of image, and Cicj and Sicj are two bytes of the color c
at the location (i, j) from the original frame and stego frame,
respectively (see Fig. 6).
MSE values obtained from video by the proposed algorithm
and modified LSB technique [8] are shown in Table 3.
In this case, the MSE value is slightly higher than the
modified LSB technique [8], but the payload is always greater,
so in terms of payload, the proposed algorithm is resulted better.
4.4. Histogram difference
The distribution of intensities in the video frame can be
shown by the simple histogram of the frame. The histogram
plots the frequency next to the bins. The bin represents the
pixel value and the frequency represents the number of times
the pixel value appeared in the video frame. The range of
the pixel value is 0–255. The histogram of the original video
frame and its corresponding stego video frame indicates the
amount of difference between the two frames and it also helps
to decide the stego video quality. In Fig. 7, after embedding the
message into the frames the histogram of the stego video frame
12 N. Kar et al. / ICT Express 4 (2018) 6–13

Table 1
Cover video file details.
Sl. No Cover video file information
Name of video file Resolution Frame/Sec No of frames
1 Sky.avi 1280 ∗ 720 25 114
2 Rhinos.avi 320 ∗ 240 15 331
3 Count.avi 640 ∗ 320 25 2893

(a) Original frame 17. (b) Stego frame 17.

(c) Original frame 93. (d) Stego frame 93.

Fig. 7. Histogram comparison of frames 17 and 93 of cover file rhinos.avi and stego file rhinosout.avi.

shows some changes. But the effect of these changes can be 4.5. Pixel correlation
watered down by abrupt scene changes. If the scene changes
An image generally has high data redundancy and thus its
in a video frame abruptly, the histogram of the frame where
pixels have high correlations with their neighboring pixels. A
the scene has changed also shows abrupt changes compared good image encryption algorithm should have the ability to
to the histogram of the frames where the scene has changed break these correlations. Mathematically, data correlation is
gradually. If an attacker obtains a stego frame, he cannot find defined by Eq.:
the original frame with which he can compare the differences. ∑ (i − µi) ( j − µj) p(i, j)
Corr = , (6)
As a result, changes occurring in the stego cover video frame ij
σi σ j
compared to the original cover video frame will not reveal much where i and j are the two data sequences, µ is the mean
information [20]. value, and σ is the standard deviation. If two sequences i and
 N. Kar et al. / ICT Express 4 (2018) 6–13
j have high correlations, their correlation value is close to 1.
Otherwise, it is close to 0. In our test, we took the cover frame
and stego frame (embedded frame) and checked the correlation
between them. The results of the correlation value of the cover
image = 1 and the stego image = 0. Hence, we can verify that
the encrypted image by this proposed algorithm has extremely
low correlation.
5. Conclusion
In this paper, we focus on video based steganography, as
video formats are widely used on the internet and also in mobile
systems. Enhanced LSB calculations can without much of a
stretch be actualized and do not visually corrupt the video to
the point of being indiscernible. Here, we are proposing a DNA-
based video steganography which involves various properties of
DNA such as DNA polymeric chain reaction and DNA cutting
properties with a linear congruential generator and a Burger
chaotic map. We made the proposed algorithm robust by use of
randomization. We randomized selection of frames, selection
of pixels for embedding data, and for the channel (RGB). This
algorithm is considered a high embedding efficiency algorithm
due to the low modification on the data that retains the same
quality of video. Qualitative analysis has been performed based
on PSNR and MSE. We have also checked histogram differ-
ences and pixel correlation of original and stego videos. All
obtained experimental results indicate a high standard and a
good visual quality for the stego video, which is not likely to
be suspicious.
Conflict of interest
The authors declare that there is no conflict of interest in this
paper.
References
[1] N. Provos, P. Honeyman, Hide and seek: An introduction to steganogra-
phy, IEEE Secur. Privacy 99 (3) (2003) 32–44.
[2] A. Gehani, T. LaBean, J. Reif, Dna-based cryptography, Lecture Notes in
Comput. Sci. 2950 (2003) 167–188.
[3] Khalifa Amal, Ahmed Atito, High-capacity DNA-based steganography,
in: 2012 8th International Conference on Informatics and Systems (IN-
FOS), IEEE, 2012, p. BIO-76.
13
[4] P. Das, S. Deb, N. Kar, B. Bhattacharya, An Improved DNA based dual
cover steganography, Procedia Comput. Sci. 46 (2015) 604–611.
[5] S. Li, G. Chen, X. Zheng, Chaos-based encryption for digital image
and video, Multimedia Encryption and Authentication Techniques and
Applications 129.
[6] P. Das, N. Kar, A dna based image steganography using 2d chaotic map,
in: International Conference on Electronics and Communication Systems
(ICECS), IEEE, 2014, pp. 1–5.
[7] M.E. Eltahir, L.M. Kiah, B.B. Zaidan, A.A. Zaidan, High rate video
streaming steganography, in: International Conference on Information
Management and Engineering, in: ICIME’09, IEEE, 2009, pp. 550–553.
[8] M. Ramalingam, Stego machine-video steganography using modified lsb
algorithm, World Acad. Sci. Eng. Technol. 74 (2011) 502–505.
[9] Y. Cao, H. Zhang, X. Zhao, H. Yu, Video steganography based on
optimized motion estimation perturbation, in: Proceedings of the 3rd
ACM Workshop on Information Hiding and Multimedia Security, ACM,
2015, pp. 25–31.
[10] Z. Li-Yi, Z. Wei-Dong, et al., A novel steganography algorithm based on
motion vector and matrix encoding, in: IEEE 3rd International Confer-
ence on Communication Software and Networks (ICCSN), IEEE, 2011,
pp. 406–409.
[11] H.M. Kelash, O.F.A. Wahab, O.A. Elshakankiry, H.S. El-sayed, Hiding
data in video sequences using steganography algorithms, in: International
Conference on ICT Convergence (ICTC), IEEE, 2013, pp. 353–358.
[12] B. Shimanovsky, J. Feng, M. Potkonjak, Hiding data in dna, in: Interna-
tional Workshop on Information Hiding, Springer, 2002, pp. 373–386.
[13] H.J. Shiu, Ka-Lok Ng, Jywe-Fei Fang, Richard C.T. Lee, Chien-Hung
Huang, Data hiding methods based upon DNA sequences, Inform. Sci.
180 (11) (2010) 2196–2208.
[14] V.I. Risca, DNA-based steganography, Cryptologia 25 (1) (2001) 37–49.
[15] M. Saeb, E. El-Abd, M.E. El-Zanaty, On covert data communication chan-
nels employing dna recombinant and mutagenesis-based steganographic
techniques, in: Proceedings of the 2007 annual International Conference
on Computer Engineering and Applications, 2007.
[16] B. Ramalingam, R. Amirtharaj, J.B. Balaguru R, Lcc-lsb-fpga stego - a
reconfigurable security, J. Appl. Sci. 14 (2014) 2139–2148.
[17] M. Jain, A. Kumar, R.C. Choudhary, Improved diagonal queue medical
image steganography using chaos theory, lfsr, and rabin cryptosystem,
Brain Inform. 4 (2) (2017) 95–106.
[18] S. Dumitrescu, X. Wu, Steganalysis of lsb embedding in multimedia sig-
nals, in: Proceedings: 2002 IEEE International Conference on Multimedia
and Expo, in: ICME’02, vol. 1, IEEE, 2002, pp. 581–584.
[19] A. Almohammad, G. Ghinea, Stego image quality and the reliability of
psnr, in: 2nd International Conference on Image Processing Theory Tools
and Applications (IPTA), IEEE, 2010, pp. 215–220.
[20] B. Choudhury, R. Das, A. Baruah, A novel steganalysis method based
on histogram analysis, in: Advanced Computer and Communication En-
gineering Technology, Springer, 2015, pp. 779–789.