10.1007@s11042 020 08951 1
10.1007@s11042 020 08951 1
10.1007@s11042 020 08951 1
https://doi.org/10.1007/s11042-020-08951-1
Shaswata Saha 1 & Anuran Chakraborty 1 & Agneet Chatterjee 1 & Souvik Dhargupta 1 &
Sudipta Kr Ghosal 2 & Ram Sarkar 1
Abstract
Steganography is the method of hiding secret message information in various cover
media like text, image, audio, video and others without raising suspicion to intruders
about the existence of any such information. Exploiting Modification Direction (EMD)
based image steganography is a steganographic technique designed by Zhang and Wang
in 2006 that yields minute distortion of the cover image. In this paper, an Extended EMD
based steganography using Hashed-Weightage Array abbreviated as EEMDHW has been
proposed. 2KN-ary numbers are embedded in every K pixels of the cover image, where N
denotes the number of bits embedded in each pixel. Thus, the payload is purely variable
in this technique. Embedding is performed using dynamic weightage array. This array is
made pseudo random by applying eliminative hashing technique on the message pixels.
Experimental results show that the technique is superior to other state-of-the-art tech-
niques in terms of embedding capacity (payload) and has reduced quality distortion of
cover image. Steganalysis using RS attack shows that the embedding done by the
proposed method cannot be detected upto 3 bits per pixel (bpp) embedding rate.
1 Introduction
Since ancient times, humans have devised different ways to hide information. The word
‘steganography’ means “covered writing” [1]. The technique of hiding secret messages on
the scalp of heads, used in ancient Greece, inspired researchers working in the field of data
* Sudipta Kr Ghosal
[email protected]
1
Department of Computer Science & Engineering, Jadavpur University, Kolkata 700032, India
2
Department of Computer Science & Technology, Nalhati Government Polytechnic, Birbhum, Pin,
Nalhati 731243, India
Multimedia Tools and Applications
security and they invented steganography. One of the most popular forms of steganography is
image steganography. With the ever-increasing risk of data leaking and information stealing,
the need for information security has also been increased. Thus, more and more image
steganography techniques have been introduced in the literature.
Least Significant Bit (LSB) [2] is one of the oldest and simplest forms of steganography
techniques. This technique hides each bit of the message in the LSB of every pixel of the cover
image. It introduces a maximum distortion of only ±1 in each pixel in the cover image.
However, this technique cannot hide large amount of data. Thus, more efforts were made to
hide more data. In 2001, Wang et al. [3] devised a novel technique in which they hid the
message in the k-LSBs using Genetic Algorithm (GA). However, the usage of GA to generate
optimal solution made the process time consuming. Few years later, Chang [4] et al. proposed
a technique in which they hid data using codeword grouping. The codeword was generated
using palette generation. This codeword was modified to hide secret message. This was a
robust technique. However, it introduced a high distortion in image quality.
In 2006, Zhang et al. devised the Exploiting Modification Direction (EMD) [5] technique.
In this technique, a (2n + 1) − ary digit is hidden in n pixels, and only 1 out of n pixels would
be distorted. Thus, distortion of the image quality would be less but this can be achieved with a
lower payload. In 2016, Shen et al. proposed a technique [6] using image interpolation. The
pixel value difference between a target pixel and its neighboring pixel was used to determine
the number of bits to be embedded in the target pixel. Thus, the image quality distortion
decreased in comparison with previous techniques. In 2018, Saha et al. introduced an EMD
technique [7] using a weightage array in which the payload could be varied according to the
length of the message at reduced quality distortion. However, once the number of pixels in
each pixel group could be known, the weightage array could be easily detected. Also, the
introduction of variable payload decreased the embedding capacity.
In this paper, we have proposed an EMD based steganographic technique that is capable of
hiding data at payload as high as 4 bpp. It also achieves higher security than many state-of-the-
art techniques by decreasing distortion of the cover image and using a pseudo-random
weightage array and randomness factor K. The proposed method converts the message into
2KN-ary numbers. Then it uses a pseudo-random weightage array to embed the converted
numbers inside the cover image. The weightage array is made pseudo-random using the
eliminative hashing technique which is explained in section 3.1.2 in this paper. The extraction
process is done using the same weightage array. Once the converted 2KN-ary numbers are
extracted from the stego image, they are converted to message bits accordingly.
This paper is organized in the following manner. In section 2, it discusses about some
methods on steganography devised in the past. It then introduces our proposed technique that
overcomes the shortfalls of the previous techniques in section 3. In section 4, it analyses the
results of the proposed technique and compares its performance with some state-of-the-art
methods [2, 5–10]. Finally, it checks the robustness and the security of the proposed method in
section 5. From sections 4 and 5, it can be observed that the proposed technique outperforms
the contemporary steganographic techniques and is also immune to RS attacks.
2 Related work
Steganography has had a multitude of implementations which range from the spatial domain to
the transform domain [11, 12]. In the transform domain exist techniques which are based on
Multimedia Tools and Applications
Fast Fourier Transform [13], Curvelet Transform [14], Wavelet transform [15], VQ-
compressed indices [16] etc. One of the simplest methods employed in the spatial domain is
the LSB technique [10] whereby the LSB’s of the pixels of the cover image were used to
embed the hidden message. The LSB technique does not utilize the differing degrees of
distortion that pixels can exhibit and does not generate a high Peak Signal to Noise Ratio
(PSNR). Researchers have also used the LSB technique by combining it with other concepts
like cryptographic key generation [17, 18]. Chatterjee et al. [19] proposed an OCR-based LSB
steganography technique where, instead of embedding the secret information directly in a
cover image, the OCR features of the secret information are embedded. The message is
retrieved by extracting the OCR features from the stego image and feeding them to a pretrained
OCR model. Several multilevel security mechanisms have also been proposed in the recent
past [20, 21]. However, it is found that most of these techniques have high execution time, low
embedding capacity and reduced security.
To compensate for the same, Zhang et al. [5] developed the EMD technique which utilized
different modification directions to represent varying data leading to efficient embedding of
data. In this technique, the payload does not depend on the image i.e., for an image having the
same number of pixels, the payload is equivalent. Furthermore, in the above method, only one
pixel in a group of n pixels was required to be increased or decreased by 1. As a result, the
embedding capacity of EMD technique was significantly less as compared to other techniques.
To ameliorate the shortcomings, Chang et al. [22] developed a Sudoku based reference table
method to guide the modifications of the cover pixels. Furthermore, Mao et al. [23] contrasted
trivial reference-table based steganography by considering non-uniformity of secret digits. Yin
et al. [24] developed a second-order steganographic algorithm to conceal two secret digits in
the corresponding pixel pair. Also, in 2007, Chang et al. [25] developed a two-step EMD
technique where they grouped pixels based on their grayscale values followed by EMD which
enabled them to embed the secret message in n pixels twice at a time.
Also, an improved EMD technique using the concepts of modulus was developed in 2009
by Jung et al. [26] which increased the bpp to 2.5. But another problem that plagued EMD was
that it was much less secure than its contemporaries. To tackle this, [27] used the Hungarian
algorithm to reorganize the secret sequences to reduce the image distortion as compared to the
original EMD technique, whereby they achieved a maximum PSNR of 52.1 dB. In [28], the
authors propose a method to have better stego image quality by reducing message size. The
authors in [29] have proposed a number theory based steganography approach using Fibonacci
sequence. However, the maximum embedding capacity of this technique as given in the paper
is 0.9 bpp and the maximum PSNR is 41.035 dB at that payload which is far less than most
state-of-the-art techniques.
In 2019, Setiadi has presented an Edge Area Dilation (EAD) technique [30] to increase the
payload in LSB embedding. This technique embeds messages in the dilated edge areas of the
cover image using LSB embedding. The method increases the embedding capacity of LSB
embedding to 1.42 bpp i.e. by 18.65% and achieves an average PSNR of 47.59 dB. However,
it does not comment on its feasibility at higher payloads. Yaseen et al. [31] devised another
edge-based hybrid steganographic technique. The authors encrypted the secret image before
embedding them in the cover image for better security. However, the method cannot achieve
payloads higher than 3 bpp because it embeds 3 bits per edge pixel and 2 bits per smooth pixel.
Leng et al. [32] have proposed a Generalized EMD (GEMD) technique that uses an n-
dimensional hypercube around a group of n pixels to achieve higher payload up to 4.75 bpp.
They also propose a minimum distortion process to keep the PSNR above 30 dB. Abd El-Latif
Multimedia Tools and Applications
et al. [33] proposed an image steganography based on quantum walks. At first, the method
generates Quantum Walk Substitution Boxes (QW S-box). Then it embeds data into the cover
image using these QW S-boxes.
Compared to the standard LSB or EMD techniques, edge detection based steganography
techniques [34] have also been proposed which hide more information in the edge pixels
compared to those in homogeneous regions. This is because human eye is less prone to detect
changes in the edge regions than the homogeneous regions. In [35], the researchers have
performed steganography on RGB images. Steganography has also been used by hiding data
in compressed archives [36].
In this paper, we draw inspiration from the EMD technique aiming to maintain a low image
distortion with a high PSNR utilizing the security aspect of the weightage array. We also
attempt to do away with the payload problem of EMD and successfully surpass the 3 bpp mark
without any reduction in the image quality.
3 Proposed technique
An extended EMD based steganography method has been proposed to fabricate secret
message. In this method, the cover image is partitioned into groups of non-overlapping pixels
and then message bits are embedded into them. The method proposed here has higher and
variable payload with acceptable image quality distortion which we discuss later in the result
section.
This technique requires following two parameters:
& N = Number of bits of message to be embedded in each pixel. This makes payload of the
technique variable.
& K = Number of pixels in each group, i.e., the randomness factor.
The proposed technique has been implemented using two major phases: Embedding and
Extraction.
The Embedding phase is further divided into two segments: Pre-processing of message
information and Embedding data in the cover image.
After pre-processing, the cover image is reshaped into a row vector named as C. Then each digit of
M’ is hidden in non-overlapping units of K pixels of C. This is carried out using a weightage array
w, where w is initialized as (1, 2, 3, …, K). This array is made pseudo-dynamic throughout the
process using eliminative hashing technique with the help of our randomness factor, K. In
eliminative hashing, the digit of M’ that is just embedded decides the array w for the next digit
to be embedded.
For example, let K = 3 and w = (1, 2, 3) at the beginning. Let d = 155 be the digit that is just
embedded. Now, determining w for the next digit,
w’3 ¼ 3
Thus, the new w is (2, 1, 3). While embedding, it is also checked whether addition or
subtraction would introduce less change in the pixel value. This is done to decrease distortion
of cover image and increase the PSNR value. On repeating these steps, the stego image is
finally produced. Following is the pseudo code for the embedding algorithm.
Algorithm:
Input: Cover image C, changed message array M’.
Output: Stego image S.
Parameters: N = payload, K = randomness factor.
Step 1: Initialise w with (1, 2, 3, …, K) and pow with 2N.
Step 2: Set unit of cover pixels p = (p1, p2, p3, …, pK)and unit of stego pixels be q = (q1, q2,
q3, …, qK)
Step 3: Take a digit d of M′ and convert it to modulo-pow form. Let its digits be mi.
Step 4: For every miin d we calculate index = wi and diff = mod (pindex, pow) – mi
Step 5: If diff < (pow/2), qindexis set to pwi di f f otherwise it is set to pwi di f f þ pow.
Step 6: w for next d in M′ is calculated by eliminative hashing technique
Step 7: If there exists more d in M′, goto step 2, otherwise stop embedding.
Step 8: Finally, the stego image S is obtained.
Here, Fig. 1 shows the flowchart of the embedding algorithm.
At the receiver end, the message information is retrieved using the stego image. Similar to the
embedding algorithm, the extraction algorithm also has two segments: Extracting data from
the stego image and Retrieving actual information from the extracted data.
The stego image is reshaped into a row vector and decomposed into non-overlapping units of
K pixels. The first 2KN − ary digit is extracted from the stego image using the initial weightage
Multimedia Tools and Applications
array w, i.e. (1, 2, 3, …, K). This digit is used to change w and retrieves the next message digit.
This process goes on until all the 2KN − ary digits have been extracted. These digits are then
converted into their 8-bit equivalent to finally recreate the message information, i.e. M. Given
below is the algorithm for extracting data from the stego image in modulo- 2KNform.
Algorithm:
Input: Stego image, S.
Output: Message array containing 2KN − ary digits, M”.
Parameters: N = payload, K = randomness factor.
Step 1: Initialize w with (1, 2, 3, …, K), pow _ modwith 2N,pow with 1, i and d with 0 and
pos with 1.
Step 2: Take a pixel q of S.
Step 3: Increment i by 1 and increment d bymod(q, pow _ mod) ∗ pow. Step 4: If i becomes
N ∗ K, store d in M ′ ′posand increment pos. Set pow to 1 and both i and d to 0.
Step 5: Transform w for next q in S using eliminative hashing technique as used in
embedding procedure.
Step 6: If length of M ′ ′ is less than length of M′, goto Step 2. Otherwise stop extracting.
Step 7: M ′ ′ is the required message array represented in modulo − 2KNform.
Just like the embedding algorithm has a pre-processing segment, the extraction algorithm also
needs to be post-processed to get back the actual message from the stream of 2KN − ary digits.
The elements are converted to their binary equivalents. It is made sure that all the binary
Fig. 3 Cover images (left) and their corresponding stego images (right) after performing the embedding
algorithm at 4 bpp
Multimedia Tools and Applications
numbers have N ∗ K bits to ensure parity. This binary array is then transformed into a row
vector in row-major order. Then 8-bits are taken at a time and converted into its decimal
equivalent. This is done till all the bits have been converted.
Fig. 2 depicts the flowchart for the extraction algorithm.
The proposed technique is implemented in MATLAB using five grayscale images to evaluate
the visual imperceptibility of the embedding algorithm and to compile the results. The
following images namely, ‘Lena’, ‘Baboon’, ‘Peppers’, ‘Airplane’, ‘Sailboat’, ‘Ship’ and
‘Camera’, each of size 512 × 512 have been used as cover images. Figure 3 shows the cover
images and their respective stego images after applying the embedding algorithm.
The proposed method has been tested on payloads 1, 2, 3 and 4 bpp. The stego images after
embedding message information at above payloads in “Lena” cover image is given in Fig. 4.
The algorithm has also been tested on different values of K. Figure 5 shows the stego
images generated after applying the algorithm with K = 1 to 4.
The EEMDHW algorithm has also been implemented using images of varying sizes. The
secret message has been embedded into images of sizes 64 × 64, 128 × 128, 256 × 256 and
512 × 512. The stego images, thus generated, are presented in Fig. 6.
In order to measure the performance of this method the capacity, quality, and robustness of
the method needs to be measured [37]. The capacity and the quality of the method have been
discussed hereafter, and the robustness of the proposed method has been discussed in section
5. The metrics which are mainly used to measure the capacity and the quality of any
steganographic technique are:
& Payload: This denotes the number of bits that can be hidden in each pixel of the cover
image. Thus, more the payload of a steganographic technique, more the data that can be
hidden in a cover image of a given size. It is usually measured in bpp (bits per pixel).
& Peak Signal-to-Noise Ratio (PSNR): This is the ratio of the original data to the noise
introduced to the cover image while embedding secret data. Thus, this acts as a measure of
the quality of the generated stego image with respect to the original cover image. The
difference in a stego image and a cover image will be more imperceptible for a stego image
with more PSNR than one with less PSNR.
It has been observed that as the payload is increased, the PSNR of the stego image decreases.
This is because a greater number of bits are affected and the stego image quality decreases.
Thus, the researchers in the field of steganography aims at a possible trade-off between the two
metrics, i.e., to aim at increasing embedding capacity of the image while not sacrificing its
image quality and vice versa.
From Table 1 (a) (b) (c) (d), it can be seen that the maximum PSNR is 60.19 at N (i.e.,
payload) = 1 bpp for K = 1 using ‘Peppers’ image. On the other hand, the minimum PSNR is
33.89 which has been obtained in ‘Camera’ image, at 4 bpp for K = 4. This also means that the
PSNR has never fallen below the least acceptable level (30 dB).
The three parameters that can be changed while using this algorithm are
This paper analyses how the change in each of these parameters while keeping the other 2
constant can affect the PSNR of the stego image with respect to the cover image.
Table 2 shows the effect of embedding at payloads 1–4 bpp on ‘Lena’ cover image of size
256 × 256. K is kept constant at 3. The table shows that there is a stark decrease in the PSNR of
the stego image as the payload is increased from 55.34 dB at N = 1 bpp to 34.01 to N = 4bpp.
This can be explained by the fact that as payload is increased, a greater number of bits are
affected in each pixel and the stego pixel value differs by a greater margin from the cover pixel
value.
The EEMDHW algorithm has also been implemented on ‘Baboon’ cover images of size
256 × 256 by varying value of K while keeping N = 3 and the results are compiled in Table 3.
From Table 3, it can be observed that increasing the randomness factor, K, for security
decreases the PSNR. However, while the PSNR decreases sharply from K = 1 to K = 2, the
amount of decrease becomes less thereafter.
Multimedia Tools and Applications
Table 2 Experimental results using gray-scale images of sizes 512×512, 256×256, 128×128 and 64×64
Table 2 (continued)
The proposed method has been applied on cover images of varying sizes to study the effect
of change in image size on the PSNR value. ‘Peppers’ cover images of 4 different sizes at N =
3 and K = 3 have been used and the results are presented in Table 4. It can be seen that the
PSNR values do not differ much when size of cover image is changed.
The average PSNR for each payload in Table 1 has been compared with the average PSNR
of the stego images obtained from some state-of-the-art techniques as shown in Table 5. The
PSNR in bold indicates the highest PSNR for each payload.
The table suggests that this technique beats its contemporaries by a significant margin at
any given payload from 1bpp to 3bpp. Though the EAD method [30] has an average payload
between 1 and 2 bpp, the PSNR recorded at 1.42 bpp is below that of our proposed method at
2 bpp. However, for 4 bpp, while this method beats the simple LSB technique [10] by almost
3.5 dB, the average PSNR of the proposed method falls short of that of the OPAP method of
Chan et al. [10] by a narrow margin of 0.7 dB. However, it must also be noted that a maximum
PSNR of 34.84 dB can be attained by OPAP at 4 bpp payload. Whereas, in the proposed
method the maximum PSNR obtained at payload of 4 bpp is 37.16 dB, almost 2.3 dB more
than the highest PSNR of OPAP [10] method. The GEMD method by Leng et al. [32] achieves
a higher maximum payload of 4.75 bpp. But at 4.09 bpp, it gives a PSNR of 34.32 dB which is
comparable to the PSNR of the stego image generated by the proposed method at 4 bpp.
The embedding and extraction phases of the proposed method have a time complexity of O
(PK) each where P is the number of pixels of cover image affected by the embedding
procedure and K is the randomness factor. To put this into perspective the proposed method
and the method by Saha et al. [7] have been performed using the same system. Same cover
Table 2 Comparing results at different payloads keeping K (=3) and size of image (256 × 256) constant
1 55.34
2 48.53
3 39.50
4 34.01
Multimedia Tools and Applications
Table 3 Showing results at varying the randomness factor K. Size of image (256 × 256) and payload N (=3) are
kept constant
1 43.73
2 40.72
3 39.47
4 39.47
images of size 512 × 512 and same message images large enough to conduct full embedding
have been taken into consideration to ensure the parity. A payload of 2 bpp has been used since
that is the maximum obtained by Saha et al. The experiments have been carried out on a
computer with an Intel Core i3-380M processor running at 2.53GHz using 4 GB RAM and
Ubuntu 16.04 LTS operating system. The proposed method takes an average time of 27.53 s,
51.88 s, 76.86 s and 102.77 s to do full embedding on a cover image at 2 bpp using K = 1, 2, 3
and 4. Meanwhile, the method developed by Saha et al. completes the embedding at 32.44 s at
2 bpp and K = 4. Thus, at payload 2 bpp and K = 4, the technique proposed by Saha et al. is 3
times quicker than this method. However, it compensates for the 1 min delay by increasing the
average PSNR by 1.5 dB.
5 Security analysis
In the previous section we discussed the imperceptibility of the presence of images, i.e.,
distortion in the stego images. However, there exist algorithms that detect the presence of
secret messages in seemingly innocent images. This procedure to detect the presence of
embedding in an image is known as steganalysis. This section demonstrates the robustness
of the technique using the RS steganalysis technique [38].
The RS steganalysis technique divides the entire image in groups of n pixels. For
each group of pixels, it applies a mask in order to flip the LSBs of each pixel of the
group. At first, the algorithm applies a 0 or + 1 on each pixel value and classifies it as
regular, singular or unstable. Then it adds 0 or − 1 to each pixel and classifies it again.
It does this for all the groups in the image and counts the number of regular and
singular groups for both the positive mask and the negative mask. If these numbers are
similar in both the masks, then the probability of embedding the image is less. These
values are then used to plot curves to determine the robustness of the steganographic
technique. Each RS attack figure has 2 plots. The plot on the left is the RS plot, i.e., it
counts the number of regular and singular groups for various embedding rates. The plot
next to it is the ROC curve. It estimates the ratio of the number of pixels with their
LSBs flipped to the total number of pixels, i.e., it tries to detect the estimated message
length in terms of pixels. The less the area under the ROC curve, the less detectable it
is.
This section deals with the results of the RS steganalysis on the proposed method. The RS
attack has been performed on Lena cover images of size 256 × 256 pixels. The steganalysis
have been performed on different values of N, K and embedding rates, to study the relation (if
any) between the robustness and the payload. Plots have been created for various combinations
of N and K as shown in Fig. 7 and Fig. 8.
Multimedia Tools and Applications
64 × 64 39.61
128 × 128 39.71
256 × 256 39.62
512 × 512 39.44
Figure 7 shows the RS attack plots for different payloads at K = 2. It can be seen that the for
payloads 1–3 bpp, the Rplus and Rminus curves are very close till 50% embedding rate and so
are the Splus and the Sminus curves. However, the distance increases thereafter. However,
even after 50% embedding rate, the message length is always detected to be less than 20%, i.e.,
the amount of embedding cannot be detected by the RS attack. But, the robustness of the
technique is at stake when the embedding is done at 4bpp. The area under the ROC curve is
also high, i.e., the estimated message length is close to the original embedding rate.
Figure 8 shows the RS attack plots for different values of randomness factor K at constant
payload 2 bpp. The plots are almost similar to those in Fig. 7. As K is increased, the gap
between the Rplus and Rminus curves decreases slightly to the right of 0.5 embedding rate.
Thus, the curves show that the technique remains undetectable for embedding rate < 50%.
Although it becomes detectable thereafter, the probability decreases with increase in the
randomness factor. Also, the estimated message length always remains below 20% of total
pixels irrespective of the embedding rate, i.e., the amount of data in the image is hardly
detected correctly.
Thus, it can be said that using this technique, embedding will not be detected up to 3bpp
payload at embedding rate 50%. When the embedding rate is increased, the probability of
detection increases. However, the amount of detection is not at stake unless the payload is
increased to 4bpp.
6 Conclusion
This method takes inspiration from Saha et al.’s method [7] of hiding 2KN − ary digits using a
weightage array in a cover image. It also intends to overcome the major drawbacks of the said
method viz. less payload and image quality distortion. The weightage array w has been made
more dynamic using eliminative hashing method to make things tougher for intruders. While
this paper succeeds to achieve less image distortion for payloads 1–3 bpp, it falls short at
payload 4 bpp. With the ever-increasing amount of data and the risk of data leak, more efforts
are still needed to increase the embedding capacity of cover images with considerable
distortion of quality. For this reason, the authors also aim at increasing the average PSNR
for higher payloads (4 bpp) and at trying to work with payloads higher than those proposed in
this paper.
Unlike watermarking, where the primary intent is robustness, the primary intent of stega-
nography, especially EMD steganography, is imperceptibility. The proposed method has
succeeded in increasing the imperceptibility of the distortion is the cover image, as discussed
in the comparison of PSNR or contemporary techniques. As far as the robustness is concerned,
the proposed technique is rendered undetectable under RS attack at payloads of 1–3 bpp and
randomness factor = 1 to 4 at a maximum embedding rate of 50%. Its robustness fails at values
Multimedia Tools and Applications
Fig. 7 RS attack of “Lena” image generated by EEMDHW at N = (a) 1, (b) 2, (c) 3, (d) 4 keeping K = 2 and
size = 256 × 256
Multimedia Tools and Applications
Fig. 8 RS attack of “Lena” image generated by EEMDHW at K = (a) 1, (b) 2, (c) 3, (d) 4 keeping N = 2 and
size = 256 × 256
higher than these which inspires us to extend this proposed method and is thus deemed as the
future scope of this paper. Since the security of the technique is at stake for aforementioned
conditions, the randomness of the weightage array, i.e., the eliminative hash function and its
use in dynamic weightage array based steganography, becomes effective here as it makes the
decryption of the original message from the image difficult.
References
4. C. C. Chang, P. Tsai, and M. H. Lin, “An adaptive steganography for index-based images using codeword
grouping,” Lect. Notes Comput. Sci. (including Subser. Lect. Notes Artif. Intell. Lect. Notes Bioinformatics),
2004, doi: https://doi.org/10.1007/978-3-540-30543-9_91.
5. Zhang X, Wang S (2006) Efficient Steganographic embedding by exploiting modification direction. IEEE
Commun Lett 10(11):781–783. https://doi.org/10.1109/LCOMM.2006.060863
6. S. K. Ghosal and J. K. Mandal, “On the use of the Stirling transform in image steganography,” J. Inf. Secur.
Appl., 2019, doi: https://doi.org/10.1016/j.jisa.2018.04.003, On the use of the Stirling Transform in image
steganography.
7. S. Saha, S. K. Ghosal, A. Chakraborty, S. Dhargupta, R. Sarkar, and J. K. Mandal, “Improved exploiting
modification direction-based steganography using dynamic weightage array,” Electron. Lett., 2018, doi:
https://doi.org/10.1049/el.2017.3336, Improved exploiting modification direction-based steganography
using dynamic weightage array.
8. M. Hussain, A. W. Abdul Wahab, A. T. S. Ho, N. Javed, and K. H. Jung, “A data hiding scheme using
parity-bit pixel value differencing and improved rightmost digit replacement,” Signal Process. Image
Commun., 2017, doi: https://doi.org/10.1016/j.image.2016.10.005, A data hiding scheme using parity-bit
pixel value differencing and improved rightmost digit replacement.
9. Y. Liu, C. C. Chang, P. C. Huang, and C. Y. Hsu, “Efficient Information Hiding Based on Theory of
Numbers,” Symmetry (Basel)., vol. 10, no. 1, p. 19, Jan. 2018, doi: https://doi.org/10.3390/sym10010019.
10. C. K. Chan and L. M. Cheng, “Hiding data in images by simple LSB substitution,” Pattern Recognit., 2004,
doi: https://doi.org/10.1016/j.patcog.2003.08.007, Hiding data in images by simple LSB substitution.
11. J. Bai, C. C. Chang, T. S. Nguyen, C. Zhu, and Y. Liu, “A high payload steganographic algorithm based on
edge detection,” Displays, 2017, doi: https://doi.org/10.1016/j.displa.2016.12.004, A high payload
steganographic algorithm based on edge detection.
12. D. Chaudhary, S. Gupta, and S. Deswal, “Origin of Hybrid Security Mechanisms and Ways of
Improvement,” Int. J. Big Data Secur. Intell., 2015, doi: https://doi.org/10.21742/ijbdsi.2015.2.2.01.
13. T. Rabie, “Digital image steganography: an FFT approach,” in Communications in Computer and
Information Science, 2012, doi: https://doi.org/10.1007/978-3-642-30567-2_18.
14. A. A. Al-Ataby and F. M. Al-Naima Fiete, “High capacity image steganography based on curvelet
transform,” in Proceedings - 4th International Conference on Developments in eSystems Engineering,
DeSE 2011, 2011, doi: https://doi.org/10.1109/DeSE.2011.13.
15. E. Ghasemi, J. Shanbehzadeh, and N. Fassihi, “High capacity image steganography using wavelet transform
and genetic algorithm,” in IMECS 2011 - international MultiConference of engineers and computer
scientists 2011, 2011.
16. Z. Pan, X. Ma, and X. Deng, “New reversible full-embeddable information hiding method for vector
quantisation indices based on locally adaptive complete coding list,” IET Image Process., 2015, doi:
https://doi.org/10.1049/iet-ipr.2014.0310, New reversible full-embeddable information hiding method for
vector quantisation indices based on locally adaptive complete coding list.
17. M. A. Alia and A. A. Yahya, “Public-key steganography based on matching method,” Eur. J. Sci. Res.,
2010.
18. M. R. Islam, A. Siddiqa, M. P. Uddin, A. K. Mandal, and M. D. Hossain, “An efficient filtering based
approach improving LSB image steganography using status bit along with AES cryptography,” in 2014
International conference on informatics, Electronics and Vision, ICIEV 2014, 2014, doi: https://doi.
org/10.1109/ICIEV.2014.6850714.
19. A. Chatterjee, S. K. Ghosal, and R. Sarkar, “LSB based steganography with OCR: an intelligent amalgam-
ation,” Multimed. Tools Appl., 2020, doi: https://doi.org/10.1007/s11042-019-08472-6, LSB based
steganography with OCR: an intelligent amalgamation.
20. N. Tayal, R. Bansal, S. Dhal, and S. Gupta, “A novel hybrid security mechanism for data communication
networks,” Multimed. Tools Appl., 2017, doi: https://doi.org/10.1007/s11042-016-4111-x, A novel hybrid
security mechanism for data communication networks.
21. R. Bansal, S. Gupta, and G. Sharma, “An innovative image encryption scheme based on chaotic map and
Vigenère scheme,” Multimed. Tools Appl., 2017, doi: https://doi.org/10.1007/s11042-016-3926-9, An
innovative image encryption scheme based on chaotic map and Vigenère scheme.
22. C. C. Chang, Y. C. Chou, and D. Kieu, “An information hiding scheme using Sudoku,” in 3rd International
Conference on Innovative Computing Information and Control, ICICIC’08, 2008, doi: https://doi.
org/10.1109/ICICIC.2008.149.
23. Q. Mao, K. Bharanitharan, and C. C. Chang, “Multi-round dynamic swap optimisation for table-based
steganography,” IET Image Process., 2015, doi: https://doi.org/10.1049/iet-ipr.2015.0065, Multi-round
dynamic swap optimisation for table-based steganography.
Multimedia Tools and Applications
24. Z. X. Yin, C. C. Chang, Q. Xu, and B. Luo, “Second-order steganographic method based on adaptive
reference matrix,” IET Image Process., 2015, doi: https://doi.org/10.1049/iet-ipr.2014.0159, Second-order
steganographic method based on adaptive reference matrix.
25. C. C. Chang, W. L. Tai, and K. N. Chen, “Improvements of EMD embedding for large payloads,” in
Proceedings - 3rd International Conference on Intelligent Information Hiding and Multimedia Signal
Processing, IIHMSP 2007., 2007, doi: https://doi.org/10.1109/IIHMSP.2007.4457590.
26. K. H. Jung, “Improved Exploiting Modification Direction Method by Modulus Operation,” vol. 2, no. 1, pp.
79–88, 2009.
27. N. T. Huynh, K. Bharanitharan, and H. S. Hsueh, “Efficient Combinatorial Optimization Algorithm for
Exploiting Modification Direction in Data Embedding,” Indian J. Sci. Technol., 2016, doi: https://doi.
org/10.17485/ijst/2016/v9i28/92733.
28. A. A. Abdulla, H. Sellahewa, and S. A. Jassim, “Stego quality enhancement by message size reduction and
fibonacci bit-plane mapping,” in Lecture notes in computer science (including subseries lecture notes in
artificial intelligence and lecture notes in bioinformatics), 2014, doi: https://doi.org/10.1007/978-3-319-
14054-4_10.
29. Rehman A, Saba T, Mahmood T, Mehmood Z, Shah M, Anjum A (Dec. 2019) Data hiding technique in
steganography for information security using number theory. J Inf Sci 45(6):767–778. https://doi.
org/10.1177/0165551518816303
30. D. R. I. M. Setiadi, “Payload enhancement on least significant bit image steganography using edge area
dilation,” Int. J. Electron. Telecommun., 2019, doi: https://doi.org/10.24425/ijet.2019.126312.
31. Z. F. Yaseen and A. A. Kareem, “Image steganography based on hybrid edge detector to Hide encrypted
image using Vernam algorithm,” in SCCS 2019–2019 2nd Scientific Conference of Computer Sciences,
2019, doi: https://doi.org/10.1109/SCCS.2019.8852625.
32. H. S. Leng and H. W. Tseng, “Generalize the EMD scheme on an n-dimensional hypercube with maximum
payload,” Multimed. Tools Appl., 2019, doi: https://doi.org/10.1007/s11042-019-7228-x, Generalize the
EMD scheme on an n-dimensional hypercube with maximum payload.
33. A. A. Abd EL-Latif, B. Abd-El-Atty, and S. E. Venegas-Andraca, “A novel image steganography technique
based on quantum substitution boxes,” Opt. Laser Technol., 2019, doi: https://doi.org/10.1016/j.
optlastec.2019.03.005, A novel image steganography technique based on quantum substitution boxes.
34. S. Dhargupta, A. Chakraborty, S. K. Ghosal, S. Saha, and R. Sarkar, “Fuzzy edge detection based
steganography using modified Gaussian distribution,” Multimed. Tools Appl., Jan. 2019, doi: https://doi.
org/10.1007/s11042-018-7123-x, Fuzzy edge detection based steganography using modified Gaussian
distribution.
35. X. Liao, Y. Yu, B. Li, Z. Li, and Z. Qin, “A new payload partition strategy in color image steganography,”
IEEE Trans. Circuits Syst. Video Technol., 2019, doi: https://doi.org/10.1109/tcsvt.2019.2896270, A New
Payload Partition Strategy in Color Image Steganography.
36. B. Carpentieri, A. Castiglione, A. De Santis, F. Palmieri, and R. Pizzolante, “Compression-based steganog-
raphy,” in Concurrency computation , 2019, doi: https://doi.org/10.1002/cpe.5322, 32.
37. A. A. Abdulla, “Exploiting similarities between secret and cover images for improved embedding efficiency
and security in digital steganography,” 2015.
38. J. Fridrich, M. Goljan, and R. Du, “Detecting LSB steganography in color and gray-scale images,” IEEE
Multimed., 2001, doi: https://doi.org/10.1109/93.959097, Detecting LSB steganography in color, and gray-
scale images.
Publisher’s note Springer Nature remains neutral with regard to jurisdictional claims in published maps and
institutional affiliations.
Multimedia Tools and Applications
Shaswata Saha , is currently an undergraduate student of Computer Science and Engineering at Jadavpur
University, Kolkata, India. His research interests are Image Steganography, and image processing.
Anuran Chakraborty , is currently an undergraduate student of Computer Science and Engineering at Jadavpur
University, Kolkata, India. His research interests are Image Steganography, and image processing.
Multimedia Tools and Applications
Agneet Chatterjee is currently an undergraduate student of Computer Science and Engineering at Jadavpur
University, Kolkata, India. His research interests are Image Processing, Pattern Recognition and Machine
Learning.
Souvik Dhargupta , is currently an undergraduate student of Computer Science and Engineering at Jadavpur
University, Kolkata, India. His research interests are Image Steganography, and image processing.
Multimedia Tools and Applications
Sudipta Kr Ghosal is a Lecturer in the department of Computer Science & Technology, Nalhati Government
Polytechnic, Birbhum, India. He received his bachelor of technology in Computer science and Engineering in
2007. He received his master degree in IT (Courseware Engineering) from Jadavpur University, India, in 2010.
He received his PhD degree in 2016 from University of Kalyani, India. Dr. Ghosal has around nine years of
experience in teaching and industry. He has twenty five publications in the national / international conference
proceedings, book chapters and journals.
Ram Sarkar received his B. Tech degree in Computer Science and Engineering from University of Calcutta in
2003. He received his M.C.S.E and PhD (Engg.) degrees from Jadavpur University in 2005 and 2012
respectively. He joined Jadavpur University as an Assistant Professor in 2008 where currently he is working
as an Associate Professor. He was awarded Fulbright-Nehru Fellowship (by USIEF) for post-doctoral research at
University of Maryland, College Park, USA during 2014–15. His areas of current research interest are Document
Image Processing, Machine Learning, Soft Computing, Bioinformatics etc. He is a senior member of the IEEE,
U.S.A.