You are on page 1of 26

Multimedia Tools and Applications (2021) 80:14495–14520

https://doi.org/10.1007/s11042-020-10424-4

Secured image steganography based


on Catalan transform

Souradeep Mukhopadhyay 1 & Sabbir Hossain 1 & Sudipta Kr Ghosal 2 & Ram Sarkar
1

Received: 9 June 2020 / Revised: 21 September 2020 / Accepted: 22 December 2020 /


Published online: 25 January 2021
# The Author(s), under exclusive licence to Springer Science+Business Media, LLC part of Springer Nature 2021

Abstract
Transform domain steganography methods are always preferable than spatial domain
methods to the research community due to following advantages: better feature identifi-
cation, improved security and higher robustness. In this paper, an integer sequence named
Catalan Transform (CT) has been exploited in the image steganography domain. At the
outset, the cover image is decomposed into 2 × 2 non-overlapping blocks in row major
order. Then, each such block i.e., 4-pixel group is converted into transform domain using
CT. Secret bits are embedded into the transformed components in varying proportions
which facilitates us to achieve a payload in the range of 1 to 4 bpp. Inverse Catalan
Transform (ICT) is applied over transformed cum embedded quadruples to generate the
stego-pixels in spatial domain. Successive embedding operation over an entire image
ensures the formation of stego-image. Experimental results confirm that the Peak Signal to
Noise Ratio (PSNR) values obtained by the proposed method are always above the
acceptable level (i.e., 30 dB) and at the same time, it outperforms many state-of-the-art
methods. Other metrics such as Mean Squared Error (MSE), Structural Similarity Index
Measure (SSIM), Normalized Cross Correlation (NCC) and Global histograms with
similarity measure have also been computed which prove the effectiveness of the proposed
scheme. The robustness of the stego-images has further been tested using the StegExpose
tool and it is found that the tool is unable to detect the presence of secret data. The code of
this work is available at: https://github.com/Souradeep150/Catalan-based-Steganography.

Keywords Catalan transform . Inverse Catalan transform . Steganography . PSNR . Payload

1 Introduction

Steganography is the art of hiding secret data within an audio, video, image or text file either in
spatial or in transform domain. It is also regarded as an effective solution employed to protect secret

* Sudipta Kr Ghosal
sudipta.ghosal@gmail.com

Extended author information available on the last page of the article


14496 Multimedia Tools and Applications (2021) 80:14495–14520

or sensitive data from malicious attacks. But Steganography is different from Cryptography and
Watermarking in various aspects. Cryptography is the practice and study of techniques for secure
communication in the presence of third parties called adversaries. More generally, Cryptography is
about constructing and analysing protocols that prevent third parties or the public from reading
private messages. Steganography is the practice of concealing a file, message, image, or video within
another file, message, image, or video. Steganography requires two files: one is the message which
is to be hidden and the other is the cover file which is used to hide the data/message. Images are the
most commonly used cover media (both natural images and computer generated graphics) [35] due
to its flexible size compared to other digital media and ease of accesses in the internet. In
Watermarking, the message which is inserted to the object (image, audio or video) is related with
the object, such as the ownership status of the object. While in Steganography, the message usually
is not related with the object. In Steganography, the message which is inserted to an object is hidden
(covert), while in Watermarking it is optional (can be covert or overt). There are four important
evaluation criteria for any steganographic scheme: payload, quality, robustness and embedding
efficiency [1]. The payload (also referred as capacity) signifies the amount of data embedded within
the cover image. On the other hand, quality of stego-image is an important aspect as far as the
imperceptibility is concerned. A number of metrics are used to evaluate the image quality such as
Mean Squared Error (MSE), Structural Similarity Index Measure (SSIM), and Normalized Cross
Correlation (NCC) etc. The robustness implies the resistance against any sort of attacks or
manipulations made by the intruders. The embedding efficiency is defined as the number of
necessary changes made to the cover pixels for a given embedding rate.
There are many popular Steganographic techniques based on different techniques such as
Hashing [41], Least Significant Bit (LSB) [4], Pixel Value Decomposition [2], Discrete Cosine
Transform (DCT) [11], Binomial Transform [30], Discrete Wavelet Transform (DWT) [9] and Lah
Transform [19] etc. The LSB method replaces the least significant bit of each cover pixel by the
message bit, whereas the PVD method computes the difference of successive pairs of pixels, and
then checks in which range table of the difference value belongs to. The larger difference signifies
that the pixel is “edge”, whereas the smaller difference signifies that the pixel is “smooth”. Due to
the Human Visual System (HVS) model, the alteration made into the edge pixels reflects less
distortion than the alteration made into the smooth pixels. This implies that it is advantageous to
hide more data into the pixels which are considered as the edge. For the convenience of the readers,
the evolution of the Steganographic methods has been reported in Table 1. In this table, we have
also mentioned the domain used for hiding the information, method and dataset used along with
pros and cons of the methods. From the methods cited in Table 1, we can get an idea about the
research directions conceived by the researcher in the domain of Steganography in the last decades.
Rest of the paper is organized as follows. Section 2 and Section 3 of the paper deal with the
motivation and contribution parts. The formulation of Catalan Transform (CT) is mentioned in
Section 4. Proposed method is elaborated in Section 5. Experimental Result, comparison and analysis
have been reported in Section 6. Section 7 and Section 8 deal with the security analysis and complexity
analysis of the proposed method respectively. Conclusion is of this work drawn in Section 9.

2 Motivation

Steganographic schemes discussed in Table 1 reveal that there are a number of disadvantages
despite having a few advantages. The disadvantages are lack of security, low payload, low
robustness and high computational cost. It is worthy to mention that transform domain
Table 1 Comparative study of some Steganographic methods proposed by the researchers in last two decades

Reference Method Domain Dataset Strengths of the method Limitations of the method

Shin et al. [41] One-Time Hash Spatial Not mentioned Better speed, can be used on Lower payload,
any type of file formats. Easily detectable by
steganalysis.
Akhtar et al. [4] Compressed LSB Spatial USC-SIPI database. Cover Good hiding capacity and Not efficient with high
image size is 512×512 imperceptibility. randomness of secret data.
and secret image size
is 256×256.
Biswas et al. [11] LSB with DCT Transform Baboon -RGB image of Good carrier capacity and Computational complexity
512 ×512 image size. Stego-image Quality. is high.
Ghosal et al. [30] Binomial transform Transform The USC-SIPI Image Database Good payload, calculation Low PSNR value with
and LSB complexity is low respect to Payload
Banoci et al. [9] DWT Transform Lena, Motorbike, Apple. Good image Quality with Computational intensive for
high payload. fine analysis.
Khalid et al. [5] LSB based Steganography Spatial A RGB cover image High security Complex for secret data
Multimedia Tools and Applications (2021) 80:14495–14520

with cropping of cover image. of 512 ×512 size. extraction.


Mandal et al. [32] Stirling Transform Transform The USC-SIPI Image Database High payload, PSNR Time complexity is high.
Emam et al. [15] LSB with random pixel selection Spatial Lena, Baboon, Pepper, High maximum hiding Less secure, easily detectable
Bird, Airplane. capacity, and higher by steganalysis
visual quality.
Mandal et al. [31] 2D-Discrete Hartley transform Transform Lena, Baboon, Airplane, Less Computational complexity PSNR is not good.
Earth, Sailboat, Foster City, in Processing of real data type.
San Diego, Oakland.
Ghosal et al. [18] Stirling Transform Transform The USC-SIPI Image Database Computational complexity is Low PSNR value with respect
less as integer sequence is used. to payload
Hajizadeh et al. EMD Spatial Lena, Baboon, F16, Pepper, Good data embedding capacity Security is low.
[21] Boat, Tiffany. and visual quality
Cheonshik et al. Improved EMD Spatial Lena, Baboon, Tiffany, Airplane, Higher embedding performance Easily detectable by
[27] Pepper, Goldhill, Barbara, with good perceptual quality. steganalysis.
Boat, Zelda.
Wang et al. [43] Re-adjusted GEMD Spatial Lena, Baboon, F16, Barbara, Maintains good image quality Computational complexity is
Boat, Goldhill, Tiffany, Pepper. and prevents RS detection. high.
14497
Table 1 (continued)
14498

Reference Method Domain Dataset Strengths of the method Limitations of the method

Shet et al. [39] FGPA implementation Spatial 6 Gy scale images of 512×512 Efficient for both arbitrary No secure against various
of EMD based image size. numbers of pixel groups steganalysis.
and variable image resolution
Kuo et al. [28] A data hiding scheme based Spatial Lena, Baboon, F16, Barbara, Higher embedding rate and Computational complexity is
on square formula fully Boat, Goldhill, Elaine Tiffany, good stego quality. high, easily detectable.
EMD method. Pepper, Bridge.
Abdulla et al. [3] Improving embedding efficiency Spatial 44 grayscale images from Produces high quality Low payload
for digital steganography by USC-SIPI database stego-images and robust
exploiting similarities between against LSBR and LSBM
secret and cover images targeted steganalysis tools
Meng et al. [33] A Fusion Steganographic Spatial five images from COCO2014 Complex texture regions are Low payload. Uses multiple
Algorithm dataset identified for embedding steganography scheme.
Based on Faster R-CNN secret information
Xiang et al. [45] A Novel Linguistic Spatial 1737 cover texts (1738 synonyms Uses synonym run-length Low payload and
Steganography and 91,040 words per cover encoding method to improve embedding efficiency
Based on Synonym text) the capability of anti--
Run-Length were randomly downloaded steganalysis
Encoding from Internet
Zhang et al. [46] DCT and LDA Topic Transform Accessed images from INRIA robust against most of the Low capacity and fragile
Classification Holidays database and visual and geometrical attacks against rotation attack
ImageNet database
Seyyedi et al. [40] IWT Transform 512 ×512 grayscale images Better security in cover and Low PSNR
Barbara, Peppers, Baboon, secret image both.
Lena, Airplane and Boat
Raja et al. [37] 2×2 cell of HH band Transform Cover images: Garden, Football Good payload capacity and Lower PSNR for DWT
of DWT or IWT fans, House and Water house. improved security. and formatted images.
payload images: Cups, Lion,
Sony and Rose.
Atawneh et al. [8] Diamond Encoding, Haar, DWT Transform Grayscale images sized 512×512 Stable against various Computational complexity is
taken from SIPI, CVG and steganalysis, good control high.
RSP databases. over underflow/overflow.
Multimedia Tools and Applications (2021) 80:14495–14520
Table 1 (continued)

Reference Method Domain Dataset Strengths of the method Limitations of the method

Zhang et al. [22] Haar-IWT, Multidirectional Line Transform 512 ×512-pixel grayscale images Better time complexity Lower payload compared to
Encoding Edge detector of USC-SIPI database and [O (n)]. Stable against PSNR values.
BOSSbase_1.01 database as various steganalysis.
cover image and two 128×128
grey scale images.
Dmour et al. [6] IWT, Canny edge Detector, Both Spatial BOWS2 database, which contains Enhanced security Payload capacity is lower in
XOR and 10,000 grayscale natural images Transform domain, Greater
Transform of size 512×512. computational cost.
domains
Atta et al. [38] Wavelet Packet Transformation, Transform Grayscale Lena, Baboon, Pepper, Secure against RS-steganalysis, Computation complexity is high
Neutrosophic Set Boat, Lake, Jet, Elaine, pixel differencing and Payload is low.
Couple, Truck histogram analysis.
Jothy et al. [23] IWT Transform Lena, Baboon, Pepper, Earth, Integral transform ensures PSNR is significantly low.
Football, Moon. faster calculation. Security
Multimedia Tools and Applications (2021) 80:14495–14520

is high
Nazari et al. [34] DCT and Fuzzy Transform Baboon, Boat, Pepper, Airplane, In order to increase the Lesser payload as compared to
inference system Barbara, and Lena imperceptibility, an optimal PSNR values. Not highly
number of Fuzzy Interference secure against steganalysis.
System rules in Human
Visual System parameters are
used.
Kalita et al. [24] LSB and IWT Transform Cameraman, Baboon, Barbara, No data loss. Stable against Visual quality is low due to high
Fruits, Peppers, Lena, Sails, various electronic attacks. distortion. Payload is not
Tulips, Zelda high.
14499
14500 Multimedia Tools and Applications (2021) 80:14495–14520

methods are preferred than spatial domain methods to the research community due to the
following key points: ability to identify more features, tighter security and, high robustness.
These key points motivate the researchers around the world to design and develop new Stegano-
graphic algorithms using the traditional transformations such as DCT, DFT and DWT. In spite of
having numerous features of the above mentioned transformations, few shortfalls are also observed.
For instance, DCT produces real-valued output and makes the calculation time consuming. Moreover,
DCT based methods work only on JPEG files since it assumes a certain statistical distribution of the
cover media which is commonly found in JPEG files only. In general, the DFT coefficients are
represented as complex number and the typical computational complexity of DFT is O(n2). DWT
provides better energy compaction than DCT while IWT takes the advantage of integer based
calculation. But, the payload obtained by both DWT and IWT based methods is not significantly well.
To overcome the shortfalls, discussed in the motivation section, we have proposed a
Steganographic scheme based on CT in this paper. Unlike the existing transformations, the
entire calculation of CT is integer based, not floating-point based; hence it is faster. Other than
JPEG, a number of image formats such as BMP, PGM, PPM and TIFF etc. are being
supported by our method. Unlike DFT, CT does not generate complex value as output.
Computational complexity of CT is O(nlog(n)). Going from O(n2) of DFT to O(nlog(n)) of
CT is quite a substantial improvement as far as the computational complexity is concerned.
With respect to DWT based scheme, CT based Steganography method offers better experi-
mental results in terms of payload and PSNR which can be verified from the result given in
Table 2. The proposed method also proves its robustness against common attacks such as RS
analysis, Fusion (mean), Primary sets, Chi square test etc. which can be verified from Table 7.

3 Our contribution

The use of CT in image Steganography is novel as this is utilized for the first time in this field
to the best of our knowledge. The most prominent property of CT is that alterations in
transformed coefficients due to embedding are significantly dissimilar owing to the unique
process used to compute the coefficients, which in turn, makes the process unbreakable to the
intruders. Besides, the novelty of our proposed scheme also lies in the sharp manipulation of
these transformed coefficients through an adaptive embedding rule which ensures superior
payload, improved perceptual quality, and most significantly ease of extraction by the receiver
without compromising its data embedding ability.
Hence, the key contributions of the paper can be summarized as follows:

i) Increasing the Stego-image quality as compared to some state-of-the-art methods.


ii) The computational complexity of CT is O(nlog(n)) which is considered to be fast enough.
iii) The payload can be easily varied between 1 bpp to 4 bpp.
iv) The robustness of the scheme has also been investigated using StegExpose tool which
confirms that the scheme is robust to common steganalysis attacks.

4 Catalan transform

Let us consider p0, p1, …, pl be the pixel values in a given pixel group P. By applying CT [10]
one can compute the transformed components t0, t1, …, tlas shown in equation number (1):
Multimedia Tools and Applications (2021) 80:14495–14520 14501

Table 2 PSNR values of proposed Steganographic scheme with respect to1, 2, 3 and 4 bpp of payloads

IC Dimensions of IC Block size=2×2

PSNR w.r.t. n bpp of payload

n =1 n =2 n =3 n =4

Lena 128×128 48.62 40.98 34.28 27.87


256×256 48.61 40.93 34.26 27.91
512×512 48.62 40.93 34.23 27.93
Baboon 128×128 48.67 40.87 34.32 27.90
256×256 48.62 40.89 34.21 27.93
512×512 48.62 40.92 34.24 27.93
Pepper 128×128 48.59 40.91 34.27 27.95
256×256 48.65 40.94 34.27 27.95
512×512 48.62 40.91 34.23 27.95
Splash 128×128 48.63 40.89 34.29 27.93
256×256 48.61 40.94 34.23 27.99
512×512 48.63 40.94 34.24 27.96
Sailboat 128×128 48.53 41.03 34.24 27.87
256×256 48.60 40.88 34.26 27.91
512×512 48.64 40.90 34.23 27.92
Fishing boat 128×128 48.54 40.94 34.22 27.94
256×256 48.63 40.94 34.26 27.94
512×512 48.62 40.91 34.24 27.93
Airplane 128×128 48.55 40.94 34.21 27.91
256×256 48.64 40.90 34.22 27.91
512×512 48.64 40.92 34.24 27.93
Aerial 128×128 48.64 40.91 34.23 27.92
256×256 48.60 40.90 34.25 27.93
512×512 48.65 40.93 34.23 27.94
Tank 128×128 48.70 40.95 34.20 27.89
256×256 48.62 40.89 34.21 27.92
512×512 48.64 40.90 34.21 27.93
Stream and Bridge 128×128 48.60 40.89 34.23 27.93
256×256 48.66 40.94 34.22 27.93
512×512 48.66 40.93 34.23 27.98
C0201 128×128 48.62 40.91 34.23 27.95
256×256 48.63 40.89 34.29 27.93
512×512 48.61 40.94 34.23 27.99
C0202 128×128 48.63 40.94 34.24 27.96
256×256 48.53 41.03 34.24 27.87
512×512 48.60 40.88 34.26 27.91
C0204 128×128 48.64 40.90 34.23 27.92
256×256 48.54 40.94 34.22 27.94
512×512 48.55 40.94 34.21 27.91
C0205 128×128 48.64 40.90 34.22 27.91
256×256 48.64 40.92 34.24 27.93
512×512 48.64 40.91 34.23 27.92
C0208 128×128 48.60 40.90 34.25 27.93
256×256 48.65 40.93 34.23 27.94
512×512 48.70 40.95 34.20 27.89
C0209 128×128 48.62 40.89 34.21 27.92
256×256 48.64 40.90 34.21 27.93
512×512 48.60 40.89 34.23 27.93
C0213 128×128 48.62 40.98 34.28 27.87
256×256 48.61 40.93 34.26 27.91
512×512 48.62 40.93 34.23 27.93
C0214 128×128 48.67 40.87 34.32 27.90
256×256 48.62 40.89 34.21 27.93
512×512 48.62 40.92 34.24 27.93
C0215 128×128 48.59 40.91 34.27 27.95
256×256 48.65 40.94 34.27 27.95
512×512 48.62 40.91 34.23 27.95
C0218 128×128 48.63 40.89 34.29 27.93
256×256 48.63 40.89 34.29 27.93
512×512 48.61 40.94 34.23 27.99
Average Case 128×128 48.60 40.95 34.23 27.94
256×256 48.62 40.96 34.23 27.92
512×512 48.63 40.92 34.22 27.93
14502 Multimedia Tools and Applications (2021) 80:14495–14520

 
l k 2l−k
tl ¼ ∑ pk ð1Þ
k¼0 2l−k l−k

where, for all l, 0 ≤ l ≤ size(P) – 1.


By using equation number (1), CT is applied over 4-pixel groups (i.e. size (P) = 4) to derive
the transformed quadruples as follows:
8
>
> p if i ¼ 0
< 0
p1 if i ¼ 1
ti ¼
>
> p1 þ p2 if i ¼ 2
:
2p1 þ 2p2 þ p3 if i ¼ 3
0
Again, by applying Inverse CT (ICT) one can re-calculate pixel values p0 0 ; p0 1 ; …; pl as
shown in equation number (2):
 
l k
p0 l ¼ ∑ ð−1Þðl−k Þ tk ð2Þ
k¼0 l−k
where, for all l, 0 ≤ l ≤ size(P) – 1.
By using equation number (2), ICT is applied over transformed quadruples to re-compute
the 4-pixel groups as follows: 8
>
> t0 if i ¼ 0
<
0 t1 if i ¼ 1
pi ¼
>
> −t þ t if i¼2
: 1 2
−2t2 þ t3 if i ¼ 3
In case of no embedding, all re-computed pixel values are found to be exactly same corre-
sponding to the pixel values used prior to applying CT i.e., p′i = pi.

5 Proposed method

In this section, we have presented an original piece of work in the field of transform domain
Steganography. We have classified the proposed method into two phases: Embedding and
Extraction. The benefits of transform domain methods and the utility of CT have already been
discussed. CT is used to convert each 2 × 2 non-overlapping blocks of the Cover image from
spatial domain into transformed domain. The basic idea of CT is to generate an integer
polynomial sequence in coefficient form using the pixel values by means of a set of additions
and multiplications. The secret bits, obtained from the Secret image, are embedded into the
transformed coefficients instead of the pixel values to accomplish high robustness against
common signal processing attacks. The Embedding phase offers variable payload ranging
from 1 bpp to 4 bpp besides pertaining good visual clarity in Stego-images. Further, the
extraction phase is used to extract the secret image from the Stego-image.
The Embedding and Extraction phases are elaborated in Section 5a and Section 5b. On the
other hand, an example is given in Section 5c for easy understanding of the proposed method.

5.1 Embedding

In this method, the Cover image (IC) of dimension m × n is decomposed into 2 × 2 non-
overlapping blocks where the pixel values p0, p1, p2 and p3are arranged as one dimensional
Multimedia Tools and Applications (2021) 80:14495–14520 14503

sequence named as 4-pixel group. CT is then applied on each pixel group to convert the same
into transform domain. Secret bit-stream s is obtained from the Secret image (IH). To
accomplish an average payload of n bpp, 4n numbers of bits from secret bit stream s (as
obtained from IH) are embedded into the transformed quadruples (t0, t1, t2 and t3). ICT is
applied over the transformed cum embedded quadruples to get back Stego-pixels in the spatial
domain. Above steps are repeated until and unless the secret bit-stream gets fully embedded
and the Stego image (IS) is produced.

Figure 1 depicts the flow diagram of the embedding process in great detail.

5.2 Extraction

The Stego image (IS) of dimension m × n is decomposed into 2 × 2 non-overlapping blocks


where the stego pixels p′0, p′1, p′2 and p′3 are arranged as one dimensional sequence named as
4-pixel group. CT is then applied on each pixel group to convert the same into transform
14504 Multimedia Tools and Applications (2021) 80:14495–14520

Start

Input: Cover image (IC) and


Hidden image (IH)

Decompose IC into 2×2-image blocks Convert IH into 1D-array of


bits (s)

Construct each 4-pixel group

Apply Catalan Transform (CT)

Obtain transformed quadruples


( , , , )

Compute decimal value E


Compute F = ( + 2 × + 2 × +2 × ) mod 2 from 4n-bits of s

D=E-F

.D< -2 .D>2
D?
Otherwise
4 D=D-2
D=D+2

Set D3 = D / 2 and D = D mod 2


Set D2 = D / 2 and D=D mod 2
Set D1 = D / 2 and D0 =D mod 2

( , , , )=
( + D0, +D1, + D2, + D3)

Apply Inverse Catalan Transform (ICT)

All secret bits No


embedded?

Yes
Stego-image (IS)

Stop

Fig. 1. Flow diagram of the embedding process


Multimedia Tools and Applications (2021) 80:14495–14520 14505

domain. For an average payload of n bpp, 4n numbers of secret bits are extracted from the
transformed quadruples. ICT is applied to get back the 4-pixel group in spatial domain. This
process is repeated until and unless the entire secret bit-stream gets extracted and the Secret
image (IH) is reproduced.

Figure 2 depicts the flow diagram of the extraction process in great detail.

5.3 Example

In this section, we have explained the proposed technique with an example.

i. In case of Embedding,

Step 1. Initially, the secret image is converted into 1D array of bit-stream s.


Step 2. Let us assume the first 2 × 2 non-overlapping block of the cover image is,
 
100 200
100 100

Hence the 4-pixel group is, (p0, p1, p2, p3) = (100, 200, 100, 100)

Step 3. By applying CT, the obtained transformed components are:


ðt 0 ; t1 ; t 2 ; t3 Þ ¼ ð100; 200; 200 þ 100; 2  ð200 þ 100Þ þ 100Þ
¼ ð100; 200; 300; 700Þ

Step 4. For an average payload of 3 bpp, we embed 12-bits from secret bit-stream s and let’s
say it is 000001100100. The decimal equivalent corresponding to 12-bits of secret
bit-stream is (100)10 i.e., E = 100.
Step 5. The reference value is calculated as follows:

F = (100 + 200 * 8 + 300 * 64 + 700 * 512) mod 4096 = 2468


14506 Multimedia Tools and Applications (2021) 80:14495–14520

Start

Input: Stego image (IS)

Decompose IS into 2×2-image


blocks

Construct each 4-pixel group

Apply Catalan Transform (CT)

Obtain transformed quadruples


( , , , )

F =( +2 × +2 × +2 × ) mod 2

Represent F through 4 bits

All secret bits


extracted?
N
Y

Construct 1D-array of secret bit-stream (s)

Output: Secret image (IH)

Stop

Fig. 2 Flow diagram of the extraction process

Step 6. The deviation value is computed as follows:

D = E – F = − 2368;

Step 7. Since, D < −2048 (i.e. -211)

Set D = D + 212 = −2048 + 4096 = +1728;

Step 8. Further D is decomposed into D3, D2, D1 and D0 as follows:


Set D3 = + (1728/512) = +3 and D = + (1728mod 512) = +192.
Multimedia Tools and Applications (2021) 80:14495–14520 14507

Set D2 = + (192/64) = +3 and D = + (192mod 64) = 0.


Set D1 = 0 and D0 = 0
Step 9. On embedding, the modified embedded cum transformed components become:
ðt0 0 ; t0 1 ; t0 2 ; t0 3 Þ ¼ ð100 þ 0; 200 þ 0; 300 þ 3; 700 þ 3Þ ¼ ð100; 200; 303; 703Þ

Step 10. By applying ICT, the re-computed 4-pixel group become:


(p′0, p′1, p′2, p′3) = (100, 200, 103, 97)

The 2 × 2 non-overlapping IS block becomes:


 
100 200
103 97

ii. In case of extraction,

Step 1. The 2 × 2 non-overlapping IS block is taken as the input:


 
100 200
103 97

Hence the 4-pixel group is, (p′0, p′1, p′2, p′3) = (100, 200, 103, 97)

Step 2. By applying CT, the obtained transformed components are:

ðt 0 0 ; t 0 1 ; t0 2 ; t0 3 Þ ¼ ð100; 200; 200 þ 103; 2  ð200 þ 103Þ þ 97Þ

¼ ð100; 200; 303; 703Þ

Step 3. The reference value F′ = (t′0+ 2n×t′1+ 22n×t′2+23n×t′3) mod 24n


= (100 + 200*8 + 303*64 + 703*512) mod 4096.
= 100
Step 4. For 3 bpp, we extract 12-bits and hence, F′ is converted into 12-bits of binary stream
as 000001100100. It is updated into the empty secret bit-stream s.

The extraction process ensures the regeneration of the same secret bit-stream which was
fabricated during embedding.

6 Result, comparison and analysis

The experiment deals with 10 Gy-scale images from USC-SIPI database [44] such as (i) Lena,
(ii) Baboon, (iii) Pepper, (iv) Splash, (v) Sailboat, (vi) Stream and bridge, (vii) Tank, (viii)
Fishing boat, (ix) Aerial and (x) Airplane respectively and 10 Gy-scale images of RGB-BMP
Steganalysis Dataset [7] such as (xi) C0201, (xii) C0202, (xiii) C0204, (xiv) C0205, (xv)
C0208, (xvi) C0209, (xvii) C0213, (xviii) C0214, (xix) C0215 and (xx) C02018 respectively.
For each image, three different dimensions such as 512 × 512, 256 × 256 and 128 × 128 have
been considered. The secret image (i.e., Lena) is resized as per the convenience and is being
14508 Multimedia Tools and Applications (2021) 80:14495–14520

embedded within the cover images depicted in Fig. 3. Metrics such as PSNR, SSIM, MSE,
NCC and payload are used to determine the performance of the proposed scheme. The
variation of quality level of the IS with respect to increasing payload is the fundamental study
in our scheme.
The term PSNR is used as the ratio of quality measurement between IC and IS. The PSNR
(in dB) is defined as follows:
 
255
PSNR ¼ 20log 10 ð3Þ
√MSE
Here, the term MSE is represented as the average squared difference between each cover pixel
and its stego version. It is computed as the summations of the squared differences of all the
pixels followed by dividing it with the total number of pixels. Mathematically, the MSE can be
defined as follows:

(i) Lena (ii) Baboon (iii) Pepper (iv) Splash (v) Sailboat

(vi) Stream and (vii) Tank (viii) Fishing boat (ix) Aerial (x) Airplane
bridge

(xi) C0201 (xii) C0202 (xiii) C0204 (xiv) C0205 (xv) C0208

(xvi) C0209 (xvii) C0213 (xviii) C0214 (xix) C0215 (xx) C0218

Fig. 3 Different grayscale cover images of dimension 512 × 512


Multimedia Tools and Applications (2021) 80:14495–14520 14509

1 M −1 N −1  2
MSE ¼ ∑ ∑ IC i; j −IS i; j ð4Þ
MN i¼0 j¼0

SSIM is used to measure the similarity in the structure of IC and IS. It is mathematically
expressed as follows:
ð2μIC μIS þ c1 Þð2σIC:IS þ c2 Þ
SSIM ðIC; IS Þ ¼   ð5Þ
μ2IC þ μ2iIS þ c1 σ2IC þ σ2IS þ c2

where, μIC, μIS, σIC, σIS and σIC. IS are mean of IC and IS, standard deviation of IC and IS, the
covariance of IC and IS, respectively. Two variables c1 and c2 are constants which are used to
stabilize the division with weak denominator.
NCC, the metric to evaluate the degree of similarity (or dissimilarity) between IC and IS, is
calculated as follows:
M N M N
NCC ¼ ∑ ∑ IC j;k :IS j;k = ∑ ∑ IC 2 j;k ð6Þ
j¼1 k¼1 j¼1 k¼1

On the other hand, the payload is represented by the unit bits per pixel (abbreviated as bpp).
In Table 2, PSNR values are obtained with respect to 1, 2, 3 and 4 bpp of payload. Here, the
average PSNR obtained by our method lies in between 27.92 dB to 48.63 dB despite the
PSNR value greater than 30 dB is considered as good quality IS. Therefore, it is evident from
the results that the average PSNR at 4 bpp is around 27.9 dB i.e., below 30 dB, but still
considered by keeping 4 bpp of payload as key concern. Alternatively, if PSNR is the point of
concern to us, then we may choose payload of 1, 2 and 3 bpp only since the average PSNR
values at these payloads ranges from 34.22 dB to 48.63 dB. It is also observed from Table 2
that the variation in dimension of cover images has minimal impact on the deviation of the
PSNR values. In other words, we can say that the proposed method offers uniform PSNR
values for all 10 images, and as the payload increases, corresponding PSNR value reduces.
Other than PSNR, two popular metrics used widely in the field of image steganography to
analyse the quality of images are SSIM and NCC. Both metrics are exploited to evaluate the
magnitude of similarity between cover and ISs. The permissible values for both SSIM and
NCC ranges from 0 to 1. In Table 3 and Table 4, SSIM and NCC values of the ISs are reported
respectively from which one can easily verify that the values are close to unity and therefore,
the structural distortions are very less.
The performance of the proposed scheme is compared with other state-of-the-art methods
such as Seyyedi’s method based on IWT [40], WNLS-IWT [37], DE–DWT [8], MDLE-IWT
(T = 6) [22], IWT domain scheme [6], Atta’s scheme [38]. Results of this comparison have
been summarized in Table 5. Five well-known benchmark images namely ‘Lena’, ‘Baboon’,
‘Pepper’, ‘Airplane’ and ‘Sailboat’ have been used for this purpose. Since, the proposed
scheme offers variable payload, four PSNR values have been computed for each image at
n = 1, 2, 3, 4. In comparison with Seyyedi’s method [40], our proposed scheme offers 0.48 bpp
of enhanced average payload and 0.45 dB of average PSNR improvement while n = 2 is
considered. It is also seen from Table 5 that our method outperforms the WNLS-IWT [37] in
terms of payload and image quality because our method provides 8 times more bpp and on
average 8 dB more PSNR when we consider bpp n = 1. The average-case payloads of Atta’s
scheme [38] are 1.76 bpp and 2.5 bpp respectively for given k = 2 and k = 3. The proposed
scheme is quite similar with Atta’s scheme in terms of payload for all bpp values. Regardless
14510 Multimedia Tools and Applications (2021) 80:14495–14520

Table 3 SSIM values of proposed Steganographic scheme with respect to 1, 2, 3 and 4 bpp of payloads

IC Dimensions of IC Block size=2×2

SSIM w.r.t. n bpp of payload

n=1 n=2 n=3 n=4

Lena 128×128 0.99 0.97 0.91 0.75


256×256 0.99 0.96 0.86 0.65
512×512 0.99 0.95 0.84 0.59
Baboon 128×128 0.99 0.96 0.85 0.78
256×256 0.99 0.95 0.89 0.67
512×512 0.99 0.96 0.84 0.56
Pepper 128×128 0.99 0.95 0.85 0.59
256×256 0.99 0.97 0.89 0.78
512×512 0.99 0.96 0.92 0.67
Splash 128×128 0.99 0.95 0.88 0.74
256×256 0.99 0.97 0.84 0.65
512×512 0.99 0.95 0.87 0.58
Sailboat 128×128 0.99 0.96 0.89 0.79
256×256 0.99 0.95 0.85 0.67
512×512 0.99 0.97 0.89 0.74
Fishing boat 128×128 0.99 0.96 0.90 0.78
256×256 0.99 0.95 0.86 0.67
512×512 0.99 0.97 0.84 0.74
Airplane 128×128 0.99 0.95 0.85 0.65
256×256 0.99 0.96 0.89 0.58
512×512 0.99 0.95 0.84 0.78
Aerial 128×128 0.99 0.97 0.85 0.67
256×256 0.99 0.96 0.89 0.72
512×512 0.99 0.95 0.92 0.75
Tank 128×128 0.99 0.97 0.86 0.78
256×256 0.99 0.97 0.84 0.67
512×512 0.99 0.95 0.85 0.74
Stream and Bridge 128×128 0.99 0.96 0.89 0.65
256×256 0.99 0.95 0.84 0.58
512×512 0.99 0.97 0.85 0.78
C0201 128×128 0.99 0.97 0.89 0.78
256×256 0.99 0.96 0.92 0.67
512×512 0.99 0.95 0.88 0.74
C0202 128×128 0.99 0.97 0.84 0.65
256×256 0.99 0.95 0.87 0.58
512×512 0.99 0.96 0.89 0.79
C0204 128×128 0.99 0.95 0.85 0.67
256×256 0.99 0.97 0.89 0.74
512×512 0.99 0.97 0.84 0.67
C0205 128×128 0.99 0.95 0.85 0.74
256×256 0.99 0.96 0.89 0.65
512×512 0.99 0.95 0.84 0.58
C0208 128×128 0.99 0.97 0.85 0.78
256×256 0.99 0.97 0.89 0.78
512×512 0.99 0.96 0.92 0.67
C0209 128×128 0.99 0.95 0.88 0.74
256×256 0.99 0.97 0.84 0.65
512×512 0.99 0.95 0.87 0.58
C0213 128×128 0.99 0.96 0.89 0.79
256×256 0.99 0.95 0.85 0.67
512×512 0.99 0.96 0.89 0.65
C0214 128×128 0.99 0.95 0.84 0.58
Multimedia Tools and Applications (2021) 80:14495–14520 14511

Table 3 (continued)

IC Dimensions of IC Block size = 2 × 2

SSIM w.r.t. n bpp of payload

n=1 n=2 n=3 n=4

256×256 0.99 0.97 0.85 0.78


512×512 0.99 0.97 0.89 0.78
C0215 128×128 0.99 0.96 0.92 0.67
256×256 0.99 0.95 0.88 0.74
512×512 0.99 0.97 0.84 0.65
C0218 128×128 0.99 0.95 0.87 0.58
256×256 0.99 0.96 0.89 0.79
512×512 0.99 0.95 0.85 0.67
Average Case 128×128 0.99 0.95 0.89 0.73
256×256 0.99 0.96 0.92 0.74
512×512 0.99 0.95 0.89 0.64

of this payload enhancement, the average PSNR is slightly compromised; however, it never
falls below the acceptable level [i.e., 30 dB]. With another recent work MDLE-IWT (T = 6)
[22] we have compared our method in terms of PSNR and payload. It shows that payload is
0.17 bpp higher for us, which means 44,564 bits are more embedded by our method in 512 ×
512 image but PSNR decreases by less than 2 bpp on an average. If we compare our method
with Jothy et al. [23], then it is easily seen that our method provides approximately 4 dB more
PSNR in double payload. Our method gives approximately 5 dB greater PSNR in almost the
same bpp of payload shown in Kalita et al. [24]. From Table 5, it can be easily shown that our
method provides almost 4 times payload in almost the same PSNR when Nazari et al. [34] is
compared with our scheme. This comparative assessment ensures that our method outperforms
the existing schemes in terms of both average payload and average PSNR.

7 Security analysis

This section is very important as here we judge the efficacy of our proposed Steganographic
method. We use the StegExpose tool [12], which executes a fusion of many well-known
steganalysis methods [26] to determine to a result which assists us to take decision whether a
given IS is above the threshold or not. StegExpose performs a fast fusion of all the techniques
mentioned in Table 6 by only looking at suspicious files.
Further, as the results shown in Table 7, out of the total 20 images, all lies below the stego-
threshold. What this concludes is that the proposed algorithm creates the ISs which are robust
to steganalysis attacks and can assist external encroachment.
We have also made histogram analysis on ‘Lena’ image of dimension 128 × 128. The
histograms of cover image and all ISs at different values of payload are shown in Fig. 4. It is
seen from the histograms that no huge change occurs in histograms of the ISs with respect to
the cover image. The similarity measure, the degree of changes made in the distribution of
pixel intensities, is computed for n = 1, n = 2, n = 3 and n = 4 which ensured that the values of
the similarity measure are 0.9981, 0.9893, 0.9802 and 0.9730 respectively. Since, all these
values are close to unity, the ISs are seemingly identical with the cover image.
14512 Multimedia Tools and Applications (2021) 80:14495–14520

Table 4 NCC values of proposed Steganographic scheme with respect to 1, 2, 3 and 4 bpp of payloads

IC Dimensions of IC Block size=2×2

NCC w.r.t. n bpp of payload

n=1 n=2 n=3 n=4

Lena 128×128 0.99 0.99 0.99 0.97


256×256 0.99 0.99 0.99 0.97
512×512 0.99 0.99 0.99 0.97
Baboon 128×128 0.98 0.99 0.98 0.97
256×256 0.98 0.98 0.98 0.96
512×512 0.99 0.99 0.99 0.96
Pepper 128×128 0.98 0.99 0.99 0.97
256×256 0.99 0.98 0.98 0.98
512×512 0.98 0.97 0.99 0.97
Splash 128×128 0.99 0.98 0.99 0.97
256×256 0.99 0.98 0.97 0.98
512×512 0.98 0.97 0.99 0.97
Sailboat 128×128 0.99 0.99 0.97 0.96
256×256 0.97 0.98 0.99 0.97
512×512 0.98 0.99 0.98 0.98
Fishing boat 128×128 0.98 0.97 0.99 0.98
256×256 0.97 0.98 0.99 0.97
512×512 0.99 0.99 0.99 0.97
Airplane 128×128 0.99 0.98 0.98 0.97
256×256 0.99 0.97 0.99 0.98
512×512 0.98 0.98 0.97 0.97
Aerial 128×128 0.98 0.99 0.98 0.97
256×256 0.99 0.97 0.99 0.97
512×512 0.98 0.98 0.99 0.98
Tank 128×128 0.99 0.99 0.98 0.97
256×256 0.98 0.98 0.99 0.98
512×512 0.97 0.99 0.98 0.97
Stream and Bridge 128×128 0.99 0.99 0.97 0.98
256×256 0.98 0.97 0.98 0.98
512×512 0.99 0.98 0.97 0.97
C0201 128×128 0.99 0.99 0.99 0.97
256×256 0.99 0.99 0.99 0.97
512×512 0.98 0.99 0.98 0.97
C0202 128×128 0.98 0.98 0.98 0.96
256×256 0.99 0.99 0.99 0.96
512×512 0.98 0.99 0.99 0.97
C0204 128×128 0.99 0.98 0.98 0.98
256×256 0.98 0.97 0.99 0.97
512×512 0.99 0.97 0.99 0.97
C0205 128×128 0.98 0.98 0.99 0.98
256×256 0.99 0.99 0.98 0.97
512×512 0.98 0.98 0.99 0.98
C0208 128×128 0.97 0.99 0.98 0.97
256×256 0.99 0.99 0.97 0.98
512×512 0.98 0.97 0.98 0.98
C0209 128×128 0.99 0.98 0.97 0.97
256×256 0.99 0.99 0.99 0.97
512×512 0.99 0.99 0.99 0.97
C0213 128×128 0.98 0.99 0.98 0.97
256×256 0.98 0.98 0.98 0.96
512×512 0.99 0.99 0.99 0.96
C0214 128×128 0.99 0.99 0.97 0.96
Multimedia Tools and Applications (2021) 80:14495–14520 14513

Table 4 (continued)

IC Dimensions of IC Block size = 2 × 2

NCC w.r.t. n bpp of payload

n=1 n=2 n=3 n=4

256×256 0.97 0.98 0.99 0.97


512×512 0.98 0.99 0.98 0.98
C0215 128×128 0.98 0.97 0.99 0.98
256×256 0.97 0.98 0.99 0.97
512×512 0.98 0.98 0.99 0.98
C0218 128×128 0.97 0.99 0.98 0.97
256×256 0.98 0.98 0.99 0.98
512×512 0.97 0.99 0.98 0.97
Average Case 128×128 0.99 0.99 0.98 0.97
256×256 0.99 0.99 0.97 0.98
512×512 0.99 0.98 0.99 0.97

8 Time complexity analysis

We have analysed the complexity of the proposed method in this section. Gashkov et al. [17] proved
that the calculation of inner multiplication and addition of CT in the additive basis B+ is bounded
with upper bound O(z × log(z)) where m denotes the size of the pixel block (here z = 4). The basis
containing all monotonic linear functions {ax+by: a, b ∈Real} has lower bounds of the particular
form Ω(z × log(z)). Nevertheless, the computational complexity (T) of the proposed scheme is
obtained by decomposing the IC having F columns and G rows into non-overlapping blocks of size
2 × 2 and then adding up the time complexities of the various stages sorted below:
Embedding time complexity: In case of embedding, we have analysed the time complex-
ity step by step as follows:

i. Apply CT to each 2 × 2 block of IC to generate transformed components: O(z × log(z)).


ii. For an embedding rate of n bpp, we choose 4n bits from the secret bit stream for
embedding within transformed components. The complexity of converting 4n bits into
decimal: O(n) ≈O(z) as n ≤ z.
iii. Computation of reference value F: O(1).
iv. Computation of deviation value D: O(1).
v. Computation of D3, D2, D1, and D0 from D: O(1).
vi. Computation of transformed embedded components: O(n) ≈O(z) as n ≤ z.
vii. Apply ICT to get back 2 × 2 block of IS in spatial domain: O(z × log(z)).

The embedding time complexity in worst-case is represented as,


 
FG
T¼O  maximumfOðzÞ; Oðz  logðzÞÞ; Oð1Þ; Oð1Þ; Oð1Þ; OðzÞ; Oðz  logðzÞÞg
 z 
FG
¼O  z  logðzÞ:
z
¼ Oð F  G  logðzÞÞ
14514 Multimedia Tools and Applications (2021) 80:14495–14520

Table 5 Performance comparison of the proposed Steganographic method with some state-of-the-art methods in
terms of payload and PSNR

Image Method Payload (bpp) PSNR (dB)

Lena Seyyedi’s method based on IWT [40] 1.52 40.54


WNLS-IWT [37] 0.125 40.40
DE–DWT [8] 0.75 47.35
MDLE-IWT (T=6) [22] 1.83 42.79
IWT domain scheme [6] 0.19 53.12
Atta’s scheme [38] 1.7(k=2) 44.91
2.5(k=3) 38.77
Jothy et al. [23] 0.5 44.2
Kalita et al. [24] 1.14 43.9
Nazari et al. [34] 0.24 49.65
Proposed scheme 1 48.62
2 40.93
3 34.23
4 27.93
Baboon Seyyedi’s method based on IWT [40] 1.52 38.07
WNLS-IWT [37] 0.125 30.15
DE–DWT [8] 0.75 47.35
MDLE-IWT (T=6) [22] 1.8 42.83
IWT domain scheme [6] 0.58 47.01
Atta’s scheme [38] 1.84(k=2) 44.59
2.5(k=3) 39.15
Jothy et al. [23] 0.5 44.8
Kalita et al. [24] 1.14 43.9
Nazari et al. [24] 0.19 49.65
Proposed scheme 1 48.62
2 40.92
3 34.24
4 27.93
Pepper Seyyedi’s method based on IWT [40] 1.52 40.64
WNLS-IWT [37] 0.125 39.56
DE–DWT [8] 0.75 47.34
MDLE-IWT (T=6) [22] 1.83 42.77
IWT domain scheme [6] 0.24 52.84
Atta’s scheme [38] 1.77(k=2) 45
2.5(k=3) 39.10
Jothy et al. [23] 0.5 44.3
Kalita et al. [24] 1.14 43.9
Nazari et al. [34] 0.28 45.52
Proposed scheme 1 48.62
2 40.91
3 34.23
4 27.95
Airplane Seyyedi’s method based on IWT [40] 1.52 40.18
WNLS-IWT [37] 0.125 40.77
DE–DWT [8] 0.75 47.36
MDLE-IWT (T=6) [22] 1.85 42.58
IWT domain scheme [6] 0.07 57.48
Jothy et al. [23] 0.5 44.4
Kalita et al. [24] 1.14 43.9
Nazari et al. [34] 0.68 41.59
Atta’s scheme [38] 1.80(k=2) 44.44
2.5 (k=3) 38.02
Proposed scheme 1 48.64
2 40.92
Multimedia Tools and Applications (2021) 80:14495–14520 14515

Table 5 (continued)

Image Method Payload (bpp) PSNR (dB)

3 34.24
4 27.93
Sailboat Seyyedi’s method based on IWT [40] 1.52 39.40
WNLS-IWT [37] 0.125 37.86
DE–DWT [8] 0.75 47.34
MDLE-IWT (T=6) [22] 1.83 42.89
IWT domain scheme [6] 0.313 51.16
Atta’s scheme [38] 1.7(k=2) 44.59
2.5(k=3) 38.32
Jothy et al. [23] 0.5 44.2
Kalita et al. [24] 1.14 43.9
Nazari et al. [34] 0.36 47.99
Proposed scheme 1 48.64
2 40.90
3 34.23
4 27.92

Best-case yields only one z-pixel group i.e., (z/2 × z/2) block of IC and hence,
 
F1
T¼O  z  logðzÞ
z
¼ O ð F  logðzÞÞ:

Extraction time complexity: In this portion, we have explained the time complexity of
extraction phase step by step as follows.

i. At the receiver end, apply CT to each 2 × 2 block of IS to generate transformed compo-


nents: O(z × log(z)).
ii. Computation of reference value F: O(1).
iii. Computation of binary equivalent of F in 4n bits: O(n) ≈O(z) as n ≤ z.

Table 6 Brief description of various steganalysis methods

Method Description

Primary Sets [20] It recognizes LSB embedding by forming subsets of pixels whose cardinality differ due to
insertion of hidden message.
Sample pair analysis In this method, selected multi-sets form the state of a finite state machine as its imprecise
[14] change of state causes between these multi-sets based on LSB flipping.
RS analysis [13] It differentiates between the number of singular and regular groups for the LSB and shifted
LSB plane.
Chi-square attack It is a simple and famous method to test the robustness of a security system against the
[42] attacks. It depends on probability analysis of IS after putting the information inside it
using LSB method, then compare it with the probability analysis of original image to
check the difference between them. If the difference is near zero, it means no
information inside the image, otherwise if it is near one, it implies that the image holds
the information.
14516 Multimedia Tools and Applications (2021) 80:14495–14520

Table 7 Results of security analysis on different IS with respect to different attacks

IS Payload Steganalysis Attacks


(bpp)
RS Fusion Sample Chi Primary Secret message Above stego
analysis (mean) pairs square Sets size in bytes threshold?

Lena 1 0.022064 0.019207 0.029087 0.002092 0.023585 1685 No


2 0.009641 0.01229 0.009954 0.001298 0.028266 1078 No
3 0.045242 0.036241 0.012931 0.0013 0.08549 3180 No
4 0.046598 0.066774 0.123607 0.001648 0.095241 5859 No
Baboon 1 0.136485 0.112198 0.164013 0.001304 0.146989 9844 No
2 0.158507 0.105652 0.147533 1.11E-04 0.116459 9270 No
3 0.097686 0.059338 0.068265 1.51E-13 0.071401 5206 No
4 0.097081 0.076777 0.097057 4.34E-19 0.112971 6736 No
Pepper 1 0.025974 0.018985 0.032381 3.24E-04 0.01726 1666 No
2 0.029385 0.015233 0.028863 0.001256 0.001428 1337 No
3 0.060253 0.067079 0.067259 0.001251 0.139553 5886 No
4 0.105497 0.086687 0.153261 0.001304 NaN 7606 No
Sailboat 1 0.040796 0.040962 0.037959 9.11E-09 0.085092 3594 No
2 0.067146 0.059083 0.078104 1.08E-07 0.09108 5184 No
3 0.110545 0.080996 0.132398 4.53E-05 NaN 7107 No
4 0.031109 0.019128 0.026272 1.54E-06 NaN 1678 No
Splash 1 0.061707 0.049716 0.065968 2.50E-04 0.070939 4362 No
2 0.02384 0.007363 0.003392 8.54E-04 0.001368 646 No
3 0.076192 0.061907 0.10854 9.91E-04 NaN 5432 No
4 0.022481 0.034302 0.080055 3.71E-04 NaN 3010 No

The extraction time complexity in worst-case is represented as,


 
FG
T¼O  maximumfOðz  logðzÞÞ; Oð1Þ; OðzÞg
 z 
FG
¼O  z  logðzÞ
z
¼ Oð F  G  logðzÞÞ:

Best-case yields only one z-pixel group i.e., (z/2 × z/2) block of IC and hence,
 
F1
T ¼O  z  logðzÞ ¼ O F  logðzÞ:
z

IC Block size= 2 × 2
Histogram
n=1 n=2 n=3 n=4

200 200 200 180 180

180 180 180 160 160


160 160 160
140 140
140 140 140
120 120
120
No. of pixels

120 120
No. of pixels

No. of pixels

No. of pixels

No. of pixels

100 100
100 100 100
80 80
80 80 80
60 60
60 60 60

40 40 40
40 40

20 20 20 20 20

0 0 0 0 0
−50 0 50 100 150 200 250 300 −50 0 50 100 150 200 250 300 −50 0 50 100 150 200 250 300 −50 0 50 100 150 200 250 300 −50 0 50 100 150 200 250 300
Intensity Levels Intensity Levels Intensity Levels Intensity Levels Intensity Levels

Lena (128×128) Similarity Measure: Similarity Measure: Similarity Measure: Similarity Measure:
0.9981 0.9893 0.9802 0.9730

Fig. 4 Depicts the global histograms and similarity scores of Lena IC and ISs
Multimedia Tools and Applications (2021) 80:14495–14520 14517

Table 8 Embedding/Extraction time of proposed steganographic scheme in terms of different values of payload

IC Dimensions of IC Block size=2×2

Embedding/Extraction time w.r.t. n bpp of payload

n =1 n =2 n=3 n=4

Lena 128×128 0.64/ 0.65/ 0.70/ 0.72/


0.51 0.62 0.74 0.96
256×256 1.04/ 1.23/ 1.31/ 1.34/
0.74 1.10 1.36 1.44
512×512 1.52/ 1.76/ 1.78/ 2.24/
1.35 1.59 1.86 2.12
Baboon 128×128 0.62/ 0.64/ 0.68/ 0.78/
0.58 0.72 0.76 1.03
256×256 1.12/ 1.21/ 1.40/ 1.52/
0.94 1.01 1.37 1.43
512×512 1.52/ 1.69/ 1.85/ 2.24/
1.45 1.68 1.84 2.16

In Table 8, we have summarized the embedding and extraction time respectively for different
size of “Lena” and “Baboon” with respect to increasing payload. It gives us an understanding
of execution time of our method with respect to different ICs at 1, 2, 3 and 4 bpp of payload.
Further, to prove the superiority of proposed method, we have computed the results of basic
DCT, DFT and DWT amalgamating with LSB replacement process and compared it with our
proposed method in terms of PSNR, payload and computational time. Here, “Lena” image of
size 512 × 512 is used for IC and resized “Male” image is used for IH. In Table 9, we have
summarized the comparative results whereas the visual comparison of ISs generated by DCT,
DFT and DWT along with our proposed method is shown in Fig. 5.

9 Conclusion

In this paper, a novel Steganographic method constructed on an integral sequence called CT


has been proposed. The basic motivation of exploiting CT is the pliability of concerning
integer based computation which ensures faster handling and without mislaying any informa-
tion. Simulation results substantiate the efficacy of the suggested scheme over some state-of-
the-art scheme in terms of payload, Stego-image quality or both. In our future work, we plan to
investigate some other integral sequence based transforms which are not utilized in the domain

Table 9 Comparison of the proposed method with some transform based techniques DCT, DFT and DWT

Transform used for embedding No. of bits embedded PSNR(dB) Embedding Time (s) Extraction Time (s)

DFT [36] 400 41.40 1.03 1.14


DWT [29] 131,072 40.6 1.21 1.80
DCT [25] 262,144 20.25 1.72 2.01
Proposed method 400 60.62 0.82 0.19
131,072 50.13 1.14 0.79
262,144 48.62 1.52 1.43
14518 Multimedia Tools and Applications (2021) 80:14495–14520

DCT DFT DWT Proposed Method

PSNR : 20.25 dB( 1 PSNR : 41.4 dB PSNR : 40.6 dB (0.5 PSNR : 48.62 dB (1
bpp) (0.001 bpp) bpp) bpp)
Fig. 5 Comparison of ISs obtained through DCT, DFT, DWT and our proposed method

of Steganography till date. Some optimization techniques such as Genetic Algorithm (GA),
Particle Swarm optimization (PSO) and Grey Wolf Optimization (GWO) etc. [16] may also be
investigated in steganography domain to improve the quality and security aspects.

References

1. Abdulla AA (2015) Exploiting similarities between secret and cover images for improved embedding
efficiency and security in digital steganography. Doctoral thesis, University of Buckingham, URI:http://
bear.buckingham.ac.uk/id/eprint/149
2. Abdulla AA, Sellahewa H, Jassim SA (2014) Steganography based on pixel intensity value decomposition,
Proc. SPIE 9120, Mobile Multimedia/Image Processing, Security, and Applications, https://doi.org/10.
1117/12.2050518
3. Abdulla AA, Sellahewa H, Jassim SA (2019) Improving embedding efficiency for digital steganography by
exploiting similarities between secret and cover images. Multimed Tools Appl 8:17799–17823
4. Akhtar N, Ahamad V, Javed H (2017), A compressed LSB steganography method, Proceedings of 3rd
International Conference on Computational Intelligence & Communication Technology (CICT), 1–7, doi:
https://doi.org/10.1109/CIACT.2017.7977371, Ghaziabad
5. Alafandy KA, El-Rabaie E-SM, Faragallah OS, Elmahalawy A (2016) High security data hiding using
image cropping and LSB least significant bit steganography, in IEEE, Tangier, Morocco
6. Al-Dmour H, Al-Ani A (2016) A steganography embedding method based on edge identification and XOR
coding. Expert Syst Appl 46:293–306
7. Al-Jarrah M (2018) RGB-BMP Steganalysis Dataset, Mendeley Data, version 1, https://doi.org/10.17632/
sp4g8h7v8k.1, (accessed on 14th January, 2020)
8. Atawneh S, Almomani A, Al HB, Sumari P, Gupta B (2017) Secure and imperceptible digital image
steganographic algorithm based on diamond encoding in DWT domain. Multimed Tools Appl 76:18451–
18472
9. Bánoci V, Bugár G, Levický D (2011) A novel method of image steganography in DWT domain,
Proceedings of 21st International Conference Radioelektronika 2011, Brno :1–4, doi: https://doi.org/10.
1109/RADIOELEK.2011.5936455
10. Barry P (2005) A Catalan Transform and Related Transformations on Integer Sequences. J Integer
Sequences 8:1–24
11. Biswas R, Mukherjee S, Bandyopadhyay SK (2013) DCT Domain Encryption in LSB Steganography, in
Proceedings of 5th International Conference and Computational Intelligence and Communication Networks,
Mathura, India
12. Boehm B (2014) StegExpose - A Tool for Detecting LSB Steganography, Multimed ,Cryptography Secur,
1–11
13. Dumitrescu S, Wu X, Memon N (2002) On steganalysis of random LSB embedding in continuous-tone
images. Proceedings of International Conference on Image Processing, IEEE 3:641–644
14. Dumitrescu S, Wu X, Wang Z (2003) Detection of LSB steganography via sample pair analysis. IEEE
Trans Signal Process 51:1995–2007
15. Emam MM, Aly AA, Omara FA (2016) An Improved Image Steganography Method Based on LSB
Technique with Random Pixel Selection. Int J Adv Comput Sci Appl(IJACSA) 7:361–366
16. Gadekallu TR, Khare N, Bhattacharya S et al. (2020) Deep neural networks to predict diabetic retinopathy. J
Ambient Intell Human Comput. https://doi.org/10.1007/s12652-020-01963-7
Multimedia Tools and Applications (2021) 80:14495–14520 14519

17. Gashkov SB (2015) Arithmetic complexity of certain linear transformations. Mathematical Notes 97:531–
555
18. Ghosal SK, Mandal JK (2013) Stirling Transform based Color Image Authentication. Procedia Technol 10:
95–104
19. Ghosal SK, Mukhopadhyay S, Hossain S, Sarkar R (2020) Application of Lah transform for security and
privacy of data through information hiding in telecommunication. Trans Emerging Tel Tech, Wiley :1–20.
https://doi.org/10.1002/ett.3984
20. Golijan M, Rui D (2001) Detecting LSB steganography in color, and gray-scale images. MultimedIEEE 8:
22–28
21. Hajizadeh H, Ayatollahi A , Mirzakuchaki S (2013) A new high capacity and EMD-based image
steganography scheme in spatial domain, in 2013 21st Iranian Conference on Electrical Engineering
(ICEE), Mashhad, Iran
22. Hua Zhang LH (2019) A data hiding scheme based on multidirectional line encoding and integer wavelet
transform. Signal Process: Image Commun 78:331–344
23. Jothy N, Anusuyya S (2016) A Secure Color Image Steganography Using Integer Wavelet Transform. 10th
International Conference on Intelligent Systems and Control (ISCO)., Coimbatore, India. https://doi.org/10.
1109/ISCO.2016.7726948
24. Kalita M, Tuithung T, Majumder S (2019) A New Steganography Method Using Integer Wavelet
Transform and Least Significant Bit Substitution. Comput J 62:1639–1655. https://doi.org/10.1093/
comjnl/bxz014
25. Kamya S (2020) Watermark DCT , MATLAB Central File Exchange. Retrieved July 25, 2020. (https://
www.mathworks.com/matlabcentral/fileexchange/46866-watermark-dct)
26. Kang Y, Liu F, Yang C, Xiang L, Luo X, Wang P (2019) Color image steganalysis based on channel
gradient correlation, Int J Distributed Sens Netw., 15
27. Kim C, Shin D , Shin D , Zhang X (2011) Improved Steganographic Embedding Exploiting Modification
Direction in Multimedia Communications, in FTRA International Conference on Secure and Trust
Computing, Data Management, and Application
28. Kuo W-C (2013) A Data Hiding Scheme Based on Square Formula Fully Exploiting Modification
Directions. Smart Innov Syst Technol 21:131–139. https://doi.org/10.1007/978-3-642-35473-1_14
29. Lai C-C, Tsai C-C (2010) Digital Image Watermarking Using Discrete Wavelet Transform and Singular
Value Decomposition. IEEE Trans Inst Measurement 59:3060–3063
30. Mandal JK, Ghosal SK (2014) Binomial Transform Based Fragile Watermarking for image Authentication.
J Inf Secur Appl 19:272–281
31. Mandal JK, Ghosal S (2014) Color image authentication based on two-dimensional separable discrete
hartley transform. Adv Modelling Analysis B 57:68–87
32. Mandal JK, Ghosal SK (2018) On the use of stirling transform in image Steganography. J inf Secur appl 46.
https://doi.org/10.1016/j.jisa.2018.04.003
33. Meng R, Rice SG, Wang J, Sun X (2018) A fusion steganographic algorithm based on faster R-CNN.
Comput Materials Continua 55:1–16
34. Nazari M, Ahmadi ID (2020) A novel chaotic steganography method with three approaches for color and
grayscale images based on FIS and DCT with flexible capacity. Multimed Tools Appl 79:13693–13724.
https://doi.org/10.1007/s11042-019-08415-1
35. Peng F, Zhou DL, Long M, Sun XM (2017) Discrimination of natural images and computer generated
graphics based on multi-fractal and regression analysis. AEU-International Journal of Electronics and
Communications 71:72–81
36. Poljicak A, Mandic L, Agic D (2011) Discrete Fourier transform-based watermarking method with an
optimal implementation radius. J Electron Imaging 20
37. Raja KB, Reddy HSM (2011) Wavelet based Non LSB Steganography. Int J Adv Netw Appl 03:1203–
1209
38. Randa Atta MG (2018) A High Payload Steganography Mechanism Based on Wavelet Packet
Transformation. J Vis Commun Image R 53:42–54
39. Sathish Shet K, Aswath AR, Hanumantharaju MC, Gao X-Z (2019) Novel high-speed reconfigurable
FPGA architectures for EMD-based image steganography. Multimed Tools Appl volume,Springer 78:
18309–18338
40. Seyyedi SA, Ivanov N (2014) High Payload and secure image steganography method based on block
partioning and integer wavelet transform. Int J Secur Appl 8:183–194
41. Shin N (1999) One-Time Hash Steganography. In: Pfitzmann A. (eds) Information Hiding. IH . Lecture
Notes in Computer Science,Springer, Berlin, Heidelberg, 1768
42. Stanley CA (2005) Pairs of values and the Chi-Squared Attack, Department of Mathematics, Iowa State
University, CiteSteer,1–45
14520 Multimedia Tools and Applications (2021) 80:14495–14520

43. Wang C-C, Kuo W-C, Huang Y-C, Wuu L-C (2017) A high capacity data hiding scheme based on re-
adjusted GEMD. Multimed Tools Appl, 77:6327–6341
44. Weber AG (2019) USC-SIPI Image Database: Version 5, Original release: October 1997, Signal and Image
Processing Institute, University of Southern California, Department of Electrical Engineering. http://sipi.usc.
edu/database/ (accessed on 23rd September, 2019)
45. Xiang L, Wang X, Yang C, Liu P (2017) A Novel Linguistic Steganography Based on Synonym Run-
Length Encoding. IEICE Transactions on Information and Systems.,E100.D:313–322
46. Zhang X, Peng F, Long M (2018) Robust Coverless Image Steganography Based on DCT and LDA Topic
Classification. IEEE Trans Multimed 20:3223–3238

Publisher’s note Springer Nature remains neutral with regard to jurisdictional claims in published maps and
institutional affiliations.

Affiliations

Souradeep Mukhopadhyay 1 & Sabbir Hossain 1 & Sudipta Kr Ghosal 2 & Ram Sarkar 1

Souradeep Mukhopadhyay
souradeepmukhopadhyay99@gmail.com
Sabbir Hossain
deepsabbir1999@gmail.com

Ram Sarkar
ramjucse@gmail.com

1
Department of Computer Science and Engineering, Jadavpur University, Kolkata Pin-700032, India
2
Department of Computer Science & Technology, Nalhati Government Polytechnic, Nalhati, Birbhum, West
Bengal Pin-731243, India

You might also like