Secure image hiding in speech signal by steganography-mining and encryption

Indonesian Journal of Electrical Engineering and Computer Science
Vol. 21, No. 3, March 2021, pp. 1692~1703
ISSN: 2502-4752, DOI: 10.11591/ijeecs.v21.i3.pp1692-1703  1692
Journal homepage: http://ijeecs.iaescore.com
Secure image hiding in speech signal by steganography-mining
and encryption
Amal Hameed Khaleel, Iman Qays Abduljaleel
Computer Science Department, College of Computer Science and Information Technology, University of Basrah, Iraq
Article Info ABSTRACT
Article history:
Received Oct 4, 2020
Revised Dec 6, 2020
Accepted Dec 20, 2020
Information hiding techniques are constantly evolving due to the increased
need for security and confidentiality. This paper proposes a working
mechanism in three phases. The first phase includes scrambling the values of
the gray image depending on a series of keys that are generated using a
quantum chaotic map. The second phase generates hybrid keys by mixing a
Zaslavsky and a 3D Hanon map that are used to encrypt the gray image
values produced after the scramble. Finally, in the third phase, a new
algorithm is suggested to hide the encrypted gray image at random locations
within a speech file. This algorithm includes the LSB algorithm to determine
the hidden bits and the zero-crossing K-means algorithm in selecting
locations mining in a scattered manner so that hackers cannot easily retrieve
the hidden data of any hacked person. Also used a fractional fourier
transform to choose magnitude value as specific data to hide encoded image
data. The measures MSE, PSNR, NSCR, and UACI are using to measure the
work efficiency in the encryption algorithm, and in measuring the efficiency
of the hidden algorithm, use the measures SNR, PSNR, and MSE. The results
of the paper are encouraging and efficient compared to other algorithms that
performed the same work. Hence our results show the larger the image
dimensions used, the better the values.
Keywords:
Chaotic map
Cryptography
Data mining
Scrambling
Speech steganography
This is an open access article under the CC BY-SA license.
Corresponding Author:
Amal Hameed Khaleel
Department of Computer Science
Basrah University, Basrah, Iraq
Email: amal_albahrany@yahoo.com
1. INTRODUCTION
Information hiding, a type of steganography, is one of the evolving techniques of embedding hidden
information in digital media [1]. Digital steganography is the most effective protection technique used to
preserve the privacy and confidentiality of data during transmission [2]. Steganography is an art that conceals
data for any multimedia-based (cover medium) within the signals. It is used for invisible secret
communication between parties by hiding the existence of secret data that changes within other media.
Steganography's main goals include a high payload, enhanced robustness, and improved imperceptibility.
Steganography requirements include a hidden file (text, image, video or audio), a carrier object (cover/host)
media, a hide technique, and often a secret key and encryption technique to improve the security levels [1, 3].
Audio steganography is also one of the paper hotspots of security. Speech steganography is the way to
hide the secret data in an audio medium which makes the secret data unnoticeable intruders [4]. Audio
steganography is the most challenging technique in the world of steganography because the human auditory
system is more extensive than the human visual system (HVS). The auditory system is the second largest source
of information acquisition after the visual system. However, the diverse audio features of audio steganography,

Indonesian J Elec Eng & Comp Sci ISSN: 2502-4752 
Secure image hiding in speech signal by steganography mining… (Amal Hameed Khaleel)
1693
such as loud noises tending to hide the quieter noises, make it a prominent steganography cover medium [5].
Combining a cryptography and steganography approach offers a higher degree of protection as a key is required
to decrypt the secret message [2]. Also used The chaotic system improve robustness because it is a chaotic
series of values used to generate a random series within a specified range. All audio (speech) encryption and
speech steganography are used to secure sensitive information, but the security is somewhat different. Audio
encryption hides the 'content' of the hidden secret signal, while the audio steganography hides the hidden
'behind' the host audio and converts the secret into something understandable [6].
The researchers have suggested many hiding methods and strategies, such as steganography. The
method hiding of the image in the audio signal in [7]. The method is reliable, less complex and computation
time and the discrete cosine transform and skin detection are used. A novel scheme of data hiding in video
and audio files is suggested in [8]. The scheme used 4LSB and a phase encoding algorithm to preserve the
quality of the video file even after secret data embedding. The application of the audio steganography with
rivest cipher 4 (RC4) stream cipher encryption text is designed in [9]. The application is a windows shape
that allows users to hide and recognize their message only in “.wav” audio. New technique is presented in
[10] for data embedding uses a 10-digit number in the speech signal and embedded in frame samples of the
LSB. The unique key is used to encrypt and decrypt signal for more secure transmission. The proposed
method in [11] based on the function of an octal modulus modifying the cover samples. Three test
implementations are performed to completely retrieve the audio cover file from the stego-one without using
the position map by modifying the base of the unsigned integer (unit) function based on 8, 16, and 24 unit of
the audio cover samples. The suggested approach in [12] to audio steganography used a combination of DCT
coefficient computation and an AES encryption scheme to improve the security of the module. In [13],
proposed a technique aimed to hide secret information in a host message (receiver). It used the LSB
technique for embedding information in digital audio by replacing bits of samples in a circular manner from
LSB to MSB or vice versa to hide a sequence of bits of secret text.
In this paper, speech file is chosen as carrier media and image data is hidden inside it using three
algorithms: scrambling, encryption and hiding. This paper contains 5 sections. Section 2 discusses basic
concepts; Section 3 elaborates on the methodology; Section 4 describes the results and simulation, and
Section 5 ends with a conclusion.
2. BASIC CONCEPTS
In this paper, it used several methods and combined them in order to get good results. Below is an
explanation of the basic concepts of these methods.
2.1. Quantum chaotic map
Combining quantum key with the chaotic signals could easily develop scrambling and cryptographic
algorithms. This means that when it mixes two encoding methods (chaos cryptography and quantum
cryptography), it produces a highly secure link [14]. It calculated chaotic map with the following (1) [15, 16]:
𝐹(𝑒𝑖, 𝑑𝑖) = {
𝑑𝑖
𝑒
𝑑𝑖 = [0, 𝑒)
(1−𝑑𝑖)
(1−𝑒)
𝑑𝑖 = (𝑒, 1]
(1)
where, (ei, di), respectively, the function variables and the initial state assigned to a specific map. It obtains
values that have chaotic sequences that may be between (0-1) using the frequency F(e, d). It calculated
quantum chaotic with the following Equations (2) [17]:
Quantum Chaotic = {
𝑥𝑛+1 = 𝑑 (𝑥𝑛 − | 𝑥𝑛 |)2
− 𝑑𝑦𝑛,
𝑦𝑛+1 = −𝑦𝑛 𝑒−2𝛽
+ 𝑒−𝛽
𝑑[(2 − 𝑥𝑛 − 𝑥𝑛
∗
)𝑦𝑛 − 𝑥𝑛 𝑧𝑛
∗
− 𝑥𝑛
∗
𝑧𝑛],
𝑧𝑛+1 = −𝑧𝑛𝑒−2𝛽
+ 𝑒−𝛽
𝑑[2(1 − 𝑥𝑛
∗ )𝑧𝑛 − 2𝑥𝑛𝑦𝑛 − 𝑥𝑛].
(2)
where, x[0,1], y [0,0.1], z[0,0.2], x*=x, z*=z, [6,), and d [0,4]. ɑ represents a new quantum
fluctuation about ɑ. Effects regarding quantum corrections produced by ɑ=ɑ+ɑ, x=ɑ, y=ɑtɑ and z
=ɑɑ, x*, z* are complex conjugates of x and z respectively.  Is the particular dissipation parameter in
the specific dynamic system and the intermediate values of  in addition to yn, zn0. If the quantum
corrections yn, zn0, the logistic map is one-dimensional.

 ISSN: 2502-4752
Indonesian J Elec Eng & Comp Sci, Vol. 21, No. 3, March 2021 : 1692 - 1703
1694
2.2. Zaslavsky map
The Zaslavsky map is a discrete-time, nonlinear system. It exhibits complex behavior that is an
integral part of the algorithms for encryption. It used it in this paper to generate pseudorandom bits of keys
based on a speech encryption technique. Equations in a Zaslavsky map are as follows [18, 19]:
Zaslavsky Map = {
𝑥n+1 = mod ( 𝑥𝑛 + v(1 + µ 𝑦𝑛 ) + ԑ v µ cos (2 π𝑥𝑛), 1)
𝑦𝑛+1 = 𝑒−𝑐 (𝑦𝑛 + cos(2π𝑥𝑛))
(3)
where, v, c, ԑ are control parameters and e is exponentiation, and µ =
1−𝑒−𝑐
𝑐
. The key set for the Zaslavsky
map is {x0, y0, v, c, ԑ}.
2.3. Hénon map
The Hénon map is a complex 2D discrete-time system that exhibits chaotic behaviours. Michel
Hénon introduced this as a comprehensible approach to the Poincare map resulting from the solution of the
Lorenz complex Equations. The Hénon map is extremely sensitive to initial values, and changing parameters
and initial values can create different chaotic sequences with a large translation, suggesting that it is suitable
for creating cryptographic functions due to its ability to generate huge chaotic sequences, thus providing
excellent pseudo randomness and unpredictability [20]. The Hénon map Equations [21] are defined as
follows:
Hénon map(2𝐷) = {
𝑥𝑛+1 = 1 − 𝑎𝑥𝑛
2
+ 𝑦𝑛
𝑦𝑛+1 = 𝑏𝑥𝑛
(4)
The 3D Hénon map refers to Equations [22] as follows:
Hénon map(3𝐷) = {
𝑥𝑛+1 = 𝑎 − 𝑥𝑛
2
− 𝑏𝑧𝑛
𝑦𝑛+1 = 𝑥𝑛
𝑧𝑛+1 = 𝑦𝑛
(5)
where, a, b are state variables (two bifurcation parameters a)>0(and b (>0)). 𝑥𝑛
2
=The seed map.
2.4. Zero crossing rate (ZCR)
The zero crossing rate is a measure of the number of times the signal crosses the zero amplitude line
by transition from a positive to negative or vice versa within a certain timeframe. It calculated ZCR with the
following (6) [23, 24]:
𝑍𝐶𝑅 =
∑ |𝑠𝑔 𝑥(𝑛)−𝑠𝑔 𝑥(𝑛−1)|
𝑁
𝑛=1
2𝑁
(6)
Where,
sg x(n)=the signal of the n-th sample x(n)
N=the total number of samples in the segment.
Sg x(n) can have three potential values:+1, 0,-1 depending on the sample being positive, zero or negative.
The instantaneous accuracy is set at 20 ms because the human perceptual system is not usually accurate, also
because speech signals stay stationary for (5-20) ms.
2.5. K-Means
Data mining is a common technique for data analysis that extracts useful knowledge from massive
data. Clustering is an effective data-mining algorithm. K-Means is the most widely used method in clustering
algorithms because of its easy and simple implementation, allowing fast learning for complex datasets.
K-Means divides data into multiple clusters according to instance similarity, and has a strong dependency on
the initial k cluster centers. Centroids in the initial cluster are typically chosen randomly [25, 26]. One
calculates the distance between the sample xj (1≤ 𝑗 ≤ 𝑚) and the center of each cluster 𝜇𝑖 (1<i<k) using the
following (7) [27]:
𝑑(𝑥𝑗, 𝜇𝑖) = √∑ |𝑥𝑗𝑡 − 𝜇𝑖𝑡|
2
𝑛
𝑡=1 (7)

1695
The new mean vector is defined as follows:
𝑛𝑒𝑤 𝜇𝑖 =
1
𝑐𝑖
∑ 𝑥
𝑥∈𝑐𝑖
(8)
It used K-Means algorithm to randomize the location to be hidden. After the locations in each block
are divided into two classes, one of which has the value 1 and the other has the value 2, and the selection is
made according to the category.
2.6. Least significant bit (LSB)
LSB is the simplest method to hide data in a speech file by replacing the bit of least significance of
each bit in a message. It is good to increase the amount of data that can be covered, but it allows the amount
of noise in the speech file to increase [28].
2.7. The fractional fourier transform (FrFT)
The Fractional Fourier transform provides generalization of conventional Fourier transform. FrFT is
a generalized Fourier transform, which acts as an angle of rotation in the time-frequency plane and is also
known as a rotational Fourier transform or angular Fourier transform. FrFT is a linear operator with angle (α)
and signal (f(t)) according to the following (9)-(11) [28]:
Fα (x(t)) = ∫ 𝑓(𝑡)𝑘𝛼
+∞
−∞
(𝑢, 𝑡)𝑑𝑡 (9)
𝑘α(u, t) =
{
√
1−𝑗𝑐𝑜𝑡𝛼
2𝜋
𝑒𝑗((𝑢2+𝑡2) 2)𝑐𝑜𝑡𝛼−𝑗𝑢𝑡𝑐𝑜𝑠𝑒𝑐𝛼
⁄
𝑖𝑓 𝛼 ≠ k𝜋
𝛿(𝑢 − 𝑡) 𝑖𝑓 𝛼 = 2k𝜋
𝛿(𝑢 + 𝑡) 𝑖𝑓 𝛼 = (2k + 1)𝜋
(10)
The inverse FrFT is defined as [29]:
𝑓´(𝑡) = ∫ 𝑓𝛼(𝑢)𝑘−𝛼
+∞
−∞
(𝑢, 𝑡)𝑑𝑢 (11)
Where,
(t): represents the Dirac function, α: represents the angle of rotation (α=απ/2)
It obtains a Fourier transform if α=π/2, while it gets the same signal if α=0, so the timefrequency
representation of the signal is an intermediate value result of α:(0<α<π/2).
3. PROPOSED METHOD
The proposed algorithm (steganography-mining) consist of three phases as shown in Figure 1, while
the pre-processing is not important because we did not need to distinguish or separate the speech signal from
silence, but rather we dealt with it in the blocks form.
3.1. Scrambling phase
3.1.1. Quantum chaotic of index key algorithm
This algorithm generates new index keys for scrambling images. The steps are as follows.
a) Specify the two-dimensional size of the matrix for which you want to generate index key values, for
example 64×64, 128×128, or 256×256 values.
b) Convert the two-dimensional matrix size to a one-dimensional vector size and save it in N; for example, if
size is 64×64 values that means the vector size (N) is 4096 values.
c) Use the quantum chaotic map (2) with the specified initial values to generate a series of randomly
generated key values with length N and save the result keys in one dimension vector of length N:
x1=0.45234444336, y1=0.003453324566, z1=0.001324523564
x*=0.00186, z*=0.00398, =4.489, d=3.99
d) Arrange the one-dimensional vectors in ascending order and save each value's original index before
arranging it.
e) Save each value's result index in a one-dimensional vector for late use during scrambling.

 ISSN: 2502-4752
1696
Figure 1. Flowchart of the proposed algorithm (steganography-mining)
3.1.2. Proposed scrambling algorithm
Input: color image
Output: gray image scrambled
The steps are as follows.
a) Convert the color image to a gray image size of 64×64 or 128×128 or 256×256 pixels.
b) Cut off the image into several segmentations by dividing the image dimensional by 4 (for example 256/4)
and treat each segment independently in order to scramble the values.
c) Use the suggested algorithm (quantum chaotic of index key) to generate the index key vector with a size
equal to the size of the divided matrix of the original image.
d) Arrange the data for each index of the divided image vectors, depending on the corresponding index key
vector. In other words, arrange each value in the image vector, which contains, for example, 4096 values
(if the image size is 265×256 pixels) according to the order of the keys in the index keys vector.
e) Collect segmentation matrices to reshape the original image after scrambling its values.
f) Repeat steps 2-5 by assuming that the size of the new image segmentation is due to dividing its
dimensions by 2 (i.e. 256/2).
g) Repeat steps 2-5 again at the original image size (i.e. without cutting).
h) Note that the index key vector size that the quantum chaotic generated changes its size depending on the
size of the new segment. So it becomes ((size×size) / 2) in step 6 and (size×size) in step 7.
i) Rotate the resulting image by converting each row into a column.
j) Save the resulting image for use in the next stage to encrypt the gray image.

1697
3.2. Encrypting phase
3.2.1. The proposed hybrid zaslavsky-hénon map algorithm
This algorithm generates hybrid keys, which combine Zaslavsky and 3D Hanon maps for encrypting images.
The steps are as follows.
a) Initialize the initial values for Zaslavsky map via experimental implementation using (3) as following:
x0=0.1, y0=0.1, v=0.2, c=5, ԑ=9.
b) Initialize initial values for the parameters of a 3D Hénon map and based on the x, y values from the
Zaslavsky map, generate a sequence of random values for x, y, and z for the 3D Hénon map using (5) as
following: a=1.6, b=0.5
c) Convert the resulting fractional x values in to integer values based on the following relationship:
Xn=mod (floor (x×108), 256) (12)
d) Save the integer string of random numbers resulting in a special vector for use in the encryption phase.
3.2.1. The proposed encrypting algorithm
Input: scrambled gray image
Outputs: gray encrypted image
The steps are as follows:
a) Use the suggested hybrid Zaslavsky-Hénon algorithm to generate the hybrid keys with lengths equal to
the dimensions of the gray image (i.e., if the image is 256×256, the corresponding vector key size is
65,536 and so on).
b) Perform the XOR operation between the two values (one value from the input image matrix and another
value from the key vector), and save the resulting value to the resulting encrypted image matrix.
c) Save the resulting value matrix in an image file for use in the next stage to hide the encrypted image in a
speech file.
3.3. Hiding phase
3.3.1. Zero crossing k-means algorithm
One can use this algorithm to generate secret keys for hiding images in speech signals. It uses the
zero crossing to mine locations and reduces the database for k-means clustering, which reduces the
calculation time and makes the algorithm highly efficient. The mixing between products of the stego-key
(secret) is very large. Since the length of the key decides the depth of security because of the large key size,
then it is difficult to access the hidden data and vice versa.
Input: speech signal
Output: private secret key
The steps are as follow:
a) Read the input speech signal.
b) Divide the input signal into several blocks (i.e. each block has a size of 256 samples).
c) Calculate the coefficient of zero crossing rate for first block values according to (6)
d) Save the calculated zero crossing rate values in a vector (the length of the vector is the number of blocks
for which it divided the entered speech file).
e) Choose the 16th largest coefficient in zero crossing rate vector calculated above and save the index of
each block in the 16th largest zero crossing rate vector.
f) Convert each block sample (256 samples) in the 16th largest zero crossing rate vector to a two
dimensional matrix of 16×16 values.
g) Divide the two-dimensional matrix values into two clusters (k cluster centers=2) using the k-means
clustering algorithm, where the Euclidian distance is used.
h) Replace the first cluster with the value 1 and the second cluster with the value 2, and then combine the
two clusters to obtain a vector with a length of 16 values.
i) Repeat 6-8 on all 16th largest zero crossing rate vectors to obtain a vector with 16 values for each one,
then combine the 16 values vectors to create a new vector (A of 256 values), which is the private secret
key that is used by the sender and recipient of the speech file.
j) Send the private secret key to both the sender (for use in hide algorithms) and recipient (for use in
unhide algorithms).
3.3.2. Hide algorithm (steganography-mining)
The hide algorithm hides gray encrypted images in a speech signal.
Input: speech signal file, encrypted gray image

 ISSN: 2502-4752
1698
Output: Steg-speech file containing the hidden encrypted image
The steps are as follows:
a) Read the speech signal file. (Note: the speech signal file must be at least twice the size of the hidden
image bits, to ensure that there is enough room to hide.)
b) Divide the signal into blocks (i.e. there are 256 samples of each block size).
c) Use the zero crossing k-means algorithm to obtain a vector (A of 256 values).
d) Convert the matrix values of the gray image from a matrix to a one-dimensional vector, then convert the
values of the vector from decimal to binary, then convert the binary vector to a vector containing only
zero or one value (one bit value).
e) For each speech block which does not contain the 16th largest zero crossing rate, do the following:
 Use the FrFt transformation to obtain the magnitude and phase values.
 Convert the magnitude value from decimal to binary with length equal to 64 bits.
 Extract the bits at the locations (25-32) from the binary vector and save them in a new vector (B) of 8
bits (b1 b2 b3 b4 b5 b6 b7 b8).
 Read a value from the gray image vector (G) and a value from the vector (A) and then perform the
operations as follows:
 If A=1, do XOR (a5, a7)=R1 as the first stage. Next do XOR (R1, G)=R2 as the second stage. Then
use LSB algorithm to replace bit 8 of vector (B) with R2 value (i.e. the new values of vector (B) is
b1 b2 b3 b4 b5 b6 b7 R2).
 If A=2, do XOR (a6, a8)=R1 as the first stage. Next do XOR (R1, G)=R2 as the second stage. Then
use the LSB algorithm to replace bit 8 of vector (B) with R2 (i.e. the new values of vector (B) are b1
b2 b3 b4 b5 b6 b7 R2).
 Combine the new values of vector (B) in locations (25-32) of the magnitude value.
 Convert the value of magnitude from binary to decimal.
 Perform an inverse FrFt transform to construct the original speech signal based on the same phase value
and the new magnitude value.
f) Repeat step 6 until the end of image encrypt vector binary values.
g) Combine the resulting blocks into one speech signal file and save it in an extension (“.wav”) file.
4. RESULTS AND DISCUSSION
4.1. Setting
The experiments in this paper used both the internet and dataset of 40 standard color images taken
from the database “USC-SIPI Image Database: Miscellaneous” that includes various edgy and smooth color
images in different dimensions. It used this after converting it to gray image to reduce the time because color
images need to deal with three matrices that make up the color image. It has used speech files of the periods
3, 7, 11, 16, and 25 ms of the public dataset published by Lin et al. [30] containing more than 100 h of speech
conversation in different languages and from different genders. Blocks are stored in “.wav” format. A variety
of experiments were carried out on a PC with Pentium Intel (R) Corei7, CPU@2.60, 6.00 GB RAM, 64 bit
Windows 10 to estimate the implementation time of the proposed new encryption method; and all of the
above algorithms were coded using MATLAB R2018a software.
We assessed the effectiveness of the proposed solution to the formulated problem with the following
measures to test the quality: SNR “Signal to Noise Ratio”: used to quantify the distortion of a signal due to
noise. PSNR “Peak Signal to Noise Ratio”: This metric uses as a quality measurement between the original and
stego-medium. MSE “Mean Square Error”: It used this metric to measure the distortion in the cover after it had
hidden the data within it. NPCR “Number of Pixels Change Rate”: This is a measure to see the influence on the
encrypted image by changing a single pixel in the original image (i.e. It determined the percentage of different
pixel numbers between the two images). UACI “Unified Average Changing Intensity”: This measures the
standardized mean rate of the difference between the plain and encrypted images. The following Equations
compute SNR, PSNR, MSE, NSCR and UACI [5, 14, 15, 19, 31-33]:
𝑆𝑁𝑅 = 10 𝑙𝑜𝑔10 (
∑ 𝑜(𝑖)2
𝑛
𝑖=1
∑ (𝑜(𝑖)−𝑠(𝑖))
2
𝑛
𝑖=1
) (13)
𝑃𝑆𝑁𝑅 = 10𝑙𝑜𝑔10 [
max(𝑜(𝑖),𝑠(𝑖))2
𝑎𝑏𝑠(𝑜(𝑖)−𝑠(𝑖))2
] (14)
𝑀𝑆𝐸 = 10𝑙𝑜𝑔10 ∑ ∑
(𝑜(𝑖)−𝑠(𝑖))
2
𝑀∗𝑁
𝑛
𝑖=1
𝑚
𝑖=1 (15)

1699
𝑁𝑃𝐶𝑅 =
1
𝑁×𝑀
(∑ 𝐷(𝑖, 𝑗)
𝑖,𝑗 ) × 100% (16)
𝑈𝐴𝐶𝐼 =
1
𝑁×𝑀
(∑ |
𝑐1(𝑖,𝑗)−𝑐2(𝑖,𝑗)
255
|
𝑖,𝑗 ) × 100% (17)
𝐷(𝑖, 𝑗) = {
0 𝑐1(𝑖, 𝑗) = 𝑐2(𝑖, 𝑗)
1 𝑐1(𝑖, 𝑗) ≠ 𝑐2(𝑖, 𝑗)
(18)
Where,
n,m =the numbers of rows and columns in speech signal or image
o =the block with index number in the original speech signal, s=the block with index number in the stego
speech signal
C1 =an encrypted image without any change in the original image, C2=the encrypted image after changing
one pixel in the original image.
4.2. Experiments and results
The experimental study demonstrates the efficiency of the suggested approaches because the results
showed no difference in the listening tests between the original speech signal and the stego-speech signal
after hiding an image in it. To measure and evaluate the quality of steganography approaches, it used many
metrics to measure the overall performance of the proposed steganography techniques. The statistical metrics
that it used are SNR, PSNR and MSE. The results show that even after embedding the secret message, the
size of the speech file remains the same.
Table 1 demonstrates that the PSNR values are small, while there is a considerable difference in
MSE between the original image and encrypted gray image. The NPCR and UACI closed to 100% and
33.3% respectively. Moreover, from the results can noted the effect of the image dimensions used, where the
greater the dimension, the better the values. In Table 2, Table 3, and Table 4 we can see different speech
signal files lengths were used to demonstrate its effect to hid on the images sizes (256×256, 128×128, and
64×64) respectively. As the longer the time period, the better the results. The SNR and PSNR values are
large, but they decrease as the hidden file size increases, indicating that poor noise in higher layers does not
affect the changed bits, while the MSE is very small.
Figure 2 and Figure 3 shows all the operations performed on the image with dimensional (256×256)
and (64×64) respectively to convert the image into a gray image encoded through three scrambling
operations for the purpose of scattering the values of the original image and separating the correlation
between the values of the pixels in the image, and this process was essential because the encryption process
was based on generating the keys and making fractions between those keys and the image values after the
scrambling. While Figure 4 and Figure 5 shown the original speech signal before and after embedding the
gray image (256×256) and (64×64) respectively with length (30ms). The algorithm is completely reversible
due we used LSB to change only one bit at each location in the speech file. In principle, the hidden
information retrieved is exactly same as the original covert information.
It is not possible to compare hiding the images with the sound, because the sizes of the images and the
sound differ, as well as the research of hiding the images with the sound is little, so we made a comparison of the
image coding stage, and these results indicate that our proposed method is better as we note in the Table 5.
Table 1. Gray image encoding results in different sizes
Gray Image Dimension PSNR MSE NPCR UACI
Img1 256×256 9.8635 671.009 0.9963 0.32700
Img2 256×256 9.2372 775.088 0.9959 0.33097
Img3 256×256 8.9605 826.081 0.9956 0.32939
Img4 256×256 8.3892 942.237 0.9956 0.33114
Img5 256×256 9.1660 787.902 0.9961 0.32878
Img1 128×128 9.8252 676.950 0.9953 0.32705
Img2 128×128 9.1605 788.914 0.9949 0.32887
Img3 128×128 9.0089 816.933 0.9948 0.32913
Img4 128×128 8.3457 951.718 0.9962 0.33132
Img5 128×128 9.2021 781.385 0.9961 0.32867
Img1 64×64 9.8603 671.493 0.9956 0.32682
Img2 64×64 9.1708 787.043 0.9973 0.32884
Img3 64×64 8.9443 829.173 0.9963 0.32949
Img4 64×64 8.2884 964.340 0.9970 0.33170
Img5 64×64 9.0847 802.791 0.9970 0.32922

 ISSN: 2502-4752
1700
Table 2. Hide the different gray image size (256×256) in different speech signal lengths
Cover Length Gray Image Dimension SNR PSNR MSE
Speech1 30ms Img1 256×256 109.6296 127.9767 5.1379E-14
Speech2 40ms 256×256 112.2630 130.6101 2.8018E-14
Speech3 50ms 256×256 113.7505 132.0976 1.9893E-14
Speech4 60ms 256×256 115.1321 133.4792 1.4472E-14
Speech1 30ms Img2 256×256 109.6296 127.9767 5.1379E-14
Speech2 40ms 256×256 112.2629 130.6100 2.8019E-14
Speech3 50ms 256×256 113.7510 132.0981 1.9890E-14
Speech4 60ms 256×256 115.1325 133.4796 1.4471E-14
Speech1 30ms Img3 256×256 109.6289 127.9760 5.1387E-14
Speech2 40ms 256×256 112.2629 130.6099 2.8019E-14
Speech3 50ms 256×256 113.7500 132.0971 1.9895E-14
Speech4 60ms 256×256 115.1316 133.4787 1.4474E-14
Speech1 7ms Img1 128×128 103.6463 121.4892 1.5015E-13
Speech2 10ms 128×128 106.3241 124.6712 1.0999E-13
Speech3 20ms 128×128 107.1582 125.5053 9.0765E-14
Speech4 30ms 128×128 109.6879 128.0350 5.0694E-14
Speech1 7ms Img2 128×128 103.6462 121.4891 1.5015E-13
Speech2 10ms 128×128 106.3241 124.6712 1.0999E-13
Speech3 20ms 128×128 107.1581 125.5052 9.0767E-14
Speech4 30ms 128×128 109.6881 128.0352 5.0691E-14
Speech1 7ms Img3 128×128 103.6459 121.4889 1.5016E-13
Speech2 10ms 128×128 106.3240 124.6711 1.0999E-13
Speech3 20ms 128×128 107.1580 125.5051 9.0771E-14
Speech4 30ms 128×128 109.6879 128.0350 5.0693E-14
Speech1 2ms Img1 64×64 109.5952 124.5720 3.9420E-14
Speech2 4ms 64×64 113.9923 131.7203 2.0899E-14
Speech3 7ms 64×64 103.6591 121.5020 1.4971E-13
Speech4 10ms 64×64 106.3411 124.6882 1.0955E-13
Speech1 2ms Img2 64×64 109.5948 124.5716 3.9423E-14
Speech2 4ms 64×64 113.9898 131.7177 2.0911E-14
Speech3 7ms 64×64 103.6589 121.5019 1.4971E-13
Speech4 10ms 64×64 106.3409 124.6880 1.0956E-13
Speech1 2ms Img3 64×64 109.5939 124.5707 3.9431E-14
Speech2 4ms 64×64 113.9959 131.7239 2.0881E-14
Speech3 7ms 64×64 103.6591 121.5020 1.4971E-13
Speech4 10ms 64×64 106.3411 124.6882 1.0956E-13

1701
Figure 2. Shows all the operations (a-f) performed on the image (256×256 pixels)
Figure 3. Shows all the operations (a-f) performed on the image (64×64 pixels)
(a) Input speech signal
(b) Speech signal after hiding encrypted gray image
Figure 4. Speech signal (30 ms) before and after hiding encrypted gray image (256×256 pixels)
Table 5. The comparison of NPCR & UACI measures in encryption image algorithms
Algorithm NPCR UACI
Proposed algorithm 0.9963 0.3270
[15] 0.9963 0.3051
[34] 0.9962 0.3359
[35] 0.9962 0.3354
[36] 0.9961 0.3328
a) Gray image b) First scrambling c) Second scrambling
d) Third scrambling e) Rotated scrambling f) Encrypted scrambling
a) Gray image b) First scrambling image c) Second scrambling image
d) Third scrambling image e) Rotated scrambling image f) Encrypted scrambling image
a)Inputspeechsignal b)Speechsignalafterhidingencryptedgrayimage
a)Inputspeechsignal b)Speechsignalafterhidingencryptedgrayimage

 ISSN: 2502-4752
1702
(a) Input speech signal (b) Speech signal hiding gray encrypted image
Figure 5. Speech signal (4 ms) before and after after hiding encrypted gray image (64×64)
5. CONCLUSION
Data hiding is an important technique that embeds secret information into digital media and thereby
ensures secure transfer. In our paper, it proposes an algorithm which hides the gray image in the speech signal
file in different dimensions for both, where the proposed algorithm provides the advantages of embedding
capacity and increases robustness against hackers. It used the proposed techniques in three phases: scrambling a
gray image depending on quantum chaotic map, encrypting the gray image based on a combined Zaslavsky and
3D Hanon map, and finally, hiding the encrypted gray image using the LSB algorithm and the zero crossing k-
means algorithm. Therefore, the embedding position is completely unknown to anyone who wants to hack the
secret message and does not change the file size even after embedding. Hence, even if a person found the
hidden message and attempted to decipher it, they could only get as far as the encrypted message without any
way of being able to decrypt it and knowing the secret key. The suggested algorithm is new because of the use
of sound characteristic (zero-crossing) to determine the locations within which it wants to be hidden.
Compared to similar researches, we obtained satisfactory and good results. We used MSE, PSNR,
NPCR and UACI to measure the effectiveness of the encryption algorithm where MSE values were large,
PSNR values were small, and the NPCR and UACI closed to 100% and 33.3% respectively. while The SNR
and PSNR values were large and MSE was very small which measure the effectiveness of the hiding algorithm.
Future research will concentrate on further enhancing the hiding methods using the same algorithm on color and
medical images with additions commensurate with their difference from gray images and also dealing with
different-dimensional images.
REFERENCES
[1] F. Q. A. Alyousuf, R. Din, and A. J. Qasim, “Analysis review on spatial and transform domain technique in digital
steganography,” Bull. Electr. Eng. Informatics, vol. 9, no. 2, pp. 573-581, 2020.
[2] S. P. Rajput, K. P. Adhiya, and G. K. Patnaik, “An Efficient Audio Steganography Technique to hide Text in
Audio,” 2017 Int. Conf. Comput. Commun. Control Autom., pp. 1-6, 2017.
[3] Q. M. Hussein, “New Metrics for Steganography Algorithm Quality,” International Journal of Advanced Science
and Technology, vol. 29, no. 02, pp. 2092-2098, 2020.
[4] D. Tan, Y. Lu, X. Yan, and X. Wang, “A simple review of audio steganography,” Proc. 2019 IEEE 3rd Inf.
Technol. Networking, Electron. Autom. Control Conf. ITNEC, pp. 1409-1413, 2019. ok
[5] M. Parthasarathi, T. Shreekala, T. Nadu, and T. Nadu, “Secured Data Hiding in Audio Files Using Audio
Steganography Algorithm,” Int. J. Pure Appl. Math., vol. 114, no. 7, pp. 743-753, 2017.
[6] M. Liao, X. Dong, J. Chen, and D. Zeng, “An audio steganography based on Twi-DWT and audio-extremum
features,” Chinese Control Conf. CCC., vol. 2019-July, pp. 8882-8888, 2019. doi: 10.23919/ChiCC.2019.8866035.
[7] R. Durgarao and T. Adithya, “Hiding Image in Audio Steganography Using Discrete Cosine Transform and Skin
Detection,” IJRASET, vol. 3, no. 1, pp. 216-221, 2015.
[8] V. B. Bhagat, P. N. Kulurkar, “Audio And Video Steganography: Using Lsb And Phase Encoding Algorithm,”
IJPRET, vol. 3, no. 9, pp. 1640-1648, 2015.
[9] C. T. Jian, C. C. Wen, N. H. Binti Ab Rahman, and I. R. B. A. Hamid, “Audio Steganography with Embedded
Text,” IOP Conf. Ser. Mater. Sci. Eng., vol. 226, no. 1, pp. 1-10, 2017. doi:10.1088/1757-899X/226/1/012084.
[10] H. Chowdary N, K. Karan, K. P. Bharath and R. Kumar M, “Data Hiding in Speech Signal Using Steganography
and Encryption,” 2018 3rd IEEE International Conference on Recent Trends in Electronics, Information &
Communication Technology (RTEICT), pp. 1219-1223, 2018. doi: 10.1109/RTEICT42901.2018.9012508.
a)Inputspeechsignal b)Speechsignalhidinggrayencryptedimage
a)Inputspeechsignal b)Speechsignalhidinggrayencryptedimage

1703
[11] M. H. A. Al-Hooti, T. Ahmad, and S. Djanali, “Audio Data Hiding Using Octal Modulus Function Based Unsigned
Integer Sample Values,” 2018 Int. Conf. Comput. Eng. Netw. Intell. Multimedia, CENIM 2018, pp. 48-53, 2018.
[12] S. Bandi and H. S. Manjunatha Reddy, “Combined audio steganography and AES encryption to hide the text and
image into audio using DCT,” Int. J. Recent Technol. Eng., vol. 8, no. 3, pp. 1732-1738, 2019.
[13] S. Bahuguna, “Audio Steganography Technique Using Circular Bit Replacement,” International Journal of
Computer Engineering & Technology (IJCET), vol. 10, no. 4, pp. 17-24, 2019.
[14] F. J. Farsana and K. Gopakumar, “Speech Encryption Algorithm Based on Nonorthogonal Quantum State with
Hyperchaotic Keystreams,” Adv. Math. Phys., vol. 2020, pp. 1-12, 2020, doi: 10.1155/2020/8050934.
[15] H. M. Al-Mashhadi and I. Q. Abduljaleel, “Color image encryption using chaotic maps, triangular scrambling, with
DNA sequences,” Int. Conf. Curr. Res. Comput. Sci. Inf. Technol. ICCIT 2017, pp. 93-98, 2017.
[16] A. H. Khaleel and I. Q. Abduljaleel, “A novel technique for speech encryption based on k-means clustering and
quantum chaotic map,” Bull. Electr. Eng. Informatics, vol. 10, no. 1, pp. 160-170, 2021. DOI:
10.11591/eei.v10i1.2405.
[17] J. C. do Nascimento, R. L. C. Damasceno, G. L. de Oliveira, and R. V. Ramos, “Quantum-chaotic key distribution
in optical networks: from secrecy to implementation with logistic map,” Quantum Inf. Process., vol. 17, 2018.
https://doi.org/10.1007/s11128-018-2097-1.
[18] F. J. Farsana and K. Gopakumar, “A Novel Approach for Speech Encryption: Zaslavsky Map as Pseudo Random
Number Generator,” Procedia Comput. Sci., vol. 93, pp. 816-823, 2016.
[19] D. Riadh and R. Shaker, “Implementation of Gray Image Encryption using Multi-Level of Permutation and
Substitution,” Int. J. Appl. Inf. Syst., vol. 10, no. 1, pp. 25-30, 2015, doi: 10.5120/ijais2015451458.
[20] F. J. Farsana, V. R. Devi, and K. Gopakumar, “An audio encryption scheme based on Fast Walsh Hadamard
Transform and mixed chaotic keystreams,” Appl. Comput. Informatics, pp. 1-11, 2019.
https://doi.org/10.1016/j.aci.2019.10.001.
[21] S. Zhou, F. Xu, P. Ping, Z. Xie, and X. Lyu, “Non-square colour image scrambling based on two-dimensional sine-
logistic and hénon map,” KSII Trans. Internet Inf. Syst., vol. 11, no. 12, pp. 5963-5980, 2017. DOI:
10.3837/tiis.2017.12.015.
[22] A. Prof., “Proposed Hyperchaotic System for Image Encryption,” Int. J. Adv. Comput. Sci. Appl., vol. 7, no. 1, pp. 37-40,
2016.
[23] D. Paul, “Automated Speech Recognition of Isolated Words Using Neural Networks,” Int. J. Eng. Sci. Technol.,
vol. 3, no. 6, pp. 4993-5000, 2011.
[24] K. Zvarevashe and O. Olugbara, “Speech Emotion Recognition,” 2020.
[25] Q. Wang, X. Ouyang, and J. Zhan, “A classification algorithm based on data clustering and data reduction for
intrusion detection system over big data,” KSII Trans. Internet Inf. Syst., vol. 13, no. 7, pp. 3714-3732, 2019. DOI:
10.3837/tiis.2019.07.021.
[26] J. J. Kim, M. Ryu, and S. H. Cha, “Approximate k values using Repulsive Force without Domain Knowledge in k-
means,” KSII Trans. Internet Inf. Syst., vol. 14, no. 3, pp. 976-990, 2020. DOI: 10.3837/tiis.2020.03.004.
[27] Z. Wu, R. Li, and C. Li, “Adaptive speech information hiding method based on K-means,” IEEE Access, vol. 8, pp.
23308-23316, 2020. doi: 10.1109/ACCESS.2020.2970194.
[28] K. Saroha and P. K. Singh, “A Variant of LSB Steganography for Hiding Images in Audio,” Int. J. Comput. Appl.,
vol. 11, no. 6, pp. 12-16, 2010.
[29] B. T. Krishna, “Fractional Fourier transform: A survey,” ACM Int. Conf. Proceeding Ser., 2012, pp. 751-757, 2012.
[30] Z. Lin, Y. Huang, and J. Wang, “RNN-SM: Fast Steganalysis of VoIP Streams Using Recurrent Neural Network,”
in IEEE Transactions on Information Forensics and Security, vol. 13, no. 7, pp. 1854-1868, July 2018. doi:
10.1109/TIFS.2018.2806741.
[31] R. Ismail Abdelfattah, H. Mohamed, and M. E. Nasr, “Secure Image Encryption Scheme Based on DNA and New
Multi Chaotic Map,” J. Phys. Conf. Ser., vol. 1447, no. 1, 2020. doi:10.1088/1742-6596/1447/1/012053.
[32] R. Tanwar, K. Singh, M. Zamani, A. Verma, and P. Kumar, “An Optimized Approach for Secure Data
Transmission Using Spread Spectrum Audio Steganography, Chaos Theory, and Social Impact Theory Optimizer,”
J. Comput. Networks Commun., vol. 2019, pp. 1-10, 2019. https://doi.org/10.1155/2019/5124364.
[33] I. Q. Abduljaleel and A. H. Khaleel, “Hiding text in speech signal using K-means, LSB techniques and chaotic
maps,” Int. J. Electr. Comput. Eng., vol. 10, no. 6, pp. 5726-5735, 2020, doi: 10.11591/ijece.v10i6.pp5726-5735.
[34] P. Ramasamy, V. Ranganathan, S. Kadry, R. Damaševičius, and T. Blažauskas, “An image encryption scheme
based on block scrambling, modified zigzag transformation and key generation using Enhanced Logistic-Tent
Map,” Entropy, vol. 21, no. 7, pp. 656, 2019. doi: 10.3390/e21070656.
[35] C. Xu, J. Sun, and C. Wang, “A novel image encryption algorithm based on bit-plane matrix rotation and hyper
chaotic systems,” Multimed. Tools Appl., vol. 79, no. 9-10, pp. 5573-5593, 2020. DOI:
https://doi.org/10.1007/s11042-019-08273-x.
[36] Y. Kang, L. Huang, Y. He, X. Xiong, S. Cai, and H. Zhang, “On a symmetric image encryption algorithm based on
the peculiarity of plaintext DNA coding,” Symmetry (Basel)., vol. 12, no. 9, pp. 3-18, 2020.
https://doi.org/10.3390/sym12091393.

Secure image hiding in speech signal by steganography-mining and encryption

More Related Content

Similar to Secure image hiding in speech signal by steganography-mining and encryption (20)

More from nooriasukmaningtyas (20)

Recently uploaded (20)

Secure image hiding in speech signal by steganography-mining and encryption