Received February 10, 2021, accepted February 26, 2021, date of publication March 2, 2021, date of current version

March 11, 2021.

Digital Object Identifier 10.1109/ACCESS.2021.3063237

A New Image Encryption Algorithm for Grey and

Color Medical Images
SARA T. KAMAL 1 , KHALID M. HOSNY 2 , (Senior Member, IEEE), TAHA M. ELGINDY 3,

MOHAMED M. DARWISH 1 , AND MOSTAFA M. FOUDA 4 , (Senior Member, IEEE)

1 Department of Computer Sciences, Assiut University, Assiut 71516, Egypt
2 Department of Information Technology, Zagazig University, Zagazig 44519, Egypt
3 Department of Mathematics, Assiut University, Assiut 71516, Egypt
4 Department of Electrical and Computer Engineering, Idaho State University, Pocatello, ID 83209, USA

Corresponding author: Khalid M. Hosny ([email protected])

ABSTRACT Recently, diagnosing diseases using medical images became crucial. As these images are
transmitted through the network, they need a high level of protection. If the data in these images are liable for
unauthorized usage, this may lead to severe problems. There are different methods for securing images. One
of the most efficient techniques for securing medical images is encryption. Confusion and diffusion are the
two main steps used in encryption algorithms. This paper presents a new encryption algorithm for encrypting
both grey and color medical images. A new image splitting technique based on image blocks introduced.
Then, the image blocks scrambled using a zigzag pattern, rotation, and random permutation. Then, a chaotic
logistic map generates a key to diffuse the scrambled image. The efficiency of our proposed method in
encrypting medical images is evaluated using security analysis and time complexity. The security is tested in
entropy, histogram differential attacks, correlation coefficient, PSNR, keyspace, and sensitivity. The achieved
results show a high-performance security level reached by successful encryption of both grey and color
medical images. A comparison with various encryption methods is performed. The proposed encryption
algorithm outperformed the recent existing encryption methods in encrypting medical images.

INDEX TERMS Image encryption, chaotic logistic map, color medical images, image blocks scrambling.

I. INTRODUCTION Due to the strong correlation between the image-pixels,

With the rapid development in medical device technology, big-size images, and data redundancy, traditional encryption
it became common to diagnose various diseases using medi- algorithms are not suitable for digital images, especially
cal images. Medical images are transmitted through different medical images. Many medical image encryption algorithms
networks; therefore, securing these images became an essen- [9]–[13] were proposed to reduce correlation and redundancy.
tial topic in recent years. Safe transmission of medical images In [14], Singh et al. presented a medical image encryption
requires confidentiality, integrity, and authentication. Unau- algorithm based on an improved ElGamal encryption scheme
thorized usage of such images may lead to loss of privacy of version. The problem of data expansion is resolved, and the
patients’ data. Moreover, when these images are liable for any execution speed is improved. Hua et al. [15] proposed a new
little change, it may result in an incorrect diagnosis that could medical image encryption algorithm consisting of random
threaten patients’ lives. data insertion, high-speed scrambling, and pixel adaptive
Generally, securing digital images could be achieved by diffusion. In [16], Chen et al. proposed a generalized optical
using image steganography [1], [2], image watermarking encryption framework based on Shearlets and double ran-
[3]–[5], and image encryption [6]–[8]. Encryption is the most dom phase encoding (DRPE) for encrypting medical images.
straightforward and most efficient method to ensure med- Cao et al. [17] presented a medical image encryption algo-
ical image security via converting the plain image into an rithm using edge maps. The algorithm based on three
unreadable one using a secret key. Without having that secret main parts: bit-plane decomposition, generating a random
key, nobody can restore the plain image. Image encryption sequence, and permutation.
depends on two major operations: confusion and diffusion. Different algorithms for securing medical images are
introduced, yet they may be liable to attacks. A strong
The associate editor coordinating the review of this manuscript and correlation between neighboring pixels characterizes med-
approving it for publication was Rajeswari Sundararajan. ical images; thus, removing this correlation requires a

This work is licensed under a Creative Commons Attribution 4.0 License. For more information, see https://creativecommons.org/licenses/by/4.0/
VOLUME 9, 2021 37855
 S. T. Kamal et al.: New Image Encryption Algorithm for Grey and Color Medical Images

FIGURE 1. Medical image encryption block diagram.

permutation (scrambling) technique with a higher security The remainder of this paper is organized as follows.
level. This paper presents a new algorithm for encrypting Section 2 discusses the proposed method in detail.
medical images that include four parts: image splitting, image Section 3 demonstrates the simulation results and analyses.
scrambling, key generation, and diffusion. First, the plain Finally, the paper is concluded in section 4.
image is divided into blocks and sub-blocks using a new
image splitting technique. Second, the pixels’ arrangement II. THE PROPOSED METHOD
is changed in the blocks and sub-blocks using a zigzag pat- This section describes the proposed algorithm’s main steps
tern, rotation at a 90-degree angle, and random permutation for securing medical images in detail. In the first step,
between blocks. Third, a key is generated from the logistic the plain image is encrypted and converted into an unread-
map, where the map’s initial condition depends on the plain able image. Then, to recover the plain image, we apply the
image. Finally, image pixel values are changed using the decryption step.
secret key. A. ENCRYPTION
The contributions of this paper are summarized by: Here, our algorithm for encrypting medical images consists
1. A new technique for image splitting is proposed. of four stages. In the first stage, we perform image splitting.
2. Random permutation between blocks is applied, and Confusion (scrambling) is performed in the second stage. The
pixels substitution in each block is performed to remove third stage presents key generation based on a logistic map.
the correlation between pixels. The final stage presents the diffusion process. An illustrative
3. A logistic map is used to diffuse the scrambled image, diagram of medical image encryption is shown in Figure 1.
where the map’s initial condition is based on the plain
1) PLAIN IMAGE SPLITTING
image. Therefore, the proposed algorithm is robust
against differential attacks. The plain image is divided into non-overlapping blocks of
4. Analysis of the results proves that our algorithm gains the same size. Our algorithm is appropriate for different
a high performance in encrypting medical images than block sizes (i.e., 16, 32, and 64), and the user can select
other methods. the block size. Then, each block is either sub-divided into
sub-blocks with equal sizes or remains without splitting. The
This paper aims to: sub-blocks in each block are chosen depending on a random
1. Achieve a high level of security. number generated for each block.
2. Propose a block and sub-block image to accelerate the
entire encryption process essential in securing medi- 2) CONFUSION
cal images through their transmission via IoT (Inter- Confusion is the process of changing pixels’ arrangement
net of Things) devices for healthcare and telemedicine in the image. In our algorithm, confusion is performed for
systems. blocks and sub-blocks as follows:

37856 VOLUME 9, 2021

S. T. Kamal et al.: New Image Encryption Algorithm for Grey and Color Medical Images

FIGURE 2. Demonstration of the zigzag pattern.

The numbers, M and N, refer to the number of rows and

columns in the plain image, respectively.
2. Iterate the chaotic map (eq.1) N0 + MN times, and then
skip the first N0 elements to get a new sequence S with
size MN.
3. Calculate the key using the following formula:
   
K (i) = mod floor S (i) × 1014 , 256 ,
i = 1 : MN (3)

4) DIFFUSION
In the diffusion process, image pixel values are changed, and
then a noise image is generated. Bit-wise exclusive OR oper-
ation between the key K and the scrambled image vector is
performed to obtain the encrypted image. Detailed encryption
steps are presented in Algorithm 1. Also, the flowchart of the
proposed algorithm is shown in Figure 4.
FIGURE 3. Bifurcation diagram of the logistic map.
B. DECRYPTION
With the original key and by inverting the encryption stages,
1. The zigzag pattern is applied to both undivided blocks
we can retrieve the plain image. The decryption process is
and sub-blocks, as described in Figure 2.
described as follows:
2. Both undivided blocks and sub-blocks rotated by 90◦ . 1. Bit-wise exclusive OR operation between the key K
3. Random vector r generated where its size is equal to the and the encrypted image vector is applied to get the
number of blocks in the plain image. scrambled image.
4. Random permutation between blocks based on the 2. Return each block to its original position using
vector r is applied to get the scrambled image. vector r.
3) KEY GENERATION 3. The inverse operation of rotation and the zigzag pattern,
respectively, are applied to both undivided blocks and
The key used in the diffusion process is generated from a
sub-blocks.
logistic map. The logistic map is defined by:
Yn+1 = aYn (1 − Yn ) (1) III. SIMULATION RESULTS
In this section, the efficiency of our algorithm in encrypt-
where a is the control parameter with range 0 < a ≤ 4, Y0 ing medical images is presented. All medical images used
is the initial value, and Yn is the output sequence with in this section are displayed in Figure 5, in which the
0 < Yn < 1. The map is chaotic when a ∈ [3.57, 4]. grey images are from [18], and Levoy [19], and the color
Figure 3 shows the bifurcation diagram of the logistic map. images are from [20]–[22]. Img1, Img2, Img6, and Img7 are
The key generation steps are defined as follows: 512 × 512, while for Img3, Img4, Img5, Img8, and Img9.
1. Calculate the initial value of the logistic map that The authors executed the proposed algorithm with MATLAB
depends on the plain image P by the following equation: (R2015a) on a laptop computer equipped with Core
N
i5-2430M 2.4GH CPU, 4GB memory, and Windows 7 OS.
PM P
i=1 P(i, j) The parameters used in our algorithm are: the size of blocks
j=1 in image splitting is 16 (where n = 4), and for the logistic
Y0 = (2)
M × N × 255 map: a = 3.9, and the iteration number N0 = 1000.

VOLUME 9, 2021 37857

 S. T. Kamal et al.: New Image Encryption Algorithm for Grey and Color Medical Images

Algorithm 1 The Proposed Algorithm for Medical Image our proposed algorithm assures generating encrypted images
Encryption with high randomness.
Input: the plain image P with size M × N , parameter a of
the logistic map and N0 . B. ANALYSIS OF IMAGE HISTOGRAM
1. Divide P into an equal number of blocks, with a block The histogram presents the distribution of pixels in the image.
size h = 2n where n takes a value from 4, 5, 6 For an encrypted image, the histogram should be flat to
2. Generate a random number for each block depending prevent the attackers from guessing any image information.
on the block size 2l where l takes a random value Also, the histogram of both the encrypted image and the plain
from 2, 3, . . . n, put the generated random numbers image should not be similar. Figure 6 shows the histograms
in a vector Ri , i = 1: (MN/h2 ). of three medical images and their encrypted ones. As can be
3. For i = 1: MN/h2 do observed, the histograms of the encrypted images using our
algorithm are uniform and not similar to their corresponding
4. Divide a block into sub-blocks with the same
plain image histograms.
size or keep without dividing based on Ri .
An additional experiment was performed to confirm that
5. End for the histogram of the encrypted image is uniform. This exper-
6. Perform a Zigzag pattern and rotation with angle 90, iment is based on the chi-square test (χ 2 ) which is calculated
respectively, to both undivided blocks and sub-blocks. by [27]:
7. Generate a random vector r with size MN/h2 .
8. Random permutation of image blocks based on r is 256
X (Oi − EV )2
performed to get the scrambled image X. χ2 = (5)
EV
9. Generate the initial condition of the logistic map using i=1
(eq.2)
where Oi refers to the recurrence rate of grey value i, and
10. Iterate the chaotic map (eq.1) N0 + MN times, and
EV = O/256 is the expected frequency of each grey value.
then discard first N0 elements to get a new sequence
The value of χ(α,d)
2 is 293.2478, where the significance
S with size MN.
level α is 0.05 and the degree of freedom d is 255. The values
11. For i = 1: MN do
χ 2 of the encrypted image are presented in Table 3. All the
12. K (i) = mod (floor (S(i) × 1014 ), 256) values are less than 293, which means that the histogram
13. End for of the images encrypted with our proposed algorithm is
14. Convert the matrix X into the 1D image pixel vector uniform.
X0
16. E = X 0 ⊕ K C. ANALYSIS OF CORRELATION COEFFICIENT
17. Convert E into a 2D matrix C. Principally in the plain image, adjacent pixels show high
Output: the encrypted image C correlation as their values are nearly identical. The encryption
algorithm’s efficiency is based on generating an encrypted
image with a low correlation between adjacent pixels. Math-
ematically, the correlation coefficient between two adjacent
A. ANALYSIS OF INFORMATION ENTROPY pixels defined by the following equations:
The randomness of the image is measured by information
entropy. The mathematical definition of entropy is given by: E( (A − E (A)) (B − E (B)))
rA,B = √ (6)
D(A)D(B)
w s
X 1 1
H (m) =
X
P(mi )log2 (4) E (A) = Ai (7)
P (mi ) s
i=1 i=1
1 Xs
where P(m) is the probability of appearance of m, for D (A) = (Ai − E(A))2 (8)
s i=1
greyscale images, the maximum value of entropy is 8. When
the value of entropy is near 8, the randomness of pixels in where A and B are the grey values of two adjacent pixels, and
the image is higher. In this experiment, we encrypt the grey s is the total number of selected pairs (A, B). Table 4 presents
test medical images using the proposed algorithm and calcu- the correlation coefficient values for both grey test images
late the entropy values of the encrypted images as listed in and their encrypted ones in the horizontal (H), vertical (V),
Table 1. From the results, we can observe that all the entropy and diagonal (D) directions. All the test images have corre-
values are near 8, which indicates the true randomness of lation coefficient values close to one; however, the encrypted
the encrypted images. The first test image (i.e., Img1) is images’ correlation coefficient values are near to zero. The
encrypted using our algorithm and other encryption algo- comparison with other methods based on Img1 is listed in
rithms, as listed in Table 2. As can be seen, our proposed algo- Table 5. This experiment demonstrates that the proposed
rithm shows a higher entropy value compared to the different algorithm effectively reduces the adjacent pixels’ correlation
algorithms in Table 2. From this experiment, we conclude that in the encrypted image.

37858 VOLUME 9, 2021

S. T. Kamal et al.: New Image Encryption Algorithm for Grey and Color Medical Images

FIGURE 4. Flowchart of the proposed algorithm.

VOLUME 9, 2021 37859

 S. T. Kamal et al.: New Image Encryption Algorithm for Grey and Color Medical Images

FIGURE 5. The test images. (a) The image Img1 [18]. (b) The image Img2 [18]. (c) The image Img3 [19]. (d) The image Img4 [19]. (e) The image Img5 [19].
(f) The image Img6 [20]. (g) The image Img7 [21]. (h) The image Img8 [22]. (i) The image Img9 [22].

TABLE 1. Encrypted images entropy. The symbols E1 and E2 refer to two encrypted images from
the plain image and the modified image (made by changing
one pixel in the plain image). The image width is M. Its height
is N. Here, we study our proposed algorithm’s effectiveness
in resisting differential attacks by recording the NPCR and
UACI values between the two encrypted images in table 6.
The ideal value of NPCR is 99.6094%, and of UACI is
33.4635%. All values in Table 6 are close to their ideal values.
TABLE 2. Entropy value of our algorithm and other algorithms.
Table 7 shows a comparison between our algorithm and
other image encryption algorithms. The results show that our
proposed algorithm is highly capable of resisting differential
attacks.

E. ANALYSIS OF KEYSPACE
The keyspace of a good image encryption algorithm should be
at least 2100 . If the keyspace is not large enough, the algorithm
D. ANALYSIS OF DIFFERENTIAL ATTACK could be broken using brute-force attacks. In this algorithm,
The differential attack depends on guessing information the keyspace includes the initial condition Y0 , the control
about an image by making a slight change in the plain parameter a, and the initial iteration number N0 of the chaotic
image and encrypting both images using the same algorithm. map. Here, we consider the precision of Y0 and a to be 1016 ,
We compare both images to detect a correlation between and N0 = 103 So the total keyspace is 1035 . Therefore, our
the plain image and the encrypted image. Using a practical algorithm can resist brute-force attacks as the keyspace is
algorithm, any slight change in the plain image should pro- large enough.
duce a different encrypted image. To assess the algorithm’s
performance, the Number of Pixels Change Rate (NPCR), F. ANALYSIS OF KEY SENSITIVITY
and Unified Average Changing Intensity (UACI) used. The A practical algorithm should be susceptible to any slight
NPCR and UACI are calculated as follows: change to its secret key. Attackers can break the encryp-
M N tion algorithm using a similar key, so any small change in
1 XX the key used in the decryption step cannot reconstruct the
NPCR = D (i, j) × 100 (%) (9)
MN plain image. Here, we generate two secret keys with only
i=1 j=1
( a slight change in Y0 by modifying it to Y0 + 10∧ (−10).
0 if E1 (i, j) = E2 (i, j) ,
D (i, j) = (10) The first key used in the plain image’s encryption step is
1 if E1 (i, j) 6 = E2 (i, j) , shown in Figure 7(a), and the results are shown in Figure 7(b).
M N The second key used in the decryption step (the obtained
1 X X |E1 (i, j) − E2 (i, j)|
UACI = × 100 (%) (11) results are shown in Figure (7c)). As can be seen, the second
MN 255 key failed to retrieve the plain image. When using the first
i=1 j=1

37860 VOLUME 9, 2021

S. T. Kamal et al.: New Image Encryption Algorithm for Grey and Color Medical Images

FIGURE 6. Histogram analysis of gray images. (a) The plain image Img 2. (b) The plain image Img 3. (c) The plain image Img 5. (d) Histogram of (a).
(e) Histogram of (b). (f) Histogram of (c). (g) Histogram of the encrypted image in (a). (h) Histogram of the encrypted image in (b). (i) Histogram of the
encrypted image in (c).

TABLE 3. Chi-Square analysis. and is calculated by:

!
2552
PSNR = 10 × log10 (db) (12)
MSE
M N
1 XX
MSE = |OI (i, j) − EI(i, j)|2 (13)
MN
i=1 j=1

key in the decryption step, the plain image is reconstructed The OI refers to the original image, and EI is the encrypted
successfully, as shown in Figure 7(d). image. The lower values of PSNR indicate a significant
difference between the original and the encrypted image.
G. ANALYSIS OF ENCRYPTION EFFICIENCY Table 8 lists the PSNR values for different grey medical
Peak signal to noise ratio (PSNR) is used to measure the dif- images. From the results, we can conclude that our proposed
ference between the original image and the encrypted image algorithm is highly efficient in medical image encryption.

VOLUME 9, 2021 37861

 S. T. Kamal et al.: New Image Encryption Algorithm for Grey and Color Medical Images

FIGURE 7. The sensitivity of our algorithm to the key. (a) Plain image. (b) The encrypted image of (a) using the first key. (c) A decrypted
image of (b) using the second key. (d) The decrypted image of (b) using the first key.

TABLE 4. Correlation coefficient values. TABLE 8. PSNR analysis.

TABLE 9. The average of entropy, correlation coefficient, NPCR, UACI,

PSNR and Chi-Square values for grey medical images.

TABLE 5. Comparison of the correlation coefficient values between our TABLE 10. Maximum deviation analysis.
algorithm and other algorithms.

size 256 × 256. All the results obtained with our proposed
TABLE 6. NPCR and UACI performances. algorithm are ideal, which evidences the robustness of our
algorithm.

H. ANALYSIS OF ENCRYPTION QUALITY

1) MAXIMUM DEVIATION
The quality of encryption is evaluated by measuring the
difference in pixel values between the plain and encrypted
TABLE 7. Comparison of NPCR and UACI. images. The encryption algorithm is considered to be efficient
if this difference is significant. The maximum deviation is
calculated by
254
M0 + M255 X
D= + Mi (14)
2
i=1
where Mi the difference of histogram between the plain and
the encrypted image at index i. The high value of D indicates
Our proposed algorithm was also tested on images from the significant difference between the plain and the encrypted
Levoy [19]. The average of entropy, NPCR, UACI, correla- image. The maximum deviation values using our proposed
tion coefficient in the three directions, PSNR, and chi-square algorithm listed in table 10. Large values indicate that the
listed in Table 9. All image formats are 8-bit TIF and images encrypted using proposed algorithms are entirely

37862 VOLUME 9, 2021

S. T. Kamal et al.: New Image Encryption Algorithm for Grey and Color Medical Images

FIGURE 8. Histogram analysis of Img6. (a) Original plain image. (b) Plain image histograms of red channel. (c) plain image histograms of green
channel. (d) Plain image histograms of blue channel. (e) Encrypted image histograms of red channel. (f) Encrypted image histograms of red channel.
(g) Encrypted image histograms of blue channel.

different from the plain image that proves our algorithm’s The HC refers to the histogram of the encrypted image.
excellent performance in terms of security. A lower value DH indicates the histogram’s uniformity, which
ensures high encryption quality—results in table 11 show
2) DEVIATION FROM UNIFORM HISTOGRAM low values of DH that poof high encryption quality of the
A good encryption algorithm should produce an encrypted proposed algorithm.
image with a uniform histogram. The quality of the encryp-
tion algorithm assessed by histogram deviation, which is I. COLOR MEDICAL IMAGE TESTING RESULTS
defined by: With the advancement of medical devices’ modern technol-
ogy, color medical images became widely used in diagnosing
 diseases. Our proposed algorithm can also be applied to
 M ×N
, 0 ≤ C i ≤ 255 encrypt color medical images. Generally, color images con-
HCi = 256 (15)
 0, elsewhere tain more information than grey ones as each pixel in a color
P255   image has three values (Red, Green, and Blue). So, encrypting
Ci =0 HCi − HC color images could be done by separating the image into three

DH = (16)
M ×N channels (R, G, and B) then using the algorithm to encrypt

VOLUME 9, 2021 37863

 S. T. Kamal et al.: New Image Encryption Algorithm for Grey and Color Medical Images

TABLE 11. Deviation from a uniform histogram. entropy, histogram, correlation coefficient, differential attack,
keyspace, and key sensitivity. Results showed that the pro-
posed algorithm is efficient in encrypting both grey and color
medical images. Our algorithm compared to other recent
encryption algorithms, and the results confirm that the pro-
posed algorithm has good characteristics in encrypting both
grey and color medical images.
TABLE 12. The entropy, Correlation coefficient, and NPCR and UACI
values for color medical images in three channels. REFERENCES
[1] X. Liao, J. Yin, S. Guo, X. Li, and A. K. Sangaiah, ‘‘Medical JPEG image
steganography based on preserving inter-block dependencies,’’ Comput.
Electr. Eng., vol. 67, pp. 320–329, Apr. 2018.
[2] M. A. Usman and M. R. Usman, ‘‘Using image steganography for provid-
ing enhanced medical data security,’’ in Proc. 15th IEEE Annu. Consum.
Commun. Netw. Conf. (CCNC), Jan. 2018.
[3] K. M. Hosny, M. M. Darwish, K. Li, and A. Salah, ‘‘Parallel multi-core
CPU and GPU for fast and robust medical image watermarking,’’ IEEE
Access, vol. 6, pp. 77212–77225, Dec. 2018.
[4] K. M. Hosny and M. M. Darwish, ‘‘Robust color image watermarking
using invariant quaternion legendre-Fourier moments,’’ Multimedia Tools
Appl., vol. 77, no. 19, pp. 24727–24750, Oct. 2018.
[5] K. M. Hosny and M. M. Darwish, ‘‘Resilient color image watermarking
using accurate quaternion radial substituted chebyshev moments,’’ ACM
Trans. Multimedia Comput., Commun., Appl., vol. 15, no. 2, pp. 1–25,
Jun. 2019.
[6] A. M. Vengadapurvaja, G. Nisha, R. Aarthy, and N. Sasikaladevi, ‘‘An
efficient homomorphic medical image encryption algorithm for cloud
storage security,’’ Procedia Comput. Sci., vol. 115, pp. 643–650, 2017.
[7] J. Liu, Y. Ma, S. Li, J. Lian, and X. Zhang, ‘‘A new simple chaotic system
each channel independently. In this experiment, we used four and its application in medical image encryption,’’ Multimedia Tools Appl.,
color medical images, shown in Figure 5 (i.e., Img6, Img7, vol. 77, no. 17, pp. 22787–22808, Sep. 2018.
[8] J. Liu, S. Tang, J. Lian, Y. Ma, and X. Zhang, ‘‘A novel fourth order chaotic
Img8, and Img9), as test images. Table 12 shows the entropy system and its algorithm for medical image encryption,’’ Multidimensional
results, correlation coefficient values in the three directions, Syst. Signal Process., vol. 30, no. 4, pp. 1637–1657, Oct. 2019.
NPCR, and UACI of the three channels for the encrypted [9] K. Shankar, M. Elhoseny, E. D. Chelvi, S. K. Lakshmanaprabu, and
color medical images. All the obtained results are close to W. Wu, ‘‘An efficient optimal key based chaos function for medical image
security,’’ IEEE Access, vol. 6, pp. 77145–77154, 2018.
their ideal values. Also, the histogram analysis of the color [10] J. Chen, L. Chen, L. Y. Zhang, and Z.-L. Zhu, ‘‘Medical image cipher using
image (Img6) is shown in Figure 8. Encrypted image his- hierarchical diffusion and non-sequential encryption,’’ Nonlinear Dyn.,
tograms of the three channels are all flat and different from vol. 96, no. 1, pp. 301–322, Apr. 2019.
[11] G. Ke, H. Wang, S. Zhou, and H. Zhang, ‘‘Encryption of medical image
the plain image histograms. with most significant bit and high capacity in piecewise linear chaos
Our proposed algorithm was also tested on the dataset [21] graphics,’’ Measurement, vol. 135, pp. 385–391, Mar. 2019.
that consists of 70 melanoma and 100 naevus images. The [12] A. Belazi, M. Talha, S. Kharbech, and W. Xiang, ‘‘Novel medical image
image format is JPG, and we resize all images to 512 × 512, encryption scheme based on chaos and DNA encoding,’’ IEEE Access,
vol. 7, pp. 36667–36681, 2019.
and the averages of all images in each channel listed in [13] Z. Mishra and B. Acharya, ‘‘High throughput and low area architectures of
table 12. Therefore, our proposed algorithm is efficient in secure IoT algorithm for medical image encryption,’’ J. Inf. Secur. Appl.,
encrypting color medical images. vol. 53, Aug. 2020, Art. no. 102533.
[14] D. S. Laiphrakpam and M. S. Khumanthem, ‘‘Medical image encryp-
J. ANALYSIS OF TIME COMPLEXITY tion based on improved ElGamal encryption technique,’’ Optik, vol. 147,
pp. 88–102, Oct. 2017.
Here we estimate the time complexity in each step of the [15] Z. Hua, S. Yi, and Y. Zhou, ‘‘Medical image encryption using high-
encryption process to evaluate our proposed algorithm’s total speed scrambling and pixel adaptive diffusion,’’ Signal Process., vol. 144,
time complexity. Assume that the plain image is with size pp. 134–144, Mar. 2018.
[16] M. Chen, G. Ma, C. Tang, and Z. Lei, ‘‘Generalized optical encryption
M × N, and the block size h = 2n where n = 4. The framework based on shearlets for medical image,’’ Opt. Lasers Eng.,
time complexity for the plain image splitting and confusion vol. 128, May 2020, Art. no. 106026.
stages is O((M × N)/h2 ). For the key generation stage and [17] W. Cao, Y. Zhou, C. L. P. Chen, and L. Xia, ‘‘Medical image encryption
using edge maps,’’ Signal Process., vol. 132, pp. 96–109, Mar. 2017.
the diffusion stage, the time complexity is O(M × N). There-
[18] Category: Computed Tomography Images of Mikael Häg-
fore, the total time complexity of our proposed algorithm gström’s Brain. Accessed: Feb. 9, 2021. [Online]. Available:
is O(M × N). https://commons.wikimedia.org/wiki/Category:Computed_tomography_
images_of_Mikael_H%C3%A4ggstr%C3%B6m%27s_brain
IV. CONCLUSION [19] The Stanford Volume Data Archive. Accessed: Feb. 9, 2021. [Online].
Available: https://graphics.stanford.edu/data/voldata/
This paper introduced a new algorithm for encrypting med-
[20] RITE (Retinal Images Vessel Tree Extraction) Database. Accessed:
ical images based on image blocks and chaos. The pro- Feb. 9, 2021. [Online]. Available: https://uiowa.qualtrics.com/jfe/form/
posed algorithm’s image encryption performance tested using SV_a3mc5H4SG2B3e2p?Q_JFE=qdg

37864 VOLUME 9, 2021

S. T. Kamal et al.: New Image Encryption Algorithm for Grey and Color Medical Images

[21] Dermatology Database Used in MED-NODE. Accessed: TAHA M. ELGINDY received the B.Sc. and M.Sc.
Feb. 9, 2021. [Online]. Available: http://www.cs.rug.nl/~imaging/ degrees from Assiut University, in 1967 and 1971,
databases/melanoma_naevi/ respectively, and the Ph.D. degree from Swansea
[22] Breast Cancer Histopathological Database (BreakHis). Accessed: University (G.B), in 1977. He is currently a Pro-
Jan. 8, 2021. [Online]. Available: https://web.inf.ufpr.br/vri/databases/ fessor of operation research (OR). He is the super-
breast-cancer-histopathological-database-breakhis/ visor of more than 60 students in many universities
[23] X. Chai, Z. Gan, K. Yuan, Y. Chen, and X. Liu, ‘‘A novel image encryption
in Egypt in (OR) and numerical analysis.
scheme based on DNA sequence operations and chaotic systems,’’ Neural
Comput. Appl., vol. 31, no. 1, pp. 219–237, Jan. 2019.
[24] J. Chandrasekaran and S. J. Thiruvengadam, ‘‘A hybrid chaotic and number
theoretic approach for securing DICOM images,’’ Secur. Commun. Netw.,
vol. 2017, Jan. 2017, Art. no. 6729896.
[25] X. Wang and C. Liu, ‘‘A novel and effective image encryption algorithm
based on chaos and DNA encoding,’’ Multimedia Tools Appl., vol. 76, no. 5,
pp. 6229–6245, Mar. 2017.
[26] S. Kumar, B. Panna, and R. Kumar, ‘‘Medical image encryption using
fractional discrete cosine transform with chaotic function,’’ Med. Biol. Eng.
Comput., vol. 57, no. 11, pp. 2517–2533, 2019. MOHAMED M. DARWISH received the B.Sc.
[27] A. Banu S and R. Amirtharajan, ‘‘A robust medical image encryption (Hons.) and M.Sc. degrees in computer science
in dual domain: Chaos-DNA-IWT combined approach,’’ Med. Biol. Eng. from the Faculty of Science, Assiut University,
Comput., vol. 58, no. 7, pp. 1445–1458, Jul. 2020. Assiut, Egypt. He is currently a Lecturer with
the Department of Computer Sciences, Faculty of
Computers and Informatics, Assiut University. His
research interests include image processing and
SARA T. KAMAL received the B.Sc. and M.Sc.
data mining.
degrees in computer science from the Faculty of
Science, Assiut University, Assiut, Egypt. She is
currently an Assistant Lecturer with the Faculty of
Computers and Informatics, Assiut University.

MOSTAFA M. FOUDA (Senior Member, IEEE)

received the Ph.D. degree in information sciences
from Tohoku University, Japan, in 2011. He has
KHALID M. HOSNY (Senior Member, IEEE) worked at Tohoku University, Japan, as an Assis-
was born in Zagazig, Egypt, in 1966. He received tant Professor. He has also worked at Tennessee
the B.Sc., M.Sc., and Ph.D. degrees from Zagazig Technological University, TN, USA, as a Post-
University, Egypt, in 1988, 1994, and 2000, doctoral Research Associate. He is currently an
respectively. From 1997 to 1999, he was a Visit- Assistant Professor with Idaho State University,
ing Scholar with the University of Michigan, Ann Pocatello, ID, USA. He also holds the position
Arbor, and the University of Cincinnati, Cincin- of an Associate Professor with Benha University,
nati, USA. He is currently a Professor of infor- Egypt. He has published over 30 papers in IEEE conference proceedings
mation technology with the Faculty of Computers and journals. His research interests include cybersecurity, machine learning,
and Informatics, Zagazig University. He published blockchain, the IoT, 5G networks, and smart grid communications. He has
three edited books and more than 80 articles in international journals. His served on the technical committees of several IEEE conferences. He is also a
research interests include image processing, pattern recognition, multimedia, Reviewer in several IEEE Transactions/Magazines and an Associate Editor
and computer vision. He is a Senior Member of ACM. He is an Editor and a of IEEE ACCESS.
Scientific Reviewer of more than 40 international journals.

VOLUME 9, 2021 37865