Received March 10, 2021, accepted March 31, 2021, date of publication April 9, 2021, date of current version

June 1, 2021.
Digital Object Identifier 10.1109/ACCESS.2021.3072075

Image Encryption and Authentication

With Elliptic Curve Cryptography and
Multidimensional Chaotic Maps
PRIYANSI PARIDA 1 , CHITTARANJAN PRADHAN 1 , XIAO-ZHI GAO 2,

DIPTENDU SINHA ROY 3 , (Senior Member, IEEE), AND

RABINDRA KUMAR BARIK 4 , (Member, IEEE)
1 School of Computer Engineering, KIIT University, Bhubaneswar 751024, India
2 School of Computing, University of Eastern Finland, 70211 Kuopio, Finland
3 Department of Computer Science and Engineering, NIT Meghalaya, Shillong 793003, India
4 School of Computer Applications, KIIT University, Bhubaneswar, Odisha 751024, India

Corresponding author: Priyansi Parida ([email protected])

ABSTRACT Many researchers have used the properties of the popular Elliptic Curve Cryptography(ECC)
to devise a stronger and faster image encryption algorithm to assure the secrecy of images during online
transmission. In this paper, a robust Elliptic curve based image encryption and authentication model for
both grayscale and color images has been proposed. The model uses the secure Elliptic Curve Diffie-
Hellman(ECDH) key exchange to compute a shared session key along with the improved ElGamal encoding
scheme. 3D and 4D Arnold Cat maps are used to effectively scramble and transform the values of plain
image pixels. A well-structured digital signature is used to verify the authenticity of the encrypted image
prior to decryption. The model produces good-quality cipher images with an average entropy of 7.9993 for
grayscale and 7.99925 for the individual components of color images. The model has high average NPCR
of 99.6%, average UACI of 33.3% and low correlation for both grayscale and color images. The model has
low computational costs with minimized point multiplication operations. The proposed model is robust with
high resilience against statistical, differential, chosen-plaintext(CPA), known-plaintext(KPA) and occlusion
attacks.

INDEX TERMS Arnold cat map, chaotic map, digital signature, elliptic curve cryptography, image
encryption.

I. INTRODUCTION secret key which needs to be shared before communication.

Extensive amount of data which consists of a variety of Asymmetric encryption eliminates this issue by using two
images is transmitted back and forth between users online different keys for encryption and decryption.
everyday. Ensuring the availability of correct data to the Elliptic Curve Cryptography(ECC) is a popular asymmet-
intended receiver with guaranteed secrecy of data from other ric encryption algorithm which provides higher level security
users in the network is a challenge. Many image encryp- with smaller key sizes with reduced resource consumption
tion algorithms have been presented for fast and powerful making it an ideal choice for resource-constraint devices. The
real-time encryption of images. The encryption algorithms unpredictability introduced by chaotic maps in encryption
include both symmetric and asymmetric encryption algo- algorithms, in addition makes the model easy to implement,
rithms. Symmetric encryption is fast with low computational swift and strong against the attacker. Logistic map, Henon
complexity making it suitable for large data sets. The key map, Baker’s map, Arnold Cat map, Sine map and Lorenz
distribution and management poses a huge overhead on the system are some of the well-known chaotic systems used
symmetric encryption models where each user has his own in encryption algorithms. Chaos based image encryption
schemes show higher efficacy than the simple symmetric
The associate editor coordinating the review of this manuscript and or asymmetric image encryption schemes without it. The
approving it for publication was Ahmed Farouk . encryption schemes that use a combination of chaotic maps

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 76191
 P. Parida et al.: Image Encryption and Authentication With ECC and Multidimensional Chaotic Maps

perform better than the ones employing single chaotic map. of the proposed model in comparison to the other schemes
The use of higher dimensional chaotic maps is known to prevailing in the field of image encryption. In Section VII,
enhance the security and quality of encryption. the conclusions follow.
Chen et al. [1] proposed a symmetric image encryption
algorithm with 3D cat map and logistic map for shuffling II. PRELIMINARIES
of image data with an added layer of confusion between A. ELLIPTIC CURVE CRYPTOGRAPHY
plain-cipher image after every two rounds of shuffling. The The elliptic curve, E in Fig.1 is a non-singular algebraic
encryption also uses Chen’s chaotic system for key gener- plane curve defined over a finite field Fp that consists of a
ation. Pareek et al. [2] used the outcome of two logistic set of points satisfying the well known Weierstrass equation
maps to randomly opt for one of the eight modes designed to of form [27],
encrypt every pixel in the image. Liu and Wang [3] designed a
y2 mod p = x 3 + ax + b mod p (1)
one-time key stream cipher for color image encryption based
on Piecewise Linear chaotic map(PWLCM) to compute the where x, y, a, b ∈ Fp , and 4a3 + 27b2 mod p 6= 0 with a point
key stream for color image encryption. The generation of of infinity, O.
key stream depends on another sequence obtained from the
Chebyshev map.
Liu et al. [4] devised an encryption algorithm with DNA
encoding where the XOR and complement operation is car-
ried out using the random sequences generated by the logistic
map to encrypt individual components of the color image.
However, Liu et al. [5] proved that the algorithm can be
broken by the known-plaintext attack with only one known
pair of plain-cipher images. Luo et al. [9] used two chaotic
FIGURE 1. An Elliptic Curve E defined over a finite field Fp .
sequences generated by Tent map for the permutation and
diffusion of pixels in grayscale images. The initial seed value
of Tent map depends on the plain image. Singh and Singh [10]
1) GROUP OPERATIONS ON ELLIPTIC CURVE
proposed an Elliptic curve(EC) block based image encryption
scheme with digital signature for cipher image authentication. We assume P and Q as two unique points on the elliptic
The scheme faces the cipher data expansion issue due to curve E. The group operations on the curve E are as follows.
additional amount of pixels added during encryption. Singh • Point addition: The point addition of P and Q is given
and Singh [11] presented another refined Elliptic curve based by
algorithm which uses 2D Arnold Cat map for pixel shuffling P+Q=T (2)
and improved ElGamal encryption encoding techniques. The
scheme eliminated the data expansion issue. The line that joins the given points P and Q intersects
Xu et al. [12] designed a block based image encryption the curve E at the point (−T ). The mirror image of the
that uses a quantized chaos matrix obtained from the logistic point (−T ) with respect to the abscissa is thus obtained
map for creating X coordinate, Y coordinate and swapping to be T .
control table to swap pixels together with dynamic indexes for • Point subtraction: The point subtraction of P and Q is
diffusion. Saljoughi and Mirvaziri [22] designed an encryp- given by
tion model which uses 3D logistic maps and XOR opera- P − Q = P + (−Q) (3)
tion for row-column pixel permutation. Broumandnia [24]
presented a color image encryption that uses a 3D Modular where Q = (x, y) then (−Q) = (x, −y).
Chaotic Map(3DMCM) for pixel shuffling with exclusive • Point doubling: The addition of the curve point P to itself
OR and circular shift operation for pixel substitution. Kumar results in a point S on the curve E. This is known as point
and Girdhar [26] used the Lorenz-Rossler chaotic system doubling and is given by
for pixel diffusion and 2D logistic maps for confusion as P + P = 2P = S (4)
well as DNA cryptography for encryption of pixels of color
images. The tangent drawn at the point P intersects the curve at
The paper is organized as follows. Section II delves into (−S). The mirror image of the point,(−S) with respect
the theoretical view of our model. In Section III, the pro- to x-axis is the point S.
posed EC based image encryption algorithm and the digital • Point multiplication: The point multiplication of the
signature is described in detail. In section IV, we review the point P with a scalar, m is computed by performing m
experimental results acquired from the proposed model with repetitive point additions of P with itself.
the two scrambling methods. In Section V, we analyse the mP = P + P + P + . . . + P (m times) (5)
strength of the proposed model against various statistical and
intruder attacks. In Section VI, we discuss the performance where m is a scalar, m ∈ Zp .

76192 VOLUME 9, 2021

P. Parida et al.: Image Encryption and Authentication With ECC and Multidimensional Chaotic Maps

2) ELLIPTIC CURVE DIFFIE–HELLMAN (ECDH) KEY

EXCHANGE
The ECDH key exchange is used for secure exchange of a
shared session key between two users in a network. The key
exchange is carried out as follows.
• User A generates its random secret key, nA . A then
computes the public key PA using his private key, nA and
generator point, G on the curve as,
PA = nA ∗ G (6)
and sends it to user B.
• User B generates its random secret key, nB . B then
computes the public key PB using his private key, nB and FIGURE 2. 4D Horizontal Block Reordering of Grayscale image.

generator point, G on the curve as,

PB = nB ∗ G (7)
and sends it to user A.
• Both the users A and B compute the shared ECDH key,
ZAB as in (8) and (9).
A computes ZAB as follows.

ZAB = nA ∗ PB = nA ∗ nB ∗ G (8)
B computes ZAB as follows.

ZAB = nB ∗ PA = nB ∗ nA ∗ G (9) FIGURE 3. 4D Horizontal Block Reordering of a RGB image.

B. ARNOLD’S CAT MAP

Liu et al. [29] modified the general form of 3D Arnold’s map
Arnold’s Cat map [28] is a popular discrete chaotic map that to carry out pixel position scrambling with subsequent value
performs scrambling of image pixels by P repetitive shear transform as given in (12).
mapping on the input image to obtain the originally taken  ! ! !
0
image again. The parameter P is defined as the periodicity of  x = 1
 a
·
x
mod N
the transformation map. The cat map is known for its remark- y0 b ab + 1 y (12)
able mixing properties and high sensitivity to the initial state 
 0
p = (cx + dy + p) mod M
and control parameters of the map.
where p and p’ in equation (12) are the pixel values of
1) 2D ARNOLD’S CAT MAP the initial and remodeled images, and M is maximum pixel
The two-dimensional Arnold’s cat map can be expressed as intensity. The values of pixel intensity in an image varies from
follows. 0 to 255. The modified 3D cat map can also be extended
 0 
x 1 a
  
x to 4D mapping by reordering the blocks of pixels between
= · mod N (10) the image components. The block reordering can be carried
y0 b ab + 1 y
out either vertically [31] or as horizontally [32]. Fig.2 and
where the positive integers, a and b are the control parameters Fig.3 show the horizontal block reordering for a grayscale
and N is the dimensional value of a N × N image. image and shuffling among the individual components of a
color(RGB) image.
2) EXTENSION TO HIGHER DIMENSIONS
The 2D Arnold’s Cat map can be extended to higher dimen- III. PROPOSED MODEL
sional mappings to perform more secure and efficient trans- The proposed model is divided into three phases viz.
forms. The three-dimensional Arnold’s mapping is performed ECDH Key Exchange, Encryption/Decryption, and Sign-
with the introduction of two new control parameters, c and d ing/Verification of encrypted data. The encryption is carried
[1]. The general 3D extended form of existing 2D cat map is out on blocks of pixels instead of individual pixels to increase
as follows. efficiency and speed of encryption. The model uses the
 0    
x 1 a 0 x improved ElGamal encryption encoding [11] for encryption
 y0  =  b ab + 1 0  ·  y  mod N (11) of scrambled pixels. We assume A and B are two users in the
z0 c d 1 z network where Sender A wants to transmit data(i.e.images) to

VOLUME 9, 2021 76193

 P. Parida et al.: Image Encryption and Authentication With ECC and Multidimensional Chaotic Maps

Receiver B in a secure manner. The private and public key pair hxy = SHA256 (kxy ) (16)
of Sender,A and Receiver,B are (nA , PA ) and (nB , PB ) respec-
2) Perform hash of the encrypted image C as hC and
tively. A and B compute the ECDH session key, ZAB through
concatenate hC with hxy to obtain the combined hash
their private-public key pairs using equations (8) and (9).
value H .
A digital image is characterised by a matrix of smaller
image elements called pixels. The number of pixels that hC = SHA256 (C) (17)
can be grouped together is limited by the size of the prime H = (hC || hxy ) (18)
elliptic curve parameter(p). A 512-bit elliptic curve limits the
maximum number of pixels in a group to 64. 3) Compute the parameter, k’ as follows.
The scrambling of pixels using Arnold’s Cat map can be
Z = ZABx ⊕ ZABy (19)
performed in two ways. 0
• Scheme 1: 3D Cat map transform k = Z ⊕k (20)
• Scheme 2: 4D Cat map transform 4) Calculate the signature (R, U ) as follows.
The proposed design of the image encryption model,
as shown in the Fig. 4 and 5, is explained as follows. R = SHA256 (H ) (21)
U = ((k 0 ) − R) mod n (22)
A. ENCRYPTION
1) Record the pixel intensities, dimensional size and chan- 5) The encrypted image C is sent to the receiver along with
nel information from the original image. Adjust the digital signature,(R, U ).
values of pixels by random subtraction or addition of 1
or 2 to each pixels. C. VERIFICATION OF ENCRYPTED IMAGE
2) Choose a random integer k between 2 to (n − 1), where 1) Compute the value of kG from (R, U ) and ZAB .
n is the cyclic order of the 512-bit elliptic curve. Z = ZABx ⊕ ZABy (23)
3) Scrambling of pixels can be carried out in following
k = ((R + U ) ⊕ Z ) (24)
ways.
• Scheme 1: 3D Cat Map Scrambling kG = (k ∗ G) (25)
Shuffle the pixels for j rounds using modified 3D 2) Calculate the hash value hxy from kG.
Cat map equation in (12) with control parameters
a = 1 and b = 1. kxy = kG.x ⊕ kG.y (26)
hxy = SHA256 (kxy ) (27)
j = kGx mod P (13)
3) Perform hash of the encrypted image C as hC . Combine
where kGx = x-coordinate of kG and P is the
hC with hxy to obtain the hash value H .
period of the 3D Arnold scrambling for a given
image of size N × N. hC = SHA256 (C) (28)
• Scheme 2: 4D Cat Map Scrambling H = (hC || hxy ) (29)
After shuffling the pixels for j rounds using 3D Cat
map, perform horizontal block reordering of the 4) Calculate R0 from computed hash values. If R’ is equal
scrambled image as described in Fig.2 and Fig.3. to R received digital signature (R, U ) from Sender A,
4) Partition the scrambled image pixels into groups then the signature is verified.
of 64 each. Convert each group into a single large R0 = SHA256 (H ) (30)
integer value with 256 as base.
5) Pair up every successive two integer values to form the R0 == R (31)
plain text input list,PM . Compute cipher text list,PC as
D. DECRYPTION
follows.
1) Record the pixel intensities, size and related informa-
PC = PM + (kG + ZAB ) (14) tion from the encrypted image,C.
2) Partition the encrypted image pixels into groups
6) Convert cipher text list, PC to byte values with base
of 64 each. Convert each group into a single large
256. Ensure each list has 64 pixel values with required
integer value with 256 as base.
left zeroes padding.
3) Pair up the integer values as cipher text list,PC process
7) Convert the cipher pixel values into the cipher
and compute plain text list, PM as follows.
image, C.
PM = PC − (kG + ZAB ) (32)
B. SIGNING THE ENCRYPTED IMAGE
1) Compute the hash value hxy from kG as follows. 4) Convert calculated plain text list, PM back to byte val-
ues with base as 256 and ensure each list has 64 values
kxy = kGx ⊕ kGy (15) with necessary left zero padding.

76194 VOLUME 9, 2021

P. Parida et al.: Image Encryption and Authentication With ECC and Multidimensional Chaotic Maps

FIGURE 4. Overview of the Proposed Model.

5) Unscramble the scrambled pixels in two ways. TABLE 1. Elliptic curve parameters used in the implementation of
proposed scheme.
• Scheme 2:4D Cat map Unscrambling
Reorder the blocks of scrambled image back to
generate the 3D scrambled image.
• Scheme 1:3D Cat map Unscrambling
Unscramble the scrambled pixel values for P-j
rounds to obtain the value adjusted image.
6) Adjust the values of pixels back by random subtraction
or addition of 1 or 2 to each pixel to attain the original
image again.

IV. EXPERIMENTAL RESULTS

The proposed model is implemented using Python 3.7 with
Spyder 4.1.5 IDE on a HP laptop with system configuration
as Intel(R) Core(TM) i5-8250U [email protected] with 8GB
RAM. The Elliptic curve used in the proposed algorithm is
the secure 512-bit curve from ECC Brainpool [33]. The ellip-
tic curve parameters used for implementation are depicted
in Table 1.
cipher image with even distribution of frequencies across the
histogram graph plot. Grayscale images taken in considera-
V. SECURITY ANALYSIS
tion are the Lena, Barbara, and Mandrill. RGB images taken
Various security and statistical tests are carried out to explore
as input images include Lena, House, and Mandrill. Fig. 6
the strength of the proposed algorithm with the two imple-
and Fig.7 illustrate the plots for input plain grayscale and
mentations of the algorithm using the explained scrambling
RGB images with their cipher images from Scheme 1 and
methods, Scheme 1 and Scheme 2. The proposed model is
2 scrambling methods respectively. The cipher images show
tested on grayscale and color(RGB) images sourced from the
the uniform frequency distribution of pixels across the his-
SIPI Image Database [34]. The analyses carried out include
tograms indicating good quality of cipher images.
histogram analysis, variance analysis, chi-square test, key
space analysis, Shannon’s entropy of cipher images, mea-
B. HISTOGRAM VARIANCE ANALYSIS
suring resistance to differential attacks with NPCR(Number
Variance is a quantitative measure of resistance against statis-
of Pixel Change Rate) and UACI(Unified Average Changed
tical attacks by analysing the distribution of pixel frequencies
Intensity) parameters, similarity measurement (PSNR and
in the histogram of a grayscale image. Variance of a histogram
SSIM), resistance to intruder attacks, correlation coefficient
can be computed as follows.
analysis and computation cost comparison.
P P
1 X X (pi − pj )2
A. HISTOGRAM ANALYSIS VAR(Z ) = (33)
P2 2
i=1 j=1
Histogram of an image plots the frequency of pixel intensities
in the ordinate against the values of the pixel intensities in P is the grayscale pixel intensity value and the parameters, pi
the abscissa. An effective encryption algorithm produces a and pj are the number of pixels with intensity value i and j.

VOLUME 9, 2021 76195

 P. Parida et al.: Image Encryption and Authentication With ECC and Multidimensional Chaotic Maps

FIGURE 5. Proposed design model.

Lower is the computed variance, higher is the uniformity of where the OFi and EFi are the observed and expected values
pixel distribution of the histogram. From Table 2, we infer of frequency of pixel intensity value i.EFi is
that the cipher images obtained from the proposed model
M ×N
have lower values of variance indicating high resistance to EFi = (35)
statistical attacks. Imax
Imax is the total number of pixel intensities for a grayscale
C. CHI-SQUARE TEST ANALYSIS image. Table 3 depicts the calculated values of Chi-square
Chi-square test measures the uniformity of pixel distribu- for cipher images obtained from the proposed model with
tion in the histogram of grayscale images quantitatively. The significance level of 1% and 5% at 255 degrees of free-
Chi-square value can be evaluated as follows. dom. The ideal values of Chi-square at 1% and 5% sig-
255 nificance level are χ255,0.01
2 = 310.457 and χ255,0.05
2 =
X (OFi − EFi )2
χtest
2
= (34) 293.2478. Lower the value of Chi-square, higher is the uni-
EFi formity of pixel distribution in the histogram. At 1% and 5%
i=0

76196 VOLUME 9, 2021

P. Parida et al.: Image Encryption and Authentication With ECC and Multidimensional Chaotic Maps

FIGURE 6. (a-f):Plain grayscale images with their histograms, (g-l):Cipher images with Scheme 1
scrambling with their histograms, (m-r):cipher images with Scheme 2 scrambling with respective
histogram.

FIGURE 7. (a-f):Plain RGB images with their histograms, (g-l):Cipher images with Scheme 1 scrambling
with their histograms, (m-r):Cipher images with Scheme 2 scrambling with respective histogram.

significance levels, the cipher images obtained from Discrete Logarithm Problem(ECDLP), in feasible amount of
Scheme 1 and 2 show Chi-square values lower than the ideal time, to provide higher level of security with shorter key
values. The calculated Chi-square values satisfy the hypoth- sizes. The implementation of the proposed model uses the
esis and implies grayscale pixel uniformity in the cipher ECC Brainpool [33] standard 512-bit Elliptic curve which can
images derived from the proposed model. successfully withstand brute-force attacks.

D. KEY SPACE E. SHANNON’S ENTROPY ANALYSIS

The robustness of an algorithm depends upon the key size The Shannon’s theory [35] states that entropy is directly
of the cryptosystem. The proposed algorithm relies on the proportional to the degree of uncertainty present in the data.
inability of solving the exponentially difficult Elliptic Curve An encryption algorithm should ideally produce a cipher

VOLUME 9, 2021 76197

 P. Parida et al.: Image Encryption and Authentication With ECC and Multidimensional Chaotic Maps

TABLE 2. Variance of various grayscale images. G. SIMILARITY MEASUREMENT

The Peak Signal-to-Noise ratio(PSNR) and Structural Sim-
ilarity(SSIM) index values measure the similarity between
the plain and cipher images. The PSNR value edges nearer
to infinity as Mean Squared Error(MSE) approaches zero.
SSIM index estimates the perceived quality of images from
its structural information. Lower the PSNR and SSIM values
are between plain and cipher data, better is the quality of
encryption. The measured values of PSNR and SSIM for the
proposed schemes are given in Table 8. The SSIM values
of cipher images from the tested grayscale and RGB images
approach zero with reduced values of PSNR indicating good
quality of encryption. The Scheme 2 has lower similarity
between original and cipher images than Scheme 1 making
TABLE 3. Chi-square analysis of proposed model.
it a better mode of encryption.

Pmax
PSNR = 20 × log10 √ (39)
MSE
where Pmax is the maximum pixel intensity.

H. KNOWN-PLAINTEXT ATTACK(KPA) AND

CHOSEN-PLAINTEXT ATTACK(CPA)
Known-Plaintext and Chosen-Plaintext attacks are two attack
models for cryptanalysis where intruder can exploit the
image with entropy value of 8. The entropy values are com- knowledge of one or more plaintext-ciphertext pairs from
puted for the cipher images obtained from both implementa- an encryption method to reveal secret information. In case
tions of the model in Table 4 and 5. The cipher images of the of Known-Plaintext attacks, we assume the intruder has
tested grayscale as well as color images display better entropy access to one or more plain images and their corresponding
values than existing image encryption schemes. cipher images. The proposed algorithm generates different
and unique cipher images for the same plain image due to use
F. RESISTANCE TO DIFFERENTIAL ATTACKS of the random parameter k, value adjustment, and efficient
Differential attacks are a form of cryptanalysis to find the scrambling methods in each session of encryption.
secret key by tracing the differences in the cipher data In Chosen-Plaintext attacks, the intruder can choose certain
due to minimal changes in the plain data. The Number of plaintext images and use the respective cipher images to
Changing Pixel Rate(NPCR) and Unified Average Changed unveil the secret key. We assume the intruder chooses two
Intensity(UACI) values, computed from the cipher images special images i.e. all black and all white images as the plain
acquired from a pair of nearly similar plain images, are images for the CPA attack. From Fig. 8 and Tables 9 and 10,
used to measure the resilience of the algorithm against such we infer that the proposed method produces good quality
attacks. The ideal value of NPCR for an encryption is 100% cipher images for the special input images with low cor-
with UACI value being around 33%. From Table 6 and 7, relation, uniform cipher pixel distribution and high entropy
it can be ascertained that the cipher images derived from
the scheme approximate near-ideal values for the NPCR and
UACI parameters.
P
i,j D(i, j)
NPCR = × 100% (36)
P T 0
i,j |C(i, j) − C (i, j)|
UACI = × 100% (37)
Imax × T
where
(
0 if C(i, j) = C 0 (i, j)
D(i, j) = (38)
1 if C(i, j) 6 = C 0 (i, j)
FIGURE 8. Plain All Black and All White images with their respective
T = Total number of pixels, C and C’ are the two cipher encrypted images obtained from the proposed encryption model, and the
images, and Imax = Total number of Pixel Intensities. histograms of the encrypted images.

76198 VOLUME 9, 2021

P. Parida et al.: Image Encryption and Authentication With ECC and Multidimensional Chaotic Maps

TABLE 4. Shannon’s entropy analysis for grayscale images.

TABLE 5. Shannon’s entropy analysis for RGB images.

TABLE 6. NPCR and UACI analysis for grayscale tested images.

TABLE 7. Average values for NPCR and UACI analysis for RGB tested images.

making it difficult for the intruder to elicit secret information decrypted grayscale and color images acquired from the dis-
from the plain-cipher image pairs. torted cipher images through Scheme 1 and 2. The decrypted
images show that the proposed model can resist the occlusion
I. OCCLUSION ATTACK ANALYSIS
attacks successfully.
The encrypted images can suffer from data loss due to
varying reasons while transmitting data between sender and
receiver. The ability to recognise the original images from the J. CORRELATION COEFFICIENT ANALYSIS
distorted images signifies the strength of the proposed model Correlation coefficient is a statistical measure of strength
against such occlusion attacks. Figures 9 and 10 depict the of relationship between the original and encrypted image.

VOLUME 9, 2021 76199

 P. Parida et al.: Image Encryption and Authentication With ECC and Multidimensional Chaotic Maps

TABLE 8. Similarity analysis between two proposed schemes.

FIGURE 9. Decrypted grayscale images from Scheme 1 and 2 under

FIGURE 10. Decrypted color images from Scheme 1 and 2 under
Occlusion attack analysis. (a-d) are cipher images from Scheme 1 having
Occlusion attack analysis. (a-d) are cipher images from Scheme 1 having
25%, 50% vertical, 50% horizontal and 75% data loss. (e-h) are the
25%, 50% vertical, 50% horizontal and 75% data loss. (e-h) are the
corresponding decrypted images for Scheme 1 Occlusion analysis.(i-l) are
corresponding decrypted images for Scheme 1 Occlusion analysis.(i-l) are
cipher images from Scheme 2 having 25%, 50% vertical, 50% horizontal
cipher images from Scheme 2 having 25%, 50% vertical, 50% horizontal
and 75% data loss. (m-hp) are the corresponding decrypted images for
and 75% data loss. (m-hp) are the corresponding decrypted images for
Scheme 2 Occlusion analysis.
Scheme 2 Occlusion analysis.

TABLE 9. Histogram uniformity analysis for special input images.

vertically and diagonally. The correlations graph plots for the
grayscale Lena image and its cipher image from the proposed
model are illustrated in Fig. 11. Fig. 12 shows the correlation
graph plots for the component blue of color Lena image and
its cipher image. Table 11 and 12 display the comparison of
correlation coefficient values obtained from Scheme 1 and
Scheme 2 implementation of the tested grayscale and RGB
images with those from recent image encryption algorithms.
The correlation coefficient values for tested grayscale and
RGB images are closer to zero in comparison to the existing
A plain image has high correlation between the adjoining encryption models.
pixels. The correlation graph for the plain image as a conse-
quence is dense. A strong and desirable cipher image should K. COMPUTATIONAL COST COMPARISON
have an evenly distributed graph with low correlation val- A robust image encryption algorithm should perform encryp-
ues between the adjoining pixels. We compute correlation tion on images with minimal consumption of computing
coefficient between the plain and cipher images horizontally, resources. The time consumption of an EC based encryption

76200 VOLUME 9, 2021

P. Parida et al.: Image Encryption and Authentication With ECC and Multidimensional Chaotic Maps

TABLE 10. Performance of proposed model on special input images.

TABLE 11. Comparison of correlation coefficient values for grayscale images.

algorithm is determined by the number of time intensive point

multiplications it has. Lesser the number of point multiplica-
tions, faster is the image encryption algorithm. In Table 13,we
carried out a comparative analysis of time consumption
among recent EC based image encryption algorithms. The
proposed model has lower computational costs with lesser
number of point multiplications.

VI. DISCUSSION
The performance of the proposed model is tested on two
implementations of model, Scheme 1 using 3D Arnold Cat
map and Scheme 2 using the 4D Arnold Cat map. The
extended Arnold Cat maps add additional layers of value
substitution after the numerous rounds of pixel scrambling
which increases the security as well as quality of encryp-
tion. The model generates better-quality cipher images as
evident from the uniform pixel frequency distribution in
histogram analysis as can be seen in the Fig. 6 and Fig.7. FIGURE 11. Correlation graphs of plain and cipher grayscale LENA images
from their respective proposed Scheme 1 and Scheme 2 implementations.
The cipher images obtained are unique and distinct with Here, (a - c) are horizontal correlation graphs, (c - e) are vertical
high entropy, low PSNR and SSIM values and reduced correlation graphs and (f-i) are diagonal correlation graphs.
correlation coefficient values between neighbouring pix-
els. The model has high resistance to differential attacks
with an average NPCR of 99.62% and UACI of 33.3%. Scheme 2(4D Arnold Cat Map) performs better than
From the performance comparisons between the proposed Scheme 1(3D Arnold Cat Map) scrambling in compari-
model and existing encryption schemes in Tables 4–13, son. The proposed algorithm is time-efficient with fewer
we infer that the proposed scheme has higher efficacy with time-consuming EC point multiplications. A lightweight dig-
stronger resistance against the security and statistical attacks. ital signature algorithm using the secure SHA-256 hash

VOLUME 9, 2021 76201

 P. Parida et al.: Image Encryption and Authentication With ECC and Multidimensional Chaotic Maps

TABLE 12. Comparison of correlation coefficient values for RGB images.

TABLE 13. Computational cost comparison analysis.

VII. CONCLUSION
The proposed image encryption and authentication model
intends to enhance the quality of encryption of both grayscale
and color images through the discretized chaotic 3D and
4D Arnold Cat maps. The model strengthens the qual-
ity of cipher images compared to existing schemes with
higher entropy, lower correlation, higher average NPCR
and UACI, lower PSNR and SSIM values along with
the ability to actively thwart the Chosen-Plaintext(CPA)
and Known-Plaintext attacks(KPA). The proposed model
is proven to be robust, lightweight and competent against
the statistical and cryptanalytic attacks. In future work,
the encryption model can be amended to support other forms
of real-time multimedia encryption such as audio, video and
more.

REFERENCES
FIGURE 12. Correlation graphs of Blue component of plain and cipher
color LENA images from their respective proposed Scheme 1 and [1] G. Chen, Y. Mao, and C. K. Chui, ‘‘A symmetric image encryption scheme
Scheme 2 implementations. Here, (a - c) are horizontal correlation based on 3D chaotic cat maps,’’ Chaos, Solitons Fractals, vol. 21, no. 3,
graphs, (c - e) are vertical correlation graphs and (f-i) are diagonal pp. 749–761, Jul. 2004.
correlation graphs.
[2] N. K. Pareek, V. Patidar, and K. K. Sud, ‘‘Image encryption using chaotic
logistic map,’’ Image Vis. Comput., vol. 24, no. 9, pp. 926–934, Sep. 2006.
[3] H. Liu and X. Wang, ‘‘Color image encryption based on one-time keys
function is used to verify the validity of cipher image before and robust chaotic maps,’’ Comput. Math. with Appl., vol. 59, no. 10,
decryption for faster decryption of correct cipher images. pp. 3320–3327, May 2010.

76202 VOLUME 9, 2021

P. Parida et al.: Image Encryption and Authentication With ECC and Multidimensional Chaotic Maps

[4] L. Liu, Q. Zhang, and X. Wei, ‘‘A RGB image encryption algorithm based [31] B. J. Saha, B. Jyoti, K. K. Kabi, and C. Pradhan, ‘‘A new approach on color
on DNA encoding and chaos map,’’ Comput. Electr. Eng., vol. 38, no. 5, image encryption using Arnold 4D Cat map,’’ in Computational Intelli-
pp. 1240–1248, Sep. 2012. gence in Data Mining, vol. 1. New Delhi, India: Springer, 2016. 131-138.
[5] Y. Liu, J. Tang, and T. Xie, ‘‘Cryptanalyzing a RGB image encryption [32] S. Kumar and C. Pradhan, ‘‘Color image encryption technique using
algorithm based on DNA encoding and chaos map,’’ Opt. Laser Technol., 4D logistic map,’’ in Progress in Computing, Analytics and Networking.
vol. 60, pp. 111–115, Aug. 2014. Singapore: Springer, 2020. 75-82.
[6] I. S. Sam, P. Devaraj, and R. S. Bhuvaneswaran, ‘‘A novel image cipher [33] Elliptic Curve Cryptography(ECC) Brainpool Standard Curves and
based on mixed transformed logistic maps,’’ Multimedia Tools Appl., Curve Generation. Accessed: Feb. 20, 2021. [Online]. Available:
vol. 56, no. 2, pp. 315–330, Jan. 2012. https://tools.ietf.org/html/rfc5639/
[7] G. Ye, ‘‘A block image encryption algorithm based on wave transmis- [34] USC Signal and Image Processing Insitute(SIPI) Image Database.
sion and chaotic systems,’’ Nonlinear Dyn., vol. 75, no. 3, pp. 417–427, Accessed: Feb. 20, 2021. [Online]. Available: http://sipi.usc.edu/database/
Feb. 2014. [35] C. E. Shannon, ‘‘Communication theory of secrecy systems,’’ Bell Syst.
[8] A. S. Rajput and M. Sharma, ‘‘A novel image encryption and authentication Tech. J., vol. 28, no. 4, pp. 656–715, 1949.
scheme using chaotic maps,’’ in Advances in Intelligent Informatics. Cham, [36] G.-D. Ye, X.-L. Huang, L. Y. Zhang, and Z.-X. Wang, ‘‘A self-cited pixel
Switzerland: Springer, 2015, pp. 277–286. summation based image encryption algorithm,’’ Chin. Phys. B, vol. 26,
[9] Y. Luo, L. Cao, S. Qiu, H. Lin, J. Harkin, and J. Liu, ‘‘A chaotic map- no. 1, Jan. 2017, Art. no. 010501.
control-based and the plain image-related cryptosystem,’’ Nonlinear Dyn., [37] D.-D. Liu, W. Zhang, H. Yu, and Z.-L. Zhu, ‘‘An image encryption scheme
vol. 83, no. 4, pp. 2293–2310, Mar. 2016. using self-adaptive selective permutation and inter-intra-block feedback
[10] L. D. Singh and K. M. Singh, ‘‘Image encryption using elliptic curve diffusion,’’ Signal Process., vol. 151, pp. 130–143, Oct. 2018.
cryptography,’’ Procedia Comput. Sci., vol. 54, pp. 472–481, 2015. [38] X. Chai, Z. Gan, K. Yang, Y. Chen, and X. Liu, ‘‘An image encryption
[11] D. S. Laiphrakpam and M. S. Khumanthem, ‘‘Medical image encryp- algorithm based on the memristive hyperchaotic system, cellular automata
tion based on improved ElGamal encryption technique,’’ Optik, vol. 147, and DNA sequence operations,’’ Signal Process., Image Commun., vol. 52,
pp. 88–102, Oct. 2017. pp. 6–19, Mar. 2017.
[12] L. Xu, X. Gou, Z. Li, and J. Li, ‘‘A novel chaotic image encryption [39] M. Y. Valandar, M. J. Barani, and P. Ayubi, ‘‘A fast color image encryption
algorithm using block scrambling and dynamic index based diffusion,’’ technique based on three dimensional chaotic map,’’ Optik, vol. 193,
Opt. Lasers Eng., vol. 91, pp. 41–52, Apr. 2017. Sep. 2019, Art. no. 162921.
[13] J. Wu, X. Liao, and B. Yang, ‘‘Color image encryption based on chaotic [40] A. Yaghouti Niyat, M. H. Moattar, and M. Niazi Torshiz, ‘‘Color image
systems and elliptic curve ElGamal scheme,’’ Signal Process., vol. 141, encryption based on hybrid hyper-chaotic system and cellular automata,’’
pp. 109–124, Dec. 2017. Opt. Lasers Eng., vol. 90, pp. 225–237, Mar. 2017.
[14] S. Sun, ‘‘Chaotic image encryption scheme using two-by-two deoxyri- [41] X. Wang, X. Qin, and C. Liu, ‘‘Color image encryption algorithm based
bonucleic acid complementary rules,’’ Opt. Eng., vol. 56, no. 11, 2017, on customized globally coupled map lattices,’’ Multimedia Tools Appl.,
Art. no. 116117. vol. 78, no. 5, pp. 6191–6209, Mar. 2019.
[15] X. Chai, Y. Chen, and L. Broyde, ‘‘A novel chaos-based image encryption [42] M. Kumar, A. Iqbal, and P. Kumar, ‘‘A new RGB image encryption
algorithm using DNA sequence operations,’’ Opt. Lasers Eng., vol. 88, algorithm based on DNA encoding and elliptic curve Diffie–Hellman
pp. 197–213, Jan. 2017. cryptography,’’ Signal Process., vol. 125, pp. 187–202, Aug. 2016.
[16] S. Amina and F. K. Mohamed, ‘‘An efficient and secure chaotic cipher algo- [43] Y. Niu, Z. Zhou, and X. Zhang, ‘‘An image encryption approach based
rithm for image content preservation,’’ Commun. Nonlinear Sci. Numer. on chaotic maps and genetic operations,’’ Multimedia Tools Appl., vol. 79,
Simul., vol. 60, pp. 12–32, Jul. 2018. nos. 35–36, pp. 25613–25633, Sep. 2020.
[17] Z. Hua, F. Jin, B. Xu, and H. Huang, ‘‘2D Logistic-Sine-coupling map for [44] S. Toughi, M. H. Fathi, and Y. A. Sekhavat, ‘‘An image encryption scheme
image encryption,’’ Signal Process., vol. 149, pp. 148–161, Aug. 2018. based on elliptic curve pseudo random and advanced encryption system,’’
[18] Z. Liu, T. Xia, and J. Wang, ‘‘Image encryption technique based on Signal Process., vol. 141, pp. 217–227, Dec. 2017.
new two-dimensional fractional-order discrete chaotic map and Menezes– [45] A. Kanso and M. Ghebleh, ‘‘A novel image encryption algorithm based on
Vanstone elliptic curve cryptosystem,’’ Chin. Phys. B, vol. 27, no. 3, 2018, a 3D chaotic map,’’ Commun. Nonlinear Sci. Numer. Simul., vol. 17, no. 7,
Art. no. 030502. pp. 2943–2959, Jul. 2012.
[19] D. S. Laiphrakpam and M. S. Khumanthem, ‘‘A robust image encryption [46] F. Musanna and S. Kumar, ‘‘A novel fractional order chaos-based image
scheme based on chaotic system and elliptic curve over finite field,’’ encryption using Fisher yates algorithm and 3-D cat map,’’ Multimedia
Multimedia Tools Appl., vol. 77, no. 7, pp. 8629–8652, Apr. 2018. Tools Appl., vol. 78, no. 11, pp. 14867–14895, Jun. 2019.
[20] W.-K. Lee, R. C.-W. Phan, W.-S. Yap, and B.-M. Goi, ‘‘SPRING: A novel [47] B. Norouzi, S. M. Seyedzadeh, S. Mirzakuchaki, and M. R. Mosavi,
parallel chaos-based image encryption scheme,’’ Nonlinear Dyn., vol. 92, ‘‘A novel image encryption based on hash function with only two-round
no. 2, pp. 575–593, Apr. 2018. diffusion process,’’ Multimedia Syst., vol. 20, no. 1, pp. 45–64, Feb. 2014.
[21] Y. Zhang, ‘‘The unified image encryption algorithm based on chaos and [48] H. Liu, A. Kadir, X. Sun, and Y. Li, ‘‘Chaos based adaptive double-image
cubic S-box,’’ Inf. Sci., vol. 450, pp. 361–377, Jun. 2018. encryption scheme using hash function and S-boxes,’’ Multimedia Tools
[22] A. Shokouh Saljoughi and H. Mirvaziri, ‘‘A new method for image encryp- Appl., vol. 77, no. 1, pp. 1391–1407, Jan. 2018.
tion by 3D chaotic map,’’ Pattern Anal. Appl., vol. 22, no. 1, pp. 243–257, [49] X. Wu, J. Kurths, and H. Kan, ‘‘A robust and lossless DNA encryp-
Feb. 2019. tion scheme for color images,’’ Multimedia Tools Appl., vol. 77, no. 10,
[23] Y. Luo, X. Ouyang, J. Liu, and L. Cao, ‘‘An image encryption method based pp. 12349–12376, May 2018.
on elliptic curve elgamal encryption and chaotic systems,’’ IEEE Access, [50] B. Yang and X. Liao, ‘‘A new color image encryption scheme based on
vol. 7, pp. 38507–38522, 2019. logistic map over the finite field ZN,’’ Multimedia Tools Appl., vol. 77,
[24] A. Broumandnia, ‘‘The 3D modular chaotic map to digital color image no. 16, pp. 21803–21821, Aug. 2018.
encryption,’’ Future Gener. Comput. Syst., vol. 99, pp. 489–499, Oct. 2019. [51] A. U. Rehman, X. Liao, R. Ashraf, S. Ullah, and H. Wang, ‘‘A color
[25] R. I. Abdelfatah, ‘‘Secure image transmission using chaotic-enhanced image encryption technique using exclusive-OR with DNA complemen-
elliptic curve cryptography,’’ IEEE Access, vol. 8, pp. 3875–3890, 2020. tary rules based on chaos theory and SHA-2,’’ Optik, vol. 159, pp. 348–367,
[26] V. Kumar and A. Girdhar, ‘‘A 2D logistic map and Lorenz-Rossler chaotic Apr. 2018.
system based RGB image encryption approach,’’ Multimedia Tools Appl.,
PRIYANSI PARIDA received the B.Tech. degree
vol. 80, no. 3, pp. 3749–3773, 2021.
in computer science and engineering from the
[27] N. Koblitz, ‘‘Elliptic curve cryptosystems,’’ Math. Comput., vol. 48,
no. 177, pp. 203–209, 1987.
National Institute of Technology at Rourkela,
[28] V. I. Arnol’d and A. Avez, Ergodic Problems of Classical Mechanics. New Rourkela, India, in 2018. She is currently pursu-
York, NY, USA: W. A. Benjamin, Jan. 1968. ing the M.Tech. degree in computer science and
[29] H. Liu, Z. Zhu, H. Jiang, and B. Wang, ‘‘A novel image encryption engineering with the Kalinga Institute of Indus-
algorithm based on improved 3D chaotic cat map,’’ in Proc. 9th Int. Conf. trial Technology (KIIT) Deemed to be University,
Young Comput. Scientists, Nov. 2008, pp. 3016–3021. Bhubaneswar, India. Her research interests include
[30] P. N. Khade and M. Narnaware, ‘‘3D chaotic functions for image encryp- cryptography, authentication protocols, and data
tion,’’ Int. J. Comput. Sci. Issues (IJCSI), vol. 9, no. 3, p. 323, 2012. security.

VOLUME 9, 2021 76203

 P. Parida et al.: Image Encryption and Authentication With ECC and Multidimensional Chaotic Maps

CHITTARANJAN PRADHAN received the bach- DIPTENDU SINHA ROY (Senior Member, IEEE)
elor’s, master’s, and Ph.D. degrees from the Dis- received the Ph.D. degree in engineering from
cipline of Computer Science and Engineering. the Birla Institute of Technology at Mesra, India,
He is currently working as an Associate Professor in 2010. In 2016, he joined the Department
with the School of Computer Engineering, Kalinga of Computer Science and Engineering, National
Institute of Industrial Technology (KIIT) Deemed Institute of Technology (NIT) Meghalaya, India,
to be University, Bhubaneswar, India. He has got a as an Associate Professor, where he has served as
total of 15 years of academic teaching experience the Chair for the Department of Computer Science
with more than 80 publications in reputed peer and Engineering. His current research interests
reviewed journals, edited books and conferences include software reliability, distributed and cloud
of national and international repute. He has published few books published computing, and the IoT, specifically working on application of artificial
by publishers, like LAP Lambert, IGI Global, and Elsevier. His research intelligence and machine learning for smart integrated systems.
interests include information security, image processing, deep learning, and
multimedia systems. He is also a member of various national and interna-
RABINDRA KUMAR BARIK (Member, IEEE)
tional professional societies in the field of Engineering and Research, such
received the M.Tech. and Ph.D. degrees from
as IET, IACSIT, CSI, ISCA, IAENG, and ISTE.
the Motilal Nehru National Institute of Tech-
nology Allahabad, Prayagraj, India, in 2009 and
2014, respectively. He is currently working as
an Assistant Professor with the School of Com-
puter Applications, KIIT Deemed to be University,
Bhubaneswar, India. He has published more than
20 international journals, like Springer, Elsevier,
and IGI Global. He has also published more than
30 conference papers in various top-level conferences, like Global-SIP,
CHASE, TENCON, and INDICON. He has more than 15 book chapters
XIAO-ZHI GAO received the D.Sc. (Tech.) degree on his credit. Prior to this, he has edited one book named as Cloud Com-
from the Helsinki University of Technology (now puting for Geospatial Big Data Analytics: Intelligent Edge, Fog and Mist
Aalto University), Finland, in 1999. In Jan- Computing (Springer Nature) in the series of Studies in Big Data. He is
uary 2004, he was appointed as a Docent (Adjunct doing collaborative research with The University of Texas at Dallas and
Professor) with the Helsinki University of Tech- the University of Rhode Island in the field of fog computing. His research
nology. He is currently working as a Professor interests include geospatial data science, geospatial big data infrastructure,
of data science with the University of Eastern geospatial database, geospatial cloud computing, fog computing, and IPR.
Finland, Finland. He has published more than He is a member of IAENG. He served as TPC and PC members in many
400 technical articles on refereed journals and conferences. He has received best paper awards in FICTA-2020, ICSCC-
more than 400 technical papers on international 2017, and ICECE-2017 conferences. He has selected for the MHRD Schol-
conferences, and his current Google Scholar H-index is 33. His research arship during the M.Tech. and Ph.D. degrees. He has qualified GATE-2007 in
interests include nature-inspired computing methods with their applications information technology conducted by IIT Kanpur. He is reviewing in many
in optimization, data mining, machine learning, control, signal processing, journals, like Springer, Elsevier, IEEE, and IGI Global.
and industrial electronics.

76204 VOLUME 9, 2021