Signal Processing ∎ (∎∎∎∎) ∎∎∎–∎∎∎

Contents lists available at ScienceDirect

Signal Processing
journal homepage: www.elsevier.com/locate/sigpro

A new RGB image encryption algorithm based on DNA

encoding and elliptic curve Diffie–Hellman cryptography
Manish Kumar a,n, Akhlad Iqbal c,d, Pranjal Kumar a,b
a
Department of Mathematics, Birla Institute of Technology and Science-Pilani, Hyderabad Campus, Hyderabad 500078, Telangana, India
b
Department of Electrical Engineering, Birla Institute of Technology and Science-Pilani, Hyderabad Campus, Hyderabad 500078,
Telangana, India
c
Centre for Mathematics and Statistics, School of Basic and Applied Sciences, Central University of Punjab, Bathinda, Punjab 151001, India
d
Department of Mathematics, Aligarh Muslim University, Aligarh 202002, Uttar Pradesh, India

a r t i c l e i n f o abstract

Article history: With the increasing use of media in communications, there is a need for image encryption
Received 18 August 2015 for security against attacks. In this paper, we have proposed a new algorithm for image
Received in revised form security using Elliptic Curve Cryptography (ECC) diversified with DNA encoding. The
3 January 2016
algorithm first encodes the RGB image using DNA encoding followed by asymmetric
Accepted 22 January 2016
encryption based on Elliptic Curve Diffie–Hellman Encryption (ECDHE). The proposed
algorithm is applied on standard test images for analysis. The analysis is performed on key
Keywords: spaces, key sensitivity, and statistical analysis. The results of the analysis conclude that the
DNA encoding proposed algorithm can resist exhaustive attacks and is apt for practical applications.
Elliptic curve
Finite field & 2016 Published by Elsevier B.V.
Asymmetric key
Image encryption

1. Introduction of security with a smaller key size as compared to RSA and

DES algorithms. With the recent advanced in DNA tech-
In the recent years, due to the increase in digitalization nologies, DNA computing has entered in the domain of
of media, image security has become a prime agenda for cryptography (particularly, in the field of image encryption).
secure transmission over unsecured channels. Images are Various image encryption schemes have been proposed
frequently used in diverse areas such as medical imaging, based on DNA encoding such as authors [3–13]. The com-
online teaching and advertising, etc. In real-time scenario, plimentary rule of DNA has been used to encrypt images
transmitting images through internet and storing it on [3,6]. Using DNA computation [7] have presented an image
various platforms (cloud server, hard-drive, etc.), the fun- encryption algorithm of one-time pad cryptography with
damental issue of protecting for confidentiality, integrity, DNA strands, which cannot be implemented for frequent
use (due to the use of one time pad). A coupled map lattice
and authenticity is a major concern. Since most of the data
was used to encrypt images [8], it has been designed by
is exchanged between anonymous parties. Therefore, there
using chaotic map (for simulation purpose, logistic map was
is a need for asymmetric encryption methods.
used). Recently, Zhang et al. [9] have proposed a new
Elliptic curves for the purpose of cryptography were first algorithm for image encryption based on multi-chaotic
proposed individually in [1,2]. ECDHE provides similar level maps using pseudo DNA operation rules to increase com-
plexity and ciphertext unpredictability of the algorithm;
n
Corresponding author.
[10] have used Lorenz chaotic system along with DNA
E-mail addresses: manish.math.bhu@gmail.com, computing to encrypt color image; [11] have presented a
manishkumar@hyderabad.bits-pilani.ac.in (M. Kumar). novel image fusion encryption algorithm based on DNA

http://dx.doi.org/10.1016/j.sigpro.2016.01.017
0165-1684/& 2016 Published by Elsevier B.V.

Please cite this article as: M. Kumar, et al., A new RGB image encryption algorithm based on DNA encoding and elliptic
curve Diffie–Hellman cryptography, Signal Processing (2016), http://dx.doi.org/10.1016/j.sigpro.2016.01.017i
 M. Kumar et al. / Signal Processing ∎ (∎∎∎∎) ∎∎∎–∎∎∎

sequence and hyper-chaotic system; and DNA subsequence for RGB images which is asymmetric in nature and also
based couple images encryption algorithm using chaotic provide perfect forward secrecy. The length of shifting
system has been proposed in [12]. Likewise, the authors sequence offers the users' to chose required lever of
[14] and [15] have proposed a new DNA encoding using security. The proposed algorithm is secure and immune
chaos maps. Image encryption algorithm using chaotic and against common attacks (such as known plain-text, chosen
cipher-text, cropped and noise attacks). Also, different
hyper-chaotic systems have been researched and reported
elliptic curves can be used in the algorithm to vary the
by several authors, such as [16–20].
need of encryption speed for real-time encryption. Secur-
However, DNA encoding associated with chaotic (or
ity analysis suggests that the proposed algorithm can be
hyper-chaotic) systems offer good encryption but sym-
used for secure transmission efficiently.
metric in nature and hence unable to perform perfect
forward secrecy. In order to overcome the above draw-
backs, in this paper, we have proposed a new image 2. Preliminaries
encryption algorithm using DNA computing and ECDHE
2.1. DNA encoding
Table 1
Eight rules of complementary DNA encoding. DNA sequencing is performed using four basic nucleic
acids namely, Adenine (A), Cytosine (C), Guanine (G),
Rule 1 Rule 2 Rule 3 Rule 4 Rule 5 Rule 6 Rule 7 Rule 8 Thymine (T). A and T; C and G are compliments of each
00 A 00 A 00  C 00  C 00  G 00  G 00  T 00 T
other [7]. The total number of possible combinations are
01 C 01 G 01  A 01  T 01  A 01  T 01  C 01 G 4! ¼ 24 out of which only 8 follow the complimentary rule
10 G 10 C 10  T 10  A 10  T 10  A 10  G 10 C as shown in Table 1. In DNA encoding each nucleotide is
11 T 11 T 11  G 11  G 11  C 11  C 11  A 11 A denoted with a binary number following the complimen-
tary rule, e.g.: A  00; C  01; G 10; T  11, so the decimal
Table 2 value 200 (11 00 10 00) will be represented as TAGA. The
Addition and subtraction rule for DNA. 8-bit pixel values of the image are converted into four 2-
bit DNA sequence [21].
þ A T C G – A T C G

G G C T A G G T C A
2.1.1. DNA sequences: addition and subtraction
C C A G T C C G A T
T T G A C T T A G C
Addition and subtraction phenomenon of DNA
A A T C G A A C T G sequence is very similar to traditional algebraic com-
putations. The addition and subtraction is done over

Fig. 1. DNA encoding of an RGB image.

Please cite this article as: M. Kumar, et al., A new RGB image encryption algorithm based on DNA encoding and elliptic
curve Diffie–Hellman cryptography, Signal Processing (2016), http://dx.doi.org/10.1016/j.sigpro.2016.01.017i
 M. Kumar et al. / Signal Processing ∎ (∎∎∎∎) ∎∎∎–∎∎∎ 3

modulo 4 [22], adding numbers with DNA, the addition

and subtraction are given below in Table 2 when
00 A; 11  T; 01  C; and 10  G.

2.2. Elliptic curve cryptography

ECC is a notable mechanism in the field of public key

cryptography. ECC makes use of elliptic curves in which the
variables and coefficients are elements of a finite field. As
compared to the other algorithms (like, RSA and DES) it pro-
vides the same level of security with smaller key size.
Therefore, it helps in reducing storage and transmission
requirements.

2.2.1. Elliptic curve over finite field

From [23], for every prime power pk there exists a field
k
Fp with pk elements. An elliptic curve E is the set of solu-
tions to a generalized Weierstrass equation as follows:
E: Y 2 þa1 XY þ a3 Y ¼ X 3 þ a2 x2 þ a4 X þ a6 ;
together with an extra point O. The coefficients
a1 ; a2 ,..., a6 are required to satisfy ▵ a0, where the ▵ is
defined in terms of certain quantities b2 ; b4 ; b6 ; b8 given
below:

b2 ¼ a21 þ 4a2 ; b4 ¼ 2a4 þ a1 a3 ; b6 ¼ a23 þ4a6 ;

b8 ¼ a21 a6 þ 4a2 a6  a1 a3 a4 þ a2 a23  a24 ;
2 3 2
n ¼  b2 b8  8b4  27b6 þ 9b2 b4 b6 :

If P ¼ ðx1 ; y1 Þ and Q ¼ ðx2 ; y2 Þ are points on E as shown

in Fig. 2:

(i) P ¼ Q ¼ ðx1 ; ða1 x1 þ a3 Þ  y1 Þ

(ii) If x1 ¼ x2 but P aQ , then P  Q ¼ O
(iii) If x1 a x2 and P aQ , then the line through P and Q is
y ¼ λx þ γ, where λ ¼ y2 y1 =x2  x1 ; γ ¼ y1 x2  y2 x1 =
x2  x1 .

Also, if P ¼Q, then the tangent to E at P is y ¼ λx þγ,

where: λ ¼ 3x21 þ 2a2 x1 þ a4 a1 y1 =2y1 þ a1 x1  a3 ; γ ¼ 
x31 þ a4 x1 þ2a6 a3 y1 =2y1 þ a1 x1 þ a3 . Finally, if the line
Fig. 2. Elliptic addition of two points on the curve. through P and Q (respectively the tangent at P ¼Q) is

Fig. 3. Addition of pixel values using DNA encoding.

 M. Kumar et al. / Signal Processing ∎ (∎∎∎∎) ∎∎∎–∎∎∎

y ¼ λx þ γ, then P  Q ¼ ðx3 ; y3 Þ where: x3 ¼ λ2 þ a1 λ  and defined as:

a2 x1 x2 ; y3 ¼ ða1  λÞx3  γ a3 . qffiffiffiffiffi qffiffiffiffiffi
From Hasse's theorem [23], the number of points on pk þ 1  2 pk rEðFpk Þ rpk þ 1 þ 2 pk :
the elliptic curve E over finite field Fpk denoted by EðFpk Þ

2.2.2. Elliptic curve Diffie–Hellman encryption

An elliptic curve ða1 ; a2 ; a3 ; a4 ; a6 Þ and a point on the curve
k
ðX; YÞ belonging to the field F2 are made publicly available by
an authenticating third party. Party A chooses a private value
na ð0 rna r2k  1Þ and computes Q a ¼ na ðX; YÞ and sends Qa
to Party B. Similarly, Party B chooses a private value
nb ð0 rnb r2k  1Þ and computes Q b ¼ nb ðX; YÞ and sends Qb
to Party A. Party A then computes na Q b ¼ na nb ðX; YÞ and Party
B computes nb Q a ¼ na nb ðX; YÞ for which both the values are
same and hence act as private keys for encryption.
As we know, the ECDHE offers better security than RSA
algorithm. Since ECDHE requires key exchange between par-
ties, it provides perfect forward secrecy. If information needs
to be exchanged between multiple parties, each party can
provide keys on the same curve and hence remove the need
for unique public private key pairs between each parties or a
server to moderate the exchange of information as in RSA [24]
(for instance, if information needs to be communicated
between three parties: Party A, Party B and Party C. Each of
the parties will provide their public keys Q a ; Q b and Qc and
Fig. 4. Process of shifting: 1-Down and right; 2-Down; 3-Down and left;
then compute na Q b Q c ; nb Q a Q c and nc Q a Q b , respectively.
4-Left; 5-Up and left; 6-Up; 7-Up and right; 8-Left.
Hence the common private key will be na nb nc ð; XYÞ).

Fig. 5. Interleaving of an RGB image.

Fig. 6. ECC-SUCADD_FF: Elliptic curve cryptography using successive addition over finite field.

Please cite this article as: M. Kumar, et al., A new RGB image encryption algorithm based on DNA encoding and elliptic
curve Diffie–Hellman cryptography, Signal Processing (2016), http://dx.doi.org/10.1016/j.sigpro.2016.01.017i
 M. Kumar et al. / Signal Processing ∎ (∎∎∎∎) ∎∎∎–∎∎∎ 5

3. Proposed algorithm chosen depending upon the level of security required. The
amount of shifting of pixels is obtained by using the shared
An RGB image is decomposed into three layers (Red, private keys (say, K R ; K R and KB, where the subscripts R; G and
Green, Blue) and transformed into DNA sequence matrix. To B denote each components of an RGB image) from ECC and is
scramble the image, DNA addition is performed on the layers fixed for each RGB component on each axis of the image.
as shown in Fig. 1. Circular shifting on each layers of the image The pixels of the image are interleaved to increase the
(can be seen in Fig. 4) is performed to further distort the variation in the correlation of the image across the layers
correlation of the pixel values by scrambling them indepen- in spatial domain [25].
dently in the spatial domain. This acts as a symmetric key Three different curves and points on these three curves
encryption mechanism. The length of shifting sequence can be are chosen for each component for an RGB image ECDHE.

Fig. 7. Flow chart of the proposed algorithm. (a)Encryption. (b) Decryption.

Please cite this article as: M. Kumar, et al., A new RGB image encryption algorithm based on DNA encoding and elliptic
curve Diffie–Hellman cryptography, Signal Processing (2016), http://dx.doi.org/10.1016/j.sigpro.2016.01.017i
 6 M. Kumar et al. / Signal Processing ∎ (∎∎∎∎) ∎∎∎–∎∎∎

Lookup tables are made available for each curve to speed The decryption process is similar to the encryption process
up the process. by taking the reverse order with deinterleaving of the image
Finally, the RGB image is processed through one more followed decryption of ECHDE. The image is again deinter-
round of interleaving as an added measure to partially leaved and the reordering of the image is done using the same
hide the implementation of ECDHE. octal sequence but with the negative values of the ECHDE
The process of encryption is stated the steps under- keys. The image finally undergoes DNA subtraction followed
neath (can be seen in Fig. 7a). the decoding using same rule giving the decrypted image. The
decryption algorithm is shown in Fig. 7b.
Step 1: Import the RGB image I ðm  n  3Þ, where m and
n are the row and column size, respectively.
Step 2: Each component of the RGB is decomposed into
4. Simulated results
binary numbers ðm  ðn  8ÞÞ and is encoded using
of the DNA encoding scheme to form ðm  ðn  4ÞÞ
The simulations have been performed on MATLAB
by virtue of selecting one of the eight rules.
environment 8.2 and GNU octave. For instance, a baboon
Step 3: The DNA addition operation is performed on the
image of size 512  512 with the following parameters are
components as shown in Fig. 3. RDNA ði; jÞ ¼ RðI; jÞ þ
used for encryption.
Gði; jÞ; GDNA ði; jÞ ¼ Gði; jÞ þ Bði; jÞ and BDNA ði; jÞ ¼
GDNA ði; jÞ þ Bði; jÞ. R layer: Y 2 þ 897XY þ 789Y ¼ X 3 þ 273X 2 þ 321X þ672
1.
Step 4: We generate an octal sequence and each layer A F 216 V: ð19985; 54699Þ with keys Party Apriv ¼ 120,
undergoes circular shift operation using the amount Party Bpriv ¼ 130 and Party Apub ¼ ð25918; 49043Þ and
specified by the ECDHE keys and the direction spe- Party Bpub ¼ ð58446; 19298Þ giving shared private key
cified in the octal sequence as shown in Fig. 4. K R ¼ ð13936; 47874Þ.
Step 5: The RGB components RRRGGGBBB are interleaved
2. G layer: Y 2 þ 399XY þ101Y ¼ X 3 þ487X 2 þ 9280X þ
as RGBRGBRGB as shown in Fig. 5.
11186 A F 216 V: ð33387; 32484Þ with keys Party Apriv ¼
Step 6: The image is encrypted as RDNA  ðX r ; Y r Þ over
899, Party Bpriv ¼ 1798, and Party Apub ¼ ð15516; 55029Þ
ER ða1r ; a2r ; a3r ; a4r ; a6r Þ; GDNA  ðX g ; Y g Þ over EG ða1g ;
and Party Bpub ¼ ð32826; 61971Þ giving shared private
a2g ; a3g ; a4g ; a6g Þ and BDNA  ðX b ; Y b Þ over EB ða1b ;
key K G ¼ ð65272; 38098Þ.
a2b ; a3b ; a4b ; a6b Þ and explained graphically in Fig. 6
3. B layer: Y 2 þ1210XY þ 72Y ¼ X 3 þ928X 2 þ 5249X þ
Step 7: The RGB components are again interleaved.
7892 A F 216 V: ð52477; 53228Þ with keys Party Apriv ¼
Step 8: Finally, RGB components are recombined to get
792, Party Bpriv ¼ 685, and Party Apub ¼ ð400095; 20676Þ
the encrypted image.
and Party Bpub ¼ ð57580; 21332Þ giving shared private
key K B ¼ ð48375; 32041Þ (Fig. 8).

Fig. 8. All shared private keys for elliptic curve Diffie–Hellman encryption: (a) Red component for all pixel values of KR. (b) Green component for all pixel
values of KG. (c) Blue component for all pixel values of KB.

Table 3
Octal shifting sequence.

Layer R G B R G B R G B R G B R G B

Position 1 3 5 7 8 6 4 2 3 5 6 4 5 3 4
Dest. D&R D&L U&L U&R L U R D D&L U&L U R U&L D&L U&R

Please cite this article as: M. Kumar, et al., A new RGB image encryption algorithm based on DNA encoding and elliptic
curve Diffie–Hellman cryptography, Signal Processing (2016), http://dx.doi.org/10.1016/j.sigpro.2016.01.017i
 M. Kumar et al. / Signal Processing ∎ (∎∎∎∎) ∎∎∎–∎∎∎

Fig. 9. Simulations. (a) Original Baboon image. (b) Encrypted Baboon image. (c) Decrypted Baboon image. (d) Original Lena image. (e) Encrypted Lena
image. (f) Decrypted Lena image. (g) Original hexfractal (sparse) image. (h) Eencrypted hexfractal (sparse) image. (i) Decrypted hexfractal (sparse) image.
(j) Original black image. (k) Encrypted black image. (l) Decrypted black image.

Please cite this article as: M. Kumar, et al., A new RGB image encryption algorithm based on DNA encoding and elliptic
curve Diffie–Hellman cryptography, Signal Processing (2016), http://dx.doi.org/10.1016/j.sigpro.2016.01.017i
 M. Kumar et al. / Signal Processing ∎ (∎∎∎∎) ∎∎∎–∎∎∎

Fig. 10. Key space analysis. (a) Decryption with wrong key sequence and wrong shift sequence. (b) Decryption with wrong keys and correct shift sequence.
(c) Decryption with wrong keys and wrong shift sequence. (d) Decryption with wrong key sequence and correct shift sequence. (e) Decryption with correct
keys and wrong shift sequence. (f) Correct decryption.

The octal shifting sequence can be seen in Table 3. (Number of DNA complement rule)  (Number of points
on Elliptic curve)6  (Number of possible shifting operationsn
Fig. 9 shows that the proposed algorithm is good for
encrypting color images. Since the proposed algorithm is  qffiffiffiffiffi 6
asymmetric in nature, and with the inclusion of extra layer of 8  2k þ 1  2 2k  8n r ðKey SpaceÞ r 8
security such as DNA computing and second interleaving  qffiffiffiffiffi 6
increases the robustness of the proposed algorithm. Moreover,  2k þ 1 þ 2 2k  8n :
DNA computing along with a second interleaving hide ECDHE
making it harder to compromise. For fast encryption,
For the purpose of simulation, we have taken the field F216
curve25519 proposed by Bernstein [26] can be used.
and the length of shifting sequence is fifteen (n¼15). Hence,
the key space lies between 2143:9322 to 2144:0674 , which pro-
Security analysis vides high security and robustness against brute force. The
key space can be enhanced with respect to the field size (k)
5.1. Key space analysis and the length of the shifting sequence (n).

To sustain security of images, the key space should be

large enough to resist an exhaustive attacks. The key space 5.2. Key sensitivity analysis
for the proposed algorithm is as follows:
For a viable image encryption algorithm, the decryption
should fail with the slightest change in the key should not
allow correct decryption. Various permutations in the key
space are applied and the results are shown in Fig. 10.

Please cite this article as: M. Kumar, et al., A new RGB image encryption algorithm based on DNA encoding and elliptic
curve Diffie–Hellman cryptography, Signal Processing (2016), http://dx.doi.org/10.1016/j.sigpro.2016.01.017i
 M. Kumar et al. / Signal Processing ∎ (∎∎∎∎) ∎∎∎–∎∎∎ 9

Fig. 11. Histograms analysis. (a) R component of the original image Fig. 9a. (b) G component of the original image Fig. 9a. (c) B component of the original
image Fig. 9a. (d) R component of the encrypted image Fig. 9b. (e) G component of the encrypted image Fig. 9b. (f) B component of the encrypted image
Fig. 9b. (g) R component of the decrypted image Fig. 9c. (h) G component of the decrypted image Fig. 9c. (i) B component of the decrypted image Fig. 9c.

5.3. Statistical attacks 5.4. Correlation of two adjacent pixels

5.3.1. Histogram analysis of encrypted image Correlation between two adjacent pixels can be tested in
Histograms provide a clear illustration of how the pixels four ways, by taking two vertically adjacent pixels or by
intensity is distributed. Figs. 11 and 12 show the histograms taking two horizontally adjacent pixels or by taking two
of the original (Fig. 9a and d), encrypted (Fig. 9b and e) and diagonal adjacent pixels or taking two anti-diagonal adjacent
decrypted (Fig. 9c and f) images. The histogram of the pixels in the encrypted image. We have selected 5000 ran-
encrypted image is rigorously diverse from the original dom pairs of adjacent pixels for testing purpose. The results
images' histogram. Further the encrypted images' histogram are shown in Tables 4–11. Table 12 and the correlation factor
look similar and hence cannot be used for decryption. is calculated for the original Baboon image (Fig. 9a); original

Please cite this article as: M. Kumar, et al., A new RGB image encryption algorithm based on DNA encoding and elliptic
curve Diffie–Hellman cryptography, Signal Processing (2016), http://dx.doi.org/10.1016/j.sigpro.2016.01.017i
 10 M. Kumar et al. / Signal Processing ∎ (∎∎∎∎) ∎∎∎–∎∎∎

Fig. 12. Histograms analysis. (a) R component of the original image Fig. 9d. (b) G component of the original image Fig. 9d. (c) B component of the original
image Fig. 9d. (d) R component of the encrypted image Fig. 9e. (e) G component of the encrypted image Fig. 9e. (f) B component of the encrypted image
Fig. 9e. (g) R component of the decrypted image Fig. 9f. (h) G component of the decrypted image Fig. 9f. (i) B component of the decrypted image Fig. 9f.

Table 4 where x and y are the value of the adjacent pixels and L is the
Correlation coefficients of two adjacent pixels in the plane-image and the number of samples taken Table 13. The scatter plot of the
corresponding cipher-image of Baboon image in horizontal direction.
horizontal correlation is shown in Fig. 13. Further, from the
Component of the Original image Encrypted image comparison Table 15 can be inferred that the proposed
image algorithm offers good security (Fig. 14)

Red component of the 0.928080547623810 0.018695496141379

RGB image 5.5. Mean square error
Green component of 0.862574287003796 0.006687141505165
the RGB image
Blue component of the 0.908792871958637 0.006791010171259 The Mean Square Error (MSE) between the original
RGB image image and encrypted image is calculated by the formula
given below:

1 XN XM  
Lena image (Fig. 9d) and the encrypted Baboon image MSE ¼ jf ði; jÞ  f 0 ði; jÞj2 ;
(Fig. 9b); encrypted Lena image (Fig. 9e) as N  Mn¼1m¼1

covðx; yÞ where f and f 0 are the intensity functions of decrypted

r xy ¼ pffiffiffiffiffiffipffiffiffiffiffiffi; covðx; yÞ ¼ E½ðx  EðxÞÞðy  EðyÞÞ;
Dx Dy and original images and ði; jÞ represents the position of the
pixels. Since the original and the decrypted image are the
1XL
1XL
EðxÞ ¼ xi ; Dx ¼ ðxi  EðxÞÞ2 ; same the MSE of decrypted image with respect to the
Li¼1 Li¼1
original image is zero.

Please cite this article as: M. Kumar, et al., A new RGB image encryption algorithm based on DNA encoding and elliptic
curve Diffie–Hellman cryptography, Signal Processing (2016), http://dx.doi.org/10.1016/j.sigpro.2016.01.017i
 M. Kumar et al. / Signal Processing ∎ (∎∎∎∎) ∎∎∎–∎∎∎ 11

Table 5 Table 9
Correlation coefficients of two adjacent pixels in the plane-image and the Correlation coefficients of two adjacent pixels in the plane-image and the
corresponding cipher-image of Baboon image in vertical direction. corresponding cipher-image of Lena image in vertical direction.

Component of the Original image Encrypted image Component of the Original image Encrypted image
image image

Red component of 0.865075313937753  0.006449278290027 Red component of 0.962408763131362  0.004076324725952

the RGB image the RGB image
Green component of 0.769719865770650 0.016421558279165 Green component of 0.954693507829754 0.005312032028056
the RGB image the RGB image
Blue component of 0.885904418624641 0.001296864065258 Blue component of 0.934300708191352 0.010691355134514
the RGB image the RGB image

Table 6 Table 10
Correlation coefficients of two adjacent pixels in the plane-image and the Correlation coefficients of two adjacent pixels in the plane-image and the
corresponding cipher-image of Baboon image in diagonal direction. corresponding cipher-image of Lena image in diagonal direction.

Component of the Original image Encrypted image Component of the Original image Encrypted image
image image

Red component of 0.853801965078605  0.001305310128727 Red component of 0.907966009398585  0.041028497842243

the RGB image the RGB image
Green component of 0.725686678327742 0.009244980804621 Green component 0.880408776021311  0.008522993738987
the RGB image of the RGB
Blue component of 0.842708189500707 0.017197010308042 image
the RGB image Blue component of 0.863453990135567  0.017458909511577
the RGB image

Table 7
Correlation coefficients of two adjacent pixels in the plane-image and the Table 11
corresponding cipher-image of Baboon image in anti-diagonal direction. Correlation coefficients of two adjacent pixels in the plane-image and the
corresponding cipher-image of Lena image in anti-diagonal direction.
Component of the Original image Encrypted image
image Component of the Original image Encrypted image
image
Red component of 0.853354625495345 0.005877851836138
the RGB image Red component of 0.936121841285154  0.001725957229897
Green component 0.718744447679737  0.020538983270572 the RGB image
of the RGB Green component of 0.914472528539722 0.010691070462096
image the RGB image
Blue component of 0.839496270088290  0.008756002392316 Blue component of 0.879969883231512 0.001156870814213
the RGB image the RGB image

plain-images. The low PSNR values reflect the difficulty in

Table 8
retrieving the plain-image from the cipher-image, without
Correlation coefficients of two adjacent pixels in the plane-image and the
corresponding cipher-image of Lena image in horizontal direction. the knowledge of secret key:
2552
Component of the Original image Encrypted image PSNR ¼ 20  logpffiffiffiffiffiffiffiffiffiffi:
image MSE

Red component of 0.932667029586456 0.003535161409385

the RGB image 5.7. Robustness against known-plaintext attack and chosen-
Green component of 0.922281493233349  0.009706871635486 ciphertext attack
the RGB image
Blue component of 0.893825219391821 0.018571551282381
Image encryption algorithm should be robust against
the RGB image
all types of cryptanalytic. In this subsection, we discuss the
security analysis of the proposed algorithm as known-
5.6. Peak signal to noise ratio plaintext attack and chosen-ciphertext attack. In known-
plaintext attack a cryptanalyst has an access to a plaintext
The Peak Signal Noise Ratio (PSNR) of a given color and the corresponding ciphertext and try to derive a cor-
component is the ratio of the mean square difference of relation between these two or by applying the same key
the component for the two images to the maximum mean try to decrypt ciphertext for the encrypted plaintext. In
square difference that can exist between any two images. chosen-ciphertext attack a cryptanalyst tried to find a
Greater the PSNR value, better the image quality. For matching plaintext for the ciphertext chosen randomly.
encrypted image, smaller value of PSNR is expected. PSNR This proposed algorithm is immune against the known-
values have been calculated for the red, green and blue plaintext attack, and chosen-ciphertext attack for the
components of the cipher-images with respect to their reason that decryption process depends not only on the

Please cite this article as: M. Kumar, et al., A new RGB image encryption algorithm based on DNA encoding and elliptic
curve Diffie–Hellman cryptography, Signal Processing (2016), http://dx.doi.org/10.1016/j.sigpro.2016.01.017i
 12 M. Kumar et al. / Signal Processing ∎ (∎∎∎∎) ∎∎∎–∎∎∎

Table 12
MSE and PSNR of the encrypted image with respect to the original image of Baboon image.

Component of the image MSE PSNR

Red component of the RGB image 4.902866021067581e þ08  38.739700949687872

Green component of the RGB image 4.894815488766505e þ 08  38.732563949902605
Blue component of the RGB image 4.908507314666850e þ 08  38.744695118938409

Table 13
MSE and PSNR of the encrypted image with respect to the original image of Lena image.

Component of the image MSE PSNR

Red component of the RGB image 1.481197079758423e þ09  43.575324862807250

Green component of the RGB image 1.480929875633255e þ 09  43.574541336110620
Blue component of the RGB image 1.469906628458771e þ 09  43.542093874632386

Fig. 13. Horizontal correlation of the baboon image. (a), (b) and (c) are the RGB plots of the original image. (d), (e) and (f) are the RGB plots of the
encrypted image.

combination of the possible correct keys but also on the using the correct set of keys. One image contains cropping
correct permutation of the key sequence. If attacker knows on different layers (R, G, and B) in the image, whereas the
about all the possible exact keys, but unaware about cor- second image contains cropping on sections where there is
rect order of the key sequence, attacker cannot decipher total loss of data. The main objective of the analysis is to
the image correctly. make sure that the decrypted image can be recognized
visually. From Fig. 15, we can see that the proposed algo-
5.8. Cropped attack analysis rithm is robust against cropped attack.

The cropped attack analysis has been performed to 5.9. Transmission noise analysis
check the robustness against data loss for real-time
applications of the proposed algorithm [16,27,28]. Two Transmission loss analysis has been performed to ana-
different types of cropped images have been decrypted lyze the robustness of the algorithm to the noise that

Please cite this article as: M. Kumar, et al., A new RGB image encryption algorithm based on DNA encoding and elliptic
curve Diffie–Hellman cryptography, Signal Processing (2016), http://dx.doi.org/10.1016/j.sigpro.2016.01.017i
 M. Kumar et al. / Signal Processing ∎ (∎∎∎∎) ∎∎∎–∎∎∎ 13

Fig. 14. Horizontal correlation of the Lena image. (a), (b) and (c) are the RGB plots of the original image. (d), (e) and (f) are the RGB plots of the
encrypted image.

Fig. 15. Cropped attack experiment. (a) Encrypted image with different locations from each layer (R, G, and B) is cropped. (b) Decrypted result.
(c) Encrypted image with same locations from each layer (R, G, and B) is cropped. (d) Decrypted result.
 M. Kumar et al. / Signal Processing ∎ (∎∎∎∎) ∎∎∎–∎∎∎

Fig. 16. Transmission noise attack analysis. (a) Noise attack image. (b) Decrypted result.

Table 14
Comparison of authors [7,14,15,25,29–32] with our approach. computational complexity of the proposed algorithm is
◯ðM  N  3Þ and is linear in nature and the encryption
S.no. authors [7,14,15,25,29–32] Our approach
time depends on the size of the image. The running time of
1. [29,15] can only encrypt grey- The proposed algorithm the algorithm can be reduced by performing the encryp-
scale images encrypt RGB image tion on a DNA based computer. As well as implementing
2. [30,7] require DNA based bio- This algorithm is free from
logical experiments biological experiments
single process multiple data scheme as majority of the DNA
3. [14,25] have used symmetric The proposed approach is encoding and ECDH are independent operations.
encryption asymmetric
4. The algorithm [31] increases Original and encrypted
size of the original image images are same size
5. Large key has been used [32] Small keys are used which
6. Comparison
provides huge key space
The proposed approach has been compared with the
existing techniques in Table 14. Further, security analysis
might be picked up during real-time transmission of suggests that the proposed algorithm is efficient, robust
images. The analysis is performed by introducing salt and and can encrypt RGB images as compared to [15,29] which
pepper noise in the encrypted image synthetically. The
can only encrypt greyscale images. Table 15 shows that the
image is then decrypted using the correct set of keys and
proposed technique has superior correlation between
the decrypted image is checked for visual recognition.
pixels, hence provides robust security against statistical
From Fig. 16, it can be established that the proposed
algorithm is invulnerable against noise. attacks [10,17,25,32,33]. As we can infer from Section 5.8
(cropped attack analysis) and Section 5.9 (transmission
noise analysis) that, the proposed algorithm can resist loss
5.10. Computational complexity analysis in data during transmission and can be implemented for
real time transmission as compared to [8–18,20], in these
To evaluate the computational complexity of the pro- algorithms, chaotic maps have been implemented and any
posed algorithm, the time consuming part is the DNA loss of data can limit or prohibit visual recognition of the
encoding operations as a DNA computer has been simu- decrypted image extensively. The proposed algorithm also
lated for the proposed algorithm. Each pixel is decom-
retains the original size of the image after encryption as
posed into DNA nucleotides in ◯ðM  N  3Þ where M and
compared to [31] which increases the size after encryp-
N are the dimensions of a RGB image. Then the pixel
tion. Unlike [14,25] which are symmetric, the proposed
nucleotides now MN34 are added in
◯ðM  N  3  4Þ. Finally the image is reconstructed in algorithm is asymmetric in nature and can be imple-
complexity of ◯ðM  N  3Þ. The shifting and shuffling mented for exchange of sensitive images between anon-
operations are both done in ◯ððM þ XÞ  ðN þYÞ  3Þ as ymous parties. The MSE and PSNR obtained in the pro-
they are linear in nature giving a complexity value of posed algorithm (shown in Tables 16 and 17) are better
◯ðM  N  3Þ. Since a lookup table is created for ECDH than results of [25,33]. Therefore, the proposed algorithm
encryption, the complexity is ◯ðM  N  3Þ. Finally, the is robust than the compared algorithms.

Please cite this article as: M. Kumar, et al., A new RGB image encryption algorithm based on DNA encoding and elliptic
curve Diffie–Hellman cryptography, Signal Processing (2016), http://dx.doi.org/10.1016/j.sigpro.2016.01.017i
 M. Kumar et al. / Signal Processing ∎ (∎∎∎∎) ∎∎∎–∎∎∎ 15

Table 15
Comparison of horizontal correlation of encrypted image of Lena image with other authors [10,17,32,33].

Component of the image Proposed algorithm Zhang et al. [10] Wang et al. [17] Murillo et. al [32] Mishra et al. [33] Kumar et al. [25]

Red component of the RGB 0.003535  0.0065  0.010899 0.0135 0.0219 0.0181
image
Green component of the RGB  0.009706 0.0009  0.018110  0.0835  0.0046  0.0067
image
Blue component of the RGB 0.018571  0.0008  0.006104  0.0170  0.0211 0.0154
image

Table 16 Cho for smooth and fast handling of the manuscript. The
Comparison of MSE value of decrypted image using the proposed algo- authors would also like to thank the anonymous referees
rithm with [25,33].
for their valuable, genuine comments and suggestions to
Component of the Proposed Kumar et al. Mishra et al. [33] improve the quality of this work. One of the authors [PK]
image algorithm [25] would like to thank for the support of the research grant
provided by Department of Science and Technology, New
Red component of 0 1.281718 8:3014X10  26 Delhi, Government of India under INSPIRE grant no. 861/
the RGB image
Green component 0 0.945639 1:0696X10  25
2011.
of the RGB
image
Blue component of 0 1.233145 3:1629X10  25
the RGB image References

[1] N. Koblitz, Elliptic curve cryptosystems, Math Comput 48 (177)

Table 17 (1987) 203–209.
Comparison of PSNR value of decrypted image using the proposed algo- [2] V. Miller, Use of elliptic curves in cryptography, Lect Notes Comput
rithm with [25,33]. Sci 85 (1) (1985) 417–426.
[3] T. Head, G. Rozenberg, R.S. Bladergroen, C.K.D. Breek, P.H.
Component of the Proposed Kumar et al. Mishra et al. M. Lommerse, H.P. Spaink, Computing with DNA by operating on
image algorithm [25] [33] plasmids, Biosystems 57 (2) (2000) 87–93.
[4] Y. Niu, X.Y. Wang, An anonymous key agreement protocol based on
chaotic maps, Commun Nonlinear Sci Numer Simulat 16 (4) (2011)
Red component of 1 47.052878 298.939290
1986–1992.
the RGB image [5] K. Halvorsen, W.P. Wong, Binary DNA nanostructures for data
Green component 1 48.373549 297.838589 encryption, PLOS ONE 7 (9) (2012) e44212.
of the RGB [6] X.D. Zheng, J. Xu, W. Li Parallel, DNA arithmetic operation based on
image n-moduli set, Appl Math Comput 212 (1) (2009) 17–184.
Blue component of 1 47.220662 303.129948 [7] A. Gehani, T.H. LaBean, J.H. Reif, DNA-based cryptography, DIMACS
the RGB image series in discrete mathematics, Theoret Comput Sci 54 (1) (2000)
23–249.
[8] X.Y. Wang, Y.Q. Zhang, X.M. Bao, A novel chaotic image encryption
Conclusion scheme using DNA sequence operations, Optics Lasers Eng 73 (1)
7. (2015) 53–61.
[9] Q. Zhang, L. Liu, X. Wei, Improved algorithm for image encryption
In this paper, we have proposed a new and secure RGB based on DNA encoding and multi-chaotic maps, Int J Electron
Commun 68 (3) (2014) 186–192.
image encryption algorithm using DNA computing and [10] Q. Zhang, X. Wei, RGB color image encryption method based on
ECDHE. An image is encoded using the DNA compliment Lorenz chaotic system and DNA computation, IETE Tech Rev 30 (5)
rule and is scrambled by shifting and interleaving opera- (2013) 404–409.
[11] Q. Zhang, L. Guo, X. Wei, A novel image fusion encryption algorithm
tions, thereafter encrypted using ECDHE. The image finally based on DNA sequence operation and hyper-chaotic system, Optik
undergoes another round of interleaving giving the final 124 (2013) 3596–3600.
encrypted image. The use of DNA encoding and ECDHE [12] Q. Zhang, X. Wei, A novel couple images encryption algorithm based
on DNA subsequence operation and chaotic system, Optik 124 (23)
provide a dual layer of security. The key space of the (2013) 6276–6281.
algorithm is large and can be varied according to need. The [13] P. Zhen, G. Zhao, L. Min, X. Jin, Chaos-based image encryption
decryption is successful only when the keys are applied in scheme combining DNA coding and entropy, Multimed Tools Appl
http://dx.doi.org/10.1007/s11042-015-2573-x.
a particular order hence the knowledge of just the keys are [14] L. Liu, Q. Zhang, X. Wei, A RGB image encryption algorithm based on
insufficient for decryption. The statistical analysis shows DNA encoding and chaos map, Comput Electr Eng 38 (5) (2012)
that the algorithm provides good security and can protect 1240–1248.
[15] H. Liu, X. Wang, A. Kadir, Image encryption using DNA com-
against common attacks. Therefore, this proposed algo-
plementary rule and chaotic maps, Appl Soft Comput 12 (5) (2012)
rithm can be used for secure image transmission. 1457–1466.
[16] X.Y. Wang, L. Yang, R. Liu, A. Kadir, A chaotic image encryption
algorithm based on perceptron model, Nonlinear Dyn 62 (3) (2010)
615–621.
Acknowledgments [17] X.Y. Wang, L. Teng, X. Qin, A novel color image encryption algorithm
based on chaos, Signal Process 92 (4) (2012) 1101–1108.
[18] H.J. Liu, X.Y. Wang, Color image encryption based on one-time keys
All the authors are deeply grateful to the Editor-in- and robust chaotic maps, Comput Math Appl 59 (10) (2010)
Chief Dr. Björn Ottersten and Handling Editor Dr. Nam Ik 3320–3327.

Please cite this article as: M. Kumar, et al., A new RGB image encryption algorithm based on DNA encoding and elliptic
curve Diffie–Hellman cryptography, Signal Processing (2016), http://dx.doi.org/10.1016/j.sigpro.2016.01.017i
 16 M. Kumar et al. / Signal Processing ∎ (∎∎∎∎) ∎∎∎–∎∎∎

[19] Y.Q. Zhang, X.Y. Wang, A symmetric image encryption algorithm [27] M. Kumar, D.C. Mishra, R.K. Sharma, A first approach on an RGB
based on mixed linear-nonlinear coupled map lattice, Inform Sci 273 image encryption, Optic Lasers Eng 52 (2014) 27–34.
(10) (2014) 329–351. [28] Y. Li, F. Zhang, Y. Li, R. Tao, Asymmetric multiple-image encryption
[20] H.J. Liu, X.Y. Wang, Color image encryption using spatial bit-level based on the cascaded fractional Fourier transform, Optics Lasers
permutation and high-dimension chaotic system, Optics Commun Eng 72 (2015) 18–25.
284 (16–17) (2011) 3895–3903. [29] N. Kang, A Pseudo DNA Cryptography Method arXiv:0903.2693 [cs.
[21] H.J. Shiu, K.L. Ng, J.F. Fang, R.C.T. Lee, C.H. Huang, Data hiding CR] (2009).
methods based upon DNA sequences, Inform Sci 180 (11) (2010) [30] C.T. Celland, V. Risca, C. Bancroft, Hiding messages in DNA micro-
2196–2208. dots, Nature 399 (1999) 533–534.
[22] P. Wasiewicz, J.J. Mulawka, W.R. Rudnicki, B. Lesyng. Adding num- [31] J.B. Lima, L.F.G. Novaes, Image encryption based on the fractional
bers with DNA, IEEE International Conference on Systems Man and Fourier transform over finite fields, Signal Process 94 (2014)
Cybernetics 1 (2000) 265–270. 521–530.
[23] J. Hoffstein, J. Pipher, J.H. Silverman, An Introduction to Mathema- [32] M.A. Murillo-Escobar, C. Cruz-Hernanandez, F. Abundiz-Perez, P.
tical Cryptography, Springer Science þ Business Media, New York, M. Lopez-Gutierrez, O.R. Acosta Delcampo, A RGB image encryption
2008, 279–339. algorithm based on total plain image characteristics and chaos,
[24] R. Rivest, A. Shamir, L. Adleman, A Method for Obtaining Digital Signal Process 109 (2015) 119–131.
Signatures and Public-Key Cryptosystems, Commun ACM 21 (2) [33] D.C. Mishra, R.K. Sharma, Manish. Kumar, Kuldeep. Kumar, Security
(1978) 120–126. of color image data designed by public key cryptosystem associated
[25] M. Kumar, P. Powduri, A. Reddy, An RGB image encryption using with 2D-DWT, Fractals 22 (4) . 1450011-1 1450011-16.
diffusion process associated with chaotic map, J Inf Secur Appl 21 (1)
(2015) 20–30.
[26] D.J. Bernstein, New Diffie–Hellman Speed Records Public Key
Cryptography (PKC), 3958, Springer, Berlin, Heidelberg, 2006,
207–228.

Please cite this article as: M. Kumar, et al., A new RGB image encryption algorithm based on DNA encoding and elliptic
curve Diffie–Hellman cryptography, Signal Processing (2016), http://dx.doi.org/10.1016/j.sigpro.2016.01.017i