See discussions, stats, and author profiles for this publication at: https://www.researchgate.

net/publication/259645958

Neural Network-Based Stream Image Encryption

Article · December 2012

CITATION READS
1 1,537

3 authors:

Siddeeq Yousif Ameen Karam M. Z. Othman

Duhok Polytechnic University Northern Technical University
111 PUBLICATIONS 775 CITATIONS 9 PUBLICATIONS 15 CITATIONS

SEE PROFILE SEE PROFILE

Mohammed Al-Jammas
Ninevah University
17 PUBLICATIONS 30 CITATIONS

SEE PROFILE

Some of the authors of this publication are also working on these related projects:

Investigation and Enhancement of 5G channel coding View project

AES Cryptosystem Development Using Neural Networks View project

All content following this page was uploaded by Mohammed Al-Jammas on 19 February 2014.

The user has requested enhancement of the downloaded file.

 Neural Network-Based Stream Image Encryption
Prof. Siddeeq Y. Ameen
Department of Computer and Information Engineering
College of Electronics Engineering
University of Mosul
Mosul, Iraq, [email protected]

Karam M. Z. Othman
Electronic Technical Department
Al Dour Technical Institute,
Foundation of Tech. Education,
Tikrit, Iraq, [email protected]

Dr. Mohammed. H. Al-Jammas

Department of Computer and Information Engineering
Deputy Dean, College of Electronics Engineering,
University of Mosul,
Mosul, Iraq, [email protected]

ABSTRACT services has been enforcing security and

ensuring authorized access to sensitive data. A
This paper proposed an image encryption direct solution to overcome this drawback is to
algorithm that can be called Modified Stream use an encryption algorithm to mask the image
Image Encryption to verify the requirements of data streams. The problem of image protection
secure image transfer. The proposed algorithm is has existed since the information technologies
based on the symmetric key encryption was growing up. Image is a multimedia signal
approach. The system employs Neural Networks providing the most information to a man. Above
(NN) based pseudo random number generator 80% of information is obtained from the vision
with back propagation learning algorithm. This apperceiving. That is the main reason of the
will provide the required nonlinearity properties protecting of an image against the unauthorized
that increase the randomness statistical properties reading. Traditional symmetric encryption
of the pseudo random generator. Computer algorithm such as DES, IDEA or AES is not
simulation results based on Matlab simulation suitable for the image encryption because of
have been carried out to demonstrate the different storage properties of an image. Position
effectiveness of the learning process to get the permutation and modification of the pixels
efficient weights has been presented. The results belongs to basic methods of image encryption.
also show several experimental and statistical These methods remain open for various
analysis to demonstrate the randomness of the encryption algorithms and that is the reason why
pseudo random generator together with the the other algorithms can be used.
proposed image encryption and decryption. The Image encryption techniques involve conversion
results reflect simple and an efficient image of image into one that is difficult to understood
security system. (Lala and Sami, 2009). There are various
systems of image encryption to encrypt and
KEY WORDS decrypt data, and there is no single encryption
algorithm reacts to different kinds of images.
Stream cipher, image encryption, neural Researchers have attempted to use neural
networks, pseudo random generator. networks in cryptography. The nonlinearity
characteristics that involved in the neural
networks revive an interest to use them in image
security. Moreover, The complexity of the
1.0 INTRODUCTION implementation of the neural network can be
reduced through the use of existing hardware
Over the last decade, there has been tremendous technologies (Fathy, 2010).
development in the Internet and multimedia A stream cipher consists of Pseudo-Random
technology which has a large scale use of digital Numbers Generator (PRNG). For the encryption
images as important carriers of information of PRNG is initialized with a key and provides
communication for people. The main obstacle in its output referred to as a sequence of bits than
this widespread deployment of digital image the current pseudo-key. The chance is very
important because it destroys completely the
statistical properties in the message. Current key
generators by using linear feedback shift
registers (LFSR). LFSR method is an attempt to
simulate one-time pad by creating a long line of
important information. As with any such attempt,
when the key is shorter than the message itself,
breaking a part of the cryptogram cryptanalyst
provides information about other parts of the
cipher text. For a LFSR, a known-plaintext
attack, parts of the key sequence (Akhil et al,
2011). The key stream is combined (XOR) with
a bit of text in general, an XOR gate. To decode
the message bits cipher text is XORed with the
bit sequence corresponding again. A fault in a
single encrypted in a single bit error in plain text.
This is very useful if the error rate of
transmission is highly (A. Menezes et al, 1996).
The one-time pad is a stream cipher to be proven
safe. In this figure. Each character key is
randomly generated and used only once. The
cryptogram opponents are no further information
about the text (A. Menezes et al, 1996). Figure 1.Image encryption process.

2.0 The Modified Stream Image In Figure 1, each pixel is encrypted individually.
The RGB (red, green, blue) value of the pixel is
Encryption Technique
read and changed into 24-bit binary number (8-
bits each). The 24 bits image pixel is appended
The paper adopts an image encryption technique
with 8 bits and left shifted 10 times to form 32
proposed by Kaushik and et al (Lala and Sami,
primary ciphered pixel (Xi). The output of this
2009).. The algorithm is called Stream
stage is XORed with 32 bits secondary key (SKi)
Encryption Standard (SES). It is a symmetric
to form the intermediate ciphered pixel. The
stream cipher that encrypts each colored pixel
primary key is XORed with the intermediate
using the principle of stream cipher of random
ciphered pixel (Xi) to form Yi. Removing the
key and special functions as shown in Fig. 1.
initial 8 bits from the last output will give the 24
Since it is symmetric, the same key will be used
bits colored encrypted pixel. The previous
for both encryption and decryption with certain
operations are repeated for all digital image
caution in key transfer from the sending end to
pixels (512 * 348 pixels). The 32 random bits
the receiving end. Algorithm uses same key at
(SKi and PKi) are generated using Neural
both sending and receiving ends to resolve the
Network Pseudo Random Bits (NNPRB)
problem of key distribution. A modified version
generator. Details about the proposed NNPRB
of the SES algorithm is adopted in this paper to
will be given in next section.
further extend its ability to stand against brute
The encrypted pixel can be decrypted in the in
force attacks. This is because the SES encryption
reverse order of the encryption process described
key is altered many times during the encryption
earlier. This will ensure the symmetric
process. The modified version adopted here
encryption process. In the decryption process
keeps this feature and adds more to it in order to
shown in Figure 2, the encrypted 24 bits pixel at
make the problem of guessing the original
the output of the encryption process will be the
image, harder even if an attacker knows one
input for the decryption process. In figure 2, the
decryption key. Furthermore, the SES provides
same 24 bits is used but with the same seed
protection against high redundancy of color
integer to provide the synchronization required
images (Lala and Sami, 2009).This is very
to achieve the same generated random sequence
essential feature and need to be considered in
used in the encryption process primary key. This
any encryption algorithm and have been
step will be repeated 10 times as performed
maintained in the SES. The proposed algorithm
previously in the encryption process. The output
under investigation in this paper is a modified
from the last step is XORed with the 32 bits
version of the SES and that is why it is called
secondary key and right shifted 10 times to give
MSES. The MSES uses the architecture
the output Xi.
proposed by the SES with neural network
approach in key generation and slight
modification in key usage.
 (212) is obtained. Figure. 4. shows the modified
stream Image encryption structure.

Figure 3 The Neural Network based Random

Generator

Figure 2.Image decryption process

This process (from the reading encrypted 24 bits

pixel until the computation of Xi) will be
repeated for the whole encrypted pixels to give
the decrypted image binary file. The color image
will be recovered by inverse generation of image
from the binary file.
3.0 Neural Network Pseudo Random
Bits (NNPRB) generator
The proposed neural pseudo random bits
generator flowchart is illustrated in figure (3).It Figure 4 The Modified Stream Image Encryption
starts by generating a random 12 bits, from
Structure.
which a pattern that can be XORing in a
sequence that will never generate a similar
patterns [Karam , Mohammed H. 2011 ]. Here 4.0 MSES and NNPRB generator
{1, 2, 4, 6, 7, 9, 12} locations are selected among Evaluation
250 possibilities. These patterns possess the
randomness and unpredictable properties. Then, The NNPRB and the MSES have been simulated
an artificial neural network (ANN) that has seven using Matlab-V7. In the evaluation of the
inputs and one output is used to map the XORing NNPRB, two states have been considered, the
stage of the 7 bits input patterns. The target for training state and the key generation states. In the
each input pattern is the XORing output of the training state, efficient weights are computed
previous pattern. The Back propagation is used using back propagation learning algorithm. In
in learning process of such mapping (Howard et the learning state, the back propagation learning
al, 2006), (Jacek, 1992). To enter the 2nd algorithm is simulated using a binary data
pattern, the bits (1,2,4,6,7,9,12) of the previous representation until minimum mismatch cases
pattern are XOR-ing to generate the MSB of the between the output and the target patterns are
2nd NN input with shifting other bits upwards so reached (error =10-4). When the neural network
that: (MSB) old pattern = (MSB-1) new pattern. is learned and the weights are saved, the next
This process is repeated until the required key stage is the key generation process. In this
length which is equal to the number of inputs process, the weights are very important to run the
learned neural network and generate random
binary keys. In this work, three key sequences
are generated for each image layer using the
same process of neural network key generation
but different initial values. The key size of these
keys is (384*4096) for the R, B, G image layer.
In the evaluation of the NNPRB, the statistical
randomness properties of certain selected three
key sequences have been investigated as shown
in Table 1.

Table 1 The Comparisons of the Statistical Test

Results Between the three selected generated
keys Figure 6. The ciphered image

The obtained Values 6000

Test Key1 Key2 Key3

5000

4000

Frequency 0 0 0
3000
Serial -1.9988 -1.9988 -1.9968
Poker 6.3875 4.3245 6.3875 2000

Autocorrelation 0.0939 0.1252 0 1000

0 50 100 150 200 250

The results shown in Table 1, clearly verified the Figure 7. The histogram of R-layer of original
randomness satisfaction of all the generated Image
keys. This is because the generators passed the
randomness test recommend [Andrew et al, 1800

2001]. The next step in the evaluation is the test 1600

of MSES. The three generated keys are used to 1400

encrypt the (R, G and B) layers of the color 1200

image with size (384*512) pixel. The Figure 5 1000

and 6, show the original and encrypted images. It 800

is clear that the MESE removes completely the 600

information present in the original image. For 400

further investigation about the strength of the 200

encryption process the original and encrypted 0
images histograms have been plotted as shown in 0 50 100 150 200 250
Figs. 5-12.
Figure 8. The histogram of R-layer of ciphered
images

7000

6000

5000

4000

3000

2000

1000

0 50 100 150 200 250

Figure 5. The original Image Figure 9. The histogram of G-layer of original

images
 image (Cheddad et al, 2009). These tests involve
1800 the computation of the mean square difference
1600 between the original and the encrypted image
1400
which is given by;
1200

1000

800

600

400

200

0 50 100 150 200 250

Figure 10. The histogram of G-layer of ciphered

images
6000

5000

4000

3000
Figure 13 Samples of intensity values of RGB
2000 original image
1000

0 50 100 150 200 250

Figure 11. The histogram of B-layer of original

images

1800

1600

1400

1200

1000

800

600

400

200

0 50 100 150 200 250

Figure 14 Samples of intensity values of RGB
Figure 12. The histogram of B-layer of ciphered ciphered image
images
1 m n
2
It is clear and obvious from the results presented MSE  [  (x(i, j )  y(i, j)] (1)
mn 1 1
in Figs. 5-12 that the MSES achieve the
objective of information concealing through the where m, and n are the two dimensional image
achievement of uniform distribution of the three size and x(i,j) and y(i,j) are the original and
colors Red, Green and the Blue. The results also encrypted image pixels. The two summations are
show that there are no correlations in the performed for the dimensions "i" and "j."
information in the encrypted colored image. Therefore I(i,j) represents the value of pixel (i,j)
Samples of intensity values of the original and of images x and y. From the computed MSE, the
ciphered images are shown in figures 13 and 14. value of the Peak Signal to Noise Ratio PSNR in
These plots include separate lines for the red, dB can be computed as (Ismet and Ibrahim,
green, and blue intensities. The x-axis of the plot 2005).
represents the distance of the line measured in
pixels. The y-axis represents the intensity values PSNR  20 log(max MSE ) ( 2)
of the R,G and B layers of colored image. As
shown in figure 14 the values of R,G and B have Table 2 shows the computed values of MSE and
a random intensity and the original values have PSNR for all three color of the image.
been hidden.
Extra tests were also carried out (objective Table 2 The MSE and PNR on Peppers Image
measurement) to investigate the difference
between the original image and the encrypted
 Image Color MSE PSNR dB the Engineering Technical College , Mosul, Iraq
R 10051 18.6708 in 2006. Next, he awarded the MSc in Technical
G 11816 17.0531 Computer Enginering from the Engineering
B 12196 16.7369 Technical College, Mosul, Iraq the Engineering
5.0 CONCLUSIONS Technical College, Mosul, Iraq in 2009. From
2010-2011, Mr.. Karam worked as a lecturer at
In this paper a new Stream cipher Encryption the Information Technology Department in Gulf
based on neural network algorithm has been University in Bahrain. From 2011 until now he
proposed which encodes digital image pixel by acts as a lecturer at Electronic Department in
pixel. The pseudo random number generator is Technical Institute - Al Dour , Salah Al-Dean -
based on neural network and used to generate Iraq . Through his academic life he published
random keys. These keys are tested using the over 2 papers in field of computer engineering,
well-known statistical tests(the serial test, and information security.
frequency test, Autocorrelation test and Pocker
test). After the encryption process of the digital REFERENCES
image, a histogram of the ( R ,G and B) layers
of original and ciphered image are presented in A. Menezes, P. van Oorschot, and S.
which one can notice that there is no correlation Vanstone 1996, Handbook of Applied
in the information in the encrypted colors Cryptography, CRC Press,Inc.
histograms.
Akhil Kaushik, Satvika Khanna, Manoj
Barnela and Anant KumarJ. 2011, 'Stream
AUTHORS PROFILE Encryption Standard for Digital Images',
International Journal of Computer and
Siddeeq Y. Ameen (M’05) born in 1960 in Electrical Engineering, vol. 3, no. 2, pp.
Mousol, Iraq. He awarded BSc in Electrical and 240-243.
Electronic Engineering from the University of
Technology, Baghadad, Iraq in 1983. Next, he Karam , Mohammed H."Implementation of
awarded the MSc and PhD from Loughborough neural - cryptographic System using FPGA"
University, UK. in 1986 and 1990, respectively Journal of Engineering Science and
in the field of digital communication systems Technology ,vol. 6, no. 4 pp 411 - 428 .
and data communication. From 1990-2006,
Professor Siddeeq worked with the University of Cheddad, A., Condell, J., Curran, K and Mc
Technology in Baghdad with participation in Kevitt Cheddad 2009., 'A Hash-based
most of Baghdad’s universities. From 2006- Image Encryption Algorithm', School of
2011, he acts as a dean of college of Engineering Computing and Intelligent Systems,.
at the Gulf University in Bahrain. Now he is a a
Fathy El-Ashry 2010, Digital Image
professor of data communication and networking
Encryption, Thesis, Menofia University.
at the university of Mosul, Iraq. Through his
academic life he published over 82 papers in Howard Demuth ,Mark Beale and Martin
field of data communication and information Hagan 2006, Neural Network Toolbox 5
security. for use with MATLAB. Version 5 , Math
Mohammed H. Al-Jammas (Jun’02) born in Works Inc,.
1966 in Mousol, Iraq. He awarded BSc in
Electronic and Communication Engineering Ismet Öztürk1 and Ibrahim Sogukpınar
from the University of Mosul, Mosul, Iraq in 2005, 'Analysis and Comparison of Image
1988. Next, he awarded the MSc in Encryption Algorithm' World Academy of
Communication from the University of Mosul, Science, Engineering and Technology 3.
Mosul, Iraq in 1994, and PhD in Computer Jacek M. Zurada 1992, Introduction to
Engineering from the University of Technology, Artificial Neural Systems, West publishing
Baghadad, Iraq in 2007. From 2002-2006, Dr. Company.
Mohammed worked with the University of
Technology in Baghdad. From 2007, he acts as a Lala Krikor, Sami Baba 2009, 'Image
Assistance dean of college of Electronics Encryption Using DCT and Stream Cipher',
Engineering at the University of Mosul.. European Journal of Scientific Research
Through his academic life he published over 3 vol 32 no.1, pp.47-57.
papers in field of computer engineering, and Andrew Rukhin, Juan soto and James
information security. Nechvatal “A Statistical Test Suite for
Karam M. Z Othman (September’07) born in Random and Pseudorandom Number
1984 in Baghdad, Iraq. He awarded BSc in Generators for Cryptographic Applications
Technical Computer Engineering from “, NIST special publication 2001,
http://csrc.nist.gov/rng .

View publication stats