(IJACSA) International Journal of Advanced Computer Science and Applications,

Vol. 7, No. 10, 2016

Encryption Algorithms for Color Images: A Brief

Review of Recent Trends
Anuja P Parameshwaran Wen-Zhan Song
Department of Computer Science, Department of Computer Science,
Georgia State University (GSU) Georgia State University (GSU),
Atlanta, Georgia, USA Atlanta, Georgia, USA

Abstract—The recent years have witnessed rapid

developments in the field of image encryption algorithms for
secure color image processing. Image encryption algorithms have
been classified in different ways in the past. This paper reviews
the different image encryption algorithms developed during the
period 2007-2015, highlighting their contrasting features. At the
same time a broad classification of the said algorithms into: (1)
full encryption algorithm and (2) partial encryption algorithm,
each further sub-classified with respect to their domain
orientations (spatial, frequency and hybrid domains) have been
attempted. Efforts have also been made to cover different
algorithms useful for color images of various color spaces like
Red-Green-Blue (RGB), Hue-Saturation-Intensity (HSI), Cyan-
Magenta-Yellow (CMY) etc. Chaotic cryptosystems, various
transforms like wavelets, affine transforms etc. and visual
cryptography systems are being discussed in detail. Fig. 1. Each pixel in the gray scale image matrix can take any of the 256
values between [0,255]
Keywords—RGB; HIS; CMY; Chaotic cryptosystem; wavelets;
affine transforms and visual cryptography A given Image data will differ from the corresponding text
data in several ways. For instance, an image data is inherently
I. INTRODUCTION
redundant with the pixels highly correlated (This is generally
A. Background/Preliminary: derived from the fact that an image has smooth texture). Some
The field of encryption, which deals with information cryptosystems like the chaotic cryptosystem [1-2] make use of
security, is an active research domain in modern times. this inherent property of images for encrypting them.
Irrespective of the kind of data (multimedia or plain text data) Moreover, traditional cryptosystems used for text is not used
dealt with, security is very crucial, attracting lots of in case of images for the following two reasons: Firstly, the
researchers into this important domain. World-over, size of the image data and normal text data varies greatly, with
researchers are worried on security issues during: (1) the former being many times bulkier, going up to several
transmission of data across a secure channel and (2) the Gigabytes in size. Due to the size variation between the two
storage of data, so that it is not attacked or retrieved by kinds of data at hand, it is obvious that traditional
unauthorized parties. Encryption techniques in general help cryptosystems would take a lot more time encrypting image
manage these two aspects related to security of any data. data than normal text data, which is not a very wise choice to
go with. A second constraint while dealing with the text data
The present review focuses on enlisting and exploring arises out of the requirement that the decrypted data must be
various encryption techniques related to image data. Image is same as the original data (in terms of size and content). In
an array or a matrix representation of square picture elements contrast, small distortions in the decrypted image data are
(pixels) that is systematically arranged in rows and columns. acceptable as it always depends on human perception which is
For example, a given 2D image I can be represented as MxN different for different people. Moreover, lossy compression
matrix where M and N represent the number of rows and can be applied to image data before encryption, though this
columns respectively. Each pixel represented in the image would lead to slight loss of redundant data in the decrypted
matrix has an intensity value. In case of gray scale images the image [3].
pixel can take intensity value between the ranges [0,255].
Figure 1 shows an instance of a gray scale image matrix with In image encryption, it is important to verify the following
its intensity value representation. three parameters of transmitted digital images: integrity,
confidentiality and authenticity [4]. Some of the applications
in which image encryption algorithms come really handy are:
internet communication, multimedia systems, medical
imaging, telemedicine, military communication etc. [3].

1|P age
www.ijacsa.thesai.org
 (IJACSA) International Journal of Advanced Computer Science and Applications,
Vol. 7, No. 10, 2016

Figure 2 shows an illustration of the encryption/decryption

process of image data.

Fig. 3. Image of Lena as a binary, gray-scale and RGB color image

(courtesy: ref [9])

RGB color images consist of three separate components of

8 bits each. Thus RGB has a color depth of 24 bits in total
compared to the 8 bits of gray scaled images. There are
Fig. 2. Encryption and decryption process illustration of a single image various other color space representations of images which are
matrix
more effective than the RGB representation of images. For
Image compression is another interesting area of research instance: YIQ color space (luminance (Y), Chrominance1 (I)
that goes hand in hand with image encryption. Many and Chrominance2 (Q)), HSI color space (Hue (H), Saturation
encryption algorithms like to compress the initial image first (S) and Intensity (I)) etc. [9]. Recent experimental results of
before actually encrypting it so as to get rid of any redundant correlation indicate that YIQ color space algorithm is better
data in the beginning itself [5]. However, some researchers than RGB algorithm [9].
still prefer to encrypt data first and then compress it before C. Full Encryption Algorithms vs. Partial Encryption
transmitting [5] [6] [7] [8]. Compression before encryption Algorithms:
works better for data that is highly redundant in nature [5] [6].
The image encryption algorithms can be broadly divided
In addition to removing redundant data, compression also
into two categories: full encryption algorithms and partial
reduces the bandwidth requirement for transmission of the
encryption algorithms (selective encryption algorithms). These
encrypted image along secure channels. However, it may be
algorithms will be discussed with respect to their domain
added that not all encryption algorithms make use of
orientation, i.e., whether they are spatial domain specific,
compression. Most of the algorithms which are frequency
frequency domain specific or a mix of both (hybrid domain
domain- oriented and those that involve usage of transforms
specific) as shown in Figure 4 [4].
like DCT, affine etc. have compression either of lossy or
lossless form, inherently added to them.
B. Types of Images:
Images can be divided broadly into two categories: (1)
Gray-scale images and (2) Color-images. A simple definition
of gray scale images would be an image which has only
shades of gray and is devoid of any other color. In simpler
words gray scale images contain no color information and is
mostly referred to as one-color images [29]. The difference
between gray-scale images and colored images is that for gray
scale images, only less information is required to represent a
given pixel of the image. A given pixel of a gray scaled image
can take up any of the 256 values ranging from [0-255], which
is basically called the intensity of the pixel [9]. Fig. 4. Broad classification of Image encryption algorithms into full and
partial encryption algorithms
Color images can be picturized as a three-band
monochrome image data, where each band of the image Full encryption algorithms, as the name itself suggest,
corresponds to different color [29]. For example: RGB images deals with the image as a whole and encrypts the whole image.
corresponds to red, green and blue bands if separated into their Partial encryption algorithms encrypts only a part of the image
individual components. Figure-3 shows the famous picture of rather than encrypting the whole image. Region of Interest
Lena as a binary image (consists of black and white color (ROI) is used in selecting the part of image that needs to be
only), gray scale image (consists of shades of gray ranging encrypted in partial encryption algorithm [20]. Visual
from [0-255] i.e. range from black-shades of gray-white) and cryptography (VC) algorithms also fall into the category of
RGB colored image (consists of 3 individual components R, G partial encryption algorithms. The application areas of these
and B) [9]. two classes of encryption algorithms vary greatly. For the first
comparison, we look at the computation requirements of both
the algorithms. It is seen that in partial encryption algorithms,
the computational requirements are greatly reduced as only the
lowest portion of data is encrypted. In fact, the full encryption

2|P age
www.ijacsa.thesai.org
 (IJACSA) International Journal of Advanced Computer Science and Applications,
Vol. 7, No. 10, 2016

algorithms have greater computational complexity than partial Correlation of adjacent pixels of an encrypted image has to
encryption algorithms. For the second comparison, the overall be reduced. But many encryption algorithms on color images
time required by both sections of the algorithms is looked at. existing today are just an extension of the original encryption
Time can be further divided into two categories, (i) the algorithm on a gray scale image. The correlation analysis does
encryption-decryption time (EDT) and (ii) the actual not work out well for such group of algorithms as they apply
transmission time of the encrypted data over a secure channel. the encryption schemes on the R, G and B components
EDT for full encryption algorithms are bound to be larger as separately. While doing so, such class of algorithms
compared to partial encryption algorithms as the latter focuses conveniently neglect the correlations of the R, G and B
on encrypting-decrypting only a small region of the image [4]. components together. Most of the papers discussed here [1]
Figure 5 a) and b) shows the flow and stages of full encryption [2-15] are not an extension of the gray scale image encryption
algorithms and partial encryption algorithms. scheme, with the exception of [2]. The comparison criteria of
most of the papers [1-9] are based on the following:
1) Number of Pixel Change Rate (NPCR)–This depicts
that the more a particular cryptosystem is sensitive to
changing of the original inputted data, the more effective that
cryptosystem is to resist a statistical attack from an intruder.
The higher the value of NPCR (close to 100%), better is the
cryptosystem able to resist statistical attacks.
∑

Where, W and H represent the height and width of the

image.
2) Correlation coefficient–the correlation coefficient
between the pixels of the encrypted image should be low i.e.
close to zero as it will help resist statistical attacks. The
correlation coefficient between each pair of adjacent pixels of
an image can be calculated as:

∑
Fig. 5. a) Full encryption algorithm b) partial encryption algorithm
Where,
D. Evaluation and Comparison Criteria of Image Encryption
Algorithms:
Most of the papers discussed under the category of full ∑
encryption based algorithms under spatial domain category [1-
4] are evaluated based on the following four parameters:
∑
1) Security - Security as an evaluation parameter indicates
the confidentiality and robustness of the encryption scheme
against various attacks like statistical and differential attacks { }
√ √
[9].
2) Speed–A factor used to differentiate partial and full 3) Unified Average Changing Intensity (UACI) – UACI
encryption algorithms. Less data to encrypt implies usage of values are given so as greater is the possibility that the
less CPU time which further implies faster encryption and cryptosystem can avert a differential attack from an intruder.
decryption. This is the advantage of partial encryption
algorithms over full encryption algorithms [9]. ∑ { { {
3) Correlation – Main aim of encryption is to destroy the
correlation between adjacent pixels of the encrypted image
(which is it will be nearly zero). Image data inherently Here, W and H represent the width and height of the
processes the feature of highly correlated pixels (correlation inputted image and C (R, G, B) and C‘ (R, G, B) are encrypted
value close to one), encryption aims to destroy that [1-4] [9]. images before and after a pixel in the original inputted image
4) Key space analysis – The key space of an encryption has been changed.
space should be large enough so as to avert any brute force or 4) Entropy–This parameter deals with the idea of self-
exhaustive attacks from the intruder. Moreover the encryption information. It indicates how much of information has been
scheme must also be sensitive to the key [9]. lost in the decrypted image. The ideal value discussed in here

3|P age
www.ijacsa.thesai.org
 (IJACSA) International Journal of Advanced Computer Science and Applications,
Vol. 7, No. 10, 2016

for [1-8] is 8. If an encryption scheme has a value closer to 8 Rhouma et al [10] have made use of a piecewise linear
means it has lost very little to negligible amount of chaotic map to build their chaotic cryptosystem. The main
information. If there are M messages as m is a symbol then encryption procedure had three sub procedures which were
entropy H(m) can be defined as: interleaved with each other. While dealing with color images
(in RGB color space) the first step of the encryption algorithm
was to convert the image into its three individual vector
∑
components by scanning the original image matrix from left to
right and top to bottom. The second step involved the division
II. FULL ENCRYPTION ALGORITHMS of the phase space of the skew tent map used to build the
cryptosystem into 256 equal width intervals. A mapping
Full encryption algorithms are discussed in [10-17]. These function was also defined to help map the respective values.
papers are divided into spatial domain [10-13], frequency Finally, each of the color pixels belonging to R, G or B were
domain [14-15] or hybrid domain [16-17] based algorithms. individually encrypted by going through certain number of
[10], [11] and [13] are encryption algorithms that are based on predetermined rounds. The results of [10] showed that the key
chaotic cryptosystems. Chaotic cryptosystem is based on the space (key space = 1093) was not exhaustive enough (as in the
mathematical theory of ‗Chaos‘. In theory, chaotic case with most chaotic based cryptosystems) and it was
cryptosystems are those dynamic systems that are highly vulnerable to brute force attacks. In order to prove that the
sensitive to system parameters and initial conditions. If the encryption algorithm suggested by the authors was a secure
system parameters change by chance then the decrypted image one, they proved that the values of NPCR and UACI were
at the receiving end would be totally different from the high enough to avert any differential attacks. Information loss
original inputted image [10]. Chaotic cryptographic systems incurred by [10] was very small, almost negligible, as seen
exploit the inherent feature of bulk data capacity and high data from entropy factor calculations (entropy = 7.9551, as against
redundancy of an image [30]. The reason chaotic the ideal value of 8).
cryptosystems is most useful is because the encryption
algorithm destroys any original pattern existing in the Ahmad and Alam et al [11] also talked about encryption
reconstructed image, thus making it difficult for an intruder to and decryption of images by using a chaotic cryptosystem.
reconstruct the image based on visual perception of the The main difference between [10] and [11] is that in [11], the
graphical information. Chaotic cryptosystems are tied to two authors made use of three different chaotic maps namely, 2D
properties of good ciphers i.e. confusion and diffusion [18]. cat map, 2D coupled logistics map and 1D logistics map. The
Diffusion is based on the dependency of the output bits on the original image was initially broken down to 8x8 sized blocks
input bits whereas confusion is made possible by permuting and were then shuffled by using the 2D cat map. The control
the data sequence thus guaranteeing the relationship between parameters for the shuffling process were generated randomly
the key and cipher text to be as complex as possible [18]. The as dictated by the 2D coupled logistics map and lastly, the
secret key for chaotic cryptosystems described in [10] [11] shuffled blocks of the image was encrypted in accordance to
[13] [16] [17] are the initial conditions and the system the chaotic sequence that was generated by the 1D logistics
parameters defined. map. Unlike the work described in [10], [11] surprisingly has
a huge key space (about 10112) which was capable of
The first step for designing a chaotic cryptosystem is preventing any sort of statistical or differential attack. The
selection of a chaotic map. The dynamics of the chaotic map is information loss as measured by calculating entropy (7.9992,
determined by control parameters [18]. A particular chaotic ideal entropy value = 8) was low here too and the coefficient
cryptosystem will choose the number of chaotic maps it needs. correlation was obtained to be very low as desired (that is,
A single chaotic map has only a small key space making it 0.0095, ideally if low correlation that has to be closer to 0).
easy for intruders to perform brute force attacks to break the The work in [10] did not really calculate the coefficient
system. Thus to avoid such a situation, multidimensional correlation value, so was hard to tell whether the correlation of
couple maps can be used together so as to improve the adjacent pixels of the encrypted image was low or high.
security. A chaotic map is generally used to produce a chaotic
sequence after certain number of hops which is very important Chandel et al [12] made use of the color images found in
as this sequence controls the entire encryption process. the wang dataset. The security in [12] was increased because
Though the chaotic cryptosystems seem perfect in theory, in of two operations performed one after another: splitting
practice they also have a few limitations. First of all, the procedure followed by the encryption procedure. Both the
performance of a chaotic cryptosystem is relatively slow when procedures have their own respective keys. The encryption
compared to other cryptosystems (traditional cryptosystems). algorithm used here is not really new but simple RSA
In addition, the fact that the chaotic cryptosystem is highly encryption algorithm, which is a public key cryptography
sensitive to system parameters and initial conditions of the method. Initially, the color image from the database was split
system sounds great in theory. But in practice, it is difficult to into many portions by making use of a splitting algorithm.
get the correct decrypted image if the processors at the senders RSA encryption algorithm was then used to encrypt each of
and receivers end are different! Because of these limitations, the split pieces. On the receiving end, the inverse of RSA was
chaotic cryptosystems are very much limited in real world successfully applied to decrypt each of the split pieces and
applications [18] [19]. Moreover most of the chaotic finally all the split pieces were merged together to form the
cryptosystems are already dysfunctional [31]. final decrypted image. The keys and number of split pieces
were predetermined by the authorized sender in this

4|P age
www.ijacsa.thesai.org
 (IJACSA) International Journal of Advanced Computer Science and Applications,
Vol. 7, No. 10, 2016

encryption methodology. This work made use of histograms the methodology is valid, secure and robust in nature. Figure 6
only to measure the color bins of both the original and the shows the encryption scheme as discussed in [14].
encrypted images. The encryption scheme in [12] did not have
an exhaustive analysis of the key space nor did it calculate the
coefficient correlation between adjacent pixels of the
encrypted image. The only criteria calculated in [12] was the
entropy and it consequently proved that the information loss
was almost negligible for each of the color images considered
from the wang dataset. Since no other analysis was shown by
the authors in [12] against any attacks it is difficult to say how
effective this encryption methodology can really be. But on a
positive note, the encryption scheme in [12] can be plainly
noted for its simplicity.
Just like works in [10] and [11], Wang et al [13] was also
based on chaotic cryptosystems. This paper made use of a
single chaotic map that is the logistics map. The encryption
algorithm was very similar to that discussed in [10]. Initially a
RGB-color image was broken down to its independent R, G Fig. 6. Encryption algorithm as demonstrated in [14]
and B components. Then a permutation algorithm was
performed on each of the independent component matrices. Samson et al [15] bought in compression before the
The permutation algorithm facilitated combined row and encryption process. The compression achieved by the use of
combined column scrambling which helped the R, G and B wavelets in [15] was however lossless in nature. In order to
pixels to be mutually permuted. Lastly, a diffusion process achieve additional compression, [15] made use of lossless
was applied to the R, G and B components of the image to predictive coding alongside wavelets. The encryption scheme
give the resultant cipher image. The encryption process in [15] flowed as following: firstly, the input color image was
discussed when applied in the reverse order gives back the compressed using a wavelet of sender‘s choice, the level of
original color image. This algorithm was effective against decomposition was also set according to the sender‘s choice.
exhaustive attacks as it had a huge key space which could Then additional compression was achieved using lossless
reach up to 1056. Histogram analysis was also performed predictive coding. The second step was the encryption process
which showed that after encryption the R, G and B‘s which was achieved using the regular secure advanced hill
components of the cipher image was fairly uniform proving it cipher. This in turn involved a pair of involutory matrices (that
hard for an intruder to decrypt. Thus the methodology is the inverse of a matrix A is equal to the matrix A itself), a
described in [13] was safe against statistical attacks. The mix() function and an XOR operator. The decryption process
correlation coefficient of the cipher image was also very involved firstly the decryption algorithm followed by the
small. NPCR and UACI values calculated by [13] were at a decompression algorithms in order. The wavelet used in [15]
higher range which proved that it was secure against any was the simple haar wavelet. One of the drawbacks of the
differential attacks. The authors of [13] also proved that their methodology in [15] was that there was no analysis of the key
methodology was safe against cipher text only attacks, known space or the nature of attacks possible described in it. The only
plaintext attacks, chosen plaintext and chosen cipher text form of validity [15] has is that the decrypted image obtained
attacks. was the same as the original inputted image. Figure 7 depicts
the flow of stages in [15].
With Chen et al [14] deals with a new domain, viz.,
frequency domain. This paper made use of two transforms:
affine transforms and the gyrator transforms. The affine
transform was used twice in the encryption process. The
parameters for these transforms served as the secret key in
[14]. Initially the RGB image was broken down into its 3
independent components. A function of the affine transform
was then used to mix these R, G and B components. Basically,
the components were converted to a complex function (real
and imaginary parts) with the help of the affine transform.
The scrambled pieces of the images was then combined using
the gyrator transform. In other words the gyrator function
helped to encode and transform the complex function obtained
in the first step of using the affine transform. And lastly, the
encrypted image came into being by using the final function of
the affine transform. The reason the affine transform was
performed twice for this encryption scheme was in order to
enhance the security of the algorithm. The work in [14] has
also demonstrated some numerical simulations to prove that

5|P age
www.ijacsa.thesai.org
 (IJACSA) International Journal of Advanced Computer Science and Applications,
Vol. 7, No. 10, 2016

image from any other color space to the HSI color space. New
phase plates were generated in this encryption scheme using
the fractional fourier transform. The S component obtained a
new random phase with the help of random phase encoding
which was based on fractional fourier transform. The I
component too was transformed into two new phase plates
with the help of double random phase encoding which was
again based on fractional fourier transform and it made use of
the H component and the new random phase component. The
final step to the encryption scheme in [17] was chaos
scrambling which essentially encrypted the image. The
difference between the cipher/encrypted image in [17]
compared to all the other cipher/encrypted images in [10-16]
was that it was non-linear in nature and was a combination of
gray image and a phase matrix. Numerical simulations had
been performed which enabled to illustrate the effectiveness
and the level of security obtained by the proposed encryption
scheme in [17]. Table-1 summarizes the full encryption
algorithms discussed in this section.

TABLE I. SUMMARY OF FULL ENCRYPTION ALGORITHMS. THE WORK

DISCUSSED IN [14] & [15] BANKS ITS VALIDITY ON YHE FACT YHAT YHE
Fig. 7. Flowchart of the encryption scheme followed in [15] FINAL DECRYPTED IMAGE SHOULD BE YHE SAME AS YHE ORIGINAL INPUTTED
IMAGE (THIS IS WRT YHE VISUAL PERCEPTION ALONE)
Yu et al [16] and Zhou et al [17] discussed their
methodology in the hybrid domain. Yu et al [16] made use of Numb Color Comparis
both: the chaotic cryptosystem and the wavelet transform to er of space on
Ref Catego Methods chaoti of criteria
form its own encryption scheme. This paper made use of the No ry
Domain
used c inputt used wrt
wavelet decomposition and reconstruction processes twice in maps ed security
its encryption scheme. Initially the original color image used image achieved
underwent a 1-level wavelet decomposition. Only the low
frequency coefficients were considered of importance at this Full
Purely
encrypti RGB -NPCR
stage and they were encrypted using a chaotic based [10] on
Spatial chaotic
1 color -UACI
encryption algorithm. There were four chaotic maps used for domain cryptosyste
algorith space -Entropy
this purpose alone namely: Logistic maps, Chebyshev map, m
ms
2D Arnold map and 3D Arnold map. The next step in -Larger
encryption involved that these encrypted low frequency key space
Full
coefficients act as a key stream and be XORed with the Purely -Entropy
encrypti RGB
Spatial chaotic -
unencrypted high frequency coefficients. This step rendered [11] on
domain cryptosyste
3 color
Coefficien
all the image information held by the high frequency algorith space
m t
coefficients as hidden. The third step in the encryption process ms
correlatio
was the wavelet reconstruction which rendered all the n
encrypted low frequency portions to be equally distributed Full Splitting
among the whole image. In order to have effective encrypti procedure + RGB
Spatial
[12] on RSA - color -Entropy
confidentiality in the encryption scheme of [16], the authors algorith
domain
encryption space
decided to make use of chaotic scrambling for the ms algorithm
reconstructed image. Arnold scrambling was used in order to RGB
produce a smooth image as the final decrypted image. The color
wavelet decomposition and reconstruction processes were space
Full (Made
performed again (level-2 wavelet decomposition). The key encrypti
Purely
use of
-Huge key
space was large enough to resist brute force attacks (2128). The [13] on
Spatial chaotic
1 color
space
encryption algorithm proved efficient enough against domain cryptosyste -NPCR
algorith images
m -UACI
statistical and exhaustive attacks. Overall, methodology ms from
discussed in [16] only gave us reasonable performance, but the wang
encryption time taken by the encryption process was dataset
)
determined to be 0.266 seconds (which is still long). Affine -Security,
Full
The works from [10], [11], [12], [13], [14], [15] and [16] transforms validity
encrypti Frequen RGB
[14 (used and
dealt with image from the RGB color space. Zhou et al [17] on cy
twice) +
- color
robustness
] algorith domain space
instead dealt with the image from HSI color space only. The Gyrator of
ms
first step in the encryption scheme of [17] was converting the transforms methodol

6|P age
www.ijacsa.thesai.org
 (IJACSA) International Journal of Advanced Computer Science and Applications,
Vol. 7, No. 10, 2016

ogy The methodology of how a VC algorithm work is as

illustrated following: a secret picture needs to be shared among n
through
numerical
participants. So in order to divide this secret image among the
simulation n participants, a splitting algorithm is very important. The
s secret picture is divided into n transparencies/shares based on
Full
Wavelet the splitting algorithm in such a way that if any m shares (as
transforms determined by the sender) are placed together then the original
encrypti Frequen RGB
[15 used for image become visible to the authorised receiver. But, if fewer
on cy - color -
] compressio
algorith domain
n prior to
space than m shares are placed together than the original image is
ms not revealed to the authorised receiver. The superimposition of
encryption
Hybrid the pixels of the various shares is achieved by using a simple
Chaotic
domain
cryptosyste
logical OR operator. Thus the computational complexity of
(include the VC class of algorithms is not much [21]. The only
Full m+
s both disadvantage of the VC class of algorithms is that the quality
encrypti wavelet RGB -Huge key
[16 spatial
on
+
transform 4 color space of reconstructed image could be very poor for certain classes
] algorith [partially space
frequen of inputted images.
ms chaotic
cy
cryptosyste Wong et al [20] proposed a multi-level ROI image
domains
ms] encryption universal architecture which dealt with biometric
)
-Security, data. It made use of the multi-level encryption in addition with
Hybrid Chaotic validity the stream cipher RC4 for encryption purposes. The
domain cryptosyste and encryption scheme in [20] made use of multiple ROIs to select
(include m+ robustness
Full
s both Fractional of
the regions of interest for specific users (the authorized users
encrypti HSI in the receiving end). This implied that a particular user could
[17 spatial Fourier methodol
on 1 color only see a part of the decrypted image file which it was
] + Transform ogy
algorith space
ms
frequen [partially illustrated intended to see and not the other parts of the file which it had
cy chaotic through no authorized access to. Multiple level ROIs selected for each
domains cryptosyste numerical
) ms] simulation
image and they were encrypted using three levels of authority
s using RC4 and biometric fingerprinting matching algorithms.
If the sender had to send the image to two users at the
III. PARTIAL ENCRYPTION ALGORITHMS receiving end, it first requested the authenticated server (AS)
for encryption keys for the receivers. The AS generated the
Partial encryption algorithms are also known as selective
keys and send them to the original sender. Using all the keys
encrypted algorithms. The fundamental rule of these
obtained from AS, the sender then performed multi-level ROI
algorithms is that there must be independence of the encrypted
encryption on the main image (the encrypted image is the
regions from the unencrypted regions of the image. Since they
same but the receivers can only decrypt that part that is sent
deal with encrypting only a fraction of the whole image, the
for it and not the whole image) to form an encrypted image.
computation requirements are greatly lessened by this class of
This encrypted image was then send to the two receivers by
algorithms. Another important point to take into consideration
the sender. The users at the receiving end send their biometric
is that this class of algorithms play a very important role in
information to the AS and requested their specific keys from
real-time applications. These algorithms play an important
the AS. AS who kept a copy of the keys and the level of
role specifically in medical applications. They generally help
authority verifies the same and send the keys to the receivers
separate out information into sensitive and insensitive data
if the biometric information matched the template already
only based on perception. [20-26] are papers that fall into the
stored in the AS biometric database. Once the receivers
category of partial encryption algorithms. [20-26] are further
received their respective keys from the AS, they can decrypt
are divided into spatial domain [20-24] and frequency domain
the image file and read out the portion of the image file meant
[25-26] based algorithms.
for them. The encryption scheme implemented by the authors
Along with ROI based algorithms VC algorithms also involved implementing two ROIs and also made use of two
forms a part of the partial encryption algorithms class. Visual levels of authority. The work described in [20] made it less
cryptography is a class of encryption algorithms that does not susceptible to any kind of thefts as it made use of biometric
make use of any key. It makes use of a technique where a authentication. The encryption scheme in [20] was thus
secret image is hidden inside an image into multiple layers. So proved to be one of the most secure encryption schemes
each layer of the image essentially holds some secret discussed in this review paper.
information. At the receivers end all one has to do is align the
Abdulla and Sozan [21] worked on the subtractive color
layers and the secret information in a proper way and the
model CMY (Cyan (C), Magenta (M) and Yellow (Y)). There
original image is revealed to the receiver plainly by human
was no key involved in the encryption scheme of [21] and it
perception. The advantage of VC algorithms is that no
was purely based on visual cryptography. The original image
complex computation is required to be performed in order to
was decomposed to its three constituent components based on
get the decrypted image. And since VC algorithms do not use
the CMY color model. Now, if the C primitive matrix was
keys, there is no key management needed for these class of
considered then every pixel Pi,j in the matrix had 1/4th of its
algorithms [21].

7|P age
www.ijacsa.thesai.org
 (IJACSA) International Journal of Advanced Computer Science and Applications,
Vol. 7, No. 10, 2016

value (1/4 Pi,j) stored in another matrix which represented a

new image. Since the original image was decomposed to its C,
M and Y components, which resulted in three cover images
existing. The encryption scheme in [21] was very simple and
it was based on superimposing (made use of simple OR
logical operator) the secret image (size was equal to or smaller
than that of the cover image) onto the cover image, thus
encrypting the cover image. Three shares were generated such
that each share contained a part of the secret image. In this
paper, the number of pixels in the decrypted image were the
same as the number of pixels in the original image, which was
a rule normally followed by text data. The proposed method in
[21] was not a huge success when dark pictures with high
contrast were taken into consideration as the decrypted images
appeared corrupted with large amounts of noise. This
distortion basically happened because the secret image
generated was not distinctive. The encryption technique in
[21] thus needs to work on how to get a distinct sharp secret
image when dark high contrast images are involved. The
security of the encryption scheme in [21] depended on the Fig. 8. Encryption scheme as illustrated in [22]
color composition and the color distribution of the secret
image. This paper did not mention any detailed analysis of Oh et al [23] proposed a selective encryption algorithm
attacks. (SEA), which worked on similar lines to the Advanced
Encryption standard (AES) algorithm. Like AES, the SEA
Like [21], Patil et al [22] was also an encryption technique algorithm also dealt with block cipher. Here the core
that was based purely on visual cryptography. The work in computation went through several rounds of iteration
[22] also functioned devoid of any key and the encryption depending on the key size chosen. In short, the number of
process was very simple to execute. The encryption process iterations required was directly dependent on the key size
basically consisted of three main operations namely, sieving chosen at the beginning of the algorithm. AES was generally
(logical XOR operator implemented), division and shuffling. not considered suitable for visual data (included audio, video,
Sieving meant breaking the original RGB image to its image, text etc.) as it could have involved really long
individual components. Division meant breaking each of the computation processes. This work introduced a selector
R, G and B components to numerous number of shares. For component right at the beginning of the encryption process.
example, R could be broken to shares from R A to RZ. The selector component made the selection of ROI from the
Shuffling was the last step in the encryption process which inputted image and then it performed compression of the ROI
dealt with shuffling of the divided shares (from the division based on Huffman coding. The remaining rounds of
step) within itself. These shuffled shares were later combined encryption was very similar to the normal AES algorithm.
to generate random shares, for instance in Figure 8 there are Figure 9 and Figure 10 depicts the similarity and differences
two random shares generated. Since majority of the operations between the SEA algorithm and AES algorithm with the SEA
in [22] were achieved by using logical operators like OR, algorithm maintaining the rounds of iteration (computation)
XOR and the Mod operator, the computation cost incurred same as that as AES.
was very less and the implementation in [22] was written
purely in java ([10-21] [23-26] implementations were mainly
in MATLAB). Since encryption schemes in [21] and [22] do
not have keys involved, they do not have to bother about key
management or brute force attacks. But the main drawback of
the encryption schemes in [21] and [22] was that, generally
the quality of the decrypted or recovered image is poor. Figure
8 illustrates the encryption scheme as depicted in [22].

Fig. 9. SEA encryption scheme which is likely to AES (courtesy: ref [23])

SEA algorithm had the same security level that the AES
standard algorithm possessed. The only difference between the

8|P age
www.ijacsa.thesai.org
 (IJACSA) International Journal of Advanced Computer Science and Applications,
Vol. 7, No. 10, 2016

two seemed to be that the SEA encryption scheme had a JPEG compression encoding. The only difference between the
compressor component and a selector component to its encoding scheme in [25] and that of the JPEG compression
architecture, something that a normal AES algorithm did not encoding was that, [25] made use of quaternion discrete cosine
have. transforms (QDCT) instead of discrete cosine transforms used
in the JPEG encoding scheme. Since the quantization phase
was involved in the encryption scheme in [25], the encryption
process was rendered to be lossy in nature. Just like the JPEG
compression, the encryption scheme in [25] initially divided
the original color image into 8x8 sized blocks. The QDCT is
performed instead of normal DCT in [25]. The remaining
phases remain the same as JPEG compression algorithm. That
is, the next subsequent phase was quantization, followed by
the zigzag scanning (exploited redundancy and organized
coefficients in an array in an order of lower frequency
coefficients followed by higher frequency coefficients),
sorting of the coefficients and lastly the entropy coding and
encryption. Figure 12 and Figure 13 attempt to draw a
comparison between the encryption scheme as discussed in
[25] and the JPEG compression encoding scheme.

Fig. 10. Encryption scheme of regular AES (courtesy: ref [23])

The encryption technique in Parameshachari et al [24] was

a very simple technique. The key was generated by a random
key generator here. The inputted color image was initially
broken to sub blocks. The selection of the blocks for
encryption happened randomly using the XOR and Mod
operators. The randomly selected block along with the random
key generator gave the encrypted block. The encrypted block
was later combined with all the unencrypted blocks which in
turn produced the partially encrypted image for the encryption
scheme in [24]. The additional feature [24] had was that it
made use of Self Monitoring Analysis & Reporting Fig. 12. Encryption scheme in [25] (courtesy: ref [25])
Technology copyback system (SMART) to store long term the
encrypted images generated. Figure 11 explains the flow of
the encryption process at [24].

Fig. 11. The flow of the encryption scheme in [24](courtesy: ref [24])

Fan et al [25] and Flayh et al [26] were partial encryption

algorithms which were implemented in the frequency domain.
The encoding scheme for [25] was very much similar to the

9|P age
www.ijacsa.thesai.org
 (IJACSA) International Journal of Advanced Computer Science and Applications,
Vol. 7, No. 10, 2016

algorith
ms
Partial
encryptio
[22 Spatial ROI based RGB color
n No
] domain algorithms model
algorith
ms
Partial
encryptio (VC + ROI)
[23 Spatial Medical/Dico
n based No
] domain m data
algorith algorithms
ms
Partial Any color
encryptio (VC + ROI) space (mostly
[24 Spatial
n based RGB color No
] domain
algorith algorithms models dealt
ms with)
Uses QDCT
Partial
and very Yes. Lossy
encryptio Frequenc
[25 similar to Any color compressio
n y
] JPEG space n by use of
algorith domain
compression QDCT
Fig. 13. JPEG compression encoding scheme (courtesy: ref [28]) ms
algorithm
Yes. Mostly
Flayh et al [26] proposed a wavelet based partial Partial lossless
encryption algorithm. The first step was to select a key which [26
encryptio Frequenc
Uses Haar Any color
compressio
would be useful in the encryption process. The subsequent n y n achieved
] wavelets space
step was to select the wavelet filter that one would like to use algorith domain because of
ms Haar
in the encryption process. The encryption scheme in [26] wavelets
made use of AES or stream cipher to partially encrypt the
inputted color image. The whole idea behind using wavelets IV. FUTURE WORK
was to encrypt only the lower frequency coefficients as
The topic of ‗Image Encryption Algorithms for Color
opposed to the whole image (consists of lower as well as
Images‘ is pretty huge to cover in a single survey paper. This
higher frequencies). The higher frequency coefficients could
is a topic that has a lot of scope and is still evolving. There is a
be discarded or could be coded at a lower bit rate and the
future and possibility for the partially encryption algorithms to
lower frequency coefficients could be fully encoded using
evolve. Chaotic cryptosystem, though very new to explore,
AES or stream ciphers thus making it a partially encrypted
having been evolved only during the the last decade, still is
algorithm. If there was a necessity to hide any further details
not a strong enough system and can easily be broken. Many
in the cipher/encrypted image then, one could just perform a
other transforms in the frequency domain of the various
simple operation like smoothing over the image. The wavelet
classes of encryption schemes could be explored and their
used for the encryption purpose in [26] was the haar wavelet.
results could be noted. Visual cryptography (VC) has a lot of
The coefficient correlation of the pixels in the cipher image
scope of improvement in the near future. VC based algorithms
was nearly equal to zero which was desirable in an encryption
generally render poor quality of decrypted image, so a lot of
scheme. The main problem faced in the encryption scheme in
research is going on as to how to handle certain classes of
[26] was that, as the amount of encrypted part of the image
images in a better manner (especially dark high contrasted
decreased, the execution time decreased too, which was a
images). It is difficult to say that any particular cryptosystem
good aspect but the coefficient correlation of the pixels in the
or encryption algorithm is the safest as most of the
encrypted image had increased, which was not very desirable
cryptosystems have easily been broken into. This is what
in the encryption scheme discussed in [26]. Table-2
makes encryption an ever evolving topic which has a lot of
summarizes the partial encryption algorithms discussed in this
scope to expand in the future.
section.
V. CONCLUSION
TABLE II. SUMMARY OF PARTIAL ENCRYPTION ALGORITHMS. THE
ALGORITHMS MENTIONED IN THE TABLE ARE CAPABLE OF DEALING WITH Few significant papers on image encryption algorithms for
ANY COLOR SPACE color images were discussed during the period 2007-2015 and
Re Type of the algorithms were classified as full encryption algorithms or
Categor Class/Meth Compressi
f Domain dataset/Colo partial encryption algorithms. Traditional cryptosystems fail to
y od on used??
No r space work for bulky and voluminous data like image data and they
Partial fail to produce desirable results for real-time applications. The
encryptio
[20 Spatial ROI based Biometric algorithms in either of the classes were further divided based
n No
] domain algorithms data on whether they were spatial domain, frequency domain or
algorith
ms hybrid domain based algorithms. Based on the review one can
[21
Partial
Spatial VC based CMY color
safely conclude that full encryption based encryption schemes
encryptio No were more suitable for higher security applications while
] domain algorithms model
n partial encryption based encryption schemes were a lot

10 | P a g e
www.ijacsa.thesai.org
 (IJACSA) International Journal of Advanced Computer Science and Applications,
Vol. 7, No. 10, 2016

suitable for real-time applications as they took lesser EDT [15] Samson, Ch, and V. U. K. Sastry. "An RGB Image Encryption
time and the computational requirements and complexity were Supported by Wavelet-based Lossless Compression." International
Journal of Advanced Computer Science and Applications 3, no. 9
smaller. (2012).
REFERENCES [16] Z. Yu, Z. Zhe, Y. Haibing, P. Wenjie, and Z. Yunpeng, ―A chaos-based
[1] Arroyo, David, Rhouma Rhouma, Gonzalo Alvarez, Veronica image encryption algorithm using wavelet transform‖, in 2nd
Fernandez, and Safya Belghith. "On the skew tent map as base of a new International Conference on Advanced Computer Control, March 2010,
image chaos-based encryption scheme." In Second Workshop on Vol. 2, no. 4, pp. 217–222.
Mathematical Cryptology, pp. 113-117. 2008. [17] Zhou, Nanrun, Yixian Wang, Lihua Gong, Hong He, and Jianhua Wu.
[2] Yin, Ruming, Jian Yuan, Qiuhua Yang, Xiuming Shan, and Xiqin "Novel single-channel color image encryption algorithm based on chaos
Wang. "Gemstone: a new stream cipher using coupled map lattice." and fractional Fourier transform." Optics Communications 284, no. 12
In Information Security and Cryptology, pp. 198-214. Springer Berlin (2011): 2789-2796.
Heidelberg, 2010. [18] Arroyo, David, Rhouma Rhouma, Gonzalo Alvarez, Veronica
[3] Öztürk, İsmet, and İbrahim Soğukpınar. "Analysis and comparison of Fernandez, and Safya Belghith. "On the skew tent map as base of a new
image encryption algorithms." International Journal of Information image chaos-based encryption scheme." In Second Workshop on
Technology 1, no. 2 (2004): 108-111. Mathematical Cryptology, pp. 113-117. 2008.
[4] Jawad, Lahieb Mohammed, and Ghazali Bin Sulong. "A review of color [19] Kocarev, Ljupčo. "Chaos-based cryptography: a brief
image encryption techniques." International Journal of Computer overview." Circuits and Systems Magazine, IEEE 1, no. 3 (2001): 6-21.
Science Issues 10, no. 6 (2013): 266-275. [20] A. Wong and W. Bishop, ―Backwards Compatible, Multi-Level Region-
[5] MULUALEM, GETACHEW MEHABIE. "COMPRESSION AND of-Interest (ROI) Image Encryption Architecture with Biometric
ENCRYPTION FOR SATELLITE IMAGES: A COMPARISON Authentication‖, International Conference on Signal Processing and
BETWEEN SQUEEZE CIPHER AND SPATIAL SIMULATIONS." Multimedia Applications, July 2007, pp. 324 – 329.
(2015). [21] Abdulla, Sozan. "New Visual Cryptography Algorithm For Colored
[6] Johnson, Mark, Prakash Ishwar, Vinod Prabhakaran, Daniel Schonberg, Image." arXiv preprint arXiv:1004.4445 (2010).
and Kannan Ramchandran. "On compressing encrypted data." Signal [22] Patil, Shobha, and V. R. Udupi. "A Secure Approach to Image
Processing, IEEE Transactions on 52, no. 10 (2004): 2992-3006. Encryption of color image without using key." Received from:
[7] Klinc, Demijan, Carmit Hazay, Ashish Jagmohan, Hugo Krawczyk, and http://inpressco. com/category/ijcet (2013).
Tal Rabin. "On compression of data encrypted with block [23] Oh, Ju-Young, Dong-Il Yang, and Ki-Hwan Chon. "A selective
ciphers." Information Theory, IEEE Transactions on 58, no. 11 (2012): encryption algorithm based on AES for medical
6989-7001. information." Healthcare informatics research 16, no. 1 (2010): 22-29.
[8] Zhou, Jiantao, et al. "Designing an efficient image encryption-then- [24] Parameshachari, B. D., KM Sunjiv Soyjaudah, and Sumithra Devi KA.
compression system via prediction error clustering and random "Partial Image Encryption Algorithm Using Pixel Position Manipulation
permutation." Information Forensics and Security, IEEE Transactions Technique: The SMART Copyback System."
on 9.1 (2014): 39-50. [25] Fan, Jing, Jinwei Wang, Xingming Sun, and Ting Li. "Partial Encryption
[9] Ali A.Yassin. ―Design New Algorithm for Partial Image Encryption of Color Image Using Quaternion Discrete Cosine
Based colors Space.‖ Journal of Babylon University, Pure and Applied Transform." structure 8, no. 10 (2015).
Sciences, No.(2), Vol.(20): 2012 [26] Flayh, Nahla A., Rafat Parveen, and Syed I. Ahson. "Wavelet based
[10] R. Rhouma, D. Arroyo, and S. Belghith, ―A new color image partial image encryption." In 2009 International Multimedia, Signal
cryptosystem based on a piecewise linear chaotic map‖, in 6th Processing and Communication Technologies. 2009.
International Multi-Conference on Systems, Signals and Devices, Mar. [27] Introduction to image processing -
2009, pp. 1–6. https://www.spacetelescope.org/static/projects/fits_liberator/image_proc
[11] Musheer Ahmad, and M. Alam, ―A New Algorithm of Encryption and essing.pdf
Decryption of Images Using Chaotic Mapping‖, International Journal on [28] Image compression: JPEG, Multimedia Systems (module 4, lesson 1)
Computer Science and Engineering, vol. 2, no. 1, 2009, pp. 46–50.
[29] Umbaugh, Scott E. Digital image processing and analysis: human and
[12] Chandel, Gajendra Singh, and Pragna Patel. "Image Encryption with computer vision applications with CVIPtools. CRC press, 2010.
RSA and RGB randomized Histograms." Image 3, no. 5 (2014).
[30] Debbarma, Nikhil, Lalita Kumari, and Jagdish Lal Raheja. "2D Chaos
[13] Wang, Xingyuan, Lin Teng, and Xue Qin. "A novel colour image Based Color Image Encryption Using Pseudorandom Key
encryption algorithm based on chaos." Signal Processing 92, no. 4 Generation." Chaos 2, no. 4 (2013).
(2012): 1101-1108.
[31] Pisarchik, Alexander N., and Massimiliano Zanin. "Chaotic map
[14] H. Chen, X. Du, Z. Liu, and C. Yang, ―Color image encryption based on cryptography and security." International Journal of Computer
the affine transform and gyrator transform‖, Optics and Lasers in Research 19, no. 1 (2012): 49.
Engineering, vol. 51, no. 6, Jun. 2013, pp. 768–775.

11 | P a g e
www.ijacsa.thesai.org