Multimedia Tools and Applications (2022) 81:31349–31362

https://doi.org/10.1007/s11042-022-11985-2

A generalized Arnold’s Cat Map transformation

for image scrambling

Hakan Tora1 · Erhan Gokcay2 · Mehmet Turan3 · Mohamed Buker4

Received: 21 September 2020 / Revised: 28 July 2021 / Accepted: 3 January 2022 /
Published online: 9 April 2022
© The Author(s), under exclusive licence to Springer Science+Business Media, LLC, part of Springer Nature 2022

Abstract
This study presents a new approach to generate the transformation matrix for Arnold’s Cat
Map (ACM). Matrices of standard and modified ACM are well known by many users. Since
the structure of the possible matrices is known, one can easily select one of them and use it
to recover the image with several trials. However, the proposed method generates a larger
set of transform matrices. Thus, one will have difficulty in estimating the transform matrix
used for scrambling. There is no fixed structure for our matrix as in standard or modified
ACM, making it much harder for the transform matrix to be discovered. It is possible to use
different type, order and number of operations to generate the transform matrix. The quality
of the shuffling process and the strength against brute-force attacks of the proposed method
is tested on several benchmark images.

Keywords Image scrambling · Arnold’s Cat Map · Information security · Transformation

matrix · Chaotic maps

1 Introduction

Nowadays, images are widely exchanged through the internet. These daily used images may
be from the fields of military, medicine, transportation or they may be personal [15, 17].

Hakan Tora
[email protected]

Erhan Gokcay
[email protected]

Mehmet Turan
[email protected]

Mohamed Buker
[email protected]

1 Department of Electrical and Electronics Engineering, Atilim University, Ankara, Turkey

2 Department of Software Engineering, Atilim University, Ankara, Turkey
3 Department of Mathematics, Atilim University, Ankara, Turkey
4 Graduate School of Natural and Applied Sciences, Atilim University, Ankara, Turkey
31350 Multimedia Tools and Applications (2022) 81:31349–31362

Such multimedia tools usually contain sensitive or confidential information, which requires
them to be sent securely. Security, thus, becomes very crucial for communication through
the internet. Cryptography deals with the encryption of data such as text, images, video,
etc. When the encrypted data is an image, the related techniques can be classified into two
categories: substitution and transposition; the former is based on changing the pixel’s values.
The algorithms used for this are known as diffusion algorithms; in other words, the pixel
values are completely changed. The latter is based on shuffling the pixels’ positions. In
this respect Arnold’s Cat Map (ACM) is one of the transposition techniques that scrambles
the image pixels [8, 19]. ACM is simply built on matrix transformation. It is an iterative
procedure that starts with a given image at the beginning and ends when the given image is
reconstructed, which implies that the ACM is periodic.
ACM along with the other scrambling approaches have been used very often on images.
Lin proposed a method for image encryption that encrypts a cover image by combining
the ACM and the Hartley transform method [8]. A technique based on Fibonacci transform for
scrambling is presented in [18]. It is the same as ACM, the only difference being that the
values of the four entries in the transform matrix are four consecutive Fibonacci numbers. An
improved image encryption technique based on a chaotic 3D cat map and Turing machine
in the form of dynamic random growth technique was introduced in [9]. The encryption perfor-
mance is strengthened and the key space is enlarged in [6]; in detail, the RGB layers are shuffled
using a 3D Cat map and standard map, and, finally, the shuffled image is encrypted by perfor-
ming XOR operation. Anwar et al. proposed an encryption technique based on pixel permuta-
tion using chaotic maps [2]. An image encryption algorithm based on bit-plane scrambling and
multiple chaotic system combination was proposed in [16], where a compounding of scrambling
of pixel positions and the changing of pixel values was done to strengthen security. Dongxu et
al. made use of the Fibonacci sequence and found that Arnold transformation is related with
that sequence. They defined the so-called Fibonacci Q-transformation and its application in
the scrambling action of digital image [4]. In [12], periodicity analysis of transformation was
performed, and it was proved that the periodicity of Arnold transformation is related, but not pro-
portional, to the image size. Dyson discussed the periodicity of ACM and its relation to the size of
the image, and gave some mathematical proofs [5]. They also discussed the upper and lower
bounds for the period and explicit values for particular cases. Abood proposed an image
cryptography technique to increase security in a network environment [1]. He first obtained
a shuffled image by using RC4 encryption algorithm and then embedded it to a cover image
by employing hash based LSB technique. In [7] Kumar and Chauhan presented an image crypto-
graphy that employs matrix array symmetric key (MASK) using chaos based encryption aproach
for secure transmission of data. Thakur et al. discussed a method for medical image watermarking
based on DWT, DCT and SVD [15]. In order to improve the security of the method, they utilized
2D logistic map based chaotic encryption. Zhang and Zhang proposed a multiple-image encryp-
tion algorithm based on bit planes and chaos [17]. Their algorithm only scrambles the 5th-8th
bit planes of original images by Chen chaotic system and two-dimensional Logistic map. The defi-
nition and study of the period and chaotic properties of one-dimensional pseudo-chaotic sequen-
ces, called z-sequences, based on the two-dimensional sequences generated by the discrete Arnold’s
cat map (DACM) were introduced by Souza et al [14]. In [10], Musanna et al. introduced an
image encryption approach using 3D Arnold’s Cat Map for permutation phase of encryp-
tion and logistic map for diffusion phase. Hence, the approach makes the image completely unre-
cognisable. Shalaby et al. presented a medical encryption approach which modifies Arnold’s cat
map technique combined with the well-known Advanced Encryption Standard (AES) algo-
rithm [11]. The approach was called enhanced Arnold’s Cat map-AES-131 (ECAT-AES-131)
encryption technique.
Conventional ACM matrices that are used in the aforementioned encryption techniques
have well-known structures. One of the elements in these matrices is always set to unity.
Multimedia Tools and Applications (2022) 81:31349–31362 31351

Therefore, standard matrix can easily be predicted and the encrypted information can be
revealed. This is not desired in cases where vital information is transmitted. In this study, we
propose a new approach to generate a transformation matrix used in ACM. The proposed
technique overcomes this problem. The transformation matrix has no predefined structure
or formula in terms of matrix elements. Here, the matrix is obtained using several basic
matrix operations, such as row and column operations, rotation, transposition, swapping
the rows and columns, etc., to ensure that the determinant of the transformation matrix
is always ±1.Thus, there will be larger pool of transformation matrices than the standard
ACM. Each matrix in this pool has its own characteristics regarding the way it shuffles
the given image. In other words, each matrix requires a different number of iterations to
reconstruct the image, which is simply called the period. Each iteration between the initial
and the final state of the cycle generates a different scrambled image. The least pixel-wise
correlated image is selected as the best scrambled image. To measure the quality of the shuf-
fled image, the correlation between the adjacent pixels of the image is used. The proposed
transformation matrix is tested on color images of size N × N .
The paper is organized as follows. The background information on the standard ACM
transformation is described in Section 2. The proposed approach is presented in Section 3.
Section 4 gives the results of correlation analysis between the neighboring pixels. Finally,
the conclusion appears in Section 5.

2 Background on standard ACM

Russian mathematician Vladimir I Arnold discovered a method to scramble the positions of

matrix elements according to a 2-by-2 transformation matrix [3]. This chaotic algorithm is
known as Arnold’s cat map (ACM), which is a two-dimensional invertible chaotic map.
For an image of size N × N , ACM is defined by
x 1 1 x
= (mod N ) (2.1)
y 1 2 y
where x, y ∈ {0, 1, 2, . . . , N − 1}, (x, y) is the pixel of the original image and (x , y ) is
the mapped pixel. Note that the determinant of the transformation matrix is unity, which
guarantees that the mapping is one-to-one, hence, area preserving is ensured.
Figure 1 illustrates an example of this transformation. As can be seen, the original image
is recovered at iteration 100, which is the period of the transformation matrix used in this
example. Note that there exist many other images between the initial and the final states.
Beside the traditional ACM, several modifications have been also introduced by other
researchers. See, for example, [12, 13] and the references therein. These modified ACMs
differ from the traditional method in terms of the matrix elements. One of them is defined
as follows:
x 1 1 x
= (mod N ) (2.2)
y i i+1 y

Fig. 1 Scrambling process at different iteration numbers

31352 Multimedia Tools and Applications (2022) 81:31349–31362

Equation (2.2) allows one to select any matrix which is specified by variable i. Additionally,
Eq. (2.3) is another modified version of ACM
x 1 a x
= (mod N ) (2.3)
y b ab + 1 y
This version offers more flexibility to the user for selecting the transformation matrix.
Compared with the other two ACMs in (2.1) and (2.2), it is more difficult for attackers to
recover the image since the transformation matrix has two independent variables.

3 Generation of a new ACM transformation matrix

The proposed method introduces a new ACM transformation matrix set including the stan-
dard and modified ACM matrices as subsets. We started with the fact that the transform
matrix of traditional ACM and its variations have determinant 1. The proposed algorithm
can also generate a transformation matrix with determinant −1, which inspired us to toward
using transform matrix M in any other form with the only criteria det(M) = ±1 (mod N ).

3.1 A generalized transform matrix

It is well known that for a square matrix, the absolute value of its determinant remains
unchanged when it undergoes some matrix operations such as addition/subtraction of rows
and columns, taking the transpose and inverse of the matrix, exchanging rows and columns,
multiplication of two matrices with determinants ±1, and rotating the elements of the matrix
clockwise and counter clockwise direction. In this study, we made use of such properties
of the determinant to generate a generalized transformation matrix, which shuffles a given
image. In addition, a set of operation sequence is introduced to calculate the matrix. This
sequence is required to regenerate the matrix in the retrieving phase, and it will be called
operSeq throughout the paper. It consists of several operations of any length and order. For
instance, operSeq = {inverse, rotation, inverse, transpose, row operations, column opera-
tions, etc}. Note that the inversion operation is repeated twice in this operSeq. In general,
all operations can be used in any number of times. The proposed algorithm starts with the
1 1 2 1
initial matrices M1 = and M2 = with unity determinant and an operSeq,
1 2 1 1
which is a list of operations that determines the sequence of operations applied to the initial
matrix. By performing the operations mentioned above, the algorithm is able to generate a
large number of various matrices. The execution order, type and number of operations can
be modified for each scrambling process to make the estimation much harder for an attacker.
Only one of these matrices will be chosen according to our operSeq. The algorithm leads
a b
to a matrix Mg = , whose elements can take on integer values 0, 1, 2, . . . , N − 1.
c d
There is no predefined dependency among the elements, except that the determinant is ±1.
This differs from the modified matrix given in (2.2) and (2.3). Notice that, in traditional and
modified versions, the upper left element is always set to 1 and the number of possible trans-
form matrices is at most N 2 . However, there is no such restriction in our proposed method
and the set of matrices generated by our algorithm is simply a super set of the traditional
and modified versions. It should be pointed out that the proposed algorithm may result in a
matrix that is exactly the same as, or completely different from, the conventional version of
the ACM matrix.
Since the generalized matrix has more independent elements, estimation of this matrix
is much harder than the conventional matrices. Algorithm 1 introduces a pseudo code for
Multimedia Tools and Applications (2022) 81:31349–31362 31353

the proposed method. As can be seen, the inputs of the function generate matrix are M1,
M2, N and operSeq. The operSeq list determines the type and number of operations needed
to obtain the transformation matrix. There is no limit on the number of operations. To
increase the possible number of combinations, the algorithm starts with 2 different matri-
ces (i.e. M1 and M2). One could have more matrices instead of 2. At every loop, firstly,
operSeq(counter) is applied to M1. Then, using the next operation, matrix M2 is updated.
After that, the following operation in the list operSeq is applied to the product M1 · M2. The
scaling constant k is used in row multiplication and it is incremented at each repetition. The
algorithm returns the generated matrix Mg once the number of operations reaches the limit
indicated by the operSeq list. The derived transform matrix will make a potential attacker’s
job more difficult to recover the original image. This is because, even if the attacker knows
the algorithm, he/she also has to know the parameters of the matrix with a great number of
combinations. Otherwise, he/she will be unable to recover the original image. The pseudo
code is given in Algorithm 1.
31354 Multimedia Tools and Applications (2022) 81:31349–31362

As an example, let N = 200 and operSeq = {swapCols, swapCols, krow2 +row1 ,

inverse, swapCols, transpose, krow2 +row1 , rot 90◦ , swapRows, rot 180◦ , krow1 +row2 ,
inverse}. Here, operSeq.length = 12. The intermediate steps are given in Table 1.

3.2 ACM with a generalized transform matrix

We define a generalized ACM transformation as follows:

x x a b
= Mg (mod N ), Mg = (3.1)
y y c d
where Mg is the proposed transformation matrix obtained from Algorithm 1. Algorithm 2
presents the well-known scrambling using the ACM. The algorithm will return the original
image in a finite number of steps, called the period, P , of the transformation matrix. It is
well known that ACM generates P − 1 scrambled images, each of which is a candidate to
become a scrambled image. A shuffled image is selected based on the correlation score of
each candidate image. Thus, it is ensured that the appearance of the selected image reveals
no clue about the shuffled image. This score is simply the average of the correlation of the
pixels in adjacent rows and columns (a measure of the similarity of pixel values). More
precisely, in our algorithm, we define a vector x containing all columns except for the last
one, and a vector y containing all columns excluding the first column. Then, the correlation
coefficient of these vectors is computed. A similar coefficient is obtained for the rows.
Finally, the average of these two coefficients is taken as the correlation score.

Algorithm 3 describes the descrambling process. First, the generalized transformation

matrix needs to be regenerated using the algorithm stated in Section 3.1.

4 Correlation analysis

Scrambled images should not have any explicit similarity within any part to the original
image. One of the metrics for relieving image content is pixel correlation within an image.
Correlation among adjacent pixels measures the closeness of pixel values of an image to
their neighbour values. For any ordinary image, each pixel is noticeably correlated with its
Table 1 A sample calculation sequence of the transformation matrix

M (init) swapCols Init-M2 swapCols Multiply k·row2 +row1 Init-M1 Inverse

1 1 1 1 2 1 1 2 2 3 5 8 1 1 −1 1
1 2 2 1 1 1 1 1 3 5 3 5 2 1 2 −1
Init-M2 swapCols multiply transpose Init-M1 k·row2 +row1 Init-M2 rot 90◦
Multimedia Tools and Applications (2022) 81:31349–31362

5 8 8 5 −3 −2 −3 11 −1 1 5 −2 −3 11 11 7
3 5 5 3 11 7 −2 7 2 −1 2 −1 −2 7 −3 −2
multiply swapRows Init-M1 rot 90◦ Init-M2 k·row1 +row2 multiply inverse
61 39 25 16 5 −2 −1 2 25 16 25 16 97 190 83 10
25 16 61 39 2 −1 −2 5 61 39 161 103 155 83 45 97
31355
31356 Multimedia Tools and Applications (2022) 81:31349–31362

Fig. 2 Scrambling phase

Fig. 3 Scrambling phase

Fig. 4 Peppers image at different iterations during scrambling process

Table 2 Correlation values of original and scrambled images

Image Original image Scrambled image at best iteration Scrambled image at iteration 100

Horizontal Vertical Horizontal Vertical Horizontal Vertical

Lenna 0.9163 0.9520 0.3272 0.3151 0.4253 0.4250

Penguin 0.9174 0.9571 0.0960 0.0977 0.2853 0.2855
Peppers 0.9384 0.9447 0.1472 0.1779 0.2863 0.2851
Multimedia Tools and Applications (2022) 81:31349–31362 31357

neighbouring pixels. Figure 2 shows the correlation between the pixels in horizontal and
vertical directions for Peppers image of size 200 × 200. Figure 3 illustrates the correla-
tions for the scrambled Peppers image obtained at iteration 157 with the generated matrix
4 3
. It is clear that the neighbouring pixels in the original image are highly correlated
3 2
in both directions. On the other hand, less correlation between neighbouring pixels appears
in the scrambled image.
Figure 4 shows some scrambled images at several iteration numbers. Notice that the
worst scrambled image, which is highly similar to the original image, is at iteration 100.
Since there is no direct relationship between the quality of the scrambled image
and the length of the iterations, a random selection of the iteration number may
result in a highly correlated image, which is undesired. Table 2 presents the correla-
tion values for some selected images. The best scrambled images for Lenna, Penguin
and Peppers are obtained at different iteration numbers, which are 175, 32 and 157,
respectively. As seen, correlations at iteration 100 are higher than the ones at the best
iterations.
The scrambling process is executed over a test image by using the transform matrix
3 2
for the full period. The results are shown in Fig. 5. Pixel correlation in the original
4 3
image is more than 0.9. As seen clearly, the image at iteration 31 has the highest pixel cor-
relation, where one can significantly retrieve the original content; while the best scrambled
image is obtained at 53, which is the minimum of the correlation curve.
Furthermore, 10 images of size 200×200 are randomly selected and are scrambled using
the proposed matrices generated in Algorithm 1. Tables 3 and 4 show the results, which
confirm empirically that the set of the traditional and modified transform matrices is a subset
of our matrix space. For example, the standard ACM matrix appears in row 3 of Table 3
and the matrix used in (2.3) appears in row 1 of Table 4. Also, note that the determinants of
the matrices used in scrambling in the last two rows are not 1, but 1 mod 200. Additionally,
Table 5 exhibits more scrambled images resulted from matrices generated by the proposed
approach.

Fig. 5 Correlation scores for different iterations

Table 3 Scrambled images with our ACM matrices (Part I)
31358

Original image Pixel correlation of Best scrambled Pixel correlation of Proposed gener-
the original image image scrambled image ated transformation
matrix

23 14
0.9793 0.0555
5 3

2 1
0.9676 0.0135
3 2

1 1
0.5525 0.0015
1 2

2 −1
0.9582 0.0061
−3 2

13 8
0.9746 0.0175
5 3
Multimedia Tools and Applications (2022) 81:31349–31362
Table 4 Scrambled images with our ACM matrices (Part II)

Original image Pixel correlation of Best scrambled Pixel correlation of Proposed gener-
the original image image scrambled image ated transformation
matrix

1 4
0.8619 0.0931
5 21

8 −13
0.9961 0.0900
−3 5
Multimedia Tools and Applications (2022) 81:31349–31362

5 13
0.9757 0.0521
3 8

45 94
x 0.9391 0.0179
−4 45

143 151
0.9717 0.0183
31359

−8 151
Table 5 Some images and their best shuffled outputs

Original Scrambled Original Scrambled Original Scrambled

31360

x
Multimedia Tools and Applications (2022) 81:31349–31362
Multimedia Tools and Applications (2022) 81:31349–31362 31361

5 Conclusion

In this paper, a novel transformation matrix generation method based on the Arnold’s cat
map (ACM) is introduced. Our approach relies on a series of basic operations on matrices.
Unlike the traditional and modified ACMs, the proposed matrix does not have any limita-
tion on its elements. In other words, the elements can take any integer values as long as
the determinant of the matrix is unity in magnitude modulo N . Hence, the proposed trans-
formation matrix space contains the traditional ACM matrices. Consequently, the possible
number of combinations of the transformation matrix is significantly larger than the one in
the conventional ones. Thus, predicting the proposed matrices becomes more difficult for a
third party.

Acknowledgments The authors declare that they have equally participated in all parts of the manuscript.
Also, they would like to thank Mr. P. Danesh from the Atilim University Academic Writing and Advisory
Centre for his help in the preparation of the manuscript.

References

1. Abood MH (2017) An efficient image cryptography using hash-LSB steganography with RC4 and
pixel shuffling encryption algorithms. 2017 Annual Conference on New Trends in Information &
Communications Technology Applications (NTICT). IEEE
2. Anwar S, Meghana S (2016) A pixel permutation based image encryption technique using chaotic map.
Multimed Tools Appl 78(19):27569–27590
3. Arnold VI, Avez A (1968) Ergodic problems of classical mechanics. Benjamin, New York
4. Dongxu Q, Jiangcheng Z, Xiaoyou H (2000) A new class of scrambling transformation and its
application in the image information covering. Sci China Technol Sci 43(3):304–312
5. Dyson F, Falk H (1992) Period of a discrete cat mapping. Am Math Mon 99(7):603–614
6. Gupta K, Silakari S (2011) New approach for fast color image encryption using chaotic map. J Inf Secur
2(4):139–150
7. Kumar T, Chauhan S (2018) Image cryptography with matrix array symmetric key using chaos based
approach. Int J Comput Netw Inf Secur 10(3):60
8. Lin KT (2013) Image encryption using arnold transform technique and hartley transform domain. Ninth
International Conference on Intelligent Information Hiding and Multimedia Signal Processing, pp 84–87
9. Mohamed NA, El-Azeim MA, Zaghloul A (2016) Improving Image Encryption Using 3D Cat Map and
Turing Machine. Int J Adv Comput Sci Appl 7(1). https://doi.org/10.14569/IJACSA.2016.070129
10. Musanna F, Rani A, Kumar S (2018) Image encryption using chaotic 3-D arnold’s cat map and logistic
map. Proceedings of 2nd International Conference on Computer Vision & Image Processing. Springer,
Singapore, pp 365–378
11. Shalaby MAW, Salel MT, Elmahdy HN (2020) Enhanced Arnold’s Cat map-AES Encryption Technique
for Medical Images. 2020 2nd Novel Intelligent and Leading Emerging Sciences Conference (NILES).
IEEE
12. Shang Z, Ren H, Zhang J (2008) A block location scrambling algorithm of digital image based on Arnold
transformation. The 9th International Conference for Young Computer Scientists, pp 2942–2947
13. Sinha RK, San N, Asha B, Prasad S, Sahu SS (2018) Chaotic Image Encryption Scheme Based on
Modified Arnold Cat Map and Henon Map 2018 International Conference on Current Trends towards
Converging Technologies (ICCTCT)
14. Souza CEC, Chaves DPB, Pimentel C (2021) One-dimensional Pseudo-Chaotic Sequences Based on the
Discrete Arnold’s Cat Map Over Z3m . IEEE Trans Circ Syst II: Express Briefs 68(1):491–495
15. Thakur S, Singh AK, Ghrera SP, Elhoseny M (2019) Multi-layer security of medical data through water-
marking and chaotic encryption for tele-health applications. Multimed Tools Appl 78(3):3457–3470.
https://doi.org/10.1007/s11042-018-6263-3
16. Zhang H, Cai R (2010) Image encryption algorithm based on Bit-Plane scrambling and multiple chaotic
systems combination. International Conference on Intelligent Computing and Integrated Systems, pp
113–117
31362 Multimedia Tools and Applications (2022) 81:31349–31362

17. Zhang L, Zhang X (2020) Multiple-image encryption algorithm based on bit planes and chaos. Multimed
Tools Appl 79(29):20753–20771
18. Zou W, Huang J, Zhou C (2010) Digital image scrambling technology based on two dimension
fibonacci transformation and its periodicity. Third International Symposium on Information Science and
Engineering, pp 415–418
19. Zou J, Ward RK (2003) Introducing two new image scrambling methods. Pacific Rim Conference on
Communications Computers and Signal Processing

Publisher’s note Springer Nature remains neutral with regard to jurisdictional claims in published maps
and institutional affiliations.