2017 14th International Conference on Computer Graphics, Imaging and Visualization

A Novel Selective Encryption DWT-based Algorithm for Medical Images

Med Karim Abdmouleh, Ali Khalfallah and Med Salim Bouhlel

University of Sfax
Research Unit: Sciences and Technologies of Image and Telecommunications
Higher Institute of Biotechnology, Sfax–Tunisia
medkarim.abdmouleh@isggb.rnu.tn, ali.khalfallah@enetcom.rnu.tn, medsalim.bouhlel@enis.rnu.tn

Abstract ment of techniques for securing the transmission and stor-

The remarkable development of the medical imagery is age of medical images remains a necessity. To achieve this,
due to the large use of telecommunications and informa- one must resort to encryption and compression of images
tion technologies in the medical domain. However, since [3, 5].
the telemedicine is a medical act that must answer to stern Indeed, the weight of medical images after digitization
rules, and follow the easiness that is offered by the infor- is increasingly important. It can range from 1 to 100 Mega
matics sciences to violate the confidentiality and authen- bytes for a full Scanner or MRI scan, and even more with
ticity of medical data. To resolve this problem, a lot of the operation of the new multicore scanners. Since these
methods combining compression and encryption have been images will be transmitted via networks and/or stored in
developed in the literature to secure the transmission and medical databases, they must be compressed in order to re-
the storage of medical images. In this paper, we propose duce transmission times, first, and second to improve the
a new approach of partial encryption based on the Dis- storage and archiving capacity on the other hand [5]. In
crete Wavelet Transform (DWT) and compatible with the addition to that, in order to comply with the laws of ethics
norm JPEG2000 in order to assure an optimal and a se- and to eliminate any risk of disclosure, alteration or avail-
cure transmission and storage of medical images. This ap- ability of any secrets concerning the patient’s condition or
proach is, on one hand, faster since it permits a very im- professional secrets, the images must be encrypted before
portant gain in the encryption-decryption processing time their transmission and/or storage [5].
(only 6.25% of the DWT matrix coefficients will be en-
Currently, a multitude of methods combining compres-
crypted), and on the other hand more efficient because it
sion and encryption operations are developed [6, 7, 11, 12,
permits to get an encrypted medical image.
13, 17]. Nevertheless, these methods are either inefficient
Keywords— Telemedicine, Encryption, Compression, or unsafe or very demanding in terms of computation time
DWT, RSA, JPEG2000. [3]. Moreover, the majority of these cryptosystems are ap-
plied to old compression techniques that are considerably
1 Introduction less efficient in terms of compression ratio than the new
Telemedicine is evolving considerably due to the re- algorithms based on DWT (in particular JPEG2000) [1].
markable development of information and telecommuni- In this context, we propose in this paper a new par-
cations technologies. This sector is based essentially on tial encryption approach based on DWT and compatible
the exchange of medical information which are essentially with the JPEG2000 standard, in order to optimize and se-
images. This is due to the fact that medical imaging is cur- cure the transmission and storage of medical images in
rently considered to be the crossroads of all specialisations telemedicine. In the next section of this paper, Section 2,
since it is present in almost all medical records. However, a brief overview on the Discrete Wavelet Transform will
in view of the fact that telemedicine is a medical proce- be given first. In Section 3, our new approach, which con-
dure which must respond to a medical one, and follow- cerns the application of partial encryption in the compo-
ing the awareness of the medical community of the ease nents of the DWT matrix, will be developed and argued. In
with which computers and new telecommunication tech- this part, the outline of our approach will be presented and
nologies can infringe confidentiality and certain authors the experimental results will be detailed and commented
are in a state of opposition to the computerization of data on. Finally, the benefits of our new approach will be listed.
for ethical and legal reasons [5]. As a result, the develop- Section 4 concludes the paper.

978-1-5386-0852-4/17 $31.00 © 2017 IEEE 78

79
DOI 10.1109/CGiV.2017.10
2 Discrete Wavelet Transform (DWT) The reconstruction consists of an operation symmetrical
The DWT is the basic transformation in the JPEG2000 to that of decomposition. If there is no compression, there
algorithm [1] which is the new standard of compression must be no loss of information in the image in the sense
of images [8]. The latter was developed by the Joint Pho- of signal processing. An over-sampling is applied to each
tographic Experts Group (JPEG) to replace the old JPEG of the four thumbnails obtained after decomposition of the
standard [14]. This transformation is a technique of signal original image, and then the low-pass and high-pass filters
processing which consists of decomposing an image I, into are applied to recover the original image. Thus, the Dis-
images of lower resolution [16]. To obtain a lower level of crete Wavelet Transform is the decomposition of the image
sub-band, one proceeds as follows: into approximations and details through respectively low-
pass and high-pass digital filters. The combination of these
• The low-pass filter (H) is applied to the lines of the filters makes it possible to obtain four sub-images HH, HL,
image followed by a sub-sampling. LH and LL. These filters are called mirror filters.
Multi-resolution analyzes were introduced in the late
• The high-pass filter (G) is applied to the lines of the
1980s [10]. These analyzes consist in performing the de-
image followed by a sub-sampling
composition operation several times on only the sub-band
• On the two images obtained we do the same work, LL. Figure 3 shows the schematic diagram of the decom-
but on the columns of the image. These steps are positions in several levels.
illustrated by the diagram of Figure 1(a) [9].
3 Development of our approach
By using DWT, an image can be decomposed into any In this part, we will develop our new approach that
level sub-band images, as shown in Figure 1(b). mixes selective encryption in DWT components and cryp-
In Figure 1(b): tographic techniques of image jamming. To do this, we
will first present the schema of our approach. Next, we will
• LL: represents the data coming from two low-pass explain the principle of selective encryption in the compo-
filters, one horizontal and the other vertical. The nents of the DWT that we propose in this work. Finally,
content of j corresponds to the original image but we will enumerate the benefits of the new approach.
with a lower resolution. 3.1 Schematic diagram
• LH: represents data from two filters. The first is a Since we envisage the application of our approach on
horizontal low-pass, the other is high-pass vertical. the compression algorithm JPEG2000, we present in Fig-
Its content, therefore, corresponds to horizontal de- ure 4 the principle of our approach based on the schema of
tails. a typical coder dedicated to the compression by this norm.
This new approach consists in applying a partial encryp-
• HL: represents data from two filters: a horizontal tion in the components of the DWT matrix. To restore the
high pass and a vertical low pass. Its content, there- initial information, the quantized coefficients of the matrix
fore, corresponds to vertical details. DWT are first decoded by the entropy decoder. Then, they
are decrypted before the dequantization step using the in-
• HH: this space undergoes a high-pass filtering in
verse cryptographic procedure. Finally, IDWT (inverse of
both directions, it then contains the diagonal detail
the DWT) [4] is applied to restore the original image.
information.
3.2 Selective encryption based on DWT
The decomposition of a discrete image I provides at In a multi-resolution analysis, the decomposition level
each step four sub images, one corresponds to the lower ap- increases, and the quality of image representation by the
proximation (sub-band LL) and the other three are consti- approximation contained in the LL sub-band is less accu-
tuted by the coefficients of the wavelet decomposition and rate. Figure 5 below illustrates on the example of an X-ray
contain the lost details In the approximation. These details image the result of its decomposition for different resolu-
are, depending on the image, horizontally oriented (sub- tions on the one hand and the result of the reconstruction
band LH), vertically (sub-band HL) or diagonally (sub- after cancellation of all the sub-band other than the LLs.
band HH). The area occupied by the four thumbnails to- The basic idea of the selective encryption that we
gether is the same as that occupied by the original image. present in this part is to exploit the fact that after each de-
There is neither compression nor redundancy in the repre- composition operation, most of the information in the im-
sentation obtained [9]. age is contained in the LL sub-band, and the other bands
Figure 2 illustrates the result of the application of the Contain information characterizing details. Consequently,
Discrete Wavelet Transform on a medical image represent- the encryption of the coefficients of this part (LL) with a
ing a forearm. conventional encryption algorithm such as the Advanced

80
79
 LPF 2 LL Band

LPF 2

HPF 2 LH Band

Image

LPF 2 HL Band

HPF 2

HPF 2 HH Band
Row-Column
Row Processing Column Processing
Processing
(a) (b)

Figure 1: Diagrams of DWT image decomposition: (a) DWT image decomposition process (Single Level), (b) Analysis
DWT sub-band (Two Level)

(a) (b)

Figure 2: Example of decomposition by DWT: (a) Original image, (b) Single level decomposition

(a) (b) (c)

Figure 3: Pyramidal decomposition of an image: (a) Single level decomposition, (b) Two level decomposition, (c) Three
level decomposition

Encryption Standard (AES) [2] (the new standard in en- the whole picture.
cryption) or RSA (that was invented by R. Rivest, A.
Shamir, and L. Adleman) [15] will lead to scrambling over Taking into account the fact that the decomposition in
the JPEG2000 standard [1] is limited to 5, we will limit

81
80
 Original Selective Entropy
Image DWT Quantizer
Encryption Coder
Encrypted and Compressed
Image
Restored Selective Entropy
IDWT Dequantizer
Image Decryption Decoder

Figure 4: Schematic diagram of our new approach

R=1 R =2 R =3 R =4 R=5

LL 1 LL 2 LL 3 LL 4 LL 5

Figure 5: Illustration of the quality of the approximation of the image contained in the sub-band LL as a function of the
resolution

ourselves to level 5 so that our approach is compatible with the JPEG2000 algorithm [1]. Moreover, the increase in the
this norm. number of encrypted coefficients affects exponentially the
processing time (encryption-decryption).
From Figure 6, it can be seen that the quality of the in-
3.3 Advantages of our approach
terference introduced by the encryption of the sub-band LL
Rapidity: our new approach is very fast. In fact, only
is excellent up to level 3. But from the decomposition level
6.25% of all the information contained in the DWT
4, the reconstructed images are no longer illegible. Indeed,
matrix is encrypted. Consequently, our approach al-
it is possible to extract medical information of a general
lows a very significant gain in terms of processing
nature. This fact can be aggravated by attack by decompo-
time during the encryption-decryption process com-
sition and reconstitution of the encrypted image with can-
pared to the conventional approaches which consist
cellation of the encrypted coefficients in the sub-band LL.
in encrypting all the information contained in the im-
This attack is effective since it makes it possible to clearly
age.
visualize the texture of the image contained in the averages
and the high frequencies and to eliminate the interference Robustness: The security of our crypto-compression ap-
that exists at the level of the average intensities generated proach is equivalent to the robustness of the classi-
by the encryption of the coefficients of the sub-band LL. cal encryption algorithm that will be used to encrypt
From the results illustrated in Figures 6-7, it will be the coefficients of the sub-band LL. So if we use al-
noted that the greater the number of coefficients that ex- gorithms that are deemed safe such as RSA [15] or
ist in the encrypted sub-band, the more the resulting im- AES [2] (with the use of large keys), we can say that
age, whether after encryption or after attack, is illegible our algorithms are robust enough to guarantee a very
and hence more secure. However, encryption reduces the good degree of security.
spatial redundancy that exists between the different coeffi-
cients given the randomness of the data resulting from an
encryption operation. This affects the compression ratio of

82
81
 Encrypted Part

Original image LL 1 LL 2

LL 3 LL 4 LL 5

Figure 6: Reconstructed images after encryption of the sub-band LL for different resolutions

LL 2 LL 4

LL 3 LL 5

Encrypted images Attacked images Encrypted images Attacked images

Figure 7: Result of an attack throughby cancellation of the encrypted coefficients of the sub-band LL

Efficiency: The new crypto-compression approach pre- the image. In fact, information on the bone density
sented in this work is effective since it makes it pos- and the contours of the bone, on which any radio-
sible to prevent any attempt at correct diagnosis from logical diagnosis is based, is no longer present in the

83
82
 encrypted images (see Figure 7). This makes the ex- [7] Henry Ker-Chang Chang and Jiang-Long Liu. A
traction of useful information impossible unless it is linear quadtree compression scheme for image en-
possible to recover the initial value of the encrypted cryption. Signal Processing: Image Communication,
coefficients, which is impossible given the robust- 10(4):279–290, 1997.
ness of the conventional encryption algorithms.
[8] Raphael Grosbois, Diego Santa-Cruz, and Touradj
4 Conclusion Ebrahimi. New approach to JPEG2000 compliant re-
In this paper, a new approach to crypto-compression of gion of interest coding. In SPIE’s 45th annual meet-
medical images is presented, which consists in applying ing, Applications of Digital Image Processing XXIV,
partial encryption in the components of the DWT matrix. volume 4472, pages 267–275, August 2001.
The basic idea of this approach takes advantage of the fact
[9] Chih-Hsien Hsia, Jen-Shiun Chiang, and Jing-Ming
that after application of the DWT on an image, a large part
Guo. Discrete Wavelet Transforms - Biomedical Ap-
of the energy is concentrated in the sub-band LL.
plications, chapter Multiple Moving Objects Detec-
This approach is secure and compatible with the
tion and Tracking Using Discrete Wavelet Transform,
JPEG2000 standard. Its main advantages are speed and
pages 297–320. InTech, 2011.
efficiency. Indeed, it allows to considerably reduce the
processing time in the encryption-decryption process thus [10] J. A. Jaleel, S. Salim, and S. Archana. Textu-
ensuring the optimization of transmission and secure stor- ral features based computer aided diagnostic system
age of medical images. This is due to the fact that only for mammogram mass classification. In Interna-
6.25% of the DWT matrix coefficients are encrypted in tional Conference on Control, Instrumentation, Com-
this approach. Moreover, it is effective since it provides munication and Computational Technologies (ICCI-
encrypted images that are not interpretable medically even CCT’14), pages 806–811, July 2014.
after attack.
[11] D.W. Jones. Application of splay trees to data com-
References pression. Commun. ACM, 31(8):996–1007, August
[1] ISO/IEC 1.29.15444-1, ”JPEG2000 part i final com- 1988.
mittee version 1.0”, september 2004.
[12] Xiaobo Li, Jason Knipe, and Howard Cheng. Im-
[2] FIPS PUB 197, Advanced Encryption Stan- age compression and encryption using tree structures.
dard (AES), 2001. U.S. Department of Com- Pattern Recognition Letters, 18(11-13):1253–1259,
merce/National Institute of Standards and Technol- 1997.
ogy (NIST).
[13] Yossi Matias and Adi Shamir. Proceedings of Ad-
[3] M. K. Abdmouleh, H. Amri, A. Khalfallah, and M. S. vances in Cryptology (CRYPTO), chapter A Video
Bouhlel. A fast JPEG2000 based crypto-compression Scrambling Technique Based On Space Filling
algorithm: Application to the security for transmis- Curves, pages 398–417. Springer Berlin Heidelberg,
sion of medical images. In International Workshop Berlin, Heidelberg, 1988.
on Soft Computing Applications (SOFA’16), Arad–
Romania, August 24–26 2016. [14] W. B. Pennebaker and J. L. Mitchell. JPEG Still Im-
age Data Compression Standard. Van Nostrand Rein-
[4] Mohammed Bahoura and Hassan Ezzaidi. FPGA- hold, cop, New-York, 1993.
Implementation of Discrete Wavelet Transform with
Application to Signal Denoising. Circuits, Systems, [15] R. L. Rivest, A. Shamir, and L. Adleman. A method
and Signal Processing, 31(3):987–1015, 2012. for obtaining digital signatures and public-key cryp-
tosystems. Commun. ACM, 21(2):120–126, February
[5] Med Salim Bouhlel, Fahmi Kammoun, and Eric Gar- 1978.
cia. An efficient DCT-based crypto-compression
scheme for a secure and authentic medical image [16] M. Venkateshwar Rao and V. Bhagya Raju. Im-
transmission. Journal of Testing and Evaluation for age Resolution Enhancement Technique Using Lift-
Applied Sciences and Engineering, 34(6):459–463, ing Wavelet and Discrete Wavelet Transforms, pages
2006. 235–239. Springer Singapore, Singapore, 2016.
[6] Chin-Chen Chang, Min-Shian Hwang, and Tung- [17] Chuntao Wang, Jiangqun Ni, and Qiong Huang.
Shou Chen. A new encryption algorithm for im- A new encryption-then-compression algorithm using
age cryptosystems. Journal of Systems and Software, the rate-distortion optimization. Signal Processing:
58(2):83–91, 2001. Image Communication, 39, Part A:141–150, 2015.

84
83