High Capacity Robust Medical Image Data Hiding using CDCS with Integrity Checking

ACEEE Int. J. on Signal & Image Processing, Vol. 02, No. 01, Jan 2011

High Capacity Robust Medical Image Data Hiding
using CDCS with Integrity Checking
Sunita V. Dhavale1, and Suresh N. Mali2
1
Department of Information Technology, Marathwada Mitra Mandal’s College of Engineering, Pune, Maharashtra-411052,
India.
Email: sunitadhavale75@rediffmail.com
2
Department of Computer Engineering, Vishwakarma Institute of Technology, Pune, Maharashtra-411037, India.
Email: snmali@rediffmail.com

Abstract—While transferring electronic patient report (EPR) EPR data but also increases the perceptual quality of the
data along with corresponding medical images over network, image for the given data hiding capacity. Before
confidentiality must be assured. This can be achieved by embedding this encoded EPR data in medical image et al.
embedding EPR data in corresponding medical image itself. [3], high imperceptibility as well as robustness is achieved
However, as the size of EPR increases, security and by adaptively selecting the area of an image in which to
robustness of the embedded information becomes major issue
to monitor. Also checking the integrity of this embedded data
hide data using energy thresholding method et al. [2].
must be needed in order to assure that retrieved EPR data is Further, one must also guarantee that the region in which
original and not manipulated by different types of attacks. we have embedded sensitive and confidential EPR data is
This paper proposes high capacity, robust secured blind data not tampered by any malicious manipulations et al. [4].
hiding technique in Discrete Cosine Transform (DCT) domain Thus there is a need for integrity checking that must assure
along with integrity checking. A new coding technique called both EPR data and image has not been modified by
Class Dependent Coding Scheme (CDCS) is used to increase unauthorized person. So secure hash can be calculated over
the embedding capacity. High imperceptibility is achieved by this sensitive region and these hash bits can be embedded
adaptively selecting the efficient DCT blocks. Even a slight in diagnostically less important region et al. [4] like border
modification of stego image in embedded region as well as in
ROI (Region of Interest) can be detected at receiver so to
(black surrounding) or very low frequency region outside
confirm that attack has been done. The embedding scheme ROI of medical image using fragile spatial domain
also takes care of ROI which is diagnostically important part techniques like simple LSB substitution method. So even
of the medical images and generates security key simple cutting or cropping the image at border region can
automatically. Experimental results show that the proposed loose these embedded hash bits and attack can be
scheme exhibits high imperceptibility as well as low confirmed. The stego key that is automatically generated
perceptual variations in Stego-images. Security and based on embedding factors like randomization,
robustness have been tested against various image redundancy, interleaving, energy thresholding and JPEG
manipulation attacks. quality factor provides multiple levels of security.
Index Terms—Medical image, EPR, Data hiding, Stego image,
ROI, Robustness, Security, Integrity Check. II. PROPOSED SYSTEM
The proposed embedding scheme consists of text
I. INTRODUCTION processing phase and image processing phase as shown in
Telemedicine application requires transferring Electronic Fig. 1 Text processing phase makes the stream of EPR
Patient Report (EPR) data and corresponding medical encoded bits ready for embedding, whereas, image
images over network for further diagnostic purpose. While processing phase embeds these bits into the corresponding
sharing medical images and EPR in telemedicine et al. [1], medical image. At the time of execution of these phases the
we need to protect of both medical images and EPR data as embedding parameters (r, n, w1, w2, seed, QF, x1, y1, x2,
well as to save as much space as possible in order to reduce y2) are provided as an input that gets reflected in
the cost of storage and increase the speed of transmission. automatically generated embedding key.
Both these goals can be achieved by effective embedding A. CDCS
of EPR in corresponding medical image itself.
Proposed system assigns fixed codes in CDCS to each
Aim of proposed EPR data hiding is to increase data
character by considering their probability of frequency of
hiding capacity without perceptual degradation of the
occurrences as shown in Fig. 2 The EPR characters are
medical image along with integrity checking et al. [2, 7]. A
then categorized in three different nonoverlapping special
new CDCS coding scheme has been proposed in this paper
classes as Class-A (most frequently appearing character
that will not only reduces the number of bits to represent
set), Class-B (Average frequently appearing) and Class-C
(Less frequently appearing characters). Further, the number
Corresponding Author: Mrs. Sunita V. Dhavale, M.E.(CSE-IT), of bits needed to represent each character in the respective
MMCOE, Pune-411052, India.
classes is achieved by assuming only capital letters,
13
© 2011 ACEEE
DOI: 01.IJSIP.02.01.175


alphanumeric and few special characters. Based on
Huffman encoding, the variable length class code (CC)
have been designed to represent each class as given in
Table 1.

Figure 2. Probability of occurrences for EPR characters

Table 1: CDCS: Class CODE with Fixed Code within Each Class
Class A Class B Class C 4-Bit Code
CC=1 CC=00 CC=01
(1bit) (2bits) (2bits)
Blank M 0 0000
. U 1 0001
Figure 1. Proposed System: Multilevel security
E G 2 0010
Any character in each Class will be represented by
T Y 3 0011
only 4 bits prefixed by CC (1-bit or 2-bit). Therefore, CC
along with 4 bit character code can distinguish 48 different A P 4 0100
characters as shown in Table 1, which are sufficient to O W 5 0101
represent any EPR. Huffman coding is complex and also N B 6 0110
assigns codes with more than 32 bits for non repeating
R V 7 0111
characters whereas ASCII gives fixed length codes for the
characters. The proposed CDCS combines the advantages I K 8 1000
of both fixed length and variable length coding to get less S X 9 1001
number of bits to represent same information compared to
H J ( 1010
fixed 7-bit ASCII codes.
If N1, N2 and N3 are the total number of characters D Q ) 1011
belonging to Class-A, Class-B and Class-C respectively, L Z = 1100
Total number of bits to be embedded is given by, F , * 1101

m = ( N1 + 2 N 2 + 2 N 3 ) + 4 h C - % 1110
(1)
: _ + 1111

Where,
h = N1 + N 2 + N 3 , i.e. total number of
C. Energy Thresholding
characters in EPR file.
Percentage Bit Saving (PBS) is given by, A sequence of lower and middle frequency non-zero
⎡ ⎛ m ⎞⎤ quantized DCT coefficients of randomly generated valid
⎜
PBS = ⎢1 − ⎜ ⎟⎥ ×100 %
⎟ blocks are used to embed final processed bits. After
⎣ ⎝ 7 h ⎠⎦ (2) dividing the image into 8 x 8 non overlapping blocks two
dimensional DCT is computed for each block along with its
B. Redundancy and Interleaving
energy. The blocks having energy greater than the
Robustness against various attacks can be
achieved by adding redundancy for each bit prior to threshold energy will only be considered for embedding.
embedding et al. [7, 8]. Interleaving of bits will disperse Energy threshold (Et) is calculated as,
subsequent bits from each other throughout the image.
Hence, even if any block of Stego-image undergoes with Et = w ⋅ MVE
ˆ
(3)
attack, EPR bits can be successfully recovered from other where, ŵ = Energy threshold factor and MVE = Mean
blocks. CDCS encryption along with specified number of
Value of Energy given by,
redundancy bits added (r) and number of interleaving bits
(n) provides two security levels 1 and 2 as shown in Fig. 1.

14
© 2011 ACEEE
DOI: 01.IJSIP.02.01.175


1 z the EPR bits of information, the Stego-images gives PSNR
MVE =
z
∑E
k =1
k value more than 40dB.
(4) G. EPR Data Retrieval
where, z = Number of 8 x 8 non-overlapping blocks of the Fig. 3 gives automatically generated embedding key
image and Ek= Energy of kth block which is given by, (AGEK) during the process of embedding along with the
7 7 2 respective security levels. This embedding key has to be
Ek = ∑∑
i =1 j =1
Cij shared with the receiver through a secret channel. The
embedding key has 52 bits (n=4, r=4, seed= 16, ŵ (both w1
(5)
where Cij = Two dimensional DCT coefficients. and w2) = 8, QF = 8, x1-y1= 6 and x2-y2 =6). It is difficult
to break the key for particular combination decided by the
Classify the DCT blocks by define two different energy
embedding algorithm. The extraction algorithm consists of
thresholds w1 and w2 using (3), in such a way that, w1>> all the image processing steps that are carried out at the
w2. time of embedding final processed bits. Recalculate secure
As blocks having more energy can embed hash of those embedded VB’s as well as ROI blocks.
information bits with minimal distortion et al. [7], all Extract encrypted hash bits from LSB’s of randomly
blocks having energy more than Et1 (decided by w1) will selected pixels of LFBs and decrypt it.
only be considered for embedding and treated as Valid
Blocks (VBs) while blocks having energy lesser than Et2
(decided by w2) are selected as very Low Frequency DCT
Blocks (LFBs) where a small size hash bits can be
embedded safely without causing more distortion to
medical image.
ˆ
D. Adaptive Energy Threshold Factor ( w )
There is always a trade-off between ŵ and number of
VBs. As the value of ŵ increases, we get less number of
VBs. In this scheme, one can adaptively modify the value
of ŵ by monitoring the PSNR of reconstructed image with
respect to given value of PSNR as shown in Fig. 1. This Figure 3. Bit format of AGEK
embedding parameter ŵ gives security level 3.
Compare both hash bits if equal, then image is
E. Randomization, ROI and Quantization authenticated and we can retrieve embedded EPR data
Security level 4 is achieved by randomly selecting the safely and EPR information can be reconstructed using
VBs. The random number generator based on a seed is CDCS.
used to select random VBs. Diagnostically important area
of medical image can be defined by specifying the diagonal III. EXPERIMENTAL RESULTS
indices (X1, Y1) and (X2, Y2). The VBs coming in the We used grayscale 512 x 512 L-Spine (Lumbar Spine)
vicinity of ROI will not be considered for embedding. medical images as shown in following Fig. 4a for our
These specified indices give security level 5. The randomly experiments. Experimentation is performed to check
selected VBs are then quantized for the given value of QF. tamper detection capability of the system under various
After the process of quantization the non-zero predefined attacks (intentional and unintentional). Fig. 4b shows
DCT coefficients are considered for embedding the data. corresponding stego image, when 4 bits of EPR are
The embedding parameter QF gives security level 6. embedded per 8 x 8 DCT block. The locations chosen for
F. Embedding and Reconstruction each VB block are (2,2), (3,0), (0,3) and (0,2). The 320 bit
The embedding is carried out by suitably modifying the encrypted hash code is embedded in LSBs of 8 pixels per
DCT coefficients of the valid blocks finally selected after LFB block. The quality factor QF=50 chosen.
the process of quantization. If the bit is logically ‘zero’, the For a given Et1 (decided by w1) and Et2 (decided by w2)
coefficient is rounded to ‘even’ number, otherwise to ‘odd’ in Table 2, Fig. 5a shows corresponding VB Blocks or
number. Finally stego-image is reconstructed by applying region of images where EPR data is embedded while Fig.
inverse DCT and combining all 8 x 8 image blocks. Take 5b shows corresponding LFB Blocks or region of images
secure hash of Embedded VB’s as well as ROI blocks and where Hash data is embedded. Here w2 is selected in such
encrypt and embed these hash bits into LSB’s of randomly a way that all LFBs fall generally in border black region.
selected pixels of LFBs. This hash embedding stage used The PSNR observed for the stego images is above 40db.
for integrity checking of both embedded data and ROI Any attack like small pixel stains made in VB region or
gives security level 7. EPR embedded region, which could not be visible by
Fig. 1 shows all the steps of the proposed embedding normal eye is detected at receiver side successfully.
scheme. The experimentation shows that after embedding
15
© 2011 ACEEE
DOI: 01.IJSIP.02.01.175


A. Effect of CDCS embedded hash bits to guarantee the integrity of the
Table 3 shows comparison of CDCS and ASCII codes for medical images transmitted. Proposed CDCS can be used
various number of EPR characters. It can be observed that as effective coding scheme for EPR data hiding in medical
increase in data hiding capacity is the result of PBS with images which increases the embedding capacity and
proposed CDCS. provides better perceptual quality of Stego-image.
Effective use of redundancy and interleaving enhances the
Table 2. Et1 and Et2 choices
robustness of the scheme against various attacks like JPEG
Medical EPR w1 w2 PSNR(dB) PSNR(dB) after compression, image tampering and image manipulation.
Image Bits attack Seven layered security achieved due to redundancy,
L-Spine 2534 3705 10 44.65 40.43
interleaving, energy thresholding, random-ization, ROI
Shoulder 2098 1977 10 47.14 46.61
quantization and Hash embedding stage makes the
proposed system most secured.
Also further the stego key generated can be further
encrypted using any public key encryption algorithm and
can be transmitted in secure way to the receiver along with
stego image. Also even slight modification to embedded
region and ROI part can be detected unambiguously at
receiver side. Thus the proposed system can effectively be
used for high volume of EPR data hiding in medical images
Figure 4. Original and Stego L-Spine Medical Image with reasonable robustness and security.

REFERENCES
[1] Gonzalo Alvarez1, Shujun Li and Luis Hernandez, “Analysis
of security problems in a medical image encryption system,”
Computers in Biology and Medicine, vol. 37 (2007) 424–
427.
[2] M. Fallahpour and M. H. Sedaaghi, ”High capacity lossless
Figure 5. VB and LFB Region data hiding based on histogram modification”, IEICE
Electron. Express, Vol. 4, No. 7 (2007) 205–210
Table 3: Capacity Performance of CDCS over ASCII [3] J. Zain and Malcolm Clarke, ”Security in Telemedicine:
EPR ASCII bits CDCS bits PBS Issues in Watermarking Medical Images”, 3rd International
characters (%) Conference : Science of Electronic, Technologies of
506 3542 2814 20.55 Information and Telecommunications (2005)
[4] B. Planitz, A. Maeder, “Medical Image Watermarking: A
584 4088 3244 20.65 Study on Image Degradation”, Proc. Australian Pattern
Recognition Society Workshop on Digital Image Computing,
B. Perceptual Transparency WDIC 2005, Brisbane, Australia (2005)
[5] G. S. Pavlopoulos, D.Koutsouris, ”Multiple Image
Fig. 4a and Fig. 4b shows original and Stego 512 x 512 ‘L- Watermarking Applied to Health Information Management”,
Spine’ images respectively. The locations chosen per block IEEE Transactions on Information Technology in
are at (2,2), (3,0), (0,3) and (0,2) to embed 270 EPR Biomedicine, vol. 10.4 ( 2006) 722 – 732
characters. The ‘QF =70’ and the EPR data has been [6] K. A. Navas, S. A. Thampy, and M. Sasikumar, “EPR
embedded with ‘w1 =0.5’. The PSNR observed for the Hiding in medical images for telemedicine,” International
Stego-image is 47.5982. Proposed scheme shows PSNR Journal of Biomedical Sciences Volume 3.1 (2008) 44– 47
greater than 38 dB for embedding up to 3000 EPR bits in [7] K. Solanki, N. Jacobsen, U. Madhow, B.S.Manjunath and S.
‘L-spine’ images. Chandrashekhar, ”Robust Image-Adaptive Data hiding using
Erasure and Error Correction,” IEEE Transactions on image
C. Robustness Test processing, Volume 13, (2004) 1627–1639.
The proposed system is robust and gives very less Bit Error [8] S. N. Mali and R. M. Jalnekar., ”Imperceptible and Robust
Data Hiding using Steganography Against Image
Rate (BER) against JPEG compression attack, image Manipulation,” International Journal of Emerging
tampering attack, image manipulation attack and change in Technologies and Applications in Engineering, Technology
contrast value. and Sciences, (IJ-ETA-ETS) (2008) 84–91.
[9] G. J. Simmons, ”The prisoners’ problem and subliminal
CONCLUSIONS channel”, in Advances in Cryptology. Proceedings of Crypto
83 (D. Chaum, ed.), Plenum Press (1984) 51–67.
In this paper, we proposed a method for the secure [10] Neil F. Johnson and S. Jajodia, ”Exploring Steganography.
transmission of medical images by using both DCT and Seeing Unseen”, IEEE Computer, vol. 31.2 (1998) 26 – 34.
LSB substitution along with new CDCS encoding scheme [11] Min. Wu , ”Joint Security and Robustness Enhancement for
for EPR data in order to increase hiding capacity. First we Quantization Embedding”, IEEE Transactions 0-7803-7750-
embedded EPR data into the medical images and then we 8/03(2003)483–486.

16
© 2011 ACEEE
DOI: 01.IJSIP.02.01.175

High Capacity Robust Medical Image Data Hiding using CDCS with Integrity Checking

More Related Content

What's hot (20)

Viewers also liked (8)

Similar to High Capacity Robust Medical Image Data Hiding using CDCS with Integrity Checking (20)

More from IDES Editor (20)

Recently uploaded (20)

High Capacity Robust Medical Image Data Hiding using CDCS with Integrity Checking