IJCSI International Journal of Computer Science Issues, Vol.

11, Issue 5, No 1, September 2014

ISSN (Print): 1694-0814 | ISSN (Online): 1694-0784
www.IJCSI.org 31

Block Based Medical Image Watermarking Technique for

Tamper Detection and Recovery
Eswaraiah Rayachoti1 and Sreenivasa Reddy Edara2
1
CSE, VVIT
Guntur, Andhra Pradesh, India
2
CSE, Acharya Nagarjuna University
Guntur, Andhra Pradesh, India

Abstract embedded directly into host image while data is inserted

In this paper, we propose a novel fragile block based medical into transformed host image in frequency domain
image watermarking technique for embedding data of patient watermarking techniques [8, 9, 14]. Another
into medical image, verifying the integrity of ROI (Region of categorization of watermarking technique is reversible
Interest), detecting the tampered blocks inside ROI and
and irreversible. In reversible watermarking technique [9,
recovering original ROI with less size authentication and
recovery data and with simple mathematical calculations. In the
14, 17, 18], the host image can be recovered exactly at
proposed method, the medical image is divided into three receiver side from watermarked image. Accurate recovery
regions called ROI, RONI (Region of Non Interest) and border of host image is not possible in case of irreversible
pixels. Later, authentication data of ROI and Electronic Patient watermarking techniques [8, 15]. Reversible
Record (EPR) are compressed using Run Length Encoding watermarking is more suitable for medical images [2].
(RLE) technique and then embedded into ROI. Recovery
information of ROI is embedded inside RONI and information Four types of watermarking methods are developed to
of ROI is embedded inside border pixels. Results of protect digital images: Robust watermarking [3], Fragile
experiments conducted on several medical images reveal that Watermarking [4], Semi-Fragile Watermarking [5] and
proposed method produces high quality watermarked medical
Hybrid Watermarking [6]. Robust watermarking methods
images, identifies tampered areas inside ROI of watermarked
medical images and recovers the original ROI. are used for copyright protection of digital images as it is
Keywords: Watermarking, ROI, RONI, RLE, Tamper Detection, difficult to remove robust watermarks from digital
Recovery. images. Robust watermarks sustain intentional or
unintentional attacks like scaling, compression, cropping
and so on. Fragile watermarking techniques are best for
1. Introduction checking authentication of digital images. Any
modification or tampering removes fragile watermark
Exchange of medical images between hospitals located at from watermarked image. So, absence of watermark
remote places has become a natural practice of modern indicates that image has been tampered. Semi-Fragile
times. This exchange of medical images inflicts two watermarks survive only unintentional attacks. Hybrid
restraints for the medical images: the information has not watermarks are the amalgamation of fragile and robust
been changed by unauthorized users and there should be watermarks. Here, robust watermarks are used for privacy
evidence that the information belongs to the correct control and fragile watermarks are used for the integrity
patient [1]. On the other hand transmission of patient data control of the digital image.
and his medical image separately through commercial
networks like internet results in excessive transmission Most of the medical images contain two parts called ROI
time and cost. Watermarking is one of the techniques and RONI. From diagnosis point of view ROI part is more
used to deal with the above two concerns. important. Care should be taken while hiding data into
ROI part so that visual quality will not be degraded. At
Watermarking techniques have been classified into two the same time any tampering to ROI has to be identified
categories namely spatial domain and frequency domain. and the original ROI has to be recovered in order to avoid
This classification is based on the medium used for misdiagnosis and retransmission of medical image. The
concealing the data in an image. In spatial domain recovery data of ROI is generally embedded into RONI
watermarking techniques [7, 10, 11, 13], data is [10, 11, 13, 14, 15, 16, 19]. When any tamper is detected

Copyright (c) 2014 International Journal of Computer Science Issues. All Rights Reserved.
IJCSI International Journal of Computer Science Issues, Vol. 11, Issue 5, No 1, September 2014
ISSN (Print): 1694-0814 | ISSN (Online): 1694-0784
www.IJCSI.org 32

inside ROI of received watermarked medical image then B are tampered then it is not possible to recover original
the tampered area of ROI is replaced with the recovery image, 2) this method is not using any authentication data
data embedded inside RONI. for the entire medical image to check directly whether the
image is tampered. So, all blocks in the image have to be
In this paper, we are proposing a novel block based fragile checked one after other to detect the presence of tampers.
medical image watermarking technique to achieve the This checking process leads to wastage of time when the
following objectives. image is not tampered, 3) there is no provision for
embedding EPR of patient into the medical image.
1. Identifying the presence of tampers inside ROI.
2. Recovering the original ROI when it is tampered. Wu et al. [8] developed two block based methods. In the
3. Detecting tampers inside ROI and recovering original second method, JPEG bit-string of the selected ROI is
ROI using minimal size authentication and recovery generated and then is divided into fixed length segments.
data. Later, the medical image is divided into blocks and then
4. Avoiding the process of checking ROI of the hash bits are calculated for each block excluding the block
watermarked medical image for the presence of with ROI. This hash bits are used as authentication data
tampers when the ROI is not tampered. of the blocks. In each block of image, hash bits of the
5. Embedding EPR of patient into the medical image. block and one segment of JPEG bit-string of ROI are both
embedded using robust additive watermarking technique.
The rest of the paper is organized as follows. Section 2 Then all blocks are combined to get watermarked medical
covers review literature, proposed method is explained in image. At receiver’s end, the watermarked medical image
section 3; results are illustrated in section 4 and finally is divided into blocks as done in embedding procedure.
conclusion in section 5. From each block, hash bits of the block and a segment of
JPEG bit-string are both extracted. For each block, hash
bits are calculated and then compared with the extracted
2. Literature Review hash bits to check whether the block is tampered or not. If
the block with ROI is identified as tampered then the
Number of watermarking techniques has been developed JPEG bit-string segments extracted from all blocks are
for detecting tampers in the ROI or in the entire used to recover the ROI. Disadvantages of this method
watermarked medical image and recovering the original are: 1) it is not possible to get original ROI as JPEG bit-
ROI or the entire medical image. Zain et al. [7] proposed string of ROI is used to recover ROI when it is tampered
a block based scheme, the medical image is segmented with, 2) this method requires more number of calculations
into 8×8 blocks and then a mapping is established to generate recovery data of ROI and embedding it into all
between the blocks for embedding the recovery blocks of medical image, 3) the size of authentication data
information of each block into its corresponding mapped is large; for each block 150 bits are used, 4) there is no
block. Later, each block is further divided into four sub provision for embedding EPR of patient into the medical
blocks of 4×4 size each and then a 9-bit watermark is image.
generated for each sub block. The generated 9 bit
watermark of each sub block is embedded into LSBs of Chiang et al. [9] proposed two block based methods based
first 9 pixels of the sub block in the corresponding on symmetric key cryptosystem and modified difference
mapped block. At receiver’s end, the watermarked expansion (DE) technique. The first method has the
medical image is divided into blocks of 8×8 size and then ability to recover the whole medical image, where as the
the mapping between the blocks is calculated as done in second method has the ability to recover only ROI of
embedding procedure. Later, each block is further divided medical image. In the first method, the medical image is
into four sub blocks of 4×4 size and then a 2-level divided into 4×4 size blocks and then average of each
detection scheme is applied for detecting tampered blocks. block is calculated. Later, the averages of all blocks are
This 2-level detection scheme identifies tampered blocks. concatenated and then encrypted using two symmetric
Where Level-1 detection is applied on sub-blocks of keys k1 and k2 in order to increase the degree of security.
blocks and level-2 detection is applied on blocks. When a Then, Haar wavelet transform is applied on all blocks to
tampered block is detected, the corresponding mapped identify smooth blocks. The encrypted averages of all the
block is identified and then recovery data embedded in blocks are embedded in the identified smooth blocks. At
mapped block is extracted. This recovery data is used to the receiver’s end, the embedded data is extracted from
replace the pixels in tampered block. Major drawbacks of watermarked image and then decrypted using the keys k1
this method are: 1) if both block A and its mapped block and k2 to get the averages of all blocks. Later, averages

Copyright (c) 2014 International Journal of Computer Science Issues. All Rights Reserved.
IJCSI International Journal of Computer Science Issues, Vol. 11, Issue 5, No 1, September 2014
ISSN (Print): 1694-0814 | ISSN (Online): 1694-0784
www.IJCSI.org 33

are calculated for all blocks and then compared with Memon et al. [12] implemented a hybrid watermarking
extracted averages to detect tampered blocks. When a method. In this method, the medical image is segmented
tampered block is detected then the pixels in tampered into ROI and RONI. Then, a fragile watermark is
block are replaced with the extracted average of that embedded into LSBs of ROI. RONI is divided into blocks
block. The second method is same as the first method of size N×N and then a location map indicating
except that the bits of pixels in blocks of ROI are embeddable blocks is generated. A robust watermark is
embedded instead of averages of all blocks in entire embedded into embeddable blocks of RONI using Integer
image. Pitfalls of these schemes are: 1) in the second Wavelet Transform (IWT). Later, the location map is
method the size of authentication and recovery data is embedded into LL3 of each block using LSB substitution
large; 128 bits for each block in ROI, 2) the two methods method. Finally, ROI and RONI are combined to get
require more time for embedding data into medical image watermarked image. At receiver’s end, the watermarked
as all blocks of the medical image have to be transformed medical image is segmented into ROI and RONI. Then,
into frequency domain and then smooth blocks have to be the robust watermark is extracted from RONI and is used
identified for embedding data, 3) the two methods are not for checking authentication of image. Fragile watermark
using any authentication data for the entire ROI or the is extracted from ROI and checked visually to know
entire image to check directly whether the ROI or the presence of tampers inside ROI. Two disadvantages of
entire image is tampered. So, all blocks in the ROI or in this method are: 1) there is no specification of how the
the entire image have to be checked one after another to original ROI is recovered when the ROI is tampered, 2)
detect the presence of tampers. This checking process the time complexity of this method is more as it has to
leads to wastage of time when the image is not tampered, generate location map before embedding data.
4) there is no provision for embedding EPR of patient into
the medical image. Agung et al. [13] developed a reversible method for
medical images whose ROI size is more compared to size
Liew et al. [10, 11] developed two reversible block based of RONI. In this method, the original LSBs of all pixels in
methods. In the first method, the medical image is medical image are collected and then LSB in each pixel is
segmented into two regions: ROI and RONI. Later, ROI set to zero. Later, the medical image is segmented into
and RONI are divided into non overlapping blocks of size ROI and RONI regions. Then, ROI and RONI are divided
8×8 and 6×6 respectively. Then, a mapping is formed into blocks of size 6×6 and 6×1 respectively. A mapping
between blocks of ROI to embed recovery information of is formed between blocks of ROI for storing recovery
each block into its mapped block. Each block in ROI is information of each ROI block into its mapped ROI block.
mapped to a block in RONI. This mapping is used to The removed original LSBs are compressed using RLE
embed LSBs of pixels in a ROI block into its mapped technique and then embedded into 2 LSBs of 6×1 blocks
RONI block. Then, the method implemented by Zain et in RONI. At receiver’s end, the watermarked medical
al. [7] is applied only on ROI part of the medical image image is segmented into ROI and RONI as done in
for detecting tampers inside ROI and recovering original embedding procedure. Then, the method proposed by Zain
ROI. The LSBs of pixels inside ROI are replaced with its [7] is applied only on ROI part to detect tampers inside
original bits that were stored inside RONI to make the ROI and recover original ROI. The original LSBs that
scheme reversible. Second method is same as first method were embedded in RONI are extracted and then restored
except that the removed LSBs of pixels in blocks of ROI to their positions to get the original medical image. This
are compressed using Run Length Encoding technique method has the same drawbacks as with methods
before embedding into RONI blocks. Drawbacks of the proposed by Liew et al. [10, 11].
two methods are: 1) if both block A and its mapped block
B inside ROI are tampered then it is not possible to Qershi et al. [14] developed a reversible ROI based
recover original ROI, 2) the two methods are not using watermarking scheme. At sender’s end, the medical
any authentication data for the entire ROI to check image is segmented into ROI and RONI. Later, data of
directly whether the ROI is tampered. So, all blocks in the patient and hash value of ROI are both embedded into
ROI have to be checked one after another to detect the ROI using technique developed by Gou et al. Compressed
presence of tampers. This checking process leads to form of ROI, average values of blocks inside ROI,
wastage of time when the ROI is not tampered, 3) there is embedding map for ROI, embedding map for RONI and
no provision for embedding EPR of patient into the LSBs of pixels in a secrete area of RONI are embedded
medical image. into RONI using the technique of Tian. Finally,
information of ROI is embedded into LSBs of pixels in
secrete area. At receiver’s end, ROI information is

Copyright (c) 2014 International Journal of Computer Science Issues. All Rights Reserved.
IJCSI International Journal of Computer Science Issues, Vol. 11, Issue 5, No 1, September 2014
ISSN (Print): 1694-0814 | ISSN (Online): 1694-0784
www.IJCSI.org 34

extracted from secrete area and is used to identify ROI technique. At receiver’s end, ROI information is extracted
and RONI regions. From the identified RONI region from border pixels and is used to identify ROI and RONI
compressed form of ROI, average values of blocks inside regions. Compressed form of ROI, average intensities of
ROI, embedding map of ROI, embedding map of RONI blocks in ROI and location map of ROI are extracted from
and LSB of pixels in secrete area are extracted. Using the the identified RONI region. Using the extracted location
extracted location map of ROI, patient’s data and hash map of ROI, patient’s data and hash value of ROI are
value of ROI are extracted from ROI. Then, hash value of extracted from ROI. The procedure for detecting tampered
ROI is calculated and compared with extracted hash blocks and recovering ROI is same as in [14]. Two
value. If there is a mismatch between the two hash values disadvantages of this method are: 1) use of compressed
then the ROI is divided into 16×16 blocks. For each form of ROI as recovery information for the ROI, 2)
block, the average value is calculated and compared with applicable to only images whose size is at least 512×512.
the corresponding average value in the extracted average
values. If they are not equal then the block is marked as Deng et al. [17] developed a region-based tampering
tampered and replaced by the corresponding block of the detection and recovering method based on reversible
compressed form of ROI. Two disadvantages of this watermarking and quad-tree decomposition. In this
method are: 1) extracting the embedded data from RONI method, original image is divided into blocks with high
without knowing the embedding map of RONI, 2) use of homogeneity using quad-tree decomposition and then a
compressed form of ROI as recovery data for the ROI. recovery feature is calculated for each block using linear
interpolation of pixels. The recovery features of all blocks
Qershi et al. [15] proposed a scheme based on two are embedded as first watermark using invertible integer
dimensional difference expansion (2D-DE). At sender’s transformation. Quad-tree information as second layer
end, the medical image is divided into three regions: ROI watermark is embedded using LSB replacement. In the
pixels, RONI pixels and border pixels. Later, the authentication phase, the embedded watermark is
concatenation of patient’s data, hash value of ROI, bits of extracted and the original image is recovered. The similar
pixels inside ROI and LSBs of border pixels is linear interpolation technique is utilized to get each
compressed using Huffman coding and then embedded block's feature. The tampering detection and localization
into RONI using 2D-DE technique. This embedding can be achieved through comparing the extracted feature
generates a location map which will be concatenated with with the recomputed one. The extracted feature can be
information of ROI and then embedded into LSBs of used to recover those tampered regions with high
border pixels. At receiver’s end, from border pixels in the similarity to their original state. One drawback of this
watermarked medical image both information of ROI and scheme is exact original image cannot be recovered when
location map are extracted. Using this ROI information, it is tampered.
ROI and RONI are identified. The extracted location map
is used to extract patient’s data, hash value of ROI, bits of
pixels inside ROI and LSBs of border pixels from RONI. 3. Proposed Method
The process for detecting tampered blocks is same as the
one used in [14]. Each tampered block is replaced by the To achieve the above mentioned objectives, we propose a
corresponding block of pixels in the extracted ROI. The medical image watermarking technique in this paper.
LSBs of border pixels are replaced using the extracted
LSBs from RONI. A major drawback of this scheme is it 3.1 Division of Medical Image
is applicable to only the medical images whose ROI size
In a medical image, the ROI is the most important part
is very less (up to 12% of size of entire image).
for making diagnosis. A medical image may contain
several disjoint ROI areas and may be in different shapes.
Qershi et al. [16] developed a hybrid ROI-based method.
The ROI parts are marked by a physician or by a clinician
At sender’s end, the medical image is divided into three
interactively. Each ROI area is represented by an
regions: ROI, RONI and border pixels. Later, patient’s
enclosing polygon. The enclosing polygon is
data and hash value of ROI are embedded inside ROI
characterized by the number of vertices and their
using modified DE technique. The ROI location map
coordinates. In proposed method, the medical image is
along with compressed form of ROI and average
segmented into three regions of pixels: ROI pixels, RONI
intensities of blocks inside ROI are then embedded into
pixels and border pixels as shown in Fig. 1. In present
RONI using DWT technique. Then, size of watermark
work, we use medical images containing a single ROI.
that is inserted into RONI and ROI information are
The proposed method can also be used with medical
embedded inside border pixels using the same DWT

Copyright (c) 2014 International Journal of Computer Science Issues. All Rights Reserved.
IJCSI International Journal of Computer Science Issues, Vol. 11, Issue 5, No 1, September 2014
ISSN (Print): 1694-0814 | ISSN (Online): 1694-0784
www.IJCSI.org 35

images containing multiple ROI areas. In this method, the

outer three lines of pixels in the image are indicated as BRONI k BROI mod N b 1 (1)
border.
where Nb is the number of blocks in ROI, BRONI is block
number in RONI, k is a secrete key and is a prime number
between 1 and Nb, BROI is block number in ROI. After
mapping each ROI block to a RONI block, the average
value of each ROI block is embedded inside the
corresponding mapped RONI block.

Now, the detailed embedding algorithm is explained as

follows.
Fig. 1 Division of medical image into three regions.
3.5 Embedding Algorithm

3.2 Hash value of ROI 1. Segment the original medical image into three
regions: ROI pixels, RONI pixels and border pixels.
After selecting the ROI, the hash value of the ROI is 2. Calculate hash value (h1) of ROI using MD5.
calculated using the cryptographic hash function MD5. 3. Divide the pixels inside ROI into non overlapping
This function generates a unique code for any input and is blocks of size 4×4 each.
a one way function. Determining the input from the code 4. For each ROI block, calculate average value and use
generated by MD5 is not possible. The calculated hash it as authentication and recovery data of that block.
value of ROI is used to authenticate ROI. 5. Collect Least Significant Bits of all pixels inside ROI
and denote this collection as B.
3.3 Run Length Encoding 6. Represent the characters in EPR of patient using
ASCII code and then get binary equivalent of it, E.
After calculating hash value of ROI, the ROI is divided 7. Generate watermark w by concatenating h1, B and E.
into non overlapping blocks of size 4×4 and then average 8. Compress watermark w using RLE compression
value is calculated for each block. A watermark is technique to generate wcomp.
generated by concatenating hash value of ROI, LSBs of 9. Encrypt the watermark wcomp using a secret key k1.
pixels inside ROI and EPR of patient. This generated 10. Embed the bits of encrypted watermark into LSBs of
watermark is compressed using RLE. RLE is a simple pixels inside ROI.
lossless compression technique and is used to reduce the 11. Divide RONI into non overlapping blocks of size 3×3
size of watermark. Original data can be reconstructed each.
exactly from the compressed data. In this technique, if a 12. Assuming that the number of blocks in ROI is less
bit is repeating for number of times in sequence then that than the number of blocks in RONI, map each block
bit sequence is replaced by a count value and the bit. As in ROI to a block in RONI using Eq. (1).
an example, the binary data 000001111110000000 will be 13. For each ROI block, calculate average intensity value
interpreted as five 0’s, six 1’s, seven 0’s and it is coded as and then embed into LSBs of first 8 pixels in mapped
(101, 0), (110, 1) and (111, 0). The original binary data RONI block.
containing 18 bits is compressed to 12 bits. The 14. Encrypt the bits indicating the information of ROI
compressed watermark is encrypted using a secret key k1 using secret key k1.
to provide security. The resultant watermark is embedded 15. Embed the encrypted bits into the LSBs of border
into LSBs of ROI pixels. pixels.

3.4 Mapping between blocks of ROI and RONI

3.6 Extraction Algorithm
After embedding watermark in LSBs of ROI pixels, RONI
is divided into non overlapping blocks of size 3×3 pixels. 1. Extract the encrypted bits from the LSBs of border
Assuming that the number of blocks inside ROI is less pixels in watermarked medical image.
than the number of blocks inside RONI, for each block in 2. Decrypt the extracted bits to get information of ROI.
ROI the corresponding mapped block in RONI is 3. Identify ROI pixels and RONI pixels in watermarked
identified using Eq. (1). medical image.

Copyright (c) 2014 International Journal of Computer Science Issues. All Rights Reserved.
IJCSI International Journal of Computer Science Issues, Vol. 11, Issue 5, No 1, September 2014
ISSN (Print): 1694-0814 | ISSN (Online): 1694-0784
www.IJCSI.org 36

4. Extract the encrypted watermark from the LSBs of Perceptual Error (TPE) [22] metric. Lower value of TPE
pixels inside ROI. indicates less degradation in the watermarked image.
5. Decrypt the extracted watermark to get wcomp.
6. Decompress the wcomp to obtain the hash value (h1) of Some of the medical images used in our experiments are
ROI, LSBs (B) of pixels inside ROI and EPR (E) of shown in Fig. 2. All images are resized to 256×256 and
patient. patient data of 0.5 KB size is embedded inside ROI. A
7. Replace the LSBs of pixels inside ROI with the bits in rectangular shaped ROI is considered in each medical
B. image for simulating the proposed method. Fig. 2 also
8. Calculate hash value (h2) of the ROI using MD5. shows the watermarked images generated after embedding
9. Compare h1 with h2. If h1=h2 then the ROI is watermark into original images and the watermark
authentic and the extraction process ends. extracted or reconstructed medical images. There is no
10. If h1≠h2 then the ROI is not authentic and is significant visual difference between the original,
tampered. Proceed to next step to detect tampered watermarked and watermark extracted medical images.
blocks inside ROI and recover original ROI. Table 1 illustrates the results obtained after embedding
11. Divide ROI and RONI into blocks of size 4×4 and watermark into the three medical images shown in Fig. 2.
3×3 respectively. For each ROI block identify the Table 2 depicts the average results obtained by
mapped RONI block using Eq. (1) as in embedding watermarking the hundred medical images used in our
procedure. For each ROI block, calculate average experiments. Results shown in Tables 1 and 2 indicate
intensity and then compare it with the average that the proposed method works well for different
intensity extracted from LSBs of first 8 pixels in the modalities of medical images.
corresponding mapped RONI block. If they are not
equal then mark the block as tampered and replace A medical image watermarking technique is effective if
the pixels in this block with the extracted average the PSNR value of watermarked and reconstructed
value. medical image is greater than 40 dB [23]. In the proposed
method, The PSNR and WPSNR values of watermarked
and reconstructed medical images are above 40 dB. The
perceived change in the structural information of the
4. Experimental Results watermarked medical images is insignificant as the
MSSIM values for all modalities of images are near to 1.
We developed a MATLAB program for testing the
Similarly, the low average TPE values show less visual
performance of the proposed method. For conducting
degradation in the watermarked medical images.
experiments, we used around hundred 8-bit grayscale
medical images of different sizes and modalities like CT
The intruders are prevented from getting information of
scan, MRI scan and Ultrasound. Out of these hundred
ROI by encrypting it before embedding inside border
images, 35 medical images are of CT scan, 40 medical
pixels. If an attacker identifies the ROI region and gets
images are of MRI scan and 25 medical images are of
the LSBs of pixels inside ROI then he cannot do anything
Ultrasound. Peak Signal to Noise Ratio (PSNR) and
with that data as is encrypted by a secret key. Some of the
Weighted Peak Signal to Noise Ratio (WPSNR) [20] are
state-of-the-art techniques [9, 10, 11, 13] are not using
used to measure the distortion in the generated
any authentication data, like hash value of ROI to check
watermarked medical images.
directly whether the ROI is tampered or not. So, all blocks
inside ROI have to be checked one after the other to detect
Higher value of PSNR and WPSNR designates less
the presence of tampers. This checking process leads to
distortion in the watermarked image. Mean Structural
wastage of time when the watermarked medical image is
SIMilarity index (MSSIM) [21] metric is used to measure
not tampered. Such wastage of time is not incurred in the
the similarity between the original and the watermarked
proposed method as it is using hash value of ROI to
medical image. MSSIM value is between -1 and 1. Value
directly check whether the ROI is tampered.
1 of MSSIM indicates that the original and watermarked
images are similar. Visual degradation in the
To test the performance of proposed method in terms of
watermarked image is measured using the Total
detecting tampered blocks inside ROI and recovering

Copyright (c) 2014 International Journal of Computer Science Issues. All Rights Reserved.
IJCSI International Journal of Computer Science Issues, Vol. 11, Issue 5, No 1, September 2014
ISSN (Print): 1694-0814 | ISSN (Online): 1694-0784
www.IJCSI.org 37

Fig. 2. Original, watermarked and reconstructed medical images. From top to bottom: CT scan, MRI scan and Ultrasound images.

Table 1: Results of embedding watermark into medical images of different modalities

Size
Size Number
Size of of
Modality of w of blocks PSNR WPSNR MSSIM TPE
ROI wcomp
(bits) in ROI
(bits)
CT 200×192 42752 36348 2400 52.27 54.35 0.9347 0.0612
MRI 132×176 27584 23440 1452 57.34 58.13 0.9635 0.0445
US 104×128 17664 13072 832 60.56 63.21 0.9923 0.0221

Table 2: Performance of the proposed method

Modality of Average Average Average Average
Image PSNR WPSNR MSSIM TPE
CT Scan 51.38 53.14 0.9216 0.0604
MRI Scan 54.26 56.89 0.9714 0.0468
Ultrasound 58.76 60.79 0.9851 0.0201

original ROI, we induced a tamper inside ROI of the For testing the capability of proposed method in detecting
watermarked medical images as shown in Fig. 3. tampers at multiple locations inside ROI and recovering
Proposed method identified the tamper inside ROI and original ROI, we modified pixels at number of locations
recovered original ROI. inside ROI of watermarked medical images as shown in
Fig. 5. Proposed method detected all the tampers inside
The reconstructed medical images are shown in Fig. 4. In ROI and recovered original ROI. Fig. 6 shows the
a medical image, the LSB of pixels inside RONI and reconstructed medical images. Some of the reviewed
border are generally zero. So, the LSB of pixels inside schemes [10, 11, 13] cannot recover ROI when tampers
RONI and border are set to 0 after extracting embedded are induced by attackers at multiple locations inside ROI.
data from them. Table 3 depicts the comparison between proposed method
and the previously developed block based methods for
tamper detection and recovery.

Copyright (c) 2014 International Journal of Computer Science Issues. All Rights Reserved.
IJCSI International Journal of Computer Science Issues, Vol. 11, Issue 5, No 1, September 2014
ISSN (Print): 1694-0814 | ISSN (Online): 1694-0784
www.IJCSI.org 38

Fig. 3. Watermarked medical images (from left to right: CT scan, MRI scan and Ultrasound) with a tamper inside ROI.

Fig. 4. Recovered medical images (from left to right: CT scan, MRI scan and Ultrasound).

Fig. 5. Watermarked medical images (from left to right: CT scan, MRI scan and Ultrasound) with tampers at different locations inside ROI.

Fig. 6. Recovered medical images (from left to right: CT scan, MRI scan and Ultrasound).

Proposed method is developed based on the assumption watermarked medical image are not attacked by any noise
that the intruders generally try to modify only the or not modified by intruders or not processed by common
significant part, ROI, in the medical images during their image manipulation operations.
transmission. So, identifying changes inside ROI and
recovering original ROI must be done before using the
medical image for diagnosis and to avoid misdiagnosis. 4. Conclusions
One of the limitations of proposed method is: the RONI
and border parts are not recovered exactly as LSBs of all Proposed medical image watermarking method produces
pixels inside RONI and border are set to 0 after extracting high quality watermarked medical images. The
embedded data from them. This limitation does not affect watermarked medical images look more similar to
the efficiency of the proposed method as RONI and border original medical images as PSNR, WPSNR values of
parts of medical images are not significant for making watermarked medical images are above 50dB and MSSIM
diagnosis decisions. Another limitation is: no security for values are above 0.93. Proposed method can be used with
the ROI recovery data that is embedded inside RONI. If medical images whose ROI part is up to 62% of entire
the ROI is tampered then the original ROI can be image. Proposed method uses only 8 bit authentication
recovered only when the RONI and border part of the and recovery data for each 4×4 block inside ROI. It
identifies and localizes tampers inside ROI and recovers

Copyright (c) 2014 International Journal of Computer Science Issues. All Rights Reserved.
IJCSI International Journal of Computer Science Issues, Vol. 11, Issue 5, No 1, September 2014
ISSN (Print): 1694-0814 | ISSN (Online): 1694-0784
www.IJCSI.org 39

original ROI. When the extracted hash value of ROI For future enhancement, we try to extend the method for
matches with recalculated hash value of ROI, then the medical images whose pixels are represented using 10 or
proposed method do not check the blocks inside ROI for 12 or 16 bits and also to sustain common attacks, reduce
detecting the presence of tampers. Computational embedding distortion inside ROI and recover the pixels
complexity of proposed method is less as it uses simple inside ROI with their original bits instead of with average
mathematical calculations for generating authentication of pixels.
and recovery data, detecting tampered blocks inside ROI
and recovering original ROI.

Table 3: Comparison between reviewed schemes and proposed scheme

Is there any
provision
ROI- Size of authentication and Recovery of ROI/image when it is
Scheme for
based recovery data tampered
embedding
EPR
Not possible if a block and its
Zain No 9 bits for each 4×4 block No
mapping block both are tampered
Possible, but with only compressed
Wu Yes 150 bits for each block No
form of ROI
Chiang Yes 128 bits for each 4×4 block No Yes
10,11 Not possible if a block and its
Liew Yes 9 bits for each 4×4 block No
mapping block both are tampered
Memon Yes - Yes No
Not possible if a block and its
Agung Yes 9 bits for each 3×3 block No
mapping block both are tampered
15
Qershi Yes 128 bits for each 4×4 block Yes Yes
16 Possible, but with only compressed
Qershi Yes - Yes
form of ROI
Deng No 8 bits for each block No Yes
Our
Yes 8 bits for each 4×4 block Yes Yes
method

References IEEE EMBS Annual International Conference, 2006, pp. 3270-

[1] G. Coatrieux, J. Montagner, H. Huang and Ch. Roux, “Mixed 3273.
reversible and RONI watermarking for medical image [8] J. H. K. Wu, R.-F. Chang, C.-J. Chen, C.-L. Wang, T.-H.
reliability protection”, in 29th IEEE International Conference of Kuo,W. K. Moon and D.-R. Chen, "Tamper detection and
EMBS, Cite Internationale, Lyon, France, 2007. recovery for medical images using near-lossless information
[2] X. Luo, Q. Cheng and J. Tan, "A lossless data embedding hiding technique", Journal of Digital Imaging, Vol. 21, 2008,
scheme for medical images in application of e-Diagnosis", in pp. 59-76.
Proceedings of the 25" Annual International Conference of the [9] K.-H. Chiang, K.-C. Chang-Chien, R.-F. Chang and H.-Y.Yen,
IEEE EMBS, 2003, pp. 852-855. "Tamper detection and restoring system for medical images
[3] J.J. Eggers, R. Bauml, R. Tzschoppe and B. Girod, “Scalar using wavelet-based reversible data embedding", Journal of
SOSTA scheme for information hiding”, IEEE Transactions on Digital Imaging, Vol. 21, 2008, pp. 77-90.
Signal Processing, Vol. 151, No. 4, 2003, pp. 1003–1019. [10] S. C. Liew and J. M. Zain, “Reversible medical image
[4] N.A. Memon and S.A.M. Gilani, “NROI watermarking of watermarking for tamper detection and recovery”, 3rd IEEE
medical images for content authentication”, in Proceedings of International Conference on Computer Science and Information
12th IEEE International Mutitopic Conference (INMIC’08), Technology (ICCSIT), 2010, pp. 417-420.
Karachi, Pakistan, 2008, pp. 106–110. [11] S. C. Liew and J. M. Zain, “Reversible medical image
[5] C. Delong, “A semi-fragile watermarking for ROI image watermarking for tamper detection and recovery with Run
authentication”, International Conference on Multimedia and Length Encoding compression”, World Academy of Science,
Information Technology, 2008, pp. 302–305. Engineering and Technology, 2010, pp. 799-803.
[6] A. Giakoumaki, S. Pavlopulos and D. Koutouris, “Multiple [12] Nisar Ahmed Memon, Asmatullah chaudhry and Mushtaq
image watermaking applied to health information Ahmad, “Hybrid watermarking of medical images for ROI
management”, IEEE Transactions on Information Technology authentication and recovery”, International Journal of Computer
and Biomedicine, Vol. 10, No. 4, 2006, pp. 722–732. Mathematics, Vol.88, No.10, 2011, pp. 2057- 2071.
[7] J. M. Zain and A. R. M. Fauzi, "Medical image watermarking [13] T. Agung, B. W. AdiWijaya and F. B. Fermana, “Medical
with tamper detection and recovery", in Proceedings of the 28th image watermarking with tamper detection and recovery using

Copyright (c) 2014 International Journal of Computer Science Issues. All Rights Reserved.
IJCSI International Journal of Computer Science Issues, Vol. 11, Issue 5, No 1, September 2014
ISSN (Print): 1694-0814 | ISSN (Online): 1694-0784
www.IJCSI.org 40

reversible watermarking with LSB modification and Run [19] Hussain Nyeem, Wageeh Boles and Colin Boyd, “Ultilizing
Length Encoding compression”, IEEE International Conference least significant bit-planes of RONI pixels for medical image
on Communication, Networks and Satellite, 2012, pp. 167-171. watermarking”, International Conference on Digital Image
[14] O. M. Al-Qershi and B. E. Khoo, "Authentication and data Computing: Techniques and Applications (DICTA), IEEE,
hiding using a reversible ROI-based watermarking scheme for 2013.
DICOM images", in Proceedings of International Conference on [20] N. Ponomarenko, S. Krivenko, K. Egiazarian, J. Astola and V.
Medical Systems Engineering (ICMSE), 2009, pp. 829-834. Lukin, “Weighted MSE based Metrics for Characterization of
[15] O. M. Al-Qershi and B. E. Khoo, "ROI-based tamper detection Visual Quality of Image Denoising Methods”, Eighth
and recovery for medical images using reversible watermarking International Workshop on Video Processing and Quality
technique”, International Conference on Information Theory Metrics for Consumer Electronics – VPQM, 2014.
and Information Security (ICITIS), IEEE, 2010, pp. 151-155. [21] W. Zhou, A. C. Bovik, H. R. Sheikh and E. P. Simoncelli,
[16] O. M. Al-Qershi and B. E. Khoo, "Authentication and data “Image Quality Assessment: From Error Visibility to Structural
hiding using a hybrid ROI-based watermarking scheme for Similarity”, IEEE Transactions on Image Processing, Vol. 13,
DICOM images", Journal of Digital Imaging, Vol. 24, No. 1, No. 4, 2004, pp. 600-612.
2011, pp. 114-125. [22] A. B. Watson, “DCT quantization matrices visually optimized
[17] X. Deng, Z. Chen, F. Zeng, Y. Zhang and Y. Mao, for individual images”, SPIE Human Vision, Visual Processing
“Authentication and recovery of medical diagnostic image using and Digital Display IV, 1993, pp. 202-216.
dual reversible digital watermarking”, Journal of Nanoscience [23] K.Chen and T.V. Ramabadran, “Near-lossless compression of
and Nanotechnology, Vol. 13, No.3, 2013, pp. 2099-107. medical images through entropy coded DPCM”, IEEE
[18] Baiying Lei, Ee-Leng Tan, Siping Chen, Dong Ni, Tianfu Wang Transactions on Medical Imaging, Vol. 13, No. 3, 1994, pp.
and Haijun Lei, “Reversible watermarking scheme for medical 538–548.
image based on differential evolution”, Expert Systems with
Applications, Vol. 41, 2014, pp. 3178-3188.

Copyright (c) 2014 International Journal of Computer Science Issues. All Rights Reserved.