See discussions, stats, and author profiles for this publication at: https://www.researchgate.

net/publication/282567134

A Quick Glance at Digital Watermarking in Medical Images Technique

Classification, Requirements, Attacks and Application of Tamper Localization

Article · July 2013

DOI: 10.5963/BER0202004

CITATIONS READS

3 277

3 authors, including:

Siau-Chuin Liew Jasni Mohamad Zain

Universiti Malaysia Pahang Universiti Teknologi MARA
43 PUBLICATIONS   348 CITATIONS    161 PUBLICATIONS   1,645 CITATIONS   

SEE PROFILE SEE PROFILE

Some of the authors of this publication are also working on these related projects:

Big Data and Analytics and mapReduce View project

Reversible Medical Image Watermarking View project

All content following this page was uploaded by Siau-Chuin Liew on 06 October 2015.

The user has requested enhancement of the downloaded file.

 A Review on Digital Image Watermarking in Medical Images: Technique
Classification, Requirements, Attacks and Application of Tamper Localization.

Syifak Izhar Hisham*, Siau-C huin Liew, & Jasni Mohd Zain
Faculty of Computer Systems and Software Engineering,
Universiti Malaysia Pahang, Pahang, Malaysia
*Corresponding Author: penawar85@gmail.com , +6019-3551106 (Phone)

Abstract
Digital watermarking is seen as a necessary and important field 2.1 Spatial Domain
to be developed since the attacks and manipulations of digital
documents and media are actively done. This paper surveys and
reviews some basic theory about digital watermarking, such as
One of the earliest and most simple techniques is to
the technique classification, requirements and possible attacks embed the watermark information into the LSBs of the image
nowadays. Since all data in medical images are significantly [3, 4]. The watermarks are embedded in the last bit of
important, the demand for the authentication is high. Therefore, selected pixels of an image. Since a change in LSB
this paper also focuses on a specific application of digital image corresponds a change in one unit of image gray value, its
watermarking for medical images which is tamper localization. modification is not perceivable by human eyes [5]. This
technique does not produce serious distortion to the original
Key words: Watermarking, Tamper Localization, Medical image; however, not as robust as transform domain
Image, Tampering Detection, Review techniques and rarely survives various attacks.

1. INTRODUCTION 2.2 Transform Domain

Nowadays it is a trend to create, store and distribute Most of the transform domain techniques embed the
data in digital multimedia data format. In general, multimedia watermark information into the transform coefficients of the
brought many benefits to society. For instance, the good cover image. The transform domain techniques produces
quality of the image and voice as well, easily send or receive spectral domains where watermarking can be applied. The
messages and important documents, which make our life most popular techniques in this category are Discrete Cosine
easier and convenience. However, this revolution helps the Transform (DCT), Discrete Wavelet Transform (DWT) and
pirates to exploit these features for their own intended Discrete Fourier Transform (DFT). The advantage of these
purpose illegally [1]. techniques is they can overcome possible compression and
more robust against geometric transformation such as
Therefore, the demand for the authentication methods of rotation, scaling, translation and cropping; but, the
digital media becomes significant issue in order to ensure that disadvantage is they need a certain amount of computation
work have not been tampered with, especially for some cases which take longer time to produce [2].
like national security, medical safety, internet banking and
transfer of military information and forensic investigations. A 3. REQUIREMENTS FOR IMAGE
digital watermarking system considered as a solution for WATERMARKING METHODS
preserve the integrity, confidentiality and the authenticity of
digital media. This paper surveys the classification,
requirements, attacks and the need of the watermarking, Most image watermarking schemes try to meet the
specifically on medical images. following requirements [6, 7, 8]:

2. TECHNIQUE CLASSIFICATION OF IMAGE 3.1 Perceptibility

WATERMARKING
The perceptibility of a watermarked image can be
Watermarking techniques can be classified according to judged according to its fidelity and quality. Fidelity measures
how the watermark is embedded [1, 2]. Mainly, it is divided the similarity between images before or after watermarking
into two broad categories, spatial domain and transform [1]. A reconstructed image that is very similar to the original
domain [3]. has a high fidelity. A low fidelity reconstruction is dissimilar
or distinguishable from the original. Watermarked images
may bear visible or invisible distortion due to the embedding
process.
3.2 Robustness applied. Included in this category are skewing, image rotation
and translation.
It should be robust against typical attacks, transmission
and storage imperfections, such as compression, noise 4.3 Cryptographic Attack
addition, format conversion, bit errors, filtering and noise
reduction. For content authentication, a robust watermark The aim of this kind of attack is to break the security
may not be able to differentiate both compression and measures applied in the watermarking schemes. Once the
tampering. security measure is broken, the embedded watermark is
removed or a misleading watermark is embedded. Brute-
Fragile watermarking can be an advantage for force search is one of the techniques in this category. This
authentication purposes. If a fragile watermark is detected technique attempts to break the security of the watermark by
correctly in an image, it can be assumed that the image has using a large number of known possible measures to find
not been altered or tampered with since the watermark has meaningful secret information.
been embedded [5].
4.4 Protocol Attack
3.3 Capacity
The last category is the protocol attack. It is aim at
Sufficiently useful amount of information must be attacking the entire concept of watermarking application such
allowed by a watermarking scheme to be embedded into an as in copyright protection. The attacker adds its own
image. The amount of information embedded is subjected to watermark into an image and causes the true ownership of the
the application. A reversible watermarking scheme may need image in question.
to embed more information to allow image restoration, than a
watermark scheme that embeds hash value of an image for 5. APPLICATION OF DIGITAL IMAGE
authentication purposes [5]. WATERMARKING: TAMPER LOCALIZATION
FOR MEDICAL IMAGES
3.4 Computational Complexity
From the previous section, we can see that removal
The watermarking scheme should not be computationally attack is a crucial problem to medical images. One of the
complex especially for applications where real-time requirements of an effective watermarking based
embedding is desired. authentication system as defined by Liu and Qiu (2002) is the
ability to identify manipulated area or also known as
4. ATTACKS TO DIGITAL WATERMARKING localization where the authentication watermark should be
able to detect the location of manipulated areas, and verify
other areas as authentic.
Watermarked images may be vulnerable to various
attacks as there is no watermarking scheme that can provide
the perfect security protection needed. Voloshynovskiy et al. Guo and Zhuang (2009) proposed a reversible scheme
(2001) had classified watermark attacks into four categories with tamper localization based on difference expansion. This
as below. scheme partitions an image into certain non-overlapping
regions and appending the associated local authentication
information directly into the watermark payload. The scheme
4.1 Removal Attack
also introduces the concept of region of authentication
(ROA). ROA is a region used for integrity authentication or
The aim of removal attacks is to remove the watermark in other words, the area that needs to be protected. A ROA,
signal from the watermarked image without breaking the which can be flexibly defined by the user, is partitioned into
security of the watermarking algorithm. It does not attempt to small regions as an image block or polygonal region in a
find out the encryption techniques applied or how the multilevel hierarchical manner. The novelty of this scheme is
watermark is being embedded. This category includes that the information about the ROA is embedded as part of
compression, noising, sharpening and histogram equalization. the watermark. The ROA will be used to reconstruct the
ROA in the verification process. A hashing function is used
4.2 Geometry Attack to produce digital signatures for each image block which are
then added to the watermark payload. In order to verify the
In geometry attack, the watermark signal is distorted authenticity of the image, the process starts by comparing the
rather than being removed from the image. It is possible to signature for the whole image. If the initial verification
recover the original watermark if proper countermeasure is process fails, the ROA will be reconstructed. The signatures
for the ROA will be compared to detect any tampering.
 An interesting technique is used in the tamper tampered region can be localized to an accuracy of 4 x 4
localization process where an output image consist of pixels. The tampered block is recovered using the average
shadings of the ROA is produced. The shading is used to pixel value retrieved from the watermark. The advantage of
reflect the level of confidence in the integrity of the ROA this scheme is that it can be modified to allow applying the
where light shadings corresponds to high confidence value watermarking process to a defined ROI rather than to the
and dark shadings corresponds to low confidence value. The whole image. The recovery information of the ROI is stored
tamper localization has the accuracy of up to 32 x 40 pixels. as the exact pixel value rather than average pixel values.
Ultrasound image was used in the experiment of this Mammograms were watermarked and have the PSNR
watermarking scheme and quality of the watermarked image between 36.4 to 40.5 dB.
is crucial especially for medical diagnoses. The perceptibility
of the watermarked ultrasound image is not known as the Osamah and Khoo (2011) proposed a scheme that
measurement of the distortion level of watermarked image consists of two types of watermark. The first watermark is
was not done in the experiment. embedded into spatial domain and the second watermark is
embedded into transform domain. The image is first divided
Tan et al. (2011) also proposed a tamper localization into 16x16 pixel blocks. The first watermark consists of
watermarking scheme that uses pixel value modification in patient’s data and the hash value of the ROI and is embedded
order to allow the watermark to be reversible. The image is into the ROI itself by using a modified difference expansion
divided into 16 x 16 pixel blocks and Cyclic Redundancy technique. An embedding map of the ROI is produced to
Check (CRC) is computed for each block. Each CRC is form a second watermark together with compressed recovery
embedded into its own block and in the event that the CRC information of ROI and average value of each block in the
cannot be embedded into its own block, the remaining bits ROI. The second watermark is compressed and embedded
will be carried over to the next block. The watermarked into the region of non- interest (RONI) using a DWT
image can be verified by extracting the watermark and technique. Tamper localization is done by comparing the
comparing the CRC of each block. Any mismatch of CRC average value of each block in the ROI with the retrieved
values during comparison indicates tampering and the average value from the watermark. Tampered blocks can be
tampering localization accuracy is within 16 x 16 pixels. recovered using the compressed ROI. It was claimed that this
Medical images were watermarked and the PSNR of the scheme is robust against salt and pepper attack and cropping.
watermarked images was between 34.0 to 35.0 dB. The A watermarked ultrasound image has the PSNR of 36.7 dB.
disadvantage of this scheme is that in order to allow
reversibility, all pixel value needs to be increased by four Earlier research by Jasni and Abdul (2006) had also
pixel values during the embedding process to prevent bit produced a tamper localization and recovery watermarking. It
overflow and thus the maximum pixel value allowable in an also uses block based technique where each block consists of
image to be watermarked had been constrained. 8 x 8 pixels. Each block is then divided into sub-blocks of 4 x
4 pixels.
Both schemes proposed by Guo and Zhang (2009) and
Tan et al. (2011) operate in the spatial domain and have A three-tuple watermark embedded consists of a two-bit
tamper localization and reversible capability. The schemes authentication watermark and a seven -bit recovery
might be able to identify the area of tampering but tampered watermark for other sub-block. Average intensity of a
region cannot be recovered. Recovery of the tampered region corresponding block and its sub-blocks is calculated to
is useful in order to know exactly what had been tampered generate the authentication watermark. Average intensity of a
and the motive of the tampering. sub-block is embedded as the seven-bit recovery watermark
in another block which was predetermined in a mapping
Chiang et al. (2008) proposed a reversible tamper sequence. A parity bit is generated based on the seven-bit
localization scheme with tampered region recovery recovery watermark. Tamper localization is done by
capability. This scheme is based on a difference expansion comparing the average intensity and parity bit. Blocks that
scheme proposed by Tian (2003). It was modified to allow were marked invalid are recovered using the embedded
the watermark to be embedded into the transform domain by average intensity of the sub-block. The watermarked
using the integer Haar wavelet transform. The image is first ultrasound image has a PSNR of 54.8 dB. This scheme was
divided into blocks. The recovery information is generated by evaluated to know whether watermarked medical images
taking the average pixel value of each block and embedded as affect clinical diagnoses. The study was done by Jasni et al.
watermark. The watermark is encrypted before the (2006) by adding an additional hash function to the existing
embedding process as a security feature. The whole image watermarking scheme. Various types of medical images were
can be verified by comparing the retrieved average pixel watermarked and an ultrasound image has a PSNR of 54.2
value from the watermark with the current average pixel dB with the total watermark bits of 480K. The watermarked
value of the image. Any mismatch indicates tampering and images were assessed by radiologists and it was concluded
that watermarked images did not alter clinical diagnoses. TABLE 1
The disadvantage of this scheme is that it is not reversible. SUMMARY OF TAMPER LOCALIZATION
The original image is generally preferred by radiologist for WATERMARKING FOR MEDICAL IMAGES
diagnostic purposes [12]. Although it has been clinically Researcher Characteristic PSNR
evaluated, an option should be given to allow the watermark tamper
to be removed and the original image to be restored by Guo and
localization,reversible N/A
request. Zhuang (2009)
capability.
Tan et al. tamper localization,
Yang and Shen (2010) applied hash function to image 34.0 - 35.0 dB
(2011) reversible capability.
blocks and embed the hash values into the LSBs of the
corresponding blocks. They also used vector quantization to tamper localization,
Chiang et al.
compress an image by producing an index table which can be reversible recovery 36.4 - 40.5 dB
(2008)
used for image recovery. The index table is embedded into capability
the second and third LSB of each pixel. Each block of the Osamah and tamper localization,
image is authenticated using the embedded hash value. Index 36.7 dB
Khoo (2011) recovery capability
table is used to reconstruct an image when tampering is Jasni and tamper localization,
detected. Tampered block is recovered using blocks from the 54.8 dB
Abdul (2006) recovery capability
reconstructed image. Non-medical image was watermarked
with the PSNR of only 29.3 dB. Yang and tamper localization,
29.3 dB
Shen (2010) recovery capability
Kim et al. (2011) proposed an improved scheme which for 12-bit CT
enables the detection of a tampered region and then recovers image - 65.76
Kim et al. tamper detection,
it by using the embedded information selected as the dB, for 16-bit
(2011) recovery capability
recovery feature. To extract the recovery feature, they MRI image -
analyzed the image homogeneity using quad-tree 89.3 dB
decomposition. It divides a square image into several Liew et al. tamper localization
variable-sized blocks adaptively, and chooses the average 48.7 dB
(2012) and lossless recovery
value of each block as the feature. This method produces a
more efficient length of the feature while ensuring that the
The data were based on an experiment performed on an
visual quality of the restored image is better than the method
ultrasound image. The comparison result shows that
proposed by Chiang et al. (2008). The method had been
TALLOR-RSMA scheme is better in terms of embedding
experimented in CT image and MRI image. The PSNR for
capacity and PSNR. The tamper localization accuracy of the
12-bit CT image is 65.76 dB and is 89.3 dB for 16-bit MRI
proposed scheme is at one pixel compared with 16×16 pixels
image, while the SSIM value for 12-bit CT image is 0.9938
in the Osamah and Khoo (2011) scheme. The recovered ROI
and 0.9999 for 16-bit MRI image. The values were improved
produced by TALLOR-RSMA is also of better quality due to
from the result by Chiang et al. (2008).
exact recovery achieved.
Table 1 shows the summary of the schemes discussed
Liew et al. (2012) proposed a scheme called Tamper above.
Localization and Lossless Recovery Watermarking Scheme
with ROI Segmentation and Multilevel Authentication
6. DISCUSSION
(TALLOR-RSMA), which is the enhancement of the tamper
localization and lossless recovery (TALLOR) [22] scheme.
The quality of the watermarked images using this scheme is All schemes reviewed under tamper localization section
high, with the average PSNR of 48.7 dB for the proposed in this paper were designed specifically for usage in medical
scheme. Significantly, the recovered images were identical images. All schemes have tamper localization capability and
with the original images. This was proven after comparing a majority has recovery capability. The watermarked images
the hash values from the original and recovered images. The have the PSNR of between 29.3 to 89.3 dB. The schemes that
method used in the TALLOR-RSMA scheme proved have recovery capability operate in the transform domain are
effective in reducing the tamper localization and recovery more robust against attack but require more complex
average processing time by approximately 50% compared computation. Eventually it needs more watermarking
with the TALLOR scheme. processing time. Scheme that operates in the spatial domain
developed by Eric et al. (2012) produces watermarked image
that has high PSNR and has the reversible capability. The
A comparison between this scheme and the tamper
recovered image can get exact recovery. The schemes by
localization and recovery scheme proposed by Osamah and
Kim et al. (2011) produce the highest PSNR among those
Khoo (2011) which is described previously has been done.
studies in this paper, which is 65.76 dB for 12-bit image and [13] Chiang, K., Chang, K., Chang, R. & Yen, H., “Tamper Detection
And Restoring System For Medical Images Using Wavelet-Based
is 89.3 dB for 16-bit image.
Reversible Data Embedding”, J Digit Imaging, 21:77– 90.2008.
[14] Tian, J., “Reversible Data Embedding Using A Difference
Expansion”, Ieee Trans Circuits Syst Video Technol, 13(8):890 –
896. 2003.
7. CONCLUSION [15] Osamah, M. & Khoo, B. E., “Authentication And Data Hiding
Using A Hybrid Roi-Based Watermarking Scheme For Dicom
This paper surveys from some basic knowledge of digital Images”, J Digit Imaging, 24:114– 125. 2011.
[16] Jasni, M. Z. & Abdul, R. M. F., “Medical Image Watermarking
watermarking, includes the techniques classification, With Tamper Detection And Recovery”, Proceedings Of The
requirements, and attacks of digital watermarking. This paper 28th Annual International Conference Of The Ieee Engineering
also reviews studies about an application of watermarking for In Medicine And Biology Society, Pp.3270-3273.2006.
medical images which is tamper localization. We focus on [17] Jasni, M. Z., Abdul, R. M. F. & Azian, A. A., “Clinical
Evaluation Of Watermarked Medical Images”, Proceedings Of
tamper localization since medical image should have a The 28th Annual International Conference Of The Ieee
special security system from data manipulation and be very Engineering In Medicine And Biology Society, Pp.5459-5462.
sensitive to any distortion and attacks. There are, still, many 2006.
limitations and areas that should be developed for tamper [18] Yang, C. W. & Shen, J. J., “Recover The Tampered Image
Based On Vq Indexing”, Signal Processing, 90(1):331-343,
localization, such as other modalities of medical images 2010.
should be studied and implementation in the real data system [19] Kim, K. S., Lee, M. J., Lee, J. W., Oh, T. W. & Lee, H. Y.,
at hospital should be done and evaluated. “Region-Based Tampering Detection And Recovery Using
Homogeneity Analysis In Quality-Sensitive Imaging”, Computer
Vision And Image Understanding, 115 (2011) 1308–1323. 2011.
REFERENCES [20] Chiang, K. H., Chien, K. C. C., Chang, R. F. & Yen, H. Y.,
“Tamper Detection And Restoring System For Medical Images
[1] Cox, I. J., Miller, M. L. & Bloom, J. A., “Digital Watermarking”, Using Wavelet-Based Reversible Data Embedding”, Journal Of
Morgan Kaufmann, San Francisco, 2002. Digital Imaging. 21 (1) (2008) 77–90. 2008.
[2] Song, C., Sudirman, S., Merabti, M. & Llewellyn-Jones, D., [21] Liew, S. C., Liew, S. W. & Jasni M. Z., “Tamper Localization
“Analysis Of Digital Image Watermark Attacks”, Proceedings Of And Lossless Recovery Watermarking Scheme With Roi
The 7th Ieee Consumers Communications And Networking Segmentation And Multilevel Authentication”, J Digit Imaging,
Conference, Pp. 1-5, Jan 2010. Doi 10.1007/S10278-012-9484-4. 2012.
[3] Liew, S. C. & Jasni, M. Z., “A Review of Medical Image [22] Liew S. C. & Zain, J. M., “Tamper Localization And Lossless
Watermarking And Its Implementations”, Proceedings Of Recovery Watermarking Scheme”, Commun Comput Inf Sci,
Malaysian Technical Universities Conference On Engineering 179(1):555 – 566. 2011.
And Technology (Muceet2009), Kuantan, Pahang, Malaysi, 20-22
June 2009.
[4] Mehta, G. N., Kshirsagar, Y. & Tankariya, A., “Digital Image
Watermarking”, International Journal Of Scientific Engineering
And Technology, Volume No. 1, Issue No. 2, Pg: 169 – 174, 2012
[5] Liew, S. C. , “Tamper Localization And Recovery Watermarking
Schemes For Medical Images In Pacs”, Doctor Of Philosophy
(Computer Science) Thesis, Universiti Malaysia Pahang,
Malaysia. 2011.
[6] Cox, I. J., Miller, M. L. & Linnartz, J. M. G., “A Review Of
Watermarking Principles And Practices”. Ieee Digital Signal
Processing For Multimedia System,1:461-482, 1999.
[7] Kutter, M. & Hartung, F., “Introduction To Watermarking
Techniques”, Information Techniques For Steganography And
Digital Watermarking, 1:97-119, 1999.
[8] Meerwald, P. & Uhl, A., “Watermark Security Via Wavelet Filter
Parameterization”, Proceedings Of The International Conference
On Image Processing, Pp.1027-1030. 2001.
[9] Voloshynovskiy, S., Pereira, S., Pun, T., Eggers, J. J. & Su, J. K.,
“Attacks On Digital Watermarks: Classification, Estimation
Based Attacks, And Benchmarks”, Ieee Communications
Magazine, 39(8):118-126. 2001.
[10] Liu, T. & Qiu, Z. D., “The Survey Of Digital Watermarking-
Based Image Authentication Techniques”, In Proceedings Of The
6th International Conference On Signal Processing, 2002
[11] Guo, X. & Zhuang, T., “Lossless Watermarking For Verifying
The Integrity Of Medical Images With Tamper Localization”, J
Digit Imaging, 22 (6):620 – 628, 2009
[12] Tan, C. K, Ng, C, Xu, X, Poh, C. L, Yong, L.G. & Sheah, K.,
“Security Protection Of Dicom Medical Images Using Dual-
Layer Reversible Watermarking With Tamper Detection
Capability”, J Digit Imaging, 24(3):528–540. 2011.

View publication stats