Securing Iris Images with a Robust Watermarking Algorithm based on
Discrete Cosine Transform
Mohammed A. M. Abdullah
, S. S. Dlay and W. L. Woo
Newcastle University, School of Electrical and Electronic Engineering, NE1 7RU, Newcastle Upon Tyne, U.K.
Biometric Protection, Biometric Watermarking, Discrete Cosine Transform, Iris Recognition, Template
With the expanded use of biometric systems, the security of biometric trait is becoming increasingly important.
When biometric images are transmitted through insecure channels or stored as a raw data, they become subject
to the risk of being stolen, faked and attacked. Hence, it is imperative that robust and reliable means of
protection are implemented. Various methods of data protection are available and digital watermarking is one
such techniques. This paper presents a new method for protecting the integrity of the iris images using a
demographic text as a watermark. The watermark text is embedded in the middle band frequency region of
the iris image by interchanging three middle band coefficients pairs of the Discrete Cosine Transform (DCT).
Experimental results show that exchanging more than one pair will make middle band scheme more robust
against malicious attack along with making it resistant to image manipulation such as compression. The results
also illustrate that our watermarking algorithm does not introduce discernible decrease on iris image quality
or biometric recognition performance.
The growing need for security in recent years has
resulted in the development of personal biometrics
identification systems. Biometric is the science of es-
tablishing human identity using physical or behavior
traits. The advantages of personal identification using
biometric features are numerous, such as fraud pre-
vention and secure access control (Vacca, 2007).
Although biometrics systems offer great benefits
with respect to the traditional authentication tech-
niques, the problem of ensuring the security and in-
tegrity of the biometric data is still critical. Hence,
for a biometric system to work properly, the verifier
system must guarantee that the biometric data came
from a legitimate person at the time of enrollment (Yi-
wei et al., 2002). Encryption and watermarking can
be used to achieve this purpose. However, encryption
cannot provide security after the data is decrypted. On
the contrary, watermarking involves hiding informa-
tion into the host data for protecting its integrity, so it
can provide security even after decryption.
A number of watermarking techniques are avail-
able for embedding information securely in an image.
These can be broadly classified as transformation do-
Mohammed A. M. Abdullah is also a staff member
with University of Mosul and sponsored by the Ministry of
Higher Education in Iraq to complete his PhD.
main techniques (Yiwei et al., 2002; Deb et al., 2012;
Wang et al., 2009; Sakib et al., 2011) and spatial do-
main techniques (Mukherjee et al., 2004; Singh et al.,
2012). While the spatial domain techniques are hav-
ing least complexity and high payload they cannot
withstand low pass filtering and common image pro-
cessing attacks (Dabas and Khanna, 2013).
Recently watermarking techniques have been used
to protect biometric templates (Islam et al., 2008;
Fouad et al., 2011; Isa and Aljareh, 2012; Majumder
et al., 2013; Paunwala and Patnaik, 2014). In (Is-
lam et al., 2008) the authors proposed a protection
algorithm for the fingerprint image by watermarking
it with a password extracted from the palm print of
the same person. However, no experiments were per-
formed by the authors to show the algorithm robust-
ness against attacks. The authors in (Fouad et al.,
2011) presented a scheme for protecting the iris tem-
plate using a combination of cryptography and water-
marking. The iris image is locked with a key and em-
bedded in a cover image using combinations of Least
Significant Bit (LSB) and Discrete Wavelet Trans-
form (DWT) as a watermarking algorithm. A second
key is used to specify the embedding locations. Nev-
ertheless, the two keys (iris key and embedding key)
are required in iris extraction process. Later, (Isa and
Aljareh, 2012) proposed a watermarking algorithm
for protecting the biometric image. Face images were
Abdullah M., Dlay S. and Woo W..
Securing Iris Images with a Robust Watermarking Algorithm based on Discrete Cosine Transform.
DOI: 10.5220/0005305701080114
In Proceedings of the 10th International Conference on Computer Vision Theory and Applications (VISAPP-2015), pages 108-114
ISBN: 978-989-758-091-8
2015 SCITEPRESS (Science and Technology Publications, Lda.)
watermarked using the Cox watermarking algorithm
(Cox et al., 1997). The face image acted as the user-
name for identification while the watermark acted as
a password for authentication. However, the problem
with the Cox watermarking scheme is that the original
image is needed at the watermark detector stage.
The authors in (Majumder et al., 2013) applied
biometric watermarking by taking the DWT and the
singular value decomposition of the host image to ob-
tain an Eigen value vector. Next the iris features is ex-
tracted with DCT to obtain 200 coefficients and then
embedded in the Eigen value vector derived from the
host image. Despite of the good result reported by the
authors, the drawback of this approach is that the fea-
ture extraction algorithm for iris cannot be changed.
The work in (Paunwala and Patnaik, 2014) used the
fingerprint and iris features as watermark for a cover
image. The image is divided in blocks then each block
is transformed into a two-dimensional DCT and clas-
sified as smoother block or edge block. The biomet-
ric features are embedded in the low frequency coef-
ficients of the 8× 8 DCT blocks while the edge block
is eliminated.
Biometrics like fingerprint, face, and iris conveys
unique biological information of a person. Nowadays,
iris recognition is one of the most reliable biometric
techniques that are extensively used for personal iden-
tification. To guarantee the reliability of iris recogni-
tion, a protection method is needed. The literature
review revealed that a few research have been car-
ried out so far to enhance the security of iris biometric
through watermarking.
This paper presents a watermarking algorithm
based on exchanging 3 middle band coefficients pair
of the DCT using text data as watermarks for pro-
tecting the evidentiary integrity of iris images. The
following section describes the Middle Band Coeffi-
cient Exchange (MBCE) algorithm. Section III ex-
plains the proposed algorithm. Experimental design
and performance analysis are given in Section IV and
Section V respectively. Finally, Section VI concludes
this paper.
Watermarking techniques in DCT domain allow an
image to be divided into different frequency bands,
so embedding the watermarking information in a spe-
cific frequency band becomes much easier (Langelaar
et al., 2000). Current literature survey reveals that the
middle frequency bands are most suitable for embed-
ding the watermark because the low frequency band
carries the most visual important parts of the image
while the high frequency band is exposed to removal
through compression and noise attacks on the image.
Therefore, embedding the watermark in the middle
frequency band does not affect the visual important
parts of the image (low frequency) nor overexposing
them to removal through attacks when high frequency
components are targeted (Langelaar et al., 2000).
2.1 Watermarking Algorithm
The idea of the classical Middle Band Coefficient Ex-
change (MBCE) scheme was presented by (Zhao and
Koch, 1995). Later, (Hsu and Wu, 1996) applied the
DCT to implement the middle band coefficients em-
bedding. The algorithm encodes one-bit of a binary
watermark object into one 8 × 8 sub-block of the host
image by ensuring that the difference of two mid-band
coefficients is positive in case of the encoded value is
1. Otherwise, the two mid-band coefficients are ex-
Accordingly, after the DCT is applied to the im-
age, an 8 × 8 block dimension is taken. Each DCT
block consists of three frequency bands as illustrated
in Figure 1. F
stands for the low frequency com-
ponents of the block, while F
denotes the higher
frequency components. F
is the middle frequency
band and is chosen for embedding watermark infor-
mation. This avoids significant modifications to the
cover image while providing additional resistance to
lossy compression techniques which targets the high
frequency components (Hernandez et al., 2000).
For the frequency band FM, two locations from
the DCT block (DCT
and DCT
) are chosen
as the region for comparison. After the watermark
text is converted to a binary image, each pixel value
is checked. The coefficients are swapped if the rela-
tive size of each coefficient does not agree with the bit
that is to be encoded. Thus, if the pixel value in the bi-
nary text is 1, the DCT coefficient are swapped such
that DCT
. On the other hand, in
Figure 1: Frequency regions in 8×8 DCT block (Langelaar
et al., 2000).
case of 0, they swapped so DCT
Hence, instead of inserting any data, this scheme is
hiding watermark data by interpreting 0 or 1 with the
relative values of the two fixed locations in F
gion (DCT
and DCT
). Swapping of such
coefficients will not alter the watermarked image sig-
nificantly, due to the fact that the DCT coefficients
of middle frequencies have similar magnitudes (Zhao
and Koch, 1995; Johnson and Katzenbeisser, 2000).
In the extraction stage, the 8 × 8 DCT of the image
is taken again, so it will decode a ”1” if DCT
; otherwise it will decode a ”0” to form the
The previous scheme has a serious drawback. If only
one pair of coefficient is used to hide the watermark
data, it will become vulnerable to attack as the at-
tacker can analyze some watermarked copies of an
image to predict the location of these coefficients as
well as destroy them.
To solve this problem, three coefficient pairs are
chosen from the F
frequency band to increase re-
dundancy and make the scheme robust against differ-
ent attacks. The numbers of the pairs to be swapped
in this paper were chosen as a trade-off between com-
plexity and performance.
Moreover, to improve the robustness of the wa-
termarking algorithm, we propose to add a wa-
termark strength constant k such that DCT
> k. If coefficients do not meet these crite-
ria, a constant value will be added to satisfy the rela-
3.1 Strength of Watermark
The strength of watermark has been increased by
choosing an appropriate value of the proposed
strength constant k. Increasing k will degrade the im-
age but it will reduce the chance of errors at the detec-
tion phase. Experimental results indicate that setting
k equals to 15 is the most suitable value in the percep-
tibility versus robustness. Therefore, the test has been
conducted by keeping k=15.
3.2 Embedding Algorithm
Each 8× 8 block of image will be used to hide one bit
of watermark text. A binary text image (W ) is taken as
a watermarking object which can be interpreted as a
1D array of 1 and 0. The watermark text image carries
the person bio-information such as name, ID and date
Algorithm 1: Embedding algorithm.
Input: X,W
(X: host image, W : watermarking text)
Output: Y
(Y : watermarked image)
1: loop
2: X
= X;
{subdivide the host image (X) into blocks of
8 × 8 pixel}
3: X
DCT (i)
= 2D-DCT (X
{Compute the 2D-DCT of each 8 × 8 block of
the host image}
4: for i = 1 size(W ) do
5: if W
= 0 then
6: exchange DCT coefficients such that
locations at DCT (2, 5),DCT (3,5) and
DCT (4, 3) of the 8 × 8 sub-image will be
larger than the locations
DCT (1, 6),DCT (2,6) and DCT (5, 2)
{Now adjust the three values such that
their difference becomes larger than the
constant k, thus: }
7: if DCT (2, 5) DCT (1, 6) < k then
8: DCT (2, 5)=DCT (2,5)+k/2
9: DCT (1, 6)=DCT (1,6) k/2
10: end if
11: repeat step 7-10 for the other two
coefficients: DCT (3,5) and DCT (4, 3)
12: else if W
= 1 then
13: exchange DCT coefficients such that
locations at DCT (2, 5),DCT (3,5) and
DCT (4, 3) of the 8 × 8 sub-image will be
smaller than the locations
DCT (1, 6),DCT (2,6) and DCT (5, 2)
14: end if
{Now adjust the three values such that their
difference becomes larger than the constant
k, thus: }
15: if DCT (1, 6) DCT (2, 5) < k then
16: DCT (1, 6)=DCT (1,6)+k/2
17: DCT (1, 5)=DCT (2,5) k/2
18: end if
19: repeat step 15-18 for the other two
coefficients: DCT (2,6) and DCT (5, 5)
20: Take inversre DCT to reconstruct Y
21: end for
22: end loop
of birth. The steps of the embedding algorithm are
shown in Algorithm 1 while the flow chart is depicted
in Figure 2.
Figure 2: The flow chart of the embedding algorithm.
3.3 Detection Algorithm
Watermark extraction is the reverse procedure of wa-
termark embedding. The steps of the detection algo-
rithm are shown in Algorithm 2.
The proposed algorithm has been tested on session
one of UBIRIS V1 iris database. The images were
watermarked with 64 × 64 pixel text image shown in
Figure 3(d) after converting it to a binary image.
A robust watermarking algorithm should detect the
embedded information reliably even if the water-
marked image is degraded by different transforma-
Algorithm 2: Detection algorithm.
Input: Y
(Y : watermarked image)
Output: W
(W : binary text)
1: loop
2: Y
= Y ;
{subdivide the cover image (Y ) into blocks of
8 × 8 pixel}
3: Y
DCT (i)
= 2D-DCT (Y
{Compute the 2D-DCT of each 8 × 8 block of
the cover image}
4: if DCT (2, 5), DCT (3,5),DCT (4, 3) >
DCT (1, 6), DCT (2,6),DCT (5, 2) then
5: W (i)=1
6: else
7: W (i)=0
8: end if
9: end loop
10: reconstruct the binary text image W from W
tions. Besides robustness, a good watermarking algo-
rithm should be imperceptible to the user as well as
it should not affect the matching performance of the
biometric system badly.
In order to evaluate our method, a set of different
tests have been carried out on our proposed algorithm
as shown in the next sub-sections.
Figure 3: Perceptibility of the watermarked image; (a) Orig-
inal image, (b) Watermarked image, (c) The difference im-
age, (d) Original watermark, (e) binarized text and (f) the
extracted watermark.
5.1 Watermark Perceptibility
The perceptibility is the similarity between the orig-
inal and the watermarked image. Thus, the water-
marked object should be imperceptible to the user.
According to Figure 3(c) the difference between the
original and the watermarked iris image is not notice-
able to the naked eye without the help of image pro-
cessing techniques.
To evaluate the performance of the watermarking
algorithm, Peck Signal to Noise Ratio (PSNR) Bit Er-
ror Rate (BER) are calculated. The average PSNR
between the original iris and the watermarked iris is
37.69% and the average BER is 0.257% while the av-
erage PSNR and BER of the extracted watermarking
text are 84.25% and 0.0244% respectively.
5.2 Effect on Matching Performance
To find the effect of the watermarking on the iris
recognition performance, Maseks approach (Masek,
2003) for iris recognition has been implemented and
the Equal Error Rate (EER) is calculated for the non-
watermarked iris images. After that, the proposed wa-
termarking algorithm is applied to the same iris im-
ages and the EER is calculated again. Figure 4 illus-
trates the effect of watermarking of matching perfor-
mance in term of Receiver Operating Characteristics
(ROC) curve.
According to Figure 4, the proposed watermark-
ing algorithm hardly affects the EER across all
classes. Consequently, the recognition performance
will not be affected by the proposed watermarking
5.3 Performance Against Compression
and Noise
Apart from analyzing the change in perceptibility and
matching performance, other considerations that may
degrade the images were tested on our algorithm.
There are several types of these degradations. For
Figure 4: Effect of the proposed watermarking algorithm on
iris recognition performance.
instance, the images are compressed when transmit-
ting large image files over low bandwidth channel. In
addition the iris images can be degraded if they are
transmitted over a noisy communication channel.
To simulate these effects, image compression al-
gorithm, Joint Photograph Expert Group (JPEG) has
been employed with different quality factors (Q). In
term of the noisy channel, we applied the Additive
White Gaussian Noise (AWGN) to the iris images
with zero mean and variance equals to 10
. Even
with image compression and the added noise, the ex-
tracted text is still discernible. Figure 5 depicts the ex-
tracted watermarking texts from the manipulated iris
5.4 Performance Against Image
Manipulations and Attacks
A robust watermarking algorithm should confront dif-
ferent signal processing signal processing distortions
and attacks. A number of image manipulations were
tested on our algorithm such as median filtering, his-
togram equalization and salt & pepper (noise density
= 0.005). For each type of manipulation, the match-
ing performances of the watermarked iris images are
compared with the manipulated iris images in terms
of ROC curves and EER. Moreover, the BER and
PSNR of the extracted text are also calculated after
these manipulations.
Figure 5 depicts the effect of different image ma-
nipulations on the extracted text while Figure 6 illus-
trates the effect of these image manipulations on iris
Figure 5: Extracted watermarked text after different attacks.
Figure 6: Effect of different manipulation on iris recogni-
tion performance using the proposed scheme.
recognition performance using the proposed method.
According to Figure 6, no series loss in recognition
performance is noticed in term of EER.
In order to appreciate the efficiency of the pro-
posed method, the classical MBCE scheme (Hsu and
Wu, 1996) is implemented and the proposed strength
constant (k = 15) is introduced, then the same manip-
ulations are applied to the watermarked iris images.
Table 1 summarizes the PSNR and BER of extracted
watermark text after JPEG, median filter, histogram
equalization, Gaussian noise and salt & pepper noise
using our method and the classical method.
The proposed algorithm sustained all above image
manipulations and demonstrated that the watermark-
ing scheme is resistant to different attacks.
This paper presents a novel scheme for image water-
marking to protect the integrity of the biometric im-
age. A binary text image which accommodates the
bio data of the person to be authenticated is embedded
in the iris image by interchanging the middle band co-
efficients using DCT. Exchanging more than one pair
of the middle band coefficient make the watermark-
ing scheme robust as it is impossible to the attacker to
predict the three pairs that have been used to hide the
data in each DCT block. Concurrently, the attacker
cannot disturb all the middle coefficients as it will in-
fluence the image badly.
Experimental results indicate that the proposed al-
gorithm is resistant to the common image manipula-
tions such as JPEG compression, filtering and nois-
ing. The results also illustrate that our watermark-
ing scheme does not significantly impede iris image
quality or biometric matching performance. Empir-
ical experiments show that swapping the pairs (2,5)
and (1,6), (3,5) and (2,6), (4,3) and (5,2) in each 8× 8
DCT block are visually imperceptible and maintain
Table 1: EER and PSNR of the extracted watermark after
different manipulations suing the clasical MBCE and the
proposed algorithm.
Mnipulation type
Proposed Method Classical MBCE
JPEG (Q=80) 68.19 0.97% 53.35 5.1%
JPEG (Q=70) 60.13 2.3% 44.35 8.23%
Median filter 55.92 4.2% 44.35 9.11%
73.47 0.29% 60.67 2.83%
Gaussian noise 65.81 1.65% 62.42 2.35%
Salt & pepper 67.63 1.2% 59.31 3.87%
the iris recognition performance.
The proposed watermarking scheme is beneficial
to the biometric system in a number of ways. For
example, the biometric traits and the bio informa-
tion of an individual are usually stored in indepen-
dent databases. Digital watermarking integrates the
biometric trait with the personal information in a sin-
gle file and hence allows the data to be stored and
extracted at the same time. Moreover, the integrity of
the biometric trait can be verified from the extracted
One of the main advantages of our watermarking
scheme is that it can be readily applied to any biomet-
ric image other than the iris image. On the other hand,
the proposed algorithm does not require the original
image for watermark extraction.
Cox, I. J., Kilian, J., Leighton, F. T., and Shamoon, T.
(1997). Secure spread spectrum watermarking for
multimedia. Image Processing, IEEE Transactions
on, 6(12):1673–1687.
Dabas, P. and Khanna, K. (2013). A study on spatial and
transform domain watermarking techniques. Interna-
tional Journal of Computer Applications, 71(14):38–
41. info.
Deb, K., Al-Seraj, M. S., Hoque, M. M., and Sarkar, M.
I. H. (2012). Combined dwt-dct based digital image
watermarking technique for copyright protection. In
Electrical & Computer Engineering (ICECE), 2012
7th International Conference on, pages 458–461.
Fouad, M., El Saddik, A., Jiying, Z., and Petriu, E. (2011).
Combining cryptography and watermarking to se-
cure revocable iris templates. In Instrumentation and
Measurement Technology Conference (I2MTC), 2011
IEEE, pages 1–4. lock the iris with a key and embed
it in a photo.
Hernandez, J. R., Amado, M., and Perez-Gonzalez, F.
(2000). Dct-domain watermarking techniques for still
images: detector performance analysis and a new
structure. Image Processing, IEEE Transactions on,
Hsu, C.-T. and Wu, J.-L. (1996). Hidden signatures in im-
ages. In Image Processing, 1996. Proceedings., In-
ternational Conference on, volume 3, pages 223–226
Isa, M. R. M. and Aljareh, S. (2012). Biometric image
protection based on discrete cosine transform water-
marking technique. In Engineering and Technology
(ICET), 2012 International Conference on, pages 1–
5. Cox watermarking for face image.
Islam, M., Sayeed, M., and Samraj, A. (2008). Biomet-
ric template protection using watermarking with hid-
den password encryption. In Information Technology,
2008. ITSim 2008. International Symposium on, vol-
ume 1, pages 1–8.
Johnson, N. F. and Katzenbeisser, S. (2000). A Survey of
Steganographic Techniques. Artech House Books. PN
Langelaar, G. C., Setyawan, I., and Lagendijk, R. L. (2000).
Watermarking digital image and video data. a state-of-
the-art overview. Signal Processing Magazine, IEEE,
Majumder, S., Devi, K. J., and Sarkar, S. K. (2013). Sin-
gular value decomposition and wavelet-based iris bio-
metric watermarking. Biometrics, IET, 2(1):21–27.
hiding iris image in Lena pic with key.
Masek, L. (2003). Recognition of human iris patterns for
biometric identification. Thesis.
Mukherjee, D. P., Maitra, S., and Acton, S. T. (2004).
Spatial domain digital watermarking of multimedia
objects for buyer authentication. Multimedia, IEEE
Transactions on, 6(1):1–15.
Paunwala, M. and Patnaik, S. (2014). Biometric template
protection with dct-based watermarking. Machine Vi-
sion and Applications, 25(1):263–275.
Sakib, M. N., Alam, S. B., Sazzad, A. B. M. R., Shahnaz,
C., and Fattah, S. A. (2011). A basic digital water-
marking algorithm in discrete cosine transformation
domain. In Intelligent Systems, Modelling and Simu-
lation (ISMS), 2011 Second International Conference
on, pages 419–421. a brief dicussion of watermarking
Singh, A. K., Sharma, N., Dave, M., and Mohan, A. (2012).
A novel technique for digital image watermarking in
spatial domain. In Parallel Distributed and Grid Com-
puting (PDGC), 2012 2nd IEEE International Confer-
ence on, pages 497–501. LSB watermarking with no
Vacca, J. R. (2007). Biometric Technologies and Verifica-
tion Systems. Butterworth-Heinemann.
Wang, W., Men, A., and Chen, X. (2009). Robust image
watermarking scheme based on phase features in dft
domain and generalized radon transformations. In Im-
age and Signal Processing, 2009. CISP ’09. 2nd In-
ternational Congress on, pages 1–5. radon transfrom
with DFT watermarking.
Yiwei, W., Doherty, J. F., and Van Dyck, R. E. (2002). A
wavelet-based watermarking algorithm for ownership
verification of digital images. Image Processing, IEEE
Transactions on, 11(2):77–88.
Zhao, J. and Koch, E. (1995). Embedding robust labels into
images for copyright protection. 211634 242-251.