Securing Biometric Images using Reversible Watermarking

Sabu M. Thampi & Ann Jisma Jacob
International Journal of Image Processing (IJIP), Volume (5) : Issue (4) : 2011 382
SECURING BIOMETRIC IMAGES USING REVERSIBLE
WATERMARKING
Sabu M. Thampi sabum@rajagiritech.ac.in
Rajagiri School of Engineering and
Technology, Kochi, India
Ann Jisma Jacob annjisma@yahoo.com
Indira Gandhi National Open University,
India
ABSTRACT
Biometric security is a fast growing area. Protecting biometric data is very important since it can
be misused by attackers. In order to increase security of biometric data there are different
methods in which watermarking is widely accepted. A more acceptable, new important
development in this area is reversible watermarking in which the original image can be
completely restored and the watermark can be retrieved. But reversible watermarking in
biometrics is an understudied area. Reversible watermarking maintains high quality of biometric
data. This paper proposes Rotational Replacement of LSB as a reversible watermarking
scheme for biometric images. PSNR is the regular method used for quality measurement of
biometric data. In this paper we also show that SSIM Index is a better alternate for effective
quality assessment for reversible watermarked biometric data by comparing with the well known
reversible watermarking scheme using Difference Expansion.
Keywords : Biometric, Reversible Watermarking, Difference Expansion, SSIM Index, Image Processing
1. INTRODUCTION
Applying security along with privacy is challenging in this digital age. Conventional methods are
defeated by biometric authentication as it is more convenient for privacy protection. But along
with convenience it has introduced new threats. It is prone to attacks since the data will not be
renewed so often. Thus securing biometric data has become crucial. Though there are different
methods being introduced for this, watermarking plays an important role here in protecting data.
Watermarking hides a message or image in some data to obtain a new data so that the hidden
data is indistinguishable from the original data so that an attacker cannot remove or replace the
message from the new data. Thus it has become a major part of information hiding.
Watermarking is used to pass hidden messages in communications. Watermarking can be done
in different domains like spatial or frequency domain. In Spatial domain actual pixel values are
changed where as in the frequency domain, coefficients used in transformed image
representations are changed. There are some techniques, which uses LSBs for embedding
watermark data. In most of these, the changes made to the original data become irreversible.
There is yet another way in which the original image and the embedded payload can be
perfectly recovered later by watermarking in some particular way. This technique is called
reversible watermarking.
In Reversible watermarking, a watermark is embedded into an image and from the watermarked
image; original image and the watermark are retrieved. The retrieval process does not need the
original image and so this technique is said to be blind [7]. There are different types of
reversible watermarking based on the technique they use such as Data Compression,
Difference expansion, Histogram operation and LSB Replacement [8]. Figure 1 shows the
difference between the conventional watermarking techniques and the reversible watermarking
techniques. This paper focuses on reversible watermarking using difference expansion and LSB
replacement.

FIGURE 1: Difference between conventional and Reversible watermarking system
Quality of watermarked images plays an important role in biometrics. Since watermarking adds
noise to the data, it should be done in such a way that it will not defeat the purpose of
watermarking. There are quite a few number of techniques introduced for watermarking but the
quality of watermarking is usually measured using PSNR. PSNR is a simple measure of image
quality based on average error. This technique is known to be not very accurate [5], [6]. Despite
its drawbacks, however, PSNR is still heavily reported because it is easy to compute. This
paper introduces SSIM Index as a better way of quality measurement for biometric watermarked
images. For this, the proposed rotational replacement of LSB scheme is being compared with
the well known reversible watermarking scheme using Difference Expansion.
2. PREVIOUS WORK
Earliest Reversible watermarking scheme was invented by Barton [3]. DE is the most
established method for reverse watermarking [2]. Reversible Watermarking using DE was first
proposed by Tian [2]. Tian used regular images for his experiments. This paper uses DE
method for watermarking biometric image.
PSNR is the common quality measurement technique used for watermarked images as it is
easy to calculate and optimize. But they are not very well matched to perceived visual quality
[1]. SSIM (Structural SIMilarity) Index is based on Full-reference image quality assessment [1].
SSIM Index compares local patterns of pixel intensities that have been normalized for
luminance and contrast [1]. Since SSIM Index is designed to improve on traditional methods
like PSNR, measuring the quality using SSIM Index will add value to the reversible
watermarking technique.
In this paper, reversible watermarking is done using proposed rotational replacement of LSB
and also using Tian’s reversible watermarking technique. The quality of watermarked images is
assessed using SSIMIndex and compared.
3. WATERMARKING USING DIFFERENCE EXPANSION
Tian [2] introduced the Difference Expansion technique. In this technique, pairs of pixels of the
host image HI are identified and they are converted to a low-pass image LI using integer
transformation with integer average α and a high-pass image HI containing the pixel
differences δ . If x and y be the intensity values of a pixel-pair, then α and δ are defined as




 +
=
2
)( yx
α (1)
δ = x – y (2)
x and y can be computed from α and δ as



 +
+=
2
)1(δ
αx (3)
2
δ
α −=y (4)
By appending an information bit i to the LSB of the difference δ, a new LSB can be created. The
watermarked difference is
δw = 2δ + i (5)
The resulting pixel gray-levels are calculated from the difference (δw) and integer average α
using (3) and (4).
For an image with n-bit pixel representation, the gray levels satisfy ]12,0[, −∈ n
yx , if and only
if α and δ satisfies the following condition:
)]12),12(2min(,0[)( +−−=∈ ααδ lR n
d (6)
Where Rd(α) is called the invertible region. Combining (5) and (6) we obtain the condition for a
difference δ to undergo DE.
)(2 αδ dRi ∈+ for i = 0,1 (7)
This condition is called the expandability condition for DE. If an integer average is given and the
difference satisfies the expandability condition, it is called an expandable difference.
Another method used by Tian in his Difference Expansion technique, other than embedding
which is already discussed is replacing LSB. Here, an information bit will be used to replace the
LSB of the difference. Since the LSB is replaced in the embedding process, this cannot be
considered as lossless as the first embedding technique. However, here the information about
the true LSBs of the differences that are embedded by replacing LSB are saved and embedded
with the payload, to ensure that it is not lossy.
The LSB of a difference can be flipped without affecting its ability to invert back to the pixel
domain if and only if
)(
2
2 α
δ
dRi ∈+



for i=0, 1 (8)
This is called the changeability condition. A difference satisfying the changeability condition,
given a corresponding integer average, is called a changeable difference. An expandable
difference is also a changeable difference. A changeable location remains changeable even
after its LSB is replaced, whereas an expandable location may not be expandable after DE, but
it remains changeable.

4. WATERMARKING USING ROTATIONAL REPLACEMENT OF LSB
In common LSB replacement techniques, the LSB of the original image is replaced with the
MSB of the watermark. For this, first the number of bits needs to be required is calculated and
those many LSB are replaced with MSB of watermark. To retrieve the watermark, first get the
LSB according to the number used and then use them to create new image by changing them
to MSB. Here, the restored image will be low in quality. In Rotational replacement of LSB, same
bit will be used to store secondary LSB of the original image and the watermark.
For embedding watermark using rotational replacement of LSB, first create matrices of original
image and watermark. Assuming original image is X × Y matrix and watermark is x×y matrix.
The ratio γ = (M×N)/m×n should be greater than or equal to 8, so that watermark can be
embedded using this technique. The original image matrix is divided into blocks. Thus if γ ≥64,
the block size will be 8×8. Only one byte of watermark data can be embedded to such a block.
Consider such a block. The last row of the original host image will be replaced by the last but
one row data. This will be continued rotationally so that the first row of the original matrix will be
placed at the second row and first row will be free. One byte of the watermark data will be
placed in this row.


























11000001
10001000
11100010
00110100
11011011
01010101
10101010
11111111
+ [ ]10101010


























10001000
11100010
00110100
11011011
01010101
10101010
11111111
10101010
FIGURE 2: Rotational replacement of LSB while embedding
Extraction will be similar to embedding process where, using the γ value, blocks are
determined. Based on the size of the watermarked data, first row LSB will be used to
reconstruct the watermark. After removing the first row, the remaining rows are rotationally
replaced in the reverse order of embedding. Last row will replace the data of the same row.
5. STRUCTURAL SIMILARITY INDEX (SSIM INDEX)
SSIM Index is a framework for quality assessment based on the degradation of structural
information introduced by Zhou Wang [1] et al. For human visual system a calculation of
structural information difference can provide a good approximation to the image distortion
perceived. The product of the illumination and the reflectance gives the luminance of the
surface of an object, but the structures of the objects in the scene are independent of the
illumination [1]. SSIM Index defines the structural information in an image as those attributes
that represent the structure of objects in the scene, independent of the average luminance and
contrast [1].

FIGURE 3: Diagram showing calculation of SSIM Index
[1]
In SSIM Index, the similarity measurement is done by making comparison of luminance,
contrast and structure. Let x and y be two image signals.
Luminance of each signal is calculated as:
i
n
i
x x
N Σ=
=
1
1
µ (1)
Signal contrast is calculated as:
2
1
2
1
))(
1
1
( xi
n
i
x x
N
µσ −
−
= Σ=
(2)
The structure comparison is calculated as:
xxx σµ /)( − and yyy σµ /)( − (3)
Combining these three, the SSIM index is calculated as:
SSIM(x,y) =
))((
)2)(2(
2
22
1
22
21
CC
CC
yxyx
xyyx
+++
++
σσµµ
σµµ
Where C1 & C2 are included to avoid instability when
22
yx µµ + is very close to zero.
C1= (K1L)
2
where K1 = 0.01
C2= (K2L)
2
where K2 = 0.03
L = dynamic range of the pixel values.
6. DESIGN OVERVIEW
The novelty of this paper is in introducing Rotational Replacement of LSB as a reversible
watermarking scheme and proving SSIM Index as a better alternate for measuring quality of
biometric data. Since PSNR is easy to calculate, it is the common method used everywhere,
where as the experiments conducted show that SSIM Index can be used as a better and
effective alternate. The system uses Biometric image (fingerprint) for watermarking using
Rotational Replacement of LSB and also using Difference Expansion. These reversible
watermarking techniques reduce the noise while watermarking which is highly recommended for
biometric images. The watermarked fingerprint image is used for calculating SSIM index. SSIM
Index is calculated for watermarked images with different payload.

.
FIGURE 4 : Calculation of SSIM Index of watermarked image
7. EXPERIMENTAL SETUP AND RESULTS
The experiment was conducted using a single fingerprint image with different payloads since
the quality of image is affected more as the payload increases. The finger print was collected
from internet [http://www.nist.gov] and it is of the resolution 512*512. The experiment was
implemented in Java using Image APIs. Watermarking was done using Rotational Replacement
of LSB and DE for different payload. SSIM index for the watermarked images are found. The
goal is to prove that the SSIM Index based quality measurement is a better alternate than
PSNR for biometric image watermarking and also that Rotational replacement of LSB is a better
reverse watermarking scheme than DE.
Fig 5 shows the original image and the watermarked images used for the experiment using
rotational replacement of LSB. As it can be seen, the images are visually not tampered by
watermarking.
FIGURE 5: Watermarked samples with different payload
Experimental results show that Rotational replacement of LSB is a better reverse watermarking
scheme than DE by measuring the quality using SSIMIndex. It is also proved that the execution
time for Rotational replacement of LSB is much less than DE and payload capacity of Rotational
replacement of LSB is more than DE

The SSIM Index for all the watermarked images mentioned here were found out. Figure 6
shows that SSIMIndex for DE watermarked images has a lower value indicating more noise and
Figure 7 shows that the Rotation LSB replacement is faster than DE for a given payload.
Payload DE LSB
128 1.00000000 1.00000000
256 1.00000000 1.00000000
512 1.00000000 1.00000000
1024 1.00000000 1.00000000
2048 1.00000000 1.00000000
4096 1.00000000 1.00000000
8192 1.00000000 1.00000000
16384 1.00000000 1.00000000
32768 0.99999990 1.00000000
65536 0.99999936 1.00000000
TABLE 1: Value of SSIMIndex for different payloads using DE and RRL
Payload DE LSB
128 3954 688
256 3937 687
512 3953 719
1024 3968 672
2048 4125 703
4096 3984 672
8192 3985 703
16384 4500 687
32768 4953 703
65536 6906 672
TABLE 2: Value of execution time for different payloads using DE and RRL

FIGURE 6: Graph corresponding to Table 1
FIGURE 7: Graph corresponding to Table 2
8. CONCLUSIONS AND FUTURE WORK
Biometric images can be secured using watermarking. Reversible watermarking helps us to
have images with minimum distortion. At the same time, the original image can be retrieved for
any further use. Rotational Replacement of LSB is a high performing scheme for moderate
payload. SSIM Index is proved to be more accurate for measuring the quality of watermarked
image. Consequently, in the biometric image security field SSIM Index can be used instead of
PSNR to get better result. For reversible watermarking, there is a limitation of maximum
payload, depending on the capacity of original image being used. Our future work will be
focussing on calculating the performance of feature level extraction for a watermarked image
using rotation replacement of LSB.
REFERENCES
[1] Z. Wang, A. Conrad, H. Rahim, and E. Simoncelli, “Image Quality Assessment: From Error
Visibility to Structural Similarity,” IEEE Trans. Image Processing, Vol. 13, No. 4, April 2004.
[2] J.M.Barton, “’Method and apparatus for embedding authentication information within digital
data,’’ U.S.Patent 5 646 997.1997.
[3] J. Tian, “Reversible data embedding using a difference expansion,” IEEE Trans. Circuits
Syst. Video Technol., Vol. 13, pp. 890–896, Aug.2003.
[4] A. M. Alattar, “Reversible watermark using the difference expansion of a generalized
integer transform,” IEEE Trans. Image Process., Vol.13, no. 8, pp. 1147–1156, Aug. 2004.
[5] Multibiometric Authentication- An Overview of Recent Developments -Term Project CS574,
Spring 2003, Available: http://www.ub-net.de on April 2010.
[6] Ya-Hui Shiao, Tzong-Jer Chen, Keh-Shih Chuang, Cheng-Hsun Lin and Chun-Chao
Chuang, “Quality of Compressed Medical Images,” Journal of Digital Imaging, Vol. 20, No.
2, 149-159, DOI: 10.1007/s10278-007-9013-z.
[7] Jen-Bang Feng, Iuon-Chang Lin, Chwei-Shyong Tsai, and Yen-Ping Chu, “Reversible
Watermarking: Current Status and Key Issues,” International Journal of Network Security,
Vol.2, No.3, PP.161–171, May 2006.
[8] Lixin Luo, Zhenyong Chen, Ming Chen, Xiao Zeng, and Zhang Xiong, “Reversible Image
Watermarking Using Interpolation Technique,” IEEE Transactions on information forensics
and security,Vol.5,No.1, March 2010.

Securing Biometric Images using Reversible Watermarking

More Related Content

What's hot (19)

Similar to Securing Biometric Images using Reversible Watermarking (20)

Recently uploaded (20)

Securing Biometric Images using Reversible Watermarking