Fingerprint based Biometric Watermarking

Architecture using Integer DCT

Ayush Vashistha Dr. Amit M Joshi

M.Tech, Electronics & Communication Depart., Assistant Prof., Electronics & Communication Depart.,
Malaviya National Institute of Technology, Malaviya National Institute of Technology,
Jaipur, India Jaipur, India
20 1 4 peb 5 1 83 @mnit.ac.in amjoshi.ece@mnit.ac.in

Abstract- The recent growth in multimedia technologies has purposes. Embedding fingerprint data into host signal (digital
simplified the ways for the transmission, reproduction and image) as a watermark will provide a new level of security.
manipulation of data. This has raised the concern over security Fingerprint data can easily be retrieved and authenticate the
issues of the data transmitted over the channel. Watermarking is owner. The watermarking system should have two main
the practice of embedding information into multimedia object for
properties for authentication applications. i) The features of
authentication and ownership identification. In present time,
host signal may not be affected by embedding process. The
biometric identification has acquired much attention because of
watermarked signal should be invisible to the viewer after
its distinctiveness and reliability. The fingerprints are believed to
be one of the most popular biometric scheme. In the paper, the inserting the watermark. ii) The watermarked signal must be
image watermarking algorithm is developed where fingerprint robust so that it can able to withstand all the channel attacks.
based biometric identification is inserted for authentication. The However, the watermark extraction process should be blind
robustness and invisibility of the proposed method is confirmed where the watermark is retrieved without the requirement of
on MA TLAB platform. The payload of the proposed algorithm is host signal. AI-Gindy et al.[4] proposed the watermarking of
higher than other existing scheme where watermark is embedded signature using spatial domain technique. Many watermarking
in each three color components of an RGB image. The algorithm
algorithms have been designed but only few architectures for
is synthesized on Virtex 7 FPGA family using Xilinx ISE 14.7 to
biometric data watermarking are available. Jain and Uludag [5]
verify the hardware performance.
described the algorithm for the embedding the fingerprint
minutiae data into host image. They used amplitude
Keywords- Blind Detection, Fast Fourier Transform; Integer modulation-based watermarking method based on the blue
DCT; Normalized Corelation; Real Time.
channel watermarking [6]. B1ythe and Fridrich [7] suggested
the secure digital camera system where an Irish data is scanned,
l. INTRODUCTION then subsequently it is embedded into the image.
With the development of multimedia processing In the past few years, several grey-level image
technology, information security has emerged as the main watermarking schemes have been proposed, but their
concern. In the present era, many applications area required the application to color image watermarking schemes is scarce.
security for authentication [1]. There is a necessity of an However, some researchers proposed color image techniques
effective authentication system which helps in an ownership where the watermark was embedded into any of the three
claim of the digital content. Digital watermarking is an colored components. These approaches have a drawback of
effective technique to resolve this problem. Digital limited payload bits. Therefore, the capability of withstanding
watermarking is a technology where a watermark is embedded against the different channel attacks is inadequate. In proposed
into original content for protection of an owner's copyright. algorithm, the watermark is embedded into all the three
The basic idea is to insert some secret message into digital components making it more robust to all types of attacks. The
content which can be extracted later on by owner itself [2]. A paper presents a novel hardware approach based on fingerprint
biometric represents a distinctive, measurable and robust watermarking where binary watermark is inserted in all color
physical characteristics of a personal trait which can be useful (RGB) components of an image. The paper uses simple Integer
for identification and verification of an individual. Among all DCT implementation [8]. Fingerprint based watermark is
the biometric identities, the fingerprint is considered to be most embedded in real-time in RGB components of an image.
reliable and promising personal identification system. The The organization of the paper is as following manner: Section
fingerprint trait is chosen because of its higher accuracy and 11 explains the proposed watermarking embedding process and
reliability. A fingerprint is represented by impression of pattern extraction process. The hardware implementation of the
of valleys and ridges on fingertip. The pattern of valleys and proposed scheme is elucidated in Section Ill. Section IV
ridges along with the minutiae points is unique to each and covers the results section along with analysis. The conclusion
individual [3]. Due to this, it can be useful for authentication is derived in final section V.

978-1-5090-2597-8/16/$31.00 ©2016 IEEE 2818

Authorized licensed use limited to: UNIVERSIDADE FEDERAL DE PELOTAS. Downloaded on July 29,2024 at 13:40:14 UTC from IEEE Xplore. Restrictions apply.
 H. PROPOSED WATERMARKING METHOD A. Watermark Generation

The proposed watermarking is based on Integer DCT This unit mainly deals with generation of binary watennark
technique [9]. The integer DCT technique has reduced from fingerprint image. The image is processed through
computational complexity compare to conventional floating various steps as filtering, nonnalization and binarization. The
point methods [10]. The paper covers an efficient hardware binary output values are generated corresponding to ridges and
design of 2D integer DCT that helps in real time watermark valleys. These binary values are considered as the watennark
embedding process. The inverse Integer DCT algorithm is also during the embedding process.
designed for retrieval process of the watennark. The payload of
B. Watermark Embedding Module
the proposed algorithm is increased which helps to withstand
against several attacks. The embedding module consists of three block. First block
computes Integer 2D-DCT transfonn of 8x8 blocks. Integer
A. Watermark Embedding Algorithm DCT is used to calculate transfonnation in terms of 8 x 8 non
Step I: The fingerprint is binary image which is considered as overlapping blocks. Integer DCT is multiplier less transfonn
a biometric watermark and it has been enhanced using Fast and implemented using Shift and Add Unit (SAU) as shown in
Fourier Transfonn (FFT) filter, segmentation and binarisation. Fig.I. In Fig. l , A indicates the addition process and «

Step 2: The host image is color image and each RGB indicates shifting left. In second block, these transformed
components of host image is divided into 8x8 blocks. values are scaled to the nearest even or odd integers (after
Step 3: Calculate Integer DCT Transfonn of each 8x8 block. quantization). In last block, the inverse transform is computed
Step 4: Embed 16 binary bits of biometric bit-stream into each to give watermarked image.
8x8 block as per following rule shown in eq. (I).

J a* (1(: )
If binary bit = I, then
v»
Fo ifu,v = 0
I'(u, v)
lIeu,v) ifu,v;< 0
else

f '(u,v) = !a* e(:V»)

�
feu,v)
ifu,v = 0

ifu,v;<O
(I)
Where, Fo(x) indicates converting the value of x to most Fig.2 Sift and Add Unit (SAU)

approximate odd nwnber and FI(x) indicates converting the

value of x to most approximate even number. et is the C. Watermark Extraction Module
parameter of quantization. The value of et is selected in the
In this block, the watermarked image is again divided into
range IS < et < 3 5 . In this case, et = 24 which provides invisible
8x8 block, which are fed to 2D-DCT transform block. Even­
and robust watermarking with ease of implementation.
Odd checker is used to check the quantized data and generates
Step 5: Perform integer IDCT to have a watennarked image.
corresponding binary bit. These bits may be stored in RAM
and then subsequently are converted into fingerprint image.
B. Watermark Extraction Algorithm- The overall extraction is shown in Fig. 1

Step l : The watennarked image is converted into 8 x 8 blocks.

Step 2: Again, Integer DCT is perfonned in each block.
Step 3: The watennark bit sequence for fingerprint is retrieved

F{11(:V)}==
as following eq. (2).
Watermarked 8x8 block 2D DCT Even-Odd checker Retrieved
if 1 => data(m)= 1 Image divider Transform and Bit Generator Watermark

else
data(m)=O (2)
CLK
Step 4: Then, binary data bits are used to construct the original
watennark of fingerprint.
Fig.1 Watennark Extraction Process
Ill. HARDWARE IMPLEMENTATION OF PROPOSED
WATERMARKING METHOD The overall watermark embedding process is shown in Fig.3
The hardware implementation consists of three main modules where watermark generation and watermark embedding is
as follows. defined with all the necessary blocks.

2016 IEEE Region 10 Conference (TENCON) - Proceedings of the International Conference 2819
Authorized licensed use limited to: UNIVERSIDADE FEDERAL DE PELOTAS. Downloaded on July 29,2024 at 13:40:14 UTC from IEEE Xplore. Restrictions apply.
 Wawrmllrk GeneratIon

I Fingerprint ~ H Filtering
segmentation &
Normalisation
Binarisation H BR
Generator

eLK ~
Embeddins Process

2D Inteaer DCT

BpolntDCT
~

...... ~

BPolnt

I I DCT
Host Image Buffer
r----

I
1
Bx8 blocks divisor ~
~~
Adder
Unit
4 PolntDCT

SAU
~ Output
Adder
f-o
Unit
SAU

L '---
SAU
SAU ~

'---- '----

20 Invef'5. Intaser OCT Embeddlnc

r--- r---
8 Point
Inverse
BxB
Buffer
B point Inverse OCT r Right Shift!
Divider-

I
r----
DCT
4 Po,nt

r
Inverse OCT
Watermarked
~ I- Odd & Even f- f-

~H
Add-Sub
Imap Conversion
Tree SAU

I
SAU
SAU
SAU Left Shift!
Multiplier
'--- -

Fig. 3 Watennark Embedding Process

IV . RESULTS AND ANALYSIS (5)

The proposed algorithm is validated using MATLAB Here, MAX. is the maximum possible pixel value of the
where color image of Lena (512 x512) is considered as a test image. PSNR is measured between original host image and
image. The robustness of the algorithm is computed by watermarked image. However, NC value is always calculated
Normalized Correlation (NC) and invisibility criteria is between original watermark and extracted watermark. The
calculated using PSNR [11]. Proposed method helps to embed original Lena image, original watermark (fingerprint),
256 x 256 fingerprint watermark into 512 x 512 colored host watermarked Lena image and extracted watermark are shown
image. The definition ofNC is as follows in eq. (3): in Fig.4 [12]. PSNR and NC values without any attacks is
32.50 and 1.0000 respectively.
LL w(m,n)xw'(m,n)
.\'-1.\/-1

(3)
NC == -;====P~";;;,o;f;q";;;,n==========
L w(m,n)xw(m,n)L w'(m,n)xw'(m,n)
X-I .\/-1

p=() q=()

W(m,n) is considered as original watermark, w'(m,n) is

extracted watermark.
Mean Squared Error (MSE) is a measurement of the error
introduced between two images [12]. MSE is defmed as eq. (b) Original Fingerprint Watennark
(4)

MSE==-~-II[Io(k,l)-lw(k,l){
M N k~() I~()
(4)

I" is the original image and I", is the watermarked image

PSNR is inversely proportional to MSE and is defined as
follows in eq. (5)
PSNR==JO*)o (MAXi]
glCl MSE

== 20 * 10 ( MAXI ) blocks of integer DCT. Thus watermark is embedded in each

glO .JMSE
and every bits of 512 x 512 size color image. This high

2820 2016 IEEE Region 10 Conference (TENCON) - Proceedings of the International Conference
Authorized licensed use limited to: UNIVERSIDADE FEDERAL DE PELOTAS. Downloaded on July 29,2024 at 13:40:14 UTC from IEEE Xplore. Restrictions apply.
 payload does not affect the image quality, the proposed algorithm has excellent PSNR even after the increasing
algorithm has admirable PSNR ( greater than 30db in all payload. The higher payload will certainly help to imbibe the
channels) . Various channel noise attacks are considered as watermark against severe attacks.
shown in Table I.
V. CONCLUSION
TABLE I. PSNR AND NC FOR ALL THE COMPONENTS OF COLOR LENA IMAGE
WITH CONSIDERATION OF ATTACKS The paper is based on fingerprinting based biometric
watermarking system for image authentication application.
Lena R component GComponent B Component
The proposed algorithm has an integer 2D DCT architecture
Without PSNR�33.40 PSNR�32.32 PSNR�3I.74
Attack NC � 1 .0000 NC � 1 .0000 NC � 1 .0000 which is multiplier less and has efficient hardware
Gaussian Noise PSNR�30.05 PSNR�29.50 PSNR�29.21 performance. The embedding process is used to embed the
( mean and 0.1 NC-0.941 4 NC-0.936S NC-0.9407 watermark in each bits of every RGB components of color
Variance)
image. Thus payload of the algorithm is higher than any
Salt & Pepper PSNR�2S.01 PSNR�30.2S PSNR�29.75
(0.05 Density) existing watermarking methods. The higher number of
NC � 0.9S9S NC � 0.9353 NC � 0.9392
Clockwise rotation PSNR=19.99 PSNR=20.1 9 PSNR=22.1 O watermarking bits helps to provide the integrity of
(I degree) NC-0.9101 NC-0.9224 NC-0/9242 watermarking algorithm against various attacks.
Anti-clockwise PSNR-20.36 PSNR�20.37 PSNR-22.27
rotation NC � 1 .0000 NC � 1 .0000 NC � 1 .0000 REFERENCES
(I degree)
[I] Evans, Nicholas, Stan Z. Li, Sebastien Marcel, and Arun Ross. "Guest
editorial: Special issue on biometric spoofing and countermeasures."
The excellent values of Table I suggest that proposed algorithm IEEE Transactions on Information Forensics and Security 1 0, no. 4
has capability to survive against various attacks. The hardware (201 5): 699-702.
implementation of the algorithm is done on FPGA platform to [2] Joshi, Amit M., Vivekanand Mishra, and R. M. Patrikar. "FPGA
verify the real-time performance. The algorithm is synthesized prototyping of video watermarking for ownership verification based on
H. 264/AVC." Multimedia Tools and Applications 75, no. 6 (201 6):
on Xilinx ISE 1 4.7 platform using Virtex 7 (XC7AI00T- 3121 -3144.
2CFG324 ) FPGA Family. The device utilization and hardware
[3] Jain, Anil K., and Ajay Kumar. "Biometrics of next generation: An
utilization reports are achieved as shown in Table 11 and Table overview." Second Generation Biometrics 1 2.1 (20I0).
III respectively. The maximum clock frequency of the [4] AI-Gindy, A.; AI-Ahmad, H.; Qahwaji, R.; Tawfik, A., "A novel blind
proposed is obtained around 5 4 . 1 4 MHz for same FPGA. image watermarking technique for colour RGB images in the DCT
domain using green channel," in Communications, Computers and
TABLE n. HARDWARE UTILIZATION OF PROPOSED WATERMARKfNG Applications, 200S. Mosharaka International Conference-CCA 200S

Hardware Utilisation
[5] Jain, Anil K., and Umut Uludag. "Hiding fingerprint minutiae in
images." Proceedings of 3rd Workshop on Automatic Identification
AdderlSubtractors 954
Advanced Technologies. 2002.
Registers 279
Multiplexer 250 [6] M. Kutter, F. Jordan and F. Bossen, "Digital signature of color images
Logic Shifter 36 using amplitude modulation", Proc. SPIE EI,San Jose, Feb. 1 997, vol.
3022, pp. 5IS-526.
TABLE Ill. DEVICE UTILIZATION OF PROPOSED WATERMARKfNG [7] Blythe, Paul, and Jessica Fridrich. "Secure digital camera." Digital
Forensic Research Workshop. 2004.
LOl!ic Utilization Used Available Utilization
[S] Meher, Pramod Kumar, et al. "Efficient integer OCT architectures for
Number of Slice Registers 7463 1 26S00 5%
HEVC." Circuits and Systems for Video Technology, [BEE Transactions
Number of Slice LUTs 29767 63400 46%
on 24.1 (201 4).
Number of fully used LUT-FF pairs 341 S 33S1 2 1 0%
Number of bonded lOBs 41 1 6 210 1 960% [9] Kabi, Kunal Kumar, Bidyut Jyoti Saha, and Chittaranjan Pradhan.
"Blind digital watermarking algorithm based on OCT domain and fractal
Number of BUFG/BUFGCTRLs I 32 3%
images." IT in Business. Industry and Government (CSIBIG). 2014
Conference on. IEEE, 2014.
TABLE lV. COMPARlSM WITH OTHER ALGORlTHM
[1 0] Gupta, Gaurav, Amit Mahesh Joshi, and Kanika Sharma. "An efficient
Algorithm Payload PSNR (db) NC DCT based image watermarking scheme for protecting distribution
(per 8x8 rights." Contemporary Computing (IC3), 2015 Eighth International
block) Conference on. IEEE, 2015.

Proposed Algorithm 64 bits 33.40 1 .0000 [1 1 ] Li, Chunhua, and Zhiying Qin. "A blind digital image watermarking
AI-Gindy et a1.[4] S bit 45.00 NA algorithm based on OCT." Smart and Sustainable City 2013 (ICSSC
Gupta and Amit rlOl I bit 41 .07 1 .0000 2013). lET International Conference on. lET, 2013.
Kabi et al. [9] 4 bit 40.13 1 .0000 [1 2] Rui-mei, Zhao, Lian Hua, and Hu Bo-ning. "A blind watermarking
Li and Qin [1 1 ] I bit 3S.27 0.9769 algorithm based on OCT." intelligent Information Technology
Rui-mei et al. [1 2] I bit 40.50 1 .0000 Application, 2008. UTA'08. Second International Symposium on. Vol. 3.
(NA- Not available) IEEE, 200S.
Table-IV signifies that the proposed algorithm has highest
payload among all exiting watermarking schemes. The

2016 IEEE Region 10 Conference (TENCON) - Proceedings of the International Conference 2821
Authorized licensed use limited to: UNIVERSIDADE FEDERAL DE PELOTAS. Downloaded on July 29,2024 at 13:40:14 UTC from IEEE Xplore. Restrictions apply.