Learning Center
Plans & pricing Sign in
Sign Out

A print-and-scan resilient digital watermark for card



     A Print-and-scan Resilient Digital Watermark for
                   Card Authentication
                                             Anthony T. S. Ho and Feng Shu
                                      School of Electrical and Electronic Engineering
                                            Nanyang Technological University
                                           Nanyang Avenue, Singapore 639798

   Absbaef-In this paper, a secnm card authentication system           The host data in our watermarking system are face pho-
based on d g t l watermarking is pmented. A sequence of
             iia                                                    tographs. The size of the standard photos on the security
identification characters e.g. owner’s ID number is inserted
into the photograph, thus a linkage between the photograph          cards, such as student cards, are around 1x1 inch (we use
and its owner c m he established. The print-and-scan process is      180x130 pixels photos), which is much smaller than the host
successfdy uvemme by our transform d a e n t polarity based         images used in most image watermarking algorithms proposed
watermarking algorithm. The retrieval algorithm is capable of       in the literature. Face photos are typically very smooth, which
verifying whether the inspected card is authorized. We apply onr    are much easier to introduce degradation during watermark
algorithm for 105 photographs and the success rate achieved is
better than 96%.                                                    embedding. Fnrthermore, a multi-hit watermark is hidden into
                                                                    the smooth face photos of small size, while the photo integrity
                      I. INTRODUCTION                               must be maintained at the same time.
                                                                       Our paper is organized as as follows. In section II, we
   In recent years, the rapid development of digital watermark-     introduce the framework of our proposed card authentica-
 ing has led to a significant number of watermarking algorithm      tion system, the print-and-scan process and the Hadamard
have been proposed in the literature. Apart from copyright          transfom-based watermarking embedding and retrieval algo-
protection, which has attracted much research attention, the        rithms. Experimental results are presented in section lll, and
 other application in digital watermarking which has also           conclusions in section N.
received a considerable interest is image authentication.
   In a watermarking authentication system, a sequence of              11. O U R PROPOSED CARD AUTHENTICATION SYSTEM
identification or a visual pattern, that is, the watermark, is
                                                                    A. Framework of card authentication system
 imperceptibly embedded into the host media. The watermarked
media data generally goes thmugh some distortion chan-                 In this section, the framework of ow proposed authentica-
 nel. The recovered watermark from inspected data is used           tion system is introduced. The watermark embedding process
for authentication in different ways, such as localizing the        is illustrated in Fig. 1.
distortions([11, [2]), determining the type of attacks performed       The Owner’s photo is first scanned into the computer. The
[3] or verifying the genuineness of the data.                       scanned photo will be the host image for the embedding algo-
   In this paper, we devise a watermarking system to authen-        rithm. The watermark is a sequence of personal information (a
ticate cards with the owner’s photograph. These cards include       sequence consisting of combination of characten and numbers
student cards, identity cards, passports and driving licenses.      is used here). The embedding algorithm is based on the
Nowadays, with some sophisticated editing tools (for example,       polarity of Hadamad transform coefficients, that determines
Photoshop), and printing and scanning facilities, it is extremely   the insertion of the watermark sequence into the host image.
easy to modify or replace the original photograph, and to make      The Watermarked image and a secret key, are then generated.
fake security documents or cards [6]. It is reported that two-      The watermarked photo is printed on the owner’s security
thirds of passport frauds are due to photograph substitution        document or card. The retrieval and verification process, as
~71.                                                                depicted in Fig. 2, is performed on the scanned photo of an
   To prevent such photograph substitution frauds, in the           inspected card for authentication.
proposed watermarking system, we directly establish a linkage
between the owner’s photo and hisher personal information           B. Print-and-scan process
by imperceptibly superimposing the personal identification             There is currently very limited research has being performed
information, considered as the watermark, into the original         on the print-and-scan process. After a digital image is printed-
photograph. The main challenge in the proposed system is to         and-scanned, various distortions are introduced. Distortion
overcome the print-and-scan (PS) process, which is regarded         could occur in both the pixel values and the geometric bound-
as a combination of different attacks [8].                          ary of the rescanned image [8].

 0-7803-8185-8/03/$17.00 2003 IEEE
                      0                                        1149


                                                           Fig. 1. Embedding process.

                              +                                                                             Watermark:
                                                                                                              & &
                                                                                                             & &

                                                     Fig. 2 Retrieval and verification process.

                                       ~..~.. .. .           Printer        Scanner

                                        Dipitat                    Print-scan                     Digital
                                                                    process                       Image

                                                          Fig. 3. Print-and-scan pccess.

   Typical distortions of pixel values could be caused by the lu-           C. Spread spectrum based watermark embedding and re-
minance, contrast, gamma correction, chrominance variations                 trieval
and the blurring of adjacent pixels. Distortion of the geometric
                                                                               In this section, the proposed watermark embedding and
boundary in a DAD (digital-analogue-digital)process is caused
                                                                            retrieval algorithm is presented.
by rotation, scaling and cropping. The print-and-scan process
is shown in Fig.3.
                                                                               I ) Watermark embedding algorithm: The spread spectrum
                                                                            based watermark embedding algorithm is depicted in Fig.4.
                                                                               First, the watermark bit, either 1 or 0, is expanded
                                                                            into a binary sequence in the following manner: if the
   The PS process could he modeled and explained by dividing                watermark bit is 1, then it is represented by data w , ~ :
into two scenarios. For the first scenario, rotation and cropping           11, -1,1, -1,l, -1,1, -1, ...} and the length of this sequence
a e included in the PS process. Scenario 1 PS is usually
 r                                                                          is m; if the watermark bit is 0,then it is represented by
performed by an adversary, thus such PS is considered as a                  data wm0: {-1,1, -1,1, -1,1, -1, ...} and the length of this
malicious combined attack. While in Scenario 2, rotation and                sequence is also m.
cropping are not taken into account. To overcome Scenario                      Then we transform the original image. A Hadamard trans-
I PS, techniques capable of conquering rotation, scaling                    form is used in our algorithm, due to its simplicity and ease
and wanslation (RST) must be simultaneously included. The                   for hardware implementation [5]. For one bit of watermark,
RST attacks, however, still remain an unsolved problem in                   that has been expanded into a sequence with length of rn , we
watermarking, although some techniques were proposed in the                 need to use the same amount of coefficients for insertion. The
literature [4]. In OUT system, the PS is considered as an inherent          embedding process can he divided into two passes:
system distortion rather than a malicious attack, thus Scenario                (1)Polarity modulation;
2 PS process is involved.                                                      (2)Amplitude modulation.

                     Watermark           Watermark
                      (binaiy)           expansion
                                                                 Pass 1:
                                                               modulation             modulation                      image
                      host image         exiraction

                                             Fig. 4. Block diagram of watermark embedding algorithm.

                                   Transformed                                 Decision            Retrieved
                                    host image          extraction                                 watermark

                                                                               wl and w0

                                             Fig. 5 . Block diagram of watermark extraction algorithm

   Consider that we embed one watermark bit. At first, we                   where n is defined as a certain step size, 1 is the ceil func-
obtain the polarity sequence of the selected transform coeffi-              tion, and c and c‘ denote original Coefficients and modulated
cients, as follows:                                                         coefficents, respectively . We define the sign(c) as follows:

                                 1, if    c[i] 2 0
                                 0, if    c[i] < 0
where c[i] (15 i 5 m) denotes selected coefficients.
   For the first pass, we modulate the coefficients’ polarities                By using Equation (4), the coefficients’ amplitudes are
in order to make the correlation value between s, and w    ,
                                                                            increased to the range [Tz,Tz %I.
or wm0 satisfying the following equation:
                                                                               2) Watermark retrieval algorirhm: To extract the water-

                                                                            mark, we again perform a Hadamard transform to the scanned
                                                                            photograph, and the pseudorandom generator seed is obtained
  mT={            v=
                   I”rnD,                                                   from the secret key to find the locations of the coefficients
                                                                            containing the watermark. We define these coefficients as e, :
where T1 denotes certain threshold for the correlation. In our              . We extract the polarities of c: and denote polarity sequence
experiments, we set T1 as 0.3. If the original polarity sequence            as s:, . The extraction is performed by the correlation between
does not satisfy Equation (2),some of the coefficients’ polar-              s and w , ~ or W
                                                                             ,                O,   ,
ities would be changed until the modified polarity sequence
satisfies the condition. However, if we change the polarities
of the coefficients having large amplitude, the image quality
is easily degraded. Thus, the polarities of those transform
coefficients having smaller amplitude should be changed with
higher priority.                                                                             W T ;   =            S r n G l
   In the amplitude modulation pass, we increase the amplitude                                                                           (6)
of all the selected coefficients up to the prescribed threshold
Tz, with the polarity of coefficients maintained. After a wa-               W T ; and W T ;   measure the possibility that the embedded bit
termarked image has gone through some distortion channels,                  is 1 and 0, respectively. The decision rule of the watermark
the transform coefficients would be altered. Such alteration,               bit is given as follows. k denotes the extracted watermark bit.
however, is generally within a certain range [9]. Therefore, if
the amplitude of a coefficient is increased to be greater than the
                                                                                                        1, if W T T ~ 2 W T ;
range, it has a higher probability to maintain its polarity under
certain attacks such as JPEG compression, median filtering,                                 k=     { 0,    if m ~> CGTT;
although its amplitude would still be altered. The equation for
amplitude modulation is defined as follows.                                   For detection, the detector does not require both the original
                                                                            image and the watermark. Such watermarking schemes are
                                                                            called public watermarking schemes or blind watermarking
                                                                            schemes [lo].

                   Original         Watermarked                   scanned                                          Verification
                    photo              photo                        photo                  Watermark
                  (180x130)                                                                                          CleSlNo)
                                      (180x130)                  (215xlM))

                                                                                            AAAAA1111                 Yes


                                                                                           ccccc3333                  Yes

                       Fig. 6. Samples of experimental results.(The characters used here are not real identiry numbers.)

               111. EXPERIMENTAL RESULTS                                105 different face photographs. the print-and-scanprocess. The
   In our experiments, we used a common laser printer (HP               proposed system can be used to authenticate student cards,
LaserJet 4MV) and scanner (CanoScan N1220U). with resolu-               passports, driving licenses and other security documents or
tions set to 170 dpi (dots per inch) and 200 dpi, respectively.         cards.
For subblocking the image, an 8x8 Hadamard transform was                                               REFERENCES
used. The watermark expansion length was set to 8, and
                                                                         [I] R. B. Wolfgang and E. I. Delp, "Fragile watermarking using the VWZd
Ti in polarity modulation and Tz in amplitude modulation                     watermark", Pmceeding o the SPlUISdrT Intemtioml Conference on
were set to 0.3 and 1.5, respectively. A tradeoff between the                                                f
                                                                             Security and Wotemrking o Multimedia Contents, vol. 3657. pp. 204-
robusmess and image quality was determined in selecting T I                   213. Ian. 1999.
                                                                         [Z] 1. Hu, I. Huang, D. Huang and Y. Q. Shi. "Image kagile watermarking
and Ts.These empirical values were obtained based on both                     based an fusion of multi-resolution tamper detection," IJ?E Electronic
the acceptable image quality and robusmess against the PS                     Leuers,vol. 38, no. 24, pp 1512-1513, Nov. 2002.
process. We tested our authentication system for 105 different           [3] D. Kundur and D. Hauinakas, "Digital watermarking far telltale tamper-
                                                                              prwhng and authentication:' Pmceedings o the IEEE Special Isme on
face photographs and the success rate was better than 96%.                    Identificntion old Pmtection of Multimedin Infomtion, vol. 87, no. 7.
The failures were mainly due to manual misalignment, because                  pp. li67-1180, July 1999.
our algorithm was not so resistant to rotation and translation           [4l C:Y. Lin. M. Wu. 1. A. Blwm, 1. I. Con, M. L. Miller, and Y M. Lui,
                                                                              "Rotation, scale. and translation resilient watermarking for images," IEEE
attacks. Some examples of the experimental results are listed                 Tms. I m g e Pmcesring, vol. 10, no. 5. May 2001, pp. 767 -782.
in Fig. 6.                                                               151 A. T. S. Ho, J. Shen, and S. H. T n "Robust digital image-in-image
                                                                              watermarkine alearithm usine the fast Hadamard transform:' SPIES471h
                       IV. CONCLUSION                                        A n n u l Mee&"on     Optical ?cience ond Technology, July 2002, Seattle,
   We proposed a card authentication system based on digital                  USA.
                                                                         161 A. T. S. Ho, I. Shen, H. P. T n and 1 Wwn. "Security-pnnting
                                                                                                                  a,          .
watermarking. A linkage between the bearer's photograph and                   authentication using digital watermarking,'' Electmnic Imoging. vol. 13,
hisiher personal information is established through superim-                  "0.1, Jan. 2003.
posing the personal information directly into the photograph             171 I. Mercer, Authenlicotion News, 5 (9/10), 2001.
                                                                         181 C.-Y. Lin and S.-F. Chang, "Distortion modeling and invariant extraction
as a watermark. As such, photograph substitution frauds could                 for digital image print-and-scan process," Intl. Symp. on Multimedia
be prevented using this approach. The watermarking algorithm                  lnfomrion Pmcessing, Taipei, Taiwan, Dee. 1999.
                                                                         [Y] F. Shu, A. T. S . Ho and I. Shen, "Digital Watermarking Techniques Using
was based on the polarity of Hadamard transform coefficients,                 Polarity of Transform Coefficients:' Accepted by IASTED International
that could invisibly insert a 9-character watermark into smooth               Conference on Signol and l m g e Pmcessing,Aug. 2003. Honolulu, USA.
photographs of size 180x130 pixels. This algorithm success-              [IO] S. Katzenbeisser and A. P. F. Petitcolas, Infomtion Hiding Techniques
fully survived with a success rate achieved better than 96% for              for Sreganogrophy ond Digital Wotemrking, Artech HOUSC,          2wO.


To top