Jin-Wook Shin, Jucheng Yang, Dong-Sun Park
Dept. of Infor. & Comm. Eng., Chonbuk Nat'l University, Jeonju, Jeonbuk, 561-756, Korea
Sook Yoon
Dept. of EECS, University of California at Berkeley Berkeley, CA 94720
Keywords: Digital content, Copyright protection, Extra scrambled information, Watermarking, Fingerprinting.
Abstract: Both watermarking and fingerprinting techniques can be used for protecting digital contents with
different properties. A watermarking system may degrade the fidelity of the digital contents by
embedding watermark messages, while a fingerprinting system may have high computational
complexity to generate unique features for digital contents. In this paper, we propose a novel
copyright protection technique that combines positive features of both techniques. The proposed
technique can distribute digital images without embedding messages related with them, and save extra
scrambled information on simple fingerprints stored in a certified database. Experimental results show
that the proposed method outperforms an existing method for various signal processing attacks. The
proposed technique is also flexible and fast so that it can be used for real-time applications.
The explosive growth in Internet and its supporting
digital technology for the last decade brings the
production and spread of huge amount of
multimedia data. Since it is very easy to copy, edit,
save and/or transmit digital data, illegal operations
on digital data, such as illegal copy and distribution,
have been frequently conducted without having any
charges against those actions. Protection of
intellectual property rights, therefore, is one of the
major issues in the current Internet era.
Cryptography (Cox, 2002) can be used to protect
digital data during transmission through digital
networks from network eavesdroppers, but it is not
appropriate for cases of illegal copy and
distributions. Recently, digital watermarking
techniques are focused as the copyright protection
technique for those cases. Using this technique, we
can embed information of manufacturers or
authorized users inside the digital contents to
prevent the illegal actions. The basic requirement of
a watermarking technique is to minimize the
degradation of the original contents when
embedding a watermark message and the watermark
itself should be unperceivable to other users.
Depending on different applications, a watermarking
system should have required properties including
fidelity, robustness against various signal processing
attacks, computational complexity and cost.
Various researches have been performed on
developing efficient watermarking systems (Huang,
2004) (Schydel, 1994). Most watermarking systems
developed so far one of two types depending on the
method of embedding watermark messages : spatial
domain and frequency domain methods.
Since watermarking techniques embed
watermark messages in the digital contents, slight
degradation of quality cannot be avoided. If a digital
contents experience many signal processing
operations during transmission and distribution, the
embedded watermark message may be severely
affected and hard to recover.
A fingerprinting technique (ISO/IEC 21000-11,
2004), included in MPEG21 part11, extracts unique
features from digital contents using computer vision
techniques, saves them in a database, and uses them
when there is a need to prove the identification of
the digital contents. Since this method distributes the
Shin J., Yang J., Park D. and Yoon S. (2006).
In Proceedings of the First International Conference on Computer Vision Theory and Applications, pages 27-32
DOI: 10.5220/0001367100270032
original contents without modifying any values, it
may endure more changes during distribution.
However, this technique is usually not flexible
because it cannot embed external information.
Moreover, it is usually time-consuming due to the
high computational complexity implementing
computer vision techniques.
In this paper, we propose a novel copyright
protection technique for digital images using
positive features of both watermarking and
fingerprinting techniques. The proposed method
uses external copyright messages and original digital
contents to generate extra scrambled information.
Random number generator is used to randomly
select the elements of digital contents. The generated
extra scrambled information contains the
information of both external copyright messages and
original digital contents. The generated information
can then be stored in a certified database with initial
seeds for random number generator. The original
digital contents can be distributed without having
any changes in element values. If there is a need to
identify either the external copyright message or the
original contents, the stored extra scrambled
information can be used for identification. The
proposed technique generates and saves extra
scrambled information as a function of external
copyright message and original contents, and
distributes the original contents. Therefore, there is
no degradation in fidelity on the distributed contents.
In addition, in generating the extra scrambled
information we use a simple and fast spatial domain
insertion method so that the proposed technique has
a very low computational complexity.
This paper is organized as follows. In section 2,
related techniques including watermarking and
fingerprinting models briefly described. The
proposed technique is explained in detail in Section
3. Experimental results and conclusion are described
in Section 4, 5, respectively.
2.1 Watermarking
In typical watermarking systems, a watermark
message which contains information on
manufacturers or authorized users is embedded in
the digital contents and distributed to the outside
world. The embedded watermark message should
not be perceivable not to degrade the quality of the
digital contents. The embedded watermark message
can be extracted to resolve legal disputes on the
digital contents such as illegal copy, modification
and distribution.
Fig. 1 shows the conceptual block diagram of a
typical watermarking system. It consists of a
embedder and a detector. The embedder embeds the
watermark message to the original content and the
watermarked contents are distributed. The
watermarked contents can be modified by various
signal processing operations or malicious attacks.
The watermark detector is to detect a watermark
message form the corrupted version of the digital
contents. Various techniques have been developed
for the better performance depending on the specific
applications (Nikolaidis, 2001) (Fazam, 2001).
Figure 1 : General watermarking system.
The properties required for a watermarking
system can be varied according to the applications.
Some properties including fidelity, robustness,
computational complexity, and informed/blind
detection can be found in (Cox, 2002).
2.2 Fingerprinting
A fingerprinting technique (ISO/IEC 21000-11,
2004) is can be also used to protect digital contents
from illegal uses as in the watermarking system. The
main difference between two methods in that a
fingerprinting system extracts features from digital
contents and uses the features for identifying the
digital contents while a watermarking system
embeds an external watermark message in the digital
contents. Therefore, this fingerprinting technique
does not embed any information on the digital
The conceptual block diagram of a fingerprinting
system is shown in Fig. 2. It consists of two
fingerprint generators and a comparison part. The
technique usually employs computer vision
techniques to generate an invariant fingerprint from
the digital contents.
Figure 2: Fingerprinting system.
The generated fingerprint is stored in a database
and then it can be used for comparison whenever a
legal dispute happens. As in the figure, the
fingerprinting technique does not embed any
information in the digital contents. The digital
contents is not modified and distributed to outside so
that there can be no degradation in the contents.
Since this technique usually uses time-consuming
technique to find unique fingerprint, it is very hard
to apply them to real time applications.
The method proposed in this paper combines the
features of watermarking and fingerprinting. The
method extracts some simple information such as
intensity value from original digital contents and
distributes the original contents as in the
fingerprinting technique. In the meanwhile, it also
generates extra scrambled information with external
copyright message and the information extracted
from the contents.
Fig. 3 shows the block diagram of the method. In
the figure, original digital contents are used to
generate extra scrambled information as a function
of external copyright message and random number
sequences. The resulting extra scrambled
information and additional parameters of the
generator may be stored in a certified database for
later identification purposes. The original contents
may then be distributed without changing any
temporal or spectral information. The stored
information can be used whenever there is a need to
prove the identification of the digital contents or the
existence of the copyright message in the stored
information. The detection process uses the
information stored in the database and the same
procedure to generate the extra scrambled
information for the copyright message. In this
method, we used a pseudorandom number generator
and a spatial domain to generate to extra scrambled
3.1 Extra Scrambled Information
Extra scrambled information is a binary data that
combines the information from the external
copyright message and the original contents. It
represents the copyright message or the original
work depending on applications. In other words, it
can be used to identify the copyright message or the
original digital contents itself.
The extra scrambled information generator can
be depicted as in Fig. 4. It consists of a position
selection part using three random sequences from
the random number generator, and a binary data
generator part to combine information from the
copyright message and the selected elements in the
previous part. In the figure, for simplicity, we used a
still image as the original digital contents. In this
system, it is defined that original image is X, random
sequence value K, and copyright message W.
Figure 3: Proposed block diagram.
Figure 4: Generation of extra scrambled information.
The elements of original image, X(X
, X
), are
N-bit gray values and the elements of copyright
message, W(W
, W
), are assumed to be binary
image. Three random number sequences are
generated using initial seeds. Two are used to select
position of pixels in the input image and the last one
is used as a temporary gray value.
The pixel position selector receives two random
sequences as the coordinates of a pixel and outputs
the pixel pointed by these coordinates. The binary
data generator receives the pixels from the pixel
position selector, and the temporary gray sequences,
from the random sequence generator. Using these
values, it computes temporary binary image T as in
the Eq. 1.
== 1,...,1,0
In the equation, 0
g 2
–1, k = W
× W
Finally, the temporary binary image T is
exclusive-ORed with a binary copyright image to
generate extra scrambled information(ESI) as in Eq.
ESI (x,y) = W(x,y)
T(x,y) (2) (2)
The generated binary extra scrambled
information, ESI, is stored in a certified database
with the random number seed values for later
extracting the copyright image.
3.2 Extraction of Copyright Message
A digital contents received from Internet can be
degraded by noise or malicious attacks. Due to the
degradation of digital contents, a copyright message
embedded in extra scrambled information can also
be affected. Fig. 5 shows the extraction process.
Figure 5: Extraction of copyright message.
The method is very similar to the extra
scrambled information generator process. It
generates two random numbers using the initial
seeds from the database and uses the sequences to
select positions of pixels in the input image. It also
uses one random sequence as temporary gray values
to calculate a temporary binary image as in Eq. 1.
Using the temporary binary image and extra
scrambled information from database, we can
compute the received external copyright message W’
as in Eq. 3.
W’ = ESI(x,y)
T(x,y) (3)
In this equation, the received image is X’,
temporary image, T’, and extracted copyright
message, W’.
Performance measures in watermarking or
fingerprinting systems can be varied dependently on
the types of applications. To evaluate the
performance of the proposed method, we used a
popular benchmark test called Stirmark (Nikolaidis,
2004) (Website). This algorithm generates various
spatial signal processing images such as rotation,
cropping, median filter, adding noise, and etc. Bit
Correct Ratio (BCR) is mostly used in this paper to
show the correctness of the extracted copyright
message and it is defined in Eq. 4 (Chang, 2004).
In this equation, w and w’ are the original
watermark and the extracted watermark at the
detector side, respectively and denotes the EX-OR
In the experiment, we used the 512×512 Lena
image with 8-bit gray scale as the original digital
content and a 64 × 64 binary image as the copyright
message. Total attacked 108 images generated using
Stirmark algorithm are used to verify the proposed
method. Fig. 6 shows the original Lena, binary
copyright message and its generated extra scrambled
information, respectively.
(a) Original Lena
(b) Copyright message (c) Extra scrambled information.
Figure 6: Extraction of copyright message.
After degrading the original contents with some
signal processing attacks such as histogram
equalization, median filtering and scaling, we
evaluate the performance using the BCR. Fig. 7
shows the results extracted copyright message with
BCRs. For these typical attacks, it still shows very
high BCR under those attacks.
(a) Histogram equ. : 90.6% (b) Median(3×3) : 93.7%
(c) Rotation(2°) : 88.0% (d) Rotation(0.5°) : 93.1%
(e) Noise(20%) : 75.4% (f) Scaling(90%) : 94.2%
Figure 7: Extracting results.
The performance of the proposed method is
compared to one of the leading research results as in
Table 1. C. Chang (Chang, 2004) proposed a
wavelet transform-based watermarking system using
artificial neural networks. As seen in the table, the
proposed method shows much higher BCR
comparing to the method in the reference.
Table 1: The comparison of BCR under various attacks.
Attack Types
C. Chang
Histogram Equ. 90.67% -
Median Filter 93.7% 89.25%
JPEG 94.5% 88.43%
Scaling 94.8% 78.58%
We summarize several properties of the propose
method as follows:
Fidelity : Embedding a watermark message to
other watermark system requires the modification of
pixel values in the spatial domain or frequency
components in the frequency domain. This degrades
the fidelity of the digital contents. On the contrary,
the proposed method is not directly change the
original contents as in the fingerprinting technique,
so that it keeps the fidelity to 100%.
Required space : The proposed method
requires to have a space to store a binary scrambled
information in a certified database. The extra
scrambled information has the same size of the
copyright message.
Computational complexity : The
computations in the proposed method are the
magnitude comparison and the exclusive-OR
operation. The processing time for these operations
is less than 10 and should be very short so that
the method can be used to a real time application
Robustness : Most watermarking systems
require to have robustness against various signal
processing attacks. Various experiments are
performed to evaluate the proposed method. It has
high BCRs for histogram equalization, compress,
scaling, and small amount of rotation. On the other
hand, the BCRs are rather low for high noise and
cropping of the original image.
We propose a novel copyright protection technique
by combining features of a watermarking system and
a fingerprinting system. The proposed system
generates extra scrambled information as a function
of original contents and copyright message and
stores it in a certified database for later use. The
generated extra scrambled information in trusted
certification center or database can be used in case
of copyright dispute. The stored extra scrambled
information can be used to identify the copyright
message in many applications.
Since the proposed system can distribute the
original contents without changing any of its values,
no degradation on the contents is introduced. The
experimental results show that the proposed method
performs better than the existing leading method for
various signal processing attacks. In addition, the
computational complexity of the proposed method is
very low so that it can be used for some real time
We will further study about the extra scrambled
information search algorithms according to
increasing contents and robustness algorithms to
geometric attacks.
Cox, Ingemar J., 2002. Digital Watermarking(Multimedia
Information and System) .Morgan-Kaufmann.
H. Huang, H. Hang, and J. Pan. 2004. "An Introduction to
Watermarking Techniques," Series on Innovative
Intelligence, Vol. 7, pp.3-39, World Scientific
Publishing Co. Pte. Ltd.
R.G.V. Schydel, A.Z. Tirkel, and C.F.Osborne. 1994. ”A
Digital Watermark”, In Proceeding of the
International Conf. on Image Processing. pp. 86-90,
Austin, IEEE Press.
ISO/IEC 21000-11. Information technology-Multimedia
framework(MPEG-21)-Part11:Evaluation Tools for
Persistent Association Technologies. 2004.
A. Nikolaidis and I. Pitas. 2001. " Region-Based Image
Watermarking," IEEE Trans. on Image Processing,
Vol. 10, No. 11, pp. 1726 – 1740.
M. Fazam and S. Shahram Shirani. 2001. " A robust
multimedia watermarking technique using Zernike
transform,"in Proc. IEEE Int. Workshop Multimedia
Signal Processing, pp. 529-534.
N. Nikolaidis and I. Pitas. 2004. “Benchmarking of
Watermarking Algorithms”, Series on Innovative
Intelligence. Vol. 7, pp.315-347, World Scientific
Publishing Co. Pte. Ltd.
C. Chang and I. Lin. 2004. "Robust Image Watermarking
Systems Using Neural Networks," Series on
Innovative Intelligence, Vol. 7, pp.395-427, World
Scientific Publishing Co. Pte. Ltd.