All materials on our website are shared by users. If you have any questions about copyright issues, please report us to resolve them. We are always happy to assist you.

Share

Description

Web-oriented nonblind image watermarking procedure

Tags

Transcript

Web-oriented nonblind image watermarking procedure
Franco Frattolillo
University of SannioDepartment of EngineeringResearch Centre on Software TechnologyCorso Garibaldi, 10782100 Benevento, ItalyE-mail: frattolillo@unisannio.it
Abstract.
The advances in multimedia and networking technolo- gies have created opportunities for Internet pirates, who can copy digital contents and illegally distribute them, thus violating the legal rights of content owners. In such a situation, digital watermarking has gained popularity as a main technology to implement the copy- right protection of multimedia digital contents distributed on the In- ternet. We present a novel “nonblind” watermarking procedure for JPEG images based on the use of protected extensible markup lan- guage (XML) documents. The procedure enables the copyright owner to insert a distinct watermark code identifying the buyer into the distributed images. Furthermore, to increase the security and robustness levels of the procedure, the watermark is repeatedly em- bedded into an image in the discrete cosine transform (DCT) do- main at different frequencies and by exploiting both block classiﬁca- tion techniques and perceptual analysis. The embedded watermark is then extracted from an image according to the information con- tained in a protected XML document that is associated to the image.
© 2006 SPIE and IS&T.
DOI: 10.1117/1.2234479
1 Introduction
Digital watermarking is considered one of the main securitytechniques that can be used for discouraging the unautho-rized trading of multimedia digital contents on the Internet.It is exploited to implement copyright protection systemsbased on embedding the perceptually invisible digital sig-natures of the copyright holders in the contents to be pro-tected, thus enabling the copyright holders to prove theownership of the protected contents by extracting the in-serted signatures.
1–3
The development of watermarking procedures involvesseveral design trade-offs. In particular, watermarking pro-cedures should
1,4
:1. provide a good degree of robustness against themost common, nonmalevolent content manipula-tions, including digital-to-analog conversion anddigital format conversion2. survive intentional attacks3. be imperceptible and convey as much informationas possible4. comprise watermark embedding and retrieval pro-cesses characterized by low complexity, so as tobe effectively exploitable in a web context, whichis commonly considered their typical utilizationscenarioHowever, all of these requirements and the resulting designconstraints appear to be partly contradicting. For example,“blind” watermarking procedures
3
can be considered wellsuited to be adopted in a web context, since they exploitinsertion schemes that do not need the srcinal, nonwater-marked multimedia digital contents to retrieve the hiddeninformation. As a consequence, such procedures do notforce the distinct web entities involved in “identiﬁcationand arbitration” protocols
5–7
to exchange unprotected,large-size multimedia digital contents through the insecurecommunication channels usually characterizing theInternet.
6
Furthermore, blind watermarking procedures areusually based on ﬁngerprinting techniques that enable thecopyright owner to insert speciﬁc “anticollusion codes”able to identify the buyer within any copy of content that isdistributed.
8,9
The main aim is to make it possible to deter-mine if a user is illegally in possession of a content as wellas who has initially bought and then illegally shared it via,for example, peer-to-peer network applications.
1
However,“nonblind” watermarking procedures are typically consid-ered more robust then blind ones.
7,9
Unfortunately, differ-ently from blind ones, nonblind watermarking proceduresrequire the srcinal, unprotected multimedia digital con-tents to be able to run the watermark extraction algorithmson the corresponding protected or pirated copies. Therefore,the adoption of such watermarking procedures in a webcontext results in being very difﬁcult.This paper presents a web-oriented, nonblind water-marking procedure for JPEG images based on theuse of protected extensible markup language
XML
documents.
10,11
The procedure enables the copyright ownerto insert a distinct code identifying the buyer within eachcopy of the distributed images. Furthermore, to increase theprocedure security and robustness levels, the watermark isrepeatedly embedded into an image in the discrete cosinedomain
DCT
domain at different frequencies and by ex-ploiting both block classiﬁcation techniques and perceptualanalysis. The embedded watermark is then extracted froman image according to the information contained in a pro-tected XML document that is associated to the image. Thus,the usual security and robustness levels characterizing thenonblind watermarking schemes can be achieved withoutrequiring unprotected, large-size images to be securely ex-
Paper 05170R received Sep. 23, 2005; revised manuscript received Jan. 9,2006; accepted for publication Jan. 17, 2006; published online Aug. 23,2006.
1017-9909/2006/15
3
/033011/14/$22.00 © 2006 SPIE and IS&T.
Journal of Electronic Imaging 15(3), 033011 (Jul–Sep 2006) Journal of Electronic Imaging Jul–Sep 2006/Vol. 15(3) 033011-1
changed and stored on the Internet whenever the watermarkextraction has to be performed. In fact, keeping XML docu-ments protected or securely exchanging them in a web con-text results nowadays in being rather easy due to the secu-rity technologies developed in the ﬁeld of XML and “webservices.”
12–15
Moreover, using the XML technology makesit also easier to automate the document access in a webcontext, since XML is a standard technology well sup-ported by the Java world, and standard document parsers,such as Simple API for XML
SAX
and Document ObjectModel
DOM
parsers, are freely available.The paper is organized as the follows. Section 2 presentsthe proposed watermarking procedure. Section 3 brieﬂy de-scribes the XML documents associated to the protected im-ages. Section 4 describes the watermark extraction process.Section 5 reports on some experimental results. Section 6reports conclusion remarks.
2 Watermarking Procedure
The proposed watermarking procedure, whose scheme isshown in Fig. 1, makes it possible to insert into a JPEGimage a binary code which can be exploited to unambigu-ously identify a user. The code, which is represented by asequence
0,1
and whose length is denoted as
n
, canbe repeatedly embedded into the image in the DCT domainat different frequencies, denoted as
1
,
2
,...,
f
. In fact,each frequency
i
identiﬁes an entry in each 8
8 DCTblock of a JPEG image, and so it can range
16
from 1 to8
2
=64. Furthermore, to increase the security and robustnesslevels of the procedure, the watermark insertion is assumedto be carried out at low, middle, and high frequencies cho-sen on the basis of the image to be watermarked.In principle, all the DCT coefﬁcients at the differentfrequencies
i
, with
i
=1,...,
f
, could be modiﬁed by avalue representing a watermark information. However, inthe proposed procedure, the “perceptual capacity” of thecoefﬁcients belonging to the luminance DCT blocks is pre-liminarily estimated by exploiting both block classiﬁcationtechniques and perceptual analysis. In fact, the block clas-siﬁcation techniques
16–18
are applied to indicate the bestDCT coefﬁcients that can be altered without reducing thevisual quality. They classify each luminance DCT blockwith respect to its energy distribution by using four classi-ﬁcation masks. The possible types of classiﬁed blocks are“ﬂat,” “diagonal edge,” “horizontal edge,” “vertical edge,”and “textured block.” The result of this procedure is a ﬁrstselection of DCT coefﬁcients whose modiﬁcation has aminimal or no impact to the perceptual quality of the im-age.The perceptual analysis is then applied to calculate the“just noticeable difference”
JND
values for the DCTcoefﬁcients.
19–21
Such values are the thresholds beyondwhich any changes to the respective coefﬁcient will mostlikely be visible. Therefore, let
X
b
m
denote the coefﬁ-cient at the frequency
in the DCT block
b
m
, and letJND
b
m
denote the JND value calculated for the
X
b
m
coefﬁcient. We can calculate JND
b
m
asJND
b
m
max
C
b
m
,
C
b
m
E
b
m
g
,
1
where
C
b
m
represents the perceptual threshold of thecontrast masking and is expressed as
C
b
m
= max
t
b
m
,
X
b
m
h
t
b
m
1−
h
,
2
where
E
b
m
is the entropy value calculated over the eightneighbors of the
X
b
m
coefﬁcient,
19,20
and can be approxi-mated by the following expression:
Fig. 1
Scheme of the proposed watermarking procedure.
Frattolillo: Web-oriented nonblind image watermarking procedure Journal of Electronic Imaging Jul–Sep 2006/Vol. 15(3) 033011-2
E
b
m
X
b
m
−
u
b
m
q
.
3
In Eq.
1
g
is assumed equal to 0.5. In Eq.
2
h
is assumedequal to 0.7, and
t
b
m
is equal to
t
X
b
m
1
/
X
1
,where
X
1
is a dc coefﬁcient corresponding to the meanluminance of the display, whereas
X
b
m
1
is the dc coefﬁ-cient of the block
b
m
. In fact,
t
can be approximated bythe value
q
/2, where
q
represents the coefﬁcient of the quantization matrix corresponding
20
to the frequency
.Finally, in Eq.
3
u
b
m
is equal to round
X
b
m
/
q
.Once block classiﬁcation and perceptual analysis haveidentiﬁed the DCT coefﬁcients that could usefully host wa-termark information without reducing the image visualquality, the “choice rule” block takes charge of selectingwhich of these coefﬁcients have to be actually modiﬁed bythe insertion procedure at each of the
f
frequencies chosenon the basis of the image to be watermarked
see Fig. 1
. Infact, the insertion procedure performed at each frequency
i
, with
i
=1,...,
f
, assumes that each bit of the user se-quence
is inserted into a given image by altering a pair of selected DCT coefﬁcients associated to that frequency andcharacterized by similar values. More precisely, let
b
l
i
indicate the entry at the frequency
i
in the DCT block
l
,where
l
may vary from 1 to
t
JPEG
, which denotes the maxi-mum number of DCT blocks contained in a JPEG imagewith given characteristics. Let
i
denote the set of theDCT entries
b
l
i
whose associated values have been iden-tiﬁed by the block classiﬁcation and perceptual analysis asvalid candidates at the frequency
i
, with
i
=1,...,
f
, to hostthe bits belonging to the user sequence
. Let
i
denotethe sequences of the pairs of DCT entries
b
m
i
,
b
n
i
whose values are actually selected to host the bits of theuser sequence
at the frequency
i
, with
i
=1,...,
f
and
m
and
n
being two indices in the range from 1 to
t
JPEG
. The“choice rule” states that a pair
b
m
i
,
b
n
i
is allowed tobelong to
i
only if
X
b
m
i
X
b
n
i
,
"
i
=1,...,
f
andwith
b
m
i
and
b
n
i
belonging to
i
. This can also besynthesized by the following expression:
Fig. 2
Insertion procedure.
Frattolillo: Web-oriented nonblind image watermarking procedure Journal of Electronic Imaging Jul–Sep 2006/Vol. 15(3) 033011-3
Consequently, if
is
n
bits long, the process that selectsthe DCT coefﬁcients at the frequency
i
must choose atleast 2
n
coefﬁcients. Therefore, if the insertion frequen-cies are
f
, the total number of DCT coefﬁcients to be se-lected is 2
n
f
. Finally, the cardinality of
i
is equal to
n
.The bits of the user sequence
are inserted into animage as symbols obtained by applying the “encoding func-tion”
E
, which must be deﬁned within the watermarkingprocedure. In fact,
E
deﬁnes an encoding rule by which thebits 0 and 1 are translated to the symbols belonging to thealphabet composed by
ր
,
ց
, respectively called the upsymbol and the down symbol. Thus, a user sequence
0,1
is translated to a corresponding sequence of sym-bols
ր
,
ց
depending on the function
E
. For ex-ample, the user sequence
01101...
is translated to thesequence of symbols
րցցրց
...
, if the function
E
associates the up symbol to 0 and the down symbol to 1.Let
be a user sequence, and let
be the correspondingsequence of symbols obtained by applying an
E
function.Let
j
denote the
j
’th symbol of
, and let
1
,
2
,...,
f
be the insertion frequencies. Let
W
b
m
i
denote the water-marked DCT coefﬁcient at the frequency
i
in the block
b
m
i
, and let
i
j
denote the
j
’th pair in the set
i
.The insertion procedure, whose scheme is exempliﬁed inFig. 2, is synthetized by the following expressions:
"
i
= 1, ... ,
f
,
"
j
= 1, ... ,
n
j
=
ր
Þ
W
b
m
i
=
X
b
m
i
− JND
b
m
i
W
b
n
i
=
X
b
n
i
+ JND
b
n
i
j
=
ց
Þ
W
b
m
i
=
X
b
m
i
+ JND
b
m
i
W
b
n
i
=
X
b
n
i
− JND
b
n
i
,
b
m
i
,
b
n
i
=
i
j
,
i
j
i
which assume that the
j
’th symbol
j
derived from
isinserted in the DCT coefﬁcients identiﬁed by the
j
’th pair
i
j
=
b
m
i
,
b
n
i
belonging to
i
,
"
i
=1,...,
f
and
"
j
=1,...,
n
. In fact, since the “choice rule” imposes that
X
b
m
i
X
b
n
i
for each selected pair of DCT coefﬁcients,the insertion process attempts to maximize the differenceexisting between the coefﬁcients of the pair according tothe direction speciﬁed by the insertion symbol and by anamount that should not compromise the ﬁnal visual qualityof an image. Therefore, the insertion process should be car-ried out according to the following main rules:1. The insertion frequencies should be evenly distrib-uted among the low, middle, and high frequencies,and should be chosen so that attacks characterizedby a ﬁltering behavior on an image would end upreducing its ﬁnal visual quality. This can beachieved by selecting frequencies characterized byhigh spectrum values, which, if ﬁltered, can impairthe image.2. At each insertion frequency, the pairs of the se-lected DCT coefﬁcients should belong to spatialregions that cannot be cropped without impairingthe image.Once the symbols of the sequence
have been insertedinto the image at the chosen frequencies, to increase thesecurity and robustness levels of the watermarking proce-dure against collusion and averaging attacks,
8,9
it is neces-sary to hide the modiﬁcations made to the DCT coefﬁcientsof the image. In fact, note that both the set of the insertionfrequencies
i
and the sets
i
, with
i
=1,...,
f
, are al-ways the same for all the copies of a given image to beprotected. Consequently, the DCT coefﬁcients modiﬁed atthe different insertion frequencies remain the same for allthe copies of the image. Therefore, to prevent malicioususers from individuating the DCT coefﬁcients modiﬁed bythe insertion process, the JND values modulated by a bi-nary pseudonoise sequence
−1,1
must be added to allthe unmodiﬁed DCT coefﬁcients of a watermarked image.This addition is carried out by the following expression:
X
b
k
i
=
X
b
k
i
+
k
k
JND
b
k
i
,
i
1, ... ,
f
or
i
= 1, ... ,
f
and
b
k
i
b
m
i
,
b
k
i
b
n
i
,
b
m
i
,
b
n
i
i
,
where 0
k
0.5 is a randomly varied amplitude factor.Finally, note that the proposed procedure directly acts onJPEG compressed images. To this end, as shown in Fig. 1,the incoming JPEG image is processed, and the sequence of Huffman codes, each representing one non-zero DCT coef-ﬁcient of the image in the form of one
run, level
pair, isextracted and decoded.
16
Then, the DCT coefﬁcients se-lected by the procedure are altered by inserting the water-mark information. After the insertion, the DCT coefﬁcientsare Huffman reencoded in the form of
run, level
pairs,thus generating the watermarked version of the image.However, since the number of bits needed to encode a
run,level
pair is set by the JPEG standard and may be the sameeven for different
run, level
pairs,
16
the ﬁnal size of theimages watermarked by exploiting the proposed procedureends up differing from the size of the corresponding srci-nal images by a small amount essentially determined by thenumber
f
of the chosen insertion frequencies. In fact, theconducted tests have determined increments in the size of the watermarked images about 4 to 9%, with
f
varying inthe range
3,15
.
Frattolillo: Web-oriented nonblind image watermarking procedure Journal of Electronic Imaging Jul–Sep 2006/Vol. 15(3) 033011-4
3 XML Documents
The characteristics of the proposed watermarking proce-dure make it necessary to save the insertion informationused to protect the copies of an image to make it possible tocarry out the watermark extraction. To this end, as reportedin Sec. 1, the proposed procedure assumes that such infor-mation is stored in XML documents. In particular, the XMLdocument associated to all the watermarked copies of animage, whose internal structure is out of the scope of thepaper and is not shown for the sake of brevity, has to in-clude:1. the insertion frequencies
1
,
2
,...,
f
2. the sets
i
,
"
i
=1,...,
f
3. the encoding function
E
Fig. 3
Geometric resynchronization of “Lena.”
Frattolillo: Web-oriented nonblind image watermarking procedure Journal of Electronic Imaging Jul–Sep 2006/Vol. 15(3) 033011-5

Related Search

Image WatermarkingDigital Image Watermarking, Image ProcessingDFT watermarking, Image watermarking, DiscretImage Processing, Web security, Sortware DeveWeb Presence, Web Image, Online IdentityWeb Services and Service Oriented ArchitecturText And ImageService-oriented architecturesWeb Services CompositionWeb Development

We Need Your Support

Thank you for visiting our website and your interest in our free products and services. We are nonprofit website to share and download documents. To the running of this website, we need your help to support us.

Thanks to everyone for your continued support.

No, Thanks