`in Open Networks
`
`Watermarking Technologies
`
`Alessandro Piva
`and Franco Bartolini
`University of Florence
`
`Mauro Barni
`University of Siena
`
`Integrating cryptography with watermarking technologies
`can provide intellectual property rights protection in an open
`network environment such as the Internet.
`
`Despite the ease with which digital
`
`data owners can now transfer
`multimedia documents across the
`Internet, current technology does not let
`them protect their rights to the works. In
`fact, although the Internet permits wide-
`spread dissemination of interactive ser-
`vices such as remote database access,
`archival browsing, and electronic com-
`merce, the easy-to-copy nature of digital
`data limits data owners’ willingness to
`distribute their documents electronically.
`Thus, the need for an electronic copyright
`management system (ECMS) that protects
`intellectual property rights (IPR) in open-
`network environments continues to grow.
`Network security issues are classically
`handled through cryptography;1 howev-
`er, cryptography ensures confidentiality,
`authenticity, and integrity only when a
`message is transmitted through a public
`channel, such as an open network. It
`does not protect against unauthorized
`copying after the message has been suc-
`cessfully transmitted.
`
`Digital watermarking is an effective
`way to protect copyright of multimedia
`data even after its transmission.2,3 A
`watermark, embedded in the data, can
`uniquely identify the document’s owner
`or authorized user. The main problem
`with using watermark technology for IPR
`protection, however, is its reversibility.
`Anyone who can read or detect the
`watermark can also remove it by invert-
`ing the watermark process. Our open-net-
`work ECMS combines watermarking with
`cryptography to achieve reliable copy-
`right protection while satisfying two con-
`trasting requirements:
`
`■ Actors in ECMS transactions must be
`able to verify that the watermark
`granting their rights is truly embed-
`ded in the multimedia document.
`■ Actors (other than the author) must
`not be able to remove the watermark.
`
`In this article, we discuss digital water-
`marking and describe our integrated ECMS
`
`18
`
`MAY • JUNE 2002
`
`http://computer.org/internet/
`
`1089-7801/02/$17.00 ©2002 IEEE
`
`IEEE INTERNET COMPUTING
`
`Sony Exhibit 1035
`Sony v. MZ Audio
`
`
`
`Electronic Copyright Management Systems
`
`Managing Copyrights
`
`Electronic copyright management systems
`automatically manage issues related to trad-
`ing multimedia documents through open
`communication networks. An ECMS can be
`considered an ensemble of services, con-
`nected through a network environment,
`cooperating to allow intellectual property
`rights (IPR) protection of multimedia data.
`Several projects are under way to devel-
`op ECMSs. The most recent MPEG stan-
`dardization effort (MPEG-21), for example,
`aims to establish rules and protocols for
`permitting the legal and reliable exchange
`of IPR-sensible multimedia documents.
`We distinguish two approaches to
`designing effective ECMSs:
`
`■ preventing copyright violations (IBM’s
`Cryptolope, www-3.ibm.com/software/
`security/cryptolope, for example)
`■ tracking copyright violations (the EC-
`funded Imprimatur,www.imprimatur.net,
`for example).
`
`Both approaches require authoring tools
`
`to properly prepare multimedia documents
`before distributing them.
`
`Cryptography-based ECMSs
`In a cryptography-based ECMS, the author
`wraps the digital object in an encrypted
`system and integrates it with an application
`(the reader). Because users cannot access
`the content without the proper application,
`the owner can control how the document
`is used—for example, a user can display the
`images but not print them, or play the
`audio files but not save them.
`The main disadvantage of this approach
`is the difficulty of establishing a standard for
`embedded applications. Moreover, when a
`multimedia document finally reaches the end
`user (for example,it appears on a PC screen
`or is played by a digital recorder), it can still
`be captured and copied without constraint.
`Liquid Audio (www.liquidaudio.com) is an
`example commercial system.
`
`Watermark-based ECMS
`A watermark-based ECMS tightly and
`
`robustly embeds IPR-related information
`into purchased digital objects (the hidden
`data can be the name of the copyright
`owner or a unique code identifying the
`document).Watermarking can also be used
`to hide the identification of the authorized
`distributor or buyer (the more correct
`term for this is fingerprinting) inside the
`document. It is thus always possible to
`check the document’s legal status, and to
`track the path IPR-infringing material fol-
`lows through the network.
`A main limitation of current water-
`marking technologies is their reversibili-
`ty; that is, anyone who can read or detect
`a watermark can remove it. Only the
`effective development of asymmetric
`watermarking methods, which still seem
`far off, will overcome this intrinsic limita-
`tion. On the other hand, watermark-
`based IPR management does not require
`users to adopt a particular format for the
`watermarked multimedia content, be-
`cause IPR data are directly injected into
`the content itself.
`
`approach. We also introduce our prototype system,
`available at http://lorenzo.det.unifi.it, to show the
`approach’s viability. The sidebar, “Electronic Copy-
`right Management Systems,” discusses current tech-
`nologies for IPR management over open networks.
`
`Digital Watermarking
`In digital watermarking, a digital code, or water-
`mark, is embedded into a document so that a
`given piece of information, such as the owner’s or
`authorized consumer’s identity, is indissolubly tied
`to the data. This information can later prove own-
`ership, identify a misappropriating person, trace
`the marked document’s dissemination through the
`network, or simply inform users about the rights-
`holder or the permitted use of the data.
`Watermarking does not solve all IPR problems,
`however,4,5 and most researchers agree that the
`technology is less mature than cryptography. Still,
`its potential to provide reliable protection is
`already attracting copyright holders.
`
`Watermarking Algorithms
`Several watermarking schemes have been intro-
`duced, and a great deal of research has sought to
`
`develop data-labeling techniques that are robust
`against the most common attacks and multimedia
`processing manipulations. Little attention has been
`given to protocol-level analysis, however. The
`sidebar, “Related Copy-Deterrence Protocols,”on
`page 20, discusses some work in this area.
`Because how a watermarking algorithm recov-
`ers the watermark from the data determines which
`technique will be used in a given situation, we
`classify digital watermarking techniques by their
`decoding processes.
`
`■ Blind versus not blind. A watermarking algo-
`rithm is blind if it does not need to compare the
`marked and unmarked documents to recover
`the watermark. Conversely, a watermarking
`algorithm is not blind if it needs the original
`data to extract the information from the water-
`mark. Blind techniques are sometimes referred
`to as oblivious or private.
`■ Private versus public. A watermark is private if
`only authorized readers can detect it. Not-blind
`techniques are private because only authorized
`users can access the original data needed for
`watermark reading. We extend the concept of
`
`IEEE INTERNET COMPUTING
`
`http://computer.org/internet/
`
`MAY • JUNE 2002
`
`19
`
`
`
`Watermarking Technologies
`
`Related Copy-Deterrence Protocols
`
`Various copy-deterrence protocols com-
`bining watermarking and cryptography
`have been proposed. Lintian Qiao and Klara
`Nahrstedt propose an owner-customer
`watermarking protocol, in which a cus-
`tomer sends the owner an encrypted ver-
`sion of a predetermined code.1 After
`receiving the code, the owner embeds the
`encrypted sequence into a copy of the
`image as a watermark and transmits the
`copy to the buyer. Because no one else
`knows the decryption key, the buyer can
`prove legitimate ownership of the copy.
`The protocol does not link the customer
`to the purchased copy, however, so unau-
`
`thorized copies cannot be traced. In fact, a
`counterfeiter can claim that an unautho-
`rized copy was created by the seller or
`caused by a security leak in the system.
`Nasir Memon and Ping Wah Wong pro-
`pose a buyer-seller protocol in which the
`seller does not know the buyer’s water-
`mark, and so, cannot create copies of the
`image containing it.2 The watermarking
`protocol is based on public key cryptogra-
`phy and requires a watermark certification
`authority. This model does not let the
`buyer verify that a watermark proving
`ownership is truly embedded in the copy.
`These models do not allow each actor
`
`to check that the data exchange was car-
`ried out correctly and, at the same time,
`verify that the current holder is using the
`data legally.This is the main novelty of our
`proposed approach.
`
`References
`1. L. Qiao and K. Nahrstedt, “Watermarking
`Schemes and Protocols for Protecting Rightful
`Ownership and Customer’s Rights,’’ J.Visual Comm.
`and Image Representation, vol. 9, no. 23, Sept. 1998,
`pp. 194-210.
`2. N. Memon and P.W.Wong,“A Buyer-Seller Water-
`marking Protocol,’’ IEEE Trans. Image Processing, vol.
`10, no. 4,Apr. 2001, pp. 643-649.
`
`privateness to techniques using any mechanism
`to prevent unauthorized personnel from
`extracting the watermark. If anyone can read
`the watermark, we call it public.
`■ Readable versus detectable. We also distinguish
`between algorithms that embed a code users
`can read without knowing the content in
`advance, and those that insert a mark that can
`only be detected — that is, a user can only ver-
`ify that a given code is in the document. Water-
`marks that are encrypted before they are
`embedded are even harder to detect. Detectable
`watermarking is sometimes referred to as 1-bit
`watermarking because the detector output is
`just “yes” or “no.”
`
`Not-blind methods are more robust to attacks than
`blind methods, because the original content can be
`used in detection to estimate possible modifica-
`tions introduced by an attacker to remove the
`watermark or make it unreadable. Very often, how-
`ever, the original document is not available, mak-
`ing not-blind algorithms unsuitable for many
`practical applications. Moreover, private mecha-
`nisms tend to be significantly more robust than
`public ones: an attacker can easily remove or make
`unreadable a known watermark. Because de-
`tectable watermarks are intrinsically private, it fol-
`lows that blind, detectable systems are more robust
`than other schemes.
`
`Reversibility
`A watermark is reversible if, once read or detected,
`it can be removed from the document, or at least
`made unreadable or undetectable. Virtually all
`
`existing techniques are potentially reversible.
`Indeed, because watermarks must be invisible, the
`modification introduced by the watermarking
`process is very small and thus linearizable and con-
`sequently invertible. Therefore, anyone who can
`read or detect the watermark can also remove it.
`This conflicts with our requirement that a legal
`buyer have the right to check that his or her name
`is truly embedded in the multimedia document.
`Watermark reversibility allows a buyer who can
`check for watermark presence to also remove it,
`and possibly reuse the document illegally by
`embedding a forged watermark.
`An asymmetric watermarking algorithm might
`overcome reversibility issues.6 In asymmetric
`watermarking, watermark detection and decoding
`reveals only part of the secret used to embed the
`watermark (the public key); the private key
`remains hidden. Requiring the private key for
`watermark removal prevents reversibility prob-
`lems. Asymmetric watermarking is a very imma-
`ture field, however, and researchers are still not
`sure whether it can be used for secure public
`watermark detection. Moreover, asymmetric
`schemes embed a very small amount of informa-
`tion into a document and thus are not suited for
`complex ECMS applications. Rather, we expect
`they will be used to manage document copies,
`where a lower capacity is required.
`The ECMS presented in this article is explicitly
`designed to overcome the problems deriving from
`watermark reversibility. We assume the use of a
`detectable watermarking scheme because such
`techniques are more robust and reliable than read-
`able schemes.
`
`20
`
`MAY • JUNE 2002
`
`http://computer.org/internet/
`
`IEEE INTERNET COMPUTING
`
`
`
`An Integrated Approach
`to IPR Protection
`We have developed a watermark-based ECMS that
`integrates cryptography to compensate for the
`weaknesses of watermarking schemes and to
`achieve reliable copyright protection.
`Trading multimedia documents in an open-net-
`work environment involves many actors — the
`document author or authors, an editor, a media
`distributor, buyers, and so on. It also involves
`electronic payment issues, such as information
`security and customer privacy. To simplify our
`presentation, we limit the number of actors and
`do not address payment or privacy issues here.
`
`Transaction Model
`Figure 1 shows a simplified trading model. Annie,
`the author of a multimedia document, registers her
`document and deposits a copy of it with a collect-
`ing society. She then contacts a media distributor,
`McDarrel, who makes her document available on
`the network, where Peter accesses and buys it. For
`simplicity, we assume the CS is a trusted third
`party that will ensure that the protected docu-
`ments are traded correctly. Note that the transac-
`tion between the buyer and the media distributor
`also involves an exchange of data with the CS.
`In our approach, the document is self-con-
`tained. At any given instant it contains all the
`information needed to verify whether the current
`holder is using the data legally. No attempt is
`made to trace the document history, however,
`either by watermarking the document each time
`the owner changes, or by recording transaction
`details in a register. We take particular care to
`allow each actor to check that the data exchange
`was carried out correctly.
`The basic principle underlying our ECMS
`strategy is that the data holder’s name must be
`watermarked into the data to prove legal owner-
`ship. To ensure that a document is being used
`legally, any authorized person can check the
`watermark field the holder’s name is written in.
`We also envision a protocol-level mechanism
`that addresses the reversibility problem by pre-
`venting data holders or counterfeiters from ben-
`efiting from watermark removal: at no step of
`the transaction can a counterfeiter insert a fake
`watermark, so a counterfeiter cannot prove doc-
`ument ownership. To keep misappropriating per-
`sons from writing their names into the data, the
`ECMS assumes that the seller (or the author
`when a media distributor sells the document)
`embeds the watermark.
`
`Managing Copyrights
`
`Author
`(Annie)
`
`1
`
`Collecting
`society
`
`(3)
`
`2
`
`Media
`distributor
`(McDarrel)
`
`3
`
`(3)
`
`Buyer
`(Peter)
`
`Figure 1. A simplified transaction model. (1) An author registers a new
`document with a collecting society. (2) The author sends a copy of the
`document to a media distributor for dissemination. (3) A buyer contacts
`the media distributor and purchases a digital copy of the document.
`
`Document with embedded watermarks
`
`1st watermark
`
`Creation unique number
`
`2nd watermark
`
`Media distributor's PIN
`
`3rd watermark
`
`Purchaser's PIN
`
`Embedded at
`creation time
`
`Embedded
`before selling
`
`Embedded
`while selling
`
`Figure 2.A document with embedded watermarks. Our ECMS uses
`three watermarks: the first refers to the creation identity; the second
`contains the media distributor’s personal identification number
`(PIN); and the third identifies the buyer.
`
`Verifying Ownership Rights
`As Figure 2 shows, the document contains three
`watermarks embedded into the data at different
`times. We use blind, detectable watermarking and
`reversible watermarks. Although similar water-
`marking algorithms could be used to implement
`the proposed ECMS, it is beyond the scope of this
`article to investigate them. A companion article in
`(www.computer.org/internet/v6n3/
`IC Online
`ecms.htm) details the watermarking method used
`to implement our prototype ECMS. Figure 3 (next
`page) illustrates the transactions involved in sell-
`ing a multimedia document.
`
`Author identifier. When Annie registers a docu-
`ment in the CS, she also embeds into the data a cre-
`
`IEEE INTERNET COMPUTING
`
`http://computer.org/internet/
`
`MAY • JUNE 2002
`
`21
`
`
`
`Watermarking Technologies
`
`Creation-unique number (CUN)
`
`Protected document
`
`1st watermark
`
`A
`
`Annie’s secret key
`
`a)
`
`Creation-unique number (CUN)
`
`Protected document
`
`+
`
`McDarrel's PIN
`
`A
`
`Annie’s secret key
`
`1st watermark
`
`2nd watermark
`
`b)
`
`To McDarrel
`
`Creation-unique number (CUN)
`
`Protected document
`
`+
`
`Peter's PIN
`
`CS
`
`CS private key
`
`To Peter
`
`c)
`
`1st watermark
`
`2nd watermark
`
`3rd watermark
`
`Hash function
`
`CS
`
`CS private key
`
`Encrypted digest
`
`Figure 3.Transactions involved in selling a multimedia document. (a)
`The document author, Annie, embeds the first watermark, contain-
`ing a creation-unique number encrypted with her secret key. (b)
`Annie embeds the second watermark, which contains the CUN and
`the media distributor’s personal identifier encrypted with her private
`key. (c) The media distributor inserts the third watermark, which con-
`tains the document CUN and the buyer’s PIN encrypted with the
`collecting society’s private key.
`
`ation-unique number (CUN), which unambiguous-
`ly identifies her document. To prevent anyone from
`reading the watermark with the CUN and exploit-
`ing watermark reversibility to remove it, Annie
`encrypts the CUN before casting. We use symmet-
`ric key encryption, but we could also use an asym-
`metric scheme (for example, we could use the same
`private key used for the second watermark) at this
`stage. Annie then deposits a copy of the water-
`marked document into the CS archive. Figure 3a
`shows the steps involved in this transaction.
`The first watermark will allow a trusted control
`authority to verify the original owner of a multi-
`media document. We assume that the document can
`be identified as belonging to Annie in some other
`way (by visual inspection, for example), given that
`a detectable watermark only allows the control
`authority (CA) to check for the CUN, not to guess it.
`
`Distributor personal identifier. If Annie wants to
`sell copies of her document through a media dis-
`tributor, she embeds a second watermark into the
`document. This watermark contains a personal
`identification number (PIN) identifying the media
`distributor, McDarrel, and the document’s CUN.
`Annie encrypts the watermark string with her pri-
`vate key and a copy of the encrypted string, which
`McDarrel can use to verify that Annie really insert-
`ed his name into the document. McDarrel can use
`Annie’s public key to read the encrypted string,
`and watermark detection software to verify it.
`(Unlike with the first watermark, only an asym-
`metric cryptography scheme can be used here.)
`Figure 3b illustrates this transaction. Note that
`because McDarrel knows the watermark content,
`he can use detectable watermarking.
`Watermark reversibility is not a problem here: if
`McDarrel erases the watermark from the document,
`he cannot prove his right to sell it. In addition,
`because Annie encrypted McDarrel’s name with her
`private key, no one can counterfeit the second
`watermark. Moreover, inserting the CUN into the
`second watermark prevents McDarrel from embed-
`ding the encrypted string into other documents of
`Annie’s he does not have permission to sell. To
`prove his right to sell the document, McDarrel must
`demonstrate that the CUN contained in the second
`watermark matches the CUN in the first.
`Of course McDarrel could embed another CUN
`on behalf of a fake author into the document. To
`get the new CUN, he must deposit a copy of the
`newly watermarked document at the CS. Because
`this new CUN would be issued after the original
`one, time ordering would allow Annie to prove
`
`22
`
`MAY • JUNE 2002
`
`http://computer.org/internet/
`
`IEEE INTERNET COMPUTING
`
`
`
`authorship, even though both CUNs would be in
`the copies McDarrel distributed.
`
`Buyer confirmation. The protocol for data exchange
`between the media distributor and the buyer, Peter,
`must be as simple as possible. A complex procedure
`might result in poor service, which would diminish
`Peter’s willingness to buy documents from McDar-
`rel. Contracts between document authors and media
`distributors are far more complex, and the parties
`can spend more time checking that their rights are
`granted and copyright laws respected. McDarrel
`might want to contact the CS, for example, to ver-
`ify that the CUN embedded into the second water-
`mark is the one assigned to the document. ECMS
`end users are rarely willing to spend this extra time
`to control purchase regularity.
`Figure 3c shows the steps involved if Peter
`wants to buy Annie’s document from McDarrel’s
`Web site. So that Peter can prove his ownership of
`the document, McDarrel embeds Peter’s name into
`the data using a third watermark, which contains
`Peter’s PIN and the document’s CUN. Unlike the
`information in the second watermark, this string
`is not encrypted with the seller’s (in this case,
`media distributor’s) private key. Instead, the CS,
`acting as a trusted third party, uses its private key.
`This compels McDarrel to inform the CS that he
`has sold a copy of Annie’s document, and obliges
`him to pass revenue to Annie.
`The following embedding strategy will assure
`Peter that the CUN in the third watermark is the
`same contained in the first, without his having to
`actually read it (which he could only do if he had
`the secret key used to encrypt the CUN).
`
`1. Peter passes his PIN to McDarrel.
`2. McDarrel passes Peter’s PIN, the CUN, and a
`string with the second watermark’s content
`(that is, McDarrel’s PIN and the CUN encrypted
`with Annie’s private key) to the CS.
`3. The CS passes revenue to Annie.
`4. After encrypting the string with Peter’s PIN and
`the CUN with its private key, the CS embeds the
`second and the third watermarks into its copy
`of the document.
`5. The CS computes a digest of the watermarked
`document using a proper hash function, signs
`the digest with its private key, and sends the
`signed digest and the third, encrypted,
`watermark to McDarrel.
`6. McDarrel embeds the third watermark into the
`document and gives it, the encrypted third
`watermark, and the signed digest to Peter.
`
`Managing Copyrights
`
`Encrypted digest
`
`CS
`
`Digest
`
`CS public key
`
`Protected document
`
`MATCH
`
`Yes/No
`
`Hash function
`
`Watermark
`detector
`
`Digest
`
`Yes/No
`
`String with the encrypted
`third watermark
`
`CS
`
`Peter's PIN
`+
`CUN
`
`CS public key
`Figure 4. Verifying a purchased document.To check whether the third
`watermark contains his name, the buyer decrypts it using the CS
`public key. He then verifies that the document contains the string by
`checking the CS-signed digest against the digest for his document.
`
`To verify that McDarrel has embedded his PIN
`within the data, Peter need only decrypt the third
`watermark using the CS public key. To check
`whether the CUN embedded in the third watermark
`corresponds to that in the first, Peter can compute
`the digest of the watermarked document and con-
`firm that it corresponds to the digest computed by
`the CS. Such a digest also allows Peter to verify the
`integrity of the watermarked document — that is,
`he can confirm that McDarrel has not modified the
`original document. Figure 4 illustrates the se-
`quence of operations Peter performs to check the
`regularity of his purchase.
`It is worth noting that the multimedia document
`is exchanged across the network only at the end
`of the transaction, when Peter receives it from
`McDarrel. Limiting the amount of data travelling
`through the network saves time, especially when
`the document is very large.
`
`Protecting Data from Illegal Use
`Suppose a control authority asks Peter to prove his
`right to a digital document in its possession. Peter
`can simply give the watermarked document and the
`file with the encrypted third watermark to the con-
`trol authority. The CA first checks the encrypted
`third watermark for Peter’s PIN, then, by applying
`a watermark detection engine to the protected doc-
`ument, it verifies that the watermark with Peter’s
`PIN is actually embedded in the data. Finally, the
`CA, which knows both the true CUN and Annie’s
`secret key, can control whether the CUN contained
`in the third watermark matches the document iden-
`tity. Figure 5 illustrates this sequence of operations.
`
`IEEE INTERNET COMPUTING
`
`http://computer.org/internet/
`
`MAY • JUNE 2002
`
`23
`
`
`
`Watermarking Technologies
`
`Protected document
`
`Watermark
`detector
`
`Yes/No
`
`CS
`
`CS public key
`
`Peter's PIN
`+
`CUN
`
`String with the encrypted
`third watermark
`
`a)
`
`Protected document
`
`Watermark
`detector
`
`Yes/No
`
`CUN
`
`b)
`
`A
`
`Annie's secret key
`
`Figure 5. Checking the legality of a protected document. (a) The con-
`trol authority reads the third watermark using the CS public key to
`verify that it contains Peter’s PIN. (b) The CA matches the CUNs in
`the first and third watermarks.
`
`Protected document
`
`Watermark
`detector
`
`YES/NO
`
`String with the encrypted
`third watermark
`
`A
`
`Annie's public key
`Figure 6. Proving the right to sell. To prove his right to sell Annie’s docu-
`ment, McDarrel shows the CA that the second watermark contains his
`PIN, and that the CUNs of the first and second watermarks match.
`
`McDarrel's PIN
`+
`CUN
`
`Indeed, the CA would not really need the user’s
`file with the encrypted third watermark if it could
`get this information directly from the CS. Rather
`than storing all watermarking codes or digests, the
`CS can simply compute them whenever it needs to,
`
`provided the CA gives it the required information.
`In particular, the CS can generate the second and
`third watermark and the digest if it knows the
`media distributor’s PIN, the buyer’s PIN, the CUN,
`and the author’s identity.
`Suppose Peter wants to give a copy of the pro-
`tected document to a friend, Felix. Although he
`can make a copy of the document, he cannot
`insert the correct third watermark. Even if Felix
`has previously purchased one of Annie’s docu-
`ments from McDarrel, he cannot use the third
`watermark from this earlier purchase to water-
`mark the unauthorized copy because the CUNs
`would not match. In practice, our proposed ECMS
`prevents Felix from feigning ownership to a doc-
`ument he did not legally acquire from a media
`distributor. Of course, by controlling Felix’s unau-
`thorized copy, the CA cannot trace it back to Peter
`because it cannot read the third watermark with-
`out knowing it in advance.
`Now suppose the CA wants to check whether
`McDarrel has permission to sell some of the docu-
`ments on his Web site. McDarrel has to tell the CA
`the name of the author of the documents to be
`controlled and provide the file with the encrypted
`second watermark. The CA can then use the
`author’s public key to check whether there is a sec-
`ond watermark embedded in the document, and
`whether it contains McDarrel’s name. Figure 6
`illustrates this process.
`Of course, the CA must verify that the CUN in
`the second watermark matches the true identity of
`the controlled document. It can do this by access-
`ing the CS archives, through an offline search, or
`by comparing the CUN to the content of the first
`watermark (as shown in Figure 5b).
`
`A Java-based Prototype System
`We have implemented an ECMS prototype, similar
`to the trading model reported in Figure 1, using
`Java technology. The actors in the prototype are
`author, vendor, collecting society, and a certifica-
`tion authority in charge of issuing and authenti-
`cating the other actors’ cryptographic keys.
`The prototype uses two servers based on Jigsaw
`technology (www.w3.org/jigsaw/): the vendor
`server and the certification authority and collect-
`ing society server. It also uses a set of Java applets:
`the author’s watermark embedder, the user’s water-
`mark decoder, and the CA watermark decoder.
`Depending on the server they are connected to,
`users can perform several actions:
`
`■ Request digital certificates through the certifi-
`
`24
`
`MAY • JUNE 2002
`
`http://computer.org/internet/
`
`IEEE INTERNET COMPUTING
`
`
`
`Managing Copyrights
`
`cation authority server. To buy an image, the
`user must first complete an online form to be
`authenticated by the Laboratorio Comuni-
`cazioni e Immagini certification authority. LCI
`then generates a security certificate based on
`the user information, which the client browser
`downloads and automatically stores in its cer-
`tificate database.
`■ Purchase images in the vendor Web server
`(available to users with personal certificates
`issued by LCI). A Java applet manages all steps
`of the image purchase. As Figure 7 shows, the
`applet displays thumbnails of all images in the
`vendor database, along with their correspond-
`ing CUNs. Using the secure socket layer (SSL)
`protocol, it sends the user’s personal certificate
`to the server and receives the watermarked
`image from the server. It then runs an image
`viewer, which displays the downloaded images
`and allows the buyer to save, as separate files,
`the image, the CS-signed image digest, and the
`encrypted third watermark. To get full func-
`tionality from the prototype, users also need a
`software developers’ certificate, available on the
`vendor page, to authenticate the Java applet.
`■ Detect watermarks. Users can download a
`software package from the vendor page, which
`includes the Java classes and the code to run
`the watermark decoder viewer shown in Fig-
`ure 8. Users must provide their surname and
`name, as indicated in the personal digital cer-
`tificate obtained by the certification authori-
`ty; the image CUN; the watermarked image;
`the third watermark encrypted with the CS
`private key; and the CS-signed image digest.
`The last three items are in the files saved by
`the image viewer.
`
`Figure 7. The vendor server. A Java applet provides thumbnails of all
`images in the database along with their CUNs.
`
`Figure 8. Watermark decoder viewer. After entering the required
`data, users can run the viewer to detect the third watermark.
`
`When all data are provided, the user starts the
`watermark-detection process by clicking “check.”
`First, the system generates the buyer’s PIN using
`surname and name strings. Using the CS public
`key, it then decrypts the third watermark and
`checks for the buyer’s PIN. Next, the system
`detects the watermark presence in the image.
`Finally, the applet computes the watermarked
`image’s digest and verifies that it corresponds to
`the CS-computed digest, which it receives from the
`vendor and stores locally in a file.
`In its current state, the prototype implements
`only part of the system to demonstrate the techni-
`cal feasibility of the proposed model. The proto-
`type has been tested by users at our laboratory,
`confirming the validity of the approach.
`
`Conclusion
`The use of watermarking technology to enforce
`copyright laws needs further investigation before
`it can be applied in real-world environments. In
`addition to addressing system robustness, we need
`in-depth protocol-level analysis to clarify what
`watermarking can and cannot achieve.
`Although practical applications can impose
`very severe requirements, which current technol-
`ogy is not flexible enough to meet, copyright pro-
`tection through digital watermarking is still fea-
`sible. The ECMS described here clearly demon-
`strates the technology’s potential. The proposed
`ECMS, though somewhat inflexible, can effec-
`tively enforce copyright laws, because it combines
`a watermarking mechanism with conventional
`
`IEEE INTERNET COMPUTING
`
`http://computer.org/internet/
`
`MAY • JUNE 2002
`
`25
`
`
`
`Watermarking Technologies
`
`cryptography to assess a document’s proper or
`improper use.
`
`References
`1. B. Schneier, Applied Cryptography, John Wiley & Sons,
`New York, 1994.
`2. F. Hartung and M. Kutter, “Multimedia Watermarking Tech-
`niques,’’ Proc. IEEE, vol. 87, IEEE Press, Piscataway, N.J.,
`July 1999, pp. 1079-1107.
`3. C.I. Podilchuk and E.J. Delp, “Digital Watermarking: Algo-
`rithms and Applications,’’ IEEE Signal Processing, vol. 18,
`no. 4, July 2001, pp. 33-46.
`4. F. Mintzer, G. W. Braudaway, and M.M. Yeung, “Effective
`and Ineffective Digital Watermarks,’’ Proc. IEEE Int’l Conf.
`Image Processing (ICIP 97), vol. III, IEEE CS Press, Los
`Alamitos, Calif., Oct. 1997, pp. 223-226.
`5. S. Craver et al., “Resolving Rightful Ownership with Invis-
`ible Watermarking Techniques: Limitations, Attacks, and
`Implications,’’ IEEE J. Selected Areas in Comm., vol. 4, no.
`16, May 1998, pp. 573-586.
`6. T. Furon, I. Venturini, and P. Duhamel, “Unified Approach
`of Asymmetric Watermarking Schemes,” Security and
`Watermarking of Multimedia Contents III, P.W. Wong and
`E. Delp, eds., Proc. SPIE, vol. 4314, 2001, pp. 269-279.
`
`Alessandro Piva is a postdoctoral researcher with the Depart-
`ment of Electronics and Telecommunications at the Uni-
`versity of Florence. He graduated in electronic engineer-
`ing and received a PhD in info