Constructing a Virtual Primary Key for Fingerprinting Relational Data
Conference Proceeding Article
Agrawal and Kiernan's watermarking technique for database relations  and Li et al's fingerprinting extension  both depend critically on primary key attributes. Hence, those techniques cannot embed marks in database relations without primary key attributes. Further, the techniques are vulnerable to simple attacks that alter or delete the primary key attribute.This paper proposes a new fingerprinting scheme that does not depend on a primary key attribute. The scheme constructs virtual primary keys from the most significant bits of some of each tuple's attributes. The actual attributes that are used to construct then virtual primary key differ from tuple to tuple. Attribute selection is based on a secret key that is known to the merchant only. Further, the selection does not depend on an apriori ordering over the attributes, or on knowledge of the original relation or fingerprint codeword.The virtual primary keys are then used in fingerprinting as in previous work . Rigorous analysis shows that, with high probability, only embedded fingerprints can be detected and embedded fingerprints cannot be modified or erased by a variety of attacks. Attacks include adding, deleting, shuffling, or modifying tuples or attributes (including a primary key attribute if one exists), guessing secret keys, and colluding with other recipients of a relation.
Information Security and Trust
Proceedings of the 2003 ACM Workshop on Digital Rights Management, Washington, DC, October 27-27
City or Country
LI, Yingjiu; Swarup, Vipin; and Jajodia, Sushil.
Constructing a Virtual Primary Key for Fingerprinting Relational Data. (2003). Proceedings of the 2003 ACM Workshop on Digital Rights Management, Washington, DC, October 27-27. 133-141. Research Collection School Of Information Systems.
Available at: http://ink.library.smu.edu.sg/sis_research/541