Databases Reference
In-Depth Information
encoding for categorical types [36] by Sion et. al. in a hierarchical fashion,
for the purpose of defeating a data generalization attack of concern in this
framework. The paper then explores whether watermarking can adversely in-
terfere with binning and conclude that the interaction is safe. Experiments
were conducted aimed at validating the robustness of the proposed framework.
5.2 Fingerprinting
Another example application of Resilient Information Hiding as a tool aid-
ing rights management, is its deployment to “track” license violators by hid-
ing a specific mark inside the Work, uniquely identifying the party it was
sold/outsourced to. This application is commonly referred to as
fingerprint-
ing
. If the Work would then be found in the public domain, that mark could
be used to assess the source of the leak.
One significant matter of concern in fingerprinting are collusion attacks. In
a collusion attack, multiple attackers “collude” by obtaining multiple copies
of the same Work (e.g., by purchasing it separately under different identities)
watermarked with different marks, in the hope of “combining” the different
copies into a single un-watermarked version. Defending against this attack is
not possible in the general case when the number of colluding partners cannot
be upper bounded. If this upper bound can be determined however, several
results provide appropriate coding techniques that allow tracing even in the
case of collusion under minimal assumptions [5] [6] [13].
For relational data, the issue of fingerprinting has been discussed by Li et.
al. in [21] where they propose to deploy their multi-bit watermarking method
[19] for this very purpose. To handle collusion attacks the authors defer to
research in [5] [6] [13]. This work is discussed in more detail elsewhere in this
topic.
5.3 Tamper Detection through Fragile Watermarking
In [17] Li et. al. explore the issue of detecting malicious alterations to data
by embedding a “fragile” watermark in the data. While in this chapter we
presented watermarking as a technique deploying Information Hiding for the
purpose of rights assessment, in this context, “watermark” is attached to
a different semantics. Whereas in rights assessment, a watermark features
resilience to value-preserving data alterations, for the purposes of tamper
detection, the “watermark” will be “fragile” so as to become a detector for
exactly such alterations. The authors also propose to allow this watermark to
point at the locations where alterations have occurred in the data.
At an overview level, the method proceeds as follows. The data is parti-
tioned into secret subsets; a keyed cryptographic hash of each such subset (in
effect the traditional message authentication code MAC) is then embedded
in the group by re-ordering its items with respect to a canonical ordering,
based on a cryptographic hash of their primary key attribute. The encoding
