Databases Reference
In-Depth Information
(
k
1
,k
2
,k
1
,k
2
)
i
such that, for example, altering
w
2
will result in enforcing
w
1
.
•
Multiple Data Sources: The paper also points out that the solution handles
recovering watermarks from data derived from multiple data sources. This
scenario is of particular interest for example in the case of an equiJOIN
performed between two data sets. Because watermarks rely on a bias in
the association between attributes, they can be naturally retrieved from
such JOIN result under certain reasonable assumptions.
•
Categorical and Numerical Data Types: Watermarking at the intersection
of categorical and numerical types is also explored. It is of interest to
provide a rights assessment mechanism that could not only prove rights
but also that the associated data sets were actually produced “together”;
this is relevant for example if the intrinsic value of the data lies in the
actual
combination
of the two data types. The authors introduce initial
ideas.
•
Bijective Attribute Re-mapping: To handle a scenario in which categor-
ical attributes are re-mapped through a bijective function to a new data
domain, the authors propose to discover the inverse mapping. This is pos-
sible if the initial data domain features distinguishing properties (e.g.,
value occurrence frequency histogram) that are likely to be preserved in
the mapped result.
5 Related Work
So far we have discussed a set of relational data types and associated wa-
termarking methods enabling future rights assessment proofs. We now sur-
vey a number of related research efforts that explore Information Hiding and
Watermarking for relational data in other security contexts such as privacy
enforcement and license violators tracing.
5.1 Privacy and Rights Protection
In [4] Bertino et. al. explore issues at the intersection of two important di-
mensions in data-centric assurance, namely rights assessment and privacy, in
the broader context of medical data. A unified framework is introduced that
combines binning and watermarking techniques for the purpose of achieving
both data privacy and the ability to assert rights.
The system design borrows components from existing work. More specifi-
cally, the binning method (for
k
-anonymity) is built upon an earlier approach
of generalization and suppression by allowing a broader concept of gener-
alization. Similar to the
consumer-driven
paradigm discussed earlier in this
chapter, to ensure data usefulness, binning is constrained by usage metrics
that define maximal allowable information loss. An initial binning stage is
followed then by watermarking. The framework then deploys a version of the
