Databases Reference
In-Depth Information
keyed hash is computed on any data item
x
i
in this scheme, the last bit of
x
i
is ignored since it will be replaced with a watermark bit.
In watermark insertion, a group of data items are collected from the data
stream until a synchronization point is met and the number of elements in the
group is greater than
L
, where
L
is the lower bound of the group size. A group
hash value is computed by hashing the concatenation of all individual hash
values of data items in the group. The watermark embedder needs two buffers:
buff
0
for the current group, and
buff
1
for the next group. A watermark is
computed based on the current group hash and the group hash of the next
group. The length of the watermark is the same as the number of data items
in the group. The watermark is embedded by replacing the least significant
bits of all data items with the watermark bits, assuming that such change
does not degrade the value of data.
The watermark detection is also performed on two buffers using the same
K
,
m
,and
L
. As in watermark insertion, a watermark is computed from the
group hash value of the current group and the next group. If the computed
group hash matches the extracted watermark, the current group of data is
authentic. Otherwise, one needs to investigate the integrity of the previous
group before ascertaining the final verification result of the current group.
Since the embedded watermarks are chained, watermark detection can detect
and localize malicious modifications even if some whole groups are deleted
from the stream. The parameters
L
and
m
can be used to analyze the tradeoffs
between false detection rates and localization precision (in terms of the average
size of the group) in tamper detection [7]. The greater the
L
or
m
, the smaller
the false detection rates, and the lower the localization precision.
7.4 A Note on Watermarking XML Data
Agrawal and Kiernan's scheme has been extended by Ng and Lau to water-
mark XML data [20]. In this scheme, the owner of the XML data is responsible
for selecting the XML elements that are suitable to be locators, where a loca-
tor is defined to have a unique value that can serve as a primary key in the
watermarking process, as in Agrawal and Kiernan's scheme. The difference
between this scheme and Agrawal and Kiernan's scheme is that if a textual
value of an element is selected to embed a mark bit, one of its words is cho-
sen and replaced by a synonym function based on a well-known synonym
database WordNet. This scheme is further extended and deployed on a XML
compression system.
Gross-Amblard considered relational or XML data that are only partially
accessible through a set of parametric queries in his query-preserving water-
marking scheme [6]. The scheme modifies some numerical values in watermark
insertion in a way that the distortions introduced to the results of those para-
metric queries are small and that the watermark can be detected from the
results of those queries.
