Databases Reference
In-Depth Information
det
(attr, wm key, db primary key, subset sz,
v
false
,
v
true
, c, map[], subset bnds[])
srt attr
←
sort on normalized crypto hash
(wm key,db primary key,wm key)
read pipe
←
null
do
{
tuple
←
next tuple
(srt attr)
}
until
(
exists
idx
such that
(subset bnds[idx] == tuple))
curr subset
←
idx
while
(
not
(srt attr.
empty
()))
do
do
tuple
next tuple
(srt attr)
read pipe = read pipe.
append
(tuple)
}
until
(
exists
idx
such that
(subset bnds[idx] == tuple))
subset bin
←
(at most subset sz elements of read pipe, excluding last read)
read pipe.
remove all remaining elements but last read
()
if
(map[curr subset])
then
mark data[curr subset]
←
decode
(subset bin,
v
false
,
v
true
, confidence)
if
(mark data[curr subset] != DECODING ERROR)
then
map[curr subset]
←
true
curr subset
←
idx
return
mark data, map
←
Fig. 9.
Watermark Detection (version using subset markers shown).
ilar to (or derived from) SQL
create table
statements. In addition, integrity
constraints (e.g., such as
end time
being greater than
begin time
)canbe
expressed. A tolerance is specified for each constraint. The tolerance is the
amount of change or violation of the constraint that is acceptable. This is an
important parameter since it can be used to tailor the quality of the water-
mark at the expense of greater change in the data. As mentioned earlier, if
the tolerances are too low, it may not be possible to insert a watermark in the
data. Various forms of expression are accommodated, e.g., in terms of arbi-
trary SQL queries over the relations, with associated requirements (usability
metric functions). For example, the requirement that the result of the join
(natural or otherwise) of two relations does not change by more than 3% can
be specified.
Once usability metrics are defined and all other parameters are in place,
the watermarking module (see Figure 10) initiates the process of watermark-
ing. An undo/rollback log is kept for each atomic step performed (i.e., 1-bit
encoding) until data usability is assessed and confirmed by querying the cur-
rently active usability plugins. This allows for rollbacks in the case when data
quality is not preserved by the current atomic operation.
To validate this consumer driven design the authors perform a set of ex-
periments showing how, for example, watermarking with classification preser-
vation can be enforced through the usability metric plugin mechanisms. More-
over, the solution is proved experimentally on real data to be extremely re-
