Databases Reference
In-Depth Information
In Digital Watermarking, the actual value to be protected lies in the Works
themselves whereas pure steganography usually makes use of them as sim-
ple value “transporters”. In Watermarking, Rights Assessment is achieved by
demonstrating (with the aid of a “secret” known only to Alice - “watermark-
ing key”) that a particular Work exhibits a rare property (“hidden message”
or “watermark”). For purposes of convincing the court, this property needs to
be so
rare
that if one considers any other random Work “similar enough” to
the one in question, this property is “very improbable” to apply (i.e., bound
false-positives rate). It also has to be
relevant
,inthatitsomehowtiestoAlice
(e.g., by featuring the bit string “(c) by Alice”).
There is a threshold determining the ability to convince the court, related
to the “very improbable” assessment. This defines a main difference from
steganography: from the court's perspective, specifics of the property (e.g.,
watermark message) are not important as long as they link to Alice (e.g., by
saying “(c) by Alice”) and, she can prove “convincingly” it is she who induced
it to the (non-watermarked) original.
In watermarking the emphasis is on “detection” rather than “extraction”.
Extraction of a watermark, or bits of it, is usually a part of the detection
process but just complements the process up to the extent of increasing the
ability to convince in court. If recovering the watermark data in itself becomes
more important than detecting the actual existence of it (i.e., “yes/no” an-
swer) then, from an application point of view, this is a drift toward covert
communication and pure Information Hiding (steganography).
2.2 Consumer Driven Watermarking
An important point about watermarking should be noted. By its very nature,
a watermark modifies the item being watermarked: it inserts an indelible mark
in the Work such that (i) the insertion of the mark does not destroy the value of
the Work, i.e., it is still useful for the
intended purpose
; and (ii) it is dicult for
an adversary to remove or alter the mark beyond detection without destroying
this value. If the Work to be watermarked cannot be modified without losing
its value then a watermark cannot be inserted. The critical issue is not to
avoid alterations, but to limit them to acceptable levels with respect to the
intended use of the Work.
Thus, an important first step in inserting a watermark, i.e., by altering it,
is to identify changes that are acceptable. Naturally, the nature and level of
such change is dependent upon the application for which the data is to be
used. Clearly, the notion of value or utility of the data becomes thus central
to the watermarking process. For example, in the case of software, the value
may be in ensuring equivalent computation, whereas for natural language
text it may be in conveying the same meaning - i.e., synonym substitution
is acceptable. Similarly, for a collection of numbers, the utility of the data
may lie in the actual values, in the relative values of the numbers, or in the
distribution (e.g., normal with a certain mean). At the same time, the concept
