Databases Reference
In-Depth Information
2.3 Discrete Data vs. Multimedia
An established body of research [2] [3] [8] [11] [14] [15] [22] [24] has resulted
from work on Information Hiding and Watermarking in frameworks such as
signal processing and multimedia (e.g., images, video and audio). Here we
explore Information Hiding as a rights assessment tool for
discrete
data types.
Let us briefly explore the relationship between the challenges and tech-
niques deployed in both frameworks. Because, while the terms might be iden-
tical, the associated models, challenges and techniques are different, almost
orthogonal. Whereas in the signal processing case there usually exists a large
noise bandwidth, due to the fact that the final data consumer is likely human
- with associated limitations of the sensory system - in the case of discrete
data types this cannot be assumed and data quality assessment needs to be
closely tied with the actual watermarking process (see Section 2.2).
Another important differentiating focus is the emphasis on the actual abil-
ity to convince in court as a success metric, unlike most approaches in the
signal processing realm, centered on bandwidth. While bandwidth is a rele-
vant related metric, it does not consider important additional issues such as
malicious transforms and removal attacks. For rights assertion, the concerns
lie not as much with packing a large
amount
of information (i.e., watermark
bits) in the Works to be protected, as with being able to both
survive
removal
attacks and
convince
in court.
Maybe the most important difference between the two domains is that,
while in a majority of watermarking solutions in the multimedia framework,
the main domain transforms are signal processing primitives (e.g., Works are
mainly considered as being compositions of signals rather than strings of bits),
in our case data types are mostly discrete and are not naturally handled as
continuous signals. Because, while discrete versions of frequency transforms
can be deployed as primitives in information encoding for digital images [8],
the basis for doing so is the fact that, although digitized, images are at the
core defined by a composition of light reflection signals and are consumed
as such by the final human consumer. By contrast, arbitrary discrete data
is naturally discrete
1
and often to be ingested by a highly sensitive seman-
tic processing component, e.g., a computer rather than a perceptual system
tolerant of distortions.
2.4 Relational Data
For completeness let us briefly overview main components of a relational
model [7]. In such a model, relations between information items are explicitly
specified: data is organized as “a number of differently sized
tables
” [7] com-
posed of “related” rows/columns. A table is a collection of
rows
or records
1
Unless we consider quantum states and uncertainty arising in the spin of the
electrons flowing through the silicon.
