Database Reference
In-Depth Information
+,
"
!"#$
%&'
(
#%&'
())*
-
)
!
-
."
/
,#
,#
Fig. 4.
Architecture
different sensors that create an image of a latent fingerprint. These images and
sensor-
specific
meta data (such as resolution, sensor ID, etc.) are sent to our
Fingerprint Ver-
ification Database (FiVe DB)
. Note, that a sensor itself has no
insert
privileges on the
table containing the original sensor images. Instead, a sensor calls a
user defined func-
tion
(UDF), which tests whether the image and the embedded provenance information
hold integrity and authenticity (i.e., contains a valid watermark). If this is the case, the
UDF stores the image and inserts the sensor-specific meta data into a separate meta data
table.
Furthermore, we have a fine-grained management of meta data in FiVe DB. Meta
data that is accessed frequently by different user roles such as image resolution or
provenance-related meta data is stored in a dedicated schema (
Meta Data
). By con-
trast, meta data that is not intended for general purpose (e.g., uploader, upload time,
etc.) is inserted in meta data tables in schema
Original Sensor Images
. Additionally, we
collect meta data such as environmental parameters that the sensors cannot provide and
store these data within schema
Meta Data
as well.
After the initial capturing of a fingerprint image, typically several transformations
are performed to improve its quality or to extract some feature vectors. This is usually
done by external tools or functions. To identify a certain data item (i.e., original sensor
data or intermediate result due to transformation executed on the data), an external tool
first queries the available meta data of this data item to get its
database id
. Next, the
tool performs its transformation(s) on the data. Finally, the data item is sent back to
FiVe DB together with the information on the transformation. For this new data item
we have to know about the following two aspects:
•
Where
does it stem from, that is, which previous intermediate results possibly exist
and what is the original image data (i.e., foreign keys to these data items),
•
How
was the new data item created (i.e., the sequence of transformations).
Since every transformation (and even initial data creation) is accompanied with a
unique id of the executing unit (e.g., sensor, or tool), we can determine who changed the
data.
Furthermore,
we
obtain
information
regarding
the
source
provenance
by