Project LS-2005-WI-A
Titel Forensic writer identification: Machines versus Humans
Forensic writer identification enjoys renewed interest since 9/11
and the "antrax letters". The goal is to find the identity of
an unknown writer of a questioned document (e.g., threat letter),
given a database of handwritten samples by known suspects.
Interestingly, this application has resisted the use of artificial
intelligence, pattern-recognition and image-processing technology.
Human experts claim that they can do this job much more accurately.
However, technology has advanced considerably by recent work
in Groningen (Schomaker/Bulacu). In this project, collected data will be
analyzed, using the Beowulf cluster of >100 computers, in order
to find optimal parameters and features for realistic forensic handwritten
samples.
The ability to program (Java, C, C++) is a prerequisite.
Affinity with statistical analysis is useful.
Supervisor: Lambert Schomaker
| A questioned sample (top, orange border) and a hit list
of the forensic writer search engine. The first nearest neigbour
is in this case fortunately the correct writer (image border marked orange).
|