62 lines
3.4 KiB
Plaintext
62 lines
3.4 KiB
Plaintext
|
Excerpt from Computer Design
|
||
|
|
April 1992
|
||
|
|
FUZZY MUG SEARCH HELPS COPS CATCH CROOKS
|
||
|
|
----------------------------------------
|
||
|
|
|
||
|
|
Well known to all devotees of detective shows is the scene where
|
||
|
|
a crime victim sits for long hours paging through mug books to
|
||
|
|
try to identify some evildoer. Even when police departments
|
||
|
|
have managed to computerize their databases of known
|
||
|
|
perpetrators, the process of narrowing the identification search
|
||
|
|
to a manageable number of mug shots based on a witness's
|
||
|
|
description is a tedious one.
|
||
|
|
|
||
|
|
Knowledge Based Systems of White Plains, NY added a fuzzy front
|
||
|
|
end to the image database of a major European police department
|
||
|
|
that significantly reduced the number of look throughs witnesses
|
||
|
|
had to do before finding a set of pictures they could seriously
|
||
|
|
work with to try to identify a subject. In the past, if someone
|
||
|
|
came in and said, ``He was kind of tall and heavy-set and looked
|
||
|
|
rather young,'' the police would have only a vague idea of what
|
||
|
|
group of pictures to start showing the witness. Even if the
|
||
|
|
system were computerized, someone would have to decide where the
|
||
|
|
cutoff point was for ``rather young,'' or ``tall.'' If a person
|
||
|
|
were described at 6'1'' but was really 5'11'', the system might
|
||
|
|
not catch the out of range number even if other factors in the
|
||
|
|
description pointed to an overall match.
|
||
|
|
|
||
|
|
In addition to implementing a front end to the database that
|
||
|
|
uses fuzzy sets to describe characteristics like ``old,''
|
||
|
|
``thin,'' ``tall'' and so on, KBS built in what it calls
|
||
|
|
``perspective shifting'' and ``semantic plies.'' Perspective
|
||
|
|
shifting changes the shape of the fuzzy set representing, say,
|
||
|
|
``tall'' if the witness is for instance a 16-year-old girl or
|
||
|
|
Japanese. It allows the system to search for ``old'' from a
|
||
|
|
``young'' perspective. Semantic plies adjust the description as
|
||
|
|
in ``tall for women'' or ``heavy for Samoans.''
|
||
|
|
|
||
|
|
The success of searching for a useful set of mug shots to
|
||
|
|
examine is because perspective shifting and semantic plies
|
||
|
|
affect the degree of belief in a fuzzy concept based on the
|
||
|
|
witness's characteristics; they do not make crisp distinctions.
|
||
|
|
Thus 5 feet will have a greater degree of membership in ``tall'
|
||
|
|
for a 10-year-old than for an adult, etc. The pictures of
|
||
|
|
subjects to look at are selected on an overall degree of truth
|
||
|
|
from the combined described characteristics, which in this
|
||
|
|
instance was set at 0.38. According to KBS, the old system
|
||
|
|
required an average of 16 look-throughs to find a set that
|
||
|
|
someone could actually work with, while the fuzzy system reduced
|
||
|
|
the number of look-throughs to two.
|
||
|
|
|
||
|
|
|
||
|
|
------------------------------------------------------------
|
||
|
|
This is article is provided with permission from Computer
|
||
|
|
Design. For subscription information to Computer Design, call
|
||
|
|
Paul Westervelt at (913) 835-3161. Do not redistribute in
|
||
|
|
an form (written or electronic) without permission from
|
||
|
|
Computer Design.
|
||
|
|
|
||
|
|
This information is provided by
|
||
|
|
Aptronix FuzzyNet
|
||
|
|
408-428-1883 Data USR V.32bis
|