Researchers give high
marks to new technology for fingerprint
identification
Thursday, April 23,
2009
Determining the identity of the owners of latent prints is
often key to crime scene research, such as the owner of this latent
printed pulled from a fingernail file. A NIST test shows that
emerging software that automates the extraction of distinguishing
features from latent prints is more advanced than expected. Credit:
NIST
Overworked crime scene
investigators can take heart at the results of recent tests at the
National Institute of Standards and Technology (NIST) of new
technologies that automate the manual portion of latent fingerprint
identification. Prototype systems evaluated by NIST performed
surprisingly well for a developing technology: half of the
prototypes were accurate at least 80 percent of the time and one
had a near perfect score. Automating the manual portion of the work
frees up time for trained examiners to spend time on very difficult
images that the software has little hope of processing.
As any TV crime series fan knows, latent prints are left behind any
time someone touches something. While ubiquitous, “latents” often
include only part of the finger—maybe just a few ridges—and
sometimes are left on textured materials, adding even more
challenges.
To identify the owner, a fingerprint examiner must first carefully
mark the distinguishing features of the full or partial print,
beginning with the positions where ridges end or branch. Then the
latent is entered into a counter-terrorist or law enforcement
identification system such as the Federal Bureau of Investigation’s
Integrated Automated Fingerprint Identification System (IAFIS). The
FBI’s system compares latents against the 55 million sets of
ten-print cards taken at arrest.
The IAFIS system was a significant advance. Now the manual, mark-up
portion of latent fingerprint identification is being automated
with an emerging technology called Automatic Feature Extraction and
Matching (AFEM). NIST biometric researchers assessed prototypes
that eight vendors are developing.
In the evaluation, researchers used a data set of 835 latent prints
and 100,000 fingerprints that have been used in real case
examinations.
The AFEM software extracted the distinguishing features of the
latent prints, then compared them against 100,000 fingerprints. For
each print the software provided a list of 50 candidates that the
fingerprint specialists compared by hand. Most identities were
found within the top 10.
In order of performance, the most accurate prototypes were
furnished by NEC Corp., Cogent Inc., SPEX Forensics, Inc.,
Motorola, Inc. and L1 Identity Solutions. Results ranged from
nearly 100 percent for the most accurate product to around 80
percent for the last three listed.
The evaluations also showed a strong correlation between the number
of distinguishing features in a latent print and its ability to
match for all prototypes and that the quality of the image data
strongly influences accuracy.
“While the testing has demonstrated accuracy beyond pre-test
expectations, the potential of the technology remains undefined and
further testing is required,” said computer scientist Patrick
Grother. “In the future we will look at lower quality latent images
to understand the technology’s limitations and we will support
development of a standardized feature set that extends the one
currently used by examiners for searches.”
###
National Institute of Standards and Technology (NIST)
Thanks to National Institute of
Standards and Technology (NIST) for this article.