Increased Latent Print Accuracy: What it means for law enforcement

Increased Latent Print Accuracy:
What it means for law enforcement
Mike French, CLPE
This page intentionally left blank.
Increased Latent Print Accuracy
What it means for law enforcement
Contents
Introduction .............................................................................................................................. 1
NIST Tests show improved AFIS latent matching capability .............................................. 1
AFIS improvements help you confidently do more with less ............................................. 2
The need for human intervention........................................................................................... 3
Real-world tests of latent print auto-encoding ..................................................................... 3
i
Increased Latent Print Accuracy
What it means for law enforcement
This page intentionally left blank.
ii
Increased Latent Print Accuracy
What it means for law enforcement
Introduction
It is apparent that AFIS (Automated Fingerprint Identification Systems) have increased in accuracy over
the years, and these improvements have brought increased efficiency—regardless of who is your AFIS
provider. This article will discuss how AFIS improvements include not only tenprint but now also latent
print auto-encoding capabilities (as well as improved latent matching). These improvements can give
you the confidence to concentrate your latent expert-level staff where they are most effective, and
allow lesser-skilled staff to perform the first steps in the latent process, thereby improving efficiency
while solving more crimes. Probably the most exciting possibility of this technology advancement is the
ability to match latent prints in near real-time from the field, giving investigators immediate actionable
intelligence.
NIST Tests show improved AFIS latent matching capability
The NIST ELFT-EFS (Evaluation of Latent Fingerprint Technology-Extended Feature Set) tests are
promoted by AFIS providers as good news for the law enforcement community. But what do these NIST
test results actually mean for law enforcement day-to-day latent operations?
NIST (the National Institute of Standards & Technology) hosted the first Latent Testing Workshop in April
2006. The workshop was conducted to canvas expert opinion in the area of evaluation of automated
latent fingerprint matching algorithms.1 At that time, the auto-encoding technology could not equal the
skill of even a non-expert. Today, that gap is closing.
Two rounds of Extended Feature Set latent evaluation tests have followed
that first 2006 Workshop, in 2009 and 2012. According to the ELFT-EFS
Evaluation #1 Final Report, “One of the purposes of ELFT-EFS is to determine
the extent to which human feature markup is effective. There is a need to
know when image-only searching is adequate, and when the additional effort
of marking minutiae and other features is appropriate.”2
The chart below, reproduced from Evaluation #2, shows the search results in
which the AFIS algorithm found the matching print in the #1 position of the
match results. The ELFT test included seven tests; the chart below contains
the results of the two tests that best reflect real-world search practices:
searches of the image only, and of the image plus extended feature set, i.e. expert markup (minutiae,
ridge counts, cores/deltas, pattern class plus ridge quality maps, creases, dots, pores, etc.). For the top
three AFIS providers MorphoTrak, NEC, and Cogent, the results for “Valuable” latent prints ranged from
68.4% to 77.1% of matching prints returned in the #1 position.3
1
http://www.nist.gov/itl/iad/ig/latent.cfm
NISTIR 7775: ELFT-EFS Evaluation #1, March 2011
3
http://biometrics.nist.gov/cs_links/latent/elft-efs/IAI_2012/ELFT-EFS2_IAI_2012_Final.pdf
2
1
Increased Latent Print Accuracy
Provider
Image Only
Image + Extended
Feature Set
A
B
C
A
B
C
What it means for law enforcement
No value
(average 4
minutiae)
20.0%
8.0%
8.0%
20.0%
12.0%
20.0%
Limited value
(average 8
minutiae)
34.5%
28.3%
30.1%
35.4%
31.0%
35.4%
Valuable
(average 25
minutiae)
72.6%
68.4%
70.8%
75.9%
76.2%
77.1%
Average of all latent
prints (no value,
limited value, value)
67.2%
63.0%
65.0%
70.2%
69.9%
71.4%
As can be seen in the “No value” column, one surprising finding was the effectiveness of AFIS in
obtaining hits on poor quality latent prints that experts had pre-determined to be of “no value” and not
worth searching. For example, the matching print was found in the #1 position for 20% of those images
by provider A: MorphoTrak. Other providers obtained results that were not as good, but still
encouraging.
The lesson learned here is that such latent prints may be worth searching after all. While these “No
Value” latent prints cannot be used to justify an arrest, or used in court to prove identification, they
could enable an examiner to declare that a latent cannot be excluded as coming from an individual. They
could therefore be useful for the investigator as a lead and the first step in linking by other means a
Person of Interest to the investigation.
AFIS improvements help you confidently do more with less
As stated above, achieving “lights-out” latent AFIS capability has been a goal of the NIST ELFT-EFS test
since the outset. Is it now possible? No, full latent print automation—“lights out”—is still not
recommended (where the AFIS completes the feature extraction, performs
the search, and makes the hit/no-hit decision without human
intervention); but the results of Evaluation #2 in the chart above suggest
that you can operate in a semi-automated mode for the first steps of the
process. If you use auto encoding and auto search—“image-only”
searches—you can free up resources to concentrate your most-skilled staff
where they are most effective.
You probably have three levels of proficiency within your latent fingerprint
unit:



A beginning fingerprint analyst, able to interpret a latent print at basic level and enter it into an
AFIS
An intermediate level fingerprint analyst, capable of analysis, comparison, and evaluation of
routine latent print images
An expert, able to work with the most difficult latent print images and offer decisions across a
wide spectrum of casework
2
Increased Latent Print Accuracy
What it means for law enforcement
Improved auto-encoding AFIS capability means that the initial stages of the AFIS process (cropping the
latent from the evidence, encoding automatically, and submitting the search) will not require the same
skill level as the second part of the process (manual edit of the encoding, verification of search results).
This allows you to manage your resources so that latent prints are encoded automatically, searched
without any editing by a junior level analyst, and edited by the more skilled expert only when the imageonly search resulted in a “no-hit” and the case merits an additional review (e.g. major/serious crime).
The need for human intervention
The increase in AFIS accuracy has been accompanied by a decrease in the average number of candidates
reviewed by latent examiners. That is, AFIS technology improvements have
simultaneously increased hit rate and productivity. But we must not
eliminate the human verification step that confirms a match. It is still the
case that fully automatic search and verification may miss a number of
matches; although that number is becoming fewer and fewer as
technology improves. You will want to review image-only no-match
searches on a case by case basis according to your agency’s predefined
business rules (e.g. by case type, score threshold, etc.) as required.
Additionally, with the current emphasis on elimination of bias in
determining search results, freeing up your expert resources in the latent encoding stage will allow you
to invest more time in double blind verification to detect false positives and false negative errors.
Real-world tests of latent print auto-encoding
You may argue that the NIST tests are “lab results”—not truly indicative of real-world operations. So
what about real-world results? A MorphoTrak customer recently analyzed the performance of autoencoding versus expert (manual) encoding. Using their three-year old AFIS, they evaluated a total of 593
latent hits obtained with manually encoded latent prints. Of those hits, 521, or 87.9% of the matching
prints were ranked within the top ten respondents of the search results. Then they resubmitted all the
images for automatic encoding. The result was a 62% hit rate (368 hits), where the matching prints were
ranked within the top ten of the search results. When they added manual encoding to the remaining
prints, they achieved another 142 hits. Together these results nearly equal the results for full manual
encoding. Subsequent rounds of manual encodings yielded the remaining hits.
From NIST ELFT-EFS, and now real-world tests, the implication is clear: if auto-encoding latent prints can
achieve 70% of the matches as with expert manual encoding, then your expert latent staff can
confidently concentrate on the difficult cases, while AFIS and the junior analysts handle the first steps of
the latent search process. Based on the results they achieved as described above, that is exactly what
this MorphoTrak customer has done.
3
Increased Latent Print Accuracy
What it means for law enforcement
AFIS can efficiently link latent prints lifted as evidence at crime scenes with inked prints on file by using
advanced matching algorithms. Both image-only and human feature markup have their place.
And keep in mind that these results were on the previous generation of MorphoTrak matchers. Today’s
algorithms are more accurate—giving even better results. We encourage you to perform similar tests of
your own AFIS. You may be surprised at the result—and it might give you the confidence to rely on AFIS
latent auto-encoding to reduce your latent unit workload.
In conclusion, allowing your AFIS to perform latent auto encoding and search can help you because it:





Is effective.
Is fast and easy.
Gives a less skilled expert the opportunity to gain experience while contributing to the unit’s
productivity.
Allows the more skilled but scarce expert to concentrate on the most difficult latent prints or
important cases.
Can provide solid leads for further investigation on difficult “No Value” latent prints.
Latent print examiners will continue to be essential to the analysis and evaluation of match results.
However research and real-world results have shown that we can improve efficiency up to five-fold by
streamlining and eliminating some of the initial casework steps. We can also start to provide
investigators with near real-time suspect information from this improved technology.
4
Increased Latent Print Accuracy
What it means for law enforcement
Mike French, CLPE (Certified Latent Print Examiner), advises and validates new applications and products for
forensic, defense intelligence and civil applications, conducts expert training, and supports AFIS research and
development at MorphoTrak. Mike has over 15 years of experience in the forensic/biometric identification field. As a
latent print examiner he served as a latent print unit technical lead and trainer for crime scene processing, laboratory
development, and evidence examination. Mike’s specialties include crime scene investigation, evidence imaging, and
latent print development and matching, and biometric system accuracy testing. Mike has a bachelor’s degree from
Central Washington University and serves on the Journal of Forensic Identification Editorial Review Board and the
International Association for Identification (IAI) latent print identification subcommittee.
5
Increased Latent Print Accuracy
What it means for law enforcement
for more information:
MorphoTrak
Corporate Headquarters
113 South Columbus Street, 4th Floor
Alexandria, Virginia U.S.A. 22314
Toll-Free: (800) 601-6790
Phone: (703) 797.2600
Fax: (703) 706.9549