BUFFALO, N.Y. -- In 2007, University at Buffalo computer
scientist Sargur Srihari, Ph.D., one of the world's experts on
pattern recognition and its application to fingerprints and
handwriting, was selected by the National Academy of Sciences to
serve with other national experts on its Committee on Identifying
the Needs of the Forensic Science Community.
The committee's charge was to explore the current status of the
forensic sciences and to guide future research that would best
serve the U.S. justice system and its citizens.
Today, the NAS released the panel's findings in a report called
"Strengthening Forensic Science in the United States: A Path
Forward" that can be found at http://www.national-academies.org.
In the conversation below, Srihari, a SUNY Distinguished
Professor in the UB Department of Computer Science and Engineering,
discusses how his work in the field of pattern recognition has
influenced and guided his interest in forensic science and how the
NAS report's findings reflect the powerful potential that the field
holds given proper scientific support.
In addition to founding and directing UB's Center of Excellence
in Document Analysis and Recognition (CEDAR), Srihari is a fellow
of the Institute for Electrical and Electronics Engineers (IEEE)
and the International Association of Pattern Recognition. He
organized the first-ever international workshop on computational
forensics in the U.S. held last August at the National Academy of
Sciences. The author of more than 300 peer-reviewed publications
and the subject of numerous stories by such well-known media as ABC
News and The New York Times, he also has served on the Board of
Scientific Counselors of the National Library of Medicine.
How has serving on the NAS committee changed or influenced
your perception of how much science can be applied to
I am a believer in strong artificial intelligence. Given
sufficient time and resources, specific intelligent tasks that
humans perform can be automated. Forensic tasks are much more
challenging than reading postal addresses, for example, and, of
course, much more is at stake.
What does this report mean to you as a researcher?
The NAS report puts the focus on the need to develop more
objective methods. One of the areas that I hope we will be able to
develop is something I call "computational forensics." It is about
developing algorithms and software to perform forensic analysis.
The results of such analysis will be less prone to criticism that
forensic testimony is tainted by bias.
Hopefully, financial support of research in the forensic
sciences, which at present only the National Institute of Justice
is able to provide, will improve. Present support of National
Institute of Justice research is miniscule in comparison to support
for agencies like the National Institutes of Health and the
National Science Foundation. Currently, there are only a handful of
scientists in this field in the U.S.; so hopefully, additional
funding will attract more researchers.
Dr. Srihari, when you began your career in the 1970s, you
were researching how to get machines to read handwriting for the
U.S. Postal Service. Today, you study how to rigorously apply
science to forensics. Please describe how that evolution
My research for the past 30 years has been in the field of
computer science known as "pattern recognition." It is an area of
artificial intelligence that is about developing algorithms to get
computers to perform cognitive tasks normally performed by skilled
humans. An example of such a skill is that of recognizing
handwriting. That interest resulted in my work being funded by the
U.S. Postal Service, since they were interested in automatic
reading of handwritten addresses. At the time, it was considered a
difficult, if not impossible, task. Today, more than 95 percent of
all handwritten addresses in the U.S. are processed by
But the techniques of pattern recognition are far more general,
and forensics is a natural application for them: one where humans
perform specialized cognitive tasks and where algorithms could be
devised to provide an objective basis for their work. The Supreme
Court ruling in Daubert v. Merrell Dow, 1993, stated that "pattern
evidence" had to have a proven scientific basis if it was going to
be allowed in court. The first task we considered was developing
algorithms to compare handwriting samples. Computers allowed us to
perform large-scale tests to show that indeed this kind of
comparison can be done with a high degree of accuracy (Journal of
Forensic Sciences, July 2002). We were also able to develop
computational tools to assist the handwriting examiners.
Then my colleagues and I at CEDAR decided to look at
fingerprints, specifically friction ridge patterns, the swirling
lines that make up the print. With high-speed computers, we found
we were able to conduct large-scale tests to determine the degree
of individuality and uniqueness in a given fingerprint (Journal of
Forensic Identification, January 2008 and CEDAR Technical Report,
What surprised you most about working in the field of
The forensic examiners welcomed my participation in their
community, which was under attack. Here was a fairly small
community of underpaid and overworked individuals, whose work was
being challenged. They were eager to engage with, and learn from,
us. As an academic, it also is satisfying work because it is not a
highly lucrative commercial field, so my contributions can make a
difference even when profit-making is not the main priority.
How has the use of DNA evidence changed the whole field of
The accuracy of DNA has raised expectations for other forensic
modalities, but DNA is available only in about 10 percent of cases.
As with DNA, it would be useful for other types of evidence to
provide exact probabilities of having a random match. Such
probability calculations are to a large extent lacking.
You have been involved as an expert witness in several cases.
Can you discuss that experience?
I spoke about the degree of accuracy obtainable with handwriting
evidence using computational techniques. This has allowed
handwriting evidence to be allowed in several cases. Expert human
capability is still higher than today's computational
What are some of the biggest challenges in trying to bring
science into the courtroom?
Judges are open to hearing scientific testimony. However, they
have to make sure in a preliminary hearing that it will be
presented in a form that is understandable to a lay jury. For
example, the strength of the evidence cannot be expressed as raw
numbers or probabilities. Still, these ideas can only be simplified
Can you talk about the backlash where some defense attorneys
use the term "junk science" to discredit scientific methods in the
courtroom? Will this attitude intensify with the release of this
This report will galvanize defense attorneys' position that the
scientific basis for introducing evidence is lacking. But justice
has to be carried out. With this report, judges and juries can
become more aware that there is always an inherent error rate, and
therefore multiple types of evidence should be considered in
arriving at a final decision. Also, it has to be kept in mind that
science is always a work in progress. Newton's laws were a
reasonable explanation until they could not explain the behavior of
small particles; thus, quantum theory was introduced, which may,
one day be replaced by another theory, such as string theory.
Do you think the time is ripe for making forensics more
Yes, indeed. And applied science will have many economic
spinoffs. We saw that with U.S. Postal Service funding of our
research. Prior to those research successes, there was no funding
for such work at U.S. universities, and the field of postal
automation was left to Japanese and German companies. After the
USPS-backed research proved fruitful, American companies, such as
Northrup-Grumman -- which applied some of our technology to reading
IRS forms -- and Lockheed-Martin -- which became the prime USPS
contractor -- created many new technology jobs in the U.S. It also
allowed the USPS to remain competitive and to keep its costs