| |
 |
 Launched in 1948 by
Allen Funt, Candid Camera brought smiles to millions of people around
the world. During the show hidden cameras caught ordinary people
being themselves. There was no pretense; their reactions to the sometimes
bizarre situations set up by the Candid Camera staff were genuine
and funny. Today, using cameras to capture the actions and features
of people in a variety of settings is no laughing matter.
In fact,
the emphasis on human recognition and identification, specifically
recognizing terrorists and other security related issues, has grown
exponentially since 9-11. Authorities around the world are anxious
to quickly and accurately confirm an individual’s identity
or assess suspicious behaviors with or without that individual’s
consent or, in some cases, their knowledge. Biometrics is the tool
they are using.
A biometric
measurement can be taken of any part of the body. Ideally, the measurement
should be constant so that it doesn’t change
over time or with a person’s mood. It should be distinctive so
that no two people could exhibit the same features. As important, it
should be something that can be easily and quickly measured, cataloged,
and referenced. A young and vibrant endeavor, the Notre Dame biometrics
program is contributing substantially to biometrics research and recognition
technologies:
Fingerprints are perhaps the best known form of biometrics.
Researchers in the College of Engineering are making breakthroughs
in face, ear, gait, and iris recognition technologies. Kevin
W. Bowyer, the Schubmehl-Prein Chair of the Department
of Computer Science and Engineering, and Patrick
J. Flynn, professor of computer science and
engineering, direct the college’s biometrics efforts. “Although
we started later than computer vision groups at other universities,” says
Flynn, “we have made significant advances in this area in a
relatively short time.” Both professors stress that the program’s
success is due largely to the work of undergraduates, graduate students,
and postdoctoral scholars. “Our projects have been national
in scope with the potential to have very high impact,” he says. “We
were able to identify niches and develop sufficient data to make
credible statistical inferences in many areas.”
Face Recognition
 In spite of warnings to not judge a book by its cover, people around
the world have been judging faces since time began. Look at the
portraits in any museum. The faces reflect beauty, youth, wisdom,
age, innocence, and evil as seen through the artist’s eyes.
Today, instead of trusting what can be seen, facial recognition
systems are helping security forces distinguish the innocent from
the evil beyond what is readily visible.
As optimistic as that sounds,
there are considerable challenges current facial recognition systems
must overcome. First and foremost is the myth that face recognition
technologies are disguise proof. While there are technologies,
such as two- and three-dimensional scans and infrared and visible
light imaging, being studied in the Department of Computer Science
and Engineering that will help identify if a person is wearing
additional makeup or a prosthesis, they are not infallible. Proper
lighting is also a factor in face recognition systems, as are the
angle of the shot and distance of the subject from the camera.
 Using
two- and three-dimensional images, Notre Dame researchers map the
topography (shape and depth) of a face. The outer corner of the
eye, the tip of the nose, and the center of the chin become landmark
points as a series of photos of an individual are taken: looking
directly at the camera with a normal expression, smiling, and with
spotlights.
In fact, several sets of images of the same individual
have been taken with at least six and as many as 14 weeks between
capture sessions. Infrared images have also been recorded and added
to a gallery of images. The Notre Dame collection is one of the
largest databases of faces in the world, featuring images captured
repeatedly from students, staff, and faculty throughout the University
over a four-year period.
More than 75,000 images from the Notre
Dame collection are being used as part of the 2006 Face Recognition
Grand Challenge (FRGC), which is sponsored by the National Institute
of Standards and Technology, Department of Homeland Security Science
and Technology Directorate, the Federal Bureau of Investigations
(FBI), the Intelligence Technology Innovation Center, and the Technical
Support Working Group. Participating researchers are provided with
the images and a six-experiment challenge which incorporates three-dimensional
scans, high-resolution still images, multiple still images taken
under a variety of conditions, multi-modal face recognition, multiple
algorithms, and preprocessing algorithms. The goal of the FRGC is
to reduce the error rates in current systems so that they may be
deployed for real-world applications.
Ear Recognition
 Ears are like fingerprints in that they are unique to each individual
and, without surgical intervention, do not change shape throughout
a person’s life. For this reason, ears have been used as
biometric tools but much less frequently than faces or fingerprints.
In fact, the United States Citizenship and Immigration Services,
formerly the Immigration and Naturalization Service, used to require
that every applicant provide two identical photos showing the entire
face, including the right ear and left eye, for use in visas or
passports. This policy was changed in August 2004 to comply with
the Border Security Act of 2003. All new photos must now exhibit
a full-frontal face position in color.
 The change in photographic
requirements, however, does not negate the usefulness of the ear
as a tool in biometric recognition. In a recent Notre Dame project,
which was funded by the National Science Foundation and the Intelligence
Technology Innovation Center, Bowyer and Ping
Yan, a graduate student
in the Department of Computer Science and Engineering, captured
multiple images of ears from more than 400 individuals. One of
the largest experimental investigations of ear biometrics ever
conducted, Bowyer and Yan applied four different approaches to
test the accuracy of their methods.
Among other variables, they
studied three different landmark selection methods. A landmark
on an ear, similar to one on the ground, provides a constant point
of reference and measurement. The landmarks they used each featured
two points: the first measured the distance between the triangular
fossa and antitragus, the second between the triangular
fossa and
incisure intertragica, and the third measured the distance between
two lines ... one along the border between the ear and the face
and the other from the top of the ear to the bottom, reflecting
the size of the ear.
They then applied Principal Component Analysis,
also called “eigenear,” which
is widely used in face recognition, to test for intensity, depth,
and edge matching. Bowyer and Yan found that ears are geometrically
complex and require a complex set of algorithms to assess the collected
data.
Although the experiment was conducted under controlled circumstances,
it appears that ear recognition based on a three-dimensional approach
is more than 90 percent accurate. However, the results also suggest
that while there is no significant difference between recognition performance
using the ear versus the face, using both the ear and the face in a
multi-modal system results in a statistically significant improvement
in recognition. |
 |
Wise as an owl,
blind as a bat, clever as a fox, boarish, catlike, having horse
sense ... people are often described in relation to animals.
It’s not surprising that researchers might apply similar
terms when attempting to categorize specific groups. In fact,
in 1998 George R. Doddington, who was at that time the senior
principal scientist with SRI International and visiting scientist
at the National Institute of Standards and Technology, co-authored
a paper on performance variability in speaker recognition systems
in which he suggested that all people could be classified into
one of four groups in regards to their speech patterns and how
well they could be identified by such systems.
 Doddington’s
menagerie was based on sheep, goats, lambs, and wolves. Sheep
represented the majority of the population. Readily distinguishable
one from another, they were easy for recognition systems to identify
and model. Goats were speakers who were more difficult for a
system to recognize. A goat, for example, might not provide a
match to itself from one day to the next. Lambs adversely affected
the performance of a speech recognition system, because they
were so similar one to another and were very easy to imitate.
Wolves, according to the Doddington scale, also negatively affected
recognition systems, because they were great imitators.
Notre Dame
researchers are extending Doddington’s concept
to facial recognition systems. “It is one of the most interesting
new projects in our lab [the Computer Vision Laboratory],” says
Patrick J. Flynn, professor of computer science and engineering. “Not
only are we examining Doddington’s Zoo in the context of
facial biometrics, but students are playing a key role in the process.”
Flynn
and undergraduates Michael G. Wittman and Patrick
M. Davis are
hoping to determine if the zoo exists and to identify what percentage
of the population can be found in each of the categories. Perhaps
more interesting to the team is the possibility of applying the
zoo concept to other biometric systems, such as iris recognition
or three-dimensional face shape.
In addition to his work on this
and other vision recognition projects, Wittman, a senior in the
Department of Computer Science and Engineering, interned with the
Intelligence and Information Systems division of Raytheon in Falls
Church, Virginia, during summer 2005. He will begin working as
a full-time Raytheon employee, a software engineer, in the Internal
Biometrics research and development program later this summer.
|
|
Gait
Recognition
 What do the story of the prodigal son and Ashley Wilkes’ homecoming
in Gone with the Wind have in common? Both the young man’s father
and Melanie Wilkes identified a loved one from a distance. Their recognition
was not solely for dramatic effect; the manner in which an individual
walks can be an identifying feature. The most recent evidence of this
was described in an article titled “Visual Analysis of Gait as
a Cue to Identity,” which was published in the December 1999
issue of Applied Cognitive Psychology. The article indicated something
known to anyone waiting and watching for a loved one’s return:
humans can, with very brief exposure and under a variety of lighting
conditions, identify other individuals by their gait.
Extensive research
has been done to address how accurately a subject could be identified
by the characteristics of his or her walk: How even are the strides?
Is there a noticeable limp? Studies have also been conducted to determine
how variables, such as terrain or load might affect the way a person
walks.
Little research has been done from the opposite perspective.
For example, an important question in today’s climate might be, “Can
an analysis of gait determine if a subject is carrying a concealed
load, such as a bomb?“ This is the question James
M. Ward, a
2004 graduate of the Department of Computer Science and Engineering
now working at GE Aircraft; Michael G. Wittman, a senior in the department;
and Flynn attempted to answer in their project, “Visual Analysis
of the Effects of Load Carriage on Gait.“
In order to conduct
the project in a controlled environment, the Notre Dame team collected
all of the images on campus, at the Stepan Center and at the Rolfs
Sports Recreation Center, using an area that was 2 ft. wide by almost
40 ft. long. Team members selected subjects who were 18 to 22 years
old, between 5’6” and 6’3” tall,
and between 140 and 200 lbs. Each participant wore black clothing on
which video markers were placed, and each took a practice walk down
the project path before filming began.
As they were taped the first
time down the runway, participants simply walked the length of the
path. After donning a 31-lb. vest, a second trip was recorded. Each
subject was then fitted with a 42-lb. vest and filmed for a third time.
After
converting the video clips, software was used to develop x and y coordinates
of all markers, and individual frames were analyzed to determine if
any trends could be found. According to the data, the upper body’s
joint trajectories vary greatly from person to person, but the lower
body acts in a much more predictable way, compensating for the extra
weight. Flynn stresses that while additional data and analysis is needed
to confirm these trends, gait is definitely something that deserves
continued exploration.
Iris Recognition
 Although some of the biometric techniques depicted by Hollywood are
more science fiction than fact, iris recognition is one of the
most accurate forms of identification known to man. The probability
of the irises from two individuals being identical is estimated
at 1 in 1072. Even identical twins have unique irises.
Iris recognition
technology differs from retinal pattern technology. A colorful
organ that surrounds the pupil, the iris acts like a shutter regulating
the amount of light that the eye receives. It features fibers,
furrows, freckles, and other patterns useful in a biometrics context.
An iris is externally visible; a retina is not. Located in the
back of the eye behind the cornea, lens, iris, and pupil, the retina
is connected to the optic nerve. It helps process images. The retina
also provides accurate biometric information, because of the unique
vessel patterns located in it. Like the iris, retinal patterns
are also thought to remain constant throughout a person’s
life.
 Benefits of iris recognition include the fact that it is a non-invasive
form of technology. High-resolution cameras quickly capture, store,
and analyze images without touching a person or damaging the eye.
The challenge faced by iris recognition systems is rooted in the
initial image capture. Currently, researchers acquire images from
a stationary subject positioned within inches of a camera. In controlled
facilities, such as a corporate laboratory or military base, the
subject might sit or stand six to 12 inches away from a camera’s
lens. Many controlled facilities also ask subjects to remove their
glasses, although it is not clear how an image is affected if a subject
is wearing glasses or contacts.
The cameras in automated teller machines,
which can also capture iris images, operate best when a subject is
17 to 19 inches away from the lens. Most important in either application,
the subject must hold completely still, remaining within camera range.
There must also be proper lighting.
While these types of systems might
work well in a corporate laboratory or on a military base, where
individuals expect to be scanned for identification purposes, they
are not yet practical for public places, where people are moving
around and lighting may not be consistent from one side of a room
to another. These are some of the factors being addressed by the
Iris Challenge Evaluation (ICE), an independent evaluation of iris
recognition technology being conducted by the National Institute
of Standards and Technology.
Researchers at Notre Dame, in conjunction
with ICE sponsors (the FBI, Intelligence Technology Innovation Center,
National Institute of Justice, Technical Support Working Group, and
the Transportation Security Administration of the U.S. Department
of Homeland Security), have provided image data sets and software
that ICE participants (academia, industry, and research institutes)
will use as they assess and measure current iris recognition efforts.
The goal of the challenge is to advance iris recognition technology
so that useful images may be acquired at greater distances, from
a variety of angles, under limited lighting conditions, and with
or without the subject’s
knowledge.
Multi-modal Biometrics
 Although the adage “two heads are better than one” was
written long before the advent of biometrics, Bowyer and Flynn contend
it applies quite well. In fact, they suggest that the future of biometrics
lies in the use of multi-modal systems. For example, iris recognition
alone cannot be applied to everyone. One in 17,000 people in the
U.S. have some type of albinism, which affects the pigment in their
irises. Acquiring an accurate image of an albino’s iris is
difficult. Involuntary eye movement can also confuse iris recognition
systems. Speaker recognition systems will never work on a mute person.
Gait technologies may trigger a false positive for someone who has
back problems or has had hip replacement surgery or polio. Because
of the involuntary motion associated with Parkinson’s disease,
the faces of people afflicted with that disease are difficult to
capture. Another segment of the population wears veils for religious
reasons, which means that facial recognition systems will not work
on them either.
In these cases, should authorities violate civil liberties
with additional and more invasive searches? Or can using a series
of biometric measurements, a multi-modal approach, help identify
potential threats while maintaining personal freedoms? The benefit
of multi-modal techniques, according to Bowyer and Flynn, is that
they offer a robustness of data which provides more accurate assessments
while maintaining most of those freedoms. But there is no simple
answer. Still in its infancy, biometrics will continue to evolve
to meet the challenges of identification, verification, and security
that are prevalent in today’s world. It will continue
to help nations assess what cannot always be seen with the naked eye.
For more information about computer recognition technology at Notre
Dame, visit http://www.cse.nd.edu/~cvrl. |
 |
In
the field of biometrics, fingerprinting is the oldest and, to
date, the most successfully applied technique. The use of fingerprints
as a means of identification, such as a legal and binding signature
on official documents, is recorded as early as 1000 B.C. In 1686
Marcello Malpighi, a professor of anatomy at the University of
Bologna, was the first to document and type fingerprints. But
it wasn’t until 1880 that Dr. Henry Faulds suggested that
fingerprints could be used as a means of personal identification.
In an article published in “Nature,” he outlined
a distinct classification system and described how best to capture
and record prints.
Building upon Faulds’ work Sir Francis
Galton became the first to prove that a person’s fingerprints
remain the same throughout his life and that no two individuals
have the same prints. In fact, he calculated that the odds of
finding identical prints were approximately 1 in 64 billion.
In 1892 Galton published Fingerprints, a book detailing the types
(arch, loop, and whorl) and characteristics (minutia) of fingerprints,
which are still used today.
Fingerprints, however, were not widely
used to identify criminals until 1901, when Sir Edward Richard
Henry began training Scotland Yard investigators in the Henry
Fingerprint Classification System, a uniform system of identification
that featured 1,024 classifications.
Another milestone in law
enforcement technology occurred in 1902, when Alphonse Bertillon,
director of the Bureau of Identification of the Paris Police,
made the first identification of a criminal by matching an unknown
print found at a crime scene with a print already on file. By
1903 fingerprinting technology was being used in the United States,
and a few years later a central storage location for North American
fingerprints was established in Ottawa, Canada. At its opening
it held 2,000 sets of prints.
Congress established the Identification
Division of the Federal Bureau of Investigation (FBI) in 1924,
the basis of the bureau’s
fingerprint repository, and by 1946 the division had processed
100 million fingerprint cards. By 1971 that number had doubled.
A
computerized system of storing and cross-referencing criminal
prints, the Automated Fingerprint Identification System, was established
in the 1990s, enabling law enforcement officers to search millions
of fingerprint files in a matter of minutes. By 1999 the FBI had
phased out the use of fingerprint cards. Although the cards are
still kept on file, the computerized fingerprint records for more
than 33 million criminals can be accessed in a matter of seconds.
 In
addition to applications in law enforcement, fingerprinting technologies
are being used in child identification kits for parents. They are
also employed by companies in a variety of different applications
and markets as biometric sensors. For example, companies are spending
millions of dollars to erect firewalls and install intruder detection
systems on their computer networks, but many are also replacing
passwords with fingerprint sensors. These sensors can control laptops,
computer mice, portable hard drives, and are even being used to
personalize wireless devices. In some cell phone systems different
fingers operate different speed dials or open separate buddy lists.
Fingerprint biometrics can also control door locks and smart card
readers.
Tried and true, fingerprinting is not the ultimate biometric. Nor
can it be used to accurately identify everyone. The fingers of
bricklayers, rock climbing enthusiasts, senior citizens, and toddlers
often lack the ridges needed to produce a defined print. Other
errors, technological or human, can also plague the identification
process. For example, in May 2004 the FBI arrested Brandon Mayfield
as a material witness in the March 2004 commuter train attacks
in Madrid. FBI officials said Mayfield, an Oregon attorney who
was also Muslim, had been identified as the source of a fingerprint
on a bag of detonators connected to the attack. Mayfield was released
two weeks later, after the FBI examined the prints of a suspect
the Spanish National Police had correctly identified as the source
of the print. |
|
 |
Patrick Henry is
famous for saying, “Give me liberty or give me death.” But
he also said, “Guard with jealous attention the public
liberty. Suspect everyone who approaches that jewel. Unfortunately,
nothing will preserve it but downright force. Whenever you give
up that force, you are inevitably ruined.” In the Historical
Review of Pennsylvania, Benjamin Franklin wrote, “They
that can give up essential liberty to obtain a little temporary
safety deserve neither liberty nor safety.” Obviously,
both founding fathers valued liberty, perhaps above all else.
Discussing
his memorial tribute “Let’s Roll” in
the April 2002 issue of USA Today, Neil Young called Franklin’s
love of personal freedom into question. He said, “Benjamin
Franklin said that anyone who gives up essential liberties to preserve
freedom is a fool, but maybe he didn’t conceive of nuclear
war and dirty bombs.”
It’s probable that neither Franklin
nor Henry envisioned a day when a jet could fell a building and
kill almost 3,000 people in a matter of minutes. Neither did they
imagine a device small enough to hold in the palm of a hand but
powerful enough to record a conversation inside a building or behind
a wall. They certainly could not have conceived of removing their
shoes and placing them, along with other personal items, in a scanner
prior to boarding an aircraft. And, even though espionage was part
of their world, it’s likely they never considered the possibility
of a suicide bomber or any kind of environmentally toxic device.
Technology was as foreign to them as it is familiar to today’s
society.
The
question raised, however, is the same: Is security (safety) or
personal liberty (privacy) more important? On September 12, 2001,
most Americans would have gladly sacrificed a little convenience,
even a little privacy, to change the horror of the previous day.
Powerless to change the past, they (we) continue to search for
answers, philosophical and technological, so that this particular
piece of history can never repeat itself.
|
|
|
|
