Applications

Editor: Michael J. Potel

http://www.wildcrest.com

Graphics and Security: Exploring Visual Biometrics ____

Kirk L. Kroeker
B iometrics is the science of recognizing a person on
the basis of physical or behavioral characteristics.
Things you can carry, such as keys or ID badges, can of
You can easily forget a password, but you’ll never forget
to bring your face, finger, or eye.

course be lost, stolen, or duplicated. The same goes for Face-recognition technology
things that you know, such as passwords or personal ID As Figure 1 shows, all face-recognition technologies
numbers. Biometrics relies on who you are—on one of share certain commonalities, such as emphasizing those
any number of unique characteristics that you can’t lose sections of the face that are less susceptible to alteration,
or forget. including the upper outlines of the eye sockets, areas sur-
Most biometric systems can be set to varying degrees rounding the cheekbones, and sides of the mouth.1
of security, which gives you more flexibility to determine Facial-scan technology works well with standard PC
access levels. Increasing security in biometric systems video capture cameras and generally requires cameras
sometimes makes them more restrictive, resulting in an that can capture images at least at 320 × 240 resolution
increased false rejection rate. The net effect of false and at least 3 to 5 frames per second. More frames per
rejection rates is usually nothing more than inconve- second, along with higher resolution, will lead to better
nience. However, if security is set too low, the false performance in verification or identification, but higher
acceptance rate might increase, which turns out to be rates typically aren’t required for basic one-to-one veri-
potentially far more serious since it involves an unau- fication systems that compare your face scan to a tem-
thorized person gaining access to protected resources. plate you’ve previously stored on the verifying system.
Furthermore, many companies use biometric securi- Because such cameras cost as little as $50.00, and
ty in addition to standard passwording systems—as a demo versions of leading vendors’ software are freely
layer of additional identity verification. Of course, many available, facial recognition is one of the few biometrics
biometric systems are expensive and sacrifice some mea- with which you can experiment on a limited budget. For
sure of personal privacy. To verify your face, finger, or facial recognition at long distances—especially for crowd
iris, you must have some personal data on file in the ver- recognition systems (see Figure 2)—a strong correlation
ifying system—personal data that can be stolen or made exists between camera quality and system capabilities.2
public. But biometric systems are becoming increasing- And for large-scale one-to-many searches—where you
ly popular both as standalone security systems and as might be comparing a face scan to several thousand face
added security largely because of one trait: convenience. templates to discover somebody’s identity—processor
speed is critical. But getting started
doing one-to-one verification can be
almost as cost effective as a standard
passwording system.

Face-recognition process
As with all biometric technologies,
1 Visionics’ sample capture, feature extraction,
FaceIt face- template comparison, and matching
recognition define the process flow of facial-scan
biometric sys- technology. The sample capture
tem creating a process will generally consist of 20
face template. to 30 seconds during which a facial-
recognition system will take several
Courtesy of Visionics

pictures of the subject’s face. Ideally,

the series of pictures will incorporate
slightly different angles and facial
expressions to allow for more accu-
rate searches. After entering a sub-

2 July/August 2002 0272-1716/02/$17.00 © 2002 IEEE

ject’s general face scan, the system—no matter what ven-
dor—will typically extract the subject’s distinctive fea-
tures and create a graphic template.
The exact algorithm any given commercial system uses
to create and then later verify the templates is typically a
closely guarded secret. The template is much smaller than
the image from which it’s drawn. Whereas quality facial
images generally require 150 to 300 Kbytes, templates will
only be approximately 1 Kbyte. Visionics, one of the most
prominent biometric vendors, uses an even smaller 84-
byte template to help accelerate one-to-many searches.
Authentication follows the same protocol. Assuming
your user is cooperative, he or she stands or sits in front
of the camera for a few seconds and is either verified or

Courtesy of Visionics
rejected. This comparison is based on the similarity of
the newly created template against the template on file.
One variant of this process is the use of facial-scan tech-
nology in forensics. The templates come from static pho-
tographs of known criminals and are stored in large
databases. The system performs a one-to-many search 2 Face-recognition software used to recognize individuals in a crowd—like
of these records to determine if the detainee is using an the kind used in Las Vegas or at high-security events—typically scans
alias. If the database has only a handful of enrollees, this crowds actively and tries to match the scans with a large database of
kind of search isn’t terribly processor intensive. But as known criminals. Crowd scanning technology, like Visionics’ FaceIt software
databases grow large, into the tens and hundreds of shown here, requires high-end video capture devices and fast processors.
thousands, this task becomes more difficult. The system
might only narrow the search to several likely candi-
dates and then require human intervention at the final track them and crop them out of the scene, matching
verification stages. the face against a watch list. FaceIt can also compress a
Another variable in identification is the dynamic face print into the 84-byte template for use in smart
between the target subjects and capture device. cards, bar codes, and other limited-size storage devices.
Standard verification typically assumes a cooperative FaceIt uses what the company calls local-feature
audience, one consisting of subjects motivated to use analysis to represent facial images in terms of local
the system correctly. Facial-scan systems, depending on building blocks. Visionics developed this mathematical
the exact type of implementation, might also have to be technique based on the understanding that all facial
optimized for uncooperative subjects. Uncooperative images can be synthesized from an irreducible set of ele-
subjects are unaware that a biometric system is in place, ments, not what you might assume to be the basic ele-
or don’t care, and make no effort to be recognized. ments of the face, such as the eye, nose, or mouth. These
Facial-scan technologies are more capable of identify- elements are derived from a representative ensemble of
ing cooperative subjects. faces using statistical techniques that span multiple pix-
els and represent universal facial shapes but aren’t com-
Visionics’ FaceIt technology monly known facial features.
Visionics FaceIt technology is a face-recognition bio- According to Visionics, more facial building elements
metric system that can automatically detect human pres- exist than facial parts. However, synthesizing a given
ence, locate and track faces, extract face images, and facial image to a high degree of precision requires only a
perform identification by matching against a database small subset (12 to 40 characteristic elements) of the total
of people it has seen before. The technology is typically available set. Identity is determined not only by which
used for one-to-many searching, verification, monitor- elements are characteristic but also by the manner in
ing, and surveillance. To determine someone’s identity which they’re geometrically combined—that is, by their
in identification mode, FaceIt computes the degree of relative positions. FaceIt maps an individual’s identity
overlap between the live face print and those associat- into a mathematical formula—which the company calls
ed with known individuals stored in a database of facial a face print—that the system can match and compare to
images. The system can return a list of possible individ- others. According to Visionics, the face print resists
uals ordered in diminishing score or it can simply return changes in lighting, skin tone, eyeglasses, facial expres-
the top match and an associated confidence level. sion, and hair variations. The face print contains the infor-
In verification mode, the face print can be stored on mation that distinguishes a face from millions of others.
a smart card or in a computerized record. FaceIt match-
es the live print to the stored one. If the confidence score Fingerprint-recognition technology
exceeds a certain threshold, then the match is success- For decades, fingerprinting was the common ink-and-
ful and the system verifies the user’s identity. FaceIt can roll procedure used when booking suspects or conduct-
find human faces anywhere in the field of view and at ing criminal investigations. Today, forensic scientists use
any distance—depending on the quality of the video fingerprint applications in large-scale one-to-many
capture device being used—and it can continuously searches on databases of up to millions of fingerprints. In

IEEE Computer Graphics and Applications 3

Applications

the core of most finger-scanning

Fingerprint Features technology.5 Much like the face-
The human fingerprint consists of ridge recognition companies, each of the
patterns that are traditionally classified primary finger-scan vendors has a
according to the decades-old Henry proprietary feature-extraction

Courtesy of Kinetic Science

system: left loop, right loop, arch, whorl, mechanism they typically guard
and tented arch. Loops make up nearly because it distinguishes them from
two thirds of all fingerprints, whorls are their competitors. Generally, once a
nearly one third, and perhaps 5 to 10 fingerprint-recognition system cap-
percent are arches. These classifications are tures a quality image, it converts the
relevant in many large-scale forensic image into a usable format. If the
applications but are rarely used in A Fingerprint scan image is grayscale, the system dis-
biometric authentication. The produced by Kinetic cards areas lighter than a particular
discontinuities that interrupt the otherwise Sciences’ optical threshold and it makes darker areas
smooth flow of ridges are the basis for fingerprint scanning black. It then thins the ridges to one
most fingerprint authentication technology. pixel for precise location of endings
techniques (see Figure A). and bifurcations.
Codified in the late 1800s as Galton The point at which a ridge ends,
features,1 many types of minutiae reside in a fingerprint, including and the point where a bifurcation
begins, are the most rudimentary
■ dots (very small ridges), minutiae and are used in most fin-
■ islands (ridges slightly longer than dots, occupying a middle gerprint-recognition applications.
space between two temporarily divergent ridges), Once the point has been situated, its
■ ponds or lakes (empty spaces between two temporarily location is commonly indicated by
divergent ridges), the distance from the core, with the
■ spurs (a notch protruding from a ridge), core serving as the center point on
■ bridges (small ridges joining two longer adjacent ridges), and an x–y axis. In addition to using the
■ crossovers (two ridges that cross each other). location of minutiae, some vendors
classify minutiae by type and quality.
Other features are essential to finger-scan authentication. The core The advantage of this is that search-
is the inner point, normally in the middle of the print, around es can proceed more quickly, as a
which swirls, loops, or arches center. particularly notable minutia might
be distinctive enough to lead to a
Reference match. A vendor can also rank high-
1. A.K. Jain and F. Farrokhnia, “Unsupervised Texture Segmentation Using versus low-quality minutia and dis-
Gabor Filters,” Pattern Recognition, vol. 24, no. 12, 1991, pp. 1167-1186. card the latter.
Getting good images of these dis-
tinctive ridges and minutiae is a com-
fact, fingerprint technology is the most common bio- plicated task. The fingerprint presents only a small area
metric technology on the market.3 And there’s good rea- to take measurements and the wear of daily life, which
son this popularity. Naeem Zafar, president of Veridicom, ridge patterns show most prominently. Vendors have
a prominent biometric systems vendor, points out that developed increasingly sophisticated mechanisms to cap-
“fingerprint biometric provides a level of security at a ture the fingerprint image with sufficient detail and res-
price point and form factor that makes it most conve- olution. The main fingerprint-scanning technologies in
nient for portable devices and IT applications.” use today include optical, silicon, and ultrasound.
Although finger-scanning technology can be used on Optical technology is the oldest and most widely used.
large databases, it’s frequently used for one-to-one ver- To do an optical scan, the user typically places his or her
ification to provide system access to individual users.4 finger on a clear scanning platform, such as the one
Zafar suggests that “fingerprint authentication deliver- shown in Figure 3. In most cases, a device simply con-
ing security, disguised as convenience, will start enter- verts the image of the fingerprint with dark ridges and
ing our lives over the next two to five years.” Initially, he light valleys into a digital signal and adjusts the contrast
says, the technology will manifest itself in government automatically.
projects, aviation security, and fraud-reduction pro- Silicon technology has gained considerable accep-
grams but ultimately it will capture consumer attention tance since its introduction in the late 1990s. Most sili-
by freeing people from the password jungle. con technology relies on direct-current capacitance. The
silicon sensor acts as one plate of a capacitor and the fin-
Fingerprint-recognition process ger is the other. The software then converts the capaci-
Once a fingerprint-recognition system captures a tance between platen and finger into a digital image.
high-quality image, it takes several steps to convert the Silicon generally produces better image quality than
fingerprint’s features into a compact template (see the optical technology. Because the silicon chip comprises
sidebar “Fingerprint Features” for more information). discrete rows and columns—typically between 200 and
This process, typically known as feature extraction, is at 300 lines in each direction on a 1-cm wafer—it can

4 July/August 2002
return detailed data. Silicon chips are small enough to template and store it in memory or
be integrated into many devices that can’t accommo- on disk.
date optical technology. Like many fingerprint-biometric
Ultrasound technology, although considered perhaps technologies, the actual fingerprint
the most accurate of the finger-scan technologies, isn’t image is never stored and can’t be
yet widely used. Ultrasound can penetrate dirt and constructed from templates. To iden-
residue, countering a main drawback to optical tech- tify or verify a fingerprint, a propri-

Courtesy of Guardware Systems

nology. However, implementing ultrasound scans is still etary SEIR matching algorithm
more expensive than other fingerprint-scan technolo- compares the extracted minutiae
gies. In ultrasound scanning, a device sends a short points from the input fingerprint on
ultrasonic pulse from several different directions toward the optical module to a previously
a finger surface and then measures the response. This stored sample. The entire matching
pulse response results from the contact scattering of the process takes roughly 1 second.
ultrasonic wave on the surface of the fingertip. Based Authentication takes place either
on a set of such responses, the scanning system recon- locally or on a server, depending on
structs an image of the finger’s surface structure. system configuration. 3 The high-grade glass used in
SecuGen embeds its core technol- Guardware’s optics makes the scan-
Veridicom’s silicon technology ogy in optical modules that work ners scratch-resistant. This also
Finger-recognition software and silicon sensors like with the set of extraction and match- allows SystemsGuard, the company’s
the one shown in Figure 4—both based on technology ing algorithms developed for use fingerprint-scanning technology, to
originally developed at Bell Labs—work together to cap- with the company’s SEIR optical be built into a front desk or other
ture and match your fingerprint in Veridicom’s method. For example, the company work areas where the front of the PC
OpenTouch technology. The technology offers a modu- embeds each module in its line of fin- might be difficult to access or there’s
lar hardware and software system for collecting, gerprint PC peripheral devices and little room for a desktop unit.
enhancing, processing, and verifying fingerprint images. standalone devices produced by orig-
Veridicom’s silicon fingerprint sensor provides 500-dpi inal equipment manufacturers for
resolution. The compact sensor is, according to Veridicom, various applications.
hard and resistant to scratches, abrasion, chemicals, cor-
rosion, and impact. The sensor’s surface consists of a sili- Iris recognition
con chip containing an array of 90,000 capacitor plates The holders of the major iris-recognition patents—
with sensing circuitry at 500 dpi. The capacitor-sensing Leonard Flom, Aran Safir, and John Daugman—found-
plates create an 8-bit raster-scanned image of the ridges ed Iridian Technologies. Since commencing operations
and valleys of the finger pressed against the chip. Software in 1993, they’ve dominated the iris recognition field.6
converts this information into a video signal. Typically, a Iridian has historically focused on access control, but its
scan takes from one-tenth to one-half a second to com- current emphasis has been shifting to e-commerce, med-
plete, depending on the processor’s speed. ical records, network identification, and online banking.
Veridicom’s software then creates a template from the So accurate are the algorithms used in iris recognition
scanned image. The system instantly erases the actual that, according to the company, the entire planet could
fingerprint image and stores the minutia data, which be enrolled in an iris database with only a small chance
becomes a unique digital fingerprint template of that of false acceptance or false rejection.
person. Future fingerprint readings for that individual Iris recognition is of course based on the visible quali-
are compared against it using the fingerprint-verifica- ties of the human iris (see Figure 5, next page). Visible
tion module in Veridicom’s verification suite. To verify characteristics include rings, furrows, freckles, and the
an individual’s identity and to authorize transactions, iris corona. Iridian’s iris-recognition technology converts
the fingerprint-verification module compares a live these visible characteristics into an IrisCode, a template
reading from a finger placed on the sensor with the stored for future verification attempts. From the 11-mm
minutia data template stored for that individual. If the
data match, the individual’s identity is verified and the
transaction is authorized. If the data don’t match, the 4 Veridicom’s
transaction is rejected. FPS200 is the
latest genera-
SecuGen’s optical technology tion of silicon-
At the most basic level, all optics-based fingerprint sys- based
tems translate illuminated images of fingerprints into fingerprint
digital code for further software processing. SecuGen sensors. It’s
devices use the company’s proprietary Surface Enhanced designed for
Irregular Reflection technology to capture high-contrast, integration into
high-resolution fingerprint images. A series of SEIR algo- the smallest
©2002 Veridicom

rithms developed by SecuGen extract data from the wireless or

image, mapping the distinguishing characteristics of fin- computing
gerprint ridge ends, splits, dots, and arches. The algo- device.
rithms then convert this data into a 400-byte digital

IEEE Computer Graphics and Applications 5

Applications

to a homogenous rubber sheet that, despite its distor-

tion, retains certain consistent qualities. Regardless of
the iris’ size at any given time, the algorithm draws on
5 The human the same amount of data, and its resultant iris code is
iris. The primary stored as a 512-byte template.
visible charac- The entire iris-scanning process is brief. The camera
teristic of the normally locates the iris in a quarter second and gener-
iris is the trabec- ates the iris code within 1 second. Database search times
ular meshwork, are quick, with hundreds of thousands of records ana-

Courtesy of Iridian Technologies

the tissue that lyzed per second, depending on the computer’s speed.
gives the The iris-capture process does run into the limitations of
appearance of grayscale imaging technology, where the darkest shades
dividing the iris of iris colorations are difficult to distinguish from the
radially. pupil. But according to Iridian, the algorithm’s robust-
ness actually allows for significant variations in image
quality. The same iris might at different times produce
iris codes that vary by as much as 25 percent, which
might sound like a flaw. But according to Daugman, the
odds of a randomly selected iris code coming close to
another match are exceptionally small.
Already several iris recognition and verification appli-
cations exist. Many companies license the technology
from Iridian to create their own products. One such
product, Panasonic’s Authenticam (see Figure 7), uses
Iridian’s Private ID iris-recognition technology and
6 One of Iridian comes with I/O Software’s SecureSuite to let multiple
Technologies’ users access PCs, files, folders, applications, and pass-
iris recognition word banks. In addition to providing security for stan-
products called dard information-access applications, you can use
IrisAccess Panasonic’s camera to authenticate users for videocon-
Courtesy of Iridian Technologies

ferencing and online collaboration.

Conclusion
Biometrics technology has come a long way from sim-
pler forms of systems security. But are biometrics-based
systems more secure or do they simply require crackers
to become more proficient at breaking into systems? To
recognize your fingerprint requires that a template of
your fingerprint actually be present in the system that
diameter iris, Daugman’s algorithms provide 3.4 bits of verifies your access. If you want to pass as somebody
data per square millimeter. This information density else, presumably you’d have to either have that person’s
means that each iris can have 266 unique spots—com- finger with you or you’d need to change the verifying
pared to 10 to 60 unique spots for traditional biometric template residing in the system that verifies your print.
technologies.7 Cracking into a system and replacing a legitimate
The first step in scanning an iris is locating it with a print with your own isn’t easy to do unless the system’s
dedicated camera no more than three feet from the eye security is poor. While biometric proponents stress the
(see Figure 6). After the camera situates the eye, Iridian’s strength of their proprietary technologies or biometrics
algorithm locates the outer and inner edges of the iris in general, no system is ever completely secure. Bruce
and then proceeds to analyze it. Iridian’s algorithm uses Schneier once pointed out that all computer security is
2D Gabor wavelets8—transforms used typically in visu- like putting a wooden stake in front of your house and
alization applications—to filter and map iris segments hoping that tresspassers will run into it.9 Contrary to
into hundreds of vectors. The wavelets assign values what many biometric proponents would have us
drawn from the orientation and spatial frequency of believe—that biometric security outclasses traditional
select areas of the iris and they then form an IrisCode. forms of security—all biometric systems are, after all,
According to Daugman, the equal-error rate (the point at another form of computer security with its own set of
which the likelihood of a false accept and false reject are strengths and weaknesses.
the same) is one in 1.2 million for IrisCodes. Biometrics effectively trade some amount of privacy
When the pupil expands and contracts—something and cost effectiveness for ultimate convenience—and
that occurs naturally with any change in lighting—it these systems are certainly no less secure than standard
skews and stretches the iris. Iridian’s algorithms account passwording systems. Passwording systems are cheap.
for such alteration after locating the iris boundaries at Complex biometric scanning equipment is usually
the outer and inner edges. Daugman draws the analogy expensive. But biometrics seems to be where the indus-

6 July/August 2002
try is headed. Aside from the Orwellian connotations,
biometrics systems offer an enormous amount of con-
venience to users. And, in the present political climate, 7 Panasonic’s
it’s hard to counter the argument that we should adopt Authenticam
biometric systems simply as additional layers of securi- uses Iridian
ty on top of traditional passwording systems. ■ Technologies’
Private ID soft-
ware to offer
References one-to-many
1. H. Wechsler et al., Face Recognition: From Theory to Appli- identification
cation, Springer-Verlag, Berlin, 1998. for applications
2. P.J. Phillips et al., The Feret Evaluation Methodology for Face- such as informa-
Recognition Algorithms, NISTIR 6264, Nat’l Inst. of Stan- tion access or

Courtesy of Panasonic
dards and Technology, Gaithersburg, Md., 1998, even videocon-
http://www.itl.nist.gov/iaui/894.03/pubs.html#face. ferencing.
3. A.K. Jain et al., “An Identity-Authentication System Using
Fingerprints,” Proc. EuroSpeech 97, IEEE CS Press, Los
Alamitos, Calif., 1997, pp. 1348-1388.
4. N. Ratha et al., “A Real-Time Matching System for Large
Fingerprint Databases,” IEEE Trans. Pattern Analysis and 8. D. Gabor, “Theory of Communication,” J. Institute of Elec-
Machine Intelligence, vol. 18, no. 8, Aug. 1996, pp. 799-813. trical Engineers, vol. 93, no. 26, Nov. 1946, pp. 429-457.
5. K. Karu and A.K. Jain, “Fingerprint Classification,” Pattern 9. B. Schneier, “Cryptographic Design Vulnerabilities,” Com-
Recognition, vol. 29, no. 3, 1996, pp. 389-404. puter, vol. 31, no. 9, Sept. 1998, pp. 29-33.
6. L. Flom and A. Safir, Iris Recognition System, US patent
4,641,349, Patent and Trademark Office, Washington, Contact Kirk L. Kroeker at [email protected].
D.C., 1987.
7. J.D. Daugman, “High-Confidence Visual Recognition of Contact editor Michael Potel at [email protected].
Persons by a Test of Statistical Independence,” IEEE Trans.
Pattern Analysis and Machine Intelligence, vol. 15, no. 11, For more articles on biometrics, see the July 2002 issue of
Nov. 1993, pp. 1148-1160. Computer.

Biometric Resources Online identification, fraud prevention, security, convenience,

For more information on biometrics, check out these customer service, and other applications.
online resources: ■ Biometric Technology Today (http://www.biometrics-
today.com): Biometric Technology Today is a monthly
■ Association for Biometrics, UK (http://www.afb.org. newsletter covering the international biometrics industry.
uk): The AfB is a nonprofit organization that aims to be an It contains news analysis, case studies, commentary, and
international forum for research and development, system regular monthly surveys.
design and integration, application development, market ■ Biometrics in Human Services User Group
development, and other issues surrounding biometrics. (http://www.dss.state.ct.us/digital/faq/dihsug.htm):
■ Automatic Identification Manufacturers Global The focus of BHSUG is to provide a platform for sharing
(http://www.aimglobal.org): The AIM Global network ideas and innovations, distributing findings, identifying
is a trade association for the Automatic Identification and best practices, and recommending and creating useful
Data Capture (AIDC) industry, representing those standards for human services users and technology
involved in technologies that include barcode, radio developers.
frequency identification, card technologies, biometrics, ■ Biometrics Institute (http://www.biometricsinstitute.
radio frequency data communications, and their org): The Biometrics Institute is an independent
associated industries. organization engaged in research, analysis, and education
■ BioAPI Consortium (http://www.bioapi.org): The for biometric users, vendors, and government agencies.
BioAPI Consortium was formed to develop a widely ■ International Biometric Society (http://www.tibs.org):
available and widely accepted application programming The International Biometric Society is an international
interface for various biometric technologies. society devoted to the mathematical and statistical aspects
■ Biometric Consortium (http://www.biometrics.org): of biometrics. Biologists, mathematicians, statisticians,
The Biometric Consortium serves as the US and others interested in its objectives are invited to
Government’s focal point for research, development, become members.
test, evaluation, and application of biometric-based ■ John Daugman (http://www.cl.cam.ac.uk/~jgd1000):
personal-identification technology. John Daugman’s personal Web page offers an excellent
■ Biometric Digest (http://www.biodigest.com): overview of the history and present use of iris-recognition
Biometric Digest is a guide to the companies and people technology, including hundreds of reference sources and
providing and using biometric technology for in-depth studies.

IEEE Computer Graphics and Applications 7