hand geometry is very stable over the course of one’s adult
life, it does not provide enough distinguishing power to be
used as the only means for identification. 7 Also, some facial
recognition systems can be fooled by an appropriately sized
photo of a legitimate user.
Behavioral biometrics measure user actions over time,
that is, for each action, there must be a beginning, an
end, and a duration. Consequently, behavioral biometrics indirectly measure characteristics of the human body.
Behavioral biometrics are learned and, therefore, can be also
re-learned. However, the consensus in the literature seems
to be that after reaching a certain age, changes in behavior
become more difficult to achieve, even with specific and
sustained effort. 11 Behavioral biometrics can therefore be
regarded as valid means of identification, even though they
are neither as unique nor as permanent as their physiological counterparts. In most cases, behavioral biometrics are
used to discern a user from a small(er) pool of candidates.
One advantage is that they are less invasive and therefore
more user-friendly. For example, a system that analyses keystroke timings or speech patterns can usually do so in the
background. In contrast, an iris or fingerprint scan requires
specific user actions.
2. 2. Biometric authentication versus identification
Authentication refers to identify confirmation or verification.
When a user claims a certain identity (e.g., by inserting a
card into an ATM or entering a user ID into a terminal and
then typing in a PIN or a password) authentication entails
deciding whether the claim is correct. The goal of the biometric classifier is to compare the current sample to the
known template for that user. The classifier returns the likelihood of a match. We refer to this as a 1 : 1 comparison.
Authentication differs from identification, where the
current sample comes from an unknown user, and the job of
the biometric classifier is to match it to a known sample.
We refer to this a 1 : n comparison. Identification is further
divided into two types: open-set and closed-set. We say that
an identification is closed-set, if it is known a priori that the
user is in the classifier database, that is, the classifier must
choose the best match from a pool of candidates. Otherwise,
identification is considered open-set.
2. 3. Design goals
When designing a new biometric system it is important to
take into account lessons learned from past and current systems. Design goals for biometric systems can be found in
the literature, for example, Jain et al. 4 Our goals include, but
are not limited to:
Universal. Must be universally applicable, to the extent
required by the application. It is important for the biometric
to apply to everyone who is intended to use the system.
Unique. Must be unique within the target population. For
example, measuring someone’s height would not work as
an identification mechanism on a large scale. At the same
time, (adult) height alone can usually identify individual
Permanent. Must remain consistent over the period of
use. Very few biometrics will stay constant over a lifetime, for
example, face geometry, voice, gait, and writing. However, as
long as the biometric is consistent over the lifetime of the
system, these biometrics work well.
Unobtrusive. If the user can be identified passively, with-
out interference, the biometric is much more likely to be
Difficult to circumvent. Ideally, a user should be unable to
change the biometric at all. At a minimum, a user must not
be able to modify his biometric to match that of another user.
3. PULSE-RESPONSE BIOMETRIC
The pulse-response biometric works by applying a low voltage pulse signal to the palm of one hand and measuring the
body’s response in the palm of the other hand. The signal
travels up through the user’s arm, across the torso, and
down the other arm. The biometric is captured by measuring the response in the user’s hand. This response is then
transformed to the frequency domain via the Fast Fourier
Transform (FFT). This transformation yields the individual
frequency components (bins) of the response signal, which
form raw data that is then fed to a classifier. Working in
the frequency domain eliminates any need for aligning the
pulses when they are measured.
The main reason for the ability of this biometric to distinguish between users is due to subtle differences in body
conductivity, at different frequencies, among different people.
When a signal pulse is applied to one palm and measured
in the other, the current travels through various types of
body tissues—blood vessels, muscle, fat tissue, cartilage,
and bones—to reach the other hand. Differences in bone
structure, muscle density, fat content, and layout (and size)
of blood vessels result in slight differences in the attenuation of the signal at different frequencies. These differences
show up as differences in the magnitude of the frequency
bins after the FFT. This is what facilitates distinguishing
Pulse-response is a physiological biometric since it measures body conductivity—a physiological characteristic distinct from behavioral aspects. However, it has an attractive
property normally associated with behavioral biometrics: it
can be captured in a completely passive fashion. Although
other physiological biometrics also have this feature, for
example, face recognition, pulse-response is not easily cir-cumventable. This combination of unobtrusiveness and difficulty to circument makes it a very attractive identification
mechanism. Essentially, it offers the best properties of both
physiological and behavioral biometrics.
4. LIVENESS AND REPLAY
A common problem with many biometric systems is liveness
detection, that is, determining whether the biometric sample represents a “live” user or a replay. For example, a fingerprint reader would want to detect whether the purported
user’s fingerprint was produced by a real finger attached to
a human, as opposed to a fingerprint mold made of putty
or even a severed finger. Similarly, a face recognition system would need to make sure that it is not being fooled by a
user’s photo or a 3D replica.
In traditional biometric systems, liveness is usually