|
The exponential
growth in the processing power of
computers and the increasing
sophistication of the software that runs
on them are fueling the development of
biometrics as a viable security
technology, but the origins of biometric
authentication actually lie in
antiquity. For example, we know that
merchants in the Nile valley were using
a form of biometric identification
thousands of years ago. Traders
routinely covered long distances and
were often known to each other only
through descriptions of physiological
characteristics (scars, eye color,
height). Roman legions are said to have
tattooed mercenary soldiers to identify
them more readily (and prevent them from
deserting).
The use of fingerprints for the positive
identification of criminals dates to the
turn of the 20th Century. Sir Edward
Richard Henry established Scotland
Yard’s Fingerprint Bureau, building on
the work of Dr. Henry Faulds, Sir
Francis Galton, and Czech physiologist
Johannes Evangelista Purkinje.
We are all aware of how DNA evidence has
revolutionized forensic criminology, and
how many innocent people have been freed
and guilty people caught because of
technological advances in this field.
However, using DNA for identification
purposes is currently neither quick -
nor inexpensive.
The challenge faced by biometric
technologies is to develop the means to
authenticate a large number of
individuals quickly, accurately, and
cost-effectively. Twenty years ago, this
challenge would have been the stuff of
science fiction. Today, we are
experiencing a biometric revolution, and
feats that seemed unachievable then are
becoming routine now. |
|
 |
|
Fig.1 Transillumination
images of a hand. |
| |
|
 |
|
Fig.2 Principle of
transillumination imaging. |
| |
|
 |
|
Fig.3 Transillumination
image of a finger. |
| |
|
 |
|
Fig.4 Flow of identification
process. |
| |
|
 |
|
Fig.5 Dependence of
rejection-error rate on
correlation threshold:(a)
overall characteristic, (b)
diminishing region. |
| |
|
 |
|
Fig.6 Dependance of
acceptance-error rate on
correlation threshold: (a)
overall characteristics, (b)
appearing region. |
|
Xin Wang, Kozo Sushita* and Koichi
Shimizu Graduate School of Engineering,
Hokkaido University, Sapporo, Japan. *
Bionics Co. Ltd., Osaka, Japan.
Using near-infrared light, we can obtain
transillumination image of blood vessels
of a hand or a finger. This image can be
used for personal identification.
Recently, a practical system has been
developed based on this method. This
system is relatively simple and the
identification time is reasonably short.
In order to examine the effectiveness of
the personal identification using the
transillumination image of blood
vessels, the fundamental characteristics
of the de-veloped system were analyzed.
The dependence of the rejection and
acceptance errors on the correlation
threshold was clarified. It was also
found that there is over 10% clearance
of cor-relation threshold between the
minimum threshold of a sufficiently low
acceptance-error and the maximum
threshold of a sufficiently low
rejection-error. Therefore, we can
expect practi-cally negligible rejection
and acceptance errors by setting the
correlation threshold between these two
values. We can also control the errors
by choosing an appropriate correlation
thresh-old.
Recently, the importance of the security
in various systems has been rapidly
increasing. The personal identification
is one of the key technologies to
support the security in computer
systems, in access-controlled areas,
etc. The term “biometrics” has been used
to refer the field of statistical or
mathematical data analysis in the
biological sciences. In these days, the
term has also been used to refer the
technology devoted to the indi-vidual
identification using biological traits.
The techniques using finger prints, iris
scanning, retinal scanning and facial
recognition are well known. We have
pointed out the feasibility and the
usefulness of the technique to use the
transillumination image of a hand for
the biometrics [Shimizu, 1992]. Using
near-infrared light (700-1200 nm
wavelength) we can obtain
transillumination image of blood vessels
of a hand or a finger.
The image is useful for noninvasive
imaging of physiological function, as
well [Shimizu, 1996], [Taka, 2000]. The
pattern of blood vessels is
individual-specific and does not change
in aging. The pattern is hardly
interfered by the dirt and scars on the
surface and is not easy to imitate. The
simplicity of the hardware and the speed
of recognition are the great advantages
of this technique over other existing
methods. Recently, a practical system
has been developed and used in various
fields. To examine the effectiveness of
the proposed method, we have analyzed
the identification rate in personal
identification using the developed
system.
Figure 1 shows the typical examples of
the transillumination image of a hand.
Differ-ent patterns are observed with
different individuals. Figure 2
illustrates the principle of the imaging
part of the developed personal
identification system. A finger is
illumi-nated with near-infrared light
(950 nm wavelength) from an array of
LED’s, and the transil-lumination image
is obtained by a CCD camera through an
opti-cal filter. An example of the
transillumination image of a middle
finger is shown in Fig.3. Fig.1
Transillumination images of a hand.
After image processing, the pat-tern of
the blood vessels is stored in digital
codes. First, the patterns of the
persons to be registered are stored in
the memory region of a computer system.
When a subject inserts his finger in the
imaging part, a transillumination image
is ob-tained, processed and compared
with the stored images, auto-matically.
According to the re-sult of the
comparison, the next action is activated
such as opening a door, starting a com-puter,
etc. The flow of the proc-ess is
summarized in Fig.4. Since the
comparison is based on a correlation
operation, the processing time is
reasonably short (typically a few
seconds).
Analysis of identification rate The
effectiveness of this method is largely
dependent on the fact that no one has a
common pattern of blood vessels in a
finger. We need to examine the
reasonability of this hypothesis within
a limit of spatial resolution of a
practical system. For the analysis, 300
tran-sillumination images were used. The
images of six different fingers (index,
middle, ring fingers in both hands) were
obtained in each of 50 subjects. The
subjects were 42 males (average age of
36.7)and 8 females (average age of
27.5). All the combinations (90000
cases) were tested and the
identification rates were analyzed. Two
parameters were used in the analysis.
One is the rejection-error rate, or the
rate to accept the other person who has
to be rejected. Another is the
acceptance-error rate, or the rate to
reject the person himself who has to be
accepted. Figures 5 and 6 show the
change of these rates with the change of
the correlation threshold. As the
threshold increases, the rejection-error
rate decreases and the acceptance-error
rate increases. In this figure, we can
see the degree of the dependence of each
error on the threshold. In practice we
have to make a compromise between these
two errors. It should be noted that
generally there is over 10% clearance of
correlation threshold between the
minimum threshold of a sufficiently low
acceptance-error and the maximum
threshold of a sufficiently low
rejection-error. Therefore, we can
expect practically negligible rejection
and acceptance errors by setting the
correlation threshold between these two
values. In some applications, rejection
error is more serious than the
acceptance error. In such a case, the
correlation threshold should be chosen
as a higher value, or closer to 1. In
this way, we can control the errors by
an appropriate choice of the threshold
values according to the requirements of
a specific application.
We have proposed a technique for the
personal identification using a
transillumination image of a hand or a
finger. A practical system has been
developed and the feasibility has been
verified. This system is relatively
simple and the identification time is
reasonably short. In order to examine
the effectiveness of the personal
identification using the
transillumination image of blood
vessels, the fundamental characteristics
of the developed system were analyzed.
The dependence of the rejection and
acceptance errors on the correlation
threshold was clarified. Based on the
results of this analysis, we can expect
reasonably low error rate for practical
use. We can also control the errors by
choosing appropriate correlation
thresholds. With further improvements in
the techniques of transillumination
imaging, image processing and the
correlation operation, higher
performance of this method is expected
in the near future.
Shimizu K. Optical trans-body imaging
- Feasibility of optical CT and
functional imaging of living body,
Medicina Philosophica, 11:620-629. 1992.
Shimizu K, Yamamoto K. Imaging
of physiological functions by laser
transillumination, OSA TOPS on
Advances Optical Imaging and Photom
Migration, 2:348-352, 1996.
Taka Y, Kato Y, Shimizu K.
Transillumination imaging of
physiological functions by NIR light,
World Congress on Medical Physics and
Biomedical Engineering, (CD-ROM)
4982-14105, 2000.
|
