Note : Les descriptions sont présentées dans la langue officielle dans laquelle elles ont été soumises.
CA 02412403 2002-11-20
SYSTEM FOR IDENTITY VERIFICATION
Field
The present invention relates to a method and
apparatus for verifying the identity of a person using a
biometric, such as a signature or fingerprint.
Background
In credit card transactions, a major security problem
exists whenever credit card information is transmitted over the
Internet or telephone lines. In addition, because of the
frequency with which credit cards, passports, and other personal
documents, are lost and stolen, there exists a need to
correctly, quickly and reliably verify the identity of the
bearers of these documents.
In a typical credit card transaction, as seen in Fig.
1, a merchant 10 transmits a credit card number, the expiry date
and a purchase order over the Internet or telephone lines 12 to
a verification agent 14. The agent 14 receiving this
information accesses the cardholder's credit information and
after comparing the latter to the purchase order amount, either
accepts or rejects the transaction_ If the transaction is
CA 02412403 2002-11-20
2
accepted, an approval code is transmitted back to the merchant
via telephone line 12. Even if the transaction is accepted
there is a risk that the card is stolen and is being used
fraudulently. Accordingly, there is a need to be able to
S quickly, accurately and securely identify the bearer of the
card.
Biometrics can be used to accurately verify identity,
however, biometric information sent over the Internet or
10 telephone lines can still be intercepted and subsequently
utilized for fraudulent transactions.
Various approaches have been developed to identify
persons by biometrics, including unique gestures such as
handwriting. Such speech and handwriting recognition systems
perform recognition of something that moves, leaving a
°'trajectory'f in space and time. Typical speech recognition
systems match transformed speech against a stored
representation. Most speech recognition systems use some form
of spectral representation, such as spectral templates or Hidden
Markov Models (HMMs).
CA 02412403 2002-11-20
3
Handwriting can be analyzed in real time or after it
has been formed. "Real time" or dynamic recognition systems
identify handwriting as a user writes, identifying such things
as number of strokes, the ordering of strokes and the direction
and velocity profile of each stroke. "Real time" systems are
also interactive, allowing users to correct recognition errors,
adapt to the system, or see the immediate results of an editing
command. Most on-line tablets capture writing as a sequence of
coordinate points.
Handwriting recognition is complicated in part,
because there are many different ways of generating the same
character. For example, the four lines of the letter E can be
drawn in any order. Handwriting tablets must also take into
account character blending and merging, which is similar to the
continuous speech problem. In other words, blending and
merging make it difficult for a recognition system to determine
where one character ends and the next one begins (or in the case
of speech recognition systems, where one word ends and the next
one begins). In addition, different characters can look quite
similar and are, therefore, difficult to distinguish. Thus,
prior to performing the character recognition, handwriting
tablets pre-process the characters. Preprocessing typically
CA 02412403 2002-11-20
4
involves properly spacing the characters and filtering out noise
from the tablet. The more complicated processing occurs during
actual character recognition.
Some character recognition processes, using binary
decision trees, prune possible characters by identifying
features. Normally simple features are identified first, such
as searching for the dots above the letters "i9' and ~tj°.
Features based on both static and dynamic features can be used
for character recognition. Other character recognition
processes involve the creation of zones, which define the
directions a pen point can travel (usually eight), and define
each character in terms of a set of zones. Look-up tables or
dictionaries can be used to classify or identify the characters
based on their features or sets of zones.
.Another character recognition scheme relies on signal
processing, in which curves from unknown forms are matched
against prototype characters. They are matched as functions of
time or as Fourier coefficients. To reduce errors, elastic
matching schemes (stretching and bending drawn curves) may be
used. However, these methods are computationally intensive and,
therefore, tend to be slow and expensive.
CA 02412403 2002-11-20
Most handwriting examination tablets reveal that
recognition of dynamic features of characters is possible, as in
speech. However, for the reasons discussed above, it is easier
to recognize isolated characters than strings of characters.
Most systems lag recognition by about a second, and recognition
rates are not very high. Reported rates of 95% are achieved
only for very carefully formed writing.
For each of the types of recognition systems discussed
above, a sample input (i.e. a voice or signature sample) must be
processed and compared with a stored reference gesture in order
to verify the identity of the subject. Normally, the reference
gestures are located on a remote server and accessed by
telephone lines or the Internet. The sample input must be sent
to the remote server where it is compared to the reference
gesture. Such a procedure is obviously exposed to the risk of
security breaches. Furthermore, there is a cost associated with
the maintenance of a remote server, and processing is delayed by
the need to access the server. Accordingly, it is an object of
the present invention to provide a quick and secure on-site
method of identification, which is accurate and cost effective.
CA 02412403 2002-11-20
6
SUN~?ARY OF THE INVENTION
According to the present invention there is provided a
method, and an apparatus for carrying out the method, for
S verifying a subject's identity using signatures or other
biometrics. The first step of the method comprises recording a
reference signature. The reference signature may be recorded
by, for example, measuring two-dimensional space coordinates x
and y exerted by a writer s writing instrument on a recording
medium.
An indicium is selected from the coardinates, which
identifies a specific portion (the reference biometric) of the
reference signature, having a selected characteristic, that will
be used for comparison with an unknown signature. The reference
biometric and the indicium are then placed on a portable,
readable substrate, such as the magnetic strip on a credit card.
The indicium of the reference signature is read from
the readable substrate and the coordinates of an unknown
signature are collected. The indicium is used to locate the
portion (the extracted biometric) of the unknown signature that
corresponds to the reference biometric. Once identified, the
CA 02412403 2002-11-20
7
extracted biometric is compared to the reference biometric to
determine if they match within predetermined threshold criteria.
If the reference and extracted biometrics match, the
~ identity of the provider of the unknown signature is positively
established as being the same as that of the provider of the
reference signature (or in other words, the bearer of the credit
card). If the reference and extracted biometrics do not match,
or if no portion of the unknown signature matches the
characteristics of the indicium, the identity of the provider of
the unknown signature is not verified.
The foregoing steps are done on-site, without the need
to access a server or to send information over telephone lines
or the Internet.
Advantageously, a second indicium may be stored on the
portable, readable substrate and used in the event that no
identifiable portion of the unknown signature corresponds to the
first indicium, or the results of the first comparison using the
first indicium indicate there is no match. In the preferred
embodiment the portable, readable substrate is in the form of a
magnetic strip, however, it will be appreciated by those skilled
CA 02412403 2002-11-20
g
in the art that it may take any of a number of alternative
forms.
The present invention additionally relates to an
apparatus for implementing the above method.
The coordinates of the reference and unknawn
signatures that are recorded and measured may additionally
include time, t, and force, z, among others.
It is obvious to anyone skilled in the art that the
present invention can be adapted to verify identity by applying
the method of the present invention to reference and unknown
samples of voice, fingerprints, or other biometrics.
BRIEF DESCRIPTION OF THE DRAWINGS
Further features and advantages will be apparent from
the following detailed description, given by way of example, of
a preferred embodiment taken in conjunction with the
accompanying drawings, wherein:
CA 02412403 2002-11-20
9
Fig. 1 is a schematic diagram of a typical credit card
transaction;
Fig. 2 is a schematic diagram of the identification
scheme;
Fig. 3 is a diagram of a handwritten letter 9~a~~
showing points A and B of zero velocity; and
14 Fig. 4 is a diagram of a handwritten letter ~~ae
showing entry vector C and exit vector D of a point of zero
velocity.
DETAILED DESCRIPTION WITH REFERENCE TO THE DRAWINGS
Prior to evaluating an unknown signature, a reference
signature must first be recorded and evaluated. The reference
signature may be evaluated based on both local features and
global features. Local features are those that occur within a
localized region of a signature, for example, local maxima and
minima, loops, points of intersection, points of zero velocity,
etc. Global features are those that occur throughout the
CA 02412403 2002-11-20
signature as a whole, for example, total signature time, average
velocity of signature, length-to width ratio, etc.
If we assign the values x, y, z and t such that x is
5 the horizontal displacement, y is the vertical displacement, z
is the pressure, and t is time, then individual points of a
signature can be represented by (x, y, z, t). One can normalize
the values of x, y, and z in order to compensate for variations
in signature sizes and recording device sizes.
Next, a selected biometric feature, in this case a
local maximum, and a series of points on either side of that
feature are recorded for comparison purposes. Also recorded is
an indicium, which identifies the location of the local maximum,
or other selected biometric feature. For example, the indicium
may be the number of local maxima or points of zero velocity
preceding the selected local maximum. The signatures of
different individuals vary greatly and, therefore, depending on
the nature of the reference signature some indicia may be more
reliable than others. Therefore, it may be advisable to first
test one indicium to see if it effectively identifies the
selected local maximum. If not, another indicium can be chosen.
CA 02412403 2002-11-20
1l
Signatures are not written in precisely the same way
every time. Therefore, a given indicium may not: correctly
identify the selected local maximum in every instance.
Accordingly, it may be advisable to use two or more indicia in
parallel or to use a back-up indicium that is used in the event
the first one fails.
A reference biometric, comprising the selected local
maximum, which is chosen from within the reference signature,
and coordinates x, y, z, and time, t, over a given range on
either side of the selected local maximum are encrypted and
recorded on a portable, readable substrate such as the magnetic
strip on the back of a credit or identity card. The indicium,
which will be used to locate the corresponding local maximum
within the unknown signature, is also recorded and encrypted on
the magnetic strip.
When the identity of an unknown user is being
verified, the card is swiped through a card reader and the
indicium and reference biometric are extracted and stored
locally in memory. Next the user signs his name (the unknown
signature) on a touchpad, which records the coordinates of the
unknown signature so they can be stored locally. Suitable
CA 02412403 2002-11-20
12
touchpads have been developed by DSI Datotech Systems Inc. of
Vancouver, British Columbia. The unknown signature is first
normalized to correspond with the scale of the reference
signature. By reference to the indicium, the extracted
biometric is identified and extracted from within the unknown
signature. The extracted biometric comprises a range of values
of x, y, z, and t, corresponding to the reference biometric, and
falling within a range determined by the indicium. The
coordinates of the extracted biometric from the unknown
signature are compared with those of the reference biometric.
If the difference between the y values of the extracted
biometric and the y values of the reference biometric are within
a threshold value, then the x, z and t values will also be
compared to determine if they fall within predetermined
thresholds. If the x, y, z and t values all fall within the
allowable thresholds, the extracted biometric and the reference
biometric, and therefore the unknown signature and the reference
signature, are matched. However, if the x, y, z and/or t values
do not fall within the allowable thresholds, then there is no
match. In such instances a new indicium and/or biometric
feature may be selected and the process repeated.
Alternatively, for increased reliability, comparison of a global
biometric, such as velocity, may also be made.
CA 02412403 2002-11-20
13
Referring to Fig. 2, the identification verification
system of the present invention consists of a credit card or
identity card 26, on the back of which is a magnetic strip 28
containing a reference biometric and indicium. The credit card
26 is swiped through a first input device, in this case a credit
card swiper unit 30.
Gesture input device 20, which can be a touch pad, receives
the unknown signature and extracts position, velocity,
acceleration and force information from the unknown signature.
The gesture input device 20 and the card swiper unit 30 are
connected to the biometric extractor 22. Rather than having to
store the large amount of information that would be represented
by the average signature, the unknown signature is analyzed and
only a small portion, the extracted biometric, (which is
identified by the indicium, received from the card swiper 30) is
extracted by the biometric extractor 22.
Although not shown in Figure 2, the card swiper unit
may also be coupled to the biometric comparator 24 so that
the reference biometric may be sent directly rather than passing
through the biometric extractor 22.
CA 02412403 2002-11-20
14
The extracted biometric of the unknown signature is
transmitted to the biometric comparator 24, which also receives
the reference biometric that is stored on a magnetic strip 28 on
the back of a credit card or identity card 26, which has been
swiped through a credit card swiper unit 30. Biometric
comparator 24 compares the extracted biometric with the
reference biometric from the card 26. If the comparison by the
biometric comparator 24 results in a match, then the person
providing the unknown signature is the same person that provided
the reference biometric. The accuracy of the technique is not
100 o so it may be prudent to use one or more additional
biometrics or portions of a signature for comparison in parallel
to determine, with an adequate level of confidence, whether
there is a match. Alternatively, the identity verification
procedure can be repeated.
The biometric extractor 22 and biometric comparator 24
may be incorporated into a CPU (not shown) and the results
displayed on a monitor (not shown).
Any one of several conventional statistical analyses
can be used determine whether there is a match between the
CA 02412403 2002-11-20
extracted biometric and the reference biometric, such as a
calculation of the average of the square of the differences
between the coordinates of the extracted biometric and the
reference biometric.
5
An alternative method of comparing reference and
extracted biometrics comprises a vector analysis surrounding
points of zero velocity. In a typical signature, there are
likely a plurality of points where the velocity of the pen is
10 zero. For example, referring to Figure 3, points A and B of the
letter "a" will be points of zero velocity. Referring to Figure
4, point A will have two position vectors surrounding that
point, a vector C entering the point A and a vector D exiting
from the point A. Therefore, using these three pieces of data,
15 a given point where the pen velocity is zero will have (x1, yl,
z1) indicating the point of zero velocity, (x2, y2, z2) indicating
the entry vector, and (x3, y3, z3} indicating the exit vector.
Therefore, a given point of zero velocity, identified by indicia
as discussed above, can be used as a reference biometric to
verify the identity of the person providing an unknown
signature, by comparing the point of zero velocity, and the
associated entry and exit vectors with the corresponding point
of zero velocity and vectors of a reference signature.
CA 02412403 2002-11-20
16
Accordingly, while this invention has been described
with reference to illustrative embodiments, this description is
not intended to be construed in a limiting sense. Various
modifications of the illustrative embodiments, as well as other
embodiments of the invention, will be apparent to persons
skilled in the art upon reference to this description. It is
therefore contemplated that the appended claims will cover any
such modifications or embodiments as fall within the true scope
of the invention.