close

Вход

Забыли?

вход по аккаунту

?

JP2001318691

код для вставкиСкачать
Patent Translate
Powered by EPO and Google
Notice
This translation is machine-generated. It cannot be guaranteed that it is intelligible, accurate,
complete, reliable or fit for specific purposes. Critical decisions, such as commercially relevant or
financial decisions, should not be based on machine-translation output.
DESCRIPTION JP2001318691
[0001]
BACKGROUND OF THE INVENTION 1. Field of the Invention The present invention relates to a
personal identification device, and more particularly to a personal identification device which
constitutes a part of a security system, and more particularly to a personal identification device
which utilizes individual differences of living bodies as identification information.
[0002]
2. Description of the Related Art Personal identification devices used for entry / exit management
in a highly classified room, access control of a computer, etc. include fingerprints, iris, retinal
blood vessels, handprints, face, etc. There are known methods of utilizing such characteristics as
signature.
[0003]
Among them, the identification method using the fingerprint pattern disclosed in Japanese Patent
Application Laid-Open Nos. 6-28457, 6-28458, 6-28459, etc. It is known as a method that can
realize a high identification rate compared to other methods by using it.
[0004]
However, although the method of using a two-dimensional pattern in which a fingerprint is
optically detected is invariant to life and death, it is relatively easy to collect another person's
fingerprint pattern and copy the wrinkle pattern, and how high the discrimination is Once the
rate is realized, there is a problem that it is vulnerable to forgery using duplication.
08-05-2019
1
[0005]
In addition, it is necessary to shake hands with a prism or the like for collecting a fingerprint
pattern, and there is a problem in terms of hygiene for use by an unspecified number of people.
[0006]
The method of detecting a fingerprint by ultrasonic waves detects a two-dimensional pattern with
a sound wave whose propagation velocity is slower than that of light waves and whose
wavelength is longer than that of light waves, so the device becomes simple and its operation is
stable. There is a problem similar to the method of detecting a dimensional pattern.
[0007]
In addition, because there is an image of criminal investigation in any of the methods that use
two-dimensional patterns of fingerprints, users have high psychological resistance, and it is
difficult for the problem to spread before the technology of dispelling images before the spread
of the image. I have it.
[0008]
In order to solve this problem, even if using a three-dimensional pattern in which the shape
information of ridges that are ridges and depressions that are non-contacts and fingerprints is
added, one-dimensional addition is made to appearance information when it is clogged It is only
an effect that the forgery is somewhat complicated, the detection device is complicated, the data
processing amount is increased, the device cost and the processing time are increased, and the
problem is solved fundamentally. Can not be.
That is, there is a problem that even if it is a planar two-dimensional pattern or a threedimensional three-dimensional pattern, as long as appearance information is used, it is weak to
forgery.
[0009]
The above-mentioned conventional methods other than the method of utilizing the appearance of
these fingerprints have the following problems.
08-05-2019
2
[0010]
That is, in the method of using the bill, there is a problem that the risk of being copied is large
because the appearance is used as well as the two-dimensional pattern of the fingerprint.
[0011]
Since the method of using the iris reads the iris pattern by the camera, it is possible to sniff and
record the video signal from the camera, and inject the recorded video signal of the spear into
the video cable of the camera at the time of authentication. There is a problem that the risk of
being forged is large.
[0012]
In the method of using the shape of the face, it is necessary to keep the posture and lighting
conditions within a certain range, there is a problem that the stability is lacking and there is a
possibility that the forgery may be made to use the appearance.
[0013]
Voiceprints and signatures are likely to be stolen, and they are susceptible to the user's mental
state and health condition and lack stability.
[0014]
The method of utilizing the blood vessel pattern of the hand and the finger is certainly more
difficult in counterfeiting than the method of utilizing the appearance feature, and the
psychological burden on the user is smaller than the method of utilizing the retinal blood vessel
pattern. Since light and the like are used to use information relatively near the surface such as a
subcutaneous vein, there is a problem that the possibility of stealing a pattern by a near infrared
camera or the like remains after all.
In addition, it is difficult to obtain completely the same pattern every time because the blood
vessels of the hand and the finger are deformed due to the movement of the hand and the finger.
Therefore, there is a problem that the tolerance range at the time of collation has to be widely
adopted, and the false authentication rate becomes high.
08-05-2019
3
[0015]
SUMMARY OF THE INVENTION The present invention has been made in view of the above
circumstances, and it is extremely difficult to forge, there is no psychological resistance of the
user, and there are few physical restraint conditions at the time of use. An object of the present
invention is to provide an individual identification apparatus using biological information which
is hardly present and which is not easily influenced by the psychological condition or the health
condition.
[0016]
In order to achieve the above object, the personal identification apparatus of the invention
according to the first aspect comprises storage means for storing reference characteristic
information based on a buzzing sound of a plurality of objects; Detection means for detecting the
buzzing sound of the objects, feature extraction means for extracting the feature information
based on the buzzing sounds of the plurality of objects detected by the detecting means, and the
feature information extracted by the feature extracting means And collating means for collating
with the reference characteristic information stored in the storage means.
[0017]
According to the present invention, in order to identify an individual, a buzzing sound of a
plurality of objects is detected by detection means such as a microphone or a piezoelectric
element.
Preferably, at least one of the plurality of objects has a physical surface shape capable of
identifying an individual, such as a human hand or a toe.
In this case the physical surface shape is a fingerprint.
Therefore, by rubbing the finger and the finger, or the finger and the object, it is possible to
generate a rubbing sound unique to the person.
[0018]
08-05-2019
4
Then, the feature extraction means extracts feature information based on this accompaniment
sound.
A filtering means is further provided for filtering out components below the frequency of the
audio frequency band from the accompaniment sound detected by the detection means to extract
a signal of an ultrasonic component, and extracting feature information based on the signal of
the ultrasonic component You may do it.
At this time, the filtering means may change the filtering frequency based on a power spectrum
of a frequency lower than the audible frequency band of the mixing sound.
Furthermore, the filtering means may perform normalization processing to make the maximum
value of the crest value of the signal waveform constant.
[0019]
The collation means collates the feature information extracted by the feature extraction means
with the reference feature information stored in the storage means, and determines whether or
not both of them match.
Note that this collation result may be displayed on a display means, for example.
Further, when this personal identification apparatus is used for, for example, permission for login
of a computer, login may be permitted if both match according to the comparison result.
[0020]
As described above, since personal identification is performed using ultrasonic waves that are
generated when the fingerprint shape is in contact with another object, not the appearance
information of the fingerprint, the reliability of personal identification can be improved.
08-05-2019
5
[0021]
BEST MODE FOR CARRYING OUT THE INVENTION Embodiments of the present invention will be
described below.
[0022]
FIG. 1 shows a personal identification device 10 according to the present invention.
In this embodiment, the case where the personal identification device 10 according to the
present invention is applied to login to a remote host computer 20 will be described.
[0023]
As shown in FIG. 1, the personal identification apparatus 10 includes a microphone 12, a voltage
amplification unit 14, a filtering unit 16, a feature extraction unit 18, a determination unit 20,
and a feature recording unit 22.
The feature recording unit 22 is connected to the feature extraction unit 18, the determination
unit 20, and the computer terminal 24.
The computer terminal 24 is connected to the host computer 26 via the network 28.
[0024]
A user using a remote host computer 26 connected from the computer terminal 24 via the
network 28 inputs the ID of the user at the computer terminal 24 and also inputs a password as
necessary.
This ID is sent to the feature recording unit 22. In the feature recording unit 22, a unique feature
value corresponding to the ID of the user is recorded in advance.
08-05-2019
6
[0025]
Next, in order to authenticate the identity of the user, the fingerprint of at least one finger of that
hand or the fingerprint of at least one finger of that person's foot and the other hand in front of
the microphone 12 Alternatively, finger rubs (frictions generated by rubs between a finger and a
finger or an object) are generated by rubbing a fingerprint of a toe or a rubable object other than
a fingerprint.
[0026]
For example, as shown in FIG. 2, each finger of the index finger 30 and thumb 32 of a dominant
hand or a non-dominant hand, that is, a portion with a fingerprint facing each other, moves both
fingers by about 1 cm to 2 cm and rubs against each other. Alternatively, the index finger 30 is
touched on the housing or key top of the keyboard provided on the computer terminal 24 and
the finger is moved about 1 cm to 2 cm and rubbed to generate a finger rub.
[0027]
Here, the reason for setting the moving distance by rubbing to 1 cm to 2 cm is because it is a
desirable value for the following reason.
First, in the present invention, an individual is identified by the ultrasonic wave component of
finger rubbing sound emitted when a fingerprint is rubbed. Assuming that the central frequency
of the ultrasonic wave component is 50 kHz, the time of one wavelength is 0.02 ms. Become.
[0028]
Although it is desirable to obtain data of 100 wavelengths or more (2 ms or more) in order to
efficiently calculate the feature quantities of ultrasonic components necessary for identification,
it is preferable to use finger and finger The relative velocity may extend to several meters per
second, and for example, the lower limit of the movement distance necessary for taking 100
waves or more as 5 m per second is 1 cm.
[0029]
The upper limit value of 2 cm is derived from the belly length of the first joint where the
fingerprint of the finger of a common adult hand is present, and there is a rational basis, but if it
is gently rubbed, it is necessary. The moving distance may be shorter, and the moving path at the
08-05-2019
7
time of rubbing does not have to be a straight line, so 1 cm to 2 cm is merely a guide, and it is
not a necessary condition or a necessary condition.
[0030]
Calculation of the combination of the fingers to be rubbed when rubbing finger-to-finger belly-toface to generate finger rubs, rubbing against the thumb of the right hand the fingers other than
the thumb of the right hand and all fingers of the left hand For the fingers other than the thumb
of the right hand, only the thumb that has already been counted for the right hand can be rubbed
with all the fingers of the left hand. For one finger other than the thumb of the right hand There
are five combinations that can be rubbed, except for the overlap, and there are twenty
combinations for all fingers except the thumb of the right hand, so there are twenty-nine
combinations for the right hand.
In the same calculation, there are also 33 ways in the left hand, but when calculating the number
of combinations of the right hand, the numbers counted in combination with the finger of the left
hand overlap, so there are 33 ways in total after combining the right hand and left hand.
[0031]
However, this calculation does not take into consideration the case where there is no finger
under certain circumstances or the case where it is possible to rub fingers that do not include the
thumb only with the right or left hand by having special skills, There is no loss of generality.
When using the toes, the number of combinations is further reduced since the toes can not rub
the belly of the same toes of the left and right unless they have special skills.
[0032]
When the right and left toes are rubbed against each other, there are 25 ways in total, since the
five toes of the different toes are combined with each other on one of the left and right toes.
08-05-2019
8
When combining hands and feet, there are four ways with just the right finger for the right hand,
25 ways with the right finger and the right toe, 54 ways with the right finger and the left toe, and
54 for the left hand as well. There are 108 after all in the street. That is, in the combination of
both hands and both legs, there are 158 (33 + 25 + 108-8) ways except for the overlap in the
above combination. Even if one of the fingers to be rubbed is one, there are as many
combinations as this.
[0033]
Furthermore, in the case where the partner of rubbing is not a finger but an object that can be
rubbed, if there is a plurality of one side of the finger to be rubbed, there are cases where
rubbing can be done simultaneously with these combinations, etc. The user can be registered in
any one or a plurality of combinations of 158 or more combinations, which makes it more
difficult for the imitator to imitate.
[0034]
When the object to be abraded is not a finger, if the object is too smooth it will not generate
finger rubs or it will not be suitable for identification if it is too rough, if it is too rough or sticky
it will not move or it will move Even if it becomes difficult or finger rubs are not generated or
generated, it is not suitable for identification, but almost all solid substances that are familiar in
the surrounding environment of general life have such characteristics. Because it does not have,
it is suitable for the object.
For this reason, it is not necessary to prepare a special object, and as a result, it is not possible to
know what it is necessary for the imitator to imitate by concealing what it is intended to, so what
to imitate It does not know whether it should be stolen or used, and it has the advantage that the
imitators have the advantage of making imitation more difficult and reducing their willingness to
imitate.
[0035]
For example, in the case of an object such as an identification card that is supposed to be
limitedly owned by an individual, the target to be stolen becomes clear, but this invention does
not have such a thing. The object may be provided near the installation location of the computer
terminal, but it may be carried around. When carrying around, it has the same drawbacks as the
08-05-2019
9
identification card described above, but it is not a fixed form like an identification card, but it is
necessary to find out what is the object to be carried. It has to be done, so there is less security
like a proof of identity card. When carrying around, the object may be carried as it is, but it is
more desirable to carve in a physical surface shape such as a fingerprint which can identify the
possessed individual.
[0036]
There is no need for this physical surface shape to be correlated with the fingerprint at all, and
the probability that a plurality of persons can have the same or similar ones is sufficiently lower
than the required discrimination resolution and from the surface shape of the individual It may
be anything as long as it can be identified, but creation of a pattern becomes easier if it is similar
to a fingerprint which is said to be unequal and immutable to life.
[0037]
An example is shown in FIG. 3, in which a key ring 34 is used as a finger-sliding object and a
pattern 36 such as a fingerprint is engraved.
The reason why it is desirable to engrave a physical surface shape that can identify an individual
in this way is that it is a substitute unless the same object is obtained, even if the object of finger
rubbing is limited. It is difficult to forge the
[0038]
However, if you carry such an object, it might be better to target your finger, but the advantage
of using an object that is not a finger is that your finger may It is not always the case that it is
only when the object can not be made. There is no doubt that it is not easy to sniff and copy the
ultrasonic component of finger rubs that occur as a result of finger-to-finger rubbing, but if it is
to be sniffed and replicated, then the fingerprint is considered lifelong From what is said, anyone
can easily spoof by using the copy.
[0039]
08-05-2019
10
Of course, as described above, the risk of being falsified significantly is reduced, but not
absolutely, as compared with the method using a fingerprint image. When the finger rubs by
fingers and fingers are eavesdropped and replicated, the target used until then when it was
judged that the information was copied or the probability of being copied was high, even if it was
judged that the information was reproduced. Spoofing can be prevented in advance by replacing
it with another object that generates an ultrasonic wave different from the ultrasonic wave
generated when rubbing with an object.
[0040]
If the object to be rubbed out is replaced with another one, the copied one can not be used as a
fake, and the finger rubs generated by the new object and finger rubbing must be intercepted
and copied. That is, in the present invention, one of the major features is that it is also possible to
reduce the risk of a system that authenticates only by physical characteristics or physical
characteristics that are said to be lifelong. In the case where the user does not carry the object,
the pattern is of a type that is sufficiently specific to the operation of the computer terminal 24
to be logged in and operated, or the like so as to sufficiently reduce the probability of being
installed. If it is only one, it will not be possible to practice imitation in advance, since it can not
know the pattern unless it visits the place where the computer terminal 24 to be used is installed,
and security can be further strengthened. .
[0041]
An example of such is shown in FIG. In FIG. 4, a finger rubbing target object 40 is prepared for
the keyboard 38, and a finger rubbing sound is generated with the index finger 42. In this case, if
the finger rub object is also used as the pointing pad, the space for the generation of the finger
rub noise is also unnecessary, which is advantageous. Further, if a piezoelectric element (not
shown) is attached to the lower surface of the object to be rubbed instead of the microphone 12,
the size can be further reduced. Although a key top is also effective as a finger rubbing object, it
is necessary to use a member having wear resistance and to carry out a surface treatment since a
normal key top wears and becomes flat with use.
[0042]
By the way, finger rubs generated by such rubbing are subjected to sound wave / voltage
08-05-2019
11
conversion by the microphone 12 and passed through the voltage amplification unit 14 to the
filtering unit 16 which filters frequencies lower than the audio frequency band and the audio
frequency band. In the practice of the present invention, a microphone having an appropriate
sensitivity to ultrasonic waves in a frequency band higher than the audio frequency band is used,
but whether the ordinary microphone has no sensitivity outside the audio frequency band or low
sensitivity outside the audio frequency band It is one of the notable advantages of the present
invention that it has any advantage that eavesdropping can not be made easy.
[0043]
The reason for using the ultrasonic component of finger rubs is that it is firstly robust to noise in
the audio frequency band other than finger rubs, and secondly it is for personal identification
performed by voiceprints in quiet places such as offices etc. One of the biggest drawbacks is that
people around us aren't bothered by the noise in the audio frequency band, and thirdly it's
attenuated as the distance from the sound source is different compared to the sound wave of the
audio frequency components and components lower than the audio frequency components. It is
more difficult to eavesdropper because of its large amount and low diffraction, and fourth, it is
difficult to record and analyze the characteristics of finger rubs emitted by specific individuals by
eavesdropping because it is not audible to the human ear, After recording and analyzing the
characteristics of the finger rubs emitted by the individual with the longitudinal orientation, it is
difficult to reproduce the ultrasonic wave having the characteristics of the hemorrhoid.
[0044]
The upper limit of the frequency characteristics of the microphone may be as high as 100 kHz at
most in consideration of the power spectrum of the finger rub and the availability and price of
the microphone, and the practicability will not be lost even at several tens of kHz, but will be
described below In order to be able to reliably separate the ultrasonic components of the audible
sound and the finger rub, even if the cutoff characteristic of the filtering section 16 is not
sufficiently steep, the upper limit of the audio frequency band which usually decreases with
aging at 15 kHz to 20 kHz, ie, 30 kHz It is desirable that the frequency is 40 kHz or more.
[0045]
In the filtering unit 16, the audio frequency band and sounds lower than the audio frequency
band are filtered.
This is to increase the signal-to-noise ratio of the ultrasonic component of finger rubs to be
08-05-2019
12
identified and to perform more reliable identification.
This filter section 16 uses a passive analog filter combining at least two parts of a resistor, a
capacitor, and a coil which is an electronic passive element, or an electronic active component on
the at least two electronic passive elements. Using an active analog filter, mechanically using a
quartz oscillator or ceramic element, etc. numerically using digital processing such as discrete
Fourier transform or Z transform, or at least You may combine two or more.
[0046]
The cutoff frequency of the signal to be filtered by the filtering unit 16 does not have to be fixed,
and a time zone with a small amplitude of amplitude generated in time series in the forward and
backward paths of finger rubbing is determined as ambient noise other than finger rubbing. If
the ambient noise energy is automatically detected by measuring its power spectrum, for
example, if the ambient noise energy extends to a higher frequency band, the cutoff frequency
may be changed adaptively. It is possible to more effectively improve the signal to noise ratio.
[0047]
Before or after filtering or simultaneously with filtering, it is desirable to perform normalization
processing to make the maximum value of the peak value of the waveform constant.
The reason why normalization processing is not performed before the processing in the filtering
unit 16 is that filtering in the filtering unit 16 eliminates signals in frequency bands unnecessary
for the processing after the feature extraction unit 18, so the filtering unit 16 and before are
processed. If the normalization process is performed, the dynamic range is narrowed.
[0048]
The feature extraction unit 18 extracts the feature of the finger rub signal whose frequency is
lower than the audible frequency band or the audible frequency band from the filtering unit 16.
The feature extraction unit 18 analyzes the signal of the finger rubbing sound on either one or
both of the time axis and the frequency axis to extract a feature. As features to be extracted and
analyzed, there are so-called peak-to-peak maximum minimum width, root mean square value,
probability density function, autocorrelation function, spectrum, and the like. Generally, even if
08-05-2019
13
you try to mimic the characteristics of the finger rubs of others, the number of samples of finger
rubs obtained by wiretapping is small, and therefore often there is not enough sampling period
to imitate, as a result In particular, the autocorrelation function is effective and suitable for
picking out the forger and is preferably equipped, as the signal tends to be periodic.
[0049]
In addition, if the system configuration is such that it is entirely up to the person to be identified
that the timing of inputting the finger rub, that is, when the identification subject generates
finger rub and how many times the finger rub is to be generated. Generally, since the generation
of finger rubs is single-shot, feature value extraction by sonagram or wavelet transform is
preferable. In the case where the system side guides or forces the timing for inputting the finger
rub, the periodic analysis method is preferable by inputting the finger rub with a certain period
and a certain number of times. Here, the meaning of periodicity is different from the meaning of
becoming periodical when trying to imitate as described above. The former is intended to imitate
to detect that the disorder of the signal component is reduced, while the latter facilitates
separation of the signal component and noise.
[0050]
The features extracted by the feature extraction unit 18 are sent to the determination unit 20
and the feature recording unit 22. The feature recording unit 22 records the feature obtained
prior to the identification of the individual in a temporary storage location. If the feature quantity
recorded in this temporary storage location is data of the feature quantity of the registered
individual according to an instruction from the computer terminal 24 later receiving the result of
the determination unit 20, the registered individual identification It is used to learn and record
feature quantity data extracted for If it is data of the unregistered individual feature amount, it is
used to newly record.
[0051]
The discrimination unit 20 correlates the feature amount of the individual to be identified
extracted by the feature extraction unit 18 with the feature amount stored in the feature
recording unit 22, and is a registered individual or is not registered. It is output as a
determination result to the registered individual how much it matches with the characteristics of
08-05-2019
14
the individual. For example, assuming that three persons of A, B and C are registered, the
matching degree between the person to be identified and the registered individual is 0.85 for A,
0.81 for B, and C for Is output as 0.42. In this example, although the matching degree of A is the
highest, the determination unit 20 alone may determine whether to consider it as A or not to be
registered based on this numerical value, but other authentication means, external conditions,
etc. It is better if the information can be used, so the computer terminal 24 receiving the output
of the determination unit 20 or the host computer 26 connected with the computer terminal 24
via the network 28 may be used.
[0052]
Further, in the above example, the difference between the matching levels of A and B is 0.04, but
the judgment unit 20 may also judge whether the numerical value indicating the difference is
regarded as large or small. The computer terminal 24 or the host computer 26 to which the
computer terminal 24 is connected may be used. That is, the determination unit 20 may
determine the identification by itself, or may only provide information necessary for the
determination of the identification.
[0053]
As a method of taking correlation in the discrimination unit 20, there are a shortest distance
method and a similarity method as a pattern matching method, a Bayesian discrimination method
and a maximum likelihood method as a statistical determination method, and a neural method as
an autonomous method. Although there is a network etc., for example, the distance between the
microphone 12 and the fingertip may be different between registration and collation, and the
feature quantity extracted by the feature extraction unit 18 and the feature quantity recorded by
the feature recording unit 22 Is generally heterogeneous covariance matrix, so in the statistical
discrimination method, the Mahalanobis distance is adopted rather than adopting the Euclidean
distance between the feature of the person to be identified and the registered feature as the
explanatory variable of the discriminant function. Is preferred.
[0054]
When a neural network is used as an autonomous method, it is possible to literally learn and
construct a network configuration that provides a more optimal discrimination rate by
feedbacking correctness / incorrectness of discrimination results, but learning The higher the
identification rate, the lower the generalization performance.
08-05-2019
15
In the first place, the fact that neural networks are represented in a distributed manner in the
input / output relationship of information in the first place, that is, the fact that the entropy on
the network is increasing makes tuning of the network configuration more difficult. It is often
useful to make the representation of the network regular by the method and localize the
information representation to some extent.
[0055]
As one of such methods, it is preferable to configure a fuzzy neural network that facilitates
expressing the input / output relationship of the network in the form of if then rules. In a fuzzy
neural network, it is known that relatively small-scale configurations with several inputs and one
output are reasonably well learned and function. However, it is a fact that learning with multiple
inputs and multiple outputs takes much time to be impractical in practice, and a desirable input /
output relationship can not be obtained, but the method disclosed in JP-A-8-190483 is used. It
can be used.
[0056]
Although the entire configuration has been described above, this configuration is similarly
applied to mobile information devices such as mobile phones, hand-held computers, palm
computers, etc. which are likely to be abused by loss or theft in addition to computer login. it can.
[0057]
Next, an embodiment of the present invention will be described based on actually obtained data.
First, finger rubbing of the index finger and thumb of the dominant hand is repeated for 30
seconds in three different persons (person A, person B and person C) to generate finger rubbing
sound, and this finger rubbing sound is collected by the microphone 12 and voltage is generated.
A portion of the waveform amplified by the amplification unit 14 is shown in FIG. 5 to FIG. As
you can see from the figure, you can see that the individual's characteristics are exposed. In FIG.
5 to FIG. 7, a waveform of finger rubbing for two reciprocations is shown.
08-05-2019
16
[0058]
Next, in the filtering unit 16, normalized waveforms are shown in FIGS. 8 to 10 through
numerical filters in which the frequency below 40 kHz is -100 dB and the frequency above 40
kHz is 0 dB. The cutoff characteristics of the filter unit 16 such as cutoff frequency and cutoff
slope characteristics do not have to be as steep as shown in this embodiment, and may be passive
or active or analog filters based on a combination of these.
[0059]
However, even in this case, it is desirable that the cutoff frequency be 30 kHz to 40 kHz or more
as described above. The cutoff characteristic is preferably -6 dB / oct or more because it is
necessary to sufficiently reduce ambient noise and electromagnetic induction noise of a
commercial power source, but the phase characteristic and flatness of the cutoff are not so
serious problems, so it is necessary to examine it so much Absent.
[0060]
8 to 10 respectively correspond to FIG. 5 to FIG. 7 and “wave form of one reciprocation of
finger rubbing” “another waveform of one reciprocation of finger rub” shown in FIG. 5 to FIG.
The description of the forward path waveform "waveform return path waveform" can be used as
a description of the same place in Figs. 8-10, respectively.
[0061]
The audio frequency band filtered by the filtering unit 16 and the frequency band filtered lower
than the audio frequency are transmitted to the feature extraction unit 18.
In the feature extraction unit 18, after full-wave rectification, an envelope is obtained through a
numerical filter that sets 0 dB below 200 Hz and −100 dB or more above 200 Hz. Waveforms
obtained in this manner are shown in FIGS. In FIG. 11, it can be understood that the wave height
value of the return path is high and the rise and fall are steep if the return path is compared with
the return path. It can be seen in FIG. 12 that the forward path waveform is not unimodal but has
one or more local peaks. It can be seen in FIG. 13 that the forward and reverse paths are
08-05-2019
17
relatively smooth in rising and falling as compared with FIGS. 11 and 12.
[0062]
Generally, the forward and return paths are in a relative relationship and both can be said to be
commutative unless at least one definition of a starting point or an end point is given. The
forward path and the backward path described here can be identified from information other
than the information collected by the microphone 12 and usually can not be distinguished.
Therefore, it is necessary to select a round-trip invariant feature in the calculation of the feature
amount. Taking this into consideration, one of the features to be extracted is the ratio of the peak
value of the current mountain to the peak value of the immediately preceding mountain, ie, the
peak value of the immediately preceding mountain, by scanning along the time axis of the
waveform. The peak of the mountain at present. One of the other features is to count the peaks
that include the peaks present in one mountain. However, in order to exclude the peaks due to
disturbances such as noise from the counting object, counting is not performed if the ratio of the
height of the peak value to the valley immediately after the peak (valley height ÷ peak height) is
0.9 or more . Further, the steepness of the mountain slope is obtained as another feature value.
This will be described with reference to FIG. The steepness (s) of the mountain 100 in FIG. 14 is
obtained by obtaining the half value (h) 102 of the peak value (p) 101 of the mountain, obtaining
the time width (h) 103 of this half value, and calculating with S = p / h Do. The feature values
described above are summarized as follows.
[0063]
The discriminating unit 20 discriminates an individual by comparing the feature extracted by the
feature extracting unit 18 with the unique feature value prepared by the input of the ID. Next, an
example in which the determination unit 20 is configured by fuzzy inference having three inputs
and three outputs will be described. The fuzzy inference uses mammdani's inference method and
has the following rules.
[0064]
Rule 1: If the ratio between the current peak of the mountain and the peak of the previous
mountain is large, the person A is likely to be.
[0065]
08-05-2019
18
Rule 2: If the ratio between the current peak of the mountain and the peak of the previous
mountain is small, the person A is likely to be.
[0066]
Rule 3: If the ratio between the current peak of the mountain and the peak of the previous
mountain is medium, the person B is likely to be.
[0067]
Rule 4: If the ratio between the current peak value of the mountain and the peak value of the
previous mountain is medium, the person C is likely to be.
[0068]
Rule 5: If the ratio of the peak value of the current mountain to the peak value of the previous
mountain is large, the person A is likely to be.
[0069]
Rule 6: If the variation in the ratio between the current peak of the mountain and the peak of the
previous mountain is small, the person B is likely to be.
[0070]
Rule 7: If the variation of the ratio between the current peak of the mountain and the peak of the
previous peak is small, the person C is likely to be.
[0071]
Rule 8: If there are a large number of peaks including the peaks present in one mountain, person
B is likely to be.
[0072]
Rule 9: If the number of peaks including the peaks present in one mountain is medium, person A
is likely.
[0073]
Rule 10: If there are a small number of peaks including the peaks present in one mountain, it is
08-05-2019
19
likely that the person is C.
[0074]
Rule 11: If the steepness of the slope of the mountain is large, it is likely that the person is
person B.
[0075]
Rule 12: If the steepness of the slope of the mountain is medium, it is likely to be person A.
[0076]
Rule 13: If the steepness of the slope of the mountain is small, it is likely that the person is
person C.
[0077]
Rule 14: If the change in the steepness of the slope of the mountain is large, the possibility of
being person A is high.
[0078]
Rule 15: If there is a large change in the steepness of the slope of the mountain, it is highly likely
that the person is person B.
[0079]
Rule 16: If the variation of the steepness of the slope of the mountain is small, it is highly
probable that the person is person C.
Of these 16 rules, only the rules relating to the person to be identified by the input of the ID are
applied and integrated, non-fuzzified, and the value is compared with a predetermined threshold
to determine.
The result is sent to the computer terminal 24 to permit or deny login.
08-05-2019
20
Although the above has been described for three different people, it can be easily applied to any
number of subjects.
If there are few target persons, the input of the ID may be omitted.
In this case, all the rules of the determination unit 20 can be applied, and the person showing the
highest non-fuzzified numerical value can be sent to the computer terminal 24 as a
determination result.
[0080]
As described above, according to the present invention, since the individual identification is
performed using the accompaniment sound, it is possible to improve the reliability of the
individual identification.
[0081]
Brief description of the drawings
[0082]
1 is a schematic block diagram of a personal identification device.
[0083]
FIG. 2 is a diagram for explaining finger rubbing.
[0084]
FIG. 3 is an external view of a key ring as an example of an object to be rubbed with a finger.
[0085]
FIG. 4 is an external view of a keyboard as an example of an object to be rubbed with a finger.
[0086]
FIG. 5 is a waveform diagram of the measured sound of the accompaniment sound.
08-05-2019
21
[0087]
FIG. 6 is a waveform diagram of the measured accompaniment sound waveform.
[0088]
FIG. 7 is a waveform diagram of a waveform obtained by filtering the measured accompaniment
sound.
[0089]
FIG. 8 is a waveform diagram of a waveform obtained by filtering the measured accompaniment
sound.
[0090]
FIG. 9 is a waveform diagram of a waveform obtained by filtering the measured accompaniment
sound.
[0091]
FIG. 10 is a waveform diagram of a waveform obtained by filtering the measured accompaniment
sound.
[0092]
FIG. 11 is a waveform diagram of a waveform obtained by filtering the measured accompaniment
sound.
[0093]
FIG. 12 is a waveform diagram of a waveform obtained by filtering the measured accompaniment
sound.
[0094]
FIG. 13 is a waveform diagram of a waveform obtained by filtering the measured accompaniment
sound.
08-05-2019
22
[0095]
FIG. 14 is a diagram for describing feature values.
[0096]
Explanation of sign
[0097]
Reference Signs List 10 personal identification device 12 microphone 14 voltage amplification
unit 16 filtering unit 18 feature extraction unit 20 discrimination unit 22 feature recording unit
24 computer terminal 26 host computer 28 network 30 index finger 32 thumb 34 key ring 36
rhomboid 100 peak of mountain of waveform 100 102 peak peak half value 103 peak peak half
width
08-05-2019
23
Документ
Категория
Без категории
Просмотров
0
Размер файла
37 Кб
Теги
jp2001318691
1/--страниц
Пожаловаться на содержимое документа