JP2021086255A - Pen input personal authentication method - Google Patents

Abstract

To provide a pen input personal authentication method with false acceptance rate (FAR) and false rejection rate (FRR).SOLUTION: A pen input personal authentication method comprises the steps of: obtaining a signature trajectory by capturing coordinates of a pen tip and pen pressure information in a chronological order; calculating a geometric local distance at each point in time between a pre-recorded trajectory of a person's signature and an input trajectory; calculating a pen pressure local distance at each point in time between the pre-recorded trajectory and the input trajectory; calculating a local distance based on the geometric local distance and the pen pressure local distance; measuring a DTW distance using the local distance; calculating a normalized DTW distance; calculating a predetermined statistic from the normalized DTW distance; and identifying whether the input trajectory is the person's signature based on comparison between the statistic and a threshold value.SELECTED DRAWING: Figure 1

Description

The present invention relates to a pen input personal authentication method. More specifically, the present invention relates to a method of authenticating whether or not a person is a person based on handwriting information input to a plane and the vicinity of the upper part of the plane by an electronic pen.

It has been a long time since the demand for personal identification technology in the field of "security" has become a global trend. Personal identification methods include (1) personal memory such as passwords and PINs, (2) personal property such as credit cards, and (3) personal static features such as fingerprints and veins. , (4) Due to personal dynamic products such as signatures and voiceprints (see Table 1). Of these methods, (3) and (4) are called biometrics methods. As the conventional method, there are many of the above (1) and (2).

As a personal authentication method based on the above biometrics method, for example, there is a pen input personal authentication method according to Patent Document 1. This recognizes the handwriting information when signing a tablet with an electronic pen as (1) position trajectory information of the pen tip on the tablet plane and (2) pressure trajectory information, and these are information as necessary. After compression, the evaluation function created for this purpose is used to calculate a predetermined distance from the person's template prepared in advance, and the obtained calculated value is compared with a preset threshold value. It authenticates whether or not the person who signed the information is the person himself / herself. As a result, not only the handwriting as a visible two-dimensional image but also the passage of time and the passage of pressure are used, which makes it very difficult for a third party to imitate the signature. doing.

Patent No. 4307606

Personal identification methods that rely on possessions and stored information are not always perfect in recent years when computer crimes such as the knowledge, theft, and counterfeiting of third parties by fraudulent methods are increasing, and complement these conventional methods. The biometrics method is expected as a substitute for it. There is a demand for a biometrics-type personal authentication method that is a relatively simple method for the person to be identified and that can perform personal authentication at low cost.

Among the biometrics methods, some of the biological characteristics of an individual such as fingerprints and retina have the advantage of being immutable for life, but if the data with those characteristics is stolen by a third party, the person cannot respond. , Has the drawback. On the other hand, if it is found that the signature data has been stolen, the person himself / herself can change the signature. Signature authentication is one of the authentication methods that has been attracting attention in recent years.

False rejection rate (FRR) and false acceptance rate (FAR) are often used as evaluation indexes for authentication technology, and both of these are required to have low authentication technology. However, when configuring the authentication system, if the FRR is set to be low, the FAR will be high, or conversely, if the FRR is set to be low, the FRR will be high. Therefore, there is a demand for a signature authentication technique that simultaneously reduces these contradictory evaluation indexes as much as possible.

Therefore, the present invention provides a pen input personal authentication method having a lower false acceptance rate (FAR) and false rejection rate (FRR).

The pen input personal authentication method according to one aspect of the present invention is a pen input personal authentication method that authenticates the person or not based on the handwriting information drawn by the pen tip on the plane of the input device and in the vicinity above the plane. , The step of obtaining the signature trajectory by taking in the pen tip plane and its upper vicinity coordinates and the pen pressure information in chronological order as the stroke information, the step of extracting the feature amount of the trajectory, and the person recorded in advance. The step of calculating the geometrical local distance between the signature trajectory and the input trajectory at each time, and the pressure between the pre-recorded trajectory and the input trajectory at each time. Pre-recorded by the dynamic time warping (DTW) method using the step of calculating the local distance, the step of calculating the local distance based on the geometrical local distance and the pressure local distance, and the local distance. A step of measuring the DTW distance between the trajectory and the input trajectory, a step of calculating the normalized DTW distance obtained by normalizing the DTW distance for the input trajectory, and a predetermined step from the normalized DTW distance. Includes a step of calculating the statistic of, and a step of identifying whether the entered trajectory is the signature of the person based on the comparison between the statistic and the threshold.

According to this aspect, personal authentication with a lower false acceptance rate (FAR) and false rejection rate (FRR) can be performed based on the handwriting information on the plane of the input device and in the vicinity thereof.

In the above aspect, the step of calculating the local distance may include the step of calculating the local distance by the sum of the geometric local distance and the weighted pen pressure local distance.

In the present invention, the signature is not considered as a two-dimensional still image but as a handwriting trajectory parameterized by time, and authentication is performed based on the distance between the template signature of the person prepared in advance and the input signature to be collated. After feature extraction from the handwriting trajectory, the local distance is set as the distance between the signature vectors at each time. Setting the local distance is one of the most important factors that affect the accuracy of the authentication method. The local distance between signatures at each time has two components. One is the geometric local distance in the two-dimensional plane, and the other is the pen pressure local distance. As the former, we can think of several quantities, including the Euclidean distance of two two-dimensional position vectors. Considering the importance of pen pressure in signature authentication, calculating the product of the geometric local distance between two signatures at each time and the pen pressure local distance works reasonably well, but the fluctuation of the pen pressure local distance varies. As a whole, it may adversely affect the local distance, which in turn may affect the authentication performance. According to this aspect, while considering the pen pressure, the geometric local distance and the pen pressure local distance at each time between signatures are separated, a weighted sum is taken instead of a product, and an appropriate weight is set. , It is possible to reduce the excessive influence from the writing pressure local distance.

In the above aspect, the step of extracting the feature amount of the trajectory may include a vector from a predetermined reference point to each coordinate point.

According to this aspect, the degree of freedom in setting the reference point is increased, and various local distances can be considered.

In the above aspect, the step of calculating the geometric local distance may include the step of calculating the geometric local distance based on the angle, angular velocity and length of the vector.

In the above aspect, the step of calculating the normalized DTW distance may include the step of calculating a plurality of normalized DTW distances, which is larger than the number of previously recorded trajectories.

In the present invention, the integrated value of the local distance at each time is the distance between the two signatures, but the time required to complete the signature is long in the case of a complicated signature even for the same person, while a simple signature. In the case of, it is often short. Even if the signatures are signed by each other, the longer the signature time, the longer the distance as the integrated value, and conversely, the shorter the signature time, the shorter the distance. In addition, even if the signatures are the same, the distance between the signatures tends to be relatively large, or conversely, the distance between the signatures tends to be small. Therefore, when performing authentication by comparing the distance between signatures with a threshold value, it is necessary to set a different threshold value for each individual, which is often difficult, and therefore a distance between signatures with low individual dependence is desired. According to this aspect, individual dependence can be reduced by applying a predetermined normalization. Further, according to this aspect, a normalization distance larger than the number of registered templates can be obtained, and improvement in robustness of the method can be expected.

In the above aspect, the step of calculating the predetermined statistic includes the step of calculating the predetermined statistic from each of the plurality of normalized DTW distances, and identifies whether or not the input trajectory is the signature of the person. The step may include identifying whether or not the entered trajectory is a signature of the principal by an integrated authentication model that integrates the comparison results of a plurality of statistics and a plurality of thresholds.

According to this aspect, high-performance personal authentication can be performed by utilizing the degree of freedom of the integrated method.

In the above aspect, the step of calculating the predetermined statistic includes the step of calculating the predetermined statistic from each of the plurality of normalized DTW distances, and identifies whether or not the input trajectory is the signature of the person. The step may include inputting a plurality of statistics into one discriminator and identifying whether or not the input trajectory is the signature of the person by the output of the discriminator.

In the above embodiment, the pre-recorded trajectory includes a plurality of signature trajectories entered three or more times, and the normalized DTW distance statistic calculated for the plurality of signature trajectories is less than or equal to the threshold value. Is.

Templates are registered in many authentication methods, but the completion conditions also affect the overall accuracy. According to this aspect, when the normalized distance between the templates by the basic authentication model recorded in advance does not exceed the preset threshold value, the registration is completed to suppress the variation of the templates.

The program according to another aspect of the present invention causes a computer to execute the pen input personal authentication method described in the above aspect.

The computer-readable storage medium according to another aspect of the present invention stores the program described in the above aspect.

According to the present invention, it is possible to provide a pen input personal authentication method having a lower false acceptance rate (FAR) and false rejection rate (FRR).

It is a flowchart of the authentication process in the pen input personal authentication method which concerns on 1st Embodiment of this invention. It is a figure which shows the example of the trajectory of the signature in the pen input personal authentication method which concerns on this embodiment. It is a figure which shows the example of the reference point in the pen input personal authentication method which concerns on this embodiment. It is a figure which shows the example of the plurality of local distances calculated in the pen input personal authentication method which concerns on this embodiment. It is a figure which shows the example of two signature trajectories in the pen input personal authentication method which concerns on this embodiment. It is a figure which shows the example of the DTW distance calculated in the pen input personal authentication method which concerns on this embodiment. It is a flowchart of the authentication process in the pen input personal authentication method which concerns on 2nd Embodiment of this invention.

Embodiments of the present invention will be described with reference to the accompanying drawings. In each figure, those having the same reference numerals have the same or similar configurations.

[First Embodiment]
The pen input personal authentication method according to the present embodiment may be carried out by using a commercially available pen tablet as an input device and processing the handwriting information acquired by the pen tablet with a computer. The pen tablet measures the coordinates of the tip of the pen and the pen pressure information in chronological order and outputs them to a computer. In the pen input personal authentication method according to the present embodiment, the position information of the pen tip on the tablet and near the upper part of the tablet and P pieces (P is not particularly limited as long as it is a positive integer value, and the case of being in the air is 1). A combination of an electronic pen and a tablet capable of detecting pen pressure by dividing into the stages of.) Is used.

The pen input personal authentication method according to the present embodiment is generally performed by using a computer including a personal computer and connecting a tablet to it, but in addition, the liquid crystal display or touch panel of the computer is used instead of the tablet. It may be used for. In that case, it is preferable that the pen position near the upper part of the plane can be detected. Further, even in a smartphone equipped with a pen input function, it can be used as long as the pen position near the upper part of the plane can be detected.

FIG. 1 is a flowchart of an authentication process in the pen input personal authentication method according to the first embodiment of the present invention. In the pen input personal authentication method according to the present embodiment, the input signature 101 detects the position information on the tablet plane at the tip of the pen or near the upper part of the plane and the pen pressure divided into P steps in time series as a trajectory. Use a combination of electronic pen and tablet that can be used. The input signature 101 to be collated obtained by signing this tablet and the template signature 102 prepared in advance are input to the authentication models 1 to N (N is a natural number of 2 or more). The local distance setting 103 is set in advance for each model.

Then, after setting the reference point 104 for the input handwriting information, the feature extraction 105 in each model is performed. The data obtained by feature extraction in each model includes position information and pen pressure information on the tablet plane of the pen tip or not only on the plane but also in the upper vicinity of the plane. Based on these, the distance calculation 106 between the signature input to be collated and the plurality of templates prepared in advance is performed. Then, the obtained plurality of distances are normalized 107. A predetermined statistic 108 is obtained from the plurality of normalized distances thus obtained. The comparison 109 between this statistic and the set threshold value is performed, the plurality of comparison results are integrated by the integrated authentication model 110, and after determining whether or not the person is the person, the result output 111 is performed.

FIG. 2 is a diagram showing an example of a signature trajectory in the pen input personal authentication method according to the present embodiment. The signature trajectory is drawn on a two-dimensional plane. In the figure, the minimum value x _{min for} the x coordinate of the _{trajectory, the maximum value x max for} the x coordinate, the minimum value y _{min for the} y coordinate, and the maximum value y _max for the y coordinate are shown.

FIG. 3 is a diagram showing an example of a reference point in the pen input personal authentication method according to the present embodiment. The first example A of the reference point is a point indicated by a circle, which is ((x _max −x _min ) / 2, (y _{max −} y _min ) / 2). The second example B of the reference point is a point indicated by a rhombus, which is (x _max , y _max ). Further, the third example C of the reference point is a point indicated by a triangle, which is an arbitrary point of _{x min} or more and x _{max or less with respect to the x coordinate and y min} or more and y _max or less with respect to the y coordinate. In this way, the reference point can be set arbitrarily, and various feature quantities can be extracted by using different reference points.

In the following, the details of the personal authentication method will be described in the case of using the first example A of the reference point. The geometric local distance between the two signatures is calculated based on the _{angle θ i} defined by the following formula 1 and the length (distance from the reference point) R _{i defined by the following formula 2.}

Here, [Delta] x _i and [Delta] y _i when representing the x-coordinate of the first example A reference point X _A, the y-coordinate and Y _A, is an amount represented by Equation 3 below. Further, I represents the number of data included in the handwriting information.

Further, the geometric local distance may be calculated using the _{angular velocity ω i defined by the following equation 4.} Here, p _i indicates the value of the writing pressure. It is assumed that the pen pressure value has a value from 1 to P.

_{The angle θ i} , the length R _i, and the angular velocity ω _i calculated in this way are examples of the features of the trajectory. In this specification, the feature amounts and writing pressure about the signature of the trajectory of the prerecorded principal, the angle theta _i, represents the length R _i, the angular velocity omega _i and writing pressure p _i, the to be matched signature Trajectory As shown in Equation 5 below, the feature quantities related to the above are expressed as an angle η _j , a length S _j , an angular velocity κ _j, and a writing pressure q _j . Here, J represents the number of data included in the trajectory of the signature to be collated.

In the pen input personal authentication method according to the present embodiment, the geometric local distance and the pen pressure local distance at each time between the previously recorded trajectory of the signature of the person and the input trajectory are calculated. In the present specification, the geometrical local distance and the pen pressure local distance are generally expressed by the following mathematical formula 6. Local distance u _n, v _n, w _n are calculated for each authentication model. As described later, u _n is the local distance in the diagonal direction, v _n is the local distance in the vertical direction, w is a local range in the horizontal direction.

Local distance u _n, v _n, w _n can be as shown in general following formula 7.

In Equation 7, "1" and "2" are a family of four-coefficient functions parameterized by (ρ, φ, ψ, ξ). Further, "3" and "4" in Equation 7 are a 5-coefficient function family parametrized by (ρ, φ, τ, ψ, ξ). Each of these has many degrees of freedom, and a suitable local distance can be designed according to the purpose.

The | distance difference from the reference point | in "1" and "2" in Equation 7 is not the Euclidean distance between the two pen tip positions, but the difference in the distance from the reference point of each of the two points. In addition, the | angle difference | φ × | distance difference from the reference point | ψ that appears in the geometric local distance is called the “generalized angle / distance product”, and the geometry suitable for designing φ and ψ according to the purpose. Geometric local distances can be designed. For example, when one nib position is fixed, the second set of nib positions that satisfies "Euclidean distance between two nib positions = constant" is a circumference whose radius is the constant on the right side, for example, φ = When ψ = 1, the set of the second nib positions satisfying “generalized angle / distance product = constant” is not a circumference but a double curve.

If the local distance at each time is calculated, in order to define the total distance between two signatures, i and j may be integrated. However, since there are a plurality of routes that i and j can move, it is necessary to specify some special route, and the route that gives the minimum value of the total local distance is often used. Further, while i = 1,2, ..., I, j = 1,2, ..., J, and generally I ≠ J, the index is re-calculated (re-parameterize) and either locally. Allow the index to repeat and calculate the sum of the same number of local distances as a whole. That is to find a suitable sub-index i _s and _{j s, s = 1,2, ...} , take the sum of up to S. Since the data is to maintain its causality since the signature of the time-series data, the sub-index i _s and j _s is required to satisfy the relation _{i s ≦ i s + 1,} j s ≦ j s + 1 .. The conversion data _{(θ i, R i, ω} i, p i), (η j, S j, κ j, q j) to avoid loss _{of, i s + 1 ≦ i s} + 1, j s Constraint so that the relationship of ₊₁ ≤ j _{s +1 is satisfied.} Further, i ₁ = j ₁ = 1, i _S = I, j _S = J.

The signature to be collated, the feature amount including the angle component, angular velocity component and length component of the vector from the reference point position to each handwriting coordinate point, the writing pressure information at each handwriting coordinate point, and the person prepared in advance. Based on the template signature of, the DTW distance D _n represented by the following equation 8 is calculated.

Here, is _s , j _s and _s are the above-mentioned indexes. D _n is the distance calculated by the nth authentication model. Further, when there are N authentication models and M template signatures (however, 3 ≦ M), the normalization represented by the following equation 9 is performed for each authentication model, and the normalized DTW distance is calculated. Here, the normalized DTW distance in the _nth authentication model is represented as Dis n. The normalized DTW distance Dis _n is the DTW distance between the given signature data and the template divided by the DTW distance between the templates.

In Equation 9, θ (α) _i is the angle information of the vector from the handwriting information obtained from the α-th person's template to each handwriting coordinate (x _i , y _i). η _j is the angle information of the vector from the reference point position of the input handwriting information of the signature to be collated input at the time of authentication to each handwriting coordinate (x _j , y _j). ω (α) _i and κ (β) _j are angular velocity information defined by θ (α) _i −θ (α) _i-1 , η (β) _j − η (β) _{j-1, respectively.} .. p (α) _i is the pen pressure information of the αth person template, and q (β) _j is the corresponding input pen pressure information. In addition, γ is also used in the same meaning.

In this way, in each authentication model, the DTW distance between the input handwriting information of the signature to be collated input at the time of authentication and some personal template signature handwriting information collected in advance is calculated and normalized to sign the signature. It is possible to reduce the individual dependence of the distance.

Performing DTW distance calculation and normalization yields more normalized DTW distances than the number of template signatures. For example, when the number of template signatures M = 3, nine normalized DTW distances can be obtained. As a result, a normalized DTW distance three times as many as the number of template signatures can be obtained, and robustness is improved.

In the pen input personal authentication method according to the present embodiment, the calculated normalized DTW distance statistic, for example, the average value, the median value, the maximum value, the minimum value, or a linear combination thereof, and a preset threshold value are used. If the statistic is less than or equal to the threshold value, it is authenticated as the signature of the person.

For the template signature, registration may be permitted when the statistic of the normalized DTW distance represented by the following formula 10 is equal to or less than the threshold value by using the _{distance Dis 0} for one basic authentication model prepared in advance. It should be noted that 3 ≦ M.

By using the formula 10, _{the number of distances Dis 0} is larger than the number of registered signatures M. For example, when M = 3, the number of _{distances Dis 0 is six.}

FIG. 4 is a diagram showing an example of a plurality of local distances calculated in the pen input personal authentication method according to the present embodiment. In the figure, the template signature (signature data 1) having a length of 12 is represented as a horizontal vector, and the input handwriting information (signature data 2) having a length of 15 is represented as a vertical vector. Thick arrows shown in the figure, the local distance u _n of each diagonal represents the local distance w _n in the vertical direction of the local distance v _n and horizontally.

FIG. 5 is a diagram showing an example of two signature trajectories in the pen input personal authentication method according to the present embodiment. In this example, the signature trajectories of the same person are shown in layers.

For example, in character recognition, it is often important to correspond points with relatively steep angular changes (for example, M. Kobayashi, S. Masaki, O. Miyamoto, Y. Nakagawa, Y. Komiya, T. Matsumoto, "RAV (re-parameterized angle variations) algorithm for online handwriting recognition", International Journal on Document Analysis and Recognition, March 2001, Volume 3, Issue 3, pp 181-191) Is also considered important. For example, when φ = ψ = 1 in “1” and “2” of Equation 7, in the area shown in (a) of this example, the distance between the two points considered to correspond to the two signatures is to some extent. It has an Euclidean distance, but both are on the arrows shown in the figure, and the angle / distance product is very small.

Further, in the region shown in (a), the two points considered to correspond to the two signatures are on almost the same circumference, but the Euclidean distance has a relatively large value. In this case as well, the angle / distance product becomes very small.

In this way, by using the angle / distance product, it is possible to detect the corresponding points that are difficult to detect at the Euclidean distance relatively accurately. Similarly, in "3" and "4" of Equation 7, | angle difference | φ × | angular velocity difference | τ × | distance difference from the reference point | ψ is called "generalized angle / angular velocity / distance product", and these are called "generalized angle / angular velocity / distance product". Parameters can also be designed. The angular velocity information of the pen tip can also be an effective feature amount. Further, if ξ = 0 is set in "1" and "3" of the mathematical formula 7, it becomes a local distance function that does not consider the pen pressure difference, and can be used even in a device that cannot acquire the pen pressure.

If φ = ψ = ξ = 1 is set in "1" of the formula 7, | distance difference from the reference point | × | angle difference | × | pen pressure difference | Known as an effective feature quantity (for example, Y. Komiya and T. Matsumoto, "On-line Pen Input Signature Verification PPI (pen-Position / pen-Pressure / pen-Inclination)", IEEE Internal Symposium on Systems, See Man and Cybernetics, pp. N-41- N-46, 1999, INSPEC Accession Number: 6529578).

“2” in Equation 7 is not the product of the generalized angle / distance product and | pen pressure difference | ξ, but the sum of the generalized angle / distance product and the weighted | pen pressure difference | ξ. The purpose is to improve the robustness of the local distance by separating the geometric information and the pen pressure information. By giving an appropriate weight, it becomes less susceptible to direct fluctuations in pen pressure difference, and the local distance becomes robust. The same applies to "4" in Equation 7.

FIG. 6 is a diagram showing an example of the DTW distance calculated in the pen input personal authentication method according to the present embodiment. In the pen input personal authentication method according to the present embodiment, as shown in Equation 8, the minimization problem is solved when calculating the _{DTW distance D n.} Here, since a combined explosion occurs when all the routes are listed, the pen input personal authentication method according to the present embodiment is calculated by the dynamic time warping method (DTW method). In the DTW method, since the solution of the sequential minimization shown in the following equation 11 becomes the solution of the global minimization, it is often possible to avoid the combinatorial explosion. _{Here, D 1} (0,0) = 0 is imposed as a boundary condition.

In this way, the DTW distance is a value finally determined D _n (I, L), and the DTW distance D _n of the authentication model n. FIG. 6 schematically shows such sequential processing.

In the pen input personal authentication method according to the present embodiment, the DTW distance calculated in this way is normalized as in Equation 9, the statistic is calculated, the statistic for each authentication model is compared with the threshold value, and these are compared. The signature is authenticated using the result of integrating.

There is a high degree of freedom in the method of integrating statistics, and various methods can be considered. For example, a model with a very low FAR, that is, a model with an extremely high spoofing prevention performance can be designated as H ^*, and the FRR can be improved by performing authentication by an integrated model using a plurality of models as in the procedure shown in Equation 11 below. it can. Roughly speaking, it is expected that the "personal data" determined to be "spoofing data" will be "rescued" to improve the FRR, and as a result, further improvement in accuracy will be expected.

[Second Embodiment]
FIG. 7 is a flowchart of the authentication process in the pen input personal authentication method according to the second embodiment of the present invention. In the pen input personal authentication method according to the present embodiment, the input signature 201 is detected in chronological order as the position information on the tablet plane of the pen tip or in the vicinity of the upper part of the plane and the pen pressure divided into P stages as a trajectory. Use a combination of electronic pen and tablet that can be used. For the input signature 201 to be collated obtained by signing this tablet and the template signature 202 prepared in advance, the statistics of the distance between signatures 1 to N (N is a natural number of 2 or more) are calculated. For each signature distance, the local distance setting 203 is performed in advance.

Then, after setting the reference point 204 for the input handwriting information, feature extraction 205 in each model is performed. The data obtained by feature extraction in each model includes position information and pen pressure information on the tablet plane of the pen tip or not only on the plane but also in the upper vicinity of the plane. Based on these, the distance calculation 206 between the signature input to be collated and the plurality of templates prepared in advance is performed. Then, the obtained plurality of distances are normalized 207. From the plurality of normalized distances thus obtained, the statistic 208 (statistics 1 to N) is obtained. These statistics 1 to N are input to one discriminator 209, and the output is used to determine whether or not the person is the person himself / herself, and then the result output 210 is performed.

Personal authentication using the discriminator can be performed as in the procedure shown by the following mathematical formula 13.

As described above, according to the present invention, when extracting the handwriting distance of the person from the signature input to the tablet by the electronic pen and comparing it with the template of the person prepared in advance, the calculation is high. High performance and robustness by setting local distances, low dependence on individuals, using normalized metrics and their statistics, integrating multiple models, or making decisions based on multiple distances. It is possible to build a high-performance authentication system.

The pen-input personal authentication method of the present invention reads an execution program stored in a computer-readable storage medium, for example, ROM (Read Only Memory), RAM (Random Access Memory), or a hard disk, and executes the program to execute the computer. It can also be done above. Further, the pen input personal authentication method of the present invention can also be provided in the form of a computer or a program incorporated in a device and an electronic device controlled by the computer. For example, it can be incorporated into automatic cash payment machines, access control devices, and any other device or electronic device that requires personal authentication.

The embodiments described above are for facilitating the understanding of the present invention, and are not for limiting and interpreting the present invention. Each element included in the embodiment and its arrangement, material, condition, shape, size, and the like are not limited to those exemplified, and can be changed as appropriate. In addition, the configurations shown in different embodiments can be partially replaced or combined.

Claims (10)

This is a pen input personal authentication method that authenticates the identity of a person based on the handwriting information drawn by the pen tip on the plane of the input device and in the vicinity above it.
A step of obtaining a signature trajectory by taking in the pen tip tip on the plane and its upper vicinity coordinates and pen pressure information in chronological order as the handwriting information.
The step of extracting the feature amount of the trajectory and
A step of calculating the geometric local distance at each time between the previously recorded trajectory of the person's signature and the input trajectory.
A step of calculating the writing pressure local distance at each time between the previously recorded trajectory and the input trajectory, and
A step of calculating a local distance based on the geometric local distance and the pen pressure local distance, and
A step of measuring the DTW distance between the previously recorded trajectory and the input trajectory by the dynamic time warping (DTW) method using the local distance.
A step of calculating a normalized DTW distance obtained by normalizing the DTW distance with respect to the input trajectory, and a step of calculating the normalized DTW distance.
A step of calculating a predetermined statistic from the normalized DTW distance, and
A step of identifying whether or not the entered trajectory is the signature of the person based on the comparison between the statistic and the threshold value.
Pen input personal authentication method including. The step of calculating the local distance includes a step of calculating the local distance by the sum of the geometric local distance and the weighted pen pressure local distance.
The pen input personal authentication method according to claim 1. The step of extracting the feature amount of the trajectory includes a vector from a predetermined reference point to each of the coordinate points.
The pen input personal authentication method according to claim 1 or 2. The step of calculating the geometric local distance includes a step of calculating the geometric local distance based on the angle, angular velocity and length of the vector.
The pen input personal authentication method according to claim 3. The step of calculating the normalized DTW distance includes a step of calculating a plurality of normalized DTW distances that is larger than the number of the trajectories recorded in advance.
The pen input personal authentication method according to any one of claims 1 to 4. The step of calculating the predetermined statistic includes the step of calculating the predetermined statistic from each of the plurality of normalized DTW distances.
The step of identifying whether or not the input trajectory is the signature of the principal is performed by the integrated authentication model that integrates the comparison results of the plurality of statistics and the plurality of threshold values. Includes a step to identify whether it is a signature of
The pen input personal authentication method according to claim 5. The step of calculating the predetermined statistic includes the step of calculating the predetermined statistic from each of the plurality of normalized DTW distances.
In the step of identifying whether or not the input trajectory is the signature of the person himself / herself, a plurality of the statistics are input to one discriminator, and the output of the discriminator causes the input trajectory to be input. Including the step of identifying whether it is the signature of the person,
The pen input personal authentication method according to claim 5. The pre-recorded trajectory includes a plurality of signature trajectories entered three or more times, and the normalized DTW distance statistic calculated for the plurality of signature trajectories is less than or equal to the threshold.
The pen input personal authentication method according to any one of claims 1 to 7. A program for causing a computer to execute the pen input personal authentication method according to any one of claims 1 to 8. A computer-readable storage medium that stores the program according to claim 9.

Images