KR20200072070A - Electronic device for generating verification model used to verify a user's signature - Google Patents

Abstract

According to various embodiments, disclosed is an electronic device for generating a verification model used to verify a signature of a user. According to one embodiment of the present invention, the electronic device for generating a verification model used to verify a signature of a user comprises: a display capable of receiving a signature of a user; a memory storing a verification model for verifying the signature of the user; and a processor. The processor may be set to receive the signature of the user at least once from the user, extract a plurality of feature vectors related to characteristics of the received at least one signature of the user, generate a verification model having characteristics of a single class classification model based on the plurality of feature vectors, and perform the verification of the signature using the generated verification model.

Description

An electronic device that generates a verification model used for verification of a user's signature{ELECTRONIC DEVICE FOR GENERATING VERIFICATION MODEL USED TO VERIFY A USER'S SIGNATURE}

Various embodiments of the present invention relate to an electronic device that generates a verification model used for verification of a user's signature.

Due to the spread of smart devices and the increasing demand for security of personal information, the importance of user authentication problems in the smart environment is increasing day by day, and identification of a user on a smart phone is a representative example. User authentication methods that are commonly used in smartphones include passwords and path patterns, but they are prone to exposure to others. Recently, biometric recognition methods such as fingerprint recognition, iris recognition, and face recognition have also been introduced and used, but in the case of biometric recognition, a simple and efficient authentication method that can be used in a light authentication process is required because it may cause a user's refusal.

Signature verification is one of the methods traditionally used for identity verification. In particular, in a smart device such as a smartphone or tablet, not only a morphological factor, but also pressure can be easily obtained, and a method of using signature verification for identity verification is being implemented.

However, the signature verification system is different from the general pattern recognition problem, and it is typical that it is difficult to obtain a counterfeit signature. In general, in pattern recognition, a model is constructed using samples representing two or more classes. As in the case of general pattern recognition, it is necessary to establish a signature verification system and a signature signature to establish a signature verification system, but it is not easy to obtain a signature signature in practical applications. Therefore, in order to establish a practical signature verification system, a method of performing signature verification using only a signature without a fake signature is required.

Various embodiments of the present invention are intended to provide an electronic device that generates a verification model used for verification of a user's signature.

An electronic device generating a verification model used for verification of a user's signature according to an embodiment of the present invention includes a display capable of receiving a user's signature; A memory storing a verification model for verifying the user's signature; And a processor, wherein the processor receives a user's signature at least once from a user, extracts a plurality of feature vectors related to the characteristics of the received at least one user's signature, and the plurality of feature vectors Based on this, a verification model having the characteristics of a single class classification model may be generated and set to perform verification of a signature using the generated verification model.

In an electronic device generating a verification model used for verification of a user's signature according to various embodiments of the present disclosure, the processor may be configured to generate the verification model using a support vector machine (SVC). have.

In an electronic device generating a verification model used for verification of a user's signature according to various embodiments of the present disclosure, the plurality of feature vectors are extracted using characteristics of a signature image corresponding to the at least one user's signature Can be set.

In an electronic device generating a verification model used for verification of a user's signature according to various embodiments of the present disclosure, the signature image is implemented in a first dimension and a second dimension, and the plurality of feature vectors are the first dimension Alternatively, a feature vector associated with one of the second dimensions may be included.

In an electronic device generating a verification model used for verification of a user's signature according to various embodiments of the present disclosure, the plurality of feature vectors may include feature vectors related to the first dimension and the second dimension.

In an electronic device that generates a verification model used for verification of a user's signature according to various embodiments of the present invention, the processor checks the amount of change in capacitance of the display when receiving the user's signature, and determines the amount of change in the capacitance. Based on this, the user can determine the pressure applied to the display.

In an electronic device generating a verification model used for verification of a user's signature according to various embodiments of the present disclosure, the plurality of feature vectors may include feature vectors related to the applied pressure.

In an electronic device generating a verification model used for verification of a user's signature according to various embodiments of the present disclosure, the processor receives at least one or more other user's signatures at least once, and the received at least one other Extracting a plurality of feature vectors related to a user's signature characteristic, and generating the verification model based on a plurality of feature vectors related to the user's signature characteristic and a plurality of feature vectors related to the characteristic of the other user's signature Can be set.

In an electronic device that generates a verification model used for verification of a user's signature according to various embodiments of the present disclosure, the processor is configured to display the at least one other user's excluding counterfeit signatures having a form similar to that of the user's signature. It may be configured to generate the verification model based on a signature.

In an electronic device generating a verification model used for verification of a user's signature according to various embodiments of the present disclosure, the signature of the other user may be received from an external server.

An electronic device generating a verification model used for verification of a user's signature according to various embodiments of the present disclosure may further include a security module that stores the generated verification model.

An electronic device that generates a verification model used for verification of a user's signature according to various embodiments of the present disclosure can generate a verification model by using feature vectors for at least one signature of a user other than a counterfeit signature, so that the verification model It is possible to generate a verification model even in a situation where it is difficult to obtain a signature.

An electronic device that generates a verification model used for verification of a user's signature according to various embodiments of the present invention uses at least one signature of another user for generation of a verification model, thereby providing a verification model that can perform accurate verification. Can be created.

1 is a block diagram illustrating an electronic device that performs verification of a user's signature according to an embodiment of the present invention.
2 is a diagram illustrating a change in an error rate according to a threshold value in a handwriting recognition of a user using a verification model having a characteristic of a single class classification model in an electronic device according to various embodiments of the present disclosure.
3 is a diagram illustrating a change in an error rate according to the number of samples in a verification model generated using a signature of another user in an electronic device according to various embodiments of the present disclosure.

If described in detail with reference to the accompanying drawings the present invention. Here, repeated descriptions, well-known functions that may unnecessarily obscure the subject matter of the present invention, and detailed description of the configuration will be omitted. Embodiments of the present invention are provided to more fully describe the present invention to those skilled in the art. Therefore, the shape and size of elements in the drawings may be exaggerated for a more clear description.

Hereinafter, preferred embodiments of the present invention will be described in detail with reference to the accompanying drawings.

1 is a block diagram illustrating an electronic device that performs verification of a user's signature according to an embodiment of the present invention.

Referring to FIG. 1, an electronic device that performs verification of a user's signature according to various embodiments of the present disclosure may include a display 110, a processor 120, and a memory 130.

According to various embodiments of the present invention, the display 110 may display various screens under the control of the processor 120. The display 110 may include a touch panel capable of receiving user input and a display panel capable of displaying various screens. The touch panel and the display panel can be implemented as one integrated panel.

According to various embodiments of the present invention, the display 110 may receive a user's signature. The user can enter a signature on the display 110. The user may input a signature using a pen having a conductivity that the display 110 can detect. When the display 110 includes a separate pen sensor panel, a user can input a signature using a pen that the pen sensor panel can detect.

According to various embodiments of the present invention, a signature of a user received by the display 110 may be transmitted to the processor 120.

According to various embodiments of the present invention, the processor 120 may receive a user's signature and perform verification on the authenticity of the received user's signature. The processor 120 may receive the user's signature and perform verification on the authenticity of the user's signature using a verification model. The processor 120 may perform verification on the authenticity of the user's signature, and perform various tasks based on the verification result. For example, the processor 120 may perform various applications (eg, financial applications, etc.) based on the verification result.

According to various embodiments of the present invention, the memory 130 may store a verification model used for verification of a user's signature. The memory 130 may store the verification model in an area physically or logically separated from other data. Alternatively, the verification model may be stored in a security module physically or logically separated from the memory 130.

According to various embodiments of the present invention, the processor 120 may generate a verification model used for verification of a user's signature.

The existing verification model may be generated using the first signature data corresponding to the user's signature and the second signature data corresponding to the counterfeit signature having a shape similar to the user's signature. The existing verification model can classify whether the signature data is a user's signature or a counterfeit signature having a shape similar to the user's signature.

However, there may be a problem that accessibility to the counterfeit signature having a shape similar to that of the user is low. In order to generate a verification model with a low error rate, a sufficient number of counterfeit signatures must be secured, but it is difficult to secure counterfeit signatures. Furthermore, in order to generate a verification model with a low error rate, a large number of first signature data corresponding to a user's signature must also be secured, but there is a problem of registering a large number of signatures.

According to various embodiments of the present invention, the processor 120 may generate a verification model using data related to a plurality of signatures of a user without using a counterfeit signature. The verification model generated using data related to a plurality of signatures of a user may have characteristics of a single class classification model. The single class classification model is specifically described below in relation to the content of generating the verification model, and a specific embodiment of generating the verification model will be described below.

According to various embodiments of the present invention, the processor 120 may generate a verification model having the characteristics of the single class model described above based on a plurality of signatures of the user. The processor 120 may receive the user's signature input by the user at least once. The processor 120 may extract a plurality of feature vectors related to characteristics of the user's signature input by the user.

According to various embodiments of the present invention, the processor 120 may extract a signature image corresponding to a user's signature, and extract a plurality of feature vectors from the signature image. The plurality of feature vectors may include features related to the form of the user's signature.

According to various embodiments of the present invention, the signature image may be an image implemented in a two-dimensional form. For example, the signature image may be an image implemented in a first dimension corresponding to the X-axis and a second dimension corresponding to the Y-axis. The processor 120 may include a plurality of feature vectors for features related to the first dimension, a plurality of feature vectors for features related to the second dimension, and a plurality of feature vectors for features related to the first dimension and the second dimension. Can be extracted.

According to various embodiments of the present invention, the processor 120 may include an input speed of a plurality of lines included in a signature input by a user, an average of input speeds, a maximum value of input speeds, a minimum value of input speeds, and a maximum value of input speeds. It is possible to extract feature vectors for various features including the difference between and the minimum value, the acceleration at which the line included in the signature input by the user is drawn, the size, ratio of the signature, and the dispersion of the input speed. For example, the processor 120 may extract 56 feature vectors described in Table 1 below.

Characteristic Explanation x y One 2 average speed 3 4 (Max-Min) / duration of pen-down 5 6 average of absolute velocities 7 8 average of positive velocities 9 10 average of negative velocities 11 12 average of accelerations 13 14 average of absolute accelerations 15 16 variance of speed 17 18 variance of absolute velocity 19 20 variance of acceleration 21 22 maximum speed 23 24 maximum acceleration 25 26 minimum speed 27 28 difference between maximum and average speed 29 30 difference between positive maximum velocity
and positive average velocity 31 32 frequency of positive velocity 33 34 frequency of negative velocity 35 36 zero-crossover of velocity 37 38 average positive acceleration 39 40 difference between maximum and minimum speed 41 accumulated angles 42 accumulated angles / total duration 43 average speed 44 ratio of height and width 45 ratio of eigen values of signature shape vectors 46 zero-crossover at gravity center 47 ratio of left and right margin at gravity center 48 ratio of upper and lower margin at gravity center 49 number of strokes 50 duration of pen-up 51 duration of pen-down 52 accumulated angles / duration of pen-down 53 variance of speed 54 maximum speed 55 average-crossover of speed 56 total duration

Referring to Table 1, the processor 120 may extract 56 feature vectors. The feature vectors described in Table 1 are exemplary, other feature vectors may be extracted, and a verification model may be generated using feature vectors excluding some of the 56 feature vectors.

According to various embodiments of the present invention, the processor 120 may generate a verification model having characteristics of a single class classification model based on the extracted plurality of feature vectors.

According to various embodiments of the present disclosure, a single class classification model may mean a model that expresses an area in which data to be classified may exist as a sphere having a center point and a radius. The single class classification model can classify data based on whether or not the input data is included inside the sphere. A single class model can be expressed as Equation 1 below.

[Equation 1]

: 패널티)(R: radius, C: constant,

: penalty)

When a single class model is used, data included in a sphere implemented by a single class model among a plurality of data may be classified. When using a verification model having the characteristics of a single class model, the processor 120 may classify a signature having characteristics of a user's signature among input signatures.

According to various embodiments of the present invention, the processor 120 may use signature data of another user to reduce an error rate for a model implemented in the form of Equation 1. For example, in a model represented by a sphere, the error rate can be reduced by reducing the size of the radius.

According to various embodiments of the present invention, the processor 120 may receive signatures of users other than the user, and generate a verification model using both the signatures of the other users and the user's signature. Other people's signatures do not have a form similar to the signature of the user to be verified, but may reflect the characteristics of general writing. By using the signatures of others, it is possible to solve the problem of the lack of the number of signatures of the user that may occur when generating a verification model having the form of a single class classification model. Other people's signatures may refer to each other's genuine signature, not a signature (imitation signature) entered by others in a form similar to that of the user's signature. Other people's signatures may be input on the display 110 or may be received by a separate external server. A separate external server stores other people's signatures to reduce the error rate of the verification model implemented in the electronic device 100, and in response to a request from the electronic device 100, the electronic signature data of each other person It may mean various electronic devices that are transmitted to (100).

According to various embodiments of the present invention, the processor 120 may use a support vector machine in generating a verification model. The support vector machine may be a form of an application installed in the memory 130. The processor 120 may further reduce the error rate by generating a verification model using a support vector machine.

According to various embodiments of the present disclosure, the processor 120 may extract a feature vector related to pressure applied to the display 110 when a user performs a signature, and generate a verification model based on the feature vector related to pressure. have. The processor 120 may check the amount of change in the capacitance of the display when receiving the user's writing, and determine the pressure applied by the user to the display 110 based on the amount of change in the capacitance. The processor 120 may extract a feature vector related to the pressure applied to the display 110.

According to various embodiments of the present invention, the processor 120 may perform verification of the user's signature using the generated verification model. The processor 120 may verify whether the input signature data is the user's signature or not the user's signature using the generated verification model.

2 is a diagram illustrating a change in an error rate according to a threshold value in a signature recognition of a user using a verification model having a characteristic of a single class classification model in an electronic device according to various embodiments of the present disclosure, and FIG. In an electronic device according to various embodiments of the present disclosure, in a verification model generated using a signature of another user, a diagram showing a change in an error rate according to the number of samples.

The electronic device according to the first embodiment of the present invention receives a user's signature (signature), and uses the verification model generated based on a plurality of feature vectors related to the characteristics of the user's signature to Verification can be performed.

An electronic device according to a second embodiment of the present invention receives a user's signature (a true signature) and another user's signature (a true signature), and includes a plurality of feature vectors and other user's signatures related to the characteristics of the user's signature. The input signature may be verified using a verification model generated based on a plurality of feature vectors related to the characteristic.

The verification models (for convenience, the verification model generated according to the first embodiment is the first verification model, and the verification model generated according to the second embodiment is second verification) respectively generated using the two methods described in FIG. 1. 2), and the results of the verification are described in FIGS. 2 and 3.

The first experiment was conducted using only the first verification model without using the signature data of other users, and 20 signatures out of 100 signatures were randomly selected and used for learning. For the test, 20 signatures and 50 counterfeits were randomly selected from samples that were not used for learning. The experiment results were presented by averaging the results after repeating the same experiment 1,000 times.

2 shows a change in error when different thresholds are applied to values output through the first verification model. As the threshold changes, the type 1 error and the type 2 error show the opposite, and the sum of them represents a minimum error, which corresponds to the error presented herein. In this specification, the experimentally determined -0.5 was used as a threshold.

Table 2 compares the errors according to the verification method. When using a single class SVM, the average error was 17.47%.

Verification method Average error rate (%) Existing verification model 13.42 1st verification model to apply a predetermined threshold 17.47 First verification model using different thresholds for different users 9.62

In the case of using the verification model trained using 20 signatures as a positive sample and the remaining 8 signatures as a negative sample, the error increased when compared to the average error of 13.42%. It originated from the difference in the experimental method of the verification model. When using the existing verification model, a separate existing verification model is created for each signer, and the boundary used by each existing verification model is different. When using the first verification model, a separate first verification model is created for each signer, but the boundary is the same. That is, the same threshold was used in nine first verification models for nine signers.

When different threshold values are applied for each signer, the minimum error rate that can be obtained using the first verification model is 9.62%, which is lower than that of the existing verification model. That is, by applying a method of adaptively setting the threshold value according to the signer, a lower error can be obtained than using the existing verification model. However, in Table 2, the adaptive threshold is an experimentally determined value.

In the second experiment, another person's letter was used in the learning data. The signature of the other person's signature could not be found in a similar form to the signature to be verified, but was used as general handwriting data because it can be seen as reflecting the general writing characteristics of the signature.

FIG. 3 compares the error in the case of using N(0~20) titles for learning data, and the case of using 8 other titles along with the names of 8 other names, that is, general writing data. As can be seen from FIG. 3, it can be seen that the error is reduced when the general handwriting data is used.

Another advantage of using regular handwriting data is that the required number of signatures is small.

Number of samples Second verification model using general signature data First verification model generated without using general signature data Average error rate (%) Change in error rate
(%) Average error rate (%) Change in error rate 0 23.43 - - - One 20.42 12.85 25.19 - 2 17.54 14.10 23.32 7.42 3 16.51 5.87 21.72 6.86 4 16.40 0.67 20.97 3.45 5 16.43 -0.18 20.23 3.53 6 16.54 -0.67 19.74 2.42 7 16.50 0.24 19.40 1.72 8 16.47 0.18 18.90 2.58 9 16.49 -0.12 18.70 1.06 10 16.46 0.18 18.36 1.82 11 16.38 0.49 18.15 1.14 12 16.44 -0.37 18.03 0.66 13 16.36 0.49 17.86 0.94 14 16.38 -0.12 17.86 0.00 15 16.30 0.49 17.75 0.62 16 16.34 -0.25 17.67 0.45 17 16.30 0.24 17.50 0.96 18 16.26 0.25 17.53 -0.17 19 16.21 0.31 17.49 0.23 20 16.18 0.19 17.47 0.11

As can be seen from Table 3, in the case of using general handwriting data, the average error reduction did not show a significant change of less than 1% when the increase in the name of the signature increased when four or more signatures were used. On the other hand, when the general handwriting data was not used, the average error reduction was less than 1% when 10 or more signatures were used. In the practical application, since it is desirable to register only as few as possible, the use of general handwriting data has an effect of reducing errors and reducing the number of required names.

As described above, the electronic device that performs verification of a user's signature according to various embodiments of the present invention is not limited to the configuration and method of the above-described embodiments, and the embodiments are variously modified. All or part of each of the embodiments may be selectively combined to constitute this.

Claims (10)

In the electronic device for verifying the user's signature,
A display capable of receiving a user's signature;
A memory that stores a verification model for verifying the user's signature; And
Including a processor,
The processor
Receive the user's signature at least once from the user,
Extracting a plurality of feature vectors related to characteristics of the received at least one user's signature,
A verification model having a feature of a single class classification model is generated based on the plurality of feature vectors,
An electronic device set to perform verification of a signature using the generated verification model.
According to claim 1,
The processor
An electronic device configured to generate the verification model using a support vector machine (SVC).
According to claim 1,
The plurality of feature vectors
An electronic device set to be extracted using characteristics of a signature image corresponding to the signature of the at least one user.
According to claim 3,
The signature image
Implemented in the first dimension and the second dimension,
The plurality of feature vectors
An electronic device including a feature vector associated with one of the first dimension and the second dimension.
The method of claim 4,
The plurality of feature vectors
And a feature vector associated with the first dimension and the second dimension.
According to claim 1,
The processor
Confirm the amount of change in capacitance of the display when receiving the user's signature,
Based on the capacitance change amount, the user determines the pressure applied to the display,
The plurality of feature vectors
An electronic device comprising a feature vector associated with the applied pressure.
According to claim 1,
The processor
At least one other user's signature is received at least once,
Extracting a plurality of feature vectors related to characteristics of the received at least one other user's signature,
An electronic device configured to generate the verification model based on a plurality of feature vectors related to the user's characteristic and a plurality of feature vectors related to the characteristic of the other user's signature.
The method of claim 7,
The processor
An electronic device configured to generate the verification model based on the signatures of the at least one other user, except for counterfeit signatures having a form similar to the form of the user's signature.
The method of claim 7,
The electronic device receives the signature of the other user from an external server.
According to claim 1,
The electronic device
And a security module for storing the generated verification model.

Images