Signature handwriting authentication method, storage medium and electronic device integrating point features, local features and global features

Abstract

The application relates to an online signature authentication method integrating point features, local features and global features, and relates to the technical field of electronic signature handwriting authentication. The point features of a sample signature and a to-be-tested signature are subjected to time dynamic programming to obtain the point feature similarity of the sample signature and the to-be-tested signature, the local feature similarity of the sample signature and the to-be-tested signature is calculated based on LNPS features, the global feature similarity of the sample signature and the to-be-tested signature is calculated according to the Euclidean distance, the comprehensive similarity of the sample signature and the to-be-tested signature is calculated according to the point feature similarity, the local feature similarity and the global feature similarity, and the signature is authenticated according to the comprehensive similarity. The authentication result is more real and reliable, and the accuracy and precision of judgment in the handwriting recognition and authentication process are improved.

Description

Technical Field

[0001] This invention belongs to the field of computer information processing technology, and specifically relates to a method for comprehensively judging online authentication signatures using time dynamic programming. Background Technology

[0002] Most existing online signature authentication methods collect the x-coordinate, y-coordinate, and pressure value of the user's signature at each moment, and calculate point features such as velocity, acceleration, angle, and angular velocity based on this information. Chinese invention patents, such as application number 201310521026 "An Online Handwritten Signature Authentication Method Based on Dynamic Threshold" and application number 201610250594 "An Online Signature Identity Authentication Method and System Based on Mobile Phone Sensing," all rely on these point features and use Time Dynamic Programming (DTW) to derive the similarity between the signature to be tested and the user's sample signature. They only change the threshold selection strategy, relying solely on DTW to calculate the distance between points without considering the distance between corresponding strokes in the signature for comprehensive decision-making. Therefore, the accuracy of the judgment is not high.

[0003] Chinese invention patent application CN111461015A, entitled "A User-Independent Online Signature Authentication Method and Apparatus Based on an RNN Model," provides a user-independent online signature authentication method and apparatus based on an RNN model. The method involves splitting the registered signature and the signature to be authenticated into multiple stroke data points based on stroke count. These stroke data points are then concatenated, and a stroke RNN model is used to extract the individual stroke features from the concatenated stroke data. These stroke features are then input into a signature RNN model to extract the overall signature features, and the similarity and difference values ​​S1 and S2 corresponding to the overall signature features are calculated. Based on these values, the similarity between the registered signature and the signature to be authenticated is calculated, and the authentication result is output. However, this patent application does not explain how the strokes are segmented, how the corresponding strokes are obtained, or how the stroke data and features are acquired. Furthermore, it relies primarily on deep learning and lacks interpretability and visualization of the stroke features.

[0004] These point-feature-based methods do not consider the characteristics of each stroke in the signature, nor the overall layout and structure of the signature. These features are precisely the main characteristics we are most familiar with and accept as distinguishing genuine handwriting from forgery. We might consider a forgery because two strokes are dissimilar, or we might consider it genuine because the two signatures are remarkably similar in overall structure. The aforementioned existing methods present challenges in determining the precision and accuracy of handwriting recognition and authentication. Summary of the Invention

[0005] This invention addresses the shortcomings of existing technologies, such as insufficient consideration of signature strokes and overall structure, lack of understanding of relationships between corresponding strokes, and absence of interpretability and visualization of stroke features. It proposes a DTW-based method to calculate signature similarity based on point features, including the similarity between strokes and the structural similarity of the entire signature. This improves the accuracy of the judgment.

[0006] This invention uses a DTW (Digital Transmission Weaving) method to calculate signature similarity based on point features, including the similarity between strokes and the structural similarity of the entire signature. By combining the similarities of these point features, local features, and global features, the authenticity of handwriting can be identified more accurately and effectively, solving the problems of low accuracy and low efficiency in electronic handwriting identification.

[0007] The technical solution of this invention to solve the above-mentioned technical problems is to propose an online authentication signature method that integrates point features, local features, and global features, including the following steps: performing time dynamic programming based on the point features of the sample signature and the signature to be tested to obtain the point feature similarity between the sample signature and the signature to be tested; calculating the local feature similarity between the sample signature and the signature to be tested based on LNPS features; calculating the global feature similarity between the sample signature and the signature to be tested based on Euclidean distance; calculating the comprehensive similarity between the sample signature and the signature to be tested based on the point feature similarity, local feature similarity, and global feature similarity; and authenticating the signature based on the comprehensive similarity.

[0008] The authentication of signatures based on comprehensive similarity specifically includes: calculating the comprehensive similarity between the sample signature and the signature to be tested by weighted averaging of the similarity of local features and global features; if the similarity is greater than the threshold, the signature to be tested is confirmed to be signed by the person who signed the sample signature, and the authentication is passed; otherwise, the authentication fails. The threshold is the maximum distance or minimum similarity between different sample signatures.

[0009] The system collects the x-coordinate, y-coordinate, pressure value, velocity, acceleration, angle, and angular velocity values ​​corresponding to the electronic signature strokes at all times. These values ​​at each time point are used as point features. Based on the angular velocity among these point features, the sample signature is segmented into several strokes. The angular velocities of all point features in the sample signature strokes are compared. If the angular velocity at a certain time is greater than a predetermined angle threshold, the point corresponding to that time is taken as the endpoint of the sample signature stroke. All endpoints of the sample signature strokes are obtained, and the sample signature is segmented into several strokes based on these endpoints. Time Dynamic Programming (DTW) is performed on the point features of the sample signature and the signature to be tested to obtain the point feature similarity between the sample signature and the signature to be tested. That is, based on the endpoints of the sample signature strokes and the correspondence between the two signature points, the endpoints of the signature to be tested corresponding to them are obtained. Thus, a number of strokes corresponding one-to-one between the sample signature and the signature to be tested are obtained based on the point feature similarity.

[0010] The LNPS feature extraction method can be used to calculate the local feature similarity between the sample signature and the signature to be tested. The trajectory length of the strokes is calculated based on the angular velocity of the signature strokes. The LNPS feature of the stroke is determined by the trajectory length. The norm of the difference between the LNPS features of the sample signature and the signature to be tested is then normalized using the maximum value to obtain the distance between the corresponding strokes. The local feature similarity between the two signatures is a weighted average of the distances between all corresponding strokes of the two signatures, with the weight being the total number of points corresponding to the strokes. Specifically:

[0011] Based on the angular velocities v of the signature stroke in the x and y directions at time t. x (t),v Y (t) Call the formula Calculate the trajectory length L(X) of the signature stroke segment X to be tested, according to the formula: Calculate the LNPS feature LNPS(X) of the signature stroke segment X. Then normalize the LNPS feature difference using the norm of the LNPS feature difference, according to the formula: Calculate the distance D(X, X') between the sample signature stroke X' and the corresponding test signature stroke X, where LNPS(X) is the LNPS feature of the test signature, LNPS(X') is the LNPS feature of the sample signature, the subscript 1 indicates the first norm, and max is the maximum value.

[0012] The global feature similarity between the sample signature and the signature under test is obtained by measuring the Euclidean distance between them. The strokes of the signature are replaced by their center points, and the Euclidean distance between any two strokes within the same signature is calculated. This determines the Euclidean distance between the sample signature and the signature under test, and the global feature similarity is determined based on this Euclidean distance. Specifically, this can be done as follows:

[0013] Obtain the center point m(X) and length L(X) of the trajectory X(t) of the signature stroke X. Calculate the Euclidean distance between any two strokes X and Y in the signature to be tested using the formula: L(X, Y) = ||m(X) - m(Y)||². Similarly, calculate the Euclidean distance L(X′, Y′) between corresponding two strokes of the sample signature using the formula: Calculate the Euclidean distance between the sampled signature and the signature to be tested. The global feature similarity between the two signatures is the weighted average of the Euclidean distances between them, according to the formula... Determine the similarity weights. Where L(x) and L(y) are the trajectory lengths of any two strokes X and Y in the signature to be tested, and L(X') and L(y') are the trajectory lengths of any two strokes X' and Y' in the sample signature.

[0014] Finally, a weighted average is used to calculate the combined similarity between the sampled signature and the signature to be tested, considering point features, local features, and global features. If this similarity is greater than a threshold, the signature to be tested is from the sampled signature holder; otherwise, it is a forged signature. The threshold can be defined as the maximum distance or minimum similarity between different sampled signatures during verification.

[0015] The present invention also claims protection for a computer-readable storage medium having a computer program stored thereon, which can be loaded and run by a processor to perform the above-described online signature authentication method.

[0016] The present invention also claims protection for an electronic device comprising: one or more processors; a memory; and one or more application programs stored in the memory and configured to be loaded and run by the one or more processors to perform the online signature authentication method described above.

[0017] This invention comprehensively considers point features, local features, and global features in a signature, resulting in more authentic and reliable authentication results than traditional online signature authentication systems. It improves the accuracy and precision of judgment in handwriting recognition and authentication processes. In addition, since local and global features are more interpretable and visual than point features, the system can more intuitively display the authentication results of the signature under test and provide visual interpretability like traditional manual authentication. Attached Figure Description

[0018] Figure 1 This invention provides an online authentication flowchart based on the overall features of signature strokes. Detailed Implementation

[0019] The following detailed description of the implementation of the present invention is provided with reference to the accompanying drawings and specific examples. The following embodiments are only used to more clearly illustrate the technical solutions of the present invention and should not be construed as limiting the scope of protection of this application. Based on the embodiments of the present invention.

[0020] like Figure 1This is a flowchart of the online authentication process based on the overall features of signature strokes according to the present invention. Specifically, it includes: collecting the horizontal and vertical coordinates, pressure values, and other features of the sample signature and the signature to be tested at each moment during the signing process; calculating the velocity, acceleration, angle, and angular velocity of each point in the signature strokes; and using these values ​​as point features. Because two adjacent strokes usually have different inclination angles, this results in a significant angular change at the endpoints connecting these two strokes, meaning the angular velocity of that point is significantly greater than that of other points. Therefore, the angular velocity with a significant angular change in the point features is used to segment the strokes. The sample signature is divided into several strokes based on the angular velocity. The angular velocity of each point in the sample signature strokes is compared. If the angular velocity of a point is greater than a predetermined angle threshold (the angle threshold is preset), then that point is taken as the endpoint of the sample signature stroke. All these endpoints are obtained, and the sample signature is divided into several strokes based on the endpoints.

[0021] Based on the point features of the sample signature and the signature to be tested, a Time Dynamic Programming (DTW) operation is performed to obtain the distance between the two signatures with respect to their point features. This yields the similarity between the sample signature and the signature to be tested based on their point features, as well as the point-to-point correspondence between the two signatures. Specifically, based on the endpoints of the strokes in the sample signature and the point-to-point correspondence between the sample signature and the signature to be tested, the endpoints of the corresponding strokes in the signature to be tested are obtained. Thus, the point-to-point similarity of several strokes in the sample signature and the signature to be tested is obtained. In other words, based on the point-to-point correspondence obtained by the DTW algorithm, the corresponding stroke endpoints are obtained, and the strokes that correspond one-to-one between the sample signature and the signature to be tested are determined based on these corresponding stroke endpoints.

[0022] The LNPS (Low-Number Signature Profile Normalized by Stroke Length) feature extraction method from "Recurrent Adaptation Networks for Online Signature Verification" can be used. Local features are extracted for each stroke of the signature to be tested, and the similarity of local features between the sample signature and the signature to be tested is calculated. Further details are as follows:

[0023] The trajectory of the signature stroke segment at each time step is obtained, the velocity of the signature stroke at each time step is calculated, and the trajectory length of the stroke segment is determined. Based on the instantaneous velocities in the x and y directions at any two time steps, a second-order equation is established on the trajectory of the stroke segment. The second-order equation is normalized with the trajectory length to obtain the LNPS feature of the stroke segment. The first norm of the difference between the LNPS features of the sample signature and the corresponding stroke of the test signature is normalized with the maximum value to obtain the distance between the corresponding strokes of the sample signature and the test signature. The weighted average of the distances of all corresponding strokes of the two signatures is used to obtain the local features of the two signatures.

[0024] Specifically, the trajectory of the signature stroke segment is obtained based on the x-coordinate and y-coordinate of all signatures at all times. The x-coordinate x(t) and y-coordinate y(t) of the signature stroke at time t are collected to obtain the trajectory of the signature stroke segment x(t) = (x(t), y(t)). T , 0≤t≤T, where the signature time of stroke X is time OT. Calculate the instantaneous velocity V(t) of the signature stroke at time t based on the velocities in the x and y axes at time t, V(t) = [v...]. x (t), v y (t)] T 0 ≤ t ≤ T; Differentiate the square roots of the velocities in the x and y directions using the formula: Calculate the trajectory length L(X) of the stroke X. Establish a second-order feature on the stroke trajectory X by integrating the velocities in the x and y directions at any two moments in the time interval 0-T: The values ​​in this feature represent the area projections of the signature strokes onto the planes in the x, xy, yx, and yy directions. Normalizing the second-order feature with the stroke trajectory length L(X) yields the LNPS feature of this stroke segment:

[0025]

[0026] Based on the above method, the LNPS features LNPS(X') and LNPS(X) of the sample signature and the signature to be tested are obtained. The distance between the corresponding stroke segments of the two signatures is the first norm of the LNPS feature difference. Then, the maximum value is used for normalization, i.e., the formula is called:

[0027]

[0028] Calculate the distance D(X, X') between the corresponding stroke segments of the sample signature and the signature to be tested. The local feature similarity between the two signatures is the weighted average of the corresponding stroke segments of the two signatures, with the weight being the number of points in the corresponding stroke segments.

[0029] The global features of the sampled signature and the signature to be tested are extracted. Specifically, the position of each stroke is replaced by its center point, and then the Euclidean distance between any two strokes is calculated. The similarity of the global features is determined based on the Euclidean distance.

[0030] Determine the center points m(X) and m(Y) of the trajectories X(t) and Y(t) of any two strokes X and Y in the signature to be tested, and use the formula: L(X,Y)=||m(X)-m(Y)||2 to calculate the Euclidean distance L(X,T) between the two strokes X and Y.

[0031] Similarly, the Euclidean distance between the two strokes X' and Y' corresponding to the sample signature is calculated using the formula above. Therefore, the formula is called: Calculate the Euclidean distance between corresponding stroke segments of the sample signature and the signature to be tested. Complete the calculation of the Euclidean distance for all stroke segments of both the sample signature and the signature to be tested. The weighted average of the Euclidean distances for all stroke segments of both signatures yields the global features of the two signatures, where the final weights are: L(X) and L(y) are the trajectory lengths of any two strokes in the signature to be tested, and L(X') and L(y') are the trajectory lengths of any two strokes in the sample signature.

[0032] Finally, a weighted average of the point features, local features, and global features is used to calculate the comprehensive similarity between the sampled signature and the signature to be tested. If this similarity is greater than a threshold, the signature to be tested is determined to have been signed by the person who signed the sampled signature; otherwise, it is determined to have been signed by someone other than the person who signed the sampled signature, thus completing the authentication. The threshold is determined based on the maximum distance or minimum similarity between different sampled signatures during the verification process.

[0033] This invention provides an online signature authentication method that integrates point features, local features, and global features. It performs time dynamic programming based on the point features of a sample signature and a signature to be tested to obtain the point feature similarity between the two signatures. It then calculates the local feature similarity based on LNPS features, and the global feature similarity based on Euclidean distance. Finally, it calculates the comprehensive similarity between the sample signature and the signature based on the point feature similarity, local feature similarity, and global feature similarity, and authenticates the signature based on this comprehensive similarity. This invention provides more authentic and reliable authentication results, improving the accuracy and precision of handwriting recognition and authentication processes. This authentication method is implemented by a computer program that can be loaded and run by a processor to execute the online signature authentication method described above. It can also be used as an electronic device comprising multiple processors, memory, and application programs, configured to be loaded and run by one or more processors to execute the online signature authentication method described above.

Claims

1. A method for electronic signature handwriting authentication that integrates point features, local features, and global features, characterized in that, The process includes the following steps: Performing time dynamic programming based on the point features of the sample signature and the signature to be tested to obtain the point feature similarity between the two signatures; calculating the local feature similarity between the sample signature and the signature to be tested based on the signature contour feature LNPS normalized by stroke length; calculating the global feature similarity between the sample signature and the signature to be tested based on Euclidean distance, specifically by obtaining the center point of the signature strokes, calculating the Euclidean distance between any two strokes within the same signature based on the center point, normalizing the first norm of the difference in Euclidean distances between corresponding strokes of the two signatures using the maximum value to obtain the Euclidean distance between corresponding strokes of the two signatures, and taking a weighted average of the Euclidean distances of all corresponding strokes of the two signatures to obtain the global feature similarity between the two signatures; calculating the comprehensive similarity between the sample signature and the signature to be tested based on the point feature similarity, local feature similarity, and global feature similarity; and authenticating the signature based on the comprehensive similarity.

2. The method according to claim 1, characterized in that, The authentication of signatures based on comprehensive similarity specifically includes: calculating the comprehensive similarity between the sample signature and the signature to be tested by weighted averaging of the similarity of local features and global features; if the similarity is greater than the threshold, the signature to be tested is confirmed to be signed by the person who signed the sample signature, and the authentication is passed; otherwise, the authentication fails. The threshold is the maximum distance or minimum similarity between different sample signatures.

3. The method according to claim 1 or 2, characterized in that, The point features are the horizontal coordinate, vertical coordinate, pressure value, velocity, acceleration, angle, and angular velocity value corresponding to each moment in the strokes of the electronic signature. The angular velocity of all point features of the sample signature strokes is compared. If the angular velocity at a certain moment is greater than a predetermined angle threshold, the point corresponding to that moment is taken as the endpoint of the sample signature stroke. The sample signature is divided into several strokes according to the endpoints. Based on the endpoints of the sample signature strokes and the correspondence between the points of the sample signature and the signature to be tested, the endpoints corresponding to the signature to be tested are obtained. The similarity of several strokes corresponding to the sample signature and the signature to be tested based on point features is obtained.

4. The method according to claim 1 or 2, characterized in that, The calculation of local feature similarity between the sampled signature and the test signature based on signature contour features normalized by stroke length (LNPS) further includes: calculating the trajectory length of the strokes based on the angular velocity of the signature strokes; determining the features of the strokes based on the trajectory length; normalizing the norm of the feature difference between the sampled signature and the test signature using the maximum value to obtain the distance between the corresponding strokes of the two signatures; and obtaining the local feature similarity between the two signatures by weighted averaging of the distances between all corresponding strokes of the two signatures, with the weight being the total number of points corresponding to the strokes.

5. The method according to claim 4, characterized in that, Based on the velocity v of the signature stroke to be tested in the x and y directions at time t. x (t),v Y (t) Call the formula Calculate the trajectory length L(X) of the signature stroke segment X to be tested, based on the second-order feature formed by the area projection of the signature strokes onto x, xy, yx, yy. Call the formula: Calculate the LNPS feature LNPS(X) of the signature stroke segment X to be tested, and calculate the LNPS feature LNPS(X') of the sampled signature stroke segment X' using the same method, according to the formula: The distance D(X,X') between the corresponding strokes of the two signatures is obtained by normalizing the first norm of the difference between the LNPS features of the two signatures with the maximum value. In the formula, LNPS(X) is the LNPS feature of the signature to be tested, LNPS(X') is the LNPS feature of the sampled signature, the subscript 1 indicates the first norm, and max is the maximum value.

6. The method according to claim 5, characterized in that, To obtain the center points m(x) and m(Y) of the X and Y trajectories of any two strokes in the signature to be tested, call the formula: Calculate the Euclidean distance between the X and Y segments of the signature to be tested, and calculate the Euclidean distance between corresponding two strokes of the sample signature using the same method. Call the formula: The Euclidean distance between corresponding strokes of two signatures is obtained by normalizing the norm of the Euclidean distance difference between the two signatures using the maximum value. The weighted average of the Euclidean distances between all corresponding strokes of the two signatures is the global feature similarity between the two signatures, where the similarity weights are... L(x) and L(y) are the trajectory lengths of any two strokes X and Y in the signature to be tested, and L(X') and L(y') are the trajectory lengths of any two strokes X' and Y' in the sample signature.

7. A computer-readable storage medium, characterized in that, It stores a computer program that can be loaded and run by a processor to perform the electronic signature handwriting authentication method according to any one of claims 1 to 6.

8. An electronic device, characterized in that, The electronic device includes: one or more processors; a memory; and one or more applications stored in the memory and configured to be loaded and run by the one or more processors to perform the electronic signature handwriting authentication method according to any one of claims 1 to 6.

Images