Online signature authentication method and related equipment based on deep learning and matching matrix

Abstract

This invention discloses an online signature authentication method and related equipment based on deep learning and matching matrices, comprising: acquiring time-series data containing signature information from multiple devices in multiple batches; segmenting the time-series data of the signature into several strokes according to the stroke start and end points; processing the strokes to extract algebraic features through a path mapping algorithm, and processing the strokes to extract manual features through manual operators; obtaining shallow features representing signature habits based on algebraic and manual features; compensating or truncating the shallow features of the sample signature and the shallow features of the signature to be verified to standard lengths, and calculating the corresponding matching matrix using matrix multiplication; processing the matching matrix using an attention module to obtain a spatial attention map; using the spatial attention map as input to a residual network to extract feature vectors representing user identity; concatenating the feature vectors of the sample signature and the signature to be verified and then feeding them into a fully connected network to output the authentication result of the signature to be verified.

Description

Technical Field

[0001] This invention relates to the field of biometric recognition technology, specifically to online signature authentication methods, and particularly to an online signature authentication method and related equipment based on deep learning and matching matrices. Background Technology

[0002] Handwritten signature authentication is a traditional biometric identification technology that has been used for some time. It primarily uses the signer's handwriting habits to authenticate their identity. Because each person's handwriting habits are influenced by factors such as intelligence, gender, personality, and education, signatures are unique. Therefore, this method can determine the writer's identity through details of the signature, such as cursive writing habits, pen strokes, and the fluency of the strokes.

[0003] Currently, the mainstream signature authentication methods include: 1. Offline signature authentication methods: Unlike online signature authentication methods, offline signature authentication methods convert the signature into an image and then use subsequent algorithms for identity authentication. The subsequent authentication algorithms can be broadly divided into traditional methods and deep learning-based methods. 1) Offline signature authentication methods based on traditional methods: This method does not require purchasing specialized input devices to obtain information such as the writer's writing speed and pressure. Therefore, it has a wide range of applications. Also, because this method is based on traditional signature authentication, the authentication speed is relatively fast, meeting the requirement of high efficiency in signature authentication. For example, Zhang Yongliang et al. designed an authentication algorithm using Gabor filters and support vector machines in Chinese invention patent CN115311746A. However, using images as input will result in the loss of many detailed signature information, making it difficult to judge finely imitated signatures. 2) Offline signature authentication algorithms based on deep learning: Since the input of this method is still an image, it still has the advantages and disadvantages of method 1). However, unlike this method, this technical route uses deep learning methods for subsequent signature authentication. For example, Zhang Yan et al. used a Siamese network combined with cross-entropy loss to design an authentication algorithm in Chinese invention patent CN114155613A, and Zhang Xiaorui et al. used an EfficientNet network combined with triplet loss in Chinese invention patent CN113269136A. These algorithms have good generalization performance when there is sufficient training data. However, they require multiple signatures of the person themselves to be input into the system before subsequent authentication, which is a relatively cumbersome process. 2. Online signature authentication method: This method obtains information such as the coordinates of the signature, writing pressure, and timestamp through sensors, and uses subsequent algorithms for identity authentication. Subsequent authentication algorithms can be roughly divided into distance-based methods and machine learning-based methods. 1) Distance-based online signature authentication method: This method determines whether two signatures were written by the same person by calculating the distance between the sampled signature sequence and the signature sequence to be verified. For example, Shen Zhonghao et al. used the DTW algorithm combined with Euclidean distance metric to design a signature authentication algorithm in Chinese invention patent CN114333077A. This type of method has a very fast authentication speed. However, because the threshold for judging whether two signatures are similar needs to be set manually, and the threshold value varies in different authentication scenarios, the algorithm's generalization ability is poor. 2) Online signature authentication method based on machine learning: This method uses traditional machine learning algorithms for signature authentication. For example, Shen Zhonghao et al. designed a signature authentication algorithm using a human muscle movement model in CN115171223A. This type of method has a high authentication accuracy. However, it requires multiple signatures of the user to be input into the system before subsequent authentication, which is a cumbersome process.

[0004] Handwritten signatures are a widely used method of identity authentication, favored in formal settings such as banks, government agencies, and legal departments due to their convenience and effectiveness. However, relying on visual verification of a signature is both inefficient and inaccurate. Signature authentication methods based on traditional machine learning and modern deep learning algorithms require multiple signatures from the signatory to be pre-entered into the system for subsequent authentication. This process is cumbersome and contradicts the principle of simplicity and efficiency in signature authentication. Furthermore, signatures are susceptible to forgery. Therefore, efficiently and accurately determining signature authenticity has become a pressing issue in the field of signature authentication. Summary of the Invention

[0005] To address the shortcomings of existing technologies, this invention provides an online signature authentication method and related equipment based on deep learning and matching matrices, which can improve the efficiency and accuracy of online signature authentication.

[0006] To achieve the above objectives, the present invention provides the following technical solution:

[0007] In a first aspect, the present invention provides an online signature authentication method based on deep learning and a matching matrix, comprising:

[0008] Acquire time-series data containing signatures from multiple devices and in multiple batches;

[0009] The time-series data of the signature is divided into several strokes based on the starting and ending points of the strokes.

[0010] The strokes are processed by a path mapping algorithm to extract algebraic features, and the strokes are processed by a manual operator to extract manual features. Based on the algebraic features and the manual features, shallow features representing signature habits are obtained.

[0011] The shallow features of the sample signature and the shallow features of the signature to be verified are compensated or truncated to standard length, and the corresponding matching matrix is ​​calculated by matrix multiplication.

[0012] The matching matrix is ​​processed using an attention module to obtain a spatial attention map;

[0013] The spatial attention map is used as an input residual network to extract feature vectors representing user identity.

[0014] The feature vector of the sample signature and the feature vector of the signature to be verified are concatenated and then fed into a fully connected network to output the authentication result of the signature to be verified.

[0015] Secondly, the present invention provides an online signature authentication system based on deep learning and matching matrices, comprising:

[0016] The signature acquisition unit is used to acquire time-series data containing signatures from multiple devices and in multiple batches.

[0017] The computational processing unit is used to perform:

[0018] The time-series data of the signature is divided into several strokes based on the starting and ending points of the strokes.

[0019] The strokes are processed by a path mapping algorithm to extract algebraic features, and the strokes are processed by a manual operator to extract manual features. Based on the algebraic features and the manual features, shallow features representing signature habits are obtained.

[0020] The shallow features of the sample signature and the shallow features of the signature to be verified are compensated or truncated to standard length, and the corresponding matching matrix is ​​calculated by matrix multiplication.

[0021] The matching matrix is ​​processed using an attention module to obtain a spatial attention map;

[0022] The spatial attention map is used as an input residual network to extract feature vectors representing user identity.

[0023] The authentication output unit is used to concatenate the feature vector of the sampled signature and the feature vector of the signature to be verified, and then send them into the fully connected network to output the authentication result of the signature to be verified.

[0024] Thirdly, the present invention also provides an electronic device, including a processor and a memory;

[0025] The memory is used to store programs;

[0026] The processor executes the program to implement the method described above.

[0027] Fourthly, the present invention also provides a computer-readable storage medium storing a program that is executed by a processor to implement the methods described above.

[0028] Fifthly, the present invention also provides a computer program product or computer program comprising computer instructions stored in a computer-readable storage medium. A processor of a computer device can read the computer instructions from the computer-readable storage medium and execute the computer instructions to cause the computer device to perform the preceding method.

[0029] Compared with the prior art, the advantages of this invention are as follows:

[0030] In the stroke segmentation algorithm of the signature sequence, compared with the previous signature authentication method that connects all points in the signature sequence end to end as a single stroke, this invention first uses the distance change between continuous points, the time change between continuous points, and the discrete curvature C between continuous points to find the starting point and ending point of the stroke before extracting the writing habits, and then segments the signature into several strokes, creating the possibility for the algorithm to extract the writer's continuous writing habits and stroke information in the subsequent process.

[0031] In the method for extracting shallow features of signatures based on path mapping, compared with existing algorithms that rely solely on manual design to extract shallow features of signatures, this method abandons human priors and uses a more scientific and reasonable approach to extract shallow features of signatures.

[0032] Existing algorithms for similarity measurement using matching matrices concatenate the sample signature features and the signature to be verified before inputting them into the authentication header. This requires the authentication header to first separate the concatenated features and then calculate the similarity. This additional separation operation not only increases computational cost but also affects algorithm stability and weakens the model's generalization ability. This invention, however, utilizes matching matrices to calculate the similarity between the sample signature and the signature to be verified. This process transforms online signature authentication from a one-dimensional sequence similarity measurement problem into a two-dimensional image classification problem. This not only aids the authentication header in authentication but also bridges the gap between basic visual tasks and online signature authentication tasks, accelerating the development of this field. Attached Figure Description

[0033] To more clearly illustrate the technical solutions in the embodiments of the present invention, the accompanying drawings used in the embodiments will be briefly introduced below. Obviously, the drawings described below are only some embodiments of this application. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0034] Figure 1 This is a flowchart of the online signature authentication method in an embodiment of the present invention;

[0035] Figure 2 This is another flowchart of the online signature authentication method in this embodiment of the invention;

[0036] Figure 3 This is a schematic diagram of the signature segmentation result in an embodiment of the present invention;

[0037] Figure 4 This is a schematic diagram illustrating the process of calculating the matching matrix in an embodiment of the present invention;

[0038] Figure 5 This is a schematic diagram of the system according to an embodiment of the present invention;

[0039] Figure 6This is a schematic diagram of an electronic device according to an embodiment of the present invention. Detailed Implementation

[0040] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of this application, and not all embodiments. Based on the embodiments of this application, all other embodiments obtained by those of ordinary skill in the art without creative effort are within the scope of protection of this application.

[0041] Example:

[0042] It should be noted that the terms "first," "second," etc., in the specification, claims, and accompanying drawings of this invention are used to distinguish similar objects and are not necessarily used to describe a specific order or sequence. It should be understood that such data can be interchanged where appropriate so that the embodiments of the invention described herein can be implemented in orders other than those illustrated or described herein. Furthermore, the terms "comprising" and "having," and any variations thereof, in the embodiments of this invention are intended to cover non-exclusive inclusion; for example, a process, method, system, product, or apparatus that comprises a series of steps or units is not necessarily limited to those steps or units explicitly listed, but may include other steps or units not explicitly listed or inherent to such processes, methods, products, or apparatus.

[0043] See Figures 1 to 4 An online signature authentication method based on deep learning and matching matrices may specifically include the following steps:

[0044] Step 101: Obtain signed time-series data from multiple devices and in multiple batches.

[0045] In this step, for example, a WACOM electronic pen can be used to obtain the first batch of time-series data about the signature at a sampling frequency of 100Hz. A tablet can be used to obtain the second batch of time-series data about the signature. A mobile phone can be used to obtain the third batch of time-series data about the signature. It should be noted that each batch of data needs to include information such as the coordinates of key points, timestamps, and writing pressure. The purpose of using three different acquisition devices simultaneously is to enhance the model's generalization ability to adapt to different application scenarios.

[0046] Step 102: Divide the time-series data of the signature into several strokes based on the starting and ending points of the stroke.

[0047] In this step, some online signature authentication datasets do not provide information on the position of the starting and ending points of the strokes in the signature, causing all points in the signature to appear as if completed in one stroke. Therefore, this embodiment designs a stroke segmentation module to find the starting and ending points of the strokes in the signature and divide the signature into several strokes accordingly. This embodiment uses three attributes to distinguish the starting and ending points of the strokes, specifically:

[0048] (1) Distance change V between consecutive points d Because the speed cannot change abruptly, the distance difference between consecutive points within the same time interval is within a fixed range.

[0049] (2) Time variation V between consecutive points t Because the sampling frequency of the signature is fixed, the time interval between two consecutive points should be within a fixed range.

[0050] (3) Discrete curvature C between continuous points, which is an additional attribute for determining the starting point and ending point of the pen.

[0051] In this embodiment, parameters K and U are multiplied by the first and second attributes respectively, and then the attributes are added together to obtain the fused result F, which is used for preliminary judgment. Then, the embodiment uses the third attribute for final judgment. d V t The expressions for F are as follows:

[0052]

[0053]

[0054] F = K * V d +U*V t

[0055] Where P t and T t Let represent the point and timestamp at time t, respectively; f represents the sampling frequency; and K and U are hyperparameters. If F is greater than a preset threshold TH1, the embodiment will set P... t and P t-1 As candidate points for the starting and ending points of the stroke, P is then calculated. t-1 ,P t and P t+1 The discrete curvature of the curve composed of three points is used for further verification. If the curvature is less than the preset value TH2, then P t and P t-1 These are the starting point and the ending point of the stroke, respectively; the reverse is not true.

[0056] Experiments show that the best results are achieved when K = 0.8, U = 1.2, TH1 = 5, and TH2 = 0.0001. The corresponding segmentation results are as follows: Figure 3 As shown. The signature on the left is a visualization of the result of connecting all the dots end to end to form a single stroke.

[0057] Step 103: Process the strokes to extract algebraic features using a path mapping algorithm, and process the strokes to extract manual features using manual operators. Based on the algebraic features and the manual features, obtain shallow features representing signature habits.

[0058] In this step, based on the previous step, the example divides the signature S into n strokes, S = {s1, s2, ..., sn}. n}, each stroke s i It can be viewed as a continuous path X i Assuming the path starts from T is From time T ie When the time ends, then for path X i Perform k th The order integral can be expressed by the following formula.

[0059]

[0060] Through a sliding window of size ω on path X i Swipe up to calculate the time span X i The path mapping PS(X) on (t) i By concatenating the values ​​at each time span, we can obtain path X. i In k th The result of the integral is shown in the following formula:

[0061] I k (X i ) = [I k (X i (1)) I k (X i (2)) … I k (X i (T))]

[0062] Theoretically, path signatures are a set of infinite-order integrals. To enable computation, the embodiment truncates the path signatures to the m-th order as the final result. To ensure that the computation results at each order are of the same order, the embodiment takes X... i The dual modulus of PS m (X i Normalization is performed, and its expression is shown below.

[0063]

[0064] The final path mapping algorithm is expressed as follows over the entire signature:

[0065]

[0066] The characteristics of handcrafted design specifically include the following variables:

[0067] The speed of key point P

[0068] acceleration of key point P

[0069] The path tangent angle θ of key point P p The calculation formula is as follows:

[0070]

[0071] The velocity modulus v of key point P p The calculation formula is as follows:

[0072]

[0073] Path tangent angle θ p First derivative

[0074] velocity modulus v p First derivative

[0075] The logarithmic radius of curvature ρ of key point P p (Log curvature radius), its calculation formula is as follows:

[0076]

[0077] Total acceleration modulus a p (Total acceleration magnitude), its calculation formula is as follows:

[0078]

[0079] Count the radius of curvature ρ p The first derivative;

[0080] Total acceleration modulus a p The first derivative.

[0081] Step 104: Compensate or truncate the shallow features of the sample signature and the shallow features of the signature to be verified to standard lengths, and calculate the corresponding matching matrix using matrix multiplication.

[0082] In this step, the lengths of the shallow features extracted from different signatures are usually inconsistent, making matrix multiplication impossible. Therefore, the embodiment compensates for or truncates the handcrafted features R of the sampled signature and the handcrafted features Q of the signature to be verified to a standard length L. s Then, the corresponding matching matrix M is calculated using matrix multiplication. The process of calculating the matching matrix is ​​as follows: Figure 4 As shown. Where r i q represents the i-th handcrafted feature in R. f This represents the j-th handcrafted feature in Q.

[0083] Step 105: Use the attention module to process the matching matrix to obtain a spatial attention map.

[0084] In this step, to allow the backbone network to focus more on the effective regions in the matching matrix features, the embodiment uses a Convolutional Block Attention module (CBAM). CBAM is a simple and effective attention module that can infer attention weights sequentially along both spatial and channel dimensions.

[0085] In the channel attention module (CAM), the embodiment uses average pooling and max pooling to fuse the spatial information of the matching matrix features, allowing the pooled features to propagate forward to the fully connected layer to generate the channel attention map. Channel attention can be calculated using the following formula:

[0086]

[0087] Where σ represents the sigmoid activation function.

[0088] Unlike channel attention, spatial attention modules aim to find regions with high information density, which is crucial for determining the similarity of two signatures. The implementation uses average pooling and max pooling to fuse the channel information of the matching matrix features, concatenates them, and then feeds them into a standard convolutional module to obtain a spatial attention map. Spatial attention can be calculated using the following formula.

[0089]

[0090] In the formula, σ represents the sigmoid activation function, and conv represents the convolution operation.

[0091] Step 106: Extract feature vectors representing user identity using the spatial attention map input to the residual network.

[0092] Step 107: Concatenate the feature vector of the sample signature and the feature vector of the signature to be verified, and then send them into the fully connected network to output the authentication result of the signature to be verified.

[0093] The beneficial effects of this invention are as follows:

[0094] In the stroke segmentation algorithm of the signature sequence, compared with the previous signature authentication method that connects all points in the signature sequence end to end as a single stroke, this invention first uses the distance change between continuous points, the time change between continuous points, and the discrete curvature C between continuous points to find the starting point and ending point of the stroke before extracting the writing habits, and then segments the signature into several strokes, creating the possibility for the algorithm to extract the writer's continuous writing habits and stroke information in the subsequent process.

[0095] In the method for extracting shallow features of signatures based on path mapping, compared with existing algorithms that rely solely on manual design to extract shallow features of signatures, this method abandons human priors and uses a more scientific and reasonable approach to extract shallow features of signatures.

[0096] Existing algorithms for similarity measurement using matching matrices concatenate the sample signature features and the signature to be verified before inputting them into the authentication header. This requires the authentication header to first separate the concatenated features and then calculate the similarity. This additional separation operation not only increases computational cost but also affects algorithm stability and weakens the model's generalization ability. This invention, however, utilizes matching matrices to calculate the similarity between the sample signature and the signature to be verified. This process transforms online signature authentication from a one-dimensional sequence similarity measurement problem into a two-dimensional image classification problem. This not only aids the authentication header in authentication but also bridges the gap between basic visual tasks and online signature authentication tasks, accelerating the development of this field.

[0097] See Figure 5 Based on the same inventive concept, embodiments of the present invention also provide an online signature authentication system based on deep learning and a matching matrix, comprising:

[0098] The signature acquisition unit is used to acquire time-series data containing signatures from multiple devices and in multiple batches.

[0099] The computational processing unit is used to perform:

[0100] The time-series data of the signature is divided into several strokes based on the starting and ending points of the strokes.

[0101] The strokes are processed by a path mapping algorithm to extract algebraic features, and the strokes are processed by a manual operator to extract manual features. Based on the algebraic features and the manual features, shallow features representing signature habits are obtained.

[0102] The shallow features of the sample signature and the shallow features of the signature to be verified are compensated or truncated to standard length, and the corresponding matching matrix is ​​calculated by matrix multiplication.

[0103] The matching matrix is ​​processed using an attention module to obtain a spatial attention map;

[0104] The spatial attention map is used as an input residual network to extract feature vectors representing user identity.

[0105] The authentication output unit is used to concatenate the feature vector of the sampled signature and the feature vector of the signature to be verified, and then send them into the fully connected network to output the authentication result of the signature to be verified.

[0106] Since this system corresponds to the online signature authentication method based on deep learning and matching matrix in this embodiment of the invention, and the principle of solving the problem in this system is similar to that of this method, the implementation of this system can refer to the implementation process of the above method embodiment, and the repeated parts will not be described again.

[0107] See Figure 6 Based on the same inventive concept, embodiments of the present invention also provide an electronic device, the electronic device including a processor and a memory, the memory storing at least one instruction, at least one program, code set or instruction set, the at least one instruction, the at least one program, the code set or instruction set being loaded and executed by the processor to implement the online signature authentication method based on deep learning and matching matrix as described above.

[0108] It is understood that the memory may include random access memory (RAM) or read-only memory. Optionally, the memory may include non-transitory computer-readable storage medium. The memory can be used to store instructions, programs, code, code sets, or instruction sets. The memory may include a stored program area and a stored data area, wherein the stored program area may store instructions for implementing an operating system, instructions for at least one function, instructions for implementing the various method embodiments described above, etc.; the stored data area may store data created according to the use of the server, etc.

[0109] A processor may include one or more processing cores. The processor connects to various parts of the server via various interfaces and lines, executing instructions, programs, code sets, or instruction sets stored in memory, and accessing data stored in memory to perform various server functions and process data. Optionally, the processor may be implemented using at least one of the following hardware forms: Digital Signal Processing (DSP), Field-Programmable Gate Array (FPGA), and Programmable Logic Array (PLA). The processor may integrate one or more of the following: Central Processing Unit (CPU) and Modem. The CPU primarily handles the operating system and applications; the modem handles wireless communication. It is understood that the modem may also be implemented as a separate chip without being integrated into the processor.

[0110] Since this electronic device is the electronic device corresponding to the online signature authentication method based on deep learning and matching matrix in the embodiments of the present invention, and the principle of solving the problem by this electronic device is similar to that of this method, the implementation of this electronic device can refer to the implementation process of the above method embodiments, and the repeated parts will not be described again.

[0111] Based on the same inventive concept, embodiments of the present invention also provide a computer-readable storage medium storing at least one instruction, at least one program, code set, or instruction set, wherein the at least one instruction, the at least one program, the code set, or the instruction set is loaded and executed by a processor to implement the online signature authentication method based on deep learning and matching matrix as described above.

[0112] Those skilled in the art will understand that all or part of the steps in the various methods of the above embodiments can be implemented by a program instructing related hardware. The program can be stored in a computer-readable storage medium, including read-only memory (ROM), random access memory (RAM), programmable read-only memory (PROM), erasable programmable read-only memory (EPROM), one-time programmable read-only memory (OTPROM), electrically-erasable programmable read-only memory (EEPROM), compact disc read-only memory (CD-ROM) or other optical disc storage, disk storage, magnetic tape storage, or any other computer-readable medium capable of carrying or storing data.

[0113] Since this storage medium is the storage medium corresponding to the online signature authentication method based on deep learning and matching matrix in the embodiments of the present invention, and the principle of solving the problem by this storage medium is similar to that of this method, the implementation of this storage medium can refer to the implementation process of the above method embodiments, and the repeated parts will not be described again.

[0114] In some possible implementations, various aspects of the methods of the embodiments of the present invention can also be implemented as a program product comprising program code that, when run on a computer device, causes the computer device to perform the steps of the online signature authentication method based on deep learning and matching matrices according to various exemplary embodiments of the present application described above. The executable computer program code or "code" for performing the various embodiments can be written in high-level programming languages ​​such as C, C++, C#, Smalltalk, Java, JavaScript, Visual Basic, Structured Query Language (e.g., Transact-SQL), Perl, or in various other programming languages.

[0115] It should be understood that various parts of the present invention can be implemented in hardware, software, firmware, or a combination thereof. In the above embodiments, multiple steps or methods can be implemented in software or firmware stored in memory and executed by a suitable instruction execution system. For example, if implemented in hardware, as in another embodiment, it can be implemented using any one or a combination of the following techniques known in the art: discrete logic circuits having logic gates for implementing logical functions on data signals, application-specific integrated circuits (ASICs) having suitable combinational logic gates, programmable gate arrays (PGAs), field-programmable gate arrays (FPGAs), etc.

[0116] In the description of this specification, the references to terms such as "one embodiment," "some embodiments," "example," "specific example," or "some examples," etc., indicate that a specific feature, structure, material, or characteristic described in connection with that embodiment or example is included in at least one embodiment or example of the present invention. In this specification, the illustrative expressions of the above terms do not necessarily refer to the same embodiment or example. Furthermore, the specific features, structures, materials, or characteristics described may be combined in any suitable manner in one or more embodiments or examples. Moreover, without contradiction, those skilled in the art can combine and integrate the different embodiments or examples described in this specification, as well as the features of different embodiments or examples.

[0117] The above embodiments are merely illustrative of the technical concept and features of the present invention, and are intended to enable those skilled in the art to understand the content of the present invention and implement it accordingly. They should not be construed as limiting the scope of protection of the present invention. All equivalent changes or modifications made based on the essence of the content of the present invention should be covered within the scope of protection of the present invention.

Claims

1. An online signature authentication method based on deep learning and matching matrix, characterized in that, The method comprises the following steps: obtaining time series data of multiple devices and multiple batches containing signatures; segmenting the time series data of the signature into strokes according to the pen-down points and the pen-up points; extracting algebraic features of the strokes by a path mapping algorithm and extracting manual features of the strokes by a manual operator, and obtaining shallow features representing signature habits according to the algebraic features and the manual features; compensating or truncating the shallow features of the sample signature and the shallow features of the signature to be verified into a standard length, and calculating a corresponding matching matrix by matrix multiplication; processing the matching matrix by using an attention module to obtain a spatial attention map; inputting the spatial attention map into a residual network to extract a feature vector representing the identity of a user; concatenating the feature vector of the sample signature and the feature vector of the signature to be verified, and then inputting the concatenated feature vectors into a fully connected network to output a verification result of the signature to be verified; When processing the stroke-based algebraic features using the path mapping algorithm, the signature S is divided into n strokes, i.e., S = {s1, s2, ..., sn}. n }, each stroke s i It can be viewed as a continuous path X i Let the path start from T. is From time T ie When the time ends, then for path X i Perform k th The order integral is expressed as follows: By a sliding window of size ω over the path X i The path mapping PS(X i (t)) can be computed by sliding up the path X i The path mapping PS(X i The k th th integral is given by I k (X i )=[I k (X i (1))I k (X i (2))…I k (X i (T))]; truncating the path signature to the mth order as the final result; To make the results of each step in the same order of magnitude, take X i of the two modulus pair PS m (X i ) is normalized, the expression is as follows: the expression of the path mapping algorithm on the entire signature is as follows: When extracting handcrafted features from the strokes using handcrafted operators, the handcrafted features include the following variables: the velocity of keypoint P. acceleration of key point P The path tangent angle θ of key point P p The velocity modulus v of key point P p Path tangent angle θ p First derivative velocity modulus v p First derivative Logarithmic radius of curvature ρ of key point P: total acceleration modulus a p ; radius of curvature ρ p First derivative and total acceleration mode length a p The first derivative.

2. The online signature authentication method based on deep learning and matching matrix according to claim 1, characterized in that, The method comprises the following steps: obtaining time series data of multiple devices and multiple batches containing signatures; obtaining the first batch of time series data containing signatures at a sampling frequency of 100 Hz using a WACOM electronic pen; or / and, obtaining the second batch of time series data containing signatures using a tablet; 3.The online signature authentication method based on deep learning and matching matrix according to claim 1, characterized in that, or / and, obtaining the third batch of time series data containing signatures using a mobile phone. Let the distance between successive points vary by V d : Time variation V between successive points t : The method comprises the following steps: According to the distance change V between consecutive points d , the time change V between consecutive points t , the hyperparameters K and U are obtained according to a preset formula to obtain a fused result F, wherein, F = K * V d + U * V t where P t and T t represent the point and timestamp at time t, respectively, f represents the sampling frequency, and K and U are hyperparameters. When F is greater than the first preset threshold TH1, P t and P t-1 As candidate points for the starting and ending points of the stroke, P is then calculated. t-1 ,P t and P t+1 The discrete curvature of the curve composed of three points is further verified; when the discrete curvature is less than the second preset threshold TH2, then P t and P t-1 These are the starting point and the ending point of the stroke, respectively; the reverse is not true. segmenting the time series data of the signature into strokes according to the pen-down points and the pen-up points; Path tangent angle θ of key point P p The calculation formula is as follows: The velocity module v of the key point P p The calculation formula is as follows: Logarithmic radius of curvature p of key point P p The formula for calculating p is shown below: Total acceleration module length a p The calculation formula is as follows: 5.The online signature authentication method based on deep learning and matching matrix according to claim 1, characterized in that, discrete curvature C between consecutive points, 4. The online signature verification method based on deep learning and matching matrix according to claim 1, wherein The method comprises the following steps: The spatial information of the matching matrix features is fused by average pooling and max pooling through the channel attention module, and the pooled features are positively propagated to the full connection layer to generate a channel attention map The channel attention is calculated by the following formula: processing the features of the matching matrix by using an attention module to obtain a spatial attention map, which comprises the following steps: using a convolution block attention module to sequentially infer attention weights along the spatial and channel dimensions, wherein The channel information of the matching matrix features is fused by using average pooling and max pooling through the spatial attention module, and after being spliced, they are sent to a standard convolution module for convolution to obtain a spatial attention map The spatial attention is calculated by using the following formula: in the channel dimension, 6. An online signature authentication system based on deep learning and matching matrix, characterized by, wherein σ represents a sigmoid activation function; in the spatial dimension, wherein σ represents a sigmoid activation function, and conv represents a convolution operation. The method comprises the following steps: a signature acquisition unit configured to obtain time series data of multiple devices and multiple batches containing signatures; a calculation processing unit configured to perform the following steps: segmenting the time series data of the signature into strokes according to the pen-down points and the pen-up points; extracting algebraic features of the strokes by a path mapping algorithm and extracting manual features of the strokes by a manual operator, and obtaining shallow features representing signature habits according to the algebraic features and the manual features; compensating or truncating the shallow features of the sample signature and the shallow features of the signature to be verified into a standard length, and calculating a corresponding matching matrix by matrix multiplication; wherein, when processing the stroke extraction algebraic feature by the path mapping algorithm, the signature S is divided into n strokes, i.e. S = {s1, s2,... s n}, each stroke s i can be regarded as a continuous path X i , assuming that the path starts at time T is and ends at time T ie , then the k th order integral of the path X i is expressed by the following formula: By a sliding window of size ω over the path X i The path mapping PS(X i (t)) can be computed by sliding up the values over the time span X i and concatenating the values over each time span to obtain the path X i The k th th integral is computed as follows: I k (X i ) = [I k (X i (1))I k (X i (2))... I k (X i (T i (T))] ; processing the matching matrix by using an attention module to obtain a spatial attention map; To make the results of each step in the same order of magnitude, take X i of the two modulus pair PS m (X i ) is normalized, the expression is as follows: inputting the spatial attention map into a residual network to extract a feature vector representing the identity of a user; an authentication output unit configured to concatenate the feature vector of the sample signature and the feature vector of the signature to be verified, and then input the concatenated feature vectors into a fully connected network to output a verification result of the signature to be verified; truncating the path signature to the mth order as the final result; the expression of the path mapping algorithm on the entire signature is as follows: When the strokes are processed by a manual operator to extract the manual features, the manual features include the following variables: velocity of the key point P acceleration of the key point P path tangent angle θ of the key point P p ; velocity module length v of the key point P p ; first derivative of the path tangent angle θ p first derivative of the velocity module length v p logarithmic curvature radius p of the key point P p total acceleration module length a p ; first derivative of the logarithmic curvature radius p p and first derivative of the total acceleration module length a p .​​ 7. An electronic device, comprising: The electronic device comprises a processor and a memory, and the memory stores at least one instruction, at least one program, a code set or an instruction set, which are loaded and executed by the processor to implement the online signature authentication method based on deep learning and matching matrix as claimed in any one of claims 1 to 5.

8. A computer-readable storage medium, characterized in that, The storage medium stores at least one instruction, at least one program, a code set or an instruction set, which are loaded and executed by the processor to implement the online signature authentication method based on deep learning and matching matrix as claimed in any one of claims 1 to 5.

Images