Method, system and device for verifying user identity based on random signed electronic information and storage medium

Abstract

This invention proposes a method for verifying user identity based on randomly signed electronic information. The method includes: a sampling stage, where multi-dimensional handwritten features of any written text are acquired and reconstructed into a two-dimensional feature tensor, input into a decoupled writing content and handwriting style model, and the sampled handwriting style features are extracted and stored in a database; a verification stage, where the database generates verification text information, and multi-dimensional handwritten features of the written verification text information are acquired in real time and reconstructed into a two-dimensional feature tensor, input into the decoupled writing content and handwriting style model, and the verification handwriting style features and verification content features are obtained; the verification handwriting style features are compared with the sampled handwriting style features, and the verification content features are compared with the verification text content features to perform user identity verification. By applying multi-domain features of electronic handwriting to accurately obtain handwriting style features, achieving complete decoupling of writing content and handwriting style, and adding dynamic identity verification of pre-signed content, the security of user identity authentication is improved.

Description

Technical Field

[0001] This invention relates to the fields of artificial intelligence and information security technology, and in particular to a method for verifying user identity based on randomly signed electronic information. Background Technology

[0002] With the development of technology, electronic signature technology has been widely used as a means of personal identity verification and in various applications such as electronic document confirmation. Through specific equipment and technical means, handwritten signatures are converted into electronic form to replace traditional paper signatures or seals. Meanwhile, electronic handwriting recognition technology is also constantly evolving, capable of identifying users based on temporal information such as their writing habits, pressure, and speed. Furthermore, information security verification technology is also a crucial means of ensuring the security of electronic signatures. By verifying the content and handwriting style of the electronic signature signed by the user through algorithms or rules, signature fraud can be effectively prevented.

[0003] Existing electronic handwriting verification systems primarily rely on single-domain (single-modal) electronic signature handwriting features, such as the positional information of signature image data or the temporal information of signature sequence data. After these features are extracted, they are compared with pre-stored signature samples (signature samples) to verify the accuracy of the electronic signature content. Simultaneously, user writing habits are considered, adding verification of the electronic signature's handwriting style. However, the extracted feature information is relatively monotonous, resulting in weak generalization ability of the system in complex scenarios.

[0004] Existing electronic handwriting verification systems can only perform handwriting verification tasks for specific signed content, such as requiring the verified signature to match the sample signature. Because current technology typically only acquires handwriting style features tightly coupled with content characteristics, it can only verify the electronic signature itself, making it difficult to verify signatures written on other content. This content-specific verification method significantly reduces the randomness of the verified content, increasing the risk of forgery attacks and cracking, and failing to effectively prevent signature fraud. If a user's paper or electronic signature is leaked and forged to sign contracts or documents, the consequences could be incalculable.

[0005] Therefore, how to design a system that can perform dynamic electronic handwriting authentication through non-fixed verification content and improve the security of user identity authentication is an important problem currently facing electronic signature handwriting verification systems.

[0006] For example, the invention patent application CN118135382A, entitled "A Method for Locating Image Tampered Text Based on Feature Decoupling and Texture Comparison," discloses a method for locating image tampered text based on feature decoupling and texture comparison. This method decouples semantic features from texture features, avoiding overfitting to semantic features of text appearance unrelated to tampering during tampering detection training. It determines whether each text region is tampered by comparing the texture features of each text region with global features. This document judges whether tampering has occurred based on whether there are texture differences in different regions of the image, primarily determining whether the data format is normal or tampered. Unlike handwriting verification tasks, handwriting verification cannot be performed, and it is based on image data decoupling, involving text location information and texture information within the image, which is fundamentally different from handwriting verification.

[0007] Publication No.: CN116469176A, titled "A Text-Free Machine Handwriting Authentication Method, System, Device and Medium", discloses a text-free machine handwriting authentication method. It constructs an identification model based on a deep neural network and trains the identification model using handwriting time series data. The original time series is expanded and mapped to a high-dimensional space to decouple the content and style of the handwriting, capture global information of the handwriting time series, and enhance the learning of global features.

[0008] This method is designed for handwriting verification scenarios where the content is irrelevant, and it cannot determine whether the written content is correct. It is difficult to apply to signing scenarios, and it only uses the temporal features of the data. The information in the style features is limited, and the features obtained cannot fully reflect the writer's biometric information. Summary of the Invention

[0009] This invention addresses the aforementioned problems in existing technologies by applying multi-domain features of electronic handwriting to accurately obtain the user's handwriting style characteristics. It extracts and integrates richer multi-domain feature information from signature sequence data, achieving complete decoupling between written content and handwriting style. It adds dynamic identity verification of pre-signed and randomly signed content before signing electronic signatures, increasing the randomness and diversity of verification content, and enhancing the security of user identity authentication.

[0010] The main technical solutions of this invention include: designing a dynamic electronic handwriting authentication process with non-fixed verification content to increase the randomness of the verification content and reduce the risk of being imitated; designing a model that decouples the writing content and handwriting style so that the verification system can effectively verify random writing content.

[0011] Based on one aspect of this invention, a method for verifying user identity based on randomly signed electronic information is proposed, comprising: a sampling stage, acquiring multidimensional handwritten features when writing arbitrary text information and reshaping them into a two-dimensional feature tensor, inputting them into a decoupled writing content and handwriting style model, extracting the sampled handwriting style features and storing them in a database; and a verification stage, generating verification text information in the database, acquiring multidimensional handwritten features of the written verification text information in real time and reshaping them into a two-dimensional feature tensor, inputting them into a decoupled writing content and handwriting style model, acquiring verification handwriting style features and verification content features, comparing the verification handwriting style features with the sampled handwriting style features, and comparing the verification content features with the verification text content features. If the above comparison structures match, then the user identity is verified.

[0012] Further preferred, the acquisition of multi-dimensional handwriting features includes collecting data on handwriting points, pressure, writing speed, and pen start and stop states when a user writes text information using an electronic device, and calculating the horizontal coordinate X, vertical coordinate Y, and horizontal speed V of the sampling points. x Vertical velocity V y The path tangent angle θ, the cosine value of the path tangent angle cos(θ), the sine value sin(θ), the stroke pressure P, and the starting and ending pen states T are used as multidimensional handcrafted features, where θ = arctan(V y / V x ).

[0013] Further preferably, the reshaping into a two-dimensional feature tensor includes: applying Fast Fourier Transform (FFT) to obtain the frequency domain feature information of the multidimensional handcrafted feature Z, and calculating the period information in the frequency domain feature information through the amplitude: based on the sparsity of the frequency domain and its high-frequency noise, selecting the first k amplitude values, and calculating the period {p1,…,p} corresponding to the time domain in the frequency domain. k}, and reshape Z into a two-dimensional feature tensor Z' based on frequency and period.

[0014] Further preferably, the calculation of the periodic information in the frequency domain feature through amplitude includes calling the formula: A = Avg(Amp(FFT(Z))) using the frequency domain representation of the multidimensional feature Z and the average amplitude A of the frequency components; and calling the formula based on the average amplitude A of the frequency components: The frequencies {f1,…,f} are calculated. k}; Based on the frequencies {f1,…,f k Call the formula:

[0015]

[0016] Calculate the period {p1,…,p} in the frequency domain corresponding to the time domain. k}; Based on the frequencies {f1,…,f k} and period {p1,…,p k Call the formula:

[0017]

[0018] one-dimensional tensor Convert into k two-dimensional tensors It contains different periodic information of the sequence data; Padding is a zero-padding operation to ensure that the feature length is divisible by the period size; Reshape is a tensor reshaping operation.

[0019] Further optimization involves retrieving the standard features of the verification text from the database and comparing them with the verification content features. It also retrieves the user's handwriting style features from the sample and compares them with the verification handwriting style features. If the written content is correct and the handwriting style matches successfully, the system approves the verification request, and the user enters the electronic signature signing process.

[0020] Based on the second aspect of this application, a system for verifying user identity based on randomly signed electronic information is proposed. The system includes: a sample writing device acquiring multi-dimensional handwriting features when writing arbitrary text information; a feature processing module reshaping these features into a two-dimensional feature tensor; inputting this tensor into a decoupled writing content and handwriting style model; extracting the sample writing handwriting style features and storing them in a database; the database generating verification text information; the verification writing device acquiring multi-dimensional handwriting features of the verification text information in real time; the feature processing module reshaping these features into a two-dimensional feature tensor; inputting this tensor into a decoupled writing content and handwriting style model; obtaining verification handwriting style features and verification content features; and a verification unit comparing the verification handwriting style features with the sample writing handwriting style features and comparing the verification content features with the verification text content features generated by the database. If the above comparison structures match, the user identity is verified.

[0021] Further optimization reveals that the decoupled writing content and handwriting style model includes: a feature extraction backbone network, a content encoder, a style encoder, and a joint decoder. The feature extraction backbone network performs deep feature extraction on the multidimensional handcrafted features of the input to obtain a deep feature representation. The content encoder and style encoder are two independent deep neural network modules used to separate the content feature representation and style feature representation information from the electronic sequence data from the deep feature representation. The joint decoder is an independent deconvolutional, recurrent, or autoregressive neural network module used to combine the content feature representation and style feature representation and decode and restore the input signature data.

[0022] Further preferably, the feature processing module reshapes the multidimensional handcrafted features into a two-dimensional feature tensor by: applying Fast Fourier Transform (FFT) to obtain the frequency domain feature information of the multidimensional handcrafted feature Z, and calculating the period information in the frequency domain feature through the amplitude: based on the sparsity of the frequency domain and its high-frequency noise, selecting the first k amplitude values, and calculating the period {p1,…,p} corresponding to the time domain in the frequency domain. k}, and reshape Z into a two-dimensional feature tensor Z' based on frequency and period.

[0023] Further preferably, the generation of verification text information includes: obtaining real-time information of the electronic signature device requesting verification; converting the timestamp in the real-time information into a set of floating-point numbers; for each timestamp floating-point number, the Seed function combines the longitude, latitude, and altitude floating-point numbers in the real-time information to convert them into a random number; and using the converted random number as a data index to extract the corresponding Chinese characters from the database to form the verification text for this time.

[0024] Further optimization includes the following steps for completing user authentication: The verification unit retrieves the standard features corresponding to the verification text from the database and compares them with the verification content features; it also retrieves the user's handwriting style features from the sample and compares them with the verification handwriting style features. If the written content is correct and the handwriting style matches successfully, the system approves the verification request, and the user enters the electronic signature signing process. If the user fails to verify for a predetermined number of consecutive times during the authentication process, the account is locked, requiring the user to undergo auxiliary authentication, or a second signing verification is performed on the electronic signature content and handwriting style. If the signing verification fails more than a predetermined number of times, the user re-samples and re-enters the signing verification process.

[0025] This technical solution enhances the security, accuracy, and flexibility of electronic signatures. By verifying the content and handwriting style of dynamic, random short text information before the signing process, it increases the randomness and diversity of the verification content and completely decouples content and handwriting style, effectively preventing signature fraud and improving the security of electronic signatures. It improves accuracy by obtaining the user's handwriting style information based on multi-domain feature information, allowing for more accurate verification of the user's identity. It offers strong flexibility and adaptability; both verification content and methods can be adjusted according to security requirements, demonstrating high flexibility and adaptability. Attached Figure Description

[0026] Figure 1 This exemplary embodiment includes a flowchart of a dynamic electronic handwriting verification process that decouples writing content from handwriting style.

[0027] Figure 2 A schematic diagram of a model decoupling writing content and handwriting style in this exemplary embodiment;

[0028] Figure 3 This exemplary embodiment illustrates the time-domain and frequency-domain information of the time-series data, as well as a schematic diagram of the multi-domain reconstructed two-dimensional time-series data.

[0029] Figure 4 This is a structural block diagram of an exemplary electronic device that can be used to implement embodiments of this application. Detailed Implementation

[0030] This invention designs a model for extracting and fusing multi-domain features, as well as decoupling content and handwriting style. Temporal and frequency domain features are obtained from signature time-series data and fused together. The model includes a feature extraction backbone network, a content encoder, a style encoder, and a joint decoder. The feature extraction backbone network is used to extract deep feature representations of the data; the content encoder and style encoder extract content features and handwriting style features, respectively; the joint decoder reconstructs the input data based on the style and content features to train the model.

[0031] During the sampling phase, users write a short text message with content given by the system using an electronic signature device, collecting data such as writing points, pressure, writing speed, and pen lifting and lowering states. Multidimensional handwriting features are obtained and input into a model decoupled from the writing content and handwriting style. Handwriting style features are extracted and stored in a secure database for use in the verification phase.

[0032] The short text information can be randomly sampled by the system from the database, and it is rich and diverse, so that the system model can extract an accurate style feature representation of the user.

[0033] During the verification phase, information such as the signing time, location, and random number of the electronic signature device is obtained to generate verification text information with random content. The length and content of this text information can be adjusted according to security requirements. The user writes using the electronic signature device based on the verification text information provided by the system. The system collects the user's writing data in real time, including writing points, pressure, writing speed, and pen lifting and lowering states. Furthermore, based on the real-time collected writing data, handwriting features are calculated and input into a model decoupling the writing content and handwriting style to obtain the content features and handwriting style features of the verification data.

[0034] Furthermore, based on the content and handwriting style characteristics of the acquired verification data, the system determines whether the verification content is accurate and whether the verified handwriting style characteristics are consistent with those in the sampling stage. The system compares the verification content characteristics with the corresponding standard characteristics of the verification text in the database, and compares the verification handwriting style characteristics with those of the sampled handwriting in the database. If the written content is correct and the handwriting style matches successfully, the system verifies the user's identity, and the user proceeds to the electronic signature signing process. Otherwise, the system rejects the identity verification, and the user needs to verify their identity again before proceeding with the subsequent electronic signature signing.

[0035] Depending on the importance of the signing scenario, security alert policies and secondary verification policies can be set. These include triggering a security alert if pre-signing verification fails multiple times, and performing secondary verification of the content and handwriting style of the electronic signature during the signing process. This effectively prevents signature fraud and ensures the security of the verification process.

[0036] like Figure 1 The diagram shows the flowchart of the dynamic electronic handwriting verification process of the present invention, which decouples writing content and handwriting style. It includes: a sampling stage: the user samples a short text using a signature acquisition device to obtain multi-domain handwriting features; the user inputs the decoupled content and style model to extract the sampled handwriting style features; and the user stores the sampled handwriting style features in a database. In the verification stage, a random verification sample is generated in the database. The user writes based on the random verification sample, obtains the multi-domain handwriting features using the signature acquisition device, inputs the decoupled content and style model, verifies the writing content features and handwriting style features, and compares them with the sampled handwriting style features extracted from the database. If they pass, the user's identity verification is successful, and electronic signature signing can be implemented.

[0037] This invention designs a model that decouples writing content from handwriting style, enabling dynamic verification of electronic handwriting data with different writing content. The model includes a feature extraction backbone network, a content encoder, a style encoder, and a joint decoder. The feature extraction backbone network extracts deep feature representations of handwriting features; the content encoder and style encoder extract content features and handwriting style features from the deep feature representations, respectively; the joint decoder reconstructs the input signature data features based on style and content features. Furthermore, a method for extracting and fusing multi-domain handwriting features is proposed. This method obtains and fuses temporal and frequency domain feature information from signature time-series data, increasing the effective information in the features and thus improving the model's representational ability.

[0038] like Figure 2 The diagram illustrates a model for decoupling handwriting content and handwriting style in this exemplary embodiment. The model for decoupling handwriting content and handwriting style includes: a feature extraction backbone network, a content encoder, a style encoder, and a joint decoder. Specifically, ① the feature extraction backbone network performs deep feature extraction on the input handwriting features to obtain a deep feature representation of the handwriting data. For example, the system inputs a two-dimensional feature tensor Z' that integrates multi-domain information into a well-trained feature extraction backbone network to obtain a deep representation of the handwriting data. The backbone network can be a convolutional deep neural network such as ResNet or EfficientNet, or a self-attention-based deep neural network such as Transformer or Mamba.

[0039] ② The content encoder and style encoder are two independent deep convolutional neural network modules used to separate the content feature representation and style feature representation information of handwriting data from the depth representation. During the user sampling phase, the system inputs the depth representation of the sampled handwriting data into a well-trained handwriting style encoder to model the user's handwriting style and store its sampled handwriting style feature information in the database. During the user verification phase, the system inputs the depth representation of the verification handwriting data into both the well-trained handwriting style encoder and content encoder. The verification handwriting style representation is used to determine whether it is consistent with the sampled handwriting style representation in the database, and the verification handwriting content representation is used to determine whether it is consistent with the standard content representation in the database.

[0040] ③ The joint decoder is not limited to a single network structure and can be a deep deconvolutional neural network module, a recurrent neural network module, or an autoregressive network module. It combines content feature representations and style feature representations to decode and reconstruct the input handwriting feature data, primarily used in the model training process. The method for combining content feature representations and style feature representations is not limited to a single technique. It can involve concatenating content feature representations and style feature representations along feature dimensions to form a new representation vector; or using static weights or learning different weights through an attention mechanism to weight and fuse the content feature representations and style feature representations.

[0041] The invention will be further described in detail below with a specific embodiment. During the sampling stage, the user's handwriting style is modeled. During the verification stage, the dynamic verification data undergoes content verification and handwriting style verification. After successful verification, the user can proceed to the electronic signature signing process; otherwise, further identity verification is required. Furthermore, the system's verification strategy can be set according to the signing scenario, such as a security alarm strategy or a secondary verification strategy. Specific implementation steps include:

[0042] During the sampling phase, multi-domain features are collected and fused. Users first write a complete set of short text information (Text1, ..., Text2) using an electronic signature device, with content provided by the system. n This set of short texts was randomly sampled from the database by the system, possessing richness and diversity. For a single short text message (Text), the signing device synchronously collects the writing point (x) during the user's writing process. i ,y i ), stroke pressure p i Writing speed v i , Pen lifting and lowering state t i Data such as i, where i represents the i-th sampling point in the sequence.

[0043] Collect data and calculate the x-coordinate (X), y-coordinate (Y), and horizontal velocity (V) of the handwritten information sampling points. x Vertical velocity V yPath tangent angle θ = arctan(V) y / V x The set of handcrafted features includes multi-dimensional (e.g., 9-dimensional in this embodiment) features such as the cosine value of the path tangent angle (cos(θ)), the sine value (sin(θ)), the stroke pressure P, and the pen start and end states T. These handcrafted features are abbreviated as... Where L is the sequence length.

[0044] Furthermore, a Fast Fourier Transform (FFT) is applied to obtain the frequency domain representation of the handcrafted feature Z, and the average amplitude A of the frequency components is obtained through amplitude calculation (AMP) and mean calculation (AVG).

[0045] A = Avg(Amp(FFT(Z))),

[0046] In this process, AMP calculates the amplitude in the complex representation of the frequency domain, while AVG calculates the average amplitude of the frequency components of all features to obtain the overall frequency-amplitude matrix of the data. Due to the sparsity of the frequency domain and its relationship with high-frequency noise, only the first k amplitude values ​​with sufficient high-frequency noise distribution are selected to obtain the corresponding frequencies {f1,…,f...}. k}:

[0047]

[0048] Where k is a hyperparameter controlling the number of cycles, which can be set based on data distribution or prior knowledge. In this embodiment, k = 4; f * This represents the positive frequency component.

[0049] Based on the obtained frequencies {f1,…,f k The period {p1,…,p} corresponding to the frequency domain is calculated. k}:

[0050]

[0051] Where, p i f represents the i-th maximum frequency component. i The corresponding period size. Based on p i One-dimensional data of length L can be divided into f data containing periodic information. i ×p i Two-dimensional data.

[0052] Based on the above method, using frequencies {f1,…,f k} and period {p1,…,p k The feature Z is reshaped into k two-dimensional feature tensors Z', where each two-dimensional feature tensor Z' contains the original time-domain feature information and the newly added frequency-domain feature information:

[0053]

[0054] Padding is a zero-padding operation where the feature length is divisible by the period size; Reshape is a tensor reshaping operation that reshapes a one-dimensional tensor. Convert into k two-dimensional tensors Its two-dimensional tensor contains different periodic information of the sequence data.

[0055] like Figure 3 The diagram illustrates the temporal and frequency domain information of the time-series data and the multi-domain reconstruction of two-dimensional time-series data in this exemplary embodiment. This fusion method enables the features to contain two dimensions of feature information: locality between adjacent time points (intra-period variation) and locality between adjacent periods (period-period variation), resulting in rich temporal and frequency domain information in the handwriting features. In the handwriting verification task of this example, the hyperparameter k controlling the number of periods is set to 4. The four largest frequency components {f1, f2, f3, f4} and their corresponding period sizes {p1, p2, p3, p4} are obtained using the above method. Finally, the original one-dimensional handwritten temporal features are... Based on the size p of each cycle i Reshaped into four two-dimensional fusion features of different period sizes. By statistically analyzing the frequency components of the features, multiple periodic information of the data is obtained, and this periodic information is integrated into the original time-domain features, making it easier for the model to learn richer information in the frequency and time domains.

[0056] During the verification phase, a verification text with random content is first generated using the GPS positioning data built into the electronic signature device. The length of the verification text can be adjusted according to security requirements; in this embodiment, we set the length to 6 characters. Specifically, the real-time information of the requested verification time, such as longitude, latitude, altitude, and timestamp, is obtained from the GPS of the electronic signature device and converted into a set of random numbers; in this embodiment, this is 6 random numbers.

[0057] (random number 1, ..., random number 6) = Seed(longitude, latitude, altitude, Time(timestamp)) i The system uses the time stamps in the range i{1,…,6}, where i is in milliseconds. When a user requests verification, the system retrieves six consecutive timestamps or timestamps that conform to certain rules. The `Time` function converts these six timestamps into six floating-point numbers. For each timestamp floating-point number, the `Seed` function combines the longitude, latitude, and altitude floating-point data and converts them into a random number according to certain rules. After generating six random numbers, the system uses these six random numbers as data indices to extract six corresponding Chinese characters from the Chinese character library in the database to form the verification text.

[0058] Furthermore, users write on the electronic signature device based on the verification text information provided by the system. The system collects user writing data in real time, including writing points, stroke pressure, writing speed, and pen start and stop states. Using the collected data, the system calculates a 9-dimensional handwritten feature Z and obtains a 2D feature tensor Z' that integrates multi-domain information.

[0059] Input Z' into a well-trained model that decouples the writing content and handwriting style to obtain the content feature information and handwriting style feature information of the validation data.

[0060] Furthermore, the system retrieves the standard features corresponding to the verification text from the database and compares them with the verification content features. It also retrieves the user's handwriting style features from the database and compares them with the verification handwriting style features. If the written content is correct and the handwriting style matches successfully, the system approves the verification request, and the user proceeds to the electronic signature signing process. Otherwise, the system denies the user's identity verification and refuses their participation in the subsequent electronic signature signing process, requiring the user to undergo identity verification again.

[0061] In this embodiment, we use cosine similarity to measure the degree of matching between two feature vectors. That is, when the matching degree exceeds a set threshold, the verification and the sample are considered to be consistent (e.g., the optimal threshold can be set to 0.9). At the same time, the comparison strategy is not limited to cosine similarity. Any method that measures feature distance can be used as a comparison strategy, such as Euclidean distance, Manhattan distance, etc.

[0062] Assuming the signing scenario has high security requirements, the system adds a security alert policy. Specifically, if a user fails to verify five times consecutively during the identity verification process, the system will lock the account and require the user to perform auxiliary identity verification, such as fingerprint or facial recognition. If the signing scenario has the highest security requirements, the system will add a two-factor authentication policy. That is, after passing dynamic identity verification, the user enters the formal electronic signature signing process, where the system will perform a secondary verification of the electronic signature content and handwriting style. If the content is correct and the style matches the sample content, the verification passes and the signing process ends; if the verification fails, the user needs to re-sign the electronic signature for verification; if the signing verification fails more than five times consecutively, the user can choose auxiliary identity authentication, re-submit a sample, and then re-enter the signing process for formal signing verification.

[0063] This invention decouples the written content from the handwriting style, improving the accuracy and reliability of the verification system. It has significant application potential in fields requiring precise user identification, such as forensic identification and secure access control. Furthermore, by dynamically generating verification content to authenticate users and adjusting verification strategies according to security requirements, this invention increases the randomness of the verification content and reduces the risk of being cracked. This makes it particularly useful in fields requiring high-security verification, such as the military, government, and scientific research. In summary, this invention's electronic signature handwriting verification system effectively prevents signature fraud and improves the security of electronic document signatures. It has significant application value for industries heavily reliant on electronic signatures, such as finance and e-commerce, and has broad application prospects in areas such as electronic document signing, online transaction authentication, financial security protection, and forensic identification.

[0064] like Figure 4 As shown, the electronic device 300 includes a computing unit 301, which can perform various appropriate actions and processes according to a computer program stored in a read-only memory (ROM) 302 or a computer program loaded from a storage unit 308 into a random access memory (RAM) 303. The RAM 303 may also store various programs and data required for the operation of the device 300. The computing unit 301, ROM 302, and RAM 303 are interconnected via a bus 304. An input / output (I / O) interface 305 is also connected to the bus 304.

[0065] Multiple components in electronic device 300 are connected to I / O interface 305, including: input unit 306, output unit 307, storage unit 308, and communication unit 309. Input unit 306 can be any type of device capable of inputting information to electronic device 300. Input unit 306 can receive input digital or character information and generate key signal inputs related to user settings and / or function control of electronic device. Output unit 307 can be any type of device capable of presenting information and may include, but is not limited to, a display, speaker, video / audio output terminal, vibrator, and / or printer. Storage unit 308 may include, but is not limited to, disk and optical disk. Communication unit 309 allows electronic device 300 to exchange information / data with other devices through computer networks such as the Internet and / or various telecommunications networks, and may include, but is not limited to, modems, network cards, infrared communication devices, wireless communication transceivers, and / or chipsets, such as Bluetooth devices, WiFi devices, WiMax devices, cellular communication devices, and / or the like.

[0066] The computing unit 301 can be a variety of general-purpose and / or special-purpose processing components with processing and computing capabilities. Some examples of the computing unit 301 include, but are not limited to, a central processing unit (CPU), a graphics processing unit (GPU), various special-purpose artificial intelligence (AI) computing chips, various computing units running machine learning model algorithms, a digital signal processor (DSP), and any suitable processor, controller, microcontroller, etc. The computing unit 301 performs the various methods and processes described above. For example, in some embodiments, the reconstruction and decomposition of the muscle movement trajectory of the signature stroke based on its original trajectory, and the decomposition of its logarithmic velocity curve, can be implemented as a computer software program, which is tangibly contained in a machine-readable medium, such as storage unit 308. In some embodiments, part or all of the computer program can be loaded and / or installed on the electronic device 300 via ROM 302 and / or communication unit 309. In some embodiments, the computing unit 301 can be configured by any other suitable means (e.g., by means of firmware) to perform a signature handwriting dynamic acquisition implementation method.

[0067] The program code used to implement the methods of this application may be written in any combination of one or more programming languages. This program code may be provided to a processor or controller of a general-purpose computer, special-purpose computer, or other programmable data processing device, such that when executed by the processor or controller, the functions / operations specified in the flowcharts and / or block diagrams are implemented. The program code may be executed entirely on a machine, partially on a machine, as a standalone software package partially on a machine and partially on a remote machine, or entirely on a remote machine or server.

[0068] In the context of this application, a machine-readable medium can be a tangible medium that may contain or store a program for use by or in conjunction with an instruction execution system, apparatus, or device. A machine-readable medium can be a machine-readable signal medium or a machine-readable storage medium. Machine-readable media can be, but is not limited to, electronic, magnetic, optical, electromagnetic, infrared, or semiconductor systems, apparatus, or devices, or any suitable combination of the foregoing. More specific examples of machine-readable storage media include electrical connections based on one or more wires, portable computer disks, hard disks, random access memory (RAM), read-only memory (ROM), erasable programmable read-only memory (EPROM or flash memory), optical fibers, portable compact disk read-only memory (CD-ROM), optical storage devices, magnetic storage devices, or any suitable combination of the foregoing.

[0069] As used in this application, the terms "machine-readable medium" and "computer-readable medium" refer to any computer program product, device, and / or apparatus (e.g., disk, optical disk, memory, programmable logic device (PLD)) for providing machine instructions and / or data to a programmable processor, including machine-readable media that receive machine instructions as machine-readable signals. The term "machine-readable signal" refers to any signal for providing machine instructions and / or data to a programmable processor.

[0070] To provide interaction with a user, the systems and techniques described herein can be implemented on a computer having: a display device for displaying information to the user (e.g., a CRT (cathode ray tube) or LCD (liquid crystal display) monitor); and a keyboard and pointing device (e.g., a mouse or trackball) through which the user provides input to the computer. Other types of devices can also be used to provide interaction with the user; for example, feedback provided to the user can be any form of sensory feedback (e.g., visual feedback, auditory feedback, or tactile feedback); and input from the user can be received in any form (including sound input, voice input, or tactile input).

[0071] The systems and technologies described herein can be implemented in computing systems that include backend components (e.g., as a data server), or computing systems that include middleware components (e.g., an application server), or computing systems that include frontend components (e.g., a user computer with a graphical user interface or web browser through which a user can interact with embodiments of the systems and technologies described herein), or any combination of such backend, middleware, or frontend components. The components of the system can be interconnected via digital data communication of any form or medium (e.g., a communication network). Examples of communication networks include local area networks (LANs), wide area networks (WANs), and the Internet.

[0072] Computer systems can include clients and servers. Clients and servers are generally located far apart and typically interact through communication networks. Client-server relationships are created by computer programs running on the respective computers and having a client-server relationship with each other.

Claims

1. A method for verifying user identity based on randomly signed electronic information, characterized in that, This includes the sampling stage, where multi-dimensional handcrafted features are acquired when writing arbitrary text information and reshaped into a two-dimensional feature tensor, which is then input into the decoupled writing content and handwriting style model. The sampled handwriting style features are then extracted and stored in the database. During the verification phase, the database generates verification text information, and the multi-dimensional handwritten features of the verification text information are acquired in real time and reconstructed into a two-dimensional feature tensor. This tensor is then input into the decoupled writing content and handwriting style model to obtain verification handwriting style features and verification content features. The verification handwriting style features are compared with the sampled handwriting style features, and the verification content features are compared with the verification text information features to complete user authentication. The reconstructing into a two-dimensional feature tensor includes applying Fast Fourier Transform to obtain multi-dimensional handwritten features. The frequency domain feature information is obtained, and the periodic information in the frequency domain feature is calculated by amplitude. Based on the distribution of high-frequency noise, the frequency components of the feature are statistically analyzed to obtain multiple periodic information of the data, and the periodic information is integrated into the original time domain feature. Reconstructed into a two-dimensional feature tensor .

2. The method according to claim 1, characterized in that, The acquisition of multidimensional handwriting features includes collecting data on handwriting points, pressure, writing speed, and pen movement when a user writes text using an electronic device, and calculating the coordinates of the sampling points. y-axis Horizontal speed Vertical velocity Path tangent angle Path tangent angle cosine value sine value Stroke pressure Pen-raising and lowering states As a multidimensional handcrafted characteristic, among which... .

3. The method according to claim 1, characterized in that, The Reconstructed into a two-dimensional feature tensor Including, calling the formula: Calculate multidimensional features The frequency domain representation and calculation of the average amplitude of the frequency components. According to the average amplitude of the frequency components Call the formula: Calculated frequency According to frequency Call the formula: Calculate the maximum Each frequency component corresponds to a time-domain period. According to the formula: ， We obtain the reconstructed two-dimensional feature tensor, where, For the filling operation with 0, Operations to reshape tensors.

4. The method according to any one of claims 1-3, characterized in that, The generation of verification text information includes obtaining real-time information about the moment the electronic signature device requests verification, converting the timestamp in the real-time information into a set of floating-point numbers, and for each timestamp floating-point number... The function combines the longitude, latitude, and altitude floating-point numbers from the real-time information and converts them into a random number. The converted random number is then used as a data index to extract the corresponding Chinese characters from the database to form the verification text.

5. The method according to any one of claims 1-3, characterized in that, The process of completing user authentication includes: if a user fails to authenticate for a predetermined number of consecutive times during the authentication process, the account is locked and the user is required to perform auxiliary authentication, or the content of the electronic signature and the handwriting style of the signature are verified a second time. If the signature verification fails for more than a predetermined number of consecutive times, the user re-submits a sample and re-enters the signature verification process.

6. A system for verifying user identity based on randomly signed electronic information, characterized in that, The process includes: a sample writing device acquiring multi-dimensional handwriting features when writing arbitrary text information; a feature processing module reshaping these features into a two-dimensional feature tensor; inputting this tensor into a decoupled writing content and handwriting style model to extract the sample writing handwriting style features and store them in a database; the database generating verification text information; the verification writing device acquiring multi-dimensional handwriting features of the verification text information in real time; the feature processing module reshaping this tensor into a two-dimensional feature tensor; inputting this tensor into a decoupled writing content and handwriting style model to obtain verification handwriting style features and verification content features; and a verification unit comparing the verification handwriting style features with the sample writing handwriting style features and comparing the verification content features with the verification text content features generated by the database. If the comparison structures match, user authentication is passed. The reshaping into a two-dimensional feature tensor includes: applying Fast Fourier Transform to obtain multi-dimensional handwriting features. The frequency domain feature information is obtained, and the periodic information in the frequency domain feature is calculated by amplitude. Based on the distribution of high-frequency noise, the frequency components of the feature are statistically analyzed to obtain multiple periodic information of the data, and the periodic information is integrated into the original time domain feature. Reconstructed into a two-dimensional feature tensor .

7. The system according to claim 6, characterized in that, The decoupled writing content and handwriting style model includes: a feature extraction backbone network, a content encoder, a style encoder, and a joint decoder. The feature extraction backbone network performs deep feature extraction on the multidimensional handcraft features of the input to obtain a deep feature representation. The content encoder and style encoder are two independent deep neural network modules used to separate the content feature representation and style feature representation information from the electronic sequence data from the deep feature representation. The joint decoder is an independent deconvolutional, recurrent, or autoregressive neural network module used to combine the content feature representation and style feature representation and decode and restore the input signature data.

8. The system according to any one of claims 6-7, characterized in that, The verification unit retrieves the standard features corresponding to the verification text from the database and compares them with the verification content features. It also retrieves the user's handwriting style features and compares them with the verification handwriting style features. If the written content is correct and the handwriting style matches successfully, the system approves the verification request and the user enters the electronic signature signing process. If the user fails verification a certain number of times in the identity verification process, the account is locked and the user is required to undergo auxiliary identity verification, or a second signing verification is performed on the electronic signature content and handwriting style. If the signing verification fails a certain number of times, the user re-submits a sample and re-enters the signing verification process.

9. An electronic device, comprising: processor; And a memory storing a program, characterized in that the program includes instructions that, when executed by the processor, cause the processor to perform the method for verifying user identity based on randomly signed electronic information according to any one of claims 1-5.

10. A non-transitory computer-readable storage medium storing computer instructions, characterized in that, in, The computer instructions are used to cause the computer to perform the method for verifying user identity based on randomly signed electronic information according to any one of claims 1-5.

Images