A text-independent online handwriting authentication method, system, device and medium

Abstract

The application discloses a text-independent handwriting identity authentication method, system, device and medium, wherein the method comprises the following steps: obtaining text-independent handwriting time sequence data; performing pressure normalization and coordinate center normalization processing on the handwriting time sequence data; constructing a discriminant model based on a deep neural network, and training the discriminant model by using the normalized handwriting time sequence data; obtaining template handwriting and test handwriting, inputting the template handwriting and the test handwriting into the trained discriminant model for feature extraction, and obtaining a first feature vector and a second feature vector; and determining whether the test handwriting is real handwriting or fake handwriting according to the first feature vector and the second feature vector. The application performs handwriting authentication in a text-independent scene, adaptively enhances the learning of style features according to the characteristics that content cannot be used in the text-independent scene but only style can be used, reduces the error rate, and can be widely applied to the fields of deep learning and pattern recognition.

Description

Technical Field

[0001] This invention relates to the fields of deep learning and pattern recognition, and in particular to a text-based handwriting authentication method, system, device, and medium. Background Technology

[0002] Handwriting is a common biometric feature, such as handwritten signatures and handwritten numbers. Handwritten handwriting contains rich personal writing characteristics and can be used for identity verification and authentication, widely applied in criminal investigation, finance, and other fields. Since the advent of the information age, with the development of computer and automation technologies, computer-based biometric authentication technologies have gradually emerged, and thus automated handwriting authentication technologies have also arisen (such as signature authentication). Compared to biometric authentication technologies such as face, iris, and fingerprint recognition, handwriting authentication has the characteristics of low privacy invasion, high user acceptance, high social acceptance, and high security and reliability.

[0003] In recent years, deep learning has developed rapidly, and deep learning-based methods have become the mainstream for online handwriting authentication, which has matured considerably. However, these methods are all for text-related scenarios (where the handwriting text used for authentication is the same), while handwriting authentication methods for text-related scenarios (where the handwriting text used for authentication is different) have been almost entirely unexplored. Summary of the Invention

[0004] In order to at least partially solve one of the technical problems existing in the prior art, the purpose of this invention is to provide a text-independent handwriting authentication method, system, device and medium.

[0005] The technical solution adopted in this invention is:

[0006] A text-based, non-corresponding handwriting authentication method includes the following steps:

[0007] Obtain text-independent handwriting time series data;

[0008] Pressure normalization and coordinate center normalization are performed on the handwriting time series data;

[0009] A deep neural network-based identification model is constructed, and the identification model is trained using normalized handwriting time series data.

[0010] Obtain template handwriting and test handwriting, input template handwriting and test handwriting into trained discrimination model for feature extraction, and obtain the first representation vector corresponding to template handwriting and the second representation vector corresponding to test handwriting.

[0011] The handwriting under test is determined to be either genuine or forged based on the first and second characterization vectors.

[0012] The discrimination model includes a dense multi-scale backbone network, a discriminative pattern feature mining module, and a global attention module.

[0013] The dense multi-scale backbone network is used to expand and map the original time series to a high-dimensional space.

[0014] The discriminative pattern feature mining module is used to decouple the content and style of handwriting;

[0015] The global attention module is used to capture global information from the handwriting time series to enhance the learning of global features.

[0016] Furthermore, the handwriting time series data consists of random number strings with different content, and the storage format of the random number strings is time series. The random number strings include the x and y coordinate information and pressure information of each writing point.

[0017] Furthermore, the pressure normalization and coordinate center normalization processing of the handwriting time series data includes:

[0018] Normalize the pressure information in the handwriting time series data;

[0019] The original x and y coordinate information in the handwriting time series data is subjected to center normalization processing; that is, the midpoint of the sequence coordinates is used as the origin to establish a new coordinate system, and other coordinates are compressed proportionally to the interval (-1,1) and mapped to the new coordinate system.

[0020] Several time functions are extracted based on the normalized x and y coordinate information and pressure information;

[0021] The time function includes at least one of the following functions: the function corresponding to the first derivative of the x-coordinate, the function corresponding to the first derivative of the y-coordinate, the function corresponding to the velocity, the function corresponding to the tangent angle between two points, the function corresponding to the cosine value of the tangent angle, the function corresponding to the sinine value of the tangent angle, the function corresponding to the first derivative of the velocity, the function corresponding to the first derivative of the tangent angle, the function corresponding to the logarithmic radius of curvature, the function corresponding to the magnitude of the velocity change, the function corresponding to the acceleration, or the function corresponding to the pressure.

[0022] Furthermore, the discriminative pattern feature mining module includes a max pooling layer, a maximum value retention layer, a feature segment mining layer, a statistical refinement layer, and a selective pooling layer;

[0023] The input to the discriminative pattern feature mining module is a high-dimensional feature vector extracted by a dense multi-scale network;

[0024] In the maximum value retention layer, the original sequence is compared with the output of the max pooling layer. The points whose original values ​​are the maximum values ​​are retained, and the other points are set to 0. The largest number of these points, p, are selected. i ;

[0025] Point p i The input is fed into the feature segment mining layer, with multiple feature points p i Expand outwards from the center. s / / 2 Length to obtain feature segment p s The length of the feature segment is l s , / / is the integer division symbol;

[0026] feature segment p s In the input statistical refinement layer, calculate all feature segments p. s The mean and standard deviation;

[0027] The output of the statistical refinement layer and the input of the max pooling layer are added together to form a residual connection. The input is selected by the pooling layer to compress the time dimension, and the output is a feature vector fl with fixed dimensions. l .

[0028] Furthermore, the global attention module includes a long short-term memory layer, a residual attention layer, and a selection pooling layer;

[0029] The input of the global attention module is a high-dimensional feature vector extracted by a dense multi-scale network;

[0030] The long short-term memory layer acquires long-distance dependency information ft in the time dimension. t ; to transfer long-distance dependent information ft t The input is fed into the residual attention layer, passes through the softmax function to obtain the score, multiplies the score with the input of the long short-term memory layer, and adds the result of the multiplication to the input to form a residual connection. The result of the residual connection is finally fed into a pooling layer to compress the time dimension, and outputs a fixed-dimensional feature vector f. g .

[0031] Furthermore, training the identification model using normalized handwriting time series data includes:

[0032] The fixed-dimensional feature vector f l and f g The last dimension is concatenated and input into a preset loss function to calculate the loss, and gradient backpropagation is performed to update the parameters; the training data does not need to be labeled, and the model is trained in a weakly supervised manner throughout the process.

[0033] Further, the step of determining whether the handwriting to be tested is genuine or forged based on the first representation vector and the second representation vector includes:

[0034] Based on the Euclidean distance between the first and second representation vectors, if the Euclidean distance is less than a preset threshold, the handwriting to be tested is determined to be genuine; otherwise, the handwriting to be tested is determined to be forged.

[0035] Another technical solution adopted in this invention is:

[0036] A text-free, computer-generated handwriting authentication system includes:

[0037] The data acquisition module is used to acquire handwriting time series data that is not related to text.

[0038] The data processing module is used to perform pressure normalization and coordinate center normalization on the handwriting time series data;

[0039] The model training module is used to construct a recognition model based on a deep neural network, and the recognition model is trained using normalized handwriting time series data.

[0040] The feature extraction module is used to obtain template handwriting and test handwriting. The template handwriting and test handwriting are input into the trained discrimination model for feature extraction to obtain the first representation vector corresponding to the template handwriting and the second representation vector corresponding to the test handwriting.

[0041] The note authentication module is used to determine whether the handwriting under test is genuine or forged based on the first and second representation vectors.

[0042] The discrimination model includes a dense multi-scale backbone network, a discriminative pattern feature mining module, and a global attention module.

[0043] The dense multi-scale backbone network is used to expand and map the original time series to a high-dimensional space.

[0044] The discriminative pattern feature mining module is used to decouple the content and style of handwriting;

[0045] The global attention module is used to capture global information from the handwriting time series to enhance the learning of global features.

[0046] Another technical solution adopted in this invention is:

[0047] A text-free handwriting authentication device includes:

[0048] At least one processor;

[0049] At least one memory for storing at least one program;

[0050] When the at least one program is executed by the at least one processor, the at least one processor implements the method described above.

[0051] Another technical solution adopted in this invention is:

[0052] A computer-readable storage medium storing a processor-executable program, which, when executed by a processor, performs the method described above.

[0053] The beneficial effects of this invention are: This invention performs handwriting authentication in text-independent scenarios, and through the discriminative pattern feature mining module, it adaptively enhances the learning of style features to address the characteristic that only style can be used in text-independent scenarios where content cannot be used. This results in a lower error rate than other cutting-edge methods. Attached Figure Description

[0054] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the following description is provided with accompanying drawings of the relevant technical solutions in the embodiments of the present invention or the prior art. It should be understood that the accompanying drawings described below are only for the purpose of clearly illustrating some embodiments of the technical solutions of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without any creative effort.

[0055] Figure 1 This is a flowchart illustrating the steps of a text-based handwriting authentication method according to an embodiment of the present invention.

[0056] Figure 2 This is a flowchart illustrating the text-free handwriting authentication method in an embodiment of the present invention.

[0057] Figure 3 This is a structural diagram of the identification model in an embodiment of the present invention. Detailed Implementation

[0058] The embodiments of the present invention are described in detail below. Examples of these embodiments are shown in the accompanying drawings, wherein the same or similar reference numerals denote the same or similar elements or elements having the same or similar functions throughout. The embodiments described below with reference to the accompanying drawings are exemplary and are only used to explain the present invention, and should not be construed as limiting the present invention. The step numbers in the following embodiments are set only for ease of explanation, and there is no limitation on the order between the steps. The execution order of each step in the embodiments can be adaptively adjusted according to the understanding of those skilled in the art.

[0059] In the description of this invention, it should be understood that the orientation descriptions, such as up, down, front, back, left, right, etc., are based on the orientation or positional relationship shown in the accompanying drawings. They are only for the convenience of describing this invention and simplifying the description, and do not indicate or imply that the device or element referred to must have a specific orientation, or be constructed and operated in a specific orientation. Therefore, they should not be construed as limiting this invention.

[0060] In the description of this invention, "several" means one or more, "more than" means two or more, "greater than," "less than," and "exceeding" are understood to exclude the stated number, while "above," "below," and "within" are understood to include the stated number. The use of "first" and "second" in the description is merely for distinguishing technical features and should not be construed as indicating or implying relative importance, or implicitly indicating the number of indicated technical features, or implicitly indicating the order of the indicated technical features.

[0061] In the description of this invention, unless otherwise explicitly defined, terms such as "set up," "install," and "connect" should be interpreted broadly, and those skilled in the art can reasonably determine the specific meaning of the above terms in this invention in conjunction with the specific content of the technical solution.

[0062] To address existing technical challenges, this invention proposes an online handwriting authentication method based on deep learning and style pattern mining. This method adaptively decouples handwriting content from style, enhancing the learning of style features. It also provides relevant convolutional kernel parameters and model training methods. This method is particularly suitable for text-independent handwriting data, enabling more robust high-dimensional representations of such data and improving the accuracy of identity authentication.

[0063] like Figure 1 and Figure 2 As shown, this embodiment provides a text-independent handwriting authentication method. This method employs deep learning and style pattern mining mechanisms to adaptively decouple the style and content of text-independent handwriting data, enhancing the representational ability of handwriting features, thereby achieving low-error-rate text-independent handwriting authentication and possessing high practicality. The method specifically includes the following steps:

[0064] S1. Obtain text-independent handwriting time series data.

[0065] Data Preparation: Obtain text-independent handwriting time-series data. In this embodiment, random number strings with different content are used as data examples; other text-independent handwriting data can also be used. The random number strings are stored in time-series format, containing the x and y coordinates and pressure information of each writing point.

[0066] S2. Perform pressure normalization and coordinate center normalization on the handwriting time series data.

[0067] Specifically, step S2 includes steps S21-S23:

[0068] S21. Normalize the pressure information in the handwriting time series data;

[0069] S22. Perform center normalization on the original x and y coordinate information in the handwriting time series data;

[0070] S23. Extract several time functions based on the normalized x and y coordinate information and pressure information.

[0071] The time function includes at least one of the following functions: the function corresponding to the first derivative of the x-coordinate, the function corresponding to the first derivative of the y-coordinate, the function corresponding to the velocity, the function corresponding to the tangent angle between two points, the function corresponding to the cosine value of the tangent angle, the function corresponding to the sinine value of the tangent angle, the function corresponding to the first derivative of the velocity, the function corresponding to the first derivative of the tangent angle, the function corresponding to the logarithmic radius of curvature, the function corresponding to the magnitude of the velocity change, the function corresponding to the acceleration, or the function corresponding to the pressure.

[0072] As an optional implementation, 12 time functions are extracted using x, y coordinates and pressure, as shown in Table 1:

[0073] Table 1

[0074]

[0075]

[0076] S3. Construct a recognition model based on a deep neural network, and train the recognition model using normalized handwriting time series data.

[0077] A deep neural network model is constructed, comprising a dense multi-scale backbone network, a discriminative pattern feature mining module, and a global attention module. The dense multi-scale backbone network expands the original time series to a high-dimensional space; the discriminative pattern feature mining module decouples style and content, adaptively mining eight of the most discriminative pattern feature segments, enhancing the modeling of local style features by focusing on writing patterns; the global attention module captures long-distance dependencies throughout the sequence through long short-term memory layers and residual mechanisms, enhancing the learning of global features. The network structure diagram of the discriminative model is shown below. Figure 3 As shown.

[0078] (1) Dense multi-scale backbone network

[0079] The structure of the dense multiscale backbone network is shown in Table 2. Tables 3 and 4 list the structures of some sub-layers, where l is the length of the input handwriting sequence.

[0080] Table 2 Dense Multiscale Backbone Network Structure

[0081]

[0082] The structures of the convolution-nonlinear-batch normalization layer are shown in Table 3, and the structures of the delay scaling layer are shown in Table 4.

[0083] Table 3 Convolutional-Nonlinear-Batch Normalized Layer Structure

[0084]

[0085]

[0086]

[0087]

[0088] (2) Discriminative pattern feature mining module

[0089] The structure of the discriminative pattern feature mining module is shown in Table 5:

[0090] Table 5. Structure of the Discriminative Pattern Feature Mining Module

[0091] Network layer Specific operations Output size Max pooling layer Number of cores: 3, step size: 1, edge padding: 1 l*512 Maximum value retention layer Retain the 8 largest points after max pooling of the original sequence. l*512 Feature segment mining layer Eight feature segments extend from the feature points. <![CDATA[8*l s *512]]> Statistical refinement layer - l*512 Select pooling layer Compress the time dimension of the sequence to 1, and output a vector. 320

[0092] The discriminative pattern feature mining module takes as input a high-dimensional feature vector extracted by a dense multi-scale network. In the maximum value retention layer, the original sequence is compared with the output of the max pooling layer. The points with the original maximum value are retained, while other points are set to 0. Then, the eight largest points p are selected. i ,i∈{1,…,8}. Let p i The input is fed into the feature segment mining layer, and then used as 8 feature points p i Expand outwards from the center. s / / 2 Length to obtain feature segment p s The length l of the feature segment s = l / / 4 / / 8 ( / / is the integer division symbol), at this time the output size of the feature segment is 8×l s ×512, these 8 segments represent the most prominent writing patterns of this handwriting (such as cursive, strokes, etc.), and are considered to be the most discernible stylistic features of the current handwriting. This allows the model to focus deeply on the writing style of the handwriting. (The last part, "p," appears to be a typo and can be left as is.) s In the input statistical refinement layer, calculate all p sThe mean and standard deviation of the input are calculated. The original sequence is subtracted from this mean and divided by the standard deviation. Then, values ​​at feature segments are multiplied by 0.9, and values ​​at non-feature segments are multiplied by 0.005. Finally, the inputs of the statistical refinement layer and the max pooling layer are added together to form a residual connection. Then, the input is used to select a pooling layer to compress the time dimension, and the output is a fixed-dimensional feature vector f. l .

[0093] (3) Global Attention Module

[0094] The structure of the global attention module is shown in Table 6:

[0095] Table 6

[0096]

[0097] The input to the global attention module is a high-dimensional feature vector extracted by a dense multi-scale network. The long short-term memory layer acquires long-distance dependency information f in the time dimension. t f t The input is fed into the residual attention layer, passes through the softmax function to obtain the score, and is multiplied by the input of the long short-term memory layer. The result of the multiplication is added to the input to form a residual connection. The result of the residual connection is finally fed into a pooling layer to compress the time dimension, and outputs a fixed-dimensional feature vector f. g .

[0098] As an optional implementation, f l and f g The data is concatenated in the last dimension and input into the loss function to calculate the loss. Gradient backpropagation is then performed to update the parameters. No labeled training data is required; the model is trained entirely using weak supervision.

[0099] Specifically, the training parameters are as follows:

[0100] Iteration Algebra: 350

[0101] Number of iterations per generation: 202

[0102] Optimizer: AdamW

[0103] Learning rate: 0.001, descent strategy is to multiply by 0.95 at the end of each generation.

[0104] Weight decay: 0.00001

[0105] S4. Obtain the template handwriting and the handwriting to be tested. Input the template handwriting and the handwriting to be tested into the trained discrimination model for feature extraction to obtain the first representation vector corresponding to the template handwriting and the second representation vector corresponding to the handwriting to be tested.

[0106] S5. Determine whether the handwriting to be tested is genuine or forged based on the first and second representation vectors.

[0107] Both template handwriting and the handwriting samples to be tested are input into a distance-based validator. The handwriting samples to be tested include both real and forged handwriting. The Euclidean distance between each template and each handwriting sample to be tested is calculated. The distance of the handwriting sample to be tested is divided by the distance of the template for regularization. If the distance of the handwriting sample to be tested is less than a threshold t, it is judged as real handwriting; otherwise, it is judged as forged handwriting. t is a variable preset threshold. By changing the threshold t, the handwriting samples will be judged as real / forged. The judgment result is compared with its original real / forged label to calculate the equal error rate.

[0108] In summary, the method of the present invention has at least the following advantages and beneficial effects compared with the prior art:

[0109] (1) Compared with previous methods that only target text-related handwriting authentication, the method proposed in this invention is specifically designed for text-independent scenarios. It can extract features that are significantly superior to other methods in text-independent scenarios, greatly reducing the error rate of handwriting authentication. It can also achieve performance similar to other cutting-edge methods in text-related scenarios. At the same time, this is the first method proposed for text-independent handwriting authentication.

[0110] (2) This invention proposes a discriminative pattern feature mining module, specifically designed to decouple style and content and adaptively enhance the learning of style features. In text-independent scenarios, the content of handwriting data is misaligned and cannot provide useful information for model learning; only style can be used for authentication. Therefore, this module specifically mines the most discriminative writing pattern segments to focus on the most prominent writing style, enhancing the modeling of style features. This is particularly suitable for text-independent handwriting authentication, whereas previous methods did not specifically enhance the learning of style features.

[0111] (3) The present invention performs comprehensive and effective preprocessing of data. Data normalization brings spatial invariance of handwriting, which is beneficial for the model to learn handwriting information. The 12 time functions further enrich the input features, making the writing features extracted by the model more comprehensive.

[0112] (4) The model training adopts a weakly supervised approach, which means that the training data does not need to be labeled, saving a lot of manpower and resources. The training loss is optimized in the vector space, which gives the model more room for adaptive adjustment, resulting in more robust performance and a lower error rate.

[0113] This embodiment also provides a text-independent handwriting authentication system, including:

[0114] The data acquisition module is used to acquire handwriting time series data that is not related to text.

[0115] The data processing module is used to perform pressure normalization and coordinate center normalization on the handwriting time series data;

[0116] The model training module is used to construct a recognition model based on a deep neural network, and the recognition model is trained using normalized handwriting time series data.

[0117] The feature extraction module is used to obtain template handwriting and test handwriting. The template handwriting and test handwriting are input into the trained discrimination model for feature extraction to obtain the first representation vector corresponding to the template handwriting and the second representation vector corresponding to the test handwriting.

[0118] The note authentication module is used to determine whether the handwriting under test is genuine or forged based on the first and second representation vectors.

[0119] The discrimination model includes a dense multi-scale backbone network, a discriminative pattern feature mining module, and a global attention module.

[0120] The dense multi-scale backbone network is used to expand and map the original time series to a high-dimensional space.

[0121] The discriminative pattern feature mining module is used to decouple the content and style of handwriting;

[0122] The global attention module is used to capture global information from the handwriting time series to enhance the learning of global features.

[0123] This embodiment of the text-free handwriting authentication system can execute the text-free handwriting authentication method provided in the method embodiment of the present invention, and can execute any combination of the implementation steps of the method embodiment, and has the corresponding functions and beneficial effects of the method.

[0124] This embodiment also provides a text-independent handwriting authentication device, including:

[0125] At least one processor;

[0126] At least one memory for storing at least one program;

[0127] When the at least one program is executed by the at least one processor, the at least one processor implements Figure 1 The method shown.

[0128] This embodiment of the text-free handwriting authentication device can execute the text-free handwriting authentication method provided in the method embodiment of the present invention, and can execute any combination of the implementation steps of the method embodiment, and has the corresponding functions and beneficial effects of the method.

[0129] This application also discloses a computer program product or computer program, which includes 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, causing the computer device to perform... Figure 1 The method shown.

[0130] This embodiment also provides a storage medium storing instructions or programs that can execute the text-free handwriting authentication method provided in the method embodiment of the present invention. When the instructions or programs are run, any combination of implementation steps of the method embodiment can be executed, and the method has the corresponding functions and beneficial effects.

[0131] In some alternative embodiments, the functions / operations mentioned in the block diagrams may not occur in the order shown in the operation diagrams. For example, depending on the functions / operations involved, two consecutively shown blocks may actually be executed substantially simultaneously, or the blocks may sometimes be executed in reverse order. Furthermore, the embodiments presented and described in the flowcharts of this invention are provided by way of example to provide a more comprehensive understanding of the technology. The disclosed methods are not limited to the operations and logic flows presented herein. Alternative embodiments are contemplated in which the order of various operations is altered and sub-operations described as part of a larger operation are executed independently.

[0132] Furthermore, although the invention has been described in the context of functional modules, it should be understood that, unless otherwise stated, one or more of the described functions and / or features may be integrated into a single physical device and / or software module, or one or more functions and / or features may be implemented in a separate physical device or software module. It is also understood that a detailed discussion of the actual implementation of each module is unnecessary for understanding the invention. Rather, given the properties, functions, and internal relationships of the various functional modules in the apparatus disclosed herein, the actual implementation of the module will be understood within the scope of conventional skill of an engineer. Therefore, those skilled in the art can implement the invention as set forth in the claims using ordinary techniques without excessive experimentation. It is also understood that the specific concepts disclosed are merely illustrative and not intended to limit the scope of the invention, which is determined by the full scope of the appended claims and their equivalents.

[0133] If the aforementioned functions are implemented as software functional units and sold or used as independent products, they can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of this invention, essentially, or the part that contributes to the prior art, or a portion of the technical solution, can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute all or part of the steps of the methods described in the various embodiments of this invention. The aforementioned storage medium includes various media capable of storing program code, such as USB flash drives, portable hard drives, read-only memory (ROM), random access memory (RAM), magnetic disks, or optical disks.

[0134] The logic and / or steps represented in the flowchart or otherwise described herein, for example, can be considered as a sequenced list of executable instructions for implementing logical functions, and can be embodied in any computer-readable medium for use by, or in conjunction with, an instruction execution system, apparatus, or device (such as a computer-based system, a processor-included system, or other system that can fetch and execute instructions from, an instruction execution system, apparatus, or device). For the purposes of this specification, "computer-readable medium" can be any means that can contain, store, communicate, propagate, or transmit programs for use by, or in conjunction with, an instruction execution system, apparatus, or device.

[0135] More specific examples of computer-readable media (a non-exhaustive list) include: electrical connections (electronic devices) having one or more wires, portable computer disk drives (magnetic devices), random access memory (RAM), read-only memory (ROM), erasable and editable read-only memory (EPROM or flash memory), fiber optic devices, and portable optical disc read-only memory (CDROM). Furthermore, computer-readable media can even be paper or other suitable media on which the program can be printed, since the program can be obtained electronically, for example, by optically scanning the paper or other medium, followed by editing, interpreting, or otherwise processing as necessary, and then stored in computer memory.

[0136] 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.

[0137] In the foregoing description of this specification, references to terms such as "one embodiment," "another embodiment," or "some embodiments" indicate that a specific feature, structure, material, or characteristic described in connection with an embodiment or example is included in at least one embodiment or example of the present invention. In this specification, 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.

[0138] Although embodiments of the invention have been shown and described, those skilled in the art will understand that various changes, modifications, substitutions and alterations can be made to these embodiments without departing from the principles and spirit of the invention, the scope of which is defined by the claims and their equivalents.

[0139] The above is a detailed description of the preferred embodiments of the present invention. However, the present invention is not limited to the above embodiments. Those skilled in the art can make various equivalent modifications or substitutions without departing from the spirit of the present invention. All such equivalent modifications or substitutions are included within the scope defined by the claims of this application.

Claims

1. A text-based, non-corresponding handwriting authentication method, characterized in that, Includes the following steps: Obtain text-independent handwriting time series data; Pressure normalization and coordinate center normalization are performed on the handwriting time series data; A deep neural network-based identification model is constructed, and the identification model is trained using normalized handwriting time series data. Obtain template handwriting and test handwriting, input template handwriting and test handwriting into trained discrimination model for feature extraction, and obtain the first representation vector corresponding to template handwriting and the second representation vector corresponding to test handwriting. The handwriting under test is determined to be either genuine or forged based on the first and second characterization vectors. The discrimination model includes a dense multi-scale backbone network, a discriminative pattern feature mining module, and a global attention module. The dense multi-scale backbone network is used to expand and map the original time series to a high-dimensional space. The discriminative pattern feature mining module is used to decouple the content and style of handwriting; The global attention module is used to capture global information from the handwriting time series to enhance the learning of global features; The discriminative pattern feature mining module includes a max pooling layer, a maximum value retention layer, a feature segment mining layer, a statistical refinement layer, and a selective pooling layer. The input to the discriminative pattern feature mining module is a high-dimensional feature vector extracted by a dense multi-scale backbone network; In the maximum value retention layer, the original sequence is compared with the output of the max pooling layer. The points whose original values ​​are the maximum values ​​are retained, and the other points are set to 0. The largest of these points are then selected. ; Point The input is fed into the feature segment mining layer, using multiple feature points. Expanding outwards from the center. Length of feature segment The length of the feature segment is , The integer division symbol; Feature segments In the input statistical refinement layer, all feature segments are calculated. The mean and standard deviation; The output of the statistical refinement layer and the input of the max pooling layer are added together to form a residual connection. The input is selected by the pooling layer to compress the time dimension, and the output is a feature vector with fixed dimensions. .

2. The text-based handwriting authentication method according to claim 1, characterized in that, The handwriting time series data consists of random number strings with different contents. The random number strings are stored in a time series format and include the x and y coordinates and pressure information of each handwriting point.

3. The text-based handwriting authentication method according to claim 1, characterized in that, The process of performing pressure normalization and coordinate center normalization on the handwriting time series data includes: Normalize the pressure information in the handwriting time series data; The original x and y coordinate information in the handwriting time series data is center-normalized. Several time functions are extracted based on the normalized x and y coordinate information and pressure information; The time function includes at least one of the following functions: the function corresponding to the first derivative of the x-coordinate, the function corresponding to the first derivative of the y-coordinate, the function corresponding to the velocity, the function corresponding to the tangent angle between two points, the function corresponding to the cosine value of the tangent angle, the function corresponding to the sinine value of the tangent angle, the function corresponding to the first derivative of the velocity, the function corresponding to the first derivative of the tangent angle, the function corresponding to the logarithmic radius of curvature, the function corresponding to the magnitude of the velocity change, the function corresponding to the acceleration, or the function corresponding to the pressure.

4. The text-based handwriting authentication method according to claim 1, characterized in that, The global attention module includes a long short-term memory layer, a residual attention layer, and a selection pooling layer; The input to the global attention module is a high-dimensional feature vector extracted by a dense multi-scale backbone network; The long short-term memory layer acquires long-distance dependency information in the time dimension. Long-distance dependent information The input is fed into the residual attention layer, passes through the softmax function to obtain an attention score, and is then multiplied by the input of the long short-term memory layer. The result of this multiplication is added to the input to form a residual connection. Finally, the result of the residual connection is fed into a pooling layer to compress the time dimension, and the output is a feature vector of fixed dimensions. .

5. The text-independent handwriting authentication method according to claim 4, characterized in that, The step of training the identification model using normalized handwriting time series data includes: fixed-dimensional feature vectors and The last dimension is concatenated and input into a preset loss function to calculate the loss, and gradient backpropagation is performed to update the parameters; the training data does not need to be labeled, and the model is trained in a weakly supervised manner throughout the process.

6. The text-based handwriting authentication method according to claim 1, characterized in that, The step of determining whether the handwriting to be tested is genuine or forged based on the first and second representation vectors includes: The Euclidean distance is calculated based on the first and second representation vectors. If the Euclidean distance is less than a preset threshold, the handwriting to be tested is determined to be genuine; otherwise, the handwriting to be tested is determined to be forged.

7. A text-independent handwriting authentication system, characterized in that, include: The data acquisition module is used to acquire handwriting time series data that is not related to text. The data processing module is used to perform pressure normalization and coordinate center normalization on the handwriting time series data; The model training module is used to construct a recognition model based on a deep neural network, and to train the recognition model using normalized handwriting time series data. The feature extraction module is used to obtain template handwriting and test handwriting. The template handwriting and test handwriting are input into the trained discrimination model for feature extraction to obtain the first representation vector corresponding to the template handwriting and the second representation vector corresponding to the test handwriting. The handwriting authentication module is used to determine whether the handwriting to be tested is genuine or forged based on the first representation vector and the second representation vector. The discrimination model includes a dense multi-scale backbone network, a discriminative pattern feature mining module, and a global attention module. The dense multi-scale backbone network is used to expand and map the original time series to a high-dimensional space. The discriminative pattern feature mining module is used to decouple the content and style of handwriting; The global attention module is used to capture global information from the handwriting time series to enhance the learning of global features; The discriminative pattern feature mining module includes a max pooling layer, a maximum value retention layer, a feature segment mining layer, a statistical refinement layer, and a selective pooling layer. The input to the discriminative pattern feature mining module is a high-dimensional feature vector extracted by a dense multi-scale backbone network; In the maximum value retention layer, the original sequence is compared with the output of the max pooling layer. The points whose original values ​​are the maximum values ​​are retained, and the other points are set to 0. The largest of these points are then selected. ; Point The input is fed into the feature segment mining layer, using multiple feature points. Expanding outwards from the center. Length of feature segment The length of the feature segment is , The integer division symbol; Feature segments In the input statistical refinement layer, all feature segments are calculated. The mean and standard deviation; The output of the statistical refinement layer and the input of the max pooling layer are added together to form a residual connection. The input is selected by the pooling layer to compress the time dimension, and the output is a feature vector with fixed dimensions. .

8. A text-independent handwriting authentication device, characterized in that, include: At least one processor; At least one memory for storing at least one program; When the at least one program is executed by the at least one processor, the at least one processor implements the method according to any one of claims 1-6.

9. A computer-readable storage medium storing a processor-executable program, characterized in that, The processor-executable program, when executed by the processor, is used to perform the method as described in any one of claims 1-6.

Images