Signature authentication method and device, computer equipment and storage medium

Abstract

The invention relates to a signature authentication method and device, computer equipment and a storage medium. The method comprises the following steps: acquiring a target signature image of a target user; wherein the target signature image comprises offline signature information of the target user; obtaining a target local feature of the target signature image according to the target signature image based on the trained local network model; inputting the target signature image into the trained global network model to obtain a target global feature of the target signature image and a target weight vector of the target local feature; and authenticating the offline signature information according to the target global feature, the target local feature, the target weight vector of the target local feature and the reference signature image of the target user. By adopting the method, precise authentication of the signature information can be realized.

Description

Technical Field

[0001] The present application relates to the field of computer technology, and in particular to a signature authentication method, apparatus, computer device and storage medium. Background Art

[0002] With the development of the Internet, network security risks are becoming increasingly serious, personal information leakage is serious, and network security issues are receiving more and more attention. How to ensure the security of network information is a problem that must be solved while the Internet is developing rapidly. Among them, identity authentication, especially signature authentication, plays a vital role.

[0003] Currently, most methods for authenticating signatures focus on overall recognition and verification of the image containing the signature information. However, this method ignores the local structural features of the signature information and reduces the accuracy of the signature authentication result. Summary of the invention

[0004] Based on this, it is necessary to provide a signature authentication method, device, computer equipment and storage medium that can accurately authenticate the signature in response to the above technical problems.

[0005] In a first aspect, the present application provides a signature authentication method, comprising:

[0006] Acquire a target signature image of a target user; wherein the target signature image includes offline signature information of the target user;

[0007] Based on the trained local network model, according to the target signature image, a target local feature of the target signature image is obtained;

[0008] Inputting the target signature image into the trained global network model to obtain the target global features of the target signature image and the target weight vector of the target local features;

[0009] The offline signature information is authenticated according to the target global feature, the target local feature, the target weight vector of the target local feature and the reference signature image of the target user.

[0010] In one embodiment, obtaining the target local features of the target signature image based on the trained local network model according to the target signature image includes:

[0011] Preprocessing the target signature image;

[0012] Slice the preprocessed target signature image;

[0013] The slice signature image obtained by slicing is input into the trained local network model to obtain the target local features of the target signature image.

[0014] In one embodiment, the preprocessing of the target signature image includes:

[0015] Based on the maximum inter-class variance method, the target signature image is binarized to obtain a binarized image;

[0016] The binary image is scaled to obtain a preprocessed target signature image; wherein the size ratio of the binary image is the same as the size ratio of the preprocessed target signature image.

[0017] In one embodiment, authenticating the offline signature information according to the target global feature, the target local feature, the target weight vector of the target local feature and the reference signature image of the target user includes:

[0018] Acquire reference global features and reference local features of the reference signature image of the target user;

[0019] Determining a global Euclidean distance of the target signature image according to the reference global feature and the target global feature;

[0020] determining a local Euclidean distance of the target signature image according to the reference local feature, the target local feature and a target weight vector of the target local feature;

[0021] The offline signature information is authenticated according to the global Euclidean distance and the local Euclidean distance.

[0022] In one embodiment, authenticating the offline signature information in the target signature image according to the global Euclidean distance and the local Euclidean distance includes:

[0023] Obtaining the intra-class distance of the reference signature image;

[0024] Taking the sum of the global Euclidean distance and the local Euclidean distance as the overall Euclidean distance of the target signature image;

[0025] The offline signature information in the target signature image is authenticated according to the overall Euclidean distance, the number of images of the reference signature image and the intra-class distance.

[0026] In one embodiment, authenticating the offline signature information in the target signature image according to the overall Euclidean distance, the number of images of the reference signature image, and the intra-class distance includes:

[0027] determining a normalized minimum distance according to the overall Euclidean distance, the number of images of the reference signature image, and the intra-class distance;

[0028] If the normalized minimum distance is less than the distance threshold, determining that the authentication result of the offline signature information is a true signature;

[0029] If the normalized minimum distance is greater than the distance threshold, it is determined that the authentication result is a forged signature.

[0030] In one embodiment, the global network model and the local network model are trained in the following manner, including:

[0031] Obtaining sample signatures of sample users; wherein the sample signatures include real signatures and forged signatures;

[0032] Determining a signature image and a signature slice image of the sample signature;

[0033] Inputting the signature image into an initial global network model to obtain global features of the signature image and a slicing weight vector obtained by adaptively slicing the signature image by the initial global network model;

[0034] Inputting the signature slice image into an initial local network model to obtain local features of the signature slice image;

[0035] Constructing a global loss function according to the global features of the signature image;

[0036] Constructing a local loss function according to the local features of the signature slice image and the slice weight vector;

[0037] Determine a total loss function according to the global loss function and the local loss function;

[0038] The global loss function, the local loss function and the total loss function are used to jointly train the initial global network model and the initial local network model to obtain the trained global network model and the local network model.

[0039] In a second aspect, the present application also provides a signature authentication device, including:

[0040] An image acquisition module, used to acquire a target signature image of a target user; wherein the target signature image contains offline signature information of the target user;

[0041] A first extraction module is used to obtain target local features of the target signature image based on the trained local network model and the target signature image;

[0042] A second extraction module is used to input the target signature image into the trained global network model to obtain the target global feature of the target signature image and the target weight vector of the target local feature;

[0043] The signature authentication module is used to authenticate the offline signature information according to the target global feature, the target local feature, the target weight vector of the target local feature and the reference signature image of the target user.

[0044] In a third aspect, the present application further provides a computer device, including a memory and a processor, wherein the memory stores a computer program, and when the processor executes the computer program, the following steps are implemented:

[0045] Acquire a target signature image of a target user; wherein the target signature image includes offline signature information of the target user;

[0046] Based on the trained local network model, according to the target signature image, a target local feature of the target signature image is obtained;

[0047] Inputting the target signature image into the trained global network model to obtain the target global features of the target signature image and the target weight vector of the target local features;

[0048] The offline signature information is authenticated according to the target global feature, the target local feature, the target weight vector of the target local feature and the reference signature image of the target user.

[0049] In a fourth aspect, the present application further provides a computer-readable storage medium having a computer program stored thereon, wherein when the computer program is executed by a processor, the following steps are implemented:

[0050] Acquire a target signature image of a target user; wherein the target signature image includes offline signature information of the target user;

[0051] Based on the trained local network model, according to the target signature image, a target local feature of the target signature image is obtained;

[0052] Inputting the target signature image into the trained global network model to obtain the target global features of the target signature image and the target weight vector of the target local features;

[0053] The offline signature information is authenticated according to the target global feature, the target local feature, the target weight vector of the target local feature and the reference signature image of the target user.

[0054] In a fifth aspect, the present application further provides a computer program product, including a computer program, which implements the following steps when executed by a processor:

[0055] Acquire a target signature image of a target user; wherein the target signature image includes offline signature information of the target user;

[0056] Based on the trained local network model, according to the target signature image, a target local feature of the target signature image is obtained;

[0057] Inputting the target signature image into the trained global network model to obtain the target global features of the target signature image and the target weight vector of the target local features;

[0058] The offline signature information is authenticated according to the target global feature, the target local feature, the target weight vector of the target local feature and the reference signature image of the target user.

[0059] The above-mentioned signature authentication method, device, computer equipment and storage medium, after obtaining the target signature image containing the offline signature information of the target user, obtain the target local features of the target signature image based on the trained local network model and the target signature image, and input the target signature image into the trained global network model to obtain the target global features and the target weight vector of the target local features of the target signature image; further, the offline signature information is authenticated based on the target global features, the target local features, the target weight vector of the target local features and the reference signature image of the target user. The above-mentioned scheme, by introducing the global network model and the local network model, extracts the target global features and the target local features of the target signature image, and obtains the target weight vector of the target local features, which is equivalent to comprehensively considering the global information and local information of the target signature image, and on the basis of assigning the target weight vector to each target local feature, ensures the accuracy of verifying the offline signature information in the target signature image. BRIEF DESCRIPTION OF THE DRAWINGS

[0060] In order to more clearly illustrate the technical solutions in the embodiments of the present application or the related technologies, the drawings required for use in the embodiments or the related technical descriptions are briefly introduced below. Obviously, the drawings described below are only some embodiments of the present application. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying creative work.

[0061] Figure 1 A schematic diagram of a flow chart of a signature authentication method in an embodiment;

[0062] Figure 2A schematic diagram of a process for determining a local feature of a target in one embodiment;

[0063] Figure 3 A schematic diagram of performing slicing processing on a target signature image in one embodiment;

[0064] Figure 4A A schematic diagram of a process of training a global network model and a local network model in one embodiment;

[0065] Figure 4B is a schematic diagram of a training process of a global network model and a local network model in another embodiment;

[0066] Figure 4C is a schematic diagram of a training process of a global network model and a local network model in yet another embodiment;

[0067] Figure 5 A schematic diagram of a process for authenticating offline signature information in an embodiment;

[0068] Figure 6 It is a schematic diagram of a process of authenticating offline signature information in another embodiment;

[0069] Figure 7 A schematic diagram of a process for authenticating offline signature information in yet another embodiment;

[0070] Figure 8 A schematic diagram of a flow chart of a signature authentication method in another embodiment;

[0071] Fig. 9 is a structural block diagram of a signature authentication device in an embodiment;

[0072] Fig.10 FIG. 4 is a diagram showing the internal structure of a computer device in one embodiment. DETAILED DESCRIPTION

[0073] In order to make the purpose, technical solution and advantages of the present application more clearly understood, the present application is further described in detail below in conjunction with the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are only used to explain the present application and are not used to limit the present application.

[0074] The signature authentication method provided in the embodiment of the present application can be applied to an application environment in which the authenticity of a user's offline signature is authenticated. The signature authentication method provided in the embodiment of the present application can be executed by a computer device, which can be a server or a terminal with powerful computing capabilities. In an exemplary embodiment, Figure 1 As shown, a signature authentication method is provided, which is described by taking the method applied to a server as an example, and specifically includes the following steps:

[0075] S101, obtaining a target signature image of a target user.

[0076] The target signature image is an image containing the offline signature information of the target user; and the offline signature is an offline handwritten signature.

[0077] Optionally, an image acquisition device, such as a camera, is used to take a photo of the target user's offline signature to obtain a target signature image, and ensure that the collected target signature image contains complete offline signature information. After the image acquisition device acquires the target signature image, the target signature image can be sent to the server via Bluetooth or wireless communication, and further, the server can obtain the target signature image of the target user.

[0078] S102, based on the trained local network model, according to the target signature image, obtain the target local features of the target signature image.

[0079] Among them, the local network model is a model for extracting local features of the target signature image.

[0080] Optionally, the target signature image can be sliced ​​according to preset rules, and the sliced ​​target signature image can be input into a trained local network model. The model parameters of the trained local network model can be used to extract features from the sliced ​​target signature image to obtain target local features of the sliced ​​target signature image.

[0081] S103, inputting the target signature image into the trained global network model to obtain a target weight vector of the target global features and the target local features of the target signature image.

[0082] The global network model is a model for extracting global features of the target signature image and adaptively selecting weights for the target local features extracted by the local network model; the target global feature is a global feature extracted from the target signature image. The target weight vector of the target local feature contains the weight of each target local feature.

[0083] It should be noted that, after slicing the target signature image, each sliced ​​signature image obtained contains different signature information, and the signature information contained in each sliced ​​signature image has different importance. Therefore, it is unfair if the local network model assigns the same weight to different target local features. Therefore, in the embodiment of the present application, the global network model considers the global features, adaptively slices the preprocessed target image, and assigns different weights to the signature information contained in different slices.

[0084] Optionally, the target signature image can be input into a global network model, which extracts global features from the preprocessed target signature image based on model parameters to obtain target global features; further, the fully connected layer in the global network model outputs a target weight vector of the target local features.

[0085] Exemplarily, in an embodiment of the present application, the global network model also adopts the structure of DenseNet-36, inputs a preprocessed target signature image of size 512×224, outputs a global feature of 244×16×7, and expands it into a one-dimensional feature of 1×27328; at the same time, a fully connected layer and a softmax activation function are added at the end of the global network model, and a target weight vector is output to provide adaptive slice selection weights for different target local features.

[0086] S104, authenticating the offline signature information according to the target global feature, the target local feature, the target weight vector of the target local feature and the reference signature image of the target user.

[0087] The reference signature image is an image containing the real signature information of the target user.

[0088] Optionally, the target local features may be weighted summed based on the target weight vector of the target local features; further, the target global features and the target local features after the weighted summation may be statistically analyzed to obtain a total feature that comprehensively considers the local features and the global features.

[0089] At the same time, the features in the target user's pre-reference signature image are extracted and compared with the total features for consistency. If the consistency comparison result is consistent, the verification result of the offline signature information is determined to be a genuine signature; correspondingly, if the consistency comparison result is inconsistent, the verification result of the offline signature information is a forged signature.

[0090] In the above signature authentication method, after obtaining the target signature image containing the offline signature information of the target user, the target local features of the target signature image are obtained based on the trained local network model and the target signature image, and the target signature image is input into the trained global network model to obtain the target global features and the target weight vector of the target local features of the target signature image; further, the offline signature information is authenticated based on the target global features, the target local features, the target weight vector of the target local features and the reference signature image of the target user. The above scheme, by introducing the global network model and the local network model, extracts the target global features and the target local features of the target signature image, and obtains the target weight vector of the target local features, which is equivalent to comprehensively considering the global information and local information of the target signature image, and on the basis of assigning the target weight vector to each target local feature, ensures the accuracy of verifying the offline signature information in the target signature image.

[0091] Optionally, in order to ensure the accuracy of the determined target local features, based on the above embodiments, in one embodiment, Figure 2 As shown, a method for determining a local feature of a target is provided, which specifically includes the following steps:

[0092] S201, preprocessing the target signature image.

[0093] Optionally, in order to ensure a better extraction effect of the trained local network model, the target signature image needs to be preprocessed, for example, denoising the multi-target signature image.

[0094] Optionally, in order to achieve the best performance effect of the preprocessed target signature image, the target signature image can be binarized based on the maximum inter-class variance method to obtain a binarized image; the binarized image can be scaled to obtain the preprocessed target signature image; wherein the size ratio of the binarized image is the same as the size ratio of the preprocessed target signature image. Specifically, based on the maximum inter-class variance method, the target signature image can be binarized to reduce the interference of background noise in the target signature image; further, the binarized image can be scaled to make the scaled target signature image meet the requirements of the local network model.

[0095] In addition, in order to prevent the offline signature information in the scaled target signature image from being deformed, it is necessary to ensure that the size ratio of the scaled target signature image (i.e., the pre-processed target signature image) is the same as the size ratio of the binarized image, for example, the aspect ratio of the pre-processed target signature image is the same as the aspect ratio of the binarized image. It is understandable that by performing pre-processing processes such as binarization and scaling on the target signature image, the background noise in the target signature image can be eliminated, so that the pre-processed target signature image can obtain a better expression effect.

[0096] It should be noted that before the target signature image is input into the trained global network model, the target signature image may also be preprocessed as described above.

[0097] S202, slicing the preprocessed target signature image.

[0098] The sliced ​​signature image is an image obtained by slicing the preprocessed target signature image, and each sliced ​​signature image contains part of the signature information in the offline signature information in the preprocessed target signature image.

[0099] Optional, such as Figure 3 , which is a schematic diagram of slicing the preprocessed target signature image. The preprocessed target signature image can be sliced ​​horizontally according to a preset fixed step size, so as to obtain a slice signature image obtained by slicing the preprocessed target signature image.

[0100] S203, inputting the slice signature image obtained by the slice processing into the local network model to obtain the target local features of the target signature image.

[0101] The target local feature is a local feature extracted from the slice signature image, and each slice signature image corresponds to a target local feature. The local network model is a model for extracting features from the slice signature image.

[0102] Optionally, each slice signature image may be input into a local network model, so that the local network model extracts features in each slice signature image according to model parameters to obtain target local features of each slice image.

[0103] Exemplarily, in an embodiment of the present application, the local network model adopts the structure of DenseNet-36 (densely connected convolutional network), and the input of the local network model is a slice signature image of size 244×244, and the output is a local feature of 244×7×7, which can be expanded into a one-dimensional feature of 1×11956, i.e., the target local feature.

[0104] In this embodiment, different weights are adaptively selected for different slice signature images through a global network model, replacing the existing operation of manually selecting weights, extracting more discriminative features for different slice signature images, and ensuring the adaptability and effectiveness of the weights of each target local feature.

[0105] Optionally, it can be seen from the above embodiments that both the global network model and the local network model can be constructed based on the structure of DenseNet-36. Figure 4A As shown, a training method for a global network model and a local network model is provided, which specifically includes the following steps:

[0106] S401, obtaining a sample signature of a sample user.

[0107] Among them, the sample signatures include real signatures and forged signatures; the real signature is the real signature of the sample user; the signature in front of the sample is forged by the sample user.

[0108] Optionally, the real signatures and forged signatures of different sample users may be selected and used as sample signatures of the sample users. It should be noted that the real signatures and forged signatures are both offline signatures of the sample users.

[0109] S402, determining a signature image and a signature slice image of a sample signature.

[0110] Among them, the signature image includes the real signature image and the forged signature image; the real signature image is an image containing the real signature of the sample user, and the forged signature image is an image containing the forged signature of the sample user. The signature slice image is an image obtained by slicing the signature image; the signature slice image includes the real signature slice image and the forged signature slice image.

[0111] Optionally, the sample signature is scanned to obtain a signature image of the sample signature; further, the real signature image and the forged signature image in the signature image are sliced ​​according to the same rule to obtain a real signature slice image and a forged signature slice image, namely, a signature slice image.

[0112] S403, inputting the signature image into the initial global network model to obtain the global features of the signature image and the slicing weight vector obtained by adaptively slicing the signature image by the initial global network model.

[0113] The global features include the global features of the real signature image and the global features of the forged signature image.

[0114] Optionally, the real signature image and the forged signature image in the signature image are respectively input into the initial global network model, so that the global network model processes the signature image based on the model parameters, and outputs the global features of the real signature image and the global features of the forged signature image, that is, the global features of the signature image, and the slicing weight vector obtained by the initial global network model by adaptively slicing the signature image.

[0115] S404, inputting the signature slice image into the initial local network model to obtain local features of the signature slice image.

[0116] The local features include local features of the real signature slice image and local features of the forged signature slice image.

[0117] Optionally, the real signature slice image and the forged signature slice image in the signature slice image can be input into the initial local network model, so that the local network model processes the signature slice image according to the model parameters, and outputs the local features of the real signature slice image and the local features of the forged signature slice image, that is, the local features of the signature slice image.

[0118] S405: construct a global loss function according to the global features of the signature image.

[0119] Among them, the global loss function is the loss function of training the initial global network model.

[0120] Optionally, a global loss function may be constructed based on preset rules and global features of the signature image.

[0121] For example, in order to ensure the accuracy and rationality of the constructed global loss function, for each sample user, a real signature image can be selected from the real signature images in the signature images as an anchor point in advance, and the global feature of the anchor point is ; Except for the selected anchor points, the global features of the real signature image are ,in, Indicates The global features of the real signature image, represents the number of real signature images; the global feature of forged signature images is ,in, Indicates The global features of the forged signature images, Indicates the number of forged signature images.

[0122] Furthermore, for the global network model, a global loss function is constructed based on the above anchor points, the global features of the real signature image, and the global features of the forged signature image. The global loss function of a sample user can be expressed by the following formula (1):

[0123] (1)

[0124] in, For the The global loss function for each user; is the spacing margin.

[0125] At the same time, the sample weight vector output by the global network model is ;in, Indicates The sample weights of the slices, is the number of slices.

[0126] S406, constructing a local loss function according to the local features of the signature slice image and the slice weight vector.

[0127] Among them, the local loss function is the loss function of training the initial local network model.

[0128] Optionally, a local loss function may also be constructed based on preset rules, according to local features of the signature slice image, and the slice weight vector.

[0129] For example, for each sample user, the real signature image selected in S405 is also used as an anchor point; then the local feature corresponding to the real signature image as the anchor point is ,in, Indicates the first The local features of the real signature slice image, represents the number of true signature slice images; except for the selected anchor points, the local features of the true signature slice images are ,in, Indicates The real signature image corresponds to The local features of the real signature slice image; the global features of the forged signature slice image corresponding to the forged signature image are ,in, Indicates The forged signature image corresponds to Local features of forged signature slice images.

[0130] Furthermore, for the local network model, a local loss function is constructed. Specifically, The local loss function of a sample user can be expressed by the following formula (2):

[0131] (2)

[0132] in, In the local network model The triplet loss function for each user; is the spacing margin.

[0133] S407, determining a total loss function according to the global loss function and the local loss function.

[0134] Among them, the total loss function is the loss function for jointly training the global network model and the local network model.

[0135] Optionally, the average of the sum of the global loss function and the local loss function of each sample user can be used as the total loss function. Specifically, the total loss function can be expressed by the following formula (3):

[0136] (3)

[0137] in, is the total loss function, is the number of sample users.

[0138] S408, using a global loss function, a local loss function and a total loss function, jointly train the initial global network model and the initial local network model to obtain a trained global network model and a trained local network model.

[0139] Optionally, a global loss function can be used to train the initial global network model, and a local loss function can be used to train the initial local network model. Furthermore, the total loss function can be used to jointly train the initial global network model and the initial local network model to obtain trained global network models and local network models.

[0140] Optional, such as Figure 4B , which is another schematic diagram of the training process of the global network model and the local network model.

[0141] For example, Figure 4CAs shown, an example of the training process of a global network model and a local network model is provided. First, the entire signature image containing the offline signature of the sample user is sliced ​​to obtain signature slices; then, the entire signature image and the signature slices are preprocessed respectively; further, the global features of the entire signature image are extracted through the global branch network (i.e., the global network model), and the weight vectors of each signature slice output after the global branch network adaptively slices the entire signature image are obtained, and the local features of the signature slices are extracted through the local branch network (i.e., the local network model). Finally, based on the above formula (1), a global loss function is constructed according to the global features of the entire signature image; at the same time, based on the above formula (2), a local loss function is constructed according to the local features of the signature slices and the weight vectors of each signature slice output by the global network branch.

[0142] Furthermore, based on the above formula (3), a total loss function is constructed according to the global loss function and the local loss function, and the global branch network (i.e., the global network model) and the local branch network (i.e., the local network model) are trained according to the global loss function, the local loss function and the total loss function.

[0143] In this embodiment, a global loss function is constructed, and a local loss function is constructed based on the slice weight vector obtained by adaptively slicing the signature image output by the global network model. The initial global network model and the initial local network model are further jointly trained based on the global loss function, the local loss function and the total loss function, thereby ensuring the effectiveness and accuracy of the trained global network model and the local network model.

[0144] Optionally, in order to accurately authenticate the offline signature information, in one embodiment, Figure 5 As shown, a method for authenticating offline signature information is provided, which specifically includes the following steps:

[0145] S501, obtaining reference global features and reference local features of a reference signature image of a target user.

[0146] Among them, the reference signature image is an image containing the real offline signature of the target user; the reference global feature is the feature obtained by extracting the reference signature image; the reference local feature is the feature extracted after slicing the reference signature image.

[0147] Optionally, the signature extraction network in the above embodiment may be used to extract global features and local features of the reference signature image to obtain reference global features and reference local features of the reference signature image of the target user.

[0148] S502: Determine the global Euclidean distance of the target signature image according to the reference global feature and the target global feature.

[0149] Among them, the global Euclidean distance is the distance between the reference global feature and the target global feature in the feature space.

[0150] Optionally, based on the distance calculation definition, the Euclidean distance between the reference global feature and the target global feature is calculated according to the reference global feature and the target global feature as the global Euclidean distance. Specifically, the global Euclidean distance can be determined by the following formula (4):

[0151] (4)

[0152] in, is the global Euclidean distance; is the target global feature; is the reference global feature.

[0153] S503, determining a local Euclidean distance of the target signature image according to the reference local feature, the target local feature and the target weight vector of the target local feature.

[0154] Among them, the local Euclidean distance is the distance between the reference local feature and the target local feature in the feature space.

[0155] Optionally, based on the distance calculation definition, combined with the target weight vector of the target local feature, the Euclidean distance between the reference local feature and the target local feature is calculated according to the reference local feature and the target local feature as the local Euclidean distance. Specifically, the local Euclidean distance can be determined by the following formula (5):

[0156] (5)

[0157] in, is the local Euclidean distance; is the number of local features of the target; is the target weight vector of the target local feature. The weight of the target feature vector; For the target feature vector; For the A reference local feature.

[0158] S504: Authenticate the offline signature information according to the global Euclidean distance and the local Euclidean distance.

[0159] Optionally, the sum of the global Euclidean distance and the local Euclidean distance may be compared with a preset authentication threshold, and the offline signature information may be authenticated based on the comparison result.

[0160] In this embodiment, by introducing reference global features and reference local features and calculating the global Euclidean distance and the local Euclidean distance, the rationality and accuracy of the authentication result of the offline signature information are guaranteed.

[0161] Optionally, based on the above embodiment, in one embodiment, as Figure 6 As shown, a method for authenticating offline signature information is provided, which specifically includes the following steps:

[0162] S601, obtaining the intra-class distance of the reference signature image.

[0163] The intra-class distance is the mean square distance between the reference signature images.

[0164] Optionally, the intra-class distance of each reference signature image may be calculated based on the definition of the intra-class distance. It should be noted that if the number of reference signature images is 1, the intra-class distance is 1.

[0165] S602: Taking the sum of the global Euclidean distance and the local Euclidean distance as the overall Euclidean distance of the target signature image.

[0166] The overall Euclidean distance is the Euclidean distance obtained by combining the global Euclidean distance and the local Euclidean distance.

[0167] Optionally, the global Euclidean distance and the local Euclidean distance may be directly added together, and the sum obtained is used as the overall Euclidean distance of the target signature image. Specifically, the overall Euclidean distance may be expressed by the following formula (6):

[0168] (6)

[0169] in, represents the overall Euclidean distance.

[0170] S603: Authenticate the offline signature information in the target signature image according to the overall Euclidean distance, the number of images of the reference signature image, and the intra-class distance.

[0171] In this embodiment, by introducing the intra-class distance and verifying the offline signature information according to the intra-class distance and the overall Euclidean distance, the accuracy and rationality of the verification result are further guaranteed.

[0172] Optionally, based on the above embodiment, in one embodiment, as Figure 7 As shown, a method for authenticating offline signature information is provided, which specifically includes the following steps:

[0173] S701, determining a normalized minimum distance according to the overall Euclidean distance, the number of images of the reference signature image, and the intra-class distance.

[0174] The normalized minimum distance is the normalized minimum distance between the overall Euclidean distance and the intra-class distance of the reference signature image.

[0175] Optionally, according to the overall Euclidean distance, the number of images of the reference signature image, and the intra-class distance, the process of determining the normalized minimum distance can be expressed by the following formula (7):

[0176] (7)

[0177] in, is the normalized minimum distance; The number of images that are reference signature images; is the intra-class distance.

[0178] S702: If the normalized minimum distance is less than the distance threshold, it is determined that the authentication result of the offline signature information is a true signature.

[0179] The distance threshold is a preset threshold, which can be set according to actual conditions or determined through experiments.

[0180] Optionally, if the normalized distance is less than the distance threshold, it means that the offline signature information in the target signature image matches the real signature information in the reference signature image, and the authentication result of the offline signature information can be determined to be a real signature.

[0181] S703: If the normalized minimum distance is greater than the distance threshold, determine that the authentication result is a forged signature.

[0182] Optionally, if the normalized distance is greater than the distance threshold, it means that the offline signature information in the target signature image does not match the real signature information in the reference signature image, and the authentication result of the offline signature information can be determined to be a forged signature.

[0183] In this embodiment, the normalized minimum distance is introduced, and the offline signature information is verified according to the normalized minimum distance and the distance threshold, thereby ensuring the accuracy and precision of the verification result.

[0184] Figure 8 FIG. 2 is a flow chart of a signature authentication method in another embodiment. Based on the above embodiment, this embodiment provides an optional example of a signature authentication method. Figure 8 The specific implementation process is as follows:

[0185] S801, obtaining a target signature image of a target user.

[0186] The target signature image contains the offline signature information of the target user.

[0187] S802, pre-processing the target signature image.

[0188] Optionally, based on the maximum inter-class variance method, the target signature image is binarized to obtain a binarized image; the binarized image is scaled to obtain a preprocessed target signature image; wherein the size ratio of the binarized image is the same as the size ratio of the preprocessed target signature image.

[0189] S803, slicing the pre-processed target signature image.

[0190] S804, inputting the slice signature image obtained by the slice processing into the local network model to obtain the target local features of the target signature image.

[0191] S805, inputting the preprocessed target signature image into the global network model to obtain a target weight vector of a target global feature and a target local feature of the target signature image.

[0192] S806, obtaining reference global features and reference local features of the reference signature image of the target user.

[0193] S807, determining the global Euclidean distance of the target signature image according to the reference global feature and the target global feature.

[0194] S808, determining a local Euclidean distance of the target signature image according to the reference local feature, the target local feature and the target weight vector of the target local feature.

[0195] S809, obtaining the intra-class distance of the reference signature image.

[0196] S810: Taking the sum of the global Euclidean distance and the local Euclidean distance as the overall Euclidean distance of the target signature image.

[0197] S811, determining a normalized minimum distance according to the overall Euclidean distance, the number of images of the reference signature image, and the intra-class distance.

[0198] S812, determine whether the normalized minimum distance is less than the distance threshold; if so, execute S813; if not, execute S814.

[0199] S813, determining that the authentication result of the offline signature information is a genuine signature.

[0200] S814, determining that the authentication result of the offline signature information is a forged signature.

[0201] The specific process of the above S801-S814 can be found in the description of the above method embodiment. The implementation principle and technical effect are similar and will not be repeated here.

[0202] It should be understood that, although the various steps in the flowcharts involved in the above-mentioned embodiments are displayed in sequence according to the indication of the arrows, these steps are not necessarily executed in sequence according to the order indicated by the arrows. Unless there is a clear explanation in this article, the execution of these steps does not have a strict order restriction, and these steps can be executed in other orders. Moreover, at least a part of the steps in the flowcharts involved in the above-mentioned embodiments can include multiple steps or multiple stages, and these steps or stages are not necessarily executed at the same time, but can be executed at different times, and the execution order of these steps or stages is not necessarily carried out in sequence, but can be executed in turn or alternately with other steps or at least a part of the steps or stages in other steps.

[0203] Based on the same inventive concept, the embodiment of the present application also provides a signature authentication device for implementing the above-mentioned signature authentication method. The implementation solution provided by the device to solve the problem is similar to the implementation solution recorded in the above-mentioned method, so the specific limitations in one or more signature authentication device embodiments provided below can refer to the limitations on the signature authentication method above, and will not be repeated here.

[0204] In one embodiment, Fig. 9 As shown, a signature authentication device 90 is provided, comprising: an image acquisition module 10, a feature determination module 20 and a signature authentication module 30, wherein:

[0205] The image acquisition module 10 is used to acquire a target signature image of a target user; wherein the target signature image includes offline signature information of the target user.

[0206] The first extraction module 20 is used to obtain the target local features of the target signature image based on the trained local network model and the target signature image.

[0207] The second extraction module 30 is used to input the target signature image into the trained global network model to obtain the target global features of the target signature image and the target weight vector of the target local features.

[0208] The signature authentication module 40 is used to authenticate the offline signature information according to the target global feature, the target local feature, the target weight vector of the target local feature and the reference signature image of the target user.

[0209] After acquiring the target signature image containing the offline signature information of the target user, the above-mentioned signature authentication device inputs the target signature image into the signature extraction network model, obtains the target local features of the target signature image based on the trained local network model and the target signature image, and inputs the target signature image into the trained global network model to obtain the target global features and the target weight vector of the target local features of the target signature image, and further authenticates the offline signature information based on the target global features, the target local features, the target weight vector of the target local features and the reference signature image of the target user. The above-mentioned scheme, by introducing the global network model and the local network model, extracts the target global features and the target local features of the target signature image, and obtains the target weight vector of the target local features, which is equivalent to comprehensively considering the global information and local information of the target signature image, and on the basis of assigning the target weight vector to each target local feature, ensures the accuracy of verifying the offline signature information in the target signature image.

[0210] In one embodiment, the first extraction module 20 includes:

[0211] The image processing unit is used to pre-process the target signature image.

[0212] The image slicing unit is used to slice the preprocessed target signature image.

[0213] The feature extraction unit is used to input the slice signature image obtained by slicing into the local network model to obtain the target local features of the target signature image.

[0214] In one embodiment, the image processing unit is specifically used for:

[0215] Based on the maximum inter-class variance method, the target signature image is binarized to obtain a binarized image; the binarized image is scaled to obtain a preprocessed target signature image; wherein the size ratio of the binarized image is the same as the size ratio of the preprocessed target signature image.

[0216] In one embodiment, the signature verification module 30 includes:

[0217] The feature acquisition unit is used to acquire the reference global features and reference local features of the reference signature image of the target user.

[0218] The first determining unit is used to determine the global Euclidean distance of the target signature image according to the reference global feature and the target global feature.

[0219] The second determining unit is used to determine the local Euclidean distance of the target signature image according to the reference local feature, the target local feature and the target weight vector of the target local feature.

[0220] The signature authentication unit is used to authenticate the offline signature information according to the global Euclidean distance and the local Euclidean distance.

[0221] In one embodiment, the signature verification unit includes:

[0222] The distance acquisition slave unit is used to obtain the intra-class distance of the reference signature image.

[0223] The distance determination slave unit is used to take the sum of the global Euclidean distance and the local Euclidean distance as the overall Euclidean distance of the target signature image.

[0224] The authentication slave unit is used to authenticate the offline signature information in the target signature image according to the overall Euclidean distance, the number of images of the reference signature image and the intra-class distance.

[0225] In one embodiment, the authentication slave unit is specifically used for:

[0226] The normalized minimum distance is determined according to the overall Euclidean distance, the number of images of the reference signature image and the intra-class distance; if the normalized minimum distance is less than the distance threshold, the authentication result of the offline signature information is determined to be a genuine signature; if the normalized minimum distance is greater than the distance threshold, the authentication result is determined to be a forged signature.

[0227] In one embodiment, the signature verification device 90 further includes:

[0228] Obtain sample signatures of sample users; wherein the sample signatures include real signatures and forged signatures; determine the signature image and signature slice image of the sample signature; input the signature image into the initial global network model to obtain the global features of the signature image, and the slice weight vector obtained by the initial global network model adaptively slicing the signature image; input the signature slice image into the initial local network model to obtain the local features of the signature slice image; construct a global loss function based on the global features of the signature image; construct a local loss function based on the local features of the signature slice image and the slice weight vector; determine the total loss function based on the global loss function and the local loss function; use the global loss function, the local loss function and the total loss function to jointly train the initial global network model and the initial local network model to obtain the trained global network model and the local network model.

[0229] Each module in the above-mentioned signature verification device can be implemented in whole or in part by software, hardware or a combination thereof. Each module can be embedded in or independent of a processor in a computer device in the form of hardware, or can be stored in a memory in a computer device in the form of software, so that the processor can call and execute the operations corresponding to each module above.

[0230] In an exemplary embodiment, a computer device is provided. The computer device may be a server, and its internal structure diagram may be as shown in FIG. Fig.10 As shown. The computer device includes a processor, a memory, an input / output interface (Input / Output, referred to as I / O) and a communication interface. The processor, the memory and the input / output interface are connected through a system bus, and the communication interface is connected to the system bus through the input / output interface. The processor of the computer device is used to provide computing and control capabilities. The memory of the computer device includes a non-volatile storage medium and an internal memory. The non-volatile storage medium stores an operating system, a computer program and a database. The internal memory provides an environment for the operation of the operating system and the computer program in the non-volatile storage medium. The database of the computer device is used to store data such as the target signature image of the target user. The input / output interface of the computer device is used to exchange information between the processor and an external device. The communication interface of the computer device is used to communicate with an external terminal through a network connection. When the computer program is executed by the processor, a signature authentication method is implemented.

[0231] Those skilled in the art will understand that Fig.10 The structure shown in the figure is only a block diagram of a part of the structure related to the solution of the present application, and does not constitute a limitation on the computer device to which the solution of the present application is applied. The specific computer device may include more or fewer components than those shown in the figure, or combine certain components, or have a different arrangement of components.

[0232] In one embodiment, a computer device is provided, including a memory and a processor, wherein a computer program is stored in the memory, and when the processor executes the computer program, the following steps are implemented:

[0233] Obtaining a target signature image of a target user; wherein the target signature image includes offline signature information of the target user;

[0234] Based on the trained local network model, the target local features of the target signature image are obtained according to the target signature image;

[0235] Input the target signature image into the trained global network model to obtain the target weight vector of the target global feature and the target local feature of the target signature image;

[0236] The offline signature information is authenticated according to the target global feature, the target local feature, the target weight vector of the target local feature and the reference signature image of the target user.

[0237] In one embodiment, when the processor executes the computer program to obtain the target local features of the target signature image based on the trained local network model according to the target signature image, the processor further implements the following steps:

[0238] Preprocessing the target signature image; slicing the preprocessed target signature image; inputting the sliced ​​signature image obtained by the slicing process into the trained local network model to obtain the target local features of the target signature image.

[0239] In one embodiment, when the processor executes the computer program to pre-process the target signature image, the processor also implements the following steps:

[0240] Based on the maximum inter-class variance method, the target signature image is binarized to obtain a binarized image; the binarized image is scaled to obtain a preprocessed target signature image; wherein the size ratio of the binarized image is the same as the size ratio of the preprocessed target signature image.

[0241] In one embodiment, when the processor executes the computer program to authenticate the offline signature information according to the target global feature, the target local feature, the target weight vector of the target local feature, and the reference signature image of the target user, the following steps are also implemented:

[0242] Obtain reference global features and reference local features of a reference signature image of a target user; determine a global Euclidean distance of a target signature image based on the reference global features and the target global features; determine a local Euclidean distance of the target signature image based on the reference local features, the target local features and the target weight vector of the target local features; authenticate offline signature information based on the global Euclidean distance and the local Euclidean distance.

[0243] In one embodiment, when the processor executes the computer program to authenticate the offline signature information according to the global Euclidean distance and the local Euclidean distance, the processor further implements the following steps:

[0244] The intra-class distance of the reference signature image is obtained; the sum of the global Euclidean distance and the local Euclidean distance is used as the overall Euclidean distance of the target signature image; and the offline signature information in the target signature image is authenticated according to the overall Euclidean distance, the number of images of the reference signature image and the intra-class distance.

[0245] In one embodiment, when the processor executes the computer program to authenticate the offline signature information in the target signature image according to the overall Euclidean distance, the number of images of the reference signature image, and the intra-class distance, the processor further implements the following steps:

[0246] The normalized minimum distance is determined according to the overall Euclidean distance, the number of images of the reference signature image and the intra-class distance; if the normalized minimum distance is less than the distance threshold, the authentication result of the offline signature information is determined to be a genuine signature; if the normalized minimum distance is greater than the distance threshold, the authentication result is determined to be a forged signature.

[0247] In one embodiment, when the processor executes the computer program, the processor further implements the following steps:

[0248] Obtain sample signatures of sample users; wherein the sample signatures include real signatures and forged signatures; determine the signature image and signature slice image of the sample signature; input the signature image into the initial global network model to obtain the global features of the signature image, and the slice weight vector obtained by the initial global network model adaptively slicing the signature image; input the signature slice image into the initial local network model to obtain the local features of the signature slice image; construct a global loss function based on the global features of the signature image; construct a local loss function based on the local features of the signature slice image and the slice weight vector; determine the total loss function based on the global loss function and the local loss function; use the global loss function, the local loss function and the total loss function to jointly train the initial global network model and the initial local network model to obtain the trained global network model and the local network model.

[0249] In one embodiment, a computer readable storage medium is provided, on which a computer program is stored, and when the computer program is executed by a processor, the following steps are implemented:

[0250] Obtaining a target signature image of a target user; wherein the target signature image includes offline signature information of the target user;

[0251] Based on the trained local network model, the target local features of the target signature image are obtained according to the target signature image;

[0252] Input the target signature image into the trained global network model to obtain the target weight vector of the target global feature and the target local feature of the target signature image;

[0253] The offline signature information is authenticated according to the target global feature, the target local feature, the target weight vector of the target local feature and the reference signature image of the target user.

[0254] In one embodiment, when the processor executes the computer program to obtain the target local features of the target signature image based on the trained local network model according to the target signature image, the processor further implements the following steps:

[0255] Preprocessing the target signature image; slicing the preprocessed target signature image; inputting the sliced ​​signature image obtained by the slicing process into the trained local network model to obtain the target local features of the target signature image.

[0256] In one embodiment, when the processor executes the computer program to pre-process the target signature image, the processor also implements the following steps:

[0257] Based on the maximum inter-class variance method, the target signature image is binarized to obtain a binarized image; the binarized image is scaled to obtain a preprocessed target signature image; wherein the size ratio of the binarized image is the same as the size ratio of the preprocessed target signature image.

[0258] In one embodiment, when the processor executes the computer program to authenticate the offline signature information according to the target global feature, the target local feature, the target weight vector of the target local feature, and the reference signature image of the target user, the following steps are also implemented:

[0259] Obtain reference global features and reference local features of a reference signature image of a target user; determine a global Euclidean distance of a target signature image based on the reference global features and the target global features; determine a local Euclidean distance of the target signature image based on the reference local features, the target local features and the target weight vector of the target local features; authenticate offline signature information based on the global Euclidean distance and the local Euclidean distance.

[0260] In one embodiment, when the processor executes the computer program to authenticate the offline signature information according to the global Euclidean distance and the local Euclidean distance, the processor further implements the following steps:

[0261] The intra-class distance of the reference signature image is obtained; the sum of the global Euclidean distance and the local Euclidean distance is used as the overall Euclidean distance of the target signature image; and the offline signature information in the target signature image is authenticated according to the overall Euclidean distance, the number of images of the reference signature image and the intra-class distance.

[0262] In one embodiment, when the processor executes the computer program to authenticate the offline signature information in the target signature image according to the overall Euclidean distance, the number of images of the reference signature image, and the intra-class distance, the processor further implements the following steps:

[0263] The normalized minimum distance is determined according to the overall Euclidean distance, the number of images of the reference signature image and the intra-class distance; if the normalized minimum distance is less than the distance threshold, the authentication result of the offline signature information is determined to be a genuine signature; if the normalized minimum distance is greater than the distance threshold, the authentication result is determined to be a forged signature.

[0264] In one embodiment, when the processor executes the computer program, the processor further implements the following steps:

[0265] Obtain sample signatures of sample users; wherein the sample signatures include real signatures and forged signatures; determine the signature image and signature slice image of the sample signature; input the signature image into the initial global network model to obtain the global features of the signature image, and the slice weight vector obtained by the initial global network model adaptively slicing the signature image; input the signature slice image into the initial local network model to obtain the local features of the signature slice image; construct a global loss function based on the global features of the signature image; construct a local loss function based on the local features of the signature slice image and the slice weight vector; determine the total loss function based on the global loss function and the local loss function; use the global loss function, the local loss function and the total loss function to jointly train the initial global network model and the initial local network model to obtain the trained global network model and the local network model.

[0266] In one embodiment, a computer program product is provided, comprising a computer program, which, when executed by a processor, implements the following steps:

[0267] Obtaining a target signature image of a target user; wherein the target signature image includes offline signature information of the target user;

[0268] Based on the trained local network model, the target local features of the target signature image are obtained according to the target signature image;

[0269] Input the target signature image into the trained global network model to obtain the target weight vector of the target global feature and the target local feature of the target signature image;

[0270] The offline signature information is authenticated according to the target global feature, the target local feature, the target weight vector of the target local feature and the reference signature image of the target user.

[0271] In one embodiment, when the processor executes the computer program to obtain the target local features of the target signature image based on the trained local network model according to the target signature image, the processor further implements the following steps:

[0272] Preprocessing the target signature image; slicing the preprocessed target signature image; inputting the sliced ​​signature image obtained by the slicing process into the trained local network model to obtain the target local features of the target signature image.

[0273] In one embodiment, when the processor executes the computer program to pre-process the target signature image, the processor also implements the following steps:

[0274] Based on the maximum inter-class variance method, the target signature image is binarized to obtain a binarized image; the binarized image is scaled to obtain a preprocessed target signature image; wherein the size ratio of the binarized image is the same as the size ratio of the preprocessed target signature image.

[0275] In one embodiment, when the processor executes the computer program to authenticate the offline signature information according to the target global feature, the target local feature, the target weight vector of the target local feature, and the reference signature image of the target user, the following steps are also implemented:

[0276] Obtain reference global features and reference local features of a reference signature image of a target user; determine a global Euclidean distance of a target signature image based on the reference global features and the target global features; determine a local Euclidean distance of the target signature image based on the reference local features, the target local features and the target weight vector of the target local features; authenticate offline signature information based on the global Euclidean distance and the local Euclidean distance.

[0277] In one embodiment, when the processor executes the computer program to authenticate the offline signature information according to the global Euclidean distance and the local Euclidean distance, the processor further implements the following steps:

[0278] The intra-class distance of the reference signature image is obtained; the sum of the global Euclidean distance and the local Euclidean distance is used as the overall Euclidean distance of the target signature image; and the offline signature information in the target signature image is authenticated according to the overall Euclidean distance, the number of images of the reference signature image and the intra-class distance.

[0279] In one embodiment, when the processor executes the computer program to authenticate the offline signature information in the target signature image according to the overall Euclidean distance, the number of images of the reference signature image, and the intra-class distance, the processor further implements the following steps:

[0280] The normalized minimum distance is determined according to the overall Euclidean distance, the number of images of the reference signature image and the intra-class distance; if the normalized minimum distance is less than the distance threshold, the authentication result of the offline signature information is determined to be a genuine signature; if the normalized minimum distance is greater than the distance threshold, the authentication result is determined to be a forged signature.

[0281] In one embodiment, when the processor executes the computer program, the processor further implements the following steps:

[0282] Obtain sample signatures of sample users; wherein the sample signatures include real signatures and forged signatures; determine the signature image and signature slice image of the sample signature; input the signature image into the initial global network model to obtain the global features of the signature image, and the slice weight vector obtained by the initial global network model adaptively slicing the signature image; input the signature slice image into the initial local network model to obtain the local features of the signature slice image; construct a global loss function based on the global features of the signature image; construct a local loss function based on the local features of the signature slice image and the slice weight vector; determine the total loss function based on the global loss function and the local loss function; use the global loss function, the local loss function and the total loss function to jointly train the initial global network model and the initial local network model to obtain the trained global network model and the local network model.

[0283] It should be noted that the user information (including but not limited to the target signature image of the target user, etc.) and data (including but not limited to data used for analysis, stored data, displayed data, etc.) involved in this application are all information and data authorized by the user or fully authorized by all parties, and the collection, use and processing of relevant data must comply with relevant regulations.

[0284] Those skilled in the art can understand that all or part of the processes in the above-mentioned embodiment methods can be completed by instructing the relevant hardware through a computer program, and the computer program can be stored in a non-volatile computer-readable storage medium. When the computer program is executed, it can include the processes of the embodiments of the above-mentioned methods. Among them, any reference to the memory, database or other medium used in the embodiments provided in this application can include at least one of non-volatile and volatile memory. Non-volatile memory can include read-only memory (ROM), magnetic tape, floppy disk, flash memory, optical memory, high-density embedded non-volatile memory, resistive random access memory (ReRAM), magnetoresistive random access memory (MRAM), ferroelectric random access memory (FRAM), phase change memory (PCM), graphene memory, etc. Volatile memory can include random access memory (RAM) or external cache memory, etc. As an illustration and not limitation, RAM can be in various forms, such as static random access memory (SRAM) or dynamic random access memory (DRAM). The database involved in each embodiment provided in this application may include at least one of a relational database and a non-relational database. Non-relational databases may include distributed databases based on blockchains, etc., but are not limited to this. The processor involved in each embodiment provided in this application may be a general-purpose processor, a central processing unit, a graphics processor, a digital signal processor, a programmable logic device, a data processing logic device based on quantum computing, etc., but are not limited to this.

[0285] The technical features of the above embodiments may be arbitrarily combined. To make the description concise, not all possible combinations of the technical features in the above embodiments are described. However, as long as there is no contradiction in the combination of these technical features, they should be considered to be within the scope of this specification.

[0286] The above-described embodiments only express several implementation methods of the present application, and the descriptions thereof are relatively specific and detailed, but they cannot be understood as limiting the scope of the present application. It should be pointed out that, for a person of ordinary skill in the art, several variations and improvements can be made without departing from the concept of the present application, and these all belong to the protection scope of the present application. Therefore, the protection scope of the present application shall be subject to the attached claims.

Claims

1. A signature authentication method, characterized in that: The method comprises: Acquire a target signature image of a target user; wherein the target signature image includes offline signature information of the target user; Based on the trained local network model, according to the target signature image, a target local feature of the target signature image is obtained; Inputting the target signature image into the trained global network model to obtain the target global features of the target signature image and the target weight vector of the target local features; The offline signature information is authenticated according to the target global feature, the target local feature, the target weight vector of the target local feature and the reference signature image of the target user.

2. The method according to claim 1, characterized in that The method of obtaining the target local features of the target signature image based on the trained local network model according to the target signature image includes: Preprocessing the target signature image; Slice the preprocessed target signature image; The slice signature image obtained by slicing is input into the trained local network model to obtain the target local features of the target signature image.

3. The method according to claim 2, characterized in that The preprocessing of the target signature image comprises: Based on the maximum inter-class variance method, the target signature image is binarized to obtain a binarized image; The binary image is scaled to obtain a preprocessed target signature image; wherein the size ratio of the binary image is the same as the size ratio of the preprocessed target signature image.

4. The method according to claim 1, characterized in that: The authenticating the offline signature information according to the target global feature, the target local feature, the target weight vector of the target local feature and the reference signature image of the target user comprises: Acquire reference global features and reference local features of the reference signature image of the target user; Determining a global Euclidean distance of the target signature image according to the reference global feature and the target global feature; determining a local Euclidean distance of the target signature image according to the reference local feature, the target local feature and a target weight vector of the target local feature; The offline signature information is authenticated according to the global Euclidean distance and the local Euclidean distance.

5. The method according to claim 4, characterized in that The authenticating the offline signature information according to the global Euclidean distance and the local Euclidean distance includes: Obtaining the intra-class distance of the reference signature image; Taking the sum of the global Euclidean distance and the local Euclidean distance as the overall Euclidean distance of the target signature image; The offline signature information in the target signature image is authenticated according to the overall Euclidean distance, the number of images of the reference signature image and the intra-class distance.

6. The method according to claim 5, characterized in that The authenticating the offline signature information in the target signature image according to the overall Euclidean distance, the number of images of the reference signature image and the intra-class distance comprises: determining a normalized minimum distance according to the overall Euclidean distance, the number of images of the reference signature image, and the intra-class distance; If the normalized minimum distance is less than the distance threshold, determining that the authentication result of the offline signature information is a true signature; If the normalized minimum distance is greater than the distance threshold, it is determined that the authentication result is a forged signature.

7. The method according to claim 1, characterized in that The global network model and the local network model are trained in the following ways, including: Obtaining sample signatures of sample users; wherein the sample signatures include real signatures and forged signatures; Determining a signature image and a signature slice image of the sample signature; Inputting the signature image into an initial global network model to obtain global features of the signature image and a slicing weight vector obtained by adaptively slicing the signature image by the initial global network model; Inputting the signature slice image into an initial local network model to obtain local features of the signature slice image; Constructing a global loss function according to the global features of the signature image; Constructing a local loss function according to the local features of the signature slice image and the slice weight vector; Determine a total loss function according to the global loss function and the local loss function; The global loss function, the local loss function and the total loss function are used to jointly train the initial global network model and the initial local network model to obtain the trained global network model and the local network model.

8. A signature authentication device, characterized in that: The device comprises: An image acquisition module, used to acquire a target signature image of a target user; wherein the target signature image contains offline signature information of the target user; A first extraction module is used to obtain target local features of the target signature image based on the trained local network model and the target signature image; A second extraction module is used to input the target signature image into the trained global network model to obtain the target global feature of the target signature image and the target weight vector of the target local feature; The signature authentication module is used to authenticate the offline signature information according to the target global feature, the target local feature, the target weight vector of the target local feature and the reference signature image of the target user.

9. A computer device comprising a memory and a processor, wherein the memory stores a computer program, wherein: When the processor executes the computer program, the steps of the method according to any one of claims 1 to 7 are implemented.

10. A computer-readable storage medium having a computer program stored thereon, characterized in that: When the computer program is executed by a processor, the steps of the method according to any one of claims 1 to 7 are implemented.

Images