Offline signature identification system and method based on SigPConvNeXt model

Abstract

The invention provides an offline signature identification system and method based on a SigPConvNeXt model. The system comprises a feature extraction network PConvNeXt1, a feature extraction network PConvNeXt2 and a classifier, wherein the feature extraction network PConvNeXt2 is twinned with the feature extraction network PConvNeXt1; the PConvNeXt1 and the PConvNeXt2 are respectively used for extracting feature vectors of two offline signature images needing to be compared, and the classifier is used for calculating the distance between the two signature images in the same feature space and judging whether the two signatures are from the same person or not according to a predetermined threshold value; each of the feature extraction networks PConvNeXt1 and PConvNeXt2 comprises an encoder and a decoder, the encoder comprises four stages and a CBAM module connected behind each stage, and the CBAM module comprises a channel attention module and a space attention module; the decoder comprises four decoding stages and a feature pyramid fusion module connected behind each decoding stage. According to the method, the ConvNeXt network is combined with the CBAM module, so that the feature extraction capability is enhanced; and a feature pyramid fusion module is introduced, so that effective fusion of features of different dimensions is realized.

Description

Technical Field

[0001] The present application relates to the technical field of signature authentication, and in particular, to an offline signature authentication system and method based on a SigPConvNeXt model. Background Art

[0002] Definitions of Abbreviations and Key Terms SigPConvNeXt: Definition: SigPConvNeXt (Signature Pyramid ConvNeXt) is an offline signature authentication model based on an improved ConvNeXt network and a twin network structure. It combines deep learning techniques, especially attention mechanism and feature pyramid fusion, to improve the accuracy and efficiency of handwritten signature authentication.

[0003] ConvNeXt: Definition: ConvNeXt is a deep convolutional network model proposed by Facebook AI and UC Berkeley. It draws on the structure and training strategy of Swin-Transformer and improves computational efficiency and accuracy through deep separable convolution and layer normalization.

[0004] CBAM (Convolutional Block Attention Module): Definition: CBAM is an attention mechanism module, which includes a channel attention module (CAM) and a spatial attention module (SAM), which is used to enhance the expressiveness of feature maps in convolutional neural networks. By focusing on more important channels and spatial locations, CBAM increases the sensitivity of the model to key features, thereby improving the performance of classification and recognition tasks.

[0005] FPN (Feature Pyramid Network): Definition: FPN is a feature fusion network structure that achieves the fusion of features of different scales through top-down lateral connections to enhance the model's ability to detect multi-scale objects. FPN can combine high-resolution information at the low level and strong semantic information at the high level, improving the accuracy of tasks such as target detection.

[0006] SVM (Support Vector Machine): Definition: Support vector machine is a supervised learning model used for classification and regression analysis. It distinguishes data points of different categories by finding the best hyperplane. SVM is known for its effectiveness in high-dimensional space and small sample size, especially for small sample learning problems.

[0007] Numpy: Definition: Numpy is an open source Python scientific computing library that provides multidimensional array objects, derived objects (such as masked arrays and matrices), and various routines for fast array operations. Numpy is a basic tool for data processing and scientific computing, especially in array operations, mathematical calculations, and data analysis.

[0008] Biometric authentication is an identity authentication technology based on individual physiological or behavioral characteristics. It uses unique biological characteristics for secure access, such as login devices, access control systems, payment verification, etc. Biometric authentication can be divided into two categories: physiological characteristics and behavioral characteristics. Physiological characteristics are related to individual physical attributes, such as fingerprint, panel, iris, palm print and voiceprint recognition, while behavioral characteristics are related to individual behavior patterns, such as typing rhythm, gait recognition and signature authentication. Everyone's biometrics are generally unique and difficult to forge, which is more secure than traditional passwords, but the collection and storage of biometric data may cause privacy and security issues. In addition, some biometric authentication may be affected by the environment and there is a risk of false positives and false negatives.

[0009] Signature authentication is one of the common biometric authentication technologies. It uses individual signature features to verify identity. It belongs to behavioral feature authentication and is widely used in finance, law, and contract signing. According to the signature collection method and the features obtained, signature authentication can be divided into two types: online signature and offline signature. Online signatures are usually written in real time on the device through a touch screen or a digital pen. They mainly analyze the dynamic features of the signature, such as speed, pressure, and stroke sequence. They are often used in scenarios such as electronic payment and mobile device unlocking. Offline signatures are static images, which are signatures completed on paper. They can be digitized by scanning, etc., but with the development of information technology, they can also be directly collected through devices such as signature pads. The focus is on analyzing the shape, size, angle and other static features of the signature. They are often used in scenarios such as contract signing and document verification.

[0010] Offline signature authentication can be divided into two modes: writer-dependent and writer-independent. The writer-dependent system needs to be retrained to complete the update with the addition of each new writer. For a consumer-based system, new consumers will join every day, which will incur huge update and storage costs. In contrast, the writer-independent mode will establish a general system to simulate the difference between real signatures and forged signatures. During the training process, the available signatures will be divided into training sets and test sets. For a single signer, his signature is divided into two types of tuples: similar (real signature, real signature) or dissimilar (real signature, forged signature). This process is applicable to all signers in the training set and test set to build training and test instances for the classifier. Therefore, the writer-independent mode is more desirable and more universal than the writer-dependent mode. This paper will also study the writer-independent signature authentication method.

[0011] The research on offline signature authentication methods is divided into two stages: manual features and deep learning. In the early stage of research, manual processing was mostly performed on static images, and various features (global feature extraction such as block codes, wavelets and Fourier sequences, etc., geometric and topological features of local attributes such as position, tangent direction, blob structure, connected components and curvature) were obtained on this basis for research. There have also been projection and contour-based methods (based on directional contours, surrounding features, grid-based methods, geometric moment-based methods and texture feature-based methods) and structural methods that consider the relationship between local features (such as graph matching and the recently proposed compact correlation features). With the rapid development of neural networks and deep learning in recent years, more and more researchers have begun to use deep learning to solve the problem of offline signature authentication. On the other hand, the twin network consists of two networks, accepting two tuples of images for feature extraction. At the same time, there is a calculation of the distance between the two images in the feature space at the top of the network. At the same time, the parameters between the two networks are shared, which in turn ensures that similar images cannot be mapped to very different positions in the feature space. The characteristics of the twin network make it very popular in different verification tasks such as signature verification and face verification.

[0012] The patent document entitled “Multi-scale network offline handwritten signature identification method based on attention mechanism” discloses a multi-scale network offline handwritten signature identification method based on attention mechanism, which includes the following steps: S1: Get the original image of offline signature: Read the original image data and convert it into Numpy format; S2: Preprocessing based on the above signature image using Gaussian filtering and OTSU algorithm: Gaussian filtering is performed on the Numpy data to remove image noise, and then the OTSU algorithm is used to perform the maximum inter-class variance method to divide the image into background and foreground; S3: Based on the preprocessed signature image data, the proposed network structure is trained: The entire neural network consists of four blocks, and two different convolution blocks are used in each block to convolve the same image or feature map to obtain multi-scale features. The polarized self-attention mechanism is used to find important features, and then feature fusion is performed through ADD or Concat;

[0013] S4: Optimize the model based on the above pre-trained model results: By analyzing the initial and subsequent training logs and test results, adjust the learning rate, decay strategy, weight decay and other parameters to speed up model convergence and improve accuracy; S5: Prediction and signature authentication based on the above-mentioned optimized pre-trained model: Input the test signature image data, obtain the predicted label through the pre-trained model, extract the classification features of the fully connected layer of the model, train the SVM model, use the pre-trained SVM for batch offline signature authentication, and obtain the final authentication result.

[0014] This patented solution has achieved good results in offline signature authentication, but it still has the following shortcomings: 1. Insufficient precision of feature extraction: The existing technology fails to fully utilize the key information in the image and has deficiencies in feature extraction. The feature extraction is not precise enough, which affects the final identification effect.

[0015] 2. Limited generalization ability: Existing technologies perform poorly in generalization across languages ​​or different types of signatures, especially in non-training languages.

[0016] 3. Room for improvement in accuracy: The existing technology still has a lot of room for improvement in accuracy, especially in the identification of complex or forged signatures, where the bit error rate may be high.

[0017] 4. Dependence on training samples: Existing technologies are highly dependent on the quantity and quality of training samples, and the problem of small sample learning has not been effectively solved.

[0018] 5. Model complexity and overfitting problems: Existing technologies have problems in model complexity control and overfitting, resulting in insufficient stability and reliability of the model in practical applications. Summary of the invention

[0019] In view of this, the present application provides an offline signature authentication system and method based on the SigPConvNeXt model to overcome the above-mentioned problems existing in the prior art.

[0020] To achieve the above objectives, the technical solutions adopted in this application are as follows: An offline signature authentication system based on a SigPConvNeXt model comprises a feature extraction network PConvNeXt1, a feature extraction network PConvNeXt2 twinned with the PConvNeXt1, and a classifier; the PConvNeXt1 and PConvNeXt2 are respectively used to extract feature vectors of two offline signature images to be compared, the classifier is used to calculate the distance between the two offline signature images in the same feature space, and then judge whether the two signatures are from the same person according to a predetermined threshold; the PConvNeXt1 and PConvNeXt2 have the same network architecture and share the same network parameters, the feature extraction networks PConvNeXt1 and PConvNeXt2 comprise an encoder and a decoder, the encoder comprises four stages and a CBAM module connected behind each stage, the CBAM module comprises a channel attention module and a spatial attention module, each stage comprises multiple convolutional layers for extracting features at different levels; the decoder comprises four decoding stages and a feature pyramid fusion module connected behind each decoding stage.

[0021] Furthermore, each decoding stage consists of a 1x1 convolution layer, a 3x3 convolution layer and an upsampling module, wherein the 1x1 convolution layer is used to adjust the number of channels of the feature map, the 3x3 convolution layer is used to extract features, and the upsampling is used to restore the spatial resolution of the feature map.

[0022] Furthermore, the feature pyramid fusion module includes a feature fusion module and a downsampling module. The feature fusion module fuses feature maps of different resolutions through a feature fusion operation; the downsampling module is used to perform a downsampling operation after feature fusion to reduce the size of the feature map.

[0023] The present application also provides an offline signature authentication method based on the SigPConvNeXt model. The method is based on the offline signature authentication system and includes: Step 1: Input processing: input two offline signature images to be compared into the feature extraction networks PConvNeXt1 and PConvNeXt2 in the offline signature authentication system respectively; Step 2: Feature extraction: Through the four stages of the encoder, the deep features of the offline signature image are gradually extracted, and the CBAM module after each stage is used to enhance the distinguishing ability of the features; Step 3: Feature fusion: In the decoder, the feature pyramid fusion module is used to integrate features of different scales and enhance the model's ability to capture features of different sizes and directions. Step 4: Output generation: After feature fusion, downsampling is performed to reduce the size of the feature map and generate a feature vector for signature authentication; Step 5: Identification task: Use the generated feature vector to calculate the distance between the two offline signature images in the same feature space through the classifier, and then determine whether the two signatures are from the same person based on a predetermined threshold.

[0024] Furthermore, in step 5, the distance between the two offline signature images in the same feature space is calculated by the classifier, specifically using the Euclidean distance to calculate the distance between the two offline signature images in the same feature space. , the specific formula is: in, and are the feature vectors of the two signature images respectively, is the dimension of the feature space, is the index, indicating the first Features.

[0025] Furthermore, the threshold determination method is an automatic threshold determination method, which specifically includes: 1) After each round of model training, segment the distance between the true-false signature pair and the true-true signature pair to establish Test values: in is the mean distance between true-true signature pairs, is the mean distance between true and false signature pairs; 2) Use each test value Perform signature authentication and calculate the authentication accuracy of the current model under each test value. Take the test value corresponding to the highest accuracy value as the model threshold.

[0026] Compared with the prior art, the beneficial effects of this application are: 1. Improve the precision of feature extraction: By combining the ConvNeXt network and the CBAM module to form a PConvNeXt network, this application can enhance the representativeness of features from two dimensions: channel and space. The channel attention (CAM) and spatial attention (SAM) of the CBAM module can adaptively emphasize important features and key areas, thereby improving the precision of feature extraction. This refined feature extraction helps to more accurately capture the subtle differences in handwritten signatures and improve the accuracy of authenticity identification.

[0027] 2. Enhance the generalization ability of the model: This application adopts a twin network structure and performs training and cross-validation on multiple data sets, which helps the model learn more general feature representations, thereby enhancing the generalization ability of the model in different data sets and language environments. This improvement in generalization ability means that the identification method of this application is not only applicable to specific data sets, but also can adapt to a wider range of application scenarios.

[0028] 3. Improve processing speed and efficiency: The improved ConvNeXt network structure and feature pyramid fusion (PFF) module can efficiently process and fuse features of different scales, reduce the consumption of computing resources, and improve processing speed and efficiency. This enables this application to respond quickly in practical applications and meet real-time or near-real-time identification needs.

[0029] 4. Improve accuracy and reduce error rate: By integrating multi-scale features and attention mechanism, this application can more comprehensively capture the characteristics of handwritten signatures, thereby improving the accuracy of identification. At the same time, the method of automatically determining the threshold can dynamically adjust the optimal threshold according to the model training results, further reducing the error rate and improving the reliability of identification.

[0030] 5. Reduce model complexity and overfitting: The PConvNeXt network of this application effectively balances the complexity and performance of the model and reduces the risk of overfitting through the combination of CBAM module and PFF module. This structure not only improves the stability of the model, but also ensures the performance of the model on unknown data.

[0031] 6. Promote the development of information security in the direction of biometric identification: This application reduces the subjectivity of manual identification and improves the fairness and objectivity of offline signature identification by improving the accuracy and efficiency of offline signature identification. This has important practical application value in the field of information security, especially in the fields of finance, law and contract signing with high security requirements. BRIEF DESCRIPTION OF THE DRAWINGS

[0032] In order to more clearly illustrate the technical solutions of the embodiments of the present application, the drawings required for use in the embodiments will be briefly introduced below. It should be understood that the following drawings only show certain embodiments of the present application and therefore should not be regarded as limiting the scope. For ordinary technicians in this field, other related drawings can be obtained based on these drawings without paying creative work.

[0033] Figure 1 This is the network structure diagram of the feature extraction network PConvNeXt of this application; Figure 2 This is the structure diagram of the offline signature authentication system based on the SigPConvNeXt model of this application; Figure 3This is a flow chart of the offline signature authentication method based on the SigPConvNeXt model of this application. DETAILED DESCRIPTION

[0034] In order to make the purpose, technical solutions and advantages of the embodiments of the present application clearer, the technical solutions in the embodiments of the present application will be clearly and completely described below in combination with the drawings in the embodiments of the present application. Obviously, the described embodiments are only part of the embodiments of the present application, not all of the embodiments.

[0035] The ConvNeXt model is a deep convolutional network model proposed by Facebook AI and UC Berkele. In recent years, transformer-based deep learning models have gradually become mainstream. The Swin-Transformer that appeared in 2021 became the best paper of ICCV that year. ConvNeXt is an improved model based on ResNet by drawing on the Swin-Transformer structure and training strategy. ConvNeXt refers to the proportional strategy of Swin-Transformer and adjusts the four blocks of ResNet-50, adjusting the number of stacking in ResNet-50 from the original (3, 4, 6, 3) to (3, 3, 9,3). ConvNeXt draws on the advantages of depthwise separable convolution, replaces the 3×3 convolution in the bottleneck layer with depthwise separable convolution, and increases the width of the network, thereby achieving improved computational efficiency and accuracy.

[0036] ConvNeXt is mainly composed of 4 stacked stages, each of which contains a downsampling module and multiple ConvNeXt Blocks. The initial layer is split using a convolutional layer with a size of 4×4 and a step size of 4, and layer normalization is used to improve the training stability and convergence speed of the neural network. After that, each stage extracts features by downsampling and stacking ConvNeXt Blocks of different times. The downsampling part is composed of layer normalization and a convolution kernel with a volume size of 2×2 and a step size of 2. Finally, the result is obtained through global average pooling, layer normalization and a fully connected layer. There are four versions of ConvNeXt (T, S, B, L). The specific implementation of this application adopts the ConvNeXt-B version, and its network structure is shown in Table 1.

[0037] Table 1 ConvNeXt-B network structure The Convolutional Block Attention Module (CBAM) consists of two independent submodules: the Channel Attention Module (CAM) and the Spatial Attention Module (SAM), which use attention mechanisms in the channel and space dimensions respectively. This not only saves parameters and computing power, but also ensures that it can be integrated into the existing network architecture as a plug-and-play module.

[0038] The Feature Pyramid Network (FPN) is implemented through top-down lateral connections. The low-level features contain rich positioning information, but this information may be lost when it is passed to the high-level, making it difficult to obtain accurate positioning information. This application uses a bidirectional feature fusion module (BiFFM), which adds a bottom-up path at the end of FPN. Through the bottom-up path enhancement, the accurate positioning information of the low-level is enhanced to the entire feature hierarchy.

[0039] like Figure 1 and 2 As shown, the present application provides an offline signature authentication system based on the SigPConvNeXt model, including a feature extraction network PConvNeXt1, a feature extraction network PConvNeXt2 twinned with the PConvNeXt1, and a classifier; the PConvNeXt1 and PConvNeXt2 are respectively used to extract feature vectors of two offline signature images that need to be compared, and the classifier is used to calculate the distance between the two offline signature images in the same feature space, and then judge whether the two signatures are from the same person according to a predetermined threshold; the PConvNeXt1 and PConvNeXt2 have the same network architecture and share the same network parameters, the feature extraction networks PConvNeXt1 and PConvNeXt2 include an encoder and a decoder, the encoder includes four stages and a CBAM module connected behind each stage, the CBAM module includes a channel attention module and a spatial attention module, each stage includes multiple convolutional layers for extracting features at different levels; the decoder includes four decoding stages and a feature pyramid fusion module connected behind each decoding stage.

[0040] The feature extraction network PConvNeXt (Pyramid ConvNeXt) of this application adds a CBAM module after each Stage of the ConvNeXt model, and uses the pyramid bidirectional feature fusion module to fuse the feature maps processed by each CBAM module to obtain the final feature map. The specific structure is as follows: Figure 1 As shown in Figure 2, the CBAM module is applied after each stage to enhance the expressiveness of the features, and the channel attention and spatial attention mechanisms enable the model to focus on more important areas in the image.

[0041] The entire SigPConvNeXt model, such as Figure 2 As shown in the figure, it is composed of two feature extraction networks PConvNeXt (PConvNeXt1 and PConvNeXt2). The two offline signature images that need to be compared are input into the trained feature extraction network to obtain two feature vectors, which are then passed to the classifier to calculate the distance between the two images in the same feature space. Then, a judgment is made based on a predetermined threshold to output whether the two signatures are from the same person.

[0042] By combining the ConvNeXt network with the CBAM (Convolutional Block Attention Module) module, the feature extraction capability is enhanced. The feature pyramid fusion (PFF) module is introduced to achieve effective fusion of features of different dimensions.

[0043] As a further implementation, each decoding stage consists of a 1x1 convolution layer, a 3x3 convolution layer and an upsampling module, wherein the 1x1 convolution layer is used to adjust the number of channels of the feature map, the 3x3 convolution layer is used to extract features, and the upsampling is used to restore the spatial resolution of the feature map.

[0044] As a further implementation, the feature pyramid fusion module includes a feature fusion module and a downsampling module. The feature fusion module fuses feature maps of different resolutions through a feature fusion operation; the downsampling module is used to perform a downsampling operation after feature fusion to reduce the size of the feature map.

[0045] After each decoding stage, feature maps of different resolutions are fused through feature fusion (P1, P2, P3, P4). After feature fusion, downsampling is performed to reduce the size of the feature map while retaining important features.

[0046] like Figure 3 As shown, the present application also provides an offline signature authentication method based on the SigPConvNeXt model, the method is based on the offline signature authentication system, and the method includes: Step 1: Input processing: input two offline signature images to be compared into the feature extraction networks PConvNeXt1 and PConvNeXt2 in the offline signature authentication system respectively; Step 2: Feature extraction: Through the four stages of the encoder, the deep features of the offline signature image are gradually extracted, and the CBAM module after each stage is used to enhance the distinguishing ability of the features; Step 3: Feature fusion: In the decoder, the feature pyramid fusion module is used to integrate features of different scales and enhance the model's ability to capture features of different sizes and directions. Step 4: Output generation: After feature fusion, downsampling is performed to reduce the size of the feature map and generate a feature vector for signature authentication; Step 5: Identification task: Use the generated feature vector to calculate the distance between the two offline signature images in the same feature space through the classifier, and then determine whether the two signatures are from the same person based on a predetermined threshold.

[0047] As a further implementation, the distance between the two offline signature images in the same feature space is calculated by the classifier in step 5, specifically using the Euclidean distance to calculate the distance between the two offline signature images in the same feature space. , the specific formula is: in, and are the feature vectors of the two signature images respectively, is the dimension of the feature space, is the index, indicating the first Features.

[0048] As a further implementation, the threshold determination method is an automatic threshold determination method, specifically comprising: 1) After each round of model training, segment the distance between the true-false signature pair and the true-true signature pair to establish Test values: in is the mean distance between true-true signature pairs, is the mean distance between true and false signature pairs; 6) Use each test value Perform signature authentication and calculate the current modulus under each verification value The identification accuracy of the model is calculated, and the test value corresponding to the highest accuracy value is taken as the model threshold.

[0049] When training the model, you can use the Generative Adversarial Network (GAN) to generate forged signature samples, increase the number of samples, and enhance the generalization ability of the model. At the same time, you can use the samples generated by GAN to train and test the model to improve the model's ability to identify forged signatures.

[0050] By automatically determining the threshold value, manual intervention is reduced and the degree of automation of the identification process is improved. The automated threshold determination method enables the present application to adapt to different identification environments and conditions, improving the convenience of operation and consistency of results.

[0051] The above are only specific implementations of the present application, but the protection scope of the present application is not limited thereto. Any technician familiar with the technical field can easily think of changes or substitutions within the technical scope disclosed in the present application, which should be included in the protection scope of the present application. Therefore, the protection scope of the present application should be based on the protection scope of the claims.

Claims

1. Offline signature authentication system based on SigPConvNeXt model, characterized in that: It includes a feature extraction network PConvNeXt1, a feature extraction network PConvNeXt2 that is twin to the PConvNeXt1, and a classifier; the PConvNeXt1 and PConvNeXt2 are respectively used to extract feature vectors of two offline signature images that need to be compared, the classifier is used to calculate the distance between the two offline signature images in the same feature space, and then determine whether the two signatures are from the same person according to a predetermined threshold; the PConvNeXt1 and PConvNeXt2 have the same network architecture and share the same network parameters, the feature extraction networks PConvNeXt1 and PConvNeXt2 include an encoder and a decoder, the encoder includes four stages and a CBAM module connected behind each stage, the CBAM module includes a channel attention module and a spatial attention module, each stage includes multiple convolutional layers for extracting features at different levels; the decoder includes four decoding stages and a feature pyramid fusion module connected behind each decoding stage.

2. The offline signature authentication system based on the SigPConvNeXt model as claimed in claim 1, characterized in that: Each decoding stage consists of a 1x1 convolution layer, a 3x3 convolution layer, and an upsampling module. The 1x1 convolution layer is used to adjust the number of channels of the feature map, the 3x3 convolution layer is used to extract features, and the upsampling is used to restore the spatial resolution of the feature map.

3. The offline signature authentication system based on SigPConvNeXt model as claimed in claim 2, characterized in that: The feature pyramid fusion module includes a feature fusion module and a downsampling module. The feature fusion module fuses feature maps of different resolutions through a feature fusion operation; the downsampling module is used to perform a downsampling operation after feature fusion to reduce the size of the feature map.

4. Offline signature authentication method based on SigPConvNeXt model, characterized in that: The method is based on the offline signature authentication system according to any one of claims 1 to 3, and the method comprises: Step 1: Input processing: input two offline signature images to be compared into the feature extraction networks PConvNeXt1 and PConvNeXt2 in the offline signature authentication system respectively; Step 2: Feature extraction: Through the four stages of the encoder, the deep features of the offline signature image are gradually extracted, and the CBAM module after each stage is used to enhance the distinguishing ability of the features; Step 3: Feature fusion: In the decoder, the feature pyramid fusion module is used to integrate features of different scales and enhance the model's ability to capture features of different sizes and directions. Step 4: Output generation: After feature fusion, downsampling is performed to reduce the size of the feature map and generate a feature vector for signature authentication; Step 5: Identification task: Use the generated feature vector to calculate the distance between the two offline signature images in the same feature space through the classifier, and then determine whether the two signatures are from the same person based on a predetermined threshold.

5. The offline signature authentication method based on the SigPConvNeXt model as claimed in claim 4, characterized in that: In step 5, the distance between the two offline signature images in the same feature space is calculated by the classifier, specifically, the distance between the two offline signature images in the same feature space is calculated by using the Euclidean distance. , the specific formula is: in, and are the feature vectors of the two signature images respectively.

6. The offline signature authentication method based on SigPConvNeXt model as claimed in claim 5, characterized in that: The method for determining the threshold is an automatic threshold determination method, which specifically includes: 1) After each round of model training, segment the distance between the true-false signature pair and the true-true signature pair to establish Test values: in is the mean distance between true-true signature pairs, is the mean distance between true and false signature pairs; 2) Use each test value Perform signature authentication and calculate the authentication accuracy of the current model under each test value. Take the test value corresponding to the highest accuracy value as the model threshold.

Images