Identity authentication method and system based on palm vein

Abstract

The invention relates to the field of identity authentication, in particular to an identity authentication method and system based on palm veins. An identity authentication system based on palm veins comprises an identity information acquisition module, a palm vein image acquisition module, an environment data acquisition module, a feature extraction module, a matching value calculation module, an authentication threshold adjustment module and an identity authentication module. According to the method, the multi-angle fusion operation is executed in the palm vein feature extraction process, so that the image used for executing feature extraction can fit the registered palm vein image, the difference caused by user operation behaviors during identity authentication is avoided, and the false alarm rate during identity authentication is reduced; when the matching value is calculated, the direct distance between the palm vein feature vector of the user and the authenticated palm vein feature vector corresponding to the identity information of the user in the identity library is calculated, and the accumulated distance caused by the change of the operation behavior of the user is also considered, so that the influence of the change of the operation behavior is weakened, and the false alarm rate in actual use is reduced.

Description

Technical Field

[0001] The present invention relates to the field of identity authentication, and in particular to an identity authentication method and system based on palm veins. Background Art

[0002] In the current field of biometrics, palm vein authentication has received widespread attention due to its high security and uniqueness. Palm vein authentication technology uses infrared imaging technology to capture the vein pattern inside the palm, which is highly unique and difficult to copy. However, although palm vein recognition systems have shown high accuracy and security in both theory and practice, they still face some significant challenges in practical applications.

[0003] First, the placement of the user's palm has a direct impact on the recognition result. Different placement angles, palm rotations, or finger openings may cause differences between the palm vein image and the image at the time of registration, thus affecting the matching accuracy. In addition, the positional deviation of the palm will also cause changes in the image capture area, further increasing the risk of recognition failure.

[0004] Secondly, environmental factors also affect the performance of the palm vein recognition system. For example, the intensity and quality of ambient light may affect the effect of infrared imaging, because the reflection and absorption of infrared light will be different under different lighting conditions. In addition, changes in ambient temperature may also affect the blood flow in the palm, thereby changing the visibility of the vein pattern. Summary of the invention

[0005] The present invention performs a multi-angle fusion operation in the process of palm vein feature extraction, so that the image used for feature extraction can fit the palm vein image at the time of registration, avoids the difference caused by the user's operation behavior during identity authentication, and reduces the false alarm rate during identity authentication; when calculating the matching value, not only the direct distance between the user's palm vein feature vector and the authenticated palm vein feature vector corresponding to the user's identity information in the identity library is calculated, but also the cumulative distance caused by the change of the user's operation behavior is considered, so as to weaken the influence of the change of the operation behavior, thereby reducing the false alarm rate in actual use; the authentication threshold is also adaptively adjusted based on the registration environment data and the current environment data, and the difference between the output palm vein feature vector and the authenticated palm vein feature vector is placed within a reasonable range, thereby reducing the false alarm situation.

[0006] A palm vein-based identity authentication method, comprising: Obtain the user's identity information and palm vein image; obtain the current environment data; The palm vein image of the user is sent to the feature extraction model for processing, and the corresponding palm vein feature vector is output; the palm vein feature vector of the user is matched with the authenticated palm vein feature vector corresponding to the identity information of the user in the identity database, and the matching value is output; the matching value is determined based on the direct distance and cumulative distance between the palm vein feature vector of the user and the authenticated palm vein feature vector corresponding to the identity information of the user in the identity database, the direct distance between the palm vein feature vector of the user and the authenticated palm vein feature vector corresponding to the identity information of the user in the identity database is used to describe the difference between the palm vein feature vector of the user and the authenticated palm vein feature vector corresponding to the identity information of the user in the identity database, and the cumulative distance between the palm vein feature vector of the user and the authenticated palm vein feature vector corresponding to the identity information of the user in the identity database is used to fit the difference between the palm vein feature vector of the user based on the change of the user's operation behavior and the authenticated palm vein feature vector corresponding to the identity information of the user in the identity database; the identity database stores the identity information of the user and its corresponding authenticated palm vein feature vector and registration environment data; Obtain the registration environment data corresponding to the user's identity information, adaptively adjust the authentication threshold based on the registration environment data and the current environment data, and then judge the output matching value with the authentication threshold. If the matching value is higher than the authentication threshold, output authentication success; otherwise, output authentication failure; to achieve identity authentication operation; The feature extraction model includes a multi-angle fusion layer, a feature extraction layer and a palm vein feature vector output layer. The multi-angle fusion layer is used to perform multi-angle feature learning on the palm vein image, and then perform a multi-angle fusion operation to construct a palm vein fusion feature map; the feature extraction layer is used to perform feature extraction operations on the palm vein fusion feature map to construct a palm vein feature vector; the palm vein feature vector output layer is used to output the palm vein feature vector.

[0007] As a preferred embodiment of the present invention, the palm vein image of the user is sent to a feature extraction model for processing, and the corresponding palm vein feature vector is output, which specifically includes the following steps: In the multi-angle fusion layer, the user's palm vein image is multiplied with N angle weight matrices to construct N palm vein simulation images, and then the N palm vein simulation images are spliced ​​according to the channels and convolution operations are performed to construct a palm vein fusion feature map; In the feature extraction layer, the palm vein fusion feature map is subjected to feature extraction operation to construct a palm vein feature vector; In the palm vein feature vector output layer, the palm vein feature vector is output.

[0008] As a preferred embodiment of the present invention, determining the matching value based on the direct distance and cumulative distance between the palm vein feature vector of the user and the authenticated palm vein feature vector corresponding to the identity information of the user in the identity database specifically includes the following steps: Calculate the direct distance f1 between the palm vein feature vector of the user and the authenticated palm vein feature vector corresponding to the user's identity information in the identity database; Construct a distance matrix D, and record the data in the distance matrix D as d(i)(i), i=1,2,3,…,I, I is the dimension of the palm vein feature vector and the authenticated palm vein feature vector, and d(i)(i) is the distance between the i-th element of the user's palm vein feature vector and the i-th element of the authenticated palm vein feature vector; Construct the cumulative distance matrix H, and record the data of the cumulative distance matrix H as h(i)(i), h(i)(i)=d(i)(i)+min[h(i-1)(i-1,h(i-1)(i),h(i)(i-1)]; Traverse the cumulative distance matrix H, and record the smallest h(i)(i) as the cumulative distance f2 between the user's palm vein feature vector and the authenticated palm vein feature vector corresponding to the user's identity information in the identity database; The matching value δ=α1f1+α2f2 is calculated, where α1 and α2 are the first weight coefficient and the second weight coefficient respectively, and α1+α2=1 is satisfied.

[0009] As a preferred embodiment of the present invention, the authentication threshold is adaptively adjusted based on the registered environment data and the current environment data, specifically comprising the following steps: The registered environment data and the current environment data are concatenated and sent to the threshold adjustment model for processing, and the adjustment value is output. The adjustment value is added to the initial threshold to obtain the authentication threshold; The threshold adjustment model is established based on BP neural network.

[0010] As a preferred embodiment of the present invention, training is performed on the feature extraction model, specifically comprising the following steps: Select a pre-trained convolutional neural network to assign parameters to the feature extraction layer and the palm vein feature vector output layer; Acquire several palm vein images; traverse all palm vein images, send the palm vein image to the pre-trained feature extraction layer and palm vein feature vector output layer for processing for each selected palm vein image, output the target palm vein feature vector, and then randomly rotate the palm vein image to construct several rotated palm vein images, and annotate all the rotated palm vein images by the target palm vein feature vector; until all palm vein images are traversed, all annotated rotated palm vein images are formed into a first training set; then send the first training set to the feature extraction model for training, take the target palm vein feature vector as the target, calculate the first loss value, and judge whether the first loss value is within the first preset range. If the first loss value is within the first preset range, output the trained feature extraction model; otherwise, continue to train the feature extraction model by the first training set.

[0011] As a preferred embodiment of the present invention, training is performed on the threshold adjustment model, specifically comprising the following steps: A plurality of threshold adjustment training samples are obtained, the threshold adjustment training samples include registration environment data, environment data and target adjustment value, and the threshold adjustment training samples are obtained through a group optimization algorithm; all the threshold adjustment training samples are combined into a second training set, and then the second training set is sent to the threshold adjustment model with parameter initialization for training, with the target adjustment value as the target, a second loss value is calculated, and it is determined whether the second loss value is within a second preset range. If the second loss value is within the second preset range, the trained threshold adjustment model is output; otherwise, the feature extraction model is continuously trained through the second training set.

[0012] As a preferred embodiment of the present invention, obtaining threshold adjustment training samples through a population optimization algorithm specifically includes the following steps: Set registration environment data and environment data; Construct several simulated adjustment values, then form all simulated adjustment values ​​into a population set, and set the maximum number of iterations; Calculate the fitness corresponding to each simulated adjustment value. The specific calculation method is as follows: calculate the authentication threshold by adding the simulated adjustment value and the initial threshold, obtain several matching samples, each matching sample includes the authenticated palm vein feature vector of the same user under the registered environment data and the palm vein feature vector under the environment data, traverse all matching samples, calculate the matching value based on each matching sample, and perform the identity authentication operation through the matching value and the authentication threshold; until all matching samples are traversed, the ratio of the matching sample corresponding to the successful authentication to all matching samples is used as the fitness corresponding to the simulated adjustment value; Based on the fitness corresponding to each simulated adjustment value, the population set is iteratively updated through the population optimization algorithm; Until the number of iterations reaches the maximum number of iterations, the simulation adjustment value with the maximum fitness is output as the target adjustment value corresponding to the registered environment data and the environment data.

[0013] A palm vein-based identity authentication system, comprising: An identity information acquisition module is used to obtain the user's identity information; A palm vein image acquisition module, used to acquire a palm vein image of a user; Environmental data acquisition module, used to obtain current environmental data; A feature extraction module is used to send the user's palm vein image to the feature extraction model for processing and output the corresponding palm vein feature vector; a matching value calculation module, used for matching the palm vein feature vector of the user with the authenticated palm vein feature vector corresponding to the identity information of the user in the identity database, and outputting a matching value; the matching value is determined based on the direct distance and cumulative distance between the palm vein feature vector of the user and the authenticated palm vein feature vector corresponding to the identity information of the user in the identity database, the matching value is used to describe the difference between the palm vein feature vector of the user and the authenticated palm vein feature vector corresponding to the identity information of the user in the identity database based on the direct distance between the palm vein feature vector of the user and the authenticated palm vein feature vector corresponding to the identity information of the user in the identity database, the matching value is used to fit the difference between the palm vein feature vector of the user based on the change of the user's operation behavior and the authenticated palm vein feature vector corresponding to the identity information of the user in the identity database based on the cumulative distance between the palm vein feature vector of the user and the authenticated palm vein feature vector corresponding to the identity information of the user in the identity database, the identity database stores the identity information of the user and its corresponding authenticated palm vein feature vector and registration environment data; An authentication threshold adjustment module is used to obtain the registration environment data corresponding to the user's identity information, and adaptively adjust the authentication threshold based on the registration environment data and the current environment data; The identity authentication module is used to judge the output matching value with the authentication threshold. If the matching value is higher than the authentication threshold, the authentication is output successfully; otherwise, the authentication is output failed, so as to realize the identity authentication operation.

[0014] The present invention has the following advantages: The present invention performs a multi-angle fusion operation in the process of palm vein feature extraction, so that the image used for feature extraction can fit the palm vein image at the time of registration, avoids the difference caused by the user's operation behavior during identity authentication, and reduces the false alarm rate during identity authentication; when calculating the matching value, not only the direct distance between the user's palm vein feature vector and the authenticated palm vein feature vector corresponding to the user's identity information in the identity library is calculated, but also the cumulative distance caused by the change of the user's operation behavior is considered, so as to weaken the influence of the change of the operation behavior, thereby reducing the false alarm rate in actual use; the authentication threshold is also adaptively adjusted based on the registration environment data and the current environment data, and the difference between the output palm vein feature vector and the authenticated palm vein feature vector is placed within a reasonable range, thereby reducing the false alarm situation. BRIEF DESCRIPTION OF THE DRAWINGS

[0015] Figure 1 The figure is a flow chart of a palm vein-based identity authentication system used in an embodiment of the present invention. DETAILED DESCRIPTION

[0016] In order to enable persons skilled in the art to better understand the technical solutions in the present invention, the technical solutions in the embodiments of the present invention will be clearly and completely described below in conjunction with the accompanying drawings in the embodiments of the present invention.

[0017] Embodiment 1, a palm vein-based identity authentication method, comprising: Obtain the user's identity information and palm vein image. The palm vein image refers to the image of the inner vein of the palm captured by near-infrared light. In the actual identity authentication process, the user needs to input his or her identity information through voice input or keyboard input. The identity information can be a name or work code, etc.; obtain the current environmental data. The current environmental data includes the temperature during identity authentication. The temperature may affect the user's blood flow, thereby affecting the clarity of the palm vein image, and may also cause the user's palm to sweat, thereby affecting the quality of the palm vein image; The palm vein image of the user is sent to the feature extraction model for processing, and the corresponding palm vein feature vector is output; the palm vein feature vector of the user is matched with the authenticated palm vein feature vector corresponding to the user's identity information in the identity library, and a matching value is output; the matching value is determined based on the direct distance and cumulative distance between the palm vein feature vector of the user and the authenticated palm vein feature vector corresponding to the user's identity information in the identity library, the direct distance between the palm vein feature vector of the user and the authenticated palm vein feature vector corresponding to the user's identity information in the identity library is used to describe the difference between the palm vein feature vector of the user and the authenticated palm vein feature vector corresponding to the user's identity information in the identity library, and the cumulative distance between the palm vein feature vector of the user and the authenticated palm vein feature vector corresponding to the user's identity information in the identity library is used to fit the difference between the palm vein feature vector of the user based on the change of user operation behavior and the authenticated palm vein feature vector corresponding to the user's identity information in the identity library, where the change of user operation behavior refers to the action of the user performing palm vein image scanning during the current identity authentication and the action of the user performing palm vein image scanning during the entry of the authenticated palm vein feature vector. Differences between scanning actions, such as different placement positions, angles and hand movements of the palm, etc. Due to these changes in operation behaviors, describing the matching value only from the dimension of the direct distance between the user's palm vein feature vector and the authenticated palm vein feature vector corresponding to the user's identity information in the identity library may result in a large deviation, thereby increasing the false alarm rate in actual use. Adding the dimension of the cumulative distance between the user's palm vein feature vector and the authenticated palm vein feature vector corresponding to the user's identity information in the identity library can weaken the impact of changes in operation behaviors, thereby reducing the false alarm rate in actual use; the identity library stores the user's identity information and its corresponding authenticated palm vein feature vector and registration environment data. Each user will perform a registration operation before performing identity authentication. The registration operation requires the user to perform a palm vein feature extraction operation, that is, to send the user's palm vein image to the feature extraction model for processing, output the corresponding palm vein feature vector, and record the palm vein feature vector as the authenticated palm vein feature vector corresponding to the user's identity information; the registration environment data is the environment data of the user when entering the authenticated palm vein feature vector; Obtain the registered environment data corresponding to the user's identity information, adaptively adjust the authentication threshold based on the registered environment data and the current environment data, and then judge the output matching value with the authentication threshold. If the matching value is higher than the authentication threshold, it means that the user's identity authentication is successful, and the authentication success is output; otherwise, it means that the user's identity authentication is unsuccessful, and the authentication failure is output; to achieve the identity authentication operation; because during the identity authentication process, there is a deviation between the current environment data and the registered environment data when the user registered the identity information, which will cause the output palm vein feature vector to be different from the authenticated palm vein feature vector, resulting in false positives during identity authentication. By dynamically adjusting the authentication threshold, the difference between the output palm vein feature vector and the authenticated palm vein feature vector can be placed within a reasonable range, thereby reducing false positives; The feature extraction model includes a multi-angle fusion layer, a feature extraction layer and a palm vein feature vector output layer, wherein the multi-angle fusion layer is used to perform multi-angle feature learning on the palm vein image, and then perform a multi-angle fusion operation to construct a palm vein fusion feature map. It should be noted that during the identity authentication process, the user's operation behavior will be different from that during registration, such as the palm placement angle, which will lead to the difference between the generated palm vein feature vector and the authenticated palm vein feature vector. Therefore, multi-angle fusion is used to simulate the rotation of the palm vein image, so that the generated palm vein feature vector can fit the authenticated palm vein feature vector, further reducing the false alarm rate; the feature extraction layer is used to perform feature extraction operations on the palm vein fusion feature map to construct a palm vein feature vector. The feature extraction layer can be constructed based on the VGG16 model to realize the feature extraction operation of the image; the palm vein feature vector output layer is used to output the palm vein feature vector; The present application performs a multi-angle fusion operation in the palm vein feature extraction process, so that the image used for feature extraction can fit the palm vein image at the time of registration, avoids the difference caused by the user's operation behavior during identity authentication, and reduces the false alarm rate during identity authentication; when calculating the matching value, not only the direct distance between the user's palm vein feature vector and the authenticated palm vein feature vector corresponding to the user's identity information in the identity library is calculated, but also the cumulative distance caused by the change of the user's operation behavior is considered, thereby weakening the impact of the change of operation behavior, thereby reducing the false alarm rate in actual use; the authentication threshold is also adaptively adjusted based on the registration environment data and the current environment data, and the difference between the output palm vein feature vector and the authenticated palm vein feature vector is placed within a reasonable range, thereby reducing the false alarm situation.

[0018] The user's palm vein image is sent to the feature extraction model for processing, and the corresponding palm vein feature vector is output, which specifically includes the following steps: In the multi-angle fusion layer, the user's palm vein image is multiplied by N angle weight matrices respectively to realize the rotation simulation of the palm vein image, and N palm vein simulation images are constructed. Then, the N palm vein simulation images are spliced ​​according to the channels and then convolution operations are performed to construct the palm vein fusion feature map; In the feature extraction layer, the palm vein fusion feature map is subjected to feature extraction operations, including convolution operations, normalization operations, activation operations, and pooling operations, etc., and specific settings are made based on the VGG16 model to construct a palm vein feature vector; In the palm vein feature vector output layer, the palm vein feature vector is output.

[0019] The matching value is determined based on the direct distance and cumulative distance between the palm vein feature vector of the user and the authenticated palm vein feature vector corresponding to the identity information of the user in the identity database, specifically including the following steps: Calculate the direct distance f1 between the palm vein feature vector of the user and the authenticated palm vein feature vector corresponding to the identity information of the user in the identity database, and the calculation method may be a Euclidean distance calculation method, a Hamming distance calculation method, or the like; Construct a distance matrix D, record the data in the distance matrix D as d(i)(i), i=1,2,3,…,I, I is the dimension of the palm vein feature vector and the authenticated palm vein feature vector, d(i)(i) is the distance between the i-th element of the user's palm vein feature vector (each data in the palm vein feature vector is recorded as an element, the same below) and the i-th element of the authenticated palm vein feature vector, and the distance calculation method can be the Euclidean distance calculation method and the Hamming distance calculation method, etc.; Construct the cumulative distance matrix H, and record the data of the cumulative distance matrix H as h(i)(i), h(i)(i)=d(i)(i)+min[h(i-1)(i-1),h(i-1)(i),h(i)(i-1)]. When constructing the cumulative distance matrix H, start from the lower right corner of the distance matrix D, from right to left, from bottom to top, and trace back to the upper left corner of the distance matrix D; Traverse the cumulative distance matrix H and record the smallest h(i)(i) as the cumulative distance f2 between the user's palm vein feature vector and the authenticated palm vein feature vector corresponding to the user's identity information in the identity database. This cumulative distance represents the distance under the best alignment, even if there is a difference between the current palm vein image and the palm vein image at the time of registration; The matching value δ=α1f1+α2f2 is calculated, where α1 and α2 are the first weight coefficient and the second weight coefficient respectively, and α1+α2=1 is satisfied. The first weight coefficient and the second weight coefficient are set by the developer.

[0020] The authentication threshold is adaptively adjusted based on the registered environment data and the current environment data, including the following steps: The registered environment data and the current environment data are concatenated and sent to the threshold adjustment model for processing, and the adjustment value is output. The adjustment value has positive and negative values. The adjustment value is added to the initial threshold to obtain the authentication threshold. The initial threshold is generally 0.85. When the adjustment value is positive, it means that the authentication threshold needs to be increased, but when the adjustment value is negative, it means that the authentication threshold needs to be lowered. The threshold adjustment model is established based on BP neural network.

[0021] Training the feature extraction model includes the following steps: Select a pre-trained convolutional neural network to assign parameters to the feature extraction layer and the palm vein feature vector output layer. The pre-training can be performed using the ImageNet dataset, and the convolutional neural network is set using the VGG16 model. Acquire several palm vein images. The palm vein images can be acquired from a public data set, such as the PolyU Palmprint Database data set, or can be actually collected. In this embodiment, the actual collection method is selected; traverse all palm vein images, and for each selected palm vein image, send the palm vein image to the pre-trained feature extraction layer and the palm vein feature vector output layer for processing, output the target palm vein feature vector, and then randomly rotate the palm vein image. It should be noted that after the selection, the background color is filled to construct several rotated palm vein images, and all the rotated palm vein images are annotated by the target palm vein feature vector; until all palm vein images are traversed, all the annotated rotated palm vein images are formed into a first training set; then send the first training set to the feature extraction model for training, take the target palm vein feature vector as the target, calculate the first loss value, and judge whether the first loss value is within the first preset range. The first preset range is set by the developer. If the first loss value is within the first preset range, the trained feature extraction model is output; otherwise, the feature extraction model is continuously trained through the first training set.

[0022] Training the threshold adjustment model includes the following steps: A plurality of threshold adjustment training samples are obtained, the threshold adjustment training samples include registration environment data, environment data and target adjustment value, and the threshold adjustment training samples are obtained through a group optimization algorithm. When the threshold adjustment training samples are obtained through the group optimization algorithm, the registration environment data and environment data are set, and the authenticated palm vein feature vector under the registration environment data is used as the target, and the palm vein feature vector under the environment data is matched to adjust the target adjustment value; all the threshold adjustment training samples are formed into a second training set, and then the second training set is sent to the parameter-initialized threshold adjustment model for training, and the target adjustment value is used as the target, and a second loss value is calculated to determine whether the second loss value is within a second preset range. The second preset range is also set by the developer. If the second loss value is within the second preset range, the trained threshold adjustment model is output; otherwise, the feature extraction model is continuously trained through the second training set.

[0023] The threshold adjustment training samples are obtained through the group optimization algorithm, which specifically includes the following steps: Set registration environment data and environment data; Construct several simulated adjustment values, and then form all simulated adjustment values ​​into a population set, set the maximum number of iterations, and when constructing simulated adjustment values, select a random number within the adjustment value range as the simulated adjustment value. The adjustment value range is set by the meal operator based on the inspection, generally [-1, 1]; Calculate the fitness corresponding to each simulated adjustment value. The specific calculation method is as follows: calculate the authentication threshold by adding the simulated adjustment value and the initial threshold, obtain several matching samples, each matching sample includes the authenticated palm vein feature vector of the same user under the registered environment data and the palm vein feature vector under the environment data, traverse all matching samples, calculate the matching value based on each matching sample, and perform the identity authentication operation through the matching value and the authentication threshold; until all matching samples are traversed, the ratio of the matching sample corresponding to the successful authentication to all matching samples is used as the fitness corresponding to the simulated adjustment value; Based on the fitness corresponding to each simulated adjustment value, the population set is iteratively updated through the sparrow search algorithm; Until the number of iterations reaches the maximum number of iterations, the simulation adjustment value with the maximum fitness is output as the target adjustment value corresponding to the registered environment data and the environment data.

[0024] Example 2, an identity authentication system based on palm vein, such as Figure 1 As shown, including: The identity information acquisition module is used to obtain the user's identity information. In the actual identity authentication process, the user needs to input his or her identity information through voice input or keyboard input. The identity information can be a name or work code, etc.; A palm vein image acquisition module is used to acquire a palm vein image of a user, where a palm vein image refers to an image of the inner veins of a palm captured by near-infrared light; The environmental data acquisition module is used to acquire the current environmental data, including the temperature during identity authentication. The temperature may affect the blood flow of the user, thereby affecting the clarity of the palm vein image, and may also cause the user's palm to sweat, thereby affecting the quality of the palm vein image; A feature extraction module is used to send the user's palm vein image to the feature extraction model for processing and output the corresponding palm vein feature vector; A matching value calculation module is used to match the palm vein feature vector of the user with the authenticated palm vein feature vector corresponding to the identity information of the user in the identity library, and output a matching value; the matching value is determined based on the direct distance and cumulative distance between the palm vein feature vector of the user and the authenticated palm vein feature vector corresponding to the identity information of the user in the identity library, and the matching value is based on the direct distance between the palm vein feature vector of the user and the authenticated palm vein feature vector corresponding to the identity information of the user in the identity library. It is used to describe the difference between the palm vein feature vector of the user and the authenticated palm vein feature vector corresponding to the identity information of the user in the identity library, and the matching value is based on the cumulative distance between the palm vein feature vector of the user and the authenticated palm vein feature vector corresponding to the identity information of the user in the identity library. It is used to fit the difference between the palm vein feature vector of the user based on the change of user operation behavior and the authenticated palm vein feature vector corresponding to the identity information of the user in the identity library. The change of user operation behavior here refers to the difference between the action of the user performing palm vein image scanning during the current identity authentication and the scanning action when the authenticated palm vein feature vector is entered. Due to these changes in operation behaviors, the description of the matching value based solely on the dimension of the direct distance between the user's palm vein feature vector and the authenticated palm vein feature vector corresponding to the user's identity information in the identity database may result in a large deviation, thereby increasing the false alarm rate in actual use. Adding the dimension of the cumulative distance between the user's palm vein feature vector and the authenticated palm vein feature vector corresponding to the user's identity information in the identity database can weaken the impact of changes in operation behaviors, thereby reducing the false alarm rate in actual use. The identity database stores the user's identity information and its corresponding authenticated palm vein feature vector and registration environment data. Each user will perform a registration operation before performing identity authentication. The registration operation requires the user to perform a palm vein feature extraction operation, that is, to send the user's palm vein image to the feature extraction model for processing, output the corresponding palm vein feature vector, and record the palm vein feature vector as the authenticated palm vein feature vector corresponding to the user's identity information. The registration environment data is the environment data of the user when entering the authenticated palm vein feature vector. An authentication threshold adjustment module is used to obtain the registration environment data corresponding to the user's identity information, and adaptively adjust the authentication threshold based on the registration environment data and the current environment data; The identity authentication module is used to judge the output matching value with the authentication threshold. If the matching value is higher than the authentication threshold, it means that the user's identity authentication is successful and the output is authentication success; otherwise, it means that the user's identity authentication is unsuccessful and the output is authentication failure, so as to realize the identity authentication operation.

[0025] It should be understood that those skilled in the art can make improvements or changes based on the above description, and all such improvements and changes should fall within the scope of protection of the appended claims of the present invention. Parts not described in detail in this specification belong to the prior art known to those skilled in the art.

Claims

1. A palm vein-based identity authentication method, characterized in that: include: Obtain the user's identity information and palm vein image; obtain the current environment data; The palm vein image of the user is sent to the feature extraction model for processing, and the corresponding palm vein feature vector is output; the palm vein feature vector of the user is matched with the authenticated palm vein feature vector corresponding to the identity information of the user in the identity database, and the matching value is output; the matching value is determined based on the direct distance and cumulative distance between the palm vein feature vector of the user and the authenticated palm vein feature vector corresponding to the identity information of the user in the identity database, the direct distance between the palm vein feature vector of the user and the authenticated palm vein feature vector corresponding to the identity information of the user in the identity database is used to describe the difference between the palm vein feature vector of the user and the authenticated palm vein feature vector corresponding to the identity information of the user in the identity database, and the cumulative distance between the palm vein feature vector of the user and the authenticated palm vein feature vector corresponding to the identity information of the user in the identity database is used to fit the difference between the palm vein feature vector of the user based on the change of the user's operation behavior and the authenticated palm vein feature vector corresponding to the identity information of the user in the identity database; the identity database stores the identity information of the user and its corresponding authenticated palm vein feature vector and registration environment data; Obtain the registration environment data corresponding to the user's identity information, adaptively adjust the authentication threshold based on the registration environment data and the current environment data, and then compare the output match value with the authentication threshold. If the match value is higher than the authentication threshold, output authentication success; Otherwise, the authentication failure is output to implement the identity authentication operation; The feature extraction model includes a multi-angle fusion layer, a feature extraction layer and a palm vein feature vector output layer. The multi-angle fusion layer is used to perform multi-angle feature learning on the palm vein image and then perform a multi-angle fusion operation to construct a palm vein fusion feature map. The feature extraction layer is used to perform a feature extraction operation on the palm vein fusion feature map to construct a palm vein feature vector. The palm vein feature vector output layer is used to output the palm vein feature vector.

2. The palm vein-based identity authentication method according to claim 1, characterized in that: The user's palm vein image is sent to the feature extraction model for processing, and the corresponding palm vein feature vector is output, which specifically includes the following steps: In the multi-angle fusion layer, the user's palm vein image is multiplied with N angle weight matrices to construct N palm vein simulation images, and then the N palm vein simulation images are spliced ​​according to the channels and convolution operations are performed to construct a palm vein fusion feature map; In the feature extraction layer, the palm vein fusion feature map is subjected to feature extraction operation to construct a palm vein feature vector; In the palm vein feature vector output layer, the palm vein feature vector is output.

3. The palm vein-based identity authentication method according to claim 2, characterized in that: The matching value is determined based on the direct distance and cumulative distance between the palm vein feature vector of the user and the authenticated palm vein feature vector corresponding to the identity information of the user in the identity database, specifically including the following steps: Calculate the direct distance f1 between the palm vein feature vector of the user and the authenticated palm vein feature vector corresponding to the user's identity information in the identity database; Construct a distance matrix D, and record the data in the distance matrix D as d(i)(i), i=1,2,3,…,I, I is the dimension of the palm vein feature vector and the authenticated palm vein feature vector, and d(i)(i) is the distance between the i-th element of the user's palm vein feature vector and the i-th element of the authenticated palm vein feature vector; Construct the cumulative distance matrix H, and record the data of the cumulative distance matrix H as h(i)(i), h(i)(i)=d(i)(i)+min[h(i-1)(i-1,h(i-1)(i),h(i)(i-1)]; Traverse the cumulative distance matrix H, and record the smallest h(i)(i) as the cumulative distance f2 between the user's palm vein feature vector and the authenticated palm vein feature vector corresponding to the user's identity information in the identity database; The matching value δ=α1f1+α2f2 is calculated, where α1 and α2 are the first weight coefficient and the second weight coefficient respectively, and α1+α2=1 is satisfied.

4. The palm vein-based identity authentication method according to claim 3, characterized in that: The authentication threshold is adaptively adjusted based on the registered environment data and the current environment data, including the following steps: The registered environment data and the current environment data are concatenated and sent to the threshold adjustment model for processing, and the adjustment value is output. The adjustment value is added to the initial threshold to obtain the authentication threshold; The threshold adjustment model is established based on BP neural network.

5. The palm vein-based identity authentication method according to claim 4, characterized in that: Training the feature extraction model includes the following steps: Select a pre-trained convolutional neural network to assign parameters to the feature extraction layer and the palm vein feature vector output layer; Acquire several palm vein images; traverse all palm vein images, send the palm vein image to the pre-trained feature extraction layer and palm vein feature vector output layer for processing for each selected palm vein image, output the target palm vein feature vector, and then randomly rotate the palm vein image to construct several rotated palm vein images, and annotate all the rotated palm vein images by the target palm vein feature vector; until all palm vein images are traversed, all annotated rotated palm vein images are formed into a first training set; then send the first training set to the feature extraction model for training, take the target palm vein feature vector as the target, calculate the first loss value, and judge whether the first loss value is within the first preset range. If the first loss value is within the first preset range, output the trained feature extraction model; otherwise, continue to train the feature extraction model by the first training set.

6. The palm vein-based identity authentication method according to claim 5, characterized in that: Training the threshold adjustment model includes the following steps: A plurality of threshold adjustment training samples are obtained, the threshold adjustment training samples include registration environment data, environment data and target adjustment value, and the threshold adjustment training samples are obtained through a group optimization algorithm; all the threshold adjustment training samples are combined into a second training set, and then the second training set is sent to the threshold adjustment model with parameter initialization for training, with the target adjustment value as the target, a second loss value is calculated, and it is determined whether the second loss value is within a second preset range. If the second loss value is within the second preset range, the trained threshold adjustment model is output; otherwise, the feature extraction model is continuously trained through the second training set.

7. The palm vein-based identity authentication method according to claim 6, characterized in that: The threshold adjustment training samples are obtained through the group optimization algorithm, which specifically includes the following steps: Set registration environment data and environment data; Construct several simulated adjustment values, then form all simulated adjustment values ​​into a population set, and set the maximum number of iterations; Calculate the fitness corresponding to each simulated adjustment value. The specific calculation method is as follows: calculate the authentication threshold by adding the simulated adjustment value and the initial threshold, obtain several matching samples, each matching sample includes the authenticated palm vein feature vector of the same user under the registered environment data and the palm vein feature vector under the environment data, traverse all matching samples, calculate the matching value based on each matching sample, and perform the identity authentication operation through the matching value and the authentication threshold; until all matching samples are traversed, the ratio of the matching sample corresponding to the successful authentication to all matching samples is used as the fitness corresponding to the simulated adjustment value; Based on the fitness corresponding to each simulated adjustment value, the population set is iteratively updated through the population optimization algorithm; Until the number of iterations reaches the maximum number of iterations, the simulation adjustment value with the maximum fitness is output as the target adjustment value corresponding to the registered environment data and the environment data.

8. A palm vein-based identity authentication system, characterized in that: The system applies a palm vein-based identity authentication method as described in any one of claims 1 to 7, including: An identity information acquisition module is used to obtain the user's identity information; A palm vein image acquisition module, used to acquire a palm vein image of a user; Environmental data acquisition module, used to obtain current environmental data; A feature extraction module is used to send the user's palm vein image to the feature extraction model for processing and output the corresponding palm vein feature vector; a matching value calculation module, used for matching the palm vein feature vector of the user with the authenticated palm vein feature vector corresponding to the identity information of the user in the identity database, and outputting a matching value; the matching value is determined based on the direct distance and cumulative distance between the palm vein feature vector of the user and the authenticated palm vein feature vector corresponding to the identity information of the user in the identity database, the matching value is used to describe the difference between the palm vein feature vector of the user and the authenticated palm vein feature vector corresponding to the identity information of the user in the identity database based on the direct distance between the palm vein feature vector of the user and the authenticated palm vein feature vector corresponding to the identity information of the user in the identity database, the matching value is used to fit the difference between the palm vein feature vector of the user based on the change of the user's operation behavior and the authenticated palm vein feature vector corresponding to the identity information of the user in the identity database based on the cumulative distance between the palm vein feature vector of the user and the authenticated palm vein feature vector corresponding to the identity information of the user in the identity database, the identity database stores the identity information of the user and its corresponding authenticated palm vein feature vector and registration environment data; An authentication threshold adjustment module is used to obtain the registration environment data corresponding to the user's identity information, and adaptively adjust the authentication threshold based on the registration environment data and the current environment data; The identity authentication module is used to judge the output matching value with the authentication threshold. If the matching value is higher than the authentication threshold, the authentication is output successfully; otherwise, the authentication is output failed, so as to realize the identity authentication operation.

Images