Biometric-based identity recognition method, apparatus, device, and medium

Abstract

The application discloses a biometric-based identity recognition method, device, equipment and medium, and belongs to the field of data processing. The method comprises the following steps: in the first stage, the biometric features of a first user in a preset user range are collected, and the first target data obtained by matching the biometric features with a registered biometric feature database is obtained; the first target data is used as sample data for model training, a classification model including a classification plane is obtained, and the first target data comprises a similarity parameter and an environmental factor parameter; in the second stage, according to the biometric features of a second user in the preset user range, the second target data obtained by matching the biometric features of the second user with the registered biometric feature database is obtained; the second target data is classified and processed by using the classification model, and a classification result is obtained, the second target data comprises a similarity parameter and an environmental factor parameter, and the classification result represents whether the second user passes the identity authentication. According to the embodiment of the application, the misjudgment rate of identity recognition can be reduced.

Description

Technical Field

[0001] This application belongs to the field of data processing, and in particular relates to a biometric-based identity recognition method, device, equipment and medium. Background Technology

[0002] Due to their uniqueness and convenience, biometrics have been widely used in the field of identity verification. Generally, a user's biometrics are obtained and matched against registered biometrics in a database to determine the similarity between the two. If a registered biometric in the database has a similarity score higher than a threshold, the user's identity verification is considered successful.

[0003] However, in practice, it is difficult to accurately set the threshold for determining whether identity verification has passed. If the threshold is set too high, it will result in false positives such as identifying the same object as different objects; if the threshold is set too low, it will result in false positives such as identifying different objects as the same object, leading to a high false positive rate in identity verification. Summary of the Invention

[0004] This application provides a biometric-based identity recognition method, apparatus, device, and medium that can reduce the false recognition rate of identity recognition.

[0005] In a first aspect, embodiments of this application provide a biometric-based identity recognition method, comprising: in a first stage, collecting at least a portion of first target data obtained by matching the biometrics of a first user within a preset user range with a registered biometric database, using the at least a portion of the first target data as sample data for model training, obtaining a classification model including a classification plane, wherein the first target data includes similarity parameters and environmental factor parameters, and the classification plane divides the sample data into two categories: identity authentication passed and identity authentication failed; in a second stage, based on the biometrics of a second user within a preset user range, acquiring second target data obtained by matching the biometrics of the second user with a registered biometric database, using the classification model to classify the second target data, obtaining a classification result, wherein the second target data includes similarity parameters and environmental factor parameters, and the classification result characterizes whether the second user has passed identity authentication.

[0006] Secondly, embodiments of this application provide a biometric-based identity recognition device, comprising: a model training module, configured to, in a first stage, collect at least a portion of first target data obtained by matching the biometric features of a first user within a preset user range with a registered biometric database, and use the at least a portion of the first target data as sample data for model training to obtain a classification model including a classification plane, wherein the first target data includes similarity parameters and environmental factor parameters, and the classification plane divides the sample data into two categories: identity authentication passed and identity authentication failed; and an identity recognition module, configured to, in a second stage, obtain second target data obtained by matching the biometric features of a second user within a preset user range with a registered biometric database, and use the classification model to classify the second target data to obtain a classification result, wherein the second target data includes similarity parameters and environmental factor parameters, and the classification result characterizes whether the second user has passed identity authentication.

[0007] Thirdly, embodiments of this application provide a terminal device, including: a processor and a memory storing computer program instructions; when the processor executes the computer program instructions, it implements the biometric-based identity recognition method of the first aspect.

[0008] Fourthly, embodiments of this application provide a computer-readable storage medium storing computer program instructions, which, when executed by a processor, implement the biometric-based identity recognition method of the first aspect.

[0009] This application provides a biometric-based identity recognition method, apparatus, device, and medium. In the first stage, during the matching process between a first user within a preset user range and a registered biometric database using biometrics, first target data including similarity parameters and environmental factor parameters is collected. This first target data is used as sample data for model training, resulting in a classification model that includes a classification plane capable of dividing the sample data into two categories: successful and unsuccessful identity recognition. In the second stage, after obtaining the classification model, second target data including similarity parameters and environmental factor parameters is obtained based on the matching of the biometrics of a second user within the preset user range and the registered biometric database. This second target data is then classified using the classification model, and the resulting classification result indicates whether the user has passed identity authentication. The classification model fully utilizes multi-dimensional elements such as similarity parameters and environmental factors obtained through machine learning algorithms. For second users within the same preset user range, the classification plane in the model can accurately represent whether the second user has passed identity authentication based on the second target data. This minimizes the possibility of a user and the registered user being the same user who fails authentication, or a user and the registered user being different users who pass authentication, thereby reducing the false positive rate of identity recognition. Attached Figure Description

[0010] To more clearly illustrate the technical solutions of the embodiments of this application, the accompanying drawings used in the embodiments of this application will be briefly introduced below. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0011] Figure 1 A flowchart illustrating a biometric-based identity recognition method provided in an embodiment of this application;

[0012] Figure 2 A schematic diagram illustrating an example of the projection of the classification plane into a two-dimensional plane, as provided in an embodiment of this application;

[0013] Figure 3 A flowchart illustrating a biometric-based identity recognition method provided in another embodiment of this application;

[0014] Figure 4 A flowchart illustrating a biometric-based identity recognition method provided in yet another embodiment of this application;

[0015] Figure 5 A logical schematic diagram illustrating an example of a biometric-based identity recognition method provided in an embodiment of this application;

[0016] Figure 6 This is a schematic diagram of the structure of a biometric-based identity recognition device provided in an embodiment of this application;

[0017] Figure 7 This is a schematic diagram of the structure of a terminal device provided in an embodiment of this application. Detailed Implementation

[0018] The features and exemplary embodiments of various aspects of this application will be described in detail below. To make the objectives, technical solutions, and advantages of this application clearer, the application will be further described in detail below with reference to the accompanying drawings and specific embodiments. It should be understood that the specific embodiments described herein are only intended to explain this application and not to limit it. For those skilled in the art, this application can be implemented without some of these specific details. The following description of the embodiments is merely to provide a better understanding of this application by illustrating examples. It should be noted that the acquisition, storage, use, and processing of information and data in the embodiments of this application are all authorized by users or relevant organizations and comply with the relevant provisions of national laws and regulations.

[0019] Due to their uniqueness and convenience, biometrics have been widely used in the field of identity verification. Generally, a user's biometrics are matched against the registered biometrics of other users in a database to obtain the similarity score. If a registered biometric in the database has a similarity score higher than a threshold, the user's identity verification is considered successful. In practice, using the threshold for identity verification can result in several scenarios: First, the user and the registered user in the database are actually the same user, and the user's identity verification succeeds (true accept). Second, the user and the registered user in the database are not actually the same user, but the user's identity verification succeeds (false accept). Third, the user and the registered user in the database are actually the same user, but the user's identity verification fails (false reject). Fourth, the user and the registered user in the database are not actually the same user, and the user's identity verification fails (true reject). The second and third scenarios are considered false positives in identity verification. In practice, it is difficult to accurately set the threshold for determining whether identity verification has succeeded. If the judgment threshold is set too high, the same object may be identified as different objects, which is the third situation mentioned above. If the judgment threshold is set too low, different objects may be identified as the same object, which is the second situation mentioned above, resulting in a high misjudgment rate for identity recognition.

[0020] To improve the accuracy of identity verification, some scenarios involve stacking verification processes. For example, a 1:n biometric verification might be performed first, followed by a 1:1 biometric verification. If both biometric verifications pass, the identity verification is considered successful. However, this process cannot reduce the possibility of a user and the registered user being the same user but failing verification. Furthermore, the more times the verification process is stacked, the lower the efficiency and the higher the cost.

[0021] This application provides a biometric-based identity recognition method, apparatus, device, and medium, applicable to user identity recognition in closed environments. It utilizes similarity parameters obtained by matching user biometrics with a registered biometric database, along with environmental factor parameters, and combines this with machine learning algorithms to train a classification model capable of effectively recognizing users in closed environments. This reduces the likelihood of users failing authentication despite being the same person as the registered user, and conversely, successfully authenticating despite being different users, thus lowering the false positive rate. Furthermore, using a classification model for user identity recognition eliminates the need for additional identity recognition processes, thereby improving efficiency and reducing costs to some extent.

[0022] The following describes the biometric-based identity recognition method, device, equipment, and medium provided in this application.

[0023] The first aspect of this application provides a biometric-based identity recognition method that can be applied to the identity recognition of terminal devices in closed scenarios. This biometric-based identity recognition method can be executed by the terminal device or the system, and the type of terminal device is not limited herein. Figure 1 A flowchart of a biometric-based identity recognition method provided in an embodiment of this application is shown below. Figure 1 As shown, the biometric-based identity recognition method may include steps S101 and S102.

[0024] In step S101, in the first stage, at least a portion of the first target data obtained by matching the biometrics of the first user within the preset user range with the registered biometric database is collected. The at least a portion of the first target data is used as sample data for model training to obtain a classification model including a classification plane.

[0025] The first stage can be the preliminary preparation stage or the trial operation stage of identity recognition, during which data for model training can be collected. The preset user scope can be a pre-defined set of users, which is a closed scope. For example, the preset user scope may include students and staff within a school, or employees within a company. The first user may include users within the preset user scope who are undergoing identity recognition in the first stage. During the identity recognition process, the terminal device can collect the biometrics of the first user in real time and match them one by one with the registered biometrics in the registered biometrics database to obtain the first target data. The registered biometrics database corresponds to the preset user scope and may include the registered biometrics of users within the preset user scope. The types of registered biometrics in the registered biometrics database are consistent with the types of biometrics collected from the first user. In this embodiment, the types of biometrics are not limited. For example, biometrics may include facial features, palm print features, palm vein features, voiceprint features, iris features, etc., and are not limited here.

[0026] The primary target data includes similarity parameters and environmental factor parameters. Similarity parameters characterize similarity, and the primary target data includes similarity parameters between the first user's biometrics and the registered biometrics. Environmental factor parameters characterize environmental factors, and the primary target data may include one or more environmental factor parameters. In some examples, environmental factor parameters may include one or more of the following: feature acquisition time information, feature acquisition lighting information, feature acquisition occlusion information, feature ambiguity information, feature size information, and feature acquisition module information. Feature acquisition time information characterizes the time at which the first user's biometrics were acquired. Feature acquisition lighting information characterizes the ambient lighting during the acquisition of the first user's biometrics. Feature acquisition occlusion information characterizes the occlusion level during the acquisition of the first user's biometrics. Feature ambiguity information characterizes the ambiguity of the acquired first user's biometrics. Feature size information characterizes the size of the acquired first user's biometrics. Feature acquisition module information characterizes the type and model of the module used to acquire the first user's biometrics.

[0027] The biometric features of the first user are matched one-to-one with the registered biometric features in the registered biometric database, resulting in multiple similarity parameters. Correspondingly, multiple first target data can be collected for a single first user. All first target data can be used as sample data, or only a portion of the first target data can be used. In some examples, the similarity parameters in the first target data may include the top m similarity parameters (ranked from highest to lowest) obtained from matching the first user's biometric features with the registered biometric database, where m is an integer greater than or equal to 2. The value of m can be set according to specific scenarios, needs, experience, etc., and is not limited here. Since the similarity parameters that cause misjudgments generally represent high similarity, selecting the first target data including the top m similarity parameters (ranked from highest to lowest) as sample data for model training has two advantages: firstly, the trained classification model is more targeted in correcting misjudgments and more accurate in identifying individuals; secondly, it also improves the training efficiency of the classification model.

[0028] The sample data has labels that characterize whether the sample data should pass authentication. These labels can be obtained based on the authentication results and user feedback from the first stage. The sample data may include labels representing first target data where the true result is successful authentication and first target data where the true result is unsuccessful authentication. This application does not limit the specific machine learning algorithm used for model training; for example, support vector machines, Bayesian algorithms, ensemble learning, etc., can be used. The trained classification model includes a classification plane. The classification plane can divide the sample data into two categories: successful authentication and unsuccessful authentication. The classification plane can have some error in dividing the sample data into two categories, allowing some sample data to exceed the constraints, but it is necessary to divide the sample data into these two categories as much as possible. The first target data used as sample data can be implemented as a vector or other data form that can represent multiple dimensions, and is not limited here. Each sample data can be considered as a scatter point in a multi-dimensional space, and the classification plane can divide the sample data into a category where the true result should be successful authentication and a category where the true result should be unsuccessful authentication.

[0029] For example, Figure 2 A schematic diagram illustrating an example of the projection of the classification plane into a two-dimensional plane, as provided in the embodiments of this application, is shown below. Figure 2 As shown, solid dots represent sample data that should have passed authentication, while hollow dots represent sample data that should have failed authentication. A clear grouping cohesion exists between the solid and hollow dots. Line L1 is the projection of the classification plane onto the two-dimensional plane, effectively distinguishing between solid and hollow dots. Solid dots are located on one side of line L1, and hollow dots on the other. Line L2 represents the similarity threshold used in related technologies to determine whether authentication has been passed. Lines L1 and L2 divide the two-dimensional plane into four regions: A1, A2, A3, and A4. Figure 2 As can be seen, if a similarity threshold is used to determine whether identity authentication passes, sample data in region A2 that should actually fail will be misclassified as successful, and sample data in region A4 that should actually pass will be misclassified as failed, resulting in a high false positive rate for identity authentication. However, using the classification plane in this embodiment can accurately obtain the identity authentication results corresponding to the sample data in region A2 and region A4, thus reducing the false positive rate.

[0030] In step S102, in the second stage, based on the biometrics of the second user within the preset user range, the second target data obtained by matching the biometrics of the second user with the registered biometric database is acquired, and the second target data is classified using a classification model to obtain the classification result.

[0031] The second stage is where the classification model is formally used for identity recognition. The second user includes users within a pre-defined user range for identity recognition in this stage. The second user can be the same as the first user or a different user; this is not limited here. During the second user's identity authentication process, the acquired biometric features of the second user are matched one-to-one with the registered biometric features in the registered biometric database to obtain the second target data.

[0032] The second target data may include similarity parameters and environmental factor parameters. The similarity parameters in the second target data include the similarity parameters between the second user's biometrics and the registered biometrics. The type of the similarity parameters in the second target data is consistent with the type of the similarity parameters in the first target data, and the type of the environmental factor parameters in the second target data is consistent with the type of the environmental factor parameters in the first target data. Please refer to the relevant descriptions in the above embodiments; they will not be repeated here.

[0033] The biometric features of the second user are matched one-to-one with those in the registered biometric database, yielding multiple similarity parameters. The parameter with the highest similarity score is the most representative of the second user's characteristics. In some examples, the similarity parameters in the second target data may include the highest similarity score obtained by matching the first user's biometric features with the registered biometric database. Using this second target data results in the highest success rate and accuracy for identity verification.

[0034] The classification result indicates whether the second user has passed authentication. The classification plane in the classification model can divide the sample data into two categories: those who have passed authentication and those who have failed authentication. For users within the preset user range, the classification result obtained by the classification plane in the classification model based on the second target data, indicating whether the second user has passed authentication, has a high accuracy.

[0035] The application areas of classification results are wide-ranging and not limited here. For example, classification results that indicate whether identity authentication has been passed can be applied to areas such as check-in, attendance tracking, access control, and payment.

[0036] In this embodiment, in the first stage, during the matching process between the first user within the preset user range and the registered biometric database using biometric features, first target data including similarity parameters and environmental factor parameters is collected. This first target data is used as sample data for model training, resulting in a classification model with a classification plane capable of dividing the sample data into two categories: successful and unsuccessful identity verification. In the second stage after obtaining the classification model, second target data including similarity parameters and environmental factor parameters is obtained based on the matching of the biometric features of the second user within the preset user range and the registered biometric database. This second target data is then classified using the classification model, and the resulting classification result characterizes whether the user has passed identity verification. The classification model fully utilizes multi-dimensional elements such as similarity parameters and environmental factors obtained through machine learning algorithms. For second users within the same preset user range, the classification plane in the model can accurately characterize whether the second user has passed identity verification based on the second target data, minimizing the possibility of users who are the same as the registered user failing identity verification or users who are not the same as the registered user passing identity verification, thereby reducing the false positive rate of identity verification.

[0037] Furthermore, the biometric-based identity recognition method of this application is applied to the terminal device side in closed scenarios. The operating environment of the terminal device side is very complex, and environmental factors participate in model training, which can also enable the trained classification model to adapt to the complex operating environment of the terminal device side, further improving the accuracy of identity recognition. Moreover, the biometric-based identity recognition method of this application has terminal portability. Different sample data are collected in different preset user ranges, i.e., different closed scenarios, which can train personalized classification models. Identity recognition in terminal devices under different closed scenarios can be optimized differently, resulting in a systematic improvement.

[0038] In some embodiments, in order to obtain a classification plane that can distinguish between authentication success and authentication failure as accurately as possible, a convex optimization method can be used to obtain the classification plane, thereby obtaining a classification model. Figure 3 A flowchart illustrating a biometric-based identity recognition method provided in another embodiment of this application. Figure 3 and Figure 1 The difference is that, Figure 1 Step S101 can be further refined as follows: Figure 3 Steps S1011 to S1014 are shown.

[0039] In step S1011, in the first stage, at least a portion of the first target data is collected by matching the biometrics of the first user within the preset user range with the registered biometric database.

[0040] In step S1012, a portion of the sample data is selected as the training sample dataset, and a convex quadratic programming model and constraints are constructed based on the training sample dataset and preset penalty parameters.

[0041] The sample data can be divided into multiple sample datasets, one or more of which are used as training sample datasets. Based on the training sample datasets and penalty parameters, a convex quadratic programming model and constraints can be constructed. The penalty parameters control the degree to which the regularization process in the machine learning algorithm adjusts the error. Based on each sample data point in the training sample dataset and the penalty parameters, a convex quadratic programming model and constraints can be constructed. The constraints are equations, which are used to derive the classification plane in subsequent processes based on the convex quadratic programming model and constraints.

[0042] For example, the expression of the convex quadratic programming model can be shown in equation (1), and the constraints can be shown in equation (2):

[0043]

[0044]

[0045] Among them, (x i y i (x) represents the i-th sample data in the training sample dataset and its corresponding label; j y j ) represents the j-th sample data in the training sample dataset and its corresponding label; α i This is the coefficient term corresponding to the i-th sample data, which includes a penalty parameter; α j The coefficient term corresponds to the j-th sample data, and this coefficient term includes a penalty parameter; st represents the constraint condition.

[0046] In step S1013, the classification plane is determined based on the convex quadratic programming model and constraints.

[0047] The optimal solution of the convex quadratic programming model can be obtained based on the convex quadratic programming model and the constraints. The parameters of the classification plane can be obtained based on the optimal solution, and then the classification plane can be obtained based on the parameters of the classification plane.

[0048] In some examples, the parameters of the classification plane may include coefficient parameters and constant parameters. Specifically, the coefficient parameters and constant parameters can be obtained by solving a convex quadratic programming model using constraints; based on the coefficient parameters and constant parameters, the classification plane is determined. Solving the convex quadratic programming model using constraints yields the optimal solution, from which the coefficient parameters and constant parameters can be calculated.

[0049] For example, if the convex quadratic programming model and constraints are as shown in equations (1) and (2) above, the optimal solution can be shown in equation (3) below. The coefficient parameters and constant parameters of the classification plane calculated based on the optimal solution can be shown in equations (4) and (5) below. The expression of the classification plane can be shown in equation (6) below.

[0050]

[0051]

[0052]

[0053] ω * ·x+b * =0 (6)

[0054] Where, α * The optimal solution for the convex quadratic programming model; ω * b is the coefficient parameter; * is a constant parameter; x is the target data required for classification in the classification plane, which can represent sample data here; the meanings of other parameters can be found in the relevant descriptions in the above embodiments, and will not be repeated here.

[0055] In step S1014, a classification model is obtained based on the classification plane.

[0056] Based on the expression of the classification plane, the expression of the classification model can be constructed. The expression of the classification model is then found to include the expression of the classification plane.

[0057] For example, if the expression for the classification plane is as shown in equation (6) above, then the expression for the classification model can be shown in equation (7) below:

[0058] f(x) = sign(ω) * ·x+b * (7)

[0059] Wherein, sign represents the sign function; the meanings of other parameters in equation (7) above can be found in the relevant descriptions in the above embodiments, and will not be repeated here.

[0060] Once the coefficient parameters and constant parameters are determined, the classification plane and classification model can be obtained. The establishment and use of the classification plane and classification model can be achieved more quickly and conveniently by saving, transmitting and using the coefficient parameters and constant parameters.

[0061] It should be noted that the sample data in the above embodiments can be sample data that has already undergone normalization and other feature engineering vectorization. The classification plane obtained through the above steps can effectively divide the sample data into two categories: those that have passed identity authentication and those that have failed identity authentication.

[0062] In some embodiments, to ensure the accuracy of the classification model, after obtaining the classification model, an accuracy verification can be performed. Only if the accuracy verification passes can the classification model be put into use to perform identity recognition based on the decisions made by the classification model. Specifically, after obtaining the classification model including the classification plane, a portion of the sample data can be selected as a verification sample dataset. The verification sample dataset is input into the classification model to obtain the classification result corresponding to the verification sample dataset. Based on the classification result corresponding to the verification sample dataset, the verification accuracy of the classification model is obtained. If the verification accuracy is lower than a preset reliable accuracy threshold, a new training sample dataset is obtained, and the model is trained using the new training sample dataset until a classification model with a verification accuracy higher than or equal to the reliable accuracy threshold is obtained. If the verification accuracy is higher than or equal to the reliable accuracy threshold, the classification model is activated.

[0063] The sample data in the validation dataset differs from that in the training dataset. The validation dataset is used to verify the accuracy of the classification model. Based on whether the accuracy of the validated model meets the usage requirements, it can be determined whether the classification model can be deployed. In some examples, a portion of the sample data can also be selected as a test dataset. The test dataset can be used to test the classification model and determine its ability to identify individuals on data other than the training dataset.

[0064] After obtaining the classification results corresponding to the verification sample dataset, the classification results and labels of the sample data in the verification sample dataset can be compared one by one. Based on the number of sample data whose classification results match the labels and the number of sample data whose classification results do not match the labels, the precision and recall of the verification sample dataset in the classification model can be calculated. Based on the precision, recall, and preset adjustment parameters of the verification sample dataset in the classification model, the verification accuracy of the classification model can be obtained. The precision of the classification model represents the accuracy of the classification results output by the classification model, which can be obtained by the ratio of the number of sample data whose classification results match the labels to the number of sample data in the verification sample dataset. The recall of the classification model represents the probability that the classification model will classify the sample data in the verification sample dataset that should pass identity authentication into the identity authentication pass class. The adjustment parameters can adjust the tendency of the verification accuracy to affect the precision and recall. For example, the verification accuracy can be calculated according to the following formula (8):

[0065]

[0066] Among them, F βTo verify accuracy; P is precision; R is recall; β is an adjustment parameter. In some examples, verification accuracy is more inclined towards recall than precision, and the adjustment parameter β can be set to a value greater than 1 to focus on optimizing cases where identity verification should have been successful but the classification model resulted in a failure. For example, the adjustment parameter β could be 1.2. In other examples, verification accuracy is more inclined towards recall than recall, and the adjustment parameter β can be set to a value less than 1.

[0067] The reliable accuracy threshold is a threshold representing the verification accuracy of a classification model that has passed accuracy validation. It can be set according to the scenario, requirements, experience, etc., and is not limited here. If the verification accuracy is higher than or equal to the reliable accuracy threshold, it means that the classification model's accuracy has met the requirements for deployment, and the terminal device can control the classification model to take effect and officially use it for identity authentication. If the verification accuracy is lower than the reliable accuracy threshold, it means that the classification model's accuracy has not yet met the requirements for deployment. It is necessary to continue training the classification model using a new training sample dataset, optimize the classification model, and then perform accuracy validation on the optimized classification model again. This process is repeated until the verification accuracy of the classification model is higher than or equal to the reliable accuracy threshold.

[0068] In some embodiments, to further improve the effectiveness of identity recognition, a portion of the second target data with low similarity can be filtered out using a similarity judgment threshold before using the classification model for identity recognition. The classification process of the second target data using the classification model in step S102 above, to obtain the classification result, can be refined as follows: when the similarity parameter in the second target data represents a similarity higher than or equal to the similarity judgment threshold, the second target data is classified using the classification model to obtain the classification result.

[0069] A similarity threshold can be used to filter out some second-target data that fails authentication. The similarity threshold can be set according to the scenario, requirements, experience, etc., and is not limited here. If the similarity parameter in the second-target data is lower than the similarity threshold, an authentication failure result can be given directly. Only when the similarity parameter in the second-target data is higher than the similarity threshold is a more refined authentication using a classification model. By filtering with a similarity threshold, the number of second-target data with excessively low similarity processed by the classification model can be reduced, thereby improving the efficiency of authentication.

[0070] In some embodiments, during the formal operation phase of the classification model, the model can be automatically optimized, and the fluctuations in the biometric characteristics of the users undergoing identity verification can be fed back in a timely manner to correct errors before they occur, thereby ensuring the accuracy of the classification model. Figure 4A flowchart illustrating a biometric-based identity recognition method provided in another embodiment of this application. Figure 4 and Figure 1 The difference is that, Figure 4 The biometric-based identity recognition method shown also includes steps S103 to S105.

[0071] In step S103, the center of the historical data set that is the same as the classification result of the second target data is obtained.

[0072] After the classification model performs identity recognition, the second target data can be assigned to the corresponding historical data set according to the classification results output by the classification model. The historical data set includes the second target data corresponding to the identity authentication performed before this identity authentication. In this embodiment, there may be two historical data sets: one historical data set includes the second target data that has passed identity authentication, and the other historical data set includes the second target data that has failed identity authentication. The center of the historical data set can characterize the cohesion standard of the second target data as historical data, and can be calculated from the second target data in the historical data set. In some examples, the center of the historical data set can be the average value of the second target data in the historical data set. For example, the center of the historical data set can be calculated according to the following formula (9):

[0073]

[0074] in, The center of the historical dataset; n is the number of second target data points in the historical dataset; x i This refers to the i-th second target data in the historical dataset.

[0075] If the classification result of the second target data in this identity authentication is that the identity authentication is successful, then obtain the center of the historical data set including the second target data that has been successfully authenticated; if the classification result of the second target data in this identity authentication is that the identity authentication is successful, then obtain the center of the historical data set including the second target data that has been successfully authenticated.

[0076] In step S104, the deviation type is determined based on the second target data, the historical data set with the same classification result as the second target data, and the center.

[0077] Second target data in the historical dataset corresponding to the same classification result share similarities. Under normal circumstances, second target data in the same historical dataset should be cohesive. By comparing the second target data involved in this identification process, the historical dataset with the same classification result, and the center of the historical dataset, the dispersion of each second target data under this classification result can be determined. Under normal circumstances, second target data corresponding to the same classification result should be cohesive. If there are second target data in the second target data corresponding to the same classification result that deviate significantly from the center of the historical dataset, then deviation has occurred. The type of deviation can be determined based on the degree of deviation between the second target data corresponding to the same classification result and the center of the historical dataset corresponding to that classification result. It should be noted that the second target data corresponding to the same classification result includes both the second target data involved in this identification process and the second target data in the historical dataset corresponding to that classification result. Different deviation types require different correction methods to ensure the accuracy of identification using the classification model.

[0078] In some examples, the deviation type may include user group offset, which is a situation where the distance between the second target data of a large number of users and the center of the historical data set exceeds a preset error distance threshold. That is, a situation occurs where the second target data of a large number of users consistently and significantly deviates from the center of the historical data set. The second target data can be added to a historical data set with the same classification result as the second target data. Based on the historical data set after adding the second target data and its center, the center distance corresponding to the historical data set after adding the second target data is calculated. If, within a first preset time period ending at the current moment, the center distance corresponding to the historical data set continuously increases with the addition of the second target data, the deviation type is determined to be user group offset.

[0079] The data center distance is an indicator for evaluating the dispersion of a dataset relative to its center; that is, it characterizes the degree of dispersion of a dataset relative to its center. The data center distance of a historical dataset is used to evaluate the degree of dispersion of a historical dataset relative to its center. This dispersion can, to some extent, reflect the similarity of the second target data in the historical dataset in the multi-dimensional space. The larger the data center distance of a historical dataset, the greater the dispersion of the historical dataset relative to its center, and the more severe the deviation of the second target data in the historical dataset from the center; the smaller the data center distance of a historical dataset, the smaller the dispersion of the historical dataset relative to its center, and the more closely the second target data in the historical dataset is aligned with the center. For example, when the center of the historical dataset is as shown in equation (9) above, the data center distance can be calculated as shown in equation (10) below:

[0080]

[0081] Among them, l k n is the distance to the data center; n is the number of second target data in the historical dataset; x i The i-th second target data in the historical data set; The center of the historical data set is denoted by k; k is the center parameter, which can be 1 here, indicating that the center of the historical data set is the centroid of the historical data set.

[0082] The duration of the first preset time period can be set according to the scenario, needs, experience, etc., and is not limited here. If the deviation type is user group offset, it is very likely that the adaptability of the classification model to users within the preset user range has decreased, resulting in a phenomenon that the second target data of a large number of users within the preset user range is seriously deviated from the center of the historical data set. Specifically, this can be manifested as the center distance of the historical data set continuously increasing with the number of identity recognitions. That is, the center distance of the historical data within the first preset time period up to the current moment continuously increases with the addition of the second target data.

[0083] In other examples, the deviation type may also include user-specific offset, which is when the distance between the second target data of an individual user and the center of the historical data set is greater than a preset error distance threshold. That is, there is a situation where the second target data of an individual user is consistently and significantly deviated from the center of the historical data set. If, within the first preset time period ending at the current time, the distance between the center of the historical data set and the center does not continuously increase with the addition of the second target data, the target distance of the second target data from the center is calculated based on the second target data and the center. If the target distance is greater than the preset error distance threshold, the historical target distance of the second user's second target data from the center of the historical data set is obtained. If, within the second preset time period ending at the current time, the historical target distance of the second user is consistently greater than the preset error distance threshold, the deviation type is determined to be user-specific offset.

[0084] User-specific offset differs from user group offset. Since it involves offsets occurring in individual users within a preset user range, it does not cause the distance between the data centers of the historical dataset and the data centers of the second target dataset to continuously increase over a period of time with the addition of the second target data. If, within the first preset time period ending at the current moment, the distance between the data centers of the historical dataset and the data centers of the second target dataset does not continuously increase with the addition of the second target data, it can be determined that no user group offset has occurred, and a judgment can be made regarding whether a user-specific offset has occurred. Whether a user-specific offset has occurred can be determined by whether a large amount of the second target data significantly deviates from the center of the historical dataset within the second preset time period ending at the current moment. The target distance can be calculated based on the centers of the second target data and the historical dataset. The target distance can include the distance between the second target data and the center of the historical dataset whose classification result is the same as that of the second target data. In some examples, the target distance can be calculated using Euclidean distance; that is, the target distance can include the Euclidean distance between the second target data and the center of the historical dataset whose classification result is the same as that of the second target data. The target distance can characterize the degree to which the second target data participating in this identification deviates from the aforementioned center, and to a certain extent, it can reflect the similarity between the second target data participating in this identification and the second target data in the historical dataset in multi-dimensional space. The greater the target distance, the greater the deviation from the aforementioned center; the smaller the target distance, the smaller the deviation from the aforementioned center.

[0085] The second preset time period can be set according to the scenario, needs, experience, etc., and is not limited here. If the deviation type is user-specific deviation, it is highly likely that there is a problem with the data of the second user, causing the second user's second target data to deviate significantly from the center of the historical data set. The preset error distance threshold is the maximum acceptable error of the target distance, which can be used to determine whether the second target data deviates significantly from the center of the historical data set. It can be set according to the scenario, needs, experience, etc., and is not limited here. If the target distance of the second target data from the center is greater than the preset error distance threshold, it means that the second target data participating in the current identity recognition has a large deviation relative to the previous second target data with the same classification result. Under normal circumstances, such a deviation will not occur. If this large deviation continues to occur for the second user, it can be determined that a user-specific deviation has occurred. Specifically, it can be manifested as the second user's historical target distance continuously exceeding the preset error distance threshold within the second preset time period ending at the current time. The historical target distance includes the target distance of the second user's second target data participating in identity recognition before the current time from the center of the corresponding historical data set.

[0086] In some examples, the deviation type can also be determined by the number of users corresponding to the second target data whose target distance is greater than a preset error distance threshold. For clarity, the concept of historical target deviation data is introduced. Historical target deviation data includes historical data where the target distance from the center is greater than the preset error distance threshold. This historical data includes the second target data obtained during identity verification prior to the current verification. The deviation type can be determined as user group deviation when the target distance between the second target data and the center of the historical data set is greater than the preset error distance threshold, and the sum of the number of second users corresponding to the second target data and the number of users corresponding to the historical target deviation data is greater than or equal to a group size threshold. Conversely, the deviation type can be determined as individual user deviation when the target distance between the second target data and the center of the historical data set is greater than the preset error distance threshold, and the sum of the number of second users corresponding to the second target data and the number of users corresponding to the historical target deviation data is less than the group size threshold.

[0087] Within a given period, the same user may perform multiple authentications. Therefore, the type of deviation should be determined by the number of users who have deviated. The group size threshold is used to determine whether the number of users who have deviated forms a group. It can be set based on the number of users within a preset user range, combined with the scenario, needs, and experience. If the sum of the number of users corresponding to the second target data and the number of users corresponding to the historical deviation data is greater than or equal to the group size threshold, it indicates that the number of users who have deviated is greater than or equal to the group size threshold, and a group has been formed. The deviation type can be determined as user group deviation. If the sum of the number of users corresponding to the second target data and the number of users corresponding to the historical deviation data is less than the group size threshold, it indicates that the number of users who have deviated is less than the group size threshold, and no group has been formed. The deviation type can be determined as individual user deviation.

[0088] In step S105, the correction action corresponding to the deviation type is performed according to the deviation type.

[0089] Different types of deviations have different causes. Corrective actions can be formulated based on the causes of the deviations. By executing these corrective actions, the identity authentication process can be adjusted to ensure the accuracy of identity authentication using the classification model.

[0090] In some examples, when the deviation type is user group offset, the correction action may include: using third target data obtained by matching the biometrics of users within a preset user range in the most recent preset time period with the registered biometric database as sample data for model training, so as to update the classification model.

[0091] The main reason for user group shift is the decreased adaptability of the classification model to users within the current preset user range. Therefore, it is necessary to optimize the classification model to improve its adaptability to users within the preset user range. Third-target data corresponding to users within the preset user range in the most recent preset time period can be used as sample data for model training to obtain a new classification plane, thus resulting in a new classification model. For example, if the classification model is applied to identity recognition in a school environment, since the biometric characteristics of teenagers may change collectively during their growth and development, the spontaneous optimization of the classification model can avoid misjudgments in identity recognition and ensure the accuracy of identity recognition.

[0092] In other examples, where the deviation type is user-specific offset, the correction action may include issuing a prompt message. This prompt message is used to remind the second user corresponding to the second target data and the user corresponding to the historical target deviation data to update their registered biometrics in the registered biometrics database.

[0093] The main reasons for user personal deviation are poor quality of registered biometric data in the registration biometric database or failure to promptly clean up biometric data from deregistered users. The prompt message can indicate that the current deviation type is user personal deviation, and can also prompt the corresponding user to update their registered biometric data in the database. Improving the quality of registered biometric data in the database can avoid misjudgments in identity verification and ensure the accuracy of identity recognition.

[0094] By judging deviation types and executing corrective actions, timely and spontaneous corrections can be made before identity recognition misjudgments occur, potential anomalies can be detected in advance, and identity recognition misjudgments can be avoided in a timely manner, ensuring the accuracy of identity recognition.

[0095] To facilitate understanding, the following will be combined with... Figure 5 The biometric-based identity recognition methods in the embodiments of this application will be described in detail. Figure 5 A logical schematic diagram of an example of a biometric-based identity recognition method provided in this application embodiment, as shown below. Figure 5 As shown, the implementation of the biometric-based identity recognition method can include three stages. The first stage is the first stage in the above embodiment, which is the initial classification model training stage. The second stage is the second stage in the above embodiment, which is the identity recognition stage where the classification model is officially running. The third stage is the feedback and optimization stage of the overall identity recognition.

[0096] The first phase may include sample collection, model training, and accuracy verification. Sample collection involves gathering data—the initial target data—during the trial run of the classification model. Model training utilizes this initial target data for preliminary model training to obtain the classification model. Accuracy verification verifies the accuracy of the classification model, determining whether it is suitable for formal operation.

[0097] The second stage may include matching, threshold decision, classification model decision, and recognition results. Matching may include environmental factor collection and similarity calculation. Second target data is obtained through environmental factor collection and similarity calculation. Threshold decision can be used to filter out some second target data with low similarity. Retaining the threshold decision improves the overall reliability of identity recognition. The second target data that passes the threshold decision enters the classification model decision, obtaining a recognition result that characterizes whether the user has passed identity authentication, i.e., the classification result in the above embodiment. In some cases, once the classification model is running stably, it can also operate independently of the threshold decision. It should be noted that the sample collection in the first stage can also utilize matching to obtain environmental factor parameters and similarity parameters.

[0098] The third stage may include sample storage, offset calculation, and feedback optimization. The second target data input into the classification model in the second stage can be stored as sample data in the database and participate in offset calculation. Based on the offset calculation results, it is determined whether an offset has occurred, and if so, the type of offset is determined, followed by feedback optimization. The feedback optimization part may re-enter the first stage to optimize the classification model and avoid systematic errors.

[0099] The second aspect of this application provides a biometric-based identity recognition device. Figure 6 This is a schematic diagram of the structure of a biometric-based identity recognition device provided in an embodiment of this application, as shown below. Figure 6 As shown, the biometric-based identity recognition device 200 may include a model training module 201 and an identity recognition module 202.

[0100] The model training module 201 can be used in the first stage to collect at least a portion of the first target data obtained by matching the biometrics of the first user within a preset user range with the registered biometric database, and use the at least a portion of the first target data as sample data for model training to obtain a classification model including a classification plane.

[0101] The primary target data includes similarity parameters and environmental factor parameters. The classification plane divides the sample data into two categories: those that have passed identity verification and those that have failed.

[0102] In some examples, the similarity parameters in the first target data include the top m similarity parameters from high to low obtained by matching the first user's biometrics with the registered biometric database, where m is an integer greater than or equal to 2.

[0103] The identity recognition module 202 can be used in the second stage to obtain the second target data obtained by matching the biometrics of the second user with the registered biometric database based on the biometrics of the second user within the preset user range, and to classify the second target data using a classification model to obtain the classification result.

[0104] The second set of target data includes similarity parameters and environmental factor parameters. The classification result indicates whether the second user has passed identity authentication.

[0105] In some examples, the similarity parameter in the second target data includes the highest similarity parameter obtained by matching the second user's biometrics with the registered biometric database.

[0106] In some examples, environmental factor parameters include one or more of the following: feature acquisition time information, feature acquisition illumination information, feature acquisition occlusion information, feature ambiguity information, feature size information, and feature acquisition module information.

[0107] In this embodiment, in the first stage, during the matching process between the first user within the preset user range and the registered biometric database using biometric features, first target data including similarity parameters and environmental factor parameters is collected. This first target data is used as sample data for model training, resulting in a classification model with a classification plane capable of dividing the sample data into two categories: successful and unsuccessful identity verification. In the second stage after obtaining the classification model, second target data including similarity parameters and environmental factor parameters is obtained based on the matching of the biometric features of the second user within the preset user range and the registered biometric database. This second target data is then classified using the classification model, and the resulting classification result characterizes whether the user has passed identity verification. The classification model fully utilizes multi-dimensional elements such as similarity parameters and environmental factors obtained through machine learning algorithms. For second users within the same preset user range, the classification plane in the model can accurately characterize whether the second user has passed identity verification based on the second target data, minimizing the possibility of users who are the same as the registered user failing identity verification or users who are not the same as the registered user passing identity verification, thereby reducing the false positive rate of identity verification.

[0108] Furthermore, the biometric-based identity recognition method of this application is applied to the terminal device side in closed scenarios. The operating environment of the terminal device side is very complex, and environmental factors participate in model training, which can also enable the trained classification model to adapt to the complex operating environment of the terminal device side, further improving the accuracy of identity recognition. Moreover, the biometric-based identity recognition method of this application has terminal portability. Different sample data are collected in different preset user ranges, i.e., different closed scenarios, which can train personalized classification models. Identity recognition in terminal devices under different closed scenarios can be optimized differently, resulting in a systematic improvement.

[0109] In some embodiments, the model training module 201 may be specifically used to: select a portion of the sample data as a training sample dataset; construct a convex quadratic programming model and constraints based on the training sample dataset and preset penalty parameters; determine a classification plane based on the convex quadratic programming model and constraints; and obtain a classification model based on the classification plane.

[0110] In some examples, the model training module 201 may be specifically used to: use constraints to solve a convex quadratic programming model to obtain coefficient parameters and constant parameters; and determine a classification plane based on the coefficient parameters and constant parameters.

[0111] In some embodiments, the model training module 201 is further configured to: select a portion of the sample data as a validation sample dataset, input the validation sample dataset into the classification model, and obtain the classification result corresponding to the validation sample dataset; obtain the validation accuracy of the classification model based on the classification result corresponding to the validation sample dataset; if the validation accuracy is lower than a preset reliable accuracy threshold, obtain a new training sample dataset and use the new training sample dataset to train the model until a classification model with a validation accuracy higher than or equal to the reliable accuracy threshold is obtained; and control the classification model to take effect if the validation accuracy is higher than or equal to the reliable accuracy threshold.

[0112] In some embodiments, the identity recognition module 202 may be specifically used to: when the similarity represented by the similarity parameter in the second target data is higher than or equal to the similarity judgment threshold, use a classification model to classify the second target data to obtain a classification result.

[0113] In some embodiments, the biometric identification device 200 may further include an offset optimization module. The offset optimization module may be used to: obtain the center of a historical data set that is the same as the classification result of the second target data; determine the deviation type based on the second target data, the historical data set that is the same as the classification result of the second target data, and the center; and perform a correction action corresponding to the deviation type.

[0114] In some examples, the deviation type includes user group offset. The offset optimization module can be specifically used to: add the second target data into a historical data set with the same classification result as the second target data; calculate the data center distance corresponding to the historical data set after adding the second target data based on the historical data set and the center; if the data center distance corresponding to the historical data set continuously increases with the addition of the second target data within a first preset time period ending at the current time, the deviation type is determined to be user group offset.

[0115] In some examples, the deviation type also includes user-specific offset. The offset optimization module can also be specifically used for: if, within a first preset time period ending at the current moment, the distance from the center of the historical data set to the center does not continuously increase with the addition of the second target data, calculating the target distance of the second target data from the center based on the second target data and the center; if the target distance is greater than a preset error distance threshold, obtaining the historical target distance of the second user's second target data from the center of the historical data set; if, within a second preset time period ending at the current moment, the historical target distance of the second user is continuously greater than the preset error distance threshold, determining the deviation type as user-specific offset.

[0116] In some examples, the offset optimization module can be specifically used to: determine the deviation type as user group offset when the sum of the number of the second user corresponding to the second target data and the number of users corresponding to the historical target deviation data is greater than or equal to the group size threshold, and the historical target deviation data includes historical data where the target distance from the deviation center is greater than a preset error distance threshold; and determine the deviation type as individual user offset when the sum of the number of the second user corresponding to the second target data and the number of users corresponding to the historical target deviation data is less than the group size threshold.

[0117] In some examples, when the deviation type is user group offset, the correction action includes: using the third target data obtained by matching the biometrics of users within a preset user range in the most recent preset time period with the registered biometric database as sample data for model training to update the classification model. The third target data includes similarity parameters and environmental factor parameters.

[0118] In some examples, when the deviation type is user-specific offset, the correction action includes: issuing a prompt message to prompt the second user corresponding to the second target data and the user corresponding to the historical target deviation data to update their registered biometrics in the registered biometrics database.

[0119] The biometric-based identity recognition method in the above embodiments can also be implemented by a system. The system may include a model training device and a terminal device. The model training device can execute the steps performed by the model training module 201 in the above embodiments, and the terminal device can execute the steps performed by the identity recognition module 202 in the above embodiments. The steps performed by the offset optimization module in the above embodiments can be completed collaboratively by the model training device and the terminal device.

[0120] A third aspect of this application also provides a terminal device. Figure 7 This is a schematic diagram of the structure of a terminal device provided in an embodiment of this application. Figure 7 As shown, the terminal device 300 includes a memory 301, a processor 302, and a computer program stored in the memory 301 and capable of running on the processor 302.

[0121] In some examples, the processor 302 described above may include a central processing unit (CPU), or an application-specific integrated circuit (ASIC), or one or more integrated circuits that may be configured to implement the embodiments of this application.

[0122] Memory 301 may include read-only memory (ROM), random access memory (RAM), disk storage media device, optical storage media device, flash memory device, electrical, optical, or other physical / tangible memory storage device. Therefore, typically, memory includes one or more tangible (non-transitory) computer-readable storage media (e.g., memory devices) encoded with software including computer-executable instructions, and when the software is executed (e.g., by one or more processors), it is operable to perform the operations described with reference to the biometric-based identification method according to embodiments of this application.

[0123] The processor 302 runs a computer program corresponding to the executable program code by reading the executable program code stored in the memory 301, so as to implement the biometric-based identity recognition method in the above embodiments.

[0124] In some examples, terminal device 300 may also include communication interface 303 and bus 304. For example, Figure 7 As shown, the memory 301, processor 302, and communication interface 303 are connected through bus 304 and complete communication with each other.

[0125] The communication interface 303 is mainly used to realize communication between various modules, devices, units and / or equipment in the embodiments of this application. Input devices and / or output devices can also be connected through the communication interface 303.

[0126] Bus 304 includes hardware, software, or both, that couples the components of terminal device 300 together. For example, and not limitingly, bus 304 may include an Accelerated Graphics Port (AGP) or other graphics bus, an Enhanced Industry Standard Architecture (EISA) bus, a Front Side Bus (FSB), a Hyper Transport (HT) interconnect, an Industry Standard Architecture (ISA) bus, an Infinite Bandwidth Interconnect, a Low Pin Count (LPC) bus, a memory bus, a Micro Channel Architecture (MCA) bus, a Peripheral Component Interconnect (PCI) bus, a PCI-Express (PCI-E) bus, a Serial Advanced Technology Attachment (SATA) bus, a Video Electronics Standards Association Local Bus (VLB) bus, or other suitable buses, or a combination of two or more of these. Where appropriate, bus 304 may include one or more buses. Although specific buses are described and illustrated in the embodiments of this application, this application considers any suitable bus or interconnection.

[0127] A fourth aspect of this application also provides a computer-readable storage medium storing computer program instructions. When executed by a processor, these computer program instructions can implement the biometric-based identity recognition method described in the above embodiments and achieve the same technical effect. To avoid repetition, further details are omitted here. The aforementioned computer-readable storage medium may include non-transitory computer-readable storage media, such as read-only memory (ROM), random access memory (RAM), magnetic disks, or optical disks, etc., and is not limited thereto.

[0128] This application also provides a computer program product. When the instructions in the computer program product are executed by the processor of an electronic device, the electronic device performs the biometric identification method described above and achieves the same technical effect. To avoid repetition, it will not be described again here.

[0129] It should be clarified that the various embodiments in this specification are described in a progressive manner, and the same or similar parts between the various embodiments can be referred to mutually. Each embodiment focuses on describing the differences from other embodiments. For the device embodiments, equipment embodiments, computer-readable storage medium embodiments, and computer program product embodiments, the relevant parts can be referred to the description section of the method embodiments. This application is not limited to the specific steps and structures described above and shown in the figures. Those skilled in the art can make various changes, modifications, and additions, or change the order of steps, after understanding the spirit of this application. Furthermore, for the sake of brevity, detailed descriptions of known methods and techniques are omitted here.

[0130] The aspects of this application have been described above with reference to flowchart illustrations and / or block diagrams of methods, apparatus (systems), and computer program products according to embodiments of this application. It should be understood that each block in the flowchart illustrations and / or block diagrams, and combinations of blocks in the flowchart illustrations and / or block diagrams, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, a special-purpose computer, or other programmable data processing apparatus to produce a machine such that these instructions, executable via the processor of the computer or other programmable data processing apparatus, enable the implementation of the functions / actions specified in one or more blocks of the flowchart illustrations and / or block diagrams. Such a processor can be, but is not limited to, a general-purpose processor, a special-purpose processor, a special application processor, or a field-programmable logic circuit. It is also understood that each block in the block diagrams and / or flowcharts, and combinations of blocks in the block diagrams and / or flowcharts, can also be implemented by dedicated hardware performing the specified functions or actions, or can be implemented by a combination of dedicated hardware and computer instructions.

[0131] Those skilled in the art will understand that the above embodiments are exemplary and not restrictive. Different technical features appearing in different embodiments can be combined to achieve beneficial effects. Based on a study of the drawings, specification, and claims, those skilled in the art should be able to understand and implement other variations of the disclosed embodiments. In the claims, the term "comprising" does not exclude other means or steps; the quantifier "a" does not exclude a plurality; the terms "first" and "second" are used to identify names and not to indicate any particular order. No reference numerals in the claims should be construed as limiting the scope of protection. The functionality of multiple parts appearing in the claims can be implemented by a single hardware or software module. The appearance of certain technical features in different dependent claims does not mean that these technical features cannot be combined to achieve beneficial effects.

Claims

1. A biometric-based identity recognition method, characterized in that, include: In the first stage, at least a portion of the first target data is collected by matching the biometrics of the first user within a preset user range with the registered biometric database. The at least portion of the first target data is used as sample data for model training to obtain a classification model including a classification plane. The first target data includes similarity parameters and environmental factor parameters. The classification plane divides the sample data into two categories: identity authentication passed and identity authentication failed. In the second stage, based on the biometrics of the second user within the preset user range, second target data is obtained by matching the biometrics of the second user with the registered biometric database. The classification model is used to classify the second target data to obtain a classification result. The second target data includes similarity parameters and environmental factor parameters. The classification result indicates whether the second user has passed identity authentication. After classifying the second target data using the classification model to obtain the classification result, the method further includes: Obtain the center of the historical data set that is identical to the classification result of the second target data; Based on the second target data, the historical data set that is the same as the classification result of the second target data, and the center, determine the deviation type; Based on the deviation type, perform the correction action corresponding to the deviation type.

2. The method according to claim 1, characterized in that, The step of using at least a portion of the first target data as sample data for model training to obtain a classification model including a classification plane includes: A portion of the sample data is selected as a training sample dataset, and a convex quadratic programming model and constraints are constructed based on the training sample dataset and preset penalty parameters. The classification plane is determined based on the convex quadratic programming model and constraints. The classification model is obtained based on the classification plane.

3. The method according to claim 2, characterized in that, The classification plane is determined based on the convex quadratic programming model and constraints, including: Using the aforementioned constraints, the convex quadratic programming model is solved to obtain the coefficient parameters and constant parameters; The classification plane is determined based on the coefficient parameters and the constant parameters.

4. The method according to claim 2, characterized in that, After training the model using at least a portion of the first target data as sample data to obtain a classification model including a classification plane, the method further includes: A portion of the sample data is selected as a validation sample dataset, and the validation sample dataset is input into the classification model to obtain the classification result corresponding to the validation sample dataset. Based on the classification results corresponding to the validation sample dataset, the validation accuracy of the classification model is obtained; If the verification accuracy is lower than a preset reliable accuracy threshold, a new training sample dataset is obtained, and the model is trained using the new training sample dataset until the classification model with a verification accuracy higher than or equal to the reliable accuracy threshold is obtained. If the verification accuracy is higher than or equal to the reliable accuracy threshold, the classification model is activated.

5. The method according to claim 1, characterized in that, The process of classifying the second target data using the classification model to obtain the classification result includes: If the similarity parameter in the second target data represents a similarity higher than or equal to the similarity judgment threshold, the classification model is used to classify the second target data to obtain the classification result.

6. The method according to claim 1, characterized in that, The deviation types include user group offset. The step of determining the deviation type based on the second target data, the historical data set with the same classification result as the second target data, and the center includes: Add the second target data to the historical data set that has the same classification result as the second target data; Based on the historical data set after adding the second target data and the center, the data center distance corresponding to the historical data set after adding the second target data is calculated; If, within a first preset time period ending at the current moment, the distance between the data center corresponding to the historical data set and the data center continuously increases with the addition of the second target data, the deviation type is determined to be user group offset.

7. The method according to claim 6, characterized in that, The deviation types also include user-specific offsets. The step of determining the deviation type based on the second target data, the historical data set with the same classification result as the second target data, and the center further includes: If, within the first preset time period ending at the current moment, the distance from the data center corresponding to the historical data set does not continuously increase with the addition of the second target data, the target distance of the second target data from the center is calculated based on the second target data and the center. If the target distance is greater than a preset error distance threshold, obtain the historical target distance of the second user's second target data from the center of the historical data set. If, within a second preset time period ending at the current moment, the historical target distance of the second user is continuously greater than the preset error distance threshold, the deviation type is determined to be user personal offset.

8. The method according to claim 1, characterized in that, When the deviation type is user group deviation, the correction action includes: using the third target data obtained by matching the biometrics of users within the preset user range in the most recent preset time period with the registered biometric database as sample data for model training, so as to update the classification model. The third target data includes similarity parameters and environmental factor parameters. When the deviation type is user-specific deviation, the correction action includes: issuing a prompt message, which prompts the second user corresponding to the second target data and the user corresponding to the historical target deviation data to update their registered biometrics in the registered biometric database.

9. The method according to any one of claims 1 to 8, characterized in that, The similarity parameters in the first target data include the top m similarity parameters from high to low obtained by matching the biometrics of the first user with the registered biometric database, where m is an integer greater than or equal to 2; The similarity parameters in the second target data include the similarity parameter with the highest similarity obtained by matching the biometrics of the second user with the registered biometric database.

10. The method according to any one of claims 1 to 8, characterized in that, Environmental factor parameters include one or more of the following: Feature acquisition time information, feature acquisition illumination information, feature acquisition occlusion information, feature ambiguity information, feature size information, and feature acquisition module information.

11. A biometric-based identity recognition device, characterized in that, include: The model training module is used in the first stage to collect at least a portion of the first target data obtained by matching the biometrics of the first user within a preset user range with the registered biometric database, and to use the at least a portion of the first target data as sample data for model training to obtain a classification model including a classification plane. The first target data includes similarity parameters and environmental factor parameters. The classification plane divides the sample data into two categories: identity authentication passed and identity authentication failed. The identity recognition module is used in the second stage to obtain second target data by matching the biometrics of the second user with the registered biometric database based on the biometrics of the second user within the preset user range, and to classify the second target data using the classification model to obtain a classification result. The second target data includes similarity parameters and environmental factor parameters, and the classification result indicates whether the second user has passed identity authentication. The offset optimization module is configured to obtain the center of a historical data set that is identical to the classification result of the second target data; and to determine the deviation type based on the second target data, the historical data set identical to the classification result of the second target data, and the center; and to perform a correction action corresponding to the deviation type based on the deviation type.

12. A terminal device, characterized in that, include: Processor and memory storing computer program instructions; When the processor executes the computer program instructions, it implements the biometric-based identity recognition method as described in any one of claims 1 to 10.

13. A computer-readable storage medium, characterized in that, The computer-readable storage medium stores computer program instructions that, when executed by a processor, implement the biometric-based identity recognition method as described in any one of claims 1 to 10.

Images