Biological characteristic authentication method, device and system, computer equipment and storage medium

Abstract

The invention relates to a biological characteristic authentication method, device and system, computer equipment, a computer readable storage medium and a computer program product. The method comprises the following steps: issuing a feature recognition model to a trusted mechanism and a plurality of application ends; receiving a plurality of registered biological characteristics sent by the trusted mechanism, and constructing a biological characteristic library based on the plurality of registered biological characteristics; receiving a to-be-authenticated biological feature sent by the target application end; and performing feature matching on the to-be-authenticated biological feature and each registered biological feature in the biological feature library, determining an authentication result of the to-be-authenticated biological feature, and feeding back the authentication result to the target application end. The feature recognition model is used for performing feature extraction on collected to-be-authenticated biological information at an application end to obtain to-be-authenticated biological features; the feature recognition model is used for performing feature extraction on registration biological information provided by each application end in a trusted mechanism to obtain a plurality of registration biological features; by adopting the method, biological information leakage can be avoided, and the security can be improved.

Description

Technical Field

[0001] The present application relates to the field of computer technology, and in particular to a biometric authentication method, apparatus, system, computer equipment, computer-readable storage medium, and computer program product. Background Art

[0002] With the rapid development of computer technology, biometric identification technology has emerged. Biometric identification technology can be used to perform identity authentication by closely combining computers with high-tech means such as optics, acoustics, biosensors and biostatistics principles.

[0003] In the traditional biometric authentication method, the server determines the identity of the object to be authenticated based on the biometric information collected by the application. During the interaction between the server and the application, it is easy to cause the leakage of biometric information. Therefore, the traditional biometric authentication method has the disadvantage of poor security. Summary of the invention

[0004] Based on this, it is necessary to provide a biometric authentication method, apparatus, system, computer device, computer-readable storage medium and computer program product that can improve security in response to the above technical problems.

[0005] In a first aspect, the present application provides a biometric authentication method. The method comprises:

[0006] The feature recognition model is issued to the trusted institution and multiple application terminals; the feature recognition model is used at the application terminal to extract features from the collected biometric information to be authenticated, and obtain the biometric features to be authenticated; the feature recognition model is used at the trusted institution to extract features from the registered biometric information provided by each of the application terminals, and obtain multiple registered biometric features;

[0007] Receiving a plurality of registered biometric features sent by the trusted institution, and building a biometric feature library based on the plurality of registered biometric features;

[0008] Determine a target application end among the multiple application ends that initiates a biometric authentication event, and receive the biometric to be authenticated sent by the target application end;

[0009] The biometric feature to be authenticated is matched with each registered biometric feature in the biometric feature database, and an authentication result of the biometric feature to be authenticated is determined and fed back to the target application end.

[0010] In the second aspect, the present application also provides a biometric authentication system. The system includes a server, a trusted institution, and multiple application terminals:

[0011] The service end sends a feature recognition model to a trusted institution and multiple application ends; the feature recognition model is used at the application end to extract features from the collected biometric information to be authenticated, and obtain the biometric features to be authenticated;

[0012] The trusted institution uses the feature recognition model to extract features from the registered biometric information provided by each of the application terminals, obtains multiple registered biometric features and sends them to the server terminal, so that the server terminal builds a biometric feature library based on the multiple registered biometric features;

[0013] The server end is also used to determine the target application end that initiates the biometric authentication event among the multiple application ends, receive the biometric to be authenticated sent by the target application end, perform feature matching on the biometric to be authenticated with each registered biometric in the biometric library, determine the authentication result of the biometric to be authenticated and feed it back to the target application end.

[0014] In a third aspect, the present application also provides a biometric authentication device. The device comprises:

[0015] The model sending module is used to send the feature recognition model to the trusted institution and multiple application terminals; the feature recognition model is used at the application terminal to extract features from the collected biometric information to be authenticated, and obtain the biometric features to be authenticated; the feature recognition model is used at the trusted institution to extract features from the registered biometric information provided by each of the application terminals, and obtain multiple registered biometric features;

[0016] A feature library construction module, used to receive a plurality of registered biometric features sent by the trusted institution, and to construct a biometric library based on the plurality of registered biometric features;

[0017] A biometric feature acquisition module to be authenticated, used to determine a target application end that initiates a biometric feature authentication event among the multiple application ends, and receive the biometric feature to be authenticated sent by the target application end;

[0018] The feature matching module is used to perform feature matching on the biometric feature to be authenticated and each registered biometric feature in the biometric feature library, determine the authentication result of the biometric feature to be authenticated and feed it back to the target application end.

[0019] In one of the embodiments, the device further comprises a security credential sending module, configured to:

[0020] Sending a security credential to the trusted institution so that the trusted institution performs security verification on the feature recognition model based on the security credential;

[0021] The feature library construction module is specifically used for:

[0022] When the feature recognition model passes the verification, a plurality of registered biometric features sent by the trusted institution are received.

[0023] In one embodiment, the feature matching module is specifically used for:

[0024] Perform feature matching on the biometric feature to be authenticated and each registered biometric feature in the biometric feature database, and determine the feature similarity between the biometric feature to be authenticated and each registered biometric feature;

[0025] In the case where there is a target registered biometric feature whose feature similarity satisfies the authentication condition, determining the authentication result of the biometric feature to be authenticated as authentication passed;

[0026] The identity identifier corresponding to the target registered biometric feature is fed back to the target application end.

[0027] In one embodiment, the device further comprises:

[0028] An initial recognition model delivery module is used to deliver the initial recognition model for biometric feature extraction to multiple application terminals;

[0029] A parameter update information acquisition module, used to obtain multiple parameter update information obtained by performing incremental training on the initial recognition model by at least a part of the multiple application terminals; the training samples of the incremental training include biological information collected during the biological feature authentication process;

[0030] The model updating module is used to update the initial recognition model based on the parameter update information to obtain a feature recognition model.

[0031] In one of the embodiments, the device also includes a model training module, which is used to: determine the type of biometric information to be identified and obtain an open source data set that matches the type of biometric information; use the open source data set to perform model training on an initial neural network to obtain an initial recognition model for biometric feature extraction.

[0032] In one embodiment, the model updating module includes:

[0033] A model parameter determination submodule, used to determine the model parameters corresponding to each of the parameter update information;

[0034] A statistical analysis submodule, for performing statistical analysis on each of the parameter update information corresponding to the model parameter for each of the model parameters, and determining an update parameter matching the model parameter;

[0035] The model updating submodule is used to update the initial recognition model based on the update parameters to obtain a feature recognition model.

[0036] In one embodiment, the statistical analysis submodule includes:

[0037] A weight determination unit, used to determine the weight of each of the parameter update information corresponding to the model parameter;

[0038] An update parameter determination unit is used to perform weighted sum processing on each parameter update information based on each weight to obtain an update parameter that matches the model parameter.

[0039] In one embodiment, the weight determination unit is specifically used for:

[0040] For each parameter update information corresponding to the model parameter, determine the number of training samples used in the process of training to obtain the parameter update information;

[0041] Based on the number of training samples corresponding to each of the parameter update information, a weight of each of the parameter update information is determined; and the weight is positively correlated with the number of training samples.

[0042] In one embodiment, the update parameter is a parameter offset; in this embodiment, the model update submodule is specifically used to:

[0043] For each of the model parameters, the initial parameter value of the model parameter is superimposed with the parameter offset of the model parameter to obtain an updated parameter value of the model parameter;

[0044] The initial recognition model is updated based on the updated parameter values ​​of the model parameters to obtain a feature recognition model.

[0045] In a fourth aspect, the present application further provides a computer device. The computer device includes a memory and a processor, the memory stores a computer program, and the processor implements the following steps when executing the computer program:

[0046] The feature recognition model is issued to the trusted institution and multiple application terminals; the feature recognition model is used at the application terminal to extract features from the collected biometric information to be authenticated, and obtain the biometric features to be authenticated; the feature recognition model is used at the trusted institution to extract features from the registered biometric information provided by each of the application terminals, and obtain multiple registered biometric features;

[0047] Receiving a plurality of registered biometric features sent by the trusted institution, and building a biometric feature library based on the plurality of registered biometric features;

[0048] Determine a target application end among the multiple application ends that initiates a biometric authentication event, and receive the biometric to be authenticated sent by the target application end;

[0049] The biometric feature to be authenticated is matched with each registered biometric feature in the biometric feature database, and an authentication result of the biometric feature to be authenticated is determined and fed back to the target application end.

[0050] In a fifth aspect, the present application further provides a computer-readable storage medium. The computer-readable storage medium stores a computer program, and when the computer program is executed by a processor, the following steps are implemented:

[0051] The feature recognition model is issued to the trusted institution and multiple application terminals; the feature recognition model is used at the application terminal to extract features from the collected biometric information to be authenticated, and obtain the biometric features to be authenticated; the feature recognition model is used at the trusted institution to extract features from the registered biometric information provided by each of the application terminals, and obtain multiple registered biometric features;

[0052] Receiving a plurality of registered biometric features sent by the trusted institution, and building a biometric feature library based on the plurality of registered biometric features;

[0053] Determine a target application end among the multiple application ends that initiates a biometric authentication event, and receive the biometric to be authenticated sent by the target application end;

[0054] The biometric feature to be authenticated is matched with each registered biometric feature in the biometric feature database, and an authentication result of the biometric feature to be authenticated is determined and fed back to the target application end.

[0055] In a sixth aspect, the present application further provides a computer program product. The computer program product includes a computer program, and when the computer program is executed by a processor, the following steps are implemented:

[0056] The feature recognition model is issued to the trusted institution and multiple application terminals; the feature recognition model is used at the application terminal to extract features from the collected biometric information to be authenticated, and obtain the biometric features to be authenticated; the feature recognition model is used at the trusted institution to extract features from the registered biometric information provided by each of the application terminals, and obtain multiple registered biometric features;

[0057] Receiving a plurality of registered biometric features sent by the trusted institution, and building a biometric feature library based on the plurality of registered biometric features;

[0058] Determine a target application end among the multiple application ends that initiates a biometric authentication event, and receive the biometric to be authenticated sent by the target application end;

[0059] The biometric feature to be authenticated is matched with each registered biometric feature in the biometric feature database, and an authentication result of the biometric feature to be authenticated is determined and fed back to the target application end.

[0060] The above-mentioned biometric authentication method, device, system, computer equipment, computer-readable storage medium and computer program product send feature recognition models to trusted institutions and multiple application terminals; receive multiple registered biometrics sent by trusted institutions, and build a biometric library based on multiple registered biometrics; determine the target application terminal that initiates the biometric authentication event among multiple application terminals, and receive the biometric to be authenticated sent by the target application terminal; perform feature matching on the biometric to be authenticated and each registered biometric in the biometric library, determine the authentication result of the biometric to be authenticated and feed it back to the target application terminal. The feature recognition model is used at the application terminal to extract features from the collected biometric information to be authenticated to obtain the biometric to be authenticated; the feature recognition model is used at the trusted institution to extract features from the registered biometric information provided by each application terminal to obtain multiple registered biometrics. That is to say, in the above-mentioned biometric authentication process, for the registered biometric information, the information collection terminal only includes the trusted institution and the application terminal that provides the registered biometric information; for the biometric information to be authenticated, the information collection terminal only includes the application terminal that collects the biometric information to be authenticated. This means that biometric authentication can be achieved without transmitting biometric information during the interaction between the trusted organization and the server, and between the application and the server. This can avoid the leakage of biometric information during the authentication interaction to a certain extent, and is conducive to improving security. BRIEF DESCRIPTION OF THE DRAWINGS

[0061] Figure 1 A diagram of an application environment of a biometric authentication method in an embodiment;

[0062] Figure 2 A schematic diagram of a biometric authentication method in one embodiment;

[0063] Figure 3 A schematic diagram of an interactive process of biometric information registration in one embodiment;

[0064] Figure 4 A schematic diagram of an interactive process of performing palmprint recognition in one embodiment;

[0065] Figure 5 A schematic diagram of a federated learning process with multiple participants in one embodiment;

[0066] Figure 6 is a flow chart of a biometric authentication method in another embodiment;

[0067] Figure 7A diagram showing an application environment of a palmprint feature authentication process in an embodiment;

[0068] Figure 8 A schematic diagram of a training process of a palmprint recognition model in one embodiment;

[0069] Fig. 9 A schematic diagram of the structure of a biometric authentication system in one embodiment;

[0070] Fig.10 is a structural block diagram of a biometric authentication device in one embodiment;

[0071] Fig.11 is an internal structure diagram of a computer device in one embodiment;

[0072] Fig.12 FIG. 4 is a diagram showing the internal structure of a computer device in another embodiment. DETAILED DESCRIPTION

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

[0074] The biometric authentication method provided in the embodiment of the present application can be applied to Figure 1 In the application environment shown, the application environment includes a server 102, a trusted organization 104, and multiple application terminals 106. Among them, the application terminal 106 and the server 102, and the application terminal 106 and the trusted organization 104 can communicate through a network. The network can be a wired network or a wireless network. Therefore, the application terminal 106 and the server 102, and the application terminal 106 and the trusted organization 104 can be directly or indirectly connected through wired or wireless communication. For example, the application terminal 106 can be indirectly connected to the server 102 through a wireless access point, or the application terminal 106 can be directly connected to the server 102 through the Internet, and this application does not limit this.

[0075] Among them, the application end 106 can be, but is not limited to, various desktop computers, laptops, smart phones, tablet computers, Internet of Things devices and portable wearable devices. The Internet of Things devices can be smart speakers, smart TVs, smart air conditioners, smart car-mounted devices, etc. Portable wearable devices can be smart watches, smart bracelets, head-mounted devices, etc. The service end 102 can be an independent physical server, or a server cluster or distributed system composed of multiple physical servers, or a cloud server that provides basic cloud computing services such as cloud services, cloud databases, cloud computing, cloud functions, cloud storage, network services, cloud communications, middleware services, domain name services, security services, CDN (Content Delivery Network), and big data and artificial intelligence platforms. The data storage system can store data that the service end 102 needs to process. The data storage system can be set up separately, integrated on the service end 102, or placed on the cloud or other servers. The trusted institution 104 can be a computer device used by a social institution with strong credibility. The trusted institution 104 can be used to store the registered biometric information provided by each application end 106. The computer device can specifically be a terminal or a server.

[0076] The embodiment of the present application can be applied to various authentication scenarios based on biometric features, such as access control, payment, and identity verification. A client related to biometric information collection can be installed on the application end 106, and the client can be software (such as a browser, video software, etc.), or a web page, a small program, etc. The server end 102 is a background server corresponding to the software, web page, small program, etc., or a server dedicated to biometric authentication. During the data distribution process, the server 102: sends a feature recognition model to a trusted institution and multiple application terminals; receives multiple registered biometric features sent by the trusted institution, and builds a biometric library based on the multiple registered biometric features; the feature recognition model is used at the application terminal to extract features from the collected biometric information to be authenticated, and obtain the biometric feature to be authenticated; the feature recognition model is used at the trusted institution to extract features from the registered biometric information provided by each application terminal, and obtain multiple registered biometric features; determines the target application terminal that initiates the biometric authentication event among the multiple application terminals, and receives the biometric feature to be authenticated sent by the target application terminal; performs feature matching on the biometric feature to be authenticated and each registered biometric feature in the biometric library, determines the authentication result of the biometric feature to be authenticated and feeds it back to the target application terminal.

[0077] In one embodiment, Figure 2 As shown, a biometric authentication method is provided, which is applied to Figure 1 The server 102 in the example is used as an example to illustrate, including the following steps:

[0078] Step S202: Send the feature recognition model to a trusted organization and multiple application terminals.

[0079] Among them, the feature recognition model is a machine learning model used to extract biometric features from biometric information and obtain the biometric features corresponding to the biometric information. The specific type of the feature recognition model is not unique, for example, it can be a convolutional neural network model (CNN), a recurrent neural network (RNN) model or a generative adversarial network (GAN) model, etc. The biometric authentication method involved in this application refers to a method for realizing the identity recognition of the authentication object based on the biometric features of the authentication object. The authentication object can be, for example, a natural person in a real scene, or a virtual character in a virtual scene, which is not limited here. Exemplarily, in a game scene, corresponding character images can be designed for different virtual characters, and the character image information is used as the biometric information to be authenticated to realize the biometric authentication of the virtual character. For ease of understanding, the following is an explanation of the case where the authentication object is a natural person.

[0080] Biometric information may include at least one of static information or behavioral information. The static information may include, for example, information such as fingerprints, palm prints, palm vein images, facial images, iris images, etc.; the behavioral information may include, for example, information such as gestures, handwriting, voice, gait, etc. Biometric features are used to characterize the characteristics of biometric information. The specific form of the biometric feature may be, for example, a matrix or a vector. Corresponding to the biometric information, the biometric feature may include, for example, at least one of static features and behavioral features. The static features may include, for example, fingerprint features, palm print features, palm vein features, facial features, iris features, etc.; the behavioral features may include, for example, gesture features, handwriting features, voice print features, gait features, etc. Taking palm print features as an example, the palm print features may include the features of at least one of the lines in the palm, back of the hand, etc. The palm print features may be represented by vectors, matrices, or other forms. The palm print features may include features of multiple dimensions such as point features, texture features, line features, and geometric features. Among them, point features refer to the skin surface features on the palm, such as the singular points and patterns formed locally on the palm papillary lines; texture features refer to the features of relatively shallow and fine textures on the palm surface; lines refer to the features of relatively deep lines on the palm surface; geometric features may include features such as depth, width, length and geometric shape.

[0081] In actual applications, for a certain service, the user needs to first obtain an identity through registration, and after completing the registration, obtain the qualification to enjoy the service through authentication. The registered biometric information refers to the biometric information submitted by the user during the registration process, and the biometric information to be authenticated refers to the biometric information submitted by the user during the authentication process. Furthermore, the feature recognition model is used at the application end to extract features from the collected biometric information to be authenticated to obtain the biometric features to be authenticated; the feature recognition model is used at the trusted institution to extract features from the registered biometric information provided by each application end to obtain multiple registered biometric features. The application end refers to the end with the biometric information collection function, usually the terminal held by the user. The trusted institution refers to the end where a third-party institution with strong credibility is located, and the risk of leakage of biometric information stored in the trusted institution is relatively low.

[0082] Specifically, the server can obtain a feature recognition model through deep learning training, and send the feature recognition model to the trusted institution and multiple application terminals respectively. The registered object can send the registered biometric information to the trusted institution through the application terminal, and the trusted institution uses the feature recognition model to extract features from each registered biometric information to obtain the registered biometric features. During the authentication application process, the application terminal can collect the biometric information to be authenticated of the authentication object, and use the feature recognition model to extract features from the biometric information to be authenticated to obtain the biometric features to be authenticated.

[0083] It can be understood that in order to ensure the accuracy of the registered biometrics, the application end can collect multiple sub-registered biometric information for the same registered object, and the collection parameters of each sub-registered biometric information are different. Then, the trusted institution extracts features from each sub-registered biometric information to obtain the registered biometrics of the registered object. Exemplarily, the trusted institution can first extract features from each sub-registered biometric information separately to obtain the respective sub-registered biometric features of each sub-registered biometric information, and finally fuse each sub-registered biometric feature to obtain the registered biometric feature of the registered object; or it can first fuse each sub-registered biometric information to obtain the registered biometric information of the registered object, and then extract features from the registered biometric information to obtain the registered biometric feature of the registered object.

[0084] Taking palm print images as an example, different palm print images may correspond to different display modes, which may include display position or display angle. The display position indicates the position of the palm print image showing the palm, such as the upper left corner area of ​​the palm, the center area of ​​the palm, etc., and the display angle indicates the angle between the palm plane and the shooting plane. Different display modes will result in different presented palm print images, and the palm print features that can be extracted will also be different. Based on this, the application end can collect the palm print image data of the registered object from multiple angles, and the trusted institution can obtain the palm print image data, extract partial palm print images from the multiple palm print images contained in the palm print image data, and then obtain the complete palm print image of the registered object by performing information fusion on the partial palm print images, and then perform feature extraction on the complete palm print image to obtain the palm print features of the registered object.

[0085] Step S204: receiving a plurality of registered biometric features sent by a trusted institution, and constructing a biometric feature database based on the plurality of registered biometric features.

[0086] Among them, the biometric database is a database including multiple registered biometrics. It can be understood that the biometric database can be associated with the service corresponding to the registered biometric. In other words, the server can provide multiple services, such as payment services and access control services, and each service can share a biometric database or use a different biometric database. When multiple services share a biometric database, the registered biometric information used by the same object in each service is the same; correspondingly, when each service uses a different biometric database, the registered biometric information used by the same object in each service is different. Exemplarily, the access control service may use a fingerprint feature database, and the payment service may use a palm print feature database.

[0087] Furthermore, the trusted institution may obtain multiple registered biometric information provided by multiple application terminals for different services, and use a feature recognition model that matches the registered biometric information to extract features from each registered biometric information, obtain multiple registered biometric features for multiple services, and send each registered biometric feature to the server that provides the corresponding service. For each server, the server receives the registered biometric features sent by the trusted institution, and builds a biometric library that matches the provided service based on each registered biometric feature. It should be noted that the application terminal may also provide the registered biometric information to the trusted institution after the server builds the biometric library, and the trusted institution extracts features from the registered biometric information to obtain the registered biometric features and sends them to the server, which receives the registered biometric features and updates the constructed biometric library based on the registered biometric features. Among them, the update method of the constructed biometric library may include, for example, adding a registered biometric feature, replacing a registered biometric feature, etc.

[0088] For example, taking the palmprint authentication scenario as an example, Figure 3 As shown, in the process of biometric information registration for the target service, the server providing the target service can send a palmprint recognition model to a trusted institution, and the object that expects to use the target service can send a registered palmprint image to the trusted institution through the application end. The trusted institution uses the palmprint recognition model to extract features from the registered palmprint image, obtains the registered palmprint features corresponding to the registered palmprint image, and sends the registered palmprint features to the server. The server receives the registered palmprint features and builds or updates a biometric feature library associated with the target service based on the registered palmprint features.

[0089] Step S206, determining a target application end that initiates a biometric authentication event among multiple application ends, and receiving the biometric to be authenticated sent by the target application end.

[0090] Among them, the biometric authentication event refers to an event based on biometric authentication, and the application end that initiates the biometric authentication event is called the target application end. The specific way to trigger a biometric authentication event is not unique. For example, the target application end can trigger a biometric authentication event by sending the biometric to be authenticated to the server end; it can also trigger a biometric authentication event by sending an authentication start instruction to the server end. Specifically, when the target application end initiates a biometric authentication event, it sends the biometric to be authenticated to the server end, and the server end responds to the biometric authentication event and receives the biometric to be authenticated sent by the target application end.

[0091] It can be understood that in actual applications, since the server provides services to multiple application terminals, the number of target application terminals that initiate biometric authentication events to the server can be one or more. In the case of multiple target application terminals, the server can store each biometric feature to be authenticated in a queue to be authenticated according to the reception time of each biometric feature to be authenticated, and then obtain the corresponding biometric feature to be authenticated from the queue to be authenticated in a first-in-first-out order, and perform corresponding authentication processing.

[0092] It should be noted that the same application end can be used by multiple objects, that is, for a certain application end, multiple registered biometric information can be provided to a trusted organization; the application end can also send multiple biometric features to be authenticated to the server end. For the same object, the application end that provides the registered biometric information and the application end that initiates the biometric authentication event can be the same or different, and it is not necessary for each application end to provide the registered biometric information to the trusted organization. Exemplarily, object A can provide the registered biometric information to the trusted organization through application end A1, and initiate a biometric authentication event to the server end through application end A2; object A can also provide the registered biometric information to the trusted organization through application end A1, and initiate a biometric authentication event to the server end. In addition, there may be a situation where multiple objects use the same application end.

[0093] Step S208, feature matching is performed on the biometric feature to be authenticated and each registered biometric feature in the biometric feature database, and an authentication result of the biometric feature to be authenticated is determined and fed back to the target application end.

[0094] The authentication result may include authentication passed or authentication failed. In the case of authentication passed, the authentication result may also include associated information of the biometric feature to be authenticated, and the associated information may include, for example, identity identification, validity period, account level, etc.

[0095] Specifically, the server can perform feature matching on the biometric feature to be authenticated and each registered biometric feature in the biometric database, determine the matching degree between the biometric feature to be authenticated and each registered biometric feature, and determine the authentication result of the biometric feature to be authenticated based on each matching degree. Finally, the authentication result is fed back to the target application. Exemplarily, the server can determine that the authentication result of the biometric feature to be authenticated is authentication passed when there is a registered biometric feature in the biometric database whose similarity with the biometric feature to be authenticated meets the similarity condition. The similarity condition can be, for example, that the similarity is greater than a similarity threshold, or that the similarity is greater than or equal to a similarity threshold.

[0096] It should be noted that during the interaction between the trusted institution, the server and the application, information encryption can be used to further improve security. For example, the server can encrypt the feature recognition model and send it to the trusted institution and multiple application terminals, and the trusted institution and each application terminal can decrypt the encrypted model to obtain the feature recognition model; for another example, the trusted institution can send the encrypted registered biometrics to the server, and the server can obtain the registered biometrics used to build the biometric database through decryption.

[0097] In the above-mentioned biometric authentication method, a feature recognition model is issued to a trusted institution and multiple application terminals; multiple registered biometrics sent by the trusted institution are received, and a biometric library is constructed based on the multiple registered biometrics; a target application terminal that initiates a biometric authentication event among multiple application terminals is determined, and the biometric to be authenticated sent by the target application terminal is received; the biometric to be authenticated is matched with each registered biometric in the biometric library, and the authentication result of the biometric to be authenticated is determined and fed back to the target application terminal. The feature recognition model is used at the application terminal to extract features from the collected biometric information to be authenticated to obtain the biometric to be authenticated; the feature recognition model is used at the trusted institution to extract features from the registered biometric information provided by each application terminal to obtain multiple registered biometrics. That is to say, in the above-mentioned biometric authentication process, for the registered biometric information, the information collection terminal only includes the trusted institution and the application terminal that provides the registered biometric information; for the biometric information to be authenticated, the information collection terminal only includes the application terminal that collects the biometric information to be authenticated. This means that biometric authentication can be achieved without transmitting biometric information during the interaction between the trusted organization and the server, and between the application and the server. This can avoid the leakage of biometric information during the authentication interaction to a certain extent, and is conducive to improving security.

[0098] In one embodiment, the biometric authentication method further includes: sending a security credential to a trusted institution so that the trusted institution performs security verification on the feature recognition model based on the security credential. In the case of this embodiment, receiving multiple registered biometrics sent by the trusted institution includes: receiving multiple registered biometrics sent by the trusted institution when the feature recognition model passes the verification.

[0099] Among them, the security credential refers to the credential used for security verification. Specifically, the server sends the security credential to the trusted institution so that the trusted institution can perform security verification on the feature recognition model based on the security credential, and when the verification is passed, the feature recognition model is used to extract features from each registered biometric information, obtain multiple registered biometric features and send them to the server. The server then receives these registered biometric features and constructs a biometric library. Furthermore, specific methods for security verification of the feature recognition model may include verifying the identity and authority of the model sender (i.e., the server), performing security analysis on the model itself, and so on. Specific methods for security analysis of the model itself may include, for example, fault tree analysis, functional risk analysis, and anti-attack analysis based on adversarial samples, and so on.

[0100] It can be understood that if the security verification fails, it means that there is a security risk in the feature recognition model or the server itself. The trusted institution can send feedback information of the verification failure to the server. The server can obtain the feedback information and make corrections based on the feedback information before resending the security credentials and feature recognition model to the trusted institution. Exemplarily, the feedback information can include the specific reason for the verification failure, which can be, for example, that there is a security risk in the feature recognition model. The server can make targeted adjustments to the feature recognition model and then resend the security credentials and the adjusted feature recognition model to the trusted institution.

[0101] In this embodiment, security verification is performed based on security credentials, and the registered biometric information is transmitted after the verification is passed. This can ensure the security of the feature recognition model, avoid the leakage of biometric information during the use of the model, and help to further improve security.

[0102] In one embodiment, step S208 includes: performing feature matching on the biometric to be authenticated and each registered biometric in the biometric database, respectively, to determine the feature similarity between the biometric to be authenticated and each registered biometric; in the case where there is a target registered biometric whose feature similarity meets the authentication condition, determining that the authentication result of the biometric to be authenticated is authentication passed; and feeding back the identity identifier corresponding to the target registered biometric to the target application end.

[0103] Among them, the identity identifier is the identification information corresponding to the registered biometric feature. The identity identifier may include at least one of numbers, letters or text. Specifically, when the server receives the biometric feature to be authenticated, it performs feature matching on the authentication biometric feature and each registered biometric feature in the biometric feature library, and determines the feature similarity between the biometric feature to be authenticated and each registered biometric feature. The feature similarity can be characterized by cosine similarity, Euclidean distance, Manhattan distance or Mahalanobis distance, for example. Correspondingly, the specific algorithm for feature matching can be an algorithm for determining indicators such as cosine similarity, Euclidean distance, Manhattan distance or Mahalanobis distance. After obtaining the feature similarity corresponding to each registered biometric feature, the server determines whether there is a target registered biometric feature that meets the authentication conditions among the registered biometric features, and if there is a target registered biometric feature, determines that the authentication result of the biometric feature to be authenticated is authentication passed, and feeds back the identity identifier corresponding to the target registered biometric feature to the target application end.

[0104] It can be understood that the target biometric feature has the largest feature similarity with the biometric feature to be authenticated. The feature similarity satisfies the authentication condition, which may mean that the maximum feature similarity is greater than the similarity threshold, or the maximum feature similarity is greater than or equal to the similarity threshold.

[0105] Take the palmprint authentication scenario as an example. Figure 3 As shown, the application collects the palmprint image of the user, stores the palmprint image in a local data set, and uses a local model to extract features from the palmprint image to obtain the palmprint features to be authenticated. The local model refers to the palmprint recognition model issued by the server. The application then sends the palmprint features to be authenticated to the server, so that the server retrieves the target registered palmprint features that match the palmprint features to be authenticated in the palmprint feature library, and returns the user ID (identification) corresponding to the target registered palmprint features. The application obtains the user ID, which is equivalent to completing a palmprint recognition and realizing authentication based on palmprint features.

[0106] In this embodiment, the authentication result is determined based on feature similarity, and the identity identifier corresponding to the target registered biometric feature that meets the authentication conditions is fed back to the target application end. The algorithm is simple and can improve the authentication efficiency while ensuring security.

[0107] In one embodiment, the process of training a feature recognition model includes: sending an initial recognition model for biometric feature extraction to multiple application terminals; obtaining multiple parameter update information obtained by incrementally training the initial recognition model from at least a portion of the multiple application terminals; and updating the initial recognition model based on each parameter update information to obtain a feature recognition model.

[0108] Among them, the training samples of incremental training include biological information collected during the biometric authentication process. The biological information may specifically include registered biological information and biological information to be authenticated. The parameter update information may, for example, be parameter update information of a certain model parameter, or may include parameter update information of each of multiple model parameters. Model parameters may, for example, include learning rate, number of hidden layers of a deep neural network, number of clusters, etc. The parameter update information may specifically be a parameter offset or an updated parameter value, etc., which is not limited here.

[0109] Specifically, the federated learning method can be applied to the training process of the feature recognition model. Federated learning is a decentralized machine learning method that allows training on multiple devices or data centers without centralizing the data set in one place. In federated learning, each participant trains the model locally, and then aggregates the model update information on the central server to form an updated global model. The central server then returns the global model to each participant for the next round of federated learning. In other words, using the federated learning method, multiple participants can collaborate to train the model without sharing data. Each participant has its own local data set, which is usually privacy-protected and will not be shared among the participants. In this way, data transmission and storage requirements can be reduced to achieve the effect of privacy protection. In addition, model training with the participation of multiple participants can improve the efficiency of model training.

[0110] In a specific embodiment, Figure 5 As shown in the figure, in the process of federated learning, the central server can first use the open source data set to train the network model or directly use the trained network model to obtain the global network model and send it to each participant. Each participant collects biological information as the data set for local network model training. Each participant can use the local data set for local training after a period of time. The local training uses back propagation to update the local model parameters. This process will obtain the slope of each parameter for the loss function, so that the new parameters can obtain a smaller loss function using the gradient descent method. After local training, the parameter update information of the model parameters will be obtained, and the parameter update information will be encrypted and transmitted to the central server for the update and iteration of the global network model. The central server will perform global iteration on the results of local training of each participant, which can be a weighted sum of the model offsets obtained by local training of each participant, and then added to the global network model to complete a global model update. After the global model update is completed, the new network model (i.e., the aggregated federated model) will be sent to each participant, allowing all participants to update the local network model to extract new data features, thereby completing a model update iteration.

[0111] The model before updating in each round of federated learning is called the initial recognition model, and the updated model is called the feature recognition model. A round of federated learning process refers to the entire process from the server sending the initial recognition model to updating the initial recognition model based on the parameter update information obtained from each application end to obtain the feature recognition model. In other words, the end condition of a round of federated learning is that the server sends the updated feature recognition model to each application end. It can be understood that the initial recognition model includes the initial feature recognition model determined by the server when each application end does not participate in the training, and the feature recognition model determined based on the previous round of federated learning when each application end participates. Then, in each round of federated learning, the server can send the initial recognition model for biometric feature extraction to multiple application ends. Then, each application end performs incremental training on the initial recognition model based on the biometric information collected by itself to obtain the corresponding parameter update information. Next, the server obtains multiple parameter update information obtained by incrementally training the initial recognition model by at least a part of the multiple application ends, and then updates the initial recognition model based on each parameter update information to obtain the feature recognition model.

[0112] Furthermore, the specific timing for the application to perform incremental training on the initial recognition model is not unique. For example, incremental training can be performed regularly at set time intervals, or incremental training can be performed when the amount of biological information collected after the last training reaches a threshold. In actual applications, since the amount of biological information collected by each application is different, there may be a situation at a certain time point where application A performs multiple incremental training on the initial recognition model, but application B does not perform incremental training on the initial recognition model. In this case, the server can determine the parameter update information corresponding to application A based on the sub-parameter update information of each incremental training provided by application A.

[0113] Exemplarily, if the training samples used in each incremental training completed by application end A in this round of federated learning include the training samples used in the previous incremental training, the server end may determine the sub-parameter update information obtained from application end A for the last time as the parameter update information provided by application end A in this round; if the training samples used in each incremental training completed by application end A in this round of federated learning do not include the training samples used in the previous incremental training, the server end may perform statistical calculations on each sub-parameter update information to obtain the parameter update information provided by application end A in this round. The specific algorithm of statistical calculation may include, for example, addition, multiplication, etc.

[0114] The specific timing for the server to update the initial recognition model based on each parameter update information is also not unique. Exemplarily, the server can update the model when the number of application terminals that have completed incremental training reaches the application terminal number threshold, and specifically can determine that the application terminal has completed incremental training when receiving parameter update information provided by the application terminal; the server can also update the model when the number of parameters of the model parameters targeted by the multiple parameter update information obtained reaches the parameter number threshold; the server can also determine the parameter offset of each model parameter based on each parameter update information, and update the model when the parameter offset of at least one model parameter reaches the offset threshold.

[0115] In the above embodiment, the federated learning method is applied to the training of the feature recognition model, which can improve the training efficiency while ensuring the security of biometric information. In addition, what is transmitted between the application end and the server end is parameter update information rather than model update, which can save information transmission costs.

[0116] In one embodiment, the biometric authentication method further includes: determining the type of biometric information to be identified, and obtaining an open source data set that matches the type of biometric information; and using the open source data set to perform model training on an initial neural network to obtain an initial recognition model for biometric extraction.

[0117] Among them, the initial recognition model in this embodiment refers to the feature recognition model that the server sends to the application for the first time. The biometric information type may include at least one of static information or behavioral information. The static information may include, for example, fingerprints, palm prints, palm vein images, facial images, iris images and other information; the behavioral information may include, for example, gestures, handwriting, voice, gait and other information. It can be understood that the training samples required for feature recognition models corresponding to different biometric information types are different. For example, the training samples required for training a palm print recognition model are palm print images, and the training samples required for training an iris feature recognition model are iris images. Based on this, the server can determine the type of biometric information that is expected to be recognized, obtain an open source data set that matches the biometric information type, and then use the open source data set to train the initial neural network model to obtain an initial recognition model for biometric feature extraction.

[0118] In this embodiment, an open source data set that matches the type of biometric information to be identified is used to train the initial neural network model to obtain an initial recognition model for biometric feature extraction, which can ensure the accuracy of the initial recognition model while improving training efficiency.

[0119] In one embodiment, the initial recognition model is updated based on each parameter update information to obtain a feature recognition model, including: determining the model parameters corresponding to each parameter update information; for each model parameter, performing statistical analysis on each parameter update information corresponding to the model parameter to determine an update parameter that matches the model parameter; based on each update parameter, the initial recognition model is updated to obtain a feature recognition model.

[0120] As mentioned above, the parameter update information obtained by the server from the application may include parameter update information corresponding to each of the multiple model parameters. Based on this, the server can determine the model parameters corresponding to each of the parameter update information, and then, for each model parameter, perform statistical analysis on the parameter update information corresponding to the model parameter to determine the update parameter that matches the model parameter. Finally, the server updates the initial recognition model based on the updated parameters to obtain a feature recognition model. Among them, the specific algorithm of statistical analysis may include, for example, summation, averaging, clustering, etc.

[0121] Exemplarily, the parameter update information obtained by the server from each application end includes parameter update information a1, a2, a3 and a4 of model parameter A, and parameter update information b1, b2 and b3 of model parameter B. The server can perform statistical analysis on a1 to a4 to determine the update parameter Aa that matches the model parameter A, perform statistical analysis on b1 to b3 to determine the update parameter Bb that matches the model parameter B, and update the initial recognition model based on the update parameters Aa and Bb to obtain a feature recognition model.

[0122] In this embodiment, for each model parameter, statistical analysis is performed on each parameter update information corresponding to the model parameter to determine a matching update parameter, thereby ensuring the accuracy of the feature recognition model obtained based on the update parameter.

[0123] In one embodiment, the update parameter is a parameter offset. In this embodiment, the initial recognition model is updated based on each update parameter to obtain a feature recognition model, including: for each model parameter, the initial parameter value of the model parameter is superimposed on the parameter offset of the model parameter to obtain an updated parameter value of the model parameter; the initial recognition model is updated based on the updated parameter value of each model parameter to obtain a feature recognition model.

[0124] The parameter offset refers to the offset based on the original parameter value. The data form of the offset can be a percentage or a specific value. Specifically, for each model parameter, the server adds the initial parameter value of the model parameter to the parameter offset of the model parameter to obtain the updated parameter value of the model parameter, and then updates the initial recognition model based on the updated parameter value of each model parameter to obtain a feature recognition model.

[0125] In this embodiment, the server adds the parameter offset on the basis of the initial parameter value to obtain the updated parameter value, so that during the federated learning process, only the offset information needs to be transmitted between the server and the application, which can reduce the amount of information transmitted and reduce the communication cost.

[0126] In one embodiment, statistical analysis is performed on each parameter update information corresponding to the model parameter to determine the update parameter that matches the model parameter, including: determining the weight of each parameter update information corresponding to the model parameter; and performing weighted summation processing on each parameter update information based on each weight to obtain the update parameter that matches the model parameter.

[0127] Specifically, for each model parameter, the server can perform weighted sum processing on each parameter update information corresponding to the model parameter to obtain an update parameter matching the model parameter. The specific method for determining the weight of each parameter update information is not unique.

[0128] In a specific embodiment, considering that the user groups facing the application end are different, the amount of biometric information that can be collected is also different. The service end can determine the number of user groups of each application end according to the type of user groups facing each application end, and then assign corresponding weights to the parameter update information provided by each application end. Among them, the user group type can include, for example, individuals, family members, community owners, shopping mall customers, etc. The weight is positively correlated with the number of user groups. The positive correlation of two variables means that when other conditions remain unchanged, the two variables change in the same direction, and when one variable changes from large to small, the other variable also changes from large to small. It can be understood that the positive correlation here means that the direction of change is consistent, but it does not require that when one variable changes a little, the other variable must also change. For example, when variable a is 10 to 20, variable b can be set to 100, and when variable a is 20 to 30, variable b is 120. In this way, the direction of change of a and b is that when a becomes larger, b also becomes larger. But when a is in the range of 10 to 20, b can be unchanged.

[0129] In a specific embodiment, determining the weights of each parameter update information corresponding to the model parameter includes: for each parameter update information corresponding to the model parameter, determining the number of training samples used in the process of training to obtain the parameter update information; based on the number of training samples corresponding to each parameter update information, determining the weights of each parameter update information.

[0130] Among them, the weight is positively correlated with the number of training samples. For the specific definition of positive correlation, please refer to the above and will not be repeated here. Specifically, the application end can send the number of training samples used to train the parameter update information while sending the parameter update information to the server end. The server can determine the number of training samples used in the process of training to obtain the parameter update information for each parameter update information corresponding to the model parameter, and then determine the weight of each parameter update information based on the number of training samples corresponding to each parameter update information. It should be noted that for an application end that completes multiple incremental trainings in a round of federated learning, the number of training samples of the application end refers to the total number of training samples used in each incremental training process.

[0131] In this embodiment, the weight of each parameter update information is determined based on the number of training samples corresponding to each parameter update information, which can ensure that the parameter update information with a larger weight has a larger number of training samples, thereby ensuring the accuracy of the update parameters obtained by weighted summation processing.

[0132] In one embodiment, Figure 6 As shown, a biometric authentication method is provided, which can be executed by a computer device, which can be Figure 1In this embodiment, the method includes the following steps:

[0133] Step S601, determining the type of biological information to be identified, and acquiring an open source data set matching the type of biological information;

[0134] Step S602, using an open source data set to perform model training on an initial neural network to obtain an initial recognition model for biometric feature extraction;

[0135] Step S603, sending an initial recognition model for extracting biometric features to multiple application terminals;

[0136] Step S604, obtaining multiple parameter update information obtained by performing incremental training on the initial recognition model by at least a part of the multiple application terminals;

[0137] Among them, the training samples of incremental training include biological information collected during the biometric authentication process;

[0138] Step S605, determining the model parameters corresponding to each parameter update information;

[0139] Step S606, for each model parameter, determining the number of training samples used in the process of training to obtain update information of each parameter;

[0140] Step S607, determining the weight of each parameter update information based on the number of training samples corresponding to each parameter update information;

[0141] Among them, the weight is positively correlated with the number of training samples;

[0142] Step S608, performing weighted sum processing on each parameter update information based on each weight to obtain a parameter offset that matches the model parameter;

[0143] Step S609, for each model parameter, superimpose the initial parameter value of the model parameter with the parameter offset of the model parameter to obtain an updated parameter value of the model parameter;

[0144] Step S610, updating the initial recognition model based on the updated parameter values ​​of each model parameter to obtain a feature recognition model.

[0145] Step S611, sending the feature recognition model to the trusted organization and multiple application terminals respectively;

[0146] Among them, the feature recognition model is used at the application end to extract features from the collected biometric information to be authenticated, and obtain the biometric features to be authenticated; the feature recognition model is used at the trusted institution to extract features from the registered biometric information provided by each application end, and obtain multiple registered biometric features;

[0147] Step S612, sending the security credential to the trusted institution, so that the trusted institution performs security verification on the feature recognition model based on the security credential;

[0148] Step S613, when the feature recognition model passes the verification, receiving multiple registered biometric features sent by a trusted institution, and building a biometric feature library based on the multiple registered biometric features;

[0149] Step S614, determining a target application end that initiates a biometric authentication event among multiple application ends, and receiving the biometric to be authenticated sent by the target application end;

[0150] Step S615, feature matching is performed on the biometric feature to be authenticated and each registered biometric feature in the biometric feature database to determine the feature similarity between the biometric feature to be authenticated and each registered biometric feature;

[0151] Step S616, if there is a target registered biometric feature whose feature similarity satisfies the authentication condition, determining the authentication result of the biometric feature to be authenticated as authentication passed;

[0152] Step S617, feeding back the identity identifier corresponding to the target registered biometric feature to the target application end.

[0153] In the above-mentioned biometric authentication process, for registered biometric information, the information collection end only includes the trusted institution and the application end that provides the registered biometric information; for the biometric information to be authenticated, the information collection end only includes the application end that collects the biometric information to be authenticated. This is equivalent to realizing biometric authentication without transmitting biometric information during the interaction between the trusted institution and the server end, and between the application end and the server end, which can avoid the leakage of biometric information during the authentication interaction process to a certain extent, and is conducive to improving security.

[0154] In a possible embodiment, the biometric authentication method can be applied to the voiceprint authentication scenario. In this scenario, the application end has the function of collecting sound information. In the process of voiceprint authentication, the server end sends a voiceprint recognition model to a trusted institution and multiple application ends. The voiceprint recognition model is used at the application end to extract features from the collected sound information to be authenticated to obtain the biometric features to be authenticated; the feature recognition model is used at the trusted institution to extract features from the registered sound information provided by each application end to obtain multiple registered voiceprint features. Then, the server end receives multiple registered voiceprint features sent by the trusted institution, and builds a voiceprint feature library based on these multiple registered voiceprint features. In the application process, the server end determines the target application end that initiates the voiceprint authentication event among multiple application ends, receives the voiceprint features to be authenticated sent by the target application end, performs feature matching on the voiceprint features to be authenticated and each registered voiceprint feature in the voiceprint feature library, determines the authentication result of the voiceprint features to be authenticated and feeds it back to the target application end.

[0155] In a possible embodiment, the biometric authentication method can be applied to palm print authentication scenarios. Figure 7 As shown, in this scenario, the application end is a palm-swiping terminal device with a palm-swiping function, the service end is a central server of a service provider, and the trusted institution is a third-party trusted institution other than the palm-swiping terminal device and the central server. In the process of palmprint feature authentication, the central server sends a palmprint recognition model to multiple palm-swiping terminal devices and a third-party trusted institution. The palmprint recognition model is used in the palm-swiping terminal device to extract features from the collected palmprint image to be authenticated, and obtain the palmprint features to be authenticated; the palmprint recognition model is used in the third-party trusted institution to extract features from the registered palmprint images provided by each palm-swiping terminal device, and obtain multiple registered palmprint features. Then, the central server receives multiple registered palmprint features sent by the third-party trusted institution, and constructs a palmprint feature library based on the multiple registered palmprint features. In the application process, the central server determines the target terminal that initiates the palmprint feature authentication event among multiple palm-swiping terminal devices, receives the palmprint features to be authenticated sent by the target terminal; the palmprint features to be authenticated are matched with each registered palmprint feature in the palmprint feature library, and the authentication result of the palmprint features to be authenticated is determined and fed back to the target terminal.

[0156] Among them, the palmprint recognition model is a model that applies deep learning technology to palmprint recognition. Its main idea is to use the deep learning model to automatically extract features from the palmprint image, and then use these features to recognize and classify the palmprint. Deep learning is a machine learning method based on neural networks. It simulates the working mode of human brain neurons through multi-layer neural networks to achieve learning and prediction of large-scale data. It can extract high-level abstract features from complex data by automatically learning features, and has been widely used in various fields, such as image recognition, speech recognition, natural language processing, etc.

[0157] like Figure 7 As shown, the palm swiping terminal device transmits the palm print features to be authenticated to the central server, the palm swiping terminal device transmits the registered palm print image to the third-party trusted institution, and the third-party trusted institution transmits the registered palm print features to the central server. The above method can be used for palm print recognition to protect the user's palm print information from unauthorized access, use, sharing or disclosure, realize privacy protection, and help improve the security of the authentication process.

[0158] Furthermore, the biometric authentication method provided in the present application can be specifically applied to scenarios such as access control, payment, and identity verification. Taking the case where the biometric feature is a palm print feature as an example, specifically, the application of palm swiping recognition in the access control scenario refers to the use of palm print recognition technology to perform identity authentication and access control authorization. The application process is as follows: the user places his palm on the palm swiping terminal device, the device collects the user's palm print picture, and extracts the palm print features of the palm print picture; the central server compares the user's palm print features with the existing features in the database to confirm whether the user's identity is legal; if the user's identity authentication is passed, the central server will turn on the access control device, and the user can enter the designated area; otherwise, the access control device will remain closed and the user cannot enter.

[0159] It can be understood that in federated learning that only updates the model, there is no need for the participation of a third-party trusted institution. However, in the palmprint feature authentication process, the service provider (i.e., the central server) needs to use the palmprint features to be authenticated to search the feature library. Therefore, when the palmprint feature extraction network is upgraded, the feature library stored in the central server also needs to be upgraded. Therefore, it is necessary to store the user's palm image in a third-party trusted institution. Whenever the palmprint recognition model is globally updated, the service provider needs to obtain the features extracted by the new model from the third-party institution that stores the palm image, thereby updating the feature library for palm image recognition. The third-party trusted institution needs to verify the relevant credentials of the service provider and determine the security of the feature extraction model to prevent the leakage of user palm image data. The third-party trusted institution needs to have the ability to ensure data security and the ability to review the security of other institutions.

[0160] In one embodiment, Figure 8As shown in the figure, a palmprint recognition model training method based on federated learning is provided. In the training process, the central server first initializes a global model using an open source data set or existing parameters, and ensures that all features in the feature library are extracted by the current global model. The global model is then sent to each terminal device, which needs to be deployed locally for palmprint feature extraction, and then starts palmprint recognition and provides palm scanning related services.

[0161] The palm recognition process is as follows: Figure 4 As shown, the terminal device needs to use the local model to extract features from the palm print, and then perform global feature retrieval on the central server. When the target feature matching the extracted palm print feature is found, the central server will return the user ID associated with the feature. The terminal device obtains the user ID, which means that the recognition is successful. In the process of providing palm swiping services, the terminal will cache the palm prints of users that have been identified locally. When the local data reaches a certain amount, the terminal device will start local model training. Since the local data set is small, the local training process does not require too many computing resources. After a round of local training, each device will obtain the model parameter offset of this round of training, and upload the offset to the central server after asymmetric encryption.

[0162] The central server needs to obtain the parameter offset uploaded by each device participating in local training and decrypt it, aggregate all offsets using a weighted sum method, and add the global parameter offset to the global model to complete the model update. Then the central server needs to use the updated model to upgrade the palm image feature in the feature library. During this process, the central server needs to transmit the updated model to a third-party trusted organization. The third-party trusted organization needs to verify the security of the model to ensure that the user's privacy data will not be leaked by the model. After the new model passes the security test, the third-party trusted organization allows the model to extract features from the user's palm image, and encrypts the palm image features extracted by the new model and sends them back to the central server. The central server will use these features to update the global feature library. After ensuring that all features in the feature library have been upgraded, the central server will send the updated global model to each terminal device, and the terminal device will redeploy the model to start a new round of palm image recognition model training.

[0163] Using the above method, federated learning can perform model training without exposing personal data, thus protecting user privacy. Compared with centralized training, federated learning can disperse the computing load and improve the training efficiency of network models. Federated learning can use data from multiple parties for training, thereby improving the performance and generalization ability of the model. Federated learning can perform calculations locally and only needs to transmit model parameters back to the central server, thereby reducing communication costs.

[0164] Based on the same inventive concept, the embodiment of the present application also provides a biometric authentication system for implementing the above-mentioned biometric authentication method. The implementation scheme for solving the problem provided by the system is similar to the implementation scheme recorded in the above-mentioned method, so the specific limitations in one or more biometric authentication system embodiments provided below can refer to the above-mentioned limitations on the biometric authentication method, and will not be repeated here.

[0165] In one embodiment, the present application also provides a biometric authentication system, such as Fig. 9 As shown, the system includes a server, a trusted institution, and multiple application terminals. The server sends a feature recognition model to the trusted institution and multiple application terminals; the feature recognition model is used at the application terminal to extract features from the collected biometric information to be authenticated, and obtain the biometric features to be authenticated; the trusted institution uses the feature recognition model to extract features from the registered biometric information provided by each application terminal, obtains multiple registered biometric features and sends them to the server, so that the server builds a biometric library based on the multiple registered biometric features; the server is also used to determine the target application terminal that initiates the biometric authentication event among the multiple application terminals, receive the biometric features to be authenticated sent by the target application terminal, perform feature matching on the biometric features to be authenticated and each registered biometric feature in the biometric library, determine the authentication result of the biometric features to be authenticated, and feed it back to the target application terminal.

[0166] Among them, the feature recognition model is a machine learning model used to extract biometric features from biometric information and obtain the biometric features corresponding to the biometric information. The biometric authentication method involved in this application refers to a method for realizing identity recognition of the authentication object based on the biometric features of the authentication object. The service end refers to the end that provides related services after the authentication is passed; the application end refers to the end with the function of collecting biometric information, usually the terminal held by the user; the trusted institution refers to the end where a third-party institution with strong credibility is located, and the risk of leakage of biometric information stored in the trusted institution is relatively low.

[0167] Specifically, the server can obtain a feature recognition model through deep learning training, and send the feature recognition model to the trusted institution and multiple application terminals respectively. The registered object can send the registered biometric information to the trusted institution through the application terminal, and the trusted institution uses the feature recognition model to extract features from each registered biometric information to obtain the registered biometric feature. During the authentication application process, the application terminal can collect the biometric information to be authenticated of the authentication object, and use the feature recognition model to extract features from the biometric information to be authenticated to obtain the biometric feature to be authenticated. The server receives multiple registered biometric features sent by the trusted institution, and builds a biometric library based on the multiple registered biometric features. During the application process, the server determines the target application terminal that initiates the biometric authentication event among multiple application terminals, receives the biometric feature to be authenticated sent by the target application terminal, performs feature matching on the biometric feature to be authenticated and each registered biometric feature in the biometric library, determines the authentication result of the biometric feature to be authenticated and feeds it back to the target application terminal.

[0168] The above-mentioned biometric authentication system configures a trusted institution to extract the registered biometric information to obtain the registered biometrics, and configures the server to send the feature recognition model to the trusted institution and the application end, so that during the biometric authentication process, for the registered biometric information, the information collection end only includes the trusted institution and the application end that provides the registered biometric information; for the biometric information to be authenticated, the information collection end only includes the application end that collects the biometric information to be authenticated. This is equivalent to realizing biometric authentication without transmitting biometric information during the interaction between the trusted institution and the server end, and between the application end and the server end, which can avoid the leakage of biometric information during the authentication interaction process to a certain extent, and is conducive to improving security.

[0169] In one embodiment, the trusted institution is specifically used to encrypt the obtained multiple registered biometric features to obtain encrypted biometric features and send them to the server; the server is specifically used to decrypt the encrypted biometric features to obtain multiple registered biometric features.

[0170] The specific algorithm for encryption processing may be, for example, a symmetric encryption algorithm or an asymmetric encryption algorithm, which is not limited here. Specifically, the trusted institution may determine the encryption key and the decryption key through negotiation during the interaction with the server. After the trusted institution uses the encryption key to encrypt multiple registered biometric features extracted using the feature recognition model, the encrypted biometric features are obtained and sent to the server. The server uses the decryption key to decrypt the received encrypted biometric features to obtain multiple registered biometric features.

[0171] In this embodiment, encrypted transmission of registered biometrics can prevent the registered biometrics from being tampered with, which is beneficial to further improve security.

[0172] In one embodiment, the trusted institution is also used to: respond to a registration biometric information maintenance event initiated by at least one application end among multiple application ends, determine the maintenance type and the selected registration biometric information that needs to be maintained; maintain the selected registration biometric information and the selected registration biometric features of the selected registration biometric information according to the maintenance type, and feed back the maintenance results to the service end and the at least one application end.

[0173] The maintenance type may be, for example, update or delete. The maintenance result may include maintenance success, maintenance failure, and updated registered biometrics generated after maintenance. It is understandable that the maintenance results fed back to the server and the application may be different. For example, the maintenance result fed back to the application that initiated the registered biometric information maintenance event may include feedback information of maintenance success or maintenance failure; the maintenance result fed back to the server may include a deletion instruction for the selected registered biometric, or an update of the registered biometric, etc.

[0174] Specifically, a user may initiate a registered biometric information maintenance event to a trusted institution through an application. For example, a user may send new registered biometric information to a trusted institution through an application to trigger a registered biometric information maintenance event; or may trigger a registered biometric information maintenance event by sending a maintenance instruction to a trusted institution. A trusted institution may respond to a registered biometric information maintenance event initiated by at least one of a plurality of application terminals, determine a maintenance type, and selected registered biometric information to be maintained, and then, according to the maintenance type, maintain the selected registered biometric information and the selected registered biometric features of the selected registered biometric information, and feed back the maintenance results to the service terminal and the application terminal that initiated the registered biometric information maintenance event.

[0175] In this embodiment, the trusted organization is also used to maintain the registered biometric information and feed back the maintenance results to the server and the application that initiates the registered biometric information maintenance event, which can ensure that the server's biometric library can be updated in real time, which is conducive to improving the accuracy of the biometric authentication results.

[0176] In one embodiment, the server is also used to send an initial recognition model for biometric feature extraction to multiple application terminals; each application terminal is also used to use the biometric information collected during the biometric feature authentication process to perform incremental training on the initial recognition model to obtain corresponding parameter update information; the server obtains at least a portion of the parameter update information determined by each of the multiple application terminals, updates the initial recognition model based on the parameter update information, and obtains a feature recognition model.

[0177] Among them, the training samples of incremental training include biological information collected during the biometric authentication process. The biological information may specifically include registered biological information and biological information to be authenticated. The parameter update information may, for example, be parameter update information of a certain model parameter, or may include parameter update information of each of multiple model parameters. Model parameters may, for example, include learning rate, number of hidden layers of a deep neural network, number of clusters, etc. The parameter update information may specifically be a parameter offset or an updated parameter value, etc., which is not limited here.

[0178] Specifically, the federated learning method can be applied to the training process of the feature recognition model. Federated learning is a decentralized machine learning method that allows training on multiple devices or data centers without centralizing the data set in one place. In federated learning, each participant trains the model locally, and then aggregates the model update information on the central server to form an updated global model. The central server then returns the global model to each participant for the next round of federated learning. In other words, using the federated learning method, multiple participants can collaborate to train the model without sharing data. Each participant has its own local data set, which is usually privacy-protected and will not be shared among the participants. In this way, data transmission and storage requirements can be reduced to achieve the effect of privacy protection. In addition, model training with the participation of multiple participants can improve the efficiency of model training.

[0179] The model before updating in each round of federated learning is called the initial recognition model, and the updated model is called the feature recognition model. A round of federated learning process refers to the entire process from the server sending the initial recognition model to updating the initial recognition model based on the parameter update information obtained from each application end to obtain the feature recognition model. In other words, the end condition of a round of federated learning is that the server sends the updated feature recognition model to each application end. It can be understood that the initial recognition model includes the initial feature recognition model determined by the server when each application end does not participate in the training, and the feature recognition model determined based on the previous round of federated learning when each application end participates. Then, in each round of federated learning, the server can send the initial recognition model for biometric feature extraction to multiple application ends. Then, each application end performs incremental training on the initial recognition model based on the biometric information collected by itself to obtain the corresponding parameter update information. Next, the server obtains multiple parameter update information obtained by incrementally training the initial recognition model by at least a part of the multiple application ends, and then updates the initial recognition model based on each parameter update information to obtain the feature recognition model.

[0180] In the above embodiment, the federated learning method is applied to the training of the feature recognition model, which can improve the training efficiency while ensuring the security of biometric information. In addition, what is transmitted between the application end and the server end is parameter update information rather than model update, which can save information transmission costs.

[0181] In one embodiment, the application end is specifically used to encrypt the determined parameter update information, obtain the encrypted update information and send it to the server end; the server end is specifically used to decrypt the encrypted update information to obtain the parameter update information.

[0182] The specific algorithm for encryption processing may be, for example, a symmetric encryption algorithm or an asymmetric encryption algorithm, which is not limited here. Specifically, during the interaction between the application end and the server end, the encryption key and the decryption key may be determined through negotiation. After the application end uses the encryption key to encrypt the determined parameter update information, the encrypted update information is obtained and sent to the server end. The server end uses the decryption key to decrypt the received encrypted update information to obtain the parameter update information.

[0183] In this embodiment, the parameter update information is encrypted and transmitted, which can prevent the parameter update information from being tampered with, and is conducive to further improving security.

[0184] It should be understood that, although the various steps in the flowcharts involved in the above-mentioned embodiments are displayed in sequence according to the indication of the arrows, these steps are not necessarily executed in sequence according to the order indicated by the arrows. Unless there is a clear explanation in this article, the execution of these steps does not have a strict order restriction, and these steps can be executed in other orders. Moreover, at least a part of the steps in the flowcharts involved in the above-mentioned embodiments can include multiple steps or multiple stages, and these steps or stages are not necessarily executed at the same time, but can be executed at different times, and the execution order of these steps or stages is not necessarily to be carried out in sequence, but can be executed in turn or alternately with other steps or at least a part of the steps or stages in other steps.

[0185] Based on the same inventive concept, the embodiment of the present application also provides a biometric authentication device for implementing the above-mentioned biometric authentication method. The implementation scheme for solving the problem provided by the device is similar to the implementation scheme recorded in the above-mentioned method, so the specific limitations in one or more biometric authentication device embodiments provided below can refer to the above-mentioned limitations on the biometric authentication method, and will not be repeated here.

[0186] In one embodiment, Fig.10As shown, a biometric authentication device 1000 is provided, comprising: a model sending module 1002, a feature library building module 1004, a biometric feature acquisition module 1006 to be authenticated, and a feature matching module 1008, wherein:

[0187] The model sending module 1002 is used to send the feature recognition model to the trusted institution and multiple application terminals; the feature recognition model is used at the application terminal to extract features from the collected biometric information to be authenticated, and obtain the biometric features to be authenticated; the feature recognition model is used at the trusted institution to extract features from the registered biometric information provided by each application terminal, and obtain multiple registered biometric features;

[0188] A feature library building module 1004 is used to receive a plurality of registered biometric features sent by a trusted institution, and build a biometric library based on the plurality of registered biometric features;

[0189] The biometric feature acquisition module 1006 to be authenticated is used to determine the target application end that initiates the biometric feature authentication event among multiple application ends, and receive the biometric feature to be authenticated sent by the target application end;

[0190] The feature matching module 1008 is used to perform feature matching on the biometric feature to be authenticated and each registered biometric feature in the biometric feature database, determine the authentication result of the biometric feature to be authenticated and feed it back to the target application end.

[0191] In one embodiment, the biometric authentication device 1000 further includes a security credential sending module, which is used to send the security credential to a trusted institution so that the trusted institution can perform security verification on the feature recognition model based on the security credential. In this embodiment, the feature library construction module 1004 is specifically used to receive multiple registered biometric features sent by the trusted institution when the feature recognition model passes the verification.

[0192] In one of the embodiments, the feature matching module 1008 is specifically used to: perform feature matching on the biometric feature to be authenticated and each registered biometric feature in the biometric feature library, and determine the feature similarity between the biometric feature to be authenticated and each registered biometric feature; when there is a target registered biometric feature whose feature similarity meets the authentication conditions, determine that the authentication result of the biometric feature to be authenticated is authentication passed; and feed back the identity identifier corresponding to the target registered biometric feature to the target application end.

[0193] In one embodiment, the biometric authentication device 1000 also includes: an initial recognition model sending module, which is used to send the initial recognition model for biometric extraction to multiple application terminals; a parameter update information acquisition module, which is used to obtain multiple parameter update information obtained by incrementally training the initial recognition model for at least some of the multiple application terminals; the training samples of the incremental training include biological information collected during the biometric authentication process; and a model update module, which is used to update the initial recognition model based on each parameter update information to obtain a feature recognition model.

[0194] In one embodiment, the biometric authentication device 1000 also includes a model training module, which is used to: determine the type of biometric information to be identified and obtain an open source data set that matches the biometric information type; use the open source data set to perform model training on an initial neural network to obtain an initial recognition model for biometric extraction.

[0195] In one embodiment, the model update module includes: a model parameter determination submodule, which is used to determine the model parameters corresponding to each parameter update information; a statistical analysis submodule, which is used to perform statistical analysis on each parameter update information corresponding to the model parameter for each model parameter, and determine the update parameter that matches the model parameter; and a model update submodule, which is used to update the initial recognition model based on each update parameter to obtain a feature recognition model.

[0196] In one embodiment, the statistical analysis submodule includes: a weight determination unit, which is used to determine the weights of each parameter update information corresponding to the model parameters; and an update parameter determination unit, which is used to perform weighted summation processing on each parameter update information based on each weight to obtain an update parameter that matches the model parameters.

[0197] In one of the embodiments, the weight determination unit is specifically used to: determine, for each parameter update information corresponding to the model parameter, the number of training samples used in the process of training to obtain the parameter update information; determine the weight of each parameter update information based on the number of training samples corresponding to each parameter update information; the weight is positively correlated with the number of training samples.

[0198] In one of the embodiments, the update parameter is a parameter offset; in the case of this embodiment, the model update submodule is specifically used to: for each model parameter, superimpose the initial parameter value of the model parameter with the parameter offset of the model parameter to obtain an updated parameter value of the model parameter; based on the updated parameter values ​​of each model parameter, the initial recognition model is updated to obtain a feature recognition model.

[0199] Each module in the above-mentioned biometric authentication device can be implemented in whole or in part by software, hardware or a combination thereof. Each of the above-mentioned modules can be embedded in or independent of a processor in a computer device in the form of hardware, or can be stored in a memory in a computer device in the form of software, so that the processor can call and execute the operations corresponding to each of the above modules.

[0200] In one embodiment, a computer device is provided. The computer device may be a server, and its internal structure diagram may be as follows: Fig.11 As shown. The computer device includes a processor, a memory, an input / output interface (Input / Output, referred to as I / O) and a communication interface. The processor, the memory and the input / output interface are connected through a system bus, and the communication interface is connected to the system bus through the input / output interface. The processor of the computer device is used to provide computing and control capabilities. The memory of the computer device includes a non-volatile storage medium and an internal memory. The non-volatile storage medium stores an operating system, a computer program and a database. The internal memory provides an environment for the operation of the operating system and the computer program in the non-volatile storage medium. The database of the computer device is used to store biometric data. The input / output interface of the computer device is used to exchange information between the processor and an external device. The communication interface of the computer device is used to communicate with an external terminal through a network connection. When the computer program is executed by the processor, a biometric authentication method is implemented.

[0201] In one embodiment, a computer device is provided. The computer device may be a terminal, and its internal structure diagram may be as follows: Fig.12As shown. The computer device includes a processor, a memory, an input / output interface, a communication interface, a display unit and an input device. The processor, the memory and the input / output interface are connected via a system bus, and the communication interface, the display unit and the input device are connected to the system bus via the input / output interface. The processor of the computer device is used to provide computing and control capabilities. The memory of the computer device includes a non-volatile storage medium and an internal memory. The non-volatile storage medium stores an operating system and a computer program. The internal memory provides an environment for the operation of the operating system and the computer program in the non-volatile storage medium. The input / output interface of the computer device is used to exchange information between the processor and an external device. The communication interface of the computer device is used to communicate with an external terminal in a wired or wireless manner, and the wireless manner can be implemented through WIFI, a mobile cellular network, NFC (near field communication) or other technologies. When the computer program is executed by the processor, a biometric authentication method is implemented. The display unit of the computer device is used to form a visually visible image, and can be a display screen, a projection device or a virtual reality imaging device. The display screen can be a liquid crystal display screen or an electronic ink display screen. The input device of the computer device can be a touch layer covered on the display screen, or a button, trackball or touchpad set on the computer device casing, or an external keyboard, touchpad or mouse, etc.

[0202] Those skilled in the art will understand that Fig.11 The structure shown in or 12 is merely a block diagram of a partial structure related to the scheme of the present application, and does not constitute a limitation on the computer device to which the scheme of the present application is applied. The specific computer device may include more or fewer components than those shown in the figure, or combine certain components, or have a different arrangement of components.

[0203] In one embodiment, a computer device is provided, including a memory and a processor, wherein a computer program is stored in the memory, and the processor implements the steps of the above-mentioned biometric authentication method when executing the computer program.

[0204] In one embodiment, a computer-readable storage medium is provided, on which a computer program is stored. When the computer program is executed by a processor, the steps of the above-mentioned biometric authentication method are implemented.

[0205] In one embodiment, a computer program product is provided, comprising a computer program, which implements the steps of the above-mentioned biometric authentication method when executed by a processor.

[0206] It should be noted that the user information (including but not limited to user device information, user personal information, etc.) and data (including but not limited to data used for analysis, stored data, displayed data, etc.) involved in this application are all information and data authorized by the user or fully authorized by all parties, and the collection, use and processing of relevant data must comply with relevant laws, regulations and standards of relevant countries and regions. In addition, the subject can choose not to authorize the subject information and related data, or refuse or conveniently refuse to push information, etc.

[0207] Those skilled in the art can understand that all or part of the processes in the above-mentioned embodiment methods can be completed by instructing the relevant hardware through a computer program, and the computer program can be stored in a non-volatile computer-readable storage medium. When the computer program is executed, it can include the processes of the embodiments of the above-mentioned methods. Among them, any reference to the memory, database or other medium used in the embodiments provided in the present application can include at least one of non-volatile and volatile memory. Non-volatile memory can include read-only memory (ROM), magnetic tape, floppy disk, flash memory, optical memory, high-density embedded non-volatile memory, resistive random access memory (ReRAM), magnetoresistive random access memory (MRAM), ferroelectric random access memory (FRAM), phase change memory (PCM), graphene memory, etc. Volatile memory can include random access memory (RAM) or external cache memory, etc. As an illustration and not limitation, RAM can be in various forms, such as static random access memory (SRAM) or dynamic random access memory (DRAM). The database involved in each embodiment provided in this application may include at least one of a relational database and a non-relational database. Non-relational databases may include distributed databases based on blockchains, etc., but are not limited to this. The processor involved in each embodiment provided in this application may be a general-purpose processor, a central processing unit, a graphics processor, a digital signal processor, a programmable logic device, a data processing logic device based on quantum computing, etc., but are not limited to this.

[0208] The technical features of the above embodiments may be combined arbitrarily. To make the description concise, not all possible combinations of the technical features in the above embodiments are described. However, as long as there is no contradiction in the combination of these technical features, they should be considered to be within the scope of this specification.

[0209] The above-described embodiments only express several implementation methods of the present application, and the descriptions thereof are relatively specific and detailed, but they cannot be understood as limiting the scope of the present application. It should be pointed out that, for a person of ordinary skill in the art, several variations and improvements can be made without departing from the concept of the present application, and these all belong to the protection scope of the present application. Therefore, the protection scope of the present application shall be subject to the attached claims.

Claims

1. A biometric authentication method, characterized in that: The method comprises: The feature recognition model is issued to the trusted institution and multiple application terminals; the feature recognition model is used at the application terminal to extract features from the collected biometric information to be authenticated, and obtain the biometric features to be authenticated; the feature recognition model is used at the trusted institution to extract features from the registered biometric information provided by each of the application terminals, and obtain multiple registered biometric features; Receiving a plurality of registered biometric features sent by the trusted institution, and building a biometric feature library based on the plurality of registered biometric features; Determine a target application end among the multiple application ends that initiates a biometric authentication event, and receive the biometric to be authenticated sent by the target application end; The biometric feature to be authenticated is matched with each registered biometric feature in the biometric feature database, and an authentication result of the biometric feature to be authenticated is determined and fed back to the target application end.

2. The method according to claim 1, characterized in that The method further comprises: Sending a security credential to the trusted institution so that the trusted institution performs security verification on the feature recognition model based on the security credential; The receiving a plurality of registered biometric features sent by the trusted institution comprises: When the feature recognition model passes the verification, a plurality of registered biometric features sent by the trusted institution are received.

3. The method according to claim 1, characterized in that The performing feature matching on the biometric feature to be authenticated and each registered biometric feature in the biometric feature database, determining the authentication result of the biometric feature to be authenticated and feeding it back to the target application end, includes: Perform feature matching on the biometric feature to be authenticated and each registered biometric feature in the biometric feature database, and determine the feature similarity between the biometric feature to be authenticated and each registered biometric feature; In the case where there is a target registered biometric feature whose feature similarity satisfies the authentication condition, determining the authentication result of the biometric feature to be authenticated as authentication passed; The identity identifier corresponding to the target registered biometric feature is fed back to the target application end.

4. The method according to any one of claims 1 to 3, characterized in that The process of training the feature recognition model includes: Sending initial recognition models for biometric feature extraction to multiple application terminals; Acquire multiple parameter update information obtained by performing incremental training on the initial recognition model by at least a part of the multiple application terminals; the training samples of the incremental training include biological information collected during the biological feature authentication process; The initial recognition model is updated based on each parameter update information to obtain a feature recognition model.

5. The method according to claim 4, characterized in that The method further comprises: Determine the type of biological information that is expected to be identified, and obtain an open source data set that matches the type of biological information; The open source data set is used to perform model training on the initial neural network to obtain an initial recognition model for biometric feature extraction.

6. The method according to claim 4, characterized in that The updating process of the initial recognition model based on each parameter updating information to obtain a feature recognition model includes: Determine the model parameters corresponding to each of the parameter update information; For each of the model parameters, statistically analyzing each of the parameter update information corresponding to the model parameter to determine an update parameter that matches the model parameter; The initial recognition model is updated based on the update parameters to obtain a feature recognition model.

7. The method according to claim 6, characterized in that The performing statistical analysis on each parameter update information corresponding to the model parameter to determine the update parameter matching the model parameter includes: Determine respective weights of each of the parameter update information corresponding to the model parameters; A weighted sum process is performed on each parameter update information based on each weight to obtain an update parameter that matches the model parameter.

8. The method according to claim 7, characterized in that The determining of the weights of the parameter update information corresponding to the model parameters includes: For each parameter update information corresponding to the model parameter, determine the number of training samples used in the process of training to obtain the parameter update information; Based on the number of training samples corresponding to each of the parameter update information, a weight of each of the parameter update information is determined; and the weight is positively correlated with the number of training samples.

9. The method according to claim 6, characterized in that The update parameter is a parameter offset; The updating process of the initial recognition model based on each of the update parameters to obtain a feature recognition model includes: For each of the model parameters, the initial parameter value of the model parameter is superimposed with the parameter offset of the model parameter to obtain an updated parameter value of the model parameter; The initial recognition model is updated based on the updated parameter values ​​of the model parameters to obtain a feature recognition model.

10. A biometric authentication system, characterized in that: The system includes a server, a trusted organization, and multiple application terminals: The server sends a feature recognition model to a trusted organization and multiple application terminals; The feature recognition model is used at the application end to extract features from the collected biometric information to be authenticated, to obtain the biometric features to be authenticated; The trusted institution uses the feature recognition model to extract features from the registered biometric information provided by each of the application terminals, obtains multiple registered biometric features and sends them to the server terminal, so that the server terminal builds a biometric feature library based on the multiple registered biometric features; The server end is also used to determine the target application end that initiates the biometric authentication event among the multiple application ends, receive the biometric to be authenticated sent by the target application end, perform feature matching on the biometric to be authenticated with each registered biometric in the biometric library, determine the authentication result of the biometric to be authenticated and feed it back to the target application end.

11. The system according to claim 10, characterized in that: The trusted institution is specifically used to encrypt the multiple registered biometric features to obtain the encrypted biometric features and send them to the server; The server is specifically used to decrypt the encrypted biometric features to obtain the multiple registered biometric features.

12. The system according to claim 10, characterized in that The trusted authority is also used to: In response to a registered biometric information maintenance event initiated by at least one of the plurality of application terminals, determining a maintenance type and selected registered biometric information that needs to be maintained; According to the maintenance type, the selected registered biometric information and the selected registered biometric feature of the selected registered biometric information are maintained, and the maintenance result is fed back to the service end and the at least one application end.

13. The system according to any one of claims 10 to 12, characterized in that: The server is also used to send an initial recognition model for extracting biometric features to the multiple application terminals; Each of the application terminals is also used to use the biological information collected during the biological feature authentication process to perform incremental training on the initial recognition model to obtain corresponding parameter update information; The server obtains parameter update information determined by at least a portion of the multiple application terminals, and updates the initial recognition model based on the parameter update information to obtain a feature recognition model.

14. The system according to claim 13, characterized in that: The application end is specifically used to encrypt the determined parameter update information, obtain the encrypted update information and send it to the server end; The server is specifically used to decrypt the encrypted update information to obtain parameter update information.

15. A biometric authentication device, characterized in that: The device comprises: The model sending module is used to send the feature recognition model to the trusted institution and multiple application terminals; the feature recognition model is used at the application terminal to extract features from the collected biometric information to be authenticated, and obtain the biometric features to be authenticated; the feature recognition model is used at the trusted institution to extract features from the registered biometric information provided by each of the application terminals, and obtain multiple registered biometric features; A feature library construction module, used to receive a plurality of registered biometric features sent by the trusted institution, and to construct a biometric feature library based on the plurality of registered biometric features; A biometric feature acquisition module to be authenticated, used to determine a target application end that initiates a biometric feature authentication event among the multiple application ends, and receive the biometric feature to be authenticated sent by the target application end; The feature matching module is used to perform feature matching on the biometric feature to be authenticated and each registered biometric feature in the biometric feature library, determine the authentication result of the biometric feature to be authenticated and feed it back to the target application end.

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

17. A computer-readable storage medium having a computer program stored thereon, characterized in that: When the computer program is executed by a processor, the steps of the method according to any one of claims 1 to 9 are implemented.

18. A computer program product, comprising a computer program, characterized in that When the computer program is executed by a processor, the steps of the method according to any one of claims 1 to 9 are implemented.

Images