Method and system for determining recognition results

Abstract

The application discloses a kind of identification result determination method, it is related to face recognition field, the method comprises: obtaining the first identification result of first model to historical face image under historical identification scene, obtain the second identification result of second model to historical face image, wherein, first model is used to determine the abnormal face image in historical face image, and second model is used to determine the abnormal behavior of historical object corresponding to historical face image;Using the feature vector of first identification result, the feature vector of second identification result and historical face image trains neural network model;After the face image collected in current scene is input into the neural network model that training is completed, the abnormal face image under current scene and / or abnormal behavior under current scene that the neural network model that training is completed outputs are obtained, and according to abnormal face image and / or abnormal behavior, the identification result is determined out.Solve how to improve the accuracy of the identification result of suspicious person.

Description

Technical Field

[0001] This application relates to the field of facial recognition technology, and more specifically, to a method and system for determining recognition results. Background Technology

[0002] Currently, the battle between offense and defense in facial recognition continues to escalate. Existing facial recognition systems, at both the device layer (terminal smart devices, such as mobile phones) and the liveness detection layer (interactive liveness detection, such as action liveness detection), are unable to intercept new attack methods implemented at the algorithm layer, typically adversarial samples, camera hijacking, and custom ROMs. This poses a serious security threat to facial recognition systems. Criminals can bypass the device layer and liveness detection layer by using camera hijacking and custom ROM devices, and then inject a facial image with added data perturbation into the algorithm layer, causing the facial recognition algorithm to misidentify the face, thereby resulting in financial losses for customers.

[0003] Therefore, in the process of facial recognition, there is a technical problem of how to improve the accuracy of the identification results of suspicious persons.

[0004] Regarding the technical problem of improving the accuracy of identifying suspicious persons during the facial recognition process, no effective solution has yet been proposed. Summary of the Invention

[0005] The main objective of this application is to provide a method and system for determining the identification result, so as to solve the technical problem of how to improve the accuracy of the identification result of suspicious persons in the process of face recognition in related technologies.

[0006] To achieve the above objectives, according to one aspect of this application, a method for determining recognition results is provided, comprising: acquiring a first recognition result of a first model on a historical face image in a historical recognition scenario; acquiring a second recognition result of a second model on the historical face image; wherein the first model is used to determine abnormal face images in the historical face image, and the second model is used to determine abnormal behavior of a historical object corresponding to the historical face image; training a neural network model using the feature vector of the first recognition result, the feature vector of the second recognition result, and the historical face image; inputting a face image collected in the current scene into the trained neural network model, acquiring the abnormal face image and / or abnormal behavior in the current scene output by the trained neural network model, and determining the recognition result based on the abnormal face image and / or abnormal behavior.

[0007] In an exemplary embodiment, before obtaining the first recognition result of the first model on historical face images in a historical recognition scenario, the method further includes: determining a first face image group based on historical transaction records, wherein the image similarity of the first face images contained in the first face image group is greater than a preset threshold; when it is determined that N face images contained in the first face image group have different face recognition fields, setting the label type of the N face images to abnormal, where N is a positive integer; generating a preset image library based on the N face images contained in the first face image group, using all face images contained in the preset image library as input samples and the first face image as output samples to train the first model.

[0008] In an exemplary embodiment, determining a first face image group based on historical transaction records includes: acquiring historical face images contained in the historical transaction records, wherein the historical face images correspond to face recognition fields; determining a first initial image from the historical face images whose image similarity is greater than a preset threshold; and generating the first face image group based on at least two first initial images.

[0009] In one exemplary embodiment, it is determined that the N face images included in the first face image group have different face recognition fields by: obtaining the number of different face recognition fields of the N face images; and if the number is greater than a first preset value, determining that the N face images included in the first face image group have different face recognition fields.

[0010] In an exemplary embodiment, before obtaining the second recognition result of the second model on the historical face images, the method further includes: determining a second face image group based on historical transaction records; generating a preset image library based on the face images included in the second face image group when it is determined that the behavior recognition field corresponding to any face image included in the second face image group exists in an abnormal behavior database; and training the second model by using all face images included in the preset image library as input samples and the second face image as output samples.

[0011] In one exemplary embodiment, determining a second face image group based on historical transaction records includes: acquiring historical face images contained in the historical transaction records, wherein the historical face images correspond to behavior recognition fields; determining a second initial image from the historical face images in which the field value corresponding to the behavior recognition field has changed; and generating the second face image group based on at least two second initial images.

[0012] In one exemplary embodiment, the method further includes: determining the number of times the second face image contained in the second face image group is invoked within the invocation time period; if the number of invocations is determined to be greater than a second preset value, then setting the label type of the second face image to abnormal; marking the historical object corresponding to the second face image as having abnormal behavior, and storing the behavior recognition field corresponding to the second face image in the abnormal behavior database.

[0013] To achieve the above objectives, according to another aspect of this application, a data processing result transmission system is provided. The system includes: a face acquisition module, wherein the face acquisition module is used to acquire face images; a face data transmission module, wherein the face data transmission module is used to transmit the acquired face images; and a recognition processing module, wherein the recognition processing module is used to obtain a first recognition result of a first model on historical face images in a historical recognition scenario, and to obtain a second recognition result of a second model on the historical face images, wherein the first model is used to determine abnormal face images in the historical face images, and the second model is used to determine abnormal behavior of historical objects corresponding to the historical face images; using the feature vector of the first recognition result, the feature vector of the second recognition result, and the historical face images... A neural network model is trained using face images. After inputting face images captured in the current scene into the trained neural network model, abnormal face images and / or abnormal behaviors in the current scene output by the trained neural network model are obtained, and a recognition result is determined based on the abnormal face images and / or abnormal behaviors. A face recognition main control module is used to respond to recognition requests and schedule the face acquisition module, the face data transmission module, and the recognition processing module. Communication channels are established between the face acquisition module, the face data transmission module, the recognition processing module, and the face recognition main control module.

[0014] This application employs the following steps: obtaining a first recognition result for historical face images in a historical recognition scenario from a first model used to identify abnormal face images in historical face images; obtaining a second recognition result for historical face images from a second model used to determine abnormal behavior of historical objects corresponding to historical face images; using the feature vectors of the first and second recognition results and the historical face images to train a neural network model; inputting face images collected in the current scene into the trained neural network model to obtain abnormal face images and / or abnormal behavior in the current scene; and then determining the recognition result based on the abnormal face images and / or abnormal behavior. This solves the technical problem of how to improve the accuracy of the recognition results for suspicious persons during the face recognition process, improves the accuracy of the recognition results, and thus improves the security of face recognition. Attached Figure Description

[0015] The accompanying drawings, which form part of this application, are used to provide a further understanding of this application. The illustrative embodiments and descriptions of this application are used to explain this application and do not constitute an undue limitation of this application. In the drawings:

[0016] Figure 1 This is a flowchart of a method for determining the identification result according to an embodiment of this application;

[0017] Figure 2 This is a schematic diagram of model training according to the embodiments of this application;

[0018] Figure 3 This is a system structure block diagram (I) of the system for determining the identification results based on the embodiments of this application;

[0019] Figure 4 This is a system structure block diagram (II) of the system for determining the identification results based on the embodiments of this application;

[0020] Figure 5 This is a system structure block diagram (III) of the system for determining the identification results based on the embodiments of this application;

[0021] Figure 6 This is a system structure block diagram (IV) of the system for determining the identification results based on the embodiments of this application;

[0022] Figure 7 This is a schematic diagram of an optional electronic device provided according to an embodiment of this application. Detailed Implementation

[0023] It should be noted that, unless otherwise specified, the embodiments and features described in this application can be combined with each other. This application will now be described in detail with reference to the accompanying drawings and embodiments.

[0024] To enable those skilled in the art to better understand the present application, the technical solutions in the embodiments of the present application will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present application, and not all embodiments. Based on the embodiments in the present application, all other embodiments obtained by those of ordinary skill in the art without creative effort should fall within the scope of protection of the present application.

[0025] It should be noted that the terms "first," "second," etc., in the specification, claims, and accompanying drawings of this application are used to distinguish similar objects and are not necessarily used to describe a specific order or sequence. It should be understood that such data can be interchanged where appropriate for the embodiments of this application described herein. Furthermore, the terms "comprising" and "having," and any variations thereof, are intended to cover non-exclusive inclusion; for example, a process, method, system, product, or apparatus that comprises a series of steps or units is not necessarily limited to those steps or units explicitly listed, but may include other steps or units not explicitly listed or inherent to such processes, methods, products, or apparatus.

[0026] The following describes this application in conjunction with the preferred implementation steps. Figure 1 This is a flowchart of the method for determining the identification result according to the embodiments of this application, such as... Figure 1 As shown, the method includes the following steps:

[0027] Step S102: Obtain the first recognition result of the first model on the historical face image in the historical recognition scenario, and obtain the second recognition result of the second model on the historical face image, wherein the first model is used to determine the abnormal face image in the historical face image, and the second model is used to determine the abnormal behavior of the historical object corresponding to the historical face image;

[0028] Step S104: Use the feature vector of the first recognition result, the feature vector of the second recognition result, and the historical face image to train a neural network model;

[0029] Step S106 involves inputting the face image collected in the current scene into the trained neural network model, obtaining the abnormal face image and / or abnormal behavior in the current scene output by the trained neural network model, and determining the recognition result based on the abnormal face image and / or abnormal behavior.

[0030] The method for determining the recognition result provided in this application embodiment obtains a first recognition result of a first model on historical face images in a historical recognition scenario, and a second recognition result of a second model on the historical face images. The first model is used to identify abnormal face images in the historical face images, and the second model is used to identify abnormal behavior of historical objects corresponding to the historical face images. A neural network model is trained using the feature vectors of the first and second recognition results and the historical face images. After inputting face images collected in the current scene into the trained neural network model, the abnormal face images and / or abnormal behaviors in the current scene output by the trained neural network model are obtained. The recognition result is then determined based on the abnormal face images and / or abnormal behaviors. This method solves the technical problem in related technologies of how to improve the accuracy of suspicious person recognition results during face recognition, thereby improving the accuracy of the recognition results and thus enhancing the security of face recognition.

[0031] Optionally, the historical recognition scenarios in step S102 above include: face recognition scenarios where a person holds a camera and takes a picture on a mobile banking device, and face recognition scenarios where a person takes a picture at a free angle when entering or leaving a branch.

[0032] Optionally, for step S104 above, the feature vectors of the first recognition result, the feature vectors of the second recognition result, and the historical face image are used to train a neural network model. Specifically, a neural network model can be trained by using the "knowledge distillation technique of teacher-student model" under the constraint of the MSE (mean squared error) loss function. The features extracted by the neural network model include the feature vectors of the first recognition result and the feature vectors of the second recognition result, and the recognition accuracy of the neural network model is higher than that of the first model and the second model.

[0033] The fusion model trained using the aforementioned knowledge distillation technique is a fusion model that has learned the features of all individual models. The features extracted by the fusion model include those of all the individual models, ensuring that the recognition accuracy is higher than that of any single model. The advantage of knowledge distillation is that it can transfer knowledge from large, complex models to smaller models, thereby maintaining high performance while reducing computational resources.

[0034] The following is combined Figure 2 The above model fusion process will be explained. Figure 2 This is a schematic diagram of model training according to the embodiments of this application, such as... Figure 2As shown, video data recognition features (equivalent to feature vectors) are extracted from transaction logs (i.e., historical transaction records) and image data from multiple scenarios. Under the constraint of the MSE (mean squared error) loss function, the feature vectors are then fused to train a neural network model that can output abnormal face images and / or abnormal behaviors in the current scenario. Figure 2 The model shown is a fraud detection and early warning model for suspicious persons. This neural network model can identify suspicious persons.

[0035] In an exemplary embodiment, before implementing step S102 above to obtain the first recognition result of the first model for historical face images in a historical recognition scenario, the following steps are also required: Step S12, determining a first face image group based on historical transaction records, wherein the image similarity of the first face images contained in the first face image group is greater than a preset threshold; Step S14, when it is determined that N face images contained in the first face image group have different face recognition fields, setting the label type of the N face images to abnormal, where N is a positive integer; Step S16, generating a preset image library based on the N face images contained in the first face image group, using all face images contained in the preset image library as input samples and the first face image as output samples to train the first model.

[0036] Optionally, in the above embodiments, the first face image group may be a similar image library containing similar face images, and the preset image library may be a suspicious face image library containing similar suspicious faces.

[0037] Optionally, in the above embodiments, the first model can also be trained in the following way: when there are multiple groups of first face images, a preset image library is generated according to the multiple groups of first face images, and the first model is trained by using all face images contained in the preset image library as input samples and the first face image as output samples.

[0038] In an exemplary embodiment, the process of determining the first face image group based on historical transaction records in step S12 above specifically includes the following steps: obtaining historical face images contained in historical transaction records, wherein the historical face images correspond to face recognition fields; determining a first initial image from the historical face images whose image similarity is greater than the preset threshold; and generating the first face image group based on at least two first initial images.

[0039] Optionally, in the above embodiments, the face recognition fields include, but are not limited to: user identifier, account, scene number, transaction area code, network type, network address, unique device identifier, face acquisition device type, device number, transaction serial number, call date, call time, reference photo image path, and on-site photo image path.

[0040] Optionally, in the above embodiments, the first face image group can be determined in the following way: for example, storing unstructured historical face image data in a file object storage. First, the face image detection framework MTCNN (Multi-Task Cascaded Convolutional Neural Network) algorithm is used to detect and locate faces. Then, the image obtained by masking the face region with a mask (such as mosaic) is used as the base image. The PatchMix algorithm (a simulation algorithm for realizing complex similarity relationships between images under unsupervised conditions) is used to find images similar to the base image. An initial threshold for similar images is set to 80%. Images with more than two similar images are classified and stored. For example, images with image paths x1, x2, and x3 are similar images and are numbered as class X. Class X similar images form a similar image library (equivalent to the first face image group). Similarly, images with image paths y1, y2, y3, y4, and y5 are similar images and are numbered as class Y. Class Y similar images form a similar image library, and so on, forming new similar image libraries.

[0041] In an exemplary embodiment, step S14, which determines that the N face images included in the first face image group have different face recognition fields, can be performed by: obtaining the number of different face recognition fields possessed by the N face images; and determining that the N face images included in the first face image group have different face recognition fields if the number is greater than a first preset value.

[0042] Optionally, in the above embodiments, when the face recognition field contains a user identifier, determining that the N face images contained in the first face image group have different face recognition fields includes: obtaining the first number of different user identifiers possessed by the N face images; and determining that the N face images contained in the first face image group have different face recognition fields if the first number is greater than a preset value. For example, using the image path of each type of image in the similar image library as an index condition, the face transaction log is recursively traversed. When the user identifier corresponding to a type of similar image is greater than or equal to 2 (i.e., more than 2 user identifiers are different), the face image is marked as a suspicious face and stored in the suspicious face image library (equivalent to a preset image library).

[0043] Optionally, in the above embodiments, when the face recognition field includes a call date, determining that the N face images contained in the first face image group have different face recognition fields includes: obtaining a second number of different call dates for the N face images; and determining that the N face images contained in the first face image group have different face recognition fields if the second number is greater than a preset value. For example, using the image path of each type of image in the similar image library as an index condition, the face transaction log is recursively traversed. When the call dates corresponding to a type of similar images are not the same, the face image is marked as a suspicious face and stored in the suspicious face image library.

[0044] Optionally, in the above embodiments, when the face recognition field includes a transaction region, determining that the N face images contained in the first face image group have different face recognition fields includes: obtaining a third number representing the different transaction regions of the N face images; and determining that the N face images contained in the first face image group have different face recognition fields if the third number is greater than a preset value. For example, using the image path of each type of image in the similar image library as an index condition, the face transaction log is recursively traversed. When the transaction regions corresponding to a type of similar images are not the same, the face image is marked as a suspicious face and stored in the suspicious face image library.

[0045] Optionally, in the above embodiments, when the face recognition field includes a device number, determining that the N face images contained in the first face image group have different face recognition fields includes: obtaining the fourth number of the different device numbers of the N face images; and determining that the N face images contained in the first face image group have different face recognition fields if the fourth number is greater than a preset value. For example, using the image path of each type of image in the similar image library as an index condition, the face transaction log is recursively traversed. When the device numbers corresponding to a type of similar images are different, the face image is marked as a suspicious face and stored in the suspicious face image library.

[0046] Optionally, in the above embodiments, the suspicious face image database obtained by the above annotation can be used as training samples, input into the Vision Transformer network framework, and the input samples are grouped into triples using the trplet loss function. The training is carried out with the principle of bringing positive samples closer and negative samples farther apart, so as to obtain a suspicious face image recognition model (equivalent to the first model).

[0047] In an exemplary embodiment, before obtaining the second recognition result of the second model on the historical face image in step S102, the following steps are also required: Step S22, determining the second face image group based on historical transaction records; Step S24, if it is determined that the behavior recognition field corresponding to any face image in the second face image group exists in the abnormal behavior database, generating a preset image library based on the face images in the second face image group; Step S26, training the second model using all face images in the preset image library as input samples and the second face image as output samples.

[0048] Optionally, in an exemplary embodiment, before obtaining the second recognition result of the second model on the historical face image in step S102 above, the following steps may also be performed: determining multiple groups of second face images based on historical transaction records, generating a preset image library based on the multiple groups of second face images, using all face images contained in the preset image library as input samples and the second face image as output samples to train the second model.

[0049] In an exemplary embodiment, the step S22 of determining the second face image group based on historical transaction records can be performed as follows: obtaining historical face images contained in historical transaction records, wherein the historical face images correspond to behavior recognition fields; determining second initial images from the historical face images whose field values ​​corresponding to the behavior recognition fields have changed; and generating the second face image group based on at least two second initial images.

[0050] Optionally, in the above embodiments, the behavior recognition fields include, but are not limited to: user identifier, account, scene number, transaction area code, network type, network address, unique device identifier, face acquisition device type, device number, transaction serial number, call date, call time, reference photo image path, and on-site photo image path.

[0051] In one exemplary embodiment, the method further includes: determining the number of times the second face image contained in the second face image group is invoked within the invocation time period; if the number of invocations is determined to be greater than a second preset value, then setting the label type of the second face image to abnormal; marking the historical object corresponding to the second face image as having abnormal behavior, and storing the behavior recognition field corresponding to the second face image in the abnormal behavior database.

[0052] Optionally, in the above embodiments, the above-mentioned call time period represents the time period during the call, which can start timing from the first call or from any call.

[0053] Optionally, in the above embodiments, for example, the above call time period is set to one hour, starting from the first call user identifier, and users corresponding to user identifiers that make more than 15 calls within an hour are marked as suspicious persons with abnormal behavior, and the user identifiers and corresponding behavior identification fields are stored in the abnormal behavior database.

[0054] Optionally, in the above embodiments, for example, the above call time period is set to one day, starting from the first call user identifier, and users corresponding to user identifiers that make more than 30 calls within a day are marked as suspicious persons with abnormal behavior, and the user identifiers and corresponding behavior identification fields are stored in the abnormal behavior database.

[0055] Optionally, in an exemplary embodiment, the method further includes: determining that the number of calls is greater than a second preset value, and that different behavior recognition fields exist when the second face image is called, then obtaining the number of different behavior recognition fields; if the number of different behavior recognition fields is greater than a third preset value, setting the label type of the second face image to abnormal, marking that the historical object corresponding to the second face image has abnormal behavior, and storing the behavior recognition fields corresponding to the second face image in an abnormal behavior database.

[0056] Optionally, in the above embodiments, when the behavior recognition field includes a device number and a user identifier, if it is determined that the number of different device numbers corresponding to the user identifier is greater than a preset value (for example, device numbers can be compared in pairs, and if they are different more than 5 times, then it is determined that the number of different device numbers is greater than a preset value), then the label type of the second face image is set to abnormal, marking that the historical object corresponding to the second face image has abnormal behavior, and the user identifier corresponding to the second face image is stored in the abnormal behavior database.

[0057] Optionally, in the above embodiments, for example, the above-mentioned call time period is set to one hour, starting from the first call user identifier. The device numbers of the calling devices corresponding to the user identifiers that make more than 5 calls within an hour are compared pairwise. If the comparison results of the device numbers are different more than 5 times, the user corresponding to the user identifier is marked as a suspicious person with abnormal behavior, and the user identifier and the corresponding behavior identification field are stored in the abnormal behavior database.

[0058] Optionally, in the above embodiments, for example, the above-mentioned call time period is set to one day, and the device numbers of the calling devices corresponding to the user identifiers that make more than 10 calls within a day are compared pairwise. If the comparison results of the device numbers are different more than 10 times, the user corresponding to the user identifier is marked as a suspicious person with abnormal behavior, and the user identifier and the corresponding behavior identification field are stored in the abnormal behavior database.

[0059] Optionally, in the above embodiments, when the behavior recognition field includes a network address and a user identifier, if the number of different network addresses is determined to be greater than a preset value, the label type of the second face image is set to abnormal, the historical object corresponding to the second face image is marked as having abnormal behavior, and the user identifier corresponding to the second face image is stored in the abnormal behavior database.

[0060] Optionally, in the above embodiments, for example, the above call time period is set to one hour, starting from the first call to the user identifier. For user identifiers that make more than 5 calls within an hour, all network addresses that initiate calls to the user identifier are compared pairwise. If the comparison results of more than 5 network addresses are different, the user corresponding to the user identifier is marked as a suspicious person with abnormal behavior, and the user identifier and the corresponding behavior identification field are stored in the abnormal behavior database.

[0061] Optionally, in the above embodiments, for example, the above call time period is set to one day. For user identifiers that make more than 10 calls within a day, all network addresses that initiate calls to the user identifier are compared pairwise. If the comparison results of more than 10 network addresses are different, the user corresponding to the user identifier is marked as a suspicious person with abnormal behavior, and the user identifier and the corresponding behavior identification field are stored in the abnormal behavior database.

[0062] Optionally, in the above embodiments, the face images corresponding to the user identifiers in the abnormal behavior database obtained by the above annotation (equivalent to all face images contained in the preset image library) can be used as training samples, subdivided into different classification tasks according to the annotation fields, and the labeled transaction log data can be classified and trained using a random forest model to obtain a customer transaction behavior recognition model (equivalent to the second model).

[0063] This application also provides a system for determining identification results. It should be noted that the system for determining identification results in this application can be used to execute the method for determining identification results provided in this application. The system for determining identification results in this application will be described below with reference to the accompanying drawings.

[0064] Figure 3This is a system structure block diagram (I) of the system for determining the identification results based on the embodiments of this application. For example... Figure 3 As shown, the system includes: a face acquisition module 31, a face data transmission module 32, a recognition processing module 33, and a face recognition main control module 34.

[0065] The face acquisition module 31 is used to acquire face images. The face data transmission module 32 is used to transmit the acquired face images. The recognition processing module 33 is used to obtain the first recognition result of the first model on the historical face images in the historical recognition scenario, and to obtain the second recognition result of the second model on the historical face images. The first model is used to identify abnormal face images in the historical face images, and the second model is used to identify abnormal behavior of the historical objects corresponding to the historical face images. The module trains a neural network model using the feature vectors of the first recognition result, the feature vectors of the second recognition result, and the historical face images. After inputting the face images acquired in the current scene into the trained neural network model, the module acquires the abnormal face images and / or abnormal behavior in the current scene output by the trained neural network model, and determines the recognition result based on the abnormal face images and / or abnormal behavior. The face recognition main control module 34 is used to respond to recognition requests and schedule the face acquisition module, the face data transmission module, and the recognition processing module; wherein, a communication channel is established between the face acquisition module, the face data transmission module, the recognition processing module, and the face recognition main control module.

[0066] Through the above system, when conducting face recognition transactions, the face recognition main control module 34 can schedule the face acquisition module 31, the face data transmission module 32, and the recognition processing module 33, so that the face acquisition module 31 drives the camera to acquire the user's face image / video data, and the face data transmission module 32 uses the face image / video data to upload the face image / video data to the recognition processing module 33 for processing, thereby determining the identification result of suspicious persons.

[0067] In an exemplary embodiment, the above-mentioned recognition processing module further includes: a first determining unit, configured to determine a first face image group based on historical transaction records, wherein the image similarity of the first face images contained in the first face image group is greater than a preset threshold; a first setting unit, configured to set the label type of the N face images to abnormal when it is determined that the N face images contained in the first face image group have different face recognition fields, wherein N is a positive integer; and a first generating unit to generate a preset image library based on the N face images contained in the first face image group, and to train the first model by using all face images contained in the preset image library as input samples and the first face image as output samples.

[0068] In an exemplary embodiment, the first determining unit is further configured to acquire historical face images contained in historical transaction records, wherein the historical face images correspond to face recognition fields; determine a first initial image from the historical face images whose image similarity is greater than the preset threshold; and generate the first face image group based on at least two first initial images.

[0069] In an exemplary embodiment, the first setting unit is further configured to: obtain the number of different face recognition fields of the N face images; and, if the number is determined to be greater than a first preset value, determine that the N face images included in the first face image group have different face recognition fields.

[0070] In one exemplary embodiment, the above-mentioned recognition processing module further includes: a second determining unit, configured to determine a second face image group based on historical transaction records; a second generating unit, configured to generate a preset image library based on the face images included in the second face image group when it is determined that the behavior recognition field corresponding to any face image included in the second face image group exists in an abnormal behavior database; and a first training unit, configured to train the second model using all face images included in the preset image library as input samples and the second face image as output samples.

[0071] In an exemplary embodiment, the second determining unit is further configured to: acquire historical face images contained in historical transaction records, wherein the historical face images correspond to behavior recognition fields; determine from the historical face images a second initial image in which the field value corresponding to the behavior recognition field changes; and generate the second face image group based on at least two second initial images.

[0072] In an exemplary embodiment, the above-mentioned recognition processing module further includes: a third determining unit, configured to determine the number of times the second face image contained in the second face image group is invoked within the invocation time period; a second setting unit, configured to set the label type of the second face image to abnormal if the number of invocations is determined to be greater than a second preset value; and a first marking unit, configured to mark the historical object corresponding to the second face image as having abnormal behavior, and store the behavior recognition field corresponding to the second face image in the abnormal behavior database.

[0073] Optionally, in one embodiment, it can be combined with Figure 4 The function of the face capture module 31 will be explained. Figure 4 This is a system structure block diagram (II) of the system for determining the identification results based on the embodiments of this application, as shown below. Figure 4As shown, the face acquisition module 31 specifically includes a face acquisition main control MCU unit 3101, a face data acquisition control unit 3102, a face data quality control unit 3103, and a customer transaction behavior data acquisition unit 3104.

[0074] The face acquisition module 31 specifically includes a face acquisition main control MCU unit 3101, a face data acquisition control unit 3102, a face data quality control unit 3103, and a customer transaction behavior data acquisition unit 3104, which can communicate wirelessly in real time or via wired internal communication.

[0075] The face acquisition module 31 is deployed on the terminal of the intelligent interactive device and is used to: drive the camera deployed on the intelligent interactive device to acquire the user's face image / video data through the face acquisition main control MCU unit 3101; drive the environmental sensor to acquire data such as ambient light intensity, background depth, face distance, multiple faces, tilt angle, and left and right tilt angle; and send the acquired face image / video data to the face data quality control unit 3103 through the face acquisition main control MCU unit 3101.

[0076] When the collected data is an image, it is necessary to determine whether the face in the image meets the following usage standards: interpupillary distance ≥ 60 pixels, no motion blur, no overly bright image, no overly dark image, and no faceless image.

[0077] When the acquired data is video, keyframes need to be extracted from the video data first, and then it needs to be determined whether the face in the image frame meets the above usage standards. For example, if the video duration is less than 5 seconds, keyframes are extracted at a frequency of 2 frames / second. If the face in the image frame meets the above usage standards, a face data processing request and an environmental parameter transmission request can be sent to the face data transmission module 32.

[0078] Optionally, in one embodiment, it can be combined with Figure 5 The function of the face data transmission module 32 will be explained below. Figure 5 This is a system structure block diagram (III) of the system for determining the identification results provided in the embodiments of this application, as shown in the following example. Figure 5 As shown, the face data transmission module 32 specifically includes a face data transmission main control unit 3201, a data upload unit 3202, and a data distribution unit 3203.

[0079] The face data transmission master control unit 3201 performs a data security check during data upload. If the uploaded data is determined to be a face image, video data, or environmental parameters, the data upload unit 3202 uploads the face data to the cloud. During data download, the face data transmission master control unit 3201 sends an image / feature query request to the cloud database via the data download unit 3203, and then sends the corresponding results to the smart interactive device or application server.

[0080] Optionally, in one embodiment, it can be combined with Figure 6 The function of the face recognition main control module 34 will be explained. Figure 6 This is a system structure block diagram (IV) of the system for determining the identification results provided in the embodiments of this application, as shown below. Figure 6 As shown, the face recognition main control module 34 specifically includes a face recognition algorithm main control unit 3401, a suspicious face detection and early warning handling unit 3402, a face registration algorithm processing unit 3403, a face 1:1 algorithm processing unit 3404, and a face 1:n algorithm processing unit 3405.

[0081] The face recognition algorithm main control unit 3401 is used to schedule the suspicious face detection and early warning processing unit 3402, the face registration algorithm processing unit 3403, the face 1:1 algorithm processing unit 3404, and the face 1:n algorithm processing unit 3405.

[0082] After the suspicious face detection and early warning processing unit 3402 completes the suspicious face detection and recognition, the face image data is sent to the corresponding algorithm service processing unit (i.e., Figure 6 The face registration algorithm processing unit 3403, the face 1:1 algorithm processing unit 3404, or the face 1:n algorithm processing unit 3405 in the middle are used for processing.

[0083] The face registration algorithm processing unit 3403 is used to perform data preprocessing, extract and model face features, and then register user information and face features into the face recognition system database.

[0084] The face 1:1 algorithm processing unit 3404 is used to perform image preprocessing, extract and model face features, compare the feature data with the face features queried from the database in a 1:1 manner, and return the comparison results.

[0085] The face 1:n algorithm processing unit 3405 is used to perform image preprocessing, extract and model face features, identify n faces corresponding to unique user identifiers in the database, and return the identification results.

[0086] It should be noted that the steps shown in the flowchart in the accompanying drawings can be executed in a computer system such as a set of computer-executable instructions, and although a logical order is shown in the flowchart, in some cases the steps shown or described may be executed in a different order than that shown here.

[0087] This application provides a computer-readable storage medium storing a program thereon, which, when executed by a processor, implements a method for determining recognition results.

[0088] This application provides a processor for running a program, wherein the program executes a method for determining the recognition result during runtime.

[0089] like Figure 7 As shown, this application embodiment provides an electronic device, which includes a processor, a memory, and a program stored in the memory and executable on the processor. When the processor executes the program, it performs the following steps: obtaining a first recognition result of a first model on a historical face image in a historical recognition scenario, obtaining a second recognition result of a second model on the historical face image, wherein the first model is used to determine abnormal face images in the historical face image, and the second model is used to determine abnormal behavior of the historical object corresponding to the historical face image; training a neural network model using the feature vector of the first recognition result, the feature vector of the second recognition result, and the historical face image; after inputting a face image collected in the current scene into the trained neural network model, obtaining the abnormal face image and / or abnormal behavior in the current scene output by the trained neural network model, and determining the recognition result based on the abnormal face image and / or abnormal behavior.

[0090] In an exemplary embodiment, the processor is further configured to implement the following steps: determining a first face image group based on historical transaction records, wherein the image similarity of the first face images contained in the first face image group is greater than a preset threshold; when it is determined that N face images contained in the first face image group have different face recognition fields, setting the label type of the N face images to abnormal, where N is a positive integer; generating a preset image library based on the N face images contained in the first face image group, using all face images contained in the preset image library as input samples and the first face image as output samples to train the first model.

[0091] In an exemplary embodiment, the processor is further configured to implement the following steps: acquiring historical face images contained in historical transaction records, wherein the historical face images correspond to face recognition fields; determining a first initial image from the historical face images whose image similarity is greater than the preset threshold; and generating a first face image group based on at least two first initial images.

[0092] In an exemplary embodiment, the processor is further configured to implement the following steps: obtaining the number of different face recognition fields possessed by the N face images; and, if the number is determined to be greater than a first preset value, determining that the N face images contained in the first face image group have different face recognition fields.

[0093] In an exemplary embodiment, the processor is further configured to implement the following steps: determining a second face image group based on historical transaction records; generating a preset image library based on the face images contained in the second face image group when it is determined that the behavior recognition field corresponding to any face image contained in the second face image group exists in an abnormal behavior database; and training the second model by using all face images contained in the preset image library as input samples and the second face image as output samples.

[0094] In an exemplary embodiment, the processor is further configured to implement the following steps: acquiring historical face images contained in historical transaction records, wherein the historical face images correspond to behavior recognition fields; determining from the historical face images a second initial image in which the field value corresponding to the behavior recognition field changes; and generating a second face image group based on at least two second initial images.

[0095] In an exemplary embodiment, the processor is further configured to implement the following steps: determining the number of times the second face image contained in the second face image group is invoked within the invocation time period; if the number of invocations is determined to be greater than a second preset value, then setting the label type of the second face image to abnormal; marking that the historical object corresponding to the second face image has abnormal behavior, and storing the behavior recognition field corresponding to the second face image in the abnormal behavior database.

[0096] The devices mentioned in this article can be servers, PCs, tablets, mobile phones, etc.

[0097] This application also provides a computer program product, which, when executed on a data processing device, is suitable for executing an initialization program having the following method steps: obtaining a first recognition result of a first model on a historical face image in a historical recognition scenario, obtaining a second recognition result of a second model on the historical face image, wherein the first model is used to determine abnormal face images in the historical face image, and the second model is used to determine abnormal behavior of historical objects corresponding to the historical face image; training a neural network model using the feature vector of the first recognition result, the feature vector of the second recognition result, and the historical face image; after inputting a face image collected in the current scene into the trained neural network model, obtaining the abnormal face image and / or abnormal behavior in the current scene output by the trained neural network model, and determining the recognition result based on the abnormal face image and / or abnormal behavior.

[0098] In an exemplary embodiment, the computer program product described above, when executed on a data processing device, is further adapted to perform the following steps: determining a first face image group based on historical transaction records, wherein the image similarity of the first face images contained in the first face image group is greater than a preset threshold; when it is determined that N face images contained in the first face image group have different face recognition fields, setting the label type of the N face images to abnormal, where N is a positive integer; generating a preset image library based on the N face images contained in the first face image group, and training the first model by using all face images contained in the preset image library as input samples and the first face image as an output sample.

[0099] In an exemplary embodiment, the computer program product described above, when executed on a data processing device, is further adapted to perform a program that includes the following steps: acquiring historical face images contained in historical transaction records, wherein the historical face images correspond to face recognition fields; determining a first initial image from the historical face images whose image similarity is greater than the preset threshold; and generating a first face image group based on at least two first initial images.

[0100] In an exemplary embodiment, the computer program product described above, when executed on a data processing device, is further adapted to perform a program that includes the following steps: obtaining the number of different face recognition fields possessed by the N face images; and, if the number is determined to be greater than a first preset value, determining that the N face images contained in the first face image group have different face recognition fields.

[0101] In an exemplary embodiment, the computer program product described above, when executed on a data processing device, is further adapted to perform the following steps: determining a second face image group based on historical transaction records; generating a preset image library based on the face images contained in the second face image group when it is determined that the behavior recognition field corresponding to any face image contained in the second face image group exists in an abnormal behavior database; and training the second model using all face images contained in the preset image library as input samples and the second face image as output samples.

[0102] In an exemplary embodiment, the computer program product described above, when executed on a data processing device, is further adapted to perform a program that includes the following steps: acquiring historical face images contained in historical transaction records, the historical face images corresponding to behavior recognition fields; determining from the historical face images a second initial image in which the field value corresponding to the behavior recognition field changes; and generating a second face image group based on at least two second initial images.

[0103] In an exemplary embodiment, the computer program product described above, when executed on a data processing device, is further adapted to perform the following steps: determining the number of times the second face image contained in the second face image group is invoked within the invocation time period; if the number of invocations is determined to be greater than a second preset value, setting the label type of the second face image to abnormal; marking the historical object corresponding to the second face image as having abnormal behavior, and storing the behavior recognition field corresponding to the second face image in the abnormal behavior database.

[0104] Those skilled in the art will understand that embodiments of this application can be provided as methods, systems, or computer program products. Therefore, this application can take the form of a completely hardware embodiment, a completely software embodiment, or an embodiment combining software and hardware aspects. Furthermore, this application can take the form of a computer program product embodied on one or more computer-usable storage media (including but not limited to disk storage, CD-ROM, optical storage, etc.) containing computer-usable program code.

[0105] This application is described with reference to flowchart illustrations and / or block diagrams of methods, apparatus (systems), and computer program products according to embodiments of this application. It will be understood that each block of the flowchart illustrations and / or block diagrams, and combinations of blocks in the flowchart illustrations and / or block diagrams, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, special-purpose computer, embedded processor, or other programmable data processing apparatus to produce a machine, such that the instructions, which execute via the processor of the computer or other programmable data processing apparatus, generate instructions for implementing the flowchart... Figure 1 One or more processes and / or boxes Figure 1 A device that provides the functions specified in one or more boxes.

[0106] These computer program instructions may also be stored in a computer-readable storage medium that can direct a computer or other programmable data processing device to function in a particular manner, such that the instructions stored in the computer-readable storage medium produce an article of manufacture including instruction means, which are implemented in a process Figure 1 One or more processes and / or boxes Figure 1 The function specified in one or more boxes.

[0107] These computer program instructions may also be loaded onto a computer or other programmable data processing equipment to cause a series of operational steps to be performed on the computer or other programmable equipment to produce a computer-implemented process, thereby providing instructions that execute on the computer or other programmable equipment for implementing the process. Figure 1 One or more processes and / or boxes Figure 1 The steps of the function specified in one or more boxes.

[0108] In a typical configuration, a computing device includes one or more processors (CPU), input / output interfaces, network interfaces, and memory.

[0109] Memory may include non-persistent memory in computer-readable media, such as random access memory (RAM) and / or non-volatile memory, such as read-only memory (ROM) or flash RAM. Memory is an example of computer-readable media.

[0110] Computer-readable media includes both permanent and non-permanent, removable and non-removable media that can store information using any method or technology. Information can be computer-readable instructions, data structures, modules of programs, or other data. Examples of computer storage media include, but are not limited to, phase-change memory (PRAM), static random access memory (SRAM), dynamic random access memory (DRAM), other types of random access memory (RAM), read-only memory (ROM), electrically erasable programmable read-only memory (EEPROM), flash memory or other memory technologies, CD-ROM, digital versatile optical disc (DVD) or other optical storage, magnetic tape, magnetic magnetic disk storage or other magnetic storage devices, or any other non-transferable medium that can be used to store information accessible by a computing device. As defined herein, computer-readable media does not include transient computer-readable media, such as modulated data signals and carrier waves.

[0111] It should also be noted that the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such process, method, article, or apparatus. Unless otherwise specified, an element defined by the phrase "comprising one..." does not exclude the presence of other identical elements in the process, method, article, or apparatus that includes that element.

[0112] Those skilled in the art will understand that embodiments of this application can be provided as methods, systems, or computer program products. Therefore, this application can take the form of a completely hardware embodiment, a completely software embodiment, or an embodiment combining software and hardware aspects. Furthermore, this application can take the form of a computer program product embodied on one or more computer-usable storage media (including, but not limited to, disk storage, CD-ROM, optical storage, etc.) containing computer-usable program code.

[0113] The above are merely embodiments of this application and are not intended to limit the scope of this application. Various modifications and variations can be made to this application by those skilled in the art. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of this application should be included within the scope of the claims of this application.

Claims

1. A method for determining an identification result, characterized in that, Applied to the recognition and processing module, including: The system obtains a first recognition result of a first model on a historical face image in a historical recognition scenario, and obtains a second recognition result of a second model on the historical face image. The first model is used to determine abnormal face images in the historical face image, and the second model is used to determine abnormal behavior of the historical object corresponding to the historical face image. The neural network model is trained using the feature vector of the first recognition result, the feature vector of the second recognition result, and the historical face image. After inputting the facial images captured in the current scene into the trained neural network model, the abnormal facial images and / or abnormal behaviors in the current scene output by the trained neural network model are obtained, and the recognition result is determined based on the abnormal facial images and / or abnormal behaviors; Before obtaining the first recognition result of the first model on historical face images in a historical recognition scenario, the method further includes: The first face image group is determined based on historical transaction records. The image similarity of the first face images contained in the first face image group is greater than a preset threshold. The first face image group is a similar image library containing similar face images. If it is determined that the N face images contained in the first face image group have different face recognition fields, the label type of the N face images is set to abnormal, where N is a positive integer; A preset image library is generated based on N face images contained in the first face image group. The first model is trained by using all face images contained in the preset image library as input samples and the first face image as output samples. The preset image library is a suspicious face image library that includes similar suspicious faces. When the face recognition field contains a user identifier, it is determined that the N face images contained in the first face image group have different face recognition fields in the following way: obtain the first number of different user identifiers of the N face images, and if it is determined that the first number is greater than a preset value, it is determined that the N face images contained in the first face image group have different face recognition fields.

2. The method according to claim 1, characterized in that, The first group of facial images was identified based on historical transaction records, including: Obtain historical facial images contained in historical transaction records, wherein the historical facial images correspond to facial recognition fields; A first initial image with an image similarity greater than the preset threshold is determined from the historical facial images; The first face image group is generated based on at least two first initial images.

3. The method according to claim 1, characterized in that, The N face images contained in the first face image group are determined to have different face recognition fields in the following way: Obtain the number of different face recognition fields possessed by the N face images; If the number is greater than a first preset value, it is determined that the N face images contained in the first face image group have different face recognition fields.

4. The method according to claim 1, characterized in that, Before obtaining the second recognition result of the second model on the historical face image, the method further includes: The second group of facial images was determined based on historical transaction records; If it is determined that the behavior recognition field corresponding to any face image contained in the second face image group exists in the abnormal behavior database, a preset image library is generated based on the face images contained in the second face image group; The second model is trained by using all face images contained in the preset image library as input samples and the second face image as output samples.

5. The method according to claim 4, characterized in that, The second set of facial images was determined based on historical transaction records, including: Obtain historical facial images contained in historical transaction records, wherein the historical facial images correspond to behavioral recognition fields; Determine a second initial image from the historical facial images in which the field value corresponding to the behavior recognition field has changed; The second face image group is generated based on at least two second initial images.

6. The method according to claim 4, characterized in that, The method further includes: Determine the number of times the second face image contained in the second face image group is invoked within the invocation time period; If it is determined that the number of calls is greater than the second preset value, then the label type of the second face image is set to abnormal; The historical object corresponding to the second face image is marked as having abnormal behavior, and the behavior recognition field corresponding to the second face image is stored in the abnormal behavior database.

7. A system for determining recognition results, characterized in that, include: A face acquisition module, wherein the face acquisition module is used to acquire face images; A face data transmission module, wherein the face data transmission module is used to transmit the acquired face images; The recognition processing module is configured to acquire a first recognition result of a first model on a historical face image in a historical recognition scenario, and acquire a second recognition result of a second model on the historical face image. The first model is used to identify abnormal face images in the historical face image, and the second model is used to identify abnormal behavior of the historical object corresponding to the historical face image. A neural network model is trained using the feature vectors of the first and second recognition results and the historical face image. After inputting a face image collected in the current scene into the trained neural network model, the module acquires the abnormal face image and / or abnormal behavior in the current scene output by the trained neural network model, and determines the recognition result based on the abnormal face image and / or abnormal behavior. A face recognition main control module, wherein the face recognition main control module is used to respond to recognition requests and schedule the face acquisition module, the face data transmission module, and the recognition processing module; The face acquisition module, the face data transmission module, the recognition processing module, and the face recognition main control module all have communication channels established between them. The recognition processing module further includes: a first determining unit, configured to determine a first face image group based on historical transaction records, wherein the image similarity of the first face images contained in the first face image group is greater than a preset threshold, and the first face image group is a similar image library containing similar face images; a first setting unit, configured to set the label type of the N face images to abnormal when it is determined that the N face images contained in the first face image group have different face recognition fields, wherein N is a positive integer; a first generating unit to generate a preset image library based on the N face images contained in the first face image group, using all face images contained in the preset image library as input samples and the first face image as output samples to train the first model, wherein the preset image library is a suspicious face image library containing similar suspicious faces; when the face recognition field contains a user identifier, it is determined that the N face images contained in the first face image group have different face recognition fields by: obtaining the first number of different user identifiers of the N face images, and determining that the N face images contained in the first face image group have different face recognition fields when it is determined that the first number is greater than a preset value.

8. A processor, characterized in that, The processor is used to run a program, wherein the program executes the method of any one of claims 1 to 6 when it runs.

9. An electronic device, characterized in that, It includes one or more processors and a memory for storing one or more programs, wherein when the one or more programs are executed by the one or more processors, the one or more processors cause the one or more processors to implement the method of any one of claims 1 to 6.

Images