A face recognition method, device, equipment and storage medium

Abstract

Embodiments of the present application provide a face recognition method, device and equipment, and a storage medium, and relate to the technical field of artificial intelligence. The method comprises: obtaining a second decision threshold corresponding to a feature comparison model in a target scene by jointly testing the feature comparison model and an auxiliary detection model. The mutual influence between multiple algorithm models in a face recognition scene is fully considered. Compared with manually setting the decision threshold of each algorithm model, the method in the present application is more suitable for the scene and changes in the scene of multi-model face recognition, thereby improving the accuracy and efficiency of the obtained decision threshold, and further improving the accuracy of multi-model face recognition.

Description

Technical Field

[0001] The embodiments of the present application relate to the field of artificial intelligence technology, and in particular to a face recognition method, apparatus, device, and storage medium. Background Art

[0002] Facial recognition technology is currently widely adopted in fields such as security and finance. To improve the accuracy of facial recognition results, a common practice is to use different models to detect facial images and obtain comprehensive recognition results. Since different scenarios have different requirements for risk control, recognition performance, and user experience for facial recognition systems, for example, scenarios with frequent liveness attacks at unattended terminals require a higher interception capability for the liveness detection model; scenarios with frequent synthetic attacks at untrusted terminals require a higher interception capability for the synthetic image detection model; and open scenarios with a high probability of false positives require a higher interception capability for the feature comparison model. To meet the customized needs of each scenario, it is necessary to configure the decision thresholds of multiple models for each scenario.

[0003] Related technologies manually configure the decision threshold of each model to adapt to changes in business scenarios. This method is not only inefficient, but also easily leads to low accuracy when using multiple models for face recognition. Summary of the Invention

[0004] The embodiments of the present application provide a face recognition method, apparatus, device, and storage medium for improving the accuracy of face recognition.

[0005] In one aspect, an embodiment of the present application provides a face recognition method, the method comprising:

[0006] Obtain the face image to be recognized in the target scene;

[0007] Using at least one auxiliary detection model to verify the face image to be recognized, and obtain at least one verification score;

[0008] If the at least one verification score is respectively greater than the first decision threshold corresponding to each of the at least one auxiliary detection models in the target scene, then using the feature comparison model to perform feature comparison on the face image to be identified to obtain a target comparison score;

[0009] If the target comparison score is greater than the second judgment threshold corresponding to the feature comparison model in the target scene, it is determined that the recognition of the face image to be identified is successful, and the second judgment threshold is obtained by jointly testing the feature comparison model and the auxiliary detection model.

[0010] In one aspect, an embodiment of the present application provides a face recognition device, comprising:

[0011] An acquisition module is used to acquire the face image to be recognized in the target scene;

[0012] a verification module, configured to verify the face image to be identified using at least one auxiliary detection model to obtain at least one verification score;

[0013] A feature comparison module, if the at least one verification score is respectively greater than the first decision threshold corresponding to each of the at least one auxiliary detection model in the target scene, performs feature comparison on the face image to be identified using the feature comparison model to obtain a target comparison score;

[0014] The recognition module determines that the recognition of the face image to be recognized is successful if the target comparison score is greater than the second judgment threshold corresponding to the feature comparison model in the target scene, and the second judgment threshold is obtained by jointly testing the feature comparison model and the auxiliary detection model.

[0015] Optionally, the identification module is specifically configured to:

[0016] Using a first sample image set, jointly testing the auxiliary detection model and the feature comparison model to obtain a predicted comparison score corresponding to each original sample image in the first sample image set;

[0017] Based on a first preset interception rate, screening out at least one valid sample image from the first sample image set;

[0018] The second decision threshold is determined based on the predicted comparison score corresponding to each of the at least one valid sample images.

[0019] Optionally, the identification module is specifically configured to:

[0020] multiplying the first preset interception rate by the number of original sample images in the first sample image set to obtain a target screening number;

[0021] Sorting the original sample images in the first sample image set in descending order of the predicted comparison scores to obtain a sample image sorting result;

[0022] According to the sample image sorting result, the target screening number of valid sample images are screened out from the sample image set.

[0023] Optionally, the identification module is specifically configured to:

[0024] The minimum predicted comparison score among the predicted comparison scores corresponding to the at least one valid sample image is used as the second decision threshold.

[0025] Optionally, the identification module is specifically configured to:

[0026] Acquire a first sample image set and a plurality of sample thresholds;

[0027] Using the first sample image set, jointly testing the auxiliary detection model and the feature comparison model to obtain an interception rate corresponding to the feature comparison model under the multiple sample thresholds;

[0028] Among the obtained interception rates, the sample threshold corresponding to the second preset interception rate is used as the second decision threshold.

[0029] Optionally, the at least one auxiliary detection model includes a basic detection model and at least one feature verification model;

[0030] Among them, the first decision threshold corresponding to the basic detection model is obtained by testing the basic detection model; the first decision threshold corresponding to each feature verification model is obtained by jointly testing the basic detection model and the corresponding feature verification model.

[0031] Optionally, the feature comparison module is further configured to:

[0032] For at least one feature validation model, perform the following steps:

[0033] Acquire a first sample image set and a plurality of sample thresholds;

[0034] Using the first sample image set, jointly testing the basic detection model and a feature verification model to obtain an interception rate corresponding to the feature verification model under the multiple sample thresholds;

[0035] Among the interception rates obtained, the sample threshold corresponding to the third preset interception rate is used as the first decision threshold corresponding to the one feature verification model.

[0036] Optionally, the first decision threshold corresponding to each of the at least one auxiliary detection models is obtained by performing a joint test on the at least one auxiliary detection model.

[0037] Optionally, an update module is further included, and the update module is specifically configured to:

[0038] updating the first sample image set to obtain a second sample image set, wherein a distribution ratio of each sample image in the second sample image set is determined according to a distribution ratio of each original sample image in the first sample image set;

[0039] The second sample image set is used to update the first decision threshold corresponding to each of the at least one auxiliary detection model in the target scene, and the second decision threshold corresponding to the feature comparison model in the target scene.

[0040] Optionally, the update module is specifically configured to:

[0041] Traverse multiple newly added abnormal sample images in the target scene, and perform the following steps each time a newly added abnormal sample image is traversed:

[0042] Performing face recognition on the newly added abnormal sample image using the at least one auxiliary detection model and the feature comparison model, and if recognition is unsuccessful, adding the newly added abnormal sample image to the first sample image set;

[0043] The first sample image set obtained after the traversal is completed is used as the second sample image set.

[0044] Optionally, the basic detection model is a quality detection model, the at least one feature verification model includes a liveness detection model and a non-synthetic image detection model, and the at least one verification score includes a quality detection score, a liveness detection score, and a synthetic image detection score;

[0045] The verification module is specifically used for:

[0046] Inputting the face image to be recognized into the quality detection model to obtain a quality detection score of the face image to be recognized;

[0047] Inputting the facial features of the face image to be identified extracted by the quality detection model into a liveness detection model to obtain a liveness detection score of the face image to be identified;

[0048] The facial features are input into a non-synthetic image detection model to obtain a non-synthetic image detection score of the face image to be identified.

[0049] On the one hand, an embodiment of the present application provides a computer device, including a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor implements the steps of the above-mentioned face recognition method when executing the program.

[0050] On the one hand, an embodiment of the present application provides a computer-readable storage medium, which stores a computer program that can be executed by a computer device. When the program is run on the computer device, the computer device executes the steps of the above-mentioned face recognition method.

[0051] In an embodiment of the present application, the second decision threshold corresponding to the feature comparison model in the target scenario is obtained by jointly testing the feature comparison model and the auxiliary detection model, which fully considers the mutual influence between multiple algorithm models in the face recognition scenario using multiple algorithm models. Compared with manually setting the decision threshold of each algorithm model separately, the method in the present application is more adaptable to the scenarios of multi-model face recognition and the changes in the scenarios, thereby improving the accuracy and efficiency of the obtained decision threshold, and thus improving the accuracy of multi-model face recognition. BRIEF DESCRIPTION OF THE DRAWINGS

[0052] In order to more clearly illustrate the technical solutions in the embodiments of the present invention, the following briefly introduces the drawings required for use in the description of the embodiments. Obviously, the drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying any creative work.

[0053] Figure 1 A schematic diagram of a system architecture provided in an embodiment of the present application;

[0054] Figure 2 A flowchart of a face recognition method provided in an embodiment of the present application;

[0055] Figure 3 A schematic diagram of a process for obtaining a second decision threshold provided in an embodiment of the present application;

[0056] Figure 4 A flowchart of a face recognition method provided in an embodiment of the present application;

[0057] Figure 5 A schematic diagram of a process for updating a first sample image set provided in an embodiment of the present application;

[0058] Figure 6 A schematic diagram of the structure of a face recognition device provided in an embodiment of the present application;

[0059] Figure 7 A schematic diagram of the structure of a computer device provided in an embodiment of the present application. DETAILED DESCRIPTION

[0060] In order to make the purpose, technical solutions and beneficial effects of the present invention more clearly understood, the present invention will be further described in detail below with reference to the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are only used to explain the present invention and are not intended to limit the present invention.

[0061] refer to Figure 1, which is a system architecture diagram applicable to an embodiment of the present application. The system architecture at least includes a terminal device 101 and a face recognition system 102.

[0062] The terminal device 101 is installed with a target application for face recognition, which may be a pre-installed client, a web application, or a small program embedded in other applications. The terminal device 101 may be a smart phone, a tablet computer, a laptop computer, a desktop computer, etc., but is not limited thereto.

[0063] Facial recognition system 102 is the backend server for the target application, providing services for the target application. Facial recognition system 102 can be a standalone physical server, a server cluster or distributed system consisting 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, content delivery networks (CDNs), and big data and artificial intelligence platforms.

[0064] The terminal device 101 and the face recognition system 102 can be connected directly or indirectly via wired or wireless communication, and this application does not impose any restrictions on this.

[0065] The terminal device 101 collects the face image to be recognized in the target scene, and sends the face image to be recognized in the target scene to the face recognition system 102.

[0066] The face recognition system 102 receives a face image to be recognized, uses at least one auxiliary detection model to verify the face image to be recognized, and obtains at least one verification score. If at least one verification score is respectively greater than the first decision threshold corresponding to at least one auxiliary detection model in the target scene, a feature comparison model is used to perform feature comparison on the face image to be recognized, and a target comparison score is obtained. If the target comparison score is greater than the second decision threshold corresponding to the feature comparison model in the target scene, it is determined that the face image to be recognized is successfully recognized. The second decision threshold is obtained by jointly testing the feature comparison model and the auxiliary detection model. In practical applications, the solution in the embodiment of the present application can be applied to any scenario where face recognition is required.

[0067] based on Figure 1 The system architecture diagram, the embodiment of the present application provides a process of a face recognition method, such as Figure 2 As shown, the process of this method is Figure 1 The facial recognition system 102 shown in FIG. 1 includes the following steps:

[0068] Step S201: Acquire a face image to be recognized in a target scene.

[0069] Specifically, this application includes multiple different business scenarios and multiple different algorithm models. The business scenarios include liveness risk scenarios, hijacking risk scenarios, identification risk scenarios, etc. The algorithm models include quality detection models, liveness detection models, non-synthetic image detection models, feature comparison models, etc.

[0070] The target scenario can be any one of multiple business scenarios. Each business scenario corresponds to one or more scenario risk levels, and each algorithm model corresponds to one or more model risk levels. Each model risk level corresponds to a decision threshold, which is related to the interception rate obtained when testing the algorithm model.

[0071] In specific implementation, in business scenarios with different risk levels, multiple algorithm models can be used for face recognition, and each of the multiple algorithm models corresponds to a model risk level.

[0072] For example, consider the following business scenarios: liveness risk, hijacking risk, and identification risk. Algorithm models include quality detection, liveness detection, non-synthetic image detection, and feature comparison. Table 1 shows the model risk levels of the algorithm models corresponding to these business scenarios at different scenario risk levels.

[0073] Table 1.

[0074]

[0075]

[0076] Each algorithm model corresponds to different decision thresholds under different model risk levels, as shown in Table 2.

[0077] Table 2.

[0078]

[0079]

[0080] Step S202: Use at least one auxiliary detection model to verify the face image to be recognized and obtain at least one verification score.

[0081] In step S203, if at least one verification score is greater than the first decision threshold corresponding to at least one auxiliary detection model in the target scene, a feature comparison model is used to perform feature comparison on the face image to be identified to obtain a target comparison score.

[0082] Specifically, the first decision threshold corresponding to each of the at least one auxiliary detection models in the target scenario may be the same threshold or different thresholds. When the target scenario corresponds to multiple scenario risk levels, the first decision threshold is the decision threshold corresponding to one scenario risk level of the auxiliary detection model in the target scenario.

[0083] For example, the auxiliary detection model is set to include a quality detection model, a living body detection model, and a non-synthetic image detection model.

[0084] As shown in Table 1, for the Level 1 scenario risk level in the liveness risk scenario, the model risk levels of the quality detection model, liveness detection model, and non-synthetic image detection model are Level 2, Level 2, and Level 1, respectively. Table 2 shows that the first decision threshold corresponding to a Level 2 quality detection model is 85 points, the first decision threshold corresponding to a Level 2 liveness detection model is 87 points, and the first decision threshold corresponding to a Level 1 non-synthetic image detection model is 86 points.

[0085] Specifically, the auxiliary models may be connected in series or in parallel.

[0086] When the auxiliary models are in a series structure, the auxiliary models at each level in series are used in turn to verify the face image to be identified. If the verification scores corresponding to the auxiliary models are greater than the first judgment threshold, the feature comparison model is used to perform feature comparison on the face image to be identified. If there is an auxiliary model whose verification score is less than or equal to the first judgment threshold during the verification process, the face recognition failure is directly output and the process ends.

[0087] When the auxiliary models are in a parallel structure, if the verification scores corresponding to the auxiliary models are greater than the first judgment threshold, the feature comparison model is used to perform feature comparison on the face image to be identified; if the verification score corresponding to any of the auxiliary models is less than the first judgment threshold, the face recognition failure is output and the process ends.

[0088] Step S204: If the target comparison score is greater than the second decision threshold corresponding to the feature comparison model in the target scene, it is determined that the recognition of the face image to be recognized is successful.

[0089] The second decision threshold is obtained by jointly testing the feature comparison model and the auxiliary detection model. When determining the second decision threshold, the feature comparison model can be jointly tested with one or more auxiliary detection models of the at least one auxiliary detection model.

[0090] If the target comparison score is less than or equal to the second decision threshold corresponding to the feature comparison model in the target scene, it is determined that the recognition of the face image to be recognized is unsuccessful, and the face image to be recognized is intercepted.

[0091] When the target scene corresponds to multiple scene risk levels, the second decision threshold is a decision threshold corresponding to a scene risk level of the feature comparison model under the target scene.

[0092] For example, the auxiliary detection models are set to include a quality detection model, a liveness detection model, and a non-synthetic image detection model. As shown in Table 1, for a Level 1 scenario risk level in a liveness risk scenario, the model risk levels of the quality detection model, liveness detection model, non-synthetic image detection model, and feature comparison model are Level 2, Level 2, Level 1, and Level 1, respectively. Table 2 shows that when the feature comparison model is Level 1, the corresponding second decision threshold is 78 points.

[0093] In an embodiment of the present application, the second decision threshold corresponding to the feature comparison model in the target scenario is obtained by jointly testing the feature comparison model and the auxiliary detection model, which fully considers the mutual influence between multiple algorithm models in the face recognition scenario using multiple algorithm models. Compared with manually setting the decision threshold of each algorithm model separately, the method in the present application is more adaptable to the scenarios of multi-model face recognition and the changes in the scenarios, thereby improving the accuracy and efficiency of the obtained decision threshold, and thus improving the accuracy of multi-model face recognition.

[0094] Optionally, in the above step S202, the embodiment of the present application adopts at least the following implementation methods to obtain the first decision threshold.

[0095] In one embodiment, the at least one auxiliary detection model includes a basic detection model and at least one feature verification model. The first decision threshold corresponding to the basic detection model is obtained by testing the basic detection model. The first decision threshold corresponding to each feature verification model is obtained by jointly testing the basic detection model and the corresponding feature verification model.

[0096] For the basic detection model, multiple sample thresholds are pre-set. The basic detection model is then tested using a first sample image set to obtain interception rates corresponding to the basic detection model at the multiple sample thresholds. Among the interception rates obtained, the sample threshold corresponding to the fourth preset interception rate is used as the first decision threshold corresponding to the basic detection model.

[0097] Specifically, the multiple sample thresholds may be set randomly or at preset intervals. By setting different fourth preset interception rates, the first decision thresholds of the basic detection model at different model risk levels may be obtained.

[0098] The first sample image set is used to test the basic capabilities of the model. The sample categories and distribution ratios in the first sample image set are determined based on the distribution of sample attributes in actual business scenarios. The first sample image set contains a rich variety of samples and a large sample size, ensuring the reliability of the test results. The sample attribute distribution in the first sample image set is close to that in actual business scenarios, ensuring that the test results are consistent with actual use.

[0099] For example, multiple sample thresholds are set for the basic detection model, namely 5 points, 15 points, 25 points, 35 points, 45 points, 55 points, 65 points, 75 points, 85 points, and 95 points. The basic detection model is tested using the first sample image set to obtain the corresponding interception rates of the basic detection model at each of the above sample thresholds, as shown in Table 3. The fourth preset interception rate is set to 99%, and Table 3 is used to query the fourth preset interception rate to obtain the first decision threshold corresponding to the basic detection model as 95 points.

[0100] Table 3.

[0101]

[0102]

[0103] In an embodiment of the present application, by setting multiple sample thresholds and determining the interception rate corresponding to the basic detection model at each sample threshold, and then directly selecting the sample threshold corresponding to the preset interception rate as the first decision threshold of the basic detection model, the efficiency of obtaining the decision threshold is improved.

[0104] Optionally, in order to obtain the first decision threshold corresponding to each feature verification model, the following steps are performed for each feature verification model:

[0105] A first sample image set and multiple sample thresholds are obtained. A basic detection model and a feature verification model are then jointly tested using the first sample image set to obtain an interception rate corresponding to the feature verification model under the multiple sample thresholds. For each interception rate obtained, the sample threshold corresponding to the third preset interception rate is used as a first decision threshold corresponding to the feature verification model.

[0106] Specifically, the multiple sample thresholds may be randomly set or may be set at preset intervals. By setting different third preset interception rates, the first decision threshold of the feature verification model at different model risk levels may be obtained.

[0107] One possible implementation method is to first determine the first decision threshold corresponding to the basic detection model, and then jointly test the basic detection model with each feature verification model to obtain the interception rate corresponding to each feature verification model under multiple sample thresholds.

[0108] For example, the first decision threshold of the basic detection model is set to 95 points. The at least one feature verification model includes a feature verification model 1 and a feature verification model 2.

[0109] Multiple sample thresholds were set for feature verification model 1, namely 5, 35, 65, 85, 95, and 98. Under the premise that the first decision threshold of the basic detection model was 95, the basic detection model and feature verification model 1 were jointly tested using the first sample image set to obtain the interception rates of feature verification model 1 at each of the above multiple sample thresholds, as shown in Table 4.

[0110] The third preset interception rate is set to 99%, and the third preset interception rate is used to query Table 4 to obtain the first decision threshold corresponding to feature verification model 1 as 98 points. The same method as above can be used to determine the first decision threshold corresponding to feature verification model 2, which will not be repeated here.

[0111] Table 4.

[0112] Sample threshold (points) Interception rate 5 10％ 35 45％ 65 72％ 85 90％ 95 98％ 98 99％

[0113] In an embodiment of the present application, the first decision threshold of the feature verification model is obtained by jointly testing the basic detection model and the feature verification model, fully considering the correlation between multiple algorithm models in the multi-model face recognition scenario, thereby improving the accuracy of the first decision threshold.

[0114] Another possible implementation involves setting a test order for at least one feature verification model and determining a first decision threshold corresponding to the basic detection model. The feature verification models are tested sequentially according to the test order. For each feature verification model, the basic detection model, the feature verification model, and the previously tested feature verification model are jointly tested to obtain the interception rate corresponding to the feature verification model at multiple sample thresholds.

[0115] For example, the first decision threshold corresponding to the basic detection model is set to 95 points. The at least one feature verification model includes feature verification model 1 and feature verification model 2. The test order of feature verification model 1 and feature verification model 2 is to test feature verification model 1 first and then test feature verification model 2.

[0116] Multiple sample thresholds were set for Feature Verification Model 1, namely 5, 35, 65, 85, 95, and 98. Under the premise that the first decision threshold corresponding to the basic detection model was 95, the basic detection model and Feature Verification Model 1 were jointly tested using the first sample image set to obtain the interception rates corresponding to Feature Verification Model 1 at each of the above sample thresholds, as shown in Table 5.

[0117] The third preset interception rate is set to 99%, and the third preset interception rate is used to query Table 5 to obtain a first decision threshold corresponding to the feature calculation model 1 of 98 points.

[0118] Table 5.

[0119]

[0120]

[0121] Multiple sample thresholds were set for feature verification model 2, namely 25, 45, 65, 85, 95, and 99. When the first decision threshold corresponding to the basic detection model was 95 and the first decision threshold corresponding to feature verification model 1 was 98, the first sample image set was used to jointly test the basic detection model, feature verification model 1, and feature verification model 2. The interception rates corresponding to feature verification model 2 at each of the above sample thresholds were obtained, as shown in Table 6.

[0122] The third preset interception rate is set to 99%, and the third preset interception rate is used to query Table 6 to obtain a first decision threshold corresponding to the feature calculation model 2 of 99 points.

[0123] Table 6.

[0124] Sample threshold (points) Interception rate 25 30％ 45 52％ 65 72％ 85 92％ 95 98％ 99 99％

[0125] In an embodiment of the present application, the first decision threshold of the feature verification model is obtained by jointly testing the basic detection model, the feature verification model, and the tested feature verification model. It not only considers the correlation between the basic detection model and the feature verification model in the multi-model face recognition scenario, but also considers the correlation between multiple feature verification models, thereby improving the accuracy of the first decision threshold.

[0126] In the second embodiment, the first decision threshold corresponding to each of the at least one auxiliary detection models is obtained by performing a joint test on the at least one auxiliary detection model.

[0127] In one possible implementation, multiple sample thresholds are set for each auxiliary detection model. The multiple sample thresholds may be set randomly or at preset intervals.

[0128] A first set of sample images is used to jointly test the auxiliary detection models to obtain the interception rate corresponding to each auxiliary detection model under multiple sample thresholds. For each auxiliary detection model, the sample threshold corresponding to the fifth preset interception rate of the auxiliary detection model is used as the first decision threshold corresponding to the auxiliary detection model.

[0129] For example, the at least one auxiliary detection model includes auxiliary detection model 1 and auxiliary detection model 2.

[0130] Multiple sample thresholds were set for auxiliary detection model 1 and auxiliary detection model 2, namely 5, 35, 65, 85, 95, and 98 points. Auxiliary detection model 1 and auxiliary detection model 2 were jointly tested using the first sample image set. The corresponding interception rates of auxiliary detection model 1 and auxiliary detection model 2 at each of the above sample thresholds were obtained, as shown in Table 7.

[0131] The fifth preset interception rate corresponding to auxiliary detection model 1 is set to 99%, and the fifth preset interception rate corresponding to auxiliary detection model 2 is set to 98%. Using the fifth preset interception rate corresponding to auxiliary detection model 1, query Table 7 to obtain the first decision threshold corresponding to feature calculation model 1 as 98 points. Using the fifth preset interception rate corresponding to auxiliary detection model 2, query Table 7 to obtain the first decision threshold corresponding to feature calculation model 2 as 95 points.

[0132] Table 7.

[0133]

[0134] In an embodiment of the present application, a first decision threshold of the auxiliary detection model is obtained by jointly testing multiple auxiliary detection models, and the correlation between multiple auxiliary detection models in the multi-model face recognition scenario is taken into account, thereby improving the accuracy of the first decision threshold.

[0135] Optionally, in the above step S204, the embodiment of the present application adopts at least the following implementation methods to obtain the second decision threshold.

[0136] Implementation method 1: The second decision threshold is obtained by the following method: Figure 3 As shown, the following steps are included:

[0137] Step S301: Using the first sample image set, the auxiliary detection model and the feature comparison model are jointly tested to obtain the predicted comparison score corresponding to each original sample image in the first sample image set.

[0138] Specifically, the auxiliary detection model can be the basic detection model or feature verification model described above.

[0139] Obtain a first sample image set and a first preset interception rate. For an original sample image from the first sample image set, input the original sample image into the auxiliary detection model and the feature comparison model for joint testing to obtain similarity scores between the original sample image and each reference face image in the face recognition library. The highest similarity score is used as the predicted comparison score for the original sample image.

[0140] For other original sample images in the first sample image set, the same steps as above are adopted to obtain the predicted comparison scores corresponding to the original sample images.

[0141] Step S302 : Based on a first preset interception rate, at least one valid sample image is screened out from the first sample image set.

[0142] Specifically, a first preset interception rate of the feature comparison model is multiplied by the number of original sample images in the first sample image set to obtain a target screening number. The original sample images in the first sample image set are sorted in descending order of predicted comparison scores to obtain a sample image sorting result. The sample image sorting result is then used to screen the target screening number of valid sample images from the sample image set.

[0143] Step S303: Determine a second decision threshold based on the predicted comparison score corresponding to at least one valid sample image.

[0144] Specifically, the minimum predicted comparison score among the predicted comparison scores corresponding to the at least one valid sample image may be used as the second decision threshold. Alternatively, another predicted comparison score among the predicted comparison scores corresponding to the at least one valid sample image may be selected as the second decision threshold.

[0145] By setting different first preset interception rates, the second decision threshold of the feature comparison model at different model risk levels can be obtained.

[0146] For example, the first sample image set includes 5,000 original sample images, and the first preset interception rate of the feature comparison model is set to 0.1%.

[0147] For each original sample image, the auxiliary detection model and feature comparison model are jointly tested to obtain similarity scores between the original sample image and each reference face image in the face recognition library. The highest similarity score is used as the predicted comparison score for the original sample image. This process is repeated until all 5,000 original sample images have corresponding predicted comparison scores.

[0148] Multiply the first preset interception rate by the number of images in the first sample image set to obtain a target screening number of 5. Sort the 5,000 original sample images in descending order of predicted comparison scores, and use the top 5 original sample images as valid sample images, as shown in Table 8.

[0149] As shown in Table 8, among the prediction comparison scores of each valid sample image, the minimum prediction comparison score is 90 points, and 90 points is used as the second decision threshold of the feature comparison model.

[0150] Table 8.

[0151] Original sample image Prediction comparison score (points) Original sample image 120 98 Original sample image 29 96 Original sample image 43 95 Original sample image 40 92 Original sample image 58 90

[0152] Implementation method 2: Obtain the second decision threshold using the following method:

[0153] Obtain a first set of sample images and multiple sample thresholds. Then, using the first set of sample images, jointly test the auxiliary detection model and the feature comparison model to obtain interception rates corresponding to the feature comparison model at the multiple sample thresholds. For each interception rate obtained, the sample threshold corresponding to the second preset interception rate is used as the second decision threshold.

[0154] Specifically, the multiple sample thresholds may be set randomly or according to preset intervals.

[0155] Using a first set of sample images, the first decision threshold corresponding to the auxiliary detection model is first determined. The auxiliary detection model and the feature comparison model are then jointly tested to obtain the interception rate corresponding to the feature comparison model at multiple sample thresholds. For each interception rate obtained, the sample threshold corresponding to the second preset interception rate is used as the second decision threshold. By setting different second preset interception rates, the second decision threshold of the feature comparison model can be obtained at different model risk levels.

[0156] For example, the first decision threshold of the auxiliary detection model is set to 98 points, and multiple sample thresholds are set for the feature comparison model at 10 points, 30 points, 60 points, 90 points, 92 points, 94 points, 96 points, and 98 points, respectively. Under the premise that the first decision threshold of the auxiliary detection model is 98 points, the auxiliary detection model and the feature comparison model are jointly tested using the first sample image set to obtain the interception rates of the feature comparison model at each of the above multiple sample thresholds, as shown in Table 9.

[0157] The second preset interception rate is set to 99%, and the second preset interception rate is used to query Table 9 to obtain a second decision threshold corresponding to the feature comparison model of 98 points.

[0158] Table 9.

[0159] Sample threshold (points) Interception rate 10 10％ 30 45％ 60 72％ 90 93％ 92 95％ 94 96％ 96 97％ 98 99％

[0160] Optionally, the at least one auxiliary detection model includes a basic detection model and at least one feature verification model. The basic detection model is a quality detection model, and the at least one feature verification model includes a liveness detection model and a non-synthetic image detection model.

[0161] See also Figure 4 , the embodiment of the present application provides a face recognition method, the process of the method is as follows Figure 1The facial recognition system 102 shown in FIG. 1 includes the following steps:

[0162] Step S401: Obtain a face image to be recognized.

[0163] Step S402: Input the face image to be recognized into a quality detection model to obtain a quality detection score of the face image to be recognized.

[0164] Specifically, the face image to be recognized is input into the quality detection model, the quality detection model extracts the facial features of the face image to be recognized, and based on the facial features, predicts the quality detection score of the face image to be recognized.

[0165] Step S403, determine whether the quality detection score is greater than the first decision threshold corresponding to the quality detection model. If so, execute step S404; otherwise, execute step S410 and end.

[0166] Step S404: input the facial features into a liveness detection model to obtain a liveness detection score for the face image to be identified.

[0167] Step S405 , determining whether the liveness detection score is greater than the first decision threshold corresponding to the liveness detection model. If so, executing step S406 ; otherwise, executing step S410 and ending.

[0168] Step S406: input the facial features into the non-synthetic image detection model to obtain the non-synthetic image detection score of the face image to be recognized.

[0169] Step S407, determine whether the non-synthetic image detection score is greater than the first decision threshold corresponding to the non-synthetic image detection model. If so, execute step S408; otherwise, execute step S410 and end.

[0170] Step S408: input the facial features into a feature comparison model to obtain a feature comparison score of the facial image to be identified.

[0171] Step S409, determine whether the feature comparison score is greater than the second decision threshold corresponding to the feature comparison model. If so, output recognition success and end; otherwise, execute step S410 and end.

[0172] Step S410: Output recognition failure.

[0173] Step S411: Output recognition success.

[0174] In the embodiments of the present application, the facial image to be recognized is sequentially input into a quality detection model, a liveness detection model, a non-synthetic image detection model, and a feature comparison model, and relevant threshold judgments are performed, thereby improving the accuracy of face recognition. Because the decision thresholds corresponding to each model are not set independently but are instead obtained through joint testing of the models, the mutual influence between the models is fully considered, thereby improving the accuracy of the obtained decision thresholds and, in turn, improving the accuracy of face recognition.

[0175] Optionally, the embodiment of the present application updates the first decision threshold of the auxiliary detection model and the second decision threshold of the feature comparison model in at least the following manner, including the following steps:

[0176] The first sample image set is updated to obtain a second sample image set. The second sample image set is used to update a first decision threshold corresponding to each of at least one auxiliary detection model in the target scene, and a second decision threshold corresponding to the feature comparison model in the target scene.

[0177] Specifically, the distribution ratio of each sample image in the second sample image set is determined based on the distribution ratio of each original sample image in the first sample image set. The distribution ratio is the ratio of the number of newly added abnormal sample images to the total number of images in the first sample image set.

[0178] At least one auxiliary detection model and feature comparison model are used to perform face recognition on a newly added abnormal sample image. If recognition is unsuccessful, the newly added abnormal sample image is added to the first sample image set. The first sample image set obtained after the traversal is completed is used as the second sample image set.

[0179] For example, if Figure 5 As shown, updating the first sample image set to obtain the second sample image set includes the following steps:

[0180] Step S501, set i=1.

[0181] Step S502 , determining whether i is less than or equal to the total number of newly added abnormal sample images in the newly added abnormal sample library, if so, executing step S503 ; otherwise, the execution ends.

[0182] Step S503: Obtain the i-th newly added abnormal sample image from the newly added abnormal sample library.

[0183] Step S504: input the newly added abnormal sample image into the quality detection model, the quality detection model extracts the target image features of the newly added abnormal sample image, and predicts the quality detection score of the newly added abnormal sample image based on the target image features.

[0184] Step S505 , determining whether the quality detection score is greater than a first decision threshold corresponding to the quality detection model; if so, executing step S506 ; otherwise, executing step S512 .

[0185] Step S506: Input the target image features into the liveness detection model to obtain the liveness detection score of the newly added abnormal sample image.

[0186] Step S507, determine whether the liveness detection score is greater than the first decision threshold corresponding to the liveness detection model. If so, execute step S508; otherwise, execute step S512.

[0187] Step S508: Input the target image features into the non-synthetic image detection model to obtain the non-synthetic image detection score of the newly added abnormal sample image.

[0188] Step S509 , determining whether the non-synthetic image detection score is greater than a first decision threshold corresponding to the non-synthetic image detection model; if so, executing step S510 ; otherwise, executing step S512 .

[0189] Step S510: Input the target image features into a feature comparison model to obtain a feature comparison score of the newly added abnormal sample image.

[0190] Step S511, determine whether the feature comparison score is greater than the second decision threshold corresponding to the feature comparison model. If so, execute step S514; otherwise, execute step S512.

[0191] Step S512: taking the ratio of the number of newly added abnormal sample images to the total number of images in the first sample image set as the distribution ratio.

[0192] Step S513, determine whether the distribution ratio is less than a preset value, if so, execute step S514; otherwise, end.

[0193] Step S514: adding the i-th newly added abnormal sample image to the first sample image set.

[0194] Step S515, set i=i+1, and execute step S502.

[0195] The first sample image set obtained after the above steps is completed is the second sample image set.

[0196] In the embodiments of the present application, by adding newly added abnormal sample images to the first sample image set, the first sample image set is updated, and newly added abnormal sample images from different target scenarios are continuously absorbed. This improves the versatility of the face recognition method of the present application and enhances the ability to prevent and control various risks. Since newly added abnormal sample images are added to the first sample image set according to the distribution ratio, the stability of the first sample image set is guaranteed.

[0197] Based on the same technical concept, the embodiment of the present application provides a face recognition device, such as Figure 6 As shown, the apparatus 600 includes:

[0198] An acquisition module 601 is used to acquire a face image to be recognized in a target scene;

[0199] A verification module 602 is configured to verify the face image to be recognized using at least one auxiliary detection model to obtain at least one verification score;

[0200] The feature comparison module 603 uses the feature comparison model to perform feature comparison on the face image to be identified to obtain a target comparison score if the at least one verification score is respectively greater than the first decision threshold corresponding to each of the at least one auxiliary detection models in the target scene.

[0201] Identification module 604, if the target comparison score is greater than the second judgment threshold corresponding to the feature comparison model in the target scene, it is determined that the recognition of the face image to be identified is successful, and the second judgment threshold is obtained by jointly testing the feature comparison model and the auxiliary detection model.

[0202] Optionally, the identification module 604 is specifically configured to:

[0203] Using a first sample image set, jointly testing the auxiliary detection model and the feature comparison model to obtain a predicted comparison score corresponding to each original sample image in the first sample image set;

[0204] Based on a first preset interception rate, screening out at least one valid sample image from the first sample image set;

[0205] The second decision threshold is determined based on the predicted comparison score corresponding to each of the at least one valid sample images.

[0206] Optionally, the identification module 604 is specifically configured to:

[0207] multiplying the first preset interception rate by the number of original sample images in the first sample image set to obtain a target screening number;

[0208] Sorting the original sample images in the first sample image set in descending order of the predicted comparison scores to obtain a sample image sorting result;

[0209] According to the sample image sorting result, the target screening number of valid sample images are screened out from the sample image set.

[0210] Optionally, the identification module 604 is specifically configured to:

[0211] The minimum predicted comparison score among the predicted comparison scores corresponding to the at least one valid sample image is used as the second decision threshold.

[0212] Optionally, the identification module 604 is specifically configured to:

[0213] Acquire a first sample image set and a plurality of sample thresholds;

[0214] Using the first sample image set, jointly testing the auxiliary detection model and the feature comparison model to obtain an interception rate corresponding to the feature comparison model under the multiple sample thresholds;

[0215] Among the obtained interception rates, the sample threshold corresponding to the second preset interception rate is used as the second decision threshold.

[0216] Optionally, the at least one auxiliary detection model includes a basic detection model and at least one feature verification model;

[0217] Among them, the first decision threshold corresponding to the basic detection model is obtained by testing the basic detection model; the first decision threshold corresponding to each feature verification model is obtained by jointly testing the basic detection model and the corresponding feature verification model.

[0218] Optionally, the feature comparison module 603 is further configured to:

[0219] For at least one feature validation model, perform the following steps:

[0220] Acquire a first sample image set and a plurality of sample thresholds;

[0221] Using the first sample image set, jointly testing the basic detection model and a feature verification model to obtain an interception rate corresponding to the feature verification model under the multiple sample thresholds;

[0222] Among the interception rates obtained, the sample threshold corresponding to the third preset interception rate is used as the first decision threshold corresponding to the one feature verification model.

[0223] Optionally, the first decision threshold corresponding to each of the at least one auxiliary detection models is obtained by performing a joint test on the at least one auxiliary detection model.

[0224] Optionally, an update module 605 is further included, and the update module 605 is specifically configured to:

[0225] updating the first sample image set to obtain a second sample image set, wherein a distribution ratio of each sample image in the second sample image set is determined according to a distribution ratio of each original sample image in the first sample image set;

[0226] The second sample image set is used to update the first decision threshold corresponding to each of the at least one auxiliary detection model in the target scene, and the second decision threshold corresponding to the feature comparison model in the target scene.

[0227] Optionally, the updating module 605 is specifically configured to:

[0228] Traverse multiple newly added abnormal sample images in the target scene, and perform the following steps each time a newly added abnormal sample image is traversed:

[0229] Performing face recognition on the newly added abnormal sample image using the at least one auxiliary detection model and the feature comparison model, and if recognition is unsuccessful, adding the newly added abnormal sample image to the first sample image set;

[0230] The first sample image set obtained after the traversal is completed is used as the second sample image set.

[0231] Optionally, the basic detection model is a quality detection model, the at least one feature verification model includes a liveness detection model and a non-synthetic image detection model, and the at least one verification score includes a quality detection score, a liveness detection score, and a synthetic image detection score;

[0232] The verification module 602 is specifically used to:

[0233] Inputting the face image to be recognized into the quality detection model to obtain a quality detection score of the face image to be recognized;

[0234] Inputting the facial features of the face image to be identified extracted by the quality detection model into a liveness detection model to obtain a liveness detection score of the face image to be identified;

[0235] The facial features are input into a non-synthetic image detection model to obtain a non-synthetic image detection score of the face image to be identified.

[0236] Based on the same technical concept, the embodiment of the present application provides a computer device, which can be a terminal or a server, such as Figure 7 As shown, it includes at least one processor 701 and a memory 702 connected to the at least one processor. The specific connection medium between the processor 701 and the memory 702 is not limited in the embodiment of the present application. Figure 7For example, the processor 701 and the memory 702 are connected via a bus. The bus can be divided into an address bus, a data bus, a control bus, and the like.

[0237] In an embodiment of the present application, the memory 702 stores instructions that can be executed by at least one processor 701. The at least one processor 701 can execute the steps included in the above-mentioned face recognition method by executing the instructions stored in the memory 702.

[0238] The processor 701 is the control center of the computer device. It can use various interfaces and lines to connect various parts of the computer device and perform face recognition by running or executing instructions stored in the memory 702 and calling data stored in the memory 702. Optionally, the processor 701 may include one or more processing units. The processor 701 may integrate an application processor and a modem processor. The application processor mainly processes the operating system, user interface, and application programs, and the modem processor mainly processes wireless communications. It is understandable that the above-mentioned modem processor may not be integrated into the processor 701. In some embodiments, the processor 701 and the memory 702 may be implemented on the same chip. In some embodiments, they may also be implemented separately on separate chips.

[0239] The processor 701 can be a general-purpose processor, such as a central processing unit (CPU), a digital signal processor, an application-specific integrated circuit (ASIC), a field programmable gate array or other programmable logic device, a discrete gate or transistor logic device, or a discrete hardware component, and can implement or execute the various methods, steps, and logic block diagrams disclosed in the embodiments of the present application. A general-purpose processor can be a microprocessor or any conventional processor. The steps of the method disclosed in the embodiments of the present application can be directly embodied as being executed by a hardware processor, or can be executed by a combination of hardware and software modules in the processor.

[0240] The memory 702 is a non-volatile computer-readable storage medium that can be used to store non-volatile software programs, non-volatile computer executable programs and modules. The memory 702 may include at least one type of storage medium, such as a flash memory, a hard disk, a multimedia card, a card-type memory, a random access memory (Random Access Memory, RAM), a static random access memory (Static Random Access Memory, SRAM), a programmable read-only memory (Programmable Read Only Memory, PROM), a read-only memory (Read Only Memory, ROM), an electrically erasable programmable read-only memory (Electrically Erasable Programmable Read-Only Memory, EEPROM), a magnetic memory, a disk, an optical disk, etc. The memory 702 is any other medium that can be used to carry or store a desired program code in the form of an instruction or data structure and can be accessed by a computer, but is not limited thereto. The memory 702 in the embodiment of the present application can also be a circuit or any other device that can realize a storage function, for storing program instructions and / or data.

[0241] Based on the same inventive concept, an embodiment of the present application provides a computer-readable storage medium, which stores a computer program that can be executed by a computer device. When the program runs on the computer device, the computer device executes the steps of the above-mentioned face recognition method.

[0242] Those skilled in the art will appreciate that the embodiments of the present application can be provided as methods, systems, or computer program products. Therefore, the present application can adopt the form of a complete hardware embodiment, a complete software embodiment, or an embodiment in combination with software and hardware. Moreover, the present application can adopt the form of a computer program product implemented on one or more computer-usable storage media (including but not limited to magnetic disk storage, CD-ROM, optical storage, etc.) that contain computer-usable program code.

[0243] The present application is described with reference to the flowcharts and / or block diagrams of the methods, devices (systems), and computer program products according to the present application. It should be understood that each process and / or block in the flowchart and / or block diagram, as well as the combination of processes and / or blocks in the flowchart and / or block diagram, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, a special-purpose computer, an embedded processor, or other programmable data processing device to produce a machine, so that the instructions executed by the processor of the computer or other programmable data processing device generate instructions for implementing the processes in the flowchart and / or block diagram. Figure 1 a process or multiple processes and / or boxes Figure 1 A device that provides the functions specified in a block or multiple blocks.

[0244] These computer program instructions may also be stored in a computer readable memory that can direct a computer or other programmable data processing device to work in a specific manner, so that the instructions stored in the computer readable memory produce an article of manufacture comprising an instruction device, which implements the process Figure 1 a process or multiple processes and / or boxes Figure 1 The function specified in one or more boxes.

[0245] These computer program instructions can also be loaded onto a computer or other programmable data processing device so that a series of operational steps are executed on the computer or other programmable device to produce a computer-implemented process, thereby providing the instructions executed on the computer or other programmable device for implementing the process. Figure 1 a process or multiple processes and / or boxes Figure 1 A step that specifies a function in one or more boxes.

[0246] Obviously, those skilled in the art may make various changes and modifications to this application without departing from the spirit and scope of this application. Thus, if these modifications and variations of this application fall within the scope of the claims of this application and their equivalents, this application is intended to include these modifications and variations.

Claims

1. A face recognition method, characterized in that: include: Acquire a face image to be recognized in a target scene; the target scene corresponds to one or more scene risk levels; Using at least one auxiliary detection model to verify the face image to be recognized, and obtain at least one verification score; If the at least one verification score is respectively greater than the first decision threshold corresponding to each of the at least one auxiliary detection models in the target scene, then using the feature comparison model to perform feature comparison on the face image to be identified to obtain a target comparison score; If the target comparison score is greater than a second decision threshold corresponding to the feature comparison model in the target scenario, it is determined that the recognition of the face image to be identified is successful, and the second decision threshold is dynamically determined by jointly testing the feature comparison model and the auxiliary detection model and combining it with a preset interception rate; wherein each algorithm model corresponds to one or more model risk levels, and each algorithm model risk level corresponds to a decision threshold; The second decision threshold is determined dynamically by jointly testing the feature comparison model and the auxiliary detection model and combining a preset interception rate, including: Using a first sample image set, jointly testing the auxiliary detection model and the feature comparison model to obtain a predicted comparison score corresponding to each original sample image in the first sample image set; Based on a first preset interception rate, screening out at least one valid sample image from the first sample image set; The second decision threshold is determined based on the predicted comparison score corresponding to each of the at least one valid sample images.

2. The method according to claim 1, wherein The step of selecting at least one valid sample image from the first sample image set based on the first preset interception rate includes: multiplying the first preset interception rate by the number of original sample images in the first sample image set to obtain a target screening number; Sorting the original sample images in the first sample image set in descending order of the predicted comparison scores to obtain a sample image sorting result; According to the sample image sorting result, the target screening number of valid sample images are screened out from the sample image set.

3. The method according to claim 1, wherein The determining the second decision threshold based on the predicted comparison score corresponding to each of the at least one valid sample images includes: The minimum predicted comparison score among the predicted comparison scores corresponding to the at least one valid sample image is used as the second decision threshold.

4. The method according to claim 1, wherein The second decision threshold is determined dynamically by jointly testing the feature comparison model and the auxiliary detection model and combining a preset interception rate, including: Acquire a first sample image set and a plurality of sample thresholds; Using the first sample image set, jointly testing the auxiliary detection model and the feature comparison model to obtain an interception rate corresponding to the feature comparison model under the multiple sample thresholds; Among the obtained interception rates, the sample threshold corresponding to the second preset interception rate is used as the second decision threshold.

5. The method according to claim 1, wherein The at least one auxiliary detection model includes a basic detection model and at least one feature verification model; Among them, the first decision threshold corresponding to the basic detection model is obtained by testing the basic detection model; the first decision threshold corresponding to each feature verification model is obtained by jointly testing the basic detection model and the corresponding feature verification model.

6. The method according to claim 5, wherein The first decision threshold corresponding to each feature verification model is obtained by jointly testing the basic detection model and the corresponding feature verification model, including: For at least one feature validation model, perform the following steps: Acquire a first sample image set and a plurality of sample thresholds; Using the first sample image set, jointly testing the basic detection model and a feature verification model to obtain an interception rate corresponding to the feature verification model under the multiple sample thresholds; Among the interception rates obtained, the sample threshold corresponding to the third preset interception rate is used as the first decision threshold corresponding to the one feature verification model.

7. The method according to claim 1, wherein The first decision threshold corresponding to each of the at least one auxiliary detection models is obtained by performing a joint test on the at least one auxiliary detection model.

8. The method according to any one of claims 1 to 6, characterized in that: Also includes: updating the first sample image set to obtain a second sample image set, wherein a distribution ratio of each sample image in the second sample image set is determined according to a distribution ratio of each original sample image in the first sample image set; The second sample image set is used to update the first decision threshold corresponding to each of the at least one auxiliary detection model in the target scene, and the second decision threshold corresponding to the feature comparison model in the target scene.

9. The method according to claim 8, wherein The updating of the first sample image set to obtain a second sample image set includes: Traverse multiple newly added abnormal sample images in the target scene, and perform the following steps each time a newly added abnormal sample image is traversed: Performing face recognition on the newly added abnormal sample image using the at least one auxiliary detection model and the feature comparison model, and if recognition is unsuccessful, adding the newly added abnormal sample image to the first sample image set; The first sample image set obtained after the traversal is completed is used as the second sample image set.

10. The method according to claim 5, wherein The basic detection model is a quality detection model, the at least one feature verification model includes a liveness detection model and a non-synthetic image detection model, and the at least one verification score includes a quality detection score, a liveness detection score, and a synthetic image detection score; The adopting at least one auxiliary detection model to verify the face image to be recognized and obtaining at least one verification score includes: Inputting the face image to be recognized into the quality detection model to obtain a quality detection score of the face image to be recognized; Inputting the facial features of the face image to be identified extracted by the quality detection model into a liveness detection model to obtain a liveness detection score of the face image to be identified; The facial features are input into a non-synthetic image detection model to obtain a non-synthetic image detection score of the face image to be identified.

11. A face recognition device, characterized in that: include: An acquisition module is used to acquire a face image to be identified in a target scene; the target scene corresponds to one or more scene risk levels; a verification module, configured to verify the face image to be identified using at least one auxiliary detection model to obtain at least one verification score; A feature comparison module, if the at least one verification score is respectively greater than the first decision threshold corresponding to each of the at least one auxiliary detection model in the target scene, performs feature comparison on the face image to be identified using the feature comparison model to obtain a target comparison score; The recognition module determines that the recognition of the face image to be recognized is successful if the target comparison score is greater than a second decision threshold corresponding to the feature comparison model in the target scenario, wherein the second decision threshold is dynamically determined by jointly testing the feature comparison model and the auxiliary detection model and combining it with a preset interception rate; wherein each algorithm model corresponds to one or more model risk levels, and each algorithm model risk level corresponds to a decision threshold; The recognition module is specifically configured to use a first sample image set to jointly test the auxiliary detection model and the feature comparison model to obtain a predicted comparison score corresponding to each original sample image in the first sample image set; Based on a first preset interception rate, screening out at least one valid sample image from the first sample image set; The second decision threshold is determined based on the predicted comparison score corresponding to each of the at least one valid sample images.

12. A computer device comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein: When the processor executes the program, the steps of the method according to any one of claims 1 to 10 are implemented.

13. A computer-readable storage medium, characterized in that It stores a computer program that can be executed by a computer device. When the program is run on the computer device, the computer device executes the steps of any one of the methods according to claims 1 to 10.

14. A computer program product, characterized in that It stores a computer program that can be executed by a computer device. When the program is run on the computer device, the computer device executes the steps of any one of the methods according to claims 1 to 10.

Images