Face recognition method and device, computer equipment and storage medium

Abstract

The embodiment of the invention provides a face recognition method and device, computer equipment and a storage medium, and belongs to the technical field of financial science and technology. The method comprises the following steps: acquiring face image sample data containing a sample object and face identification information of the face image sample data; the method comprises the following steps: inputting face image sample data into a candidate model group comprising at least two different selected face recognition models; performing feature extraction on the face image sample data through a selected face recognition model to obtain at least two different initial face features; performing fusion processing on the at least two different initial face features to obtain a face fusion feature; performing feature fusion evaluation on the face fusion features according to the face identification information to obtain feature fusion evaluation data of the candidate model group; screening out a target model group from the candidate model group according to the feature fusion evaluation data; and performing face recognition on the target face image data through the target model group. According to the embodiment of the invention, the face recognition precision can be improved.

Description

Technical Field

[0001] The present application relates to the field of financial technology, and in particular to a face recognition method and apparatus, computer equipment and storage medium. Background Art

[0002] Face recognition technology is a technology that uses computer vision technology to identify the identity or attributes of a face from an image or video. Face recognition technology is applied to various fields, and the requirements for face recognition accuracy are getting higher and higher. For example, in the financial field, face recognition is mainly used for user identity authentication, so the accuracy of face recognition affects the security of identity authentication. However, during the face recognition process, the face will be affected by factors such as lighting, posture, occlusion, and expression, which will affect the accuracy of recognition. Therefore, how to improve the accuracy of face recognition has become a technical problem that needs to be solved urgently. Summary of the invention

[0003] The main purpose of the embodiments of the present application is to propose a face recognition method and apparatus, a computer device and a storage medium, aiming to improve the accuracy of face recognition.

[0004] To achieve the above-mentioned purpose, a first aspect of an embodiment of the present application proposes a face recognition method, the method comprising:

[0005] Acquire a face image sample set; wherein the face image sample set includes: face image sample data containing sample objects and face identification information of the face image sample data, wherein the face identification information is used to identify whether the sample objects contained in the face image sample data are the same or different;

[0006] By inputting the face image sample data into a preset candidate model group; wherein the candidate model group includes at least two different selected face recognition models;

[0007] Performing feature extraction on the face image sample data by using each of the selected face recognition models to obtain at least two different preliminary face features;

[0008] Fusing at least two different preliminary facial features to obtain a facial fusion feature of each candidate model group;

[0009] Performing feature fusion evaluation on the face fusion feature according to the face identification information to obtain feature fusion evaluation data of the candidate model group; wherein the feature fusion evaluation data represents the recognition accuracy of the face fusion feature;

[0010] Screening out a target model group from the candidate model group according to the feature fusion evaluation data;

[0011] Face recognition is performed on the target face image data through the target model group.

[0012] In some embodiments, the step of fusing at least two different preliminary facial features to obtain a facial fusion feature of each candidate model group includes:

[0013] Filtering target dimension information from preset candidate dimension information;

[0014] Performing dimensionality reduction processing on each of the preliminary facial features according to the target dimension information to obtain a low-dimensional facial feature;

[0015] At least two different low-dimensional facial features are fused to obtain the facial fusion feature.

[0016] In some embodiments, fusing at least two different low-dimensional facial features to obtain the facial fusion feature includes:

[0017] Performing attention calculation on each of the low-dimensional facial features to obtain an attention weight coefficient; wherein the attention weight coefficient represents the importance of the low-dimensional facial feature;

[0018] At least two different low-dimensional facial features are fused according to the attention weight coefficient to obtain the facial fusion feature.

[0019] The step of fusing at least two different low-dimensional facial features to obtain the facial fusion feature includes:

[0020] Performing attention calculation on each of the low-dimensional facial features to obtain an attention weight coefficient; wherein the attention weight coefficient represents the importance of the low-dimensional facial feature;

[0021] At least two different low-dimensional facial features are fused according to the attention weight coefficient to obtain the facial fusion feature.

[0022] In some embodiments, performing feature fusion evaluation on the face fusion features according to the face identification information to obtain feature fusion evaluation data of the candidate model group includes:

[0023] Calculating the similarity of any two face fusion features to obtain fusion feature similarity;

[0024] Obtaining the fusion feature similarity of the sample objects identified by the face identification information being the same, to obtain a first similarity;

[0025] Obtaining the fusion feature similarity that the face identification information identifies that the sample objects are different, to obtain a second similarity;

[0026] The first similarity and the second similarity are concatenated to obtain the feature fusion evaluation data of the candidate model group.

[0027] In some embodiments, performing feature fusion evaluation on the face fusion features according to the face identification information to obtain feature fusion evaluation data of the candidate model group includes:

[0028] Acquire a sample object of the face fusion feature according to the face identification information to obtain face verification information;

[0029] Performing face recognition on the face fusion feature to obtain face recognition information; wherein the face recognition information represents the sample object matched by the face fusion feature;

[0030] The face recognition information and the face verification information are subjected to loss calculation to obtain feature fusion evaluation data of the candidate model group.

[0031] In some embodiments, before inputting the face image sample data into a preset candidate model group, the method further includes:

[0032] Constructing the candidate model group specifically includes:

[0033] Extracting features from the facial image sample data using a preset candidate facial recognition model to obtain at least two different candidate facial features;

[0034] Calculating the similarity of any two different candidate facial features to obtain the candidate feature similarity;

[0035] Selecting at least two different selected face recognition models from the candidate face recognition models according to the candidate feature similarity;

[0036] At least two different selected face recognition models are combined to obtain the candidate model group.

[0037] In some embodiments, selecting at least two different selected face recognition models from the candidate face recognition models according to the candidate feature similarity includes:

[0038] Screening the candidate face recognition models according to the candidate feature similarities to obtain a preliminary face recognition model;

[0039] Obtaining model structure information of each of the preliminary face recognition models;

[0040] At least two different selected face recognition models are screened out from the preliminary face recognition models according to the model structure information.

[0041] To achieve the above-mentioned purpose, a second aspect of an embodiment of the present application provides a face recognition device, the device comprising:

[0042] A sample acquisition module, used to acquire a face image sample set; wherein the face image sample set includes: face image sample data containing sample objects and face identification information of the face image sample data, wherein the face identification information is used to identify whether the sample objects contained in the face image sample data are the same or different;

[0043] A data input module, used for inputting the face image sample data into a preset candidate model group; wherein the candidate model group includes at least two different selected face recognition models;

[0044] A feature extraction module, used for performing feature extraction on the face image sample data through each of the selected face recognition models to obtain at least two different preliminary face features;

[0045] A fusion module, used for fusing at least two different preliminary facial features to obtain a facial fusion feature of each candidate model group;

[0046] A feature evaluation module, used to perform feature fusion evaluation on the face fusion feature according to the face identification information to obtain feature fusion evaluation data of the candidate model group; wherein the feature fusion evaluation data represents the recognition accuracy of the face fusion feature;

[0047] A screening module, used for screening out a target model group from the candidate model group according to the feature fusion evaluation data;

[0048] The face recognition module is used to perform face recognition on the target face image data through the target model group.

[0049] To achieve the above-mentioned purpose, the third aspect of an embodiment of the present application proposes a computer device, which includes a memory and a processor, the memory stores a computer program, and the processor implements the method described in the first aspect when executing the computer program.

[0050] To achieve the above-mentioned purpose, the fourth aspect of an embodiment of the present application proposes a computer-readable storage medium, wherein the computer-readable storage medium stores a computer program, and when the computer program is executed by a processor, the method described in the first aspect is implemented.

[0051] The face recognition method and apparatus, computer equipment and storage medium proposed in the present application form a candidate model group by combining at least two different selected face recognition models, extracting preliminary face features from face image sample data through each selected face recognition model, and at least two different selected face recognition models can obtain at least two different preliminary face features. The face fusion features output by each candidate model group are obtained by fusing at least two different preliminary face features. In order to evaluate the face feature fusion effect of the candidate model group, that is, to analyze the face recognition accuracy of each candidate model group, the face fusion features are evaluated through face identification information to obtain feature fusion evaluation data that characterizes the face recognition accuracy of the candidate model group. Therefore, a target model group with better recognition accuracy is selected from the candidate model group through the feature fusion evaluation data, so that face recognition is performed on the target face image data through the target model group to improve the accuracy of face recognition. BRIEF DESCRIPTION OF THE DRAWINGS

[0052] Figure 1 is a flow chart of a face recognition method provided by an embodiment of the present application;

[0053] Figure 2 is a flow chart of a face recognition method provided by another embodiment of the present application;

[0054] Figure 3 yes Figure 2 Flow chart of step S203 in FIG.

[0055] Figure 4 yes Figure 1 Flow chart of step S104 in FIG.

[0056] Figure 5 yes Figure 4 Flow chart of step S403 in FIG.

[0057] Figure 6 It is a detailed flow chart of constructing face fusion features in the face recognition method provided in the embodiment of the present application;

[0058] Figure 7 yes Figure 1 Flow chart of step S105 in FIG.

[0059] Figure 8 yes Figure 1 Another flow chart of step S105 in;

[0060] Fig. 9 is a structural schematic diagram of a face recognition device provided in an embodiment of the present application;

[0061] Fig.10 It is a schematic diagram of the hardware structure of the computer device provided in the embodiment of the present application. DETAILED DESCRIPTION

[0062] In order to make the purpose, technical solution and advantages of the present application more clearly understood, the present application is further described in detail below in conjunction with the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are only used to explain the present application and are not used to limit the present application.

[0063] It should be noted that, although the functional modules are divided in the device schematic diagram and the logical order is shown in the flowchart, in some cases, the steps shown or described may be performed in a different order than the module division in the device or the order in the flowchart. The terms "first", "second", etc. in the specification, claims and the above drawings are used to distinguish similar objects, and are not necessarily used to describe a specific order or sequence.

[0064] Unless otherwise defined, all technical and scientific terms used herein have the same meaning as those commonly understood by those skilled in the art to which this application belongs. The terms used herein are only for the purpose of describing the embodiments of this application and are not intended to limit this application.

[0065] First, some nouns involved in this application are analyzed:

[0066] Artificial intelligence (AI) is a new technical science that studies and develops theories, methods, technologies and application systems for simulating, extending and expanding human intelligence. AI is a branch of computer science that attempts to understand the essence of intelligence and produce a new type of intelligent machine that can respond in a similar way to human intelligence. Research in this field includes robotics, speech recognition, image recognition, natural language processing and expert systems. AI can simulate the information process of human consciousness and thinking. AI is also a theory, method, technology and application system that uses digital computers or machines controlled by digital computers to simulate, extend and expand human intelligence, perceive the environment, acquire knowledge and use knowledge to obtain the best results.

[0067] Face recognition: It is a biometric technology that identifies people based on their facial features. It uses a camera or camcorder to collect images or video streams containing faces, and automatically detects and tracks faces in the images, and then performs facial recognition on the detected faces. It is also commonly called portrait recognition or facial recognition.

[0068] VGGFace database: It is a large-scale face recognition dataset, containing 3.31 million images, 9131 IDs, and an average of 362.6 images. This dataset has the following characteristics: 1) There are many person IDs, and each ID contains a large number of images. 2) It covers a wide range of postures, ages, and races. 3) It minimizes noise as much as possible.

[0069] ArcFace: is an advanced face recognition algorithm that improves the model's discriminative ability by using the Additive Angular Margin Loss function. The core idea of ​​ArcFace is to maximize the inter-class distance and minimize the intra-class distance in the feature space, thereby learning highly discriminative facial features. This algorithm has achieved good results in face recognition tasks and has the advantages of simple implementation, high training efficiency, and intuitive geometric interpretation.

[0070] Curriculumface: It is a new loss function for deep face recognition, which combines the concept of curriculum learning and solves the shortcomings of interval-based and difficult sample mining methods. This method focuses on easy samples in the early stage of training and focuses on difficult samples in the later stage, and promotes model convergence and performance improvement by dynamically adjusting the importance of samples.

[0071] Cosine Similarity: It is a measure of the similarity between two non-zero vectors. It determines their similarity by calculating the cosine value of the angle between the two vectors.

[0072] Low-rank decomposition: is a method for decomposing a high-dimensional matrix into the product of several lower-rank matrices. This decomposition can reveal the intrinsic structure of the data, reduce the complexity of the data, and has applications in many fields, such as signal processing, image processing, machine learning, and data compression.

[0073] Face recognition is a process that uses computer vision technology to identify the identity or attributes of a face from an image or video. With the development of deep learning, face recognition has made significant progress and applications in security, finance, entertainment and other fields. However, face recognition still faces many challenges and problems. For example, face images are affected by factors such as lighting, posture, occlusion, and expression, which increases the difficulty of facial feature extraction and matching. On the other hand, due to the low quality, low resolution, and complex background of face images in actual scenarios, the face recognition performance of existing algorithms in complex environments is reduced. For example, under low light conditions, the contrast and clarity of face images will be reduced, making it difficult to extract facial features, affecting the accuracy of face recognition. Therefore, how to improve the accuracy of face recognition has become an urgent problem to be solved.

[0074] Based on this, the embodiments of the present application provide a face recognition method and apparatus, a computer device and a storage medium, aiming to improve the accuracy of face recognition.

[0075] The face recognition method and apparatus, computer equipment and storage medium provided in the embodiments of the present application are specifically illustrated through the following embodiments. First, the face recognition method in the embodiments of the present application is described.

[0076] The embodiments of the present application can acquire and process relevant data based on artificial intelligence technology. Among them, artificial intelligence (AI) is the theory, method, technology and application system that uses digital computers or machines controlled by digital computers to simulate, extend and expand human intelligence, perceive the environment, acquire knowledge and use knowledge to obtain the best results.

[0077] AI basic technologies generally include sensors, dedicated AI chips, cloud computing, distributed storage, big data processing technology, operation / interaction systems, mechatronics, etc. AI software technologies mainly include computer vision technology, robotics technology, biometrics technology, speech processing technology, natural language processing technology, and machine learning / deep learning.

[0078] The face recognition method provided in the embodiment of the present application relates to the field of financial technology. The face recognition method provided in the embodiment of the present application can be applied to a terminal, can be applied to a server side, or can be software running in a terminal or a server side. In some embodiments, the terminal can be a smart phone, a tablet computer, a laptop computer, a desktop computer, etc.; the server side can be configured as an independent physical server, or a server cluster or distributed system composed of multiple physical servers, or a cloud server that provides basic cloud computing services such as cloud services, cloud databases, cloud computing, cloud functions, cloud storage, network services, cloud communications, middleware services, domain name services, security services, CDN, and big data and artificial intelligence platforms; the software can be an application that implements the face recognition method, etc., but is not limited to the above forms.

[0079] The present application can be used in many general or special computer system environments or configurations. For example: personal computers, server computers, handheld or portable devices, tablet devices, multiprocessor systems, microprocessor-based systems, set-top boxes, programmable consumer electronics, network PCs, minicomputers, mainframe computers, distributed computing environments including any of the above systems or devices, etc. The present application can be described in the general context of computer-executable instructions executed by a computer, such as program modules. Generally, program modules include routines, programs, objects, components, data structures, etc. that perform specific tasks or implement specific abstract data types. The present application can also be practiced in distributed computing environments, in which tasks are performed by remote processing devices connected through a communication network. In a distributed computing environment, program modules can be located in local and remote computer storage media including storage devices.

[0080] It should be noted that in each specific implementation of the present application, when it comes to the need to perform relevant processing based on data related to user identity or characteristics such as user information, user behavior data, user historical data, and user location information, the user's permission or consent will be obtained first, and the collection, use, and processing of these data will comply with relevant laws, regulations, and standards. In addition, when the embodiment of the present application needs to obtain the user's sensitive personal information, the user's separate permission or consent will be obtained through a pop-up window or by jumping to a confirmation page. After clearly obtaining the user's separate permission or consent, the necessary user-related data for the normal operation of the embodiment of the present application will be obtained.

[0081] Figure 1 is an optional flowchart of the face recognition method provided in the embodiment of the present application. Figure 1 The method may include but is not limited to steps S101 to S107.

[0082] Step S101, obtaining a face image sample set; wherein the face image sample set includes: face image sample data containing sample objects and face identification information of the face image sample data, the face identification information is used to identify whether the sample objects contained in the face image sample data are the same or different;

[0083] Step S102, inputting the face image sample data into a preset candidate model group; wherein the candidate model group includes at least different selected face recognition models;

[0084] Step S103, extracting features from the face image sample data using each selected face recognition model to obtain at least two different preliminary face features;

[0085] Step S104, fusing at least two different preliminary facial features to obtain facial fusion features of each candidate model group;

[0086] Step S105, performing feature fusion evaluation on the face fusion features according to the face identification information to obtain feature fusion evaluation data of the candidate model group; wherein the feature fusion evaluation data represents the recognition accuracy of the face fusion features;

[0087] Step S106, selecting a target model group from the candidate model groups according to the feature fusion evaluation data;

[0088] Step S107, performing face recognition on the target face image data through the target model group.

[0089] In the steps S101 to S107 shown in the embodiment of the present application, the face image sample data with face identification information is input into the candidate model group, and the candidate model group includes at least two different selected face recognition models, and the face image sample data is subjected to feature extraction by each selected face recognition model to obtain at least two different preliminary face features, and then the at least two different preliminary face features are fused into face fusion features. Therefore, the face features extracted by at least two selected face recognition models are fused to obtain face fusion features that represent more face information, and then the face fusion features of each candidate model group are evaluated according to the face identification information to obtain feature fusion evaluation data that represents the recognition accuracy, then the fusion evaluation data can represent the feature fusion effect of the candidate model group, and indirectly represent the face recognition accuracy. Finally, based on the feature fusion evaluation data, the target model group is screened out from the candidate model group to select a target model group with higher recognition accuracy, then the target model group is used to perform face recognition on the target face image, and the face recognition accuracy can be improved. Therefore, the embodiment of the present application extracts facial features by combining at least two different facial recognition models, and the combined facial recognition models extract facial features separately and then fuse them to obtain fused facial features, and the fused facial features can represent richer facial information. At the same time, in order to select the model group with the best fusion effect, the fused facial features will be evaluated, and the model group with better fusion effect and higher recognition accuracy will be selected, so as to perform facial recognition through the model group with better recognition accuracy, thereby improving the accuracy of facial recognition.

[0090] It should be noted that the face recognition method can be applied to security, finance and entertainment fields. For example, in the financial field, in order to improve the security of financial payment, how to achieve high-precision face recognition effect is crucial. The face recognition method of this embodiment can improve the security of user verification and payment by screening out a high-precision target model group, and then the target model group is applied to face recognition in the financial payment process.

[0091] In step S101 of some embodiments, the face image sample set is used to select a model group with high recognition accuracy, and the face image sample set includes face image sample data containing sample objects and face identification information of the face image sample data, and the face image sample data can be directly obtained by shooting with an image collector, or can be obtained by collecting face image video data with an image collector and performing frame processing on the face image video data, or can be extracted from a pre-set face image database, and this embodiment does not impose specific restrictions on the method of obtaining face image sample data. For example, if the face image sample data is extracted from the face image database, preliminary face image data will be extracted from the face image database, and image parameters of the preliminary face image data will be obtained, and the preliminary face image data will be filtered according to the preset image selection rules and image parameters to obtain face image sample data whose image parameters meet the image selection rules. Among them, the image selection rules can be pixel rate threshold, face integrity, etc.

[0092] In this embodiment, the face image database is the VGGFace database, and the VGGFace database is a large-scale face recognition database that stores face image data of multiple sample objects, and face image data can be directly downloaded from the VGGFace database through a browser as face image sample data. It should be noted that the face image database can be the VGGFace database, or it can be the FERET database, the CMU Multi-PIE database, and the YALE database, and the FERET database, the CMU Multi-PIE database, and the YALE database are all commonly used face image databases, and this embodiment does not specifically limit the face image database.

[0093] As disclosed above, the face image sample set includes face identification information of the face image sample data, and the face identification information is used to identify whether the sample objects contained in the face image sample data are the same or different. It should be noted that the face identification information is specifically the identity identification of the sample object, and the same or different identity identification is used to determine whether the sample objects contained in the face image sample data are the same or different. For example, if the face image sample set contains 20,000 face image sample data, and the face identification information in the 20,000 face image sample data can identify 5 sample objects, then 2,000 face image sample data with the same face identification information are set, so that the target model group with better recognition accuracy can be selected more accurately.

[0094] In step S102 of some embodiments, multiple candidate model groups are set, and the combination structure of the selected face recognition models in each candidate model group is different. In this embodiment, it is mainly analyzed whether different combinations of selected face recognition models have better recognition effects than a single face recognition model, and the selected face recognition model combination with better recognition accuracy is used as the target model group.

[0095] For example, if there are three different selected face recognition models, namely ArcFace_50, ArcFace_101, and curricularface, and ArcFace_50, ArcFace_101, and curricularface are all face recognition models with better recognition performance in the field of face recognition. Specifically, if there are three selected face recognition models, the combination structure of the candidate model group can be: ArcFace_50+ArcFace_101, ArcFace_50+curricularface, ArcFace_101+curricularface, and ArcFace_50+ArcFace_101+curricularface.

[0096] Before step S102 in some embodiments, the face recognition method further includes constructing a candidate model group to set the candidate model group in advance for evaluation processing and select a target model group with better recognition accuracy.

[0097] Please refer to Figure 2 In some embodiments, constructing a candidate model group may include but is not limited to steps S201 to S204:

[0098] Step S201, extracting features from facial image sample data using a preset candidate facial recognition model to obtain at least two different candidate facial features;

[0099] Step S202, calculating the similarity of any two different candidate facial features to obtain the candidate feature similarity;

[0100] Step S203, selecting at least two different selected face recognition models from the candidate face recognition models according to the candidate feature similarity;

[0101] Step S204: combine at least two different selected face recognition models to obtain a candidate model group.

[0102] In step S201 of some embodiments, multiple candidate face recognition models are set, and each candidate face recognition model performs feature extraction on the face image sample data, and feature extraction is to extract the potential face features in the face image sample data, which is also the face distinguishing feature for face recognition. It should be noted that the candidate face features are represented by vectors. In this embodiment, the length of the candidate face features is 512. The length of the candidate face features in other embodiments can be set as required, and this embodiment does not make specific restrictions.

[0103] In step S202 of some embodiments, the candidate facial features are facial features output by each candidate facial recognition model. In order to determine the correlation between the features output by any two candidate facial recognition models, the candidate feature similarity is obtained by performing similarity calculation on any two different candidate facial features. Specifically, in this embodiment, the candidate feature similarity is calculated using a cosine similarity calculation method for any two different candidate facial features, and the cosine similarity calculation method is: , where are the similarities of two different candidate features, is the candidate feature similarity.

[0104] For example, if the candidate face recognition models are ArcFace_50, ArcFace_101, and curriculumface, multiple candidate feature similarities are obtained by calculating the cosine similarity of the candidate face features output by any two candidate face recognition models. Because there are multiple face image sample data, there are also multiple candidate feature similarities between any two candidate face recognition models. The final candidate feature similarity is determined by averaging the candidate feature similarities between any two candidate face recognition models. It should be noted that the candidate feature similarities between ArcFace_50, ArcFace_101, and curriculumface are as follows.

[0105] As shown in Table 1:

[0106] ArcFace_50 ArcFace_101 curricularface ArcFace_50 1.000000 0.003967 0.0044940 ArcFace_101 0.003967 1.000000 0.001913 curricularface 0.0044940 0.001913 1.000000

[0107] Table 1

[0108] It can be seen from Table 1 that the correlation between the candidate facial features of two different candidate facial recognition models is close to 0, while the correlation between the candidate facial features of the same candidate facial recognition model is 1, which means that the facial features extracted by different candidate facial recognition models are close to having no correlation. Therefore, different candidate facial recognition models can be combined into a candidate model group to improve the effect of facial recognition.

[0109] In steps S203 to S204 of some embodiments, the candidate feature similarity is used to characterize the correlation of output features between different candidate face recognition models, so at least two different selected face recognition models are selected from the candidate face recognition models through the candidate feature similarity, that is, at least two selected face recognition models whose feature fusion can improve the face recognition effect are selected. It should be noted that after selecting at least two different selected face recognition models, by combining at least two different selected face recognition models into a candidate model group, a candidate model group that can improve the face recognition effect is constructed.

[0110] In steps S201 to S204 shown in the present embodiment, the candidate facial features output by each candidate facial recognition model are subjected to similarity calculation to obtain the candidate feature similarity, and the feature correlation between two different candidate facial recognition models is characterized by the candidate feature similarity. At least two different selected facial recognition models can be selected from the candidate facial recognition models according to the candidate feature similarity, and the at least two different selected facial recognition models can be combined into a candidate model group to construct a candidate model group having a better facial recognition effect than a single facial recognition model.

[0111] See also Figure 3 In some embodiments, step S203 may include but is not limited to steps S301 to S303:

[0112] Step S301, screening candidate face recognition models according to candidate feature similarities to obtain a preliminary face recognition model;

[0113] Step S302, obtaining model structure information of each preliminary face recognition model;

[0114] Step S303, selecting at least two different selected face recognition models from the preliminary face recognition models according to the model structure information.

[0115] In step S301 of some embodiments, at least two preliminary face recognition models are screened out from the candidate face recognition models. It should be noted that the screening of the preliminary face recognition models is to use the candidate face recognition models whose candidate feature similarity is less than the preset similarity threshold as the preliminary face recognition models, and the candidate feature similarity is less than the preset similarity threshold, indicating that the feature correlation between the at least two preliminary face recognition models screened out is not high, then the features output by any at least two preliminary face recognition models will be fused to obtain richer facial features, so any combination of at least two preliminary face recognition models has better recognition accuracy than a single face recognition model.

[0116] In step S302 of some embodiments, the model structure information characterizes the network structure of each preliminary face recognition model. The model structure information can be used to determine whether the preliminary face recognition models are the same, so as to select different selected face recognition models based on the model structure information.

[0117] In step S303 of some embodiments, selection is made among preliminary face recognition models based on model structure information, that is, preliminary face recognition models with different model structure information are selected as selected face recognition models to obtain at least two different selected face recognition models.

[0118] In steps S301 to S303 shown in the present embodiment, a preliminary face recognition model with low feature correlation is first screened from the candidate face recognition models according to the candidate feature similarity, and then different preliminary face recognition models are selected as at least two different selected face recognition models based on the model structure information, so as to select at least two different selected face recognition models that can be combined to improve the face recognition effect.

[0119] In step S103 of some embodiments, each selected face recognition model performs feature extraction on the face image sample data, that is, extracts the latent vector of the face image sample data as the preliminary face feature. Because at least two selected face recognition models are different, the preliminary face features of each selected face recognition model are different, so as to obtain at least two different preliminary face features.

[0120] For example, if the candidate model group includes three selected face recognition models, namely ArcFace_50+ArcFace_101+curricularface, feature extraction is performed on the same face image sample data through ArcFace_50, ArcFace_101, and curricularface respectively to obtain three preliminary face features. The three preliminary face features can clearly show that the face features extracted by different selected face recognition models are unrelated. Then, the preliminary face features with different face features are fused to obtain face fusion features that represent richer face features.

[0121] See also Figure 4 In some embodiments, step S104 may include but is not limited to steps S401 to S403:

[0122] Step S401, filtering out target dimension information from preset candidate dimension information;

[0123] Step S402, performing dimensionality reduction processing on each preliminary facial feature according to the target dimension information to obtain a low-dimensional facial feature;

[0124] Step S403: fusing at least two different low-dimensional facial features to obtain a fused facial feature.

[0125] It should be noted that, in addition to the face recognition model combination being able to improve the face recognition effect, a fusion operation between preliminary face features can also be set to determine the impact of different fusion operations on the face recognition effect. Therefore, in this embodiment, the preliminary face features are subjected to dimensionality reduction processing and then fused to improve the fusion effect between features.

[0126] In step S401 of some embodiments, target dimension information is selected from the candidate dimension information, and the target dimension information represents the dimension after the dimensionality reduction of the preliminary facial features. It should be noted that the selection of the target dimension information is to randomly select the target dimension information from the candidate dimension information.

[0127] In step S402 of some embodiments, the preliminary facial information is subjected to dimensionality reduction processing according to the target dimensionality information. In this embodiment, the preliminary facial information is subjected to a low-rank decomposition operation according to the target dimensionality information, that is, the high-dimensional preliminary facial features are decomposed into several low-dimensional low-dimensional facial features. For example, if the dimension of the preliminary facial features is 512*2, and the target dimensionality information is 256 and 768, the preliminary facial features are subjected to dimensionality reduction by using a low-rank decomposition operation to obtain two low-dimensional facial features of length 256 and 768.

[0128] In steps S401 to S403 shown in this embodiment, the fusion effect of facial features can be improved by reducing the dimension of the preliminary facial features according to the target dimension information and then fusing them, rather than directly fusing the preliminary facial features, thereby improving the facial feature fusion effect and thus improving the facial recognition accuracy.

[0129] See also Figure 5 In some embodiments, step S403 may include but is not limited to steps S501 to S502:

[0130] Step S501, performing attention calculation on each low-dimensional facial feature to obtain an attention weight coefficient; wherein the attention weight coefficient represents the importance of the low-dimensional facial feature;

[0131] Step S502: fusing at least two different low-dimensional facial features according to the attention weight coefficient to obtain a facial fusion feature.

[0132] In step S501 of some embodiments, the low-dimensional facial features are facial features output by the selected facial recognition model, and the facial features extracted by different selected facial recognition models represent different importance of facial information. In order to judge the importance of facial information represented by each low-dimensional facial feature, an attention calculation is performed on the low-dimensional facial features to obtain an attention weight coefficient. It should be noted that this embodiment uses an attention mechanism to perform an attention calculation on the low-dimensional facial features to obtain an attention weight coefficient. The higher the attention weight coefficient, the more important the facial information represented by the low-dimensional facial features is.

[0133] In step S502 of some embodiments, each low-dimensional facial feature corresponds to an attention weight coefficient, and at least two different low-dimensional facial features are fused according to the attention weight coefficient. The facial fusion feature can be obtained by weighted summing up the at least two different low-dimensional facial features according to the attention weight coefficient, or by weighted averaging the at least two different low-dimensional facial features according to the attention weight coefficient. This embodiment does not limit the specific operation of fusing at least two different low-dimensional facial features into a facial fusion feature.

[0134] Specifically, Figure 6 As shown, if the candidate model group includes two selected face recognition models, and the two selected face recognition models are ArcFace_101 and curricularface, ArcFace_101 is used to extract the preliminary face features from the face image sample data, which is defined as Arc_Latent1, and the low-dimensional face features corresponding to ArcFace_101 are obtained after the preliminary face features are reduced in dimension, which is defined as Arc_Latent2. Cur_Latent1 is used to extract the preliminary face features from the face image sample data, and the low-dimensional face features corresponding to curricularface are obtained after the preliminary face features are reduced in dimension, which is defined as Cur_Latent2. By performing attention calculation on the low-dimensional face features of Arc_Latent2 and Cur_Latent2 respectively, the attention weight coefficients are obtained, which are X1 and X2 respectively, and then the two low-dimensional face features are fused based on the attention weight coefficients, that is, the face fusion feature is obtained according to the following formula (1).

[0135] Fus_Latent=X1*Arc_Latent2+X2*Cur_Latent2; (1)

[0136] Where Fus_Latent is the face fusion feature, Arc_Latent is the low-dimensional face feature corresponding to ArcFace_101, Cur_Latent2 is the low-dimensional face feature corresponding to curricularface, X1 is the attention weight coefficient corresponding to Arc_Latent2, and X2 is the attention weight coefficient corresponding to Cur_Latent2.

[0137] In steps S501 to S502 shown in this embodiment, attention calculation is performed on each low-dimensional facial feature to obtain an attention weight coefficient that represents the importance of facial information. Then, at least two different low-dimensional facial features are fused into a facial fusion feature according to the attention weight coefficient to obtain a facial fusion feature that more accurately represents the facial information.

[0138] See also Figure 7 In some embodiments, step S105 may include but is not limited to steps S701 to S704:

[0139] Step S701, calculating the similarity of any two face fusion features to obtain the fusion feature similarity;

[0140] Step S702, obtaining the fusion feature similarity of the sample objects with the same face identification information identification, and obtaining a first similarity;

[0141] Step S703, obtaining the similarity of fused features of sample objects with different facial identification information identifications, and obtaining a second similarity;

[0142] Step S704: concatenate the first similarity and the second similarity to obtain the fusion evaluation data of the candidate model group.

[0143] In step S701 of some embodiments, the face fusion feature is a face feature extracted and fused by at least two different selected face recognition models from a face image sample data, and a face image sample set contains multiple face image sample data, so the similarity of any two face fusion features is calculated to obtain the fusion feature similarity. It should be noted that the face fusion features of the same sample object are similar, while the face fusion features of different sample objects are different, so the fusion feature similarity can characterize the recognition accuracy of the same candidate model group on the same and different sample objects.

[0144] Specifically, in the optimal theoretical case, the similarity of fused features of the same sample object is close to 1, and conversely, the similarity of fused features between different sample objects is close to 0.

[0145] In steps S702 to S703 of some embodiments, it can be determined based on the face identification information whether the fused feature similarity is represented as the feature similarity between the same sample object, so the fused feature similarity for the sample objects with the same face identification information is taken as the first similarity, and the fused feature similarity for the sample objects with different face identification information is taken as the second similarity, so that it can be determined whether the candidate model group can accurately fuse the facial features through the first similarity and the second similarity.

[0146] For example, there are 5 sample objects, which are defined as sample object P1, sample object P2, sample object P3, sample object P4 and sample object P5 respectively. There are 2000 face image recognition data of the same sample object, so there are 2000 face fusion features of the same sample object, and the fusion feature similarity is 1000, so the average value of the fusion feature similarity of one sample object is taken, and then the average fusion feature similarity of the 5 sample objects is averaged to obtain the first similarity. At the same time, for the 5 sample objects, different sample objects can be combined to obtain 9 sample object groups: (P1, P2), (P1, P3), (P1, P4), (P1, P5), (P2, P3), (P2, P4), (P2, P5), (P3, P4), (P3, P5), then the fusion feature similarity of each combination is averaged, and then the average fusion feature similarity of the 9 sample object groups is averaged to obtain the second similarity.

[0147] Specifically, if the combination structure of the candidate model group can be: ArcFace_50+ArcFace_101, ArcFace_50+curricularface, ArcFace_101+curricularface and ArcFace_50+ArcFace_101+curricularface, the first similarities and the second similarities corresponding to different candidate model groups are determined as shown in Table 2:

[0148] Candidate model group First similarity Second similarity ArcFace_50+ArcFace_101+curricularface 0.601518 0.049245 ArcFace_50+ArcFace_101 0.469791 0.030376 ArcFace_50+curricularface 0.604461 0.061646 ArcFace_101+curricularface 0.595087 0.049628

[0149] Table 2

[0150] From Table 2, we can see that the candidate model group is the fusion of ArcFace_50 and curricularface, or the combination of ArcFace_101 and curricularface.

[0151] In step S704 of some embodiments, the first similarity and the second similarity are spliced, and the splicing can be done by weighted summation, average value, weighted average, and summation. This embodiment does not specifically limit the splicing method of the first similarity and the second similarity. It should be noted that this embodiment splices the fusion feature similarities corresponding to the same sample object and different sample objects as feature fusion evaluation data, which also characterizes the commonality of the fusion features of the candidate model group when the sample object is the same, and the difference of the fusion features between different sample objects, which can reflect the face recognition accuracy of the candidate model group.

[0152] In steps S701 to S704 shown in this embodiment, the fusion feature similarity is obtained by first calculating the similarity between any two face fusion features, and then the fusion feature similarity of the same sample object is used as the first similarity, and the fusion feature similarity of different sample objects is used as the second similarity. Finally, the first similarity and the second similarity are spliced ​​into feature fusion evaluation data, so as to determine the fusion performance of the output features of the candidate model group through the fusion feature similarity between the same and different sample objects, which indirectly reflects the face recognition accuracy of the candidate model group.

[0153] See also Figure 8 In some embodiments, step S105 may also include but is not limited to steps S801 to S803:

[0154] Step S801, obtaining a sample object of face fusion features according to face identification information to obtain face verification information;

[0155] Step S802, performing face recognition on the face fusion feature to obtain face recognition information; wherein the face recognition information represents a sample object matched by the face fusion feature;

[0156] Step S803, performing loss calculation on the face recognition information and the face verification information to obtain feature fusion evaluation data of the candidate model group.

[0157] In step S801 of some embodiments, the face identification information can be used to determine which sample object the face fusion feature originates from, so as to determine the face verification information. It should be noted that each face fusion feature originates from a face image sample data, and the face image sample data contains a sample object, and the face identification information identifies whether the sample objects contained in the face image sample data are the same or different, so that the face identification information can be used to determine which sample object the face image sample data belongs to.

[0158] In step S802 of some embodiments, face recognition is performed on the face fusion features to obtain face recognition information, and the face recognition information indicates which sample object the face fusion features correspond to. Specifically, a classifier is trained based on a face image sample set, and the classifier can accurately classify the sample objects involved in the face image sample data, and face recognition is performed on the face fusion features by using the classifier to output face recognition information that identifies the sample object. For example, if the face image sample set consists of face image sample data of 5 sample objects, the classifier can directly identify the sample object information as face recognition information based on the face fusion features by learning the face features of the 5 sample objects.

[0159] In step S803 of some embodiments, loss data is obtained by performing loss calculation on the face recognition information and the face verification information, and the loss data is used as feature fusion evaluation data to directly characterize the face recognition accuracy of the candidate model group. It should be noted that the loss calculation can be implemented using loss functions such as mean square error, mean absolute error, cross entropy loss, and logarithmic loss, and this embodiment does not impose specific restrictions on the loss calculation.

[0160] In steps S801 to S803 illustrated in the present embodiment, face recognition is performed based on face fusion features to obtain face recognition information, and loss calculation is performed based on face verification information and face recognition information corresponding to the face fusion features to obtain feature fusion evaluation data. Therefore, the feature fusion evaluation data can directly characterize the face recognition accuracy of the candidate model group.

[0161] In step S106 of some embodiments, the candidate model group with the largest feature fusion evaluation data is used as the target model group. Then the target model group is the model group with the highest recognition accuracy among the candidate model groups, which can achieve more accurate face recognition.

[0162] In step S107 of some embodiments, the target face image data is a face image collected in real time. Specifically, the target face image data can be directly obtained through a face collector, or a video frame image containing a face can be selected from collected face video data as the target face image data.

[0163] For example, if the application scenario is a financial payment system, the target face image data is the face image data collected by the financial payment system. By using the target model group to perform face recognition on the target face image data, high-precision face recognition can be achieved, improving the security of financial payment. At the same time, if the target model group is used for face recognition in the security field, the security of security verification can also be improved.

[0164] The embodiment of the present application first extracts multiple face image sample data and face identification information of the face image sample data from the face image database, and then combines the face image sample data and the face identification information into a face image sample set. By inputting multiple face image sample data into the candidate model group respectively, and the candidate model group includes at least two different selected face recognition models, each selected face recognition model performs feature extraction on each face image sample data to obtain preliminary face features. At the same time, select target dimension information from the candidate dimension information, reduce the preliminary face features to low-dimensional face features according to the target dimension information, and then perform attention calculation on each low-dimensional face feature to obtain an attention weight coefficient, and fuse at least two different low-dimensional face features into face fusion features according to the attention weight coefficient to obtain the face fusion features of each candidate model group. In order to judge the fusion effect of the face fusion features output by the candidate model group, that is, whether the face fusion features can improve the face recognition accuracy. The fusion feature similarity is obtained by calculating the cosine similarity of any two face fusion features obtained from the same candidate model group, and then the fusion feature similarity of the sample objects with the same face identification information is used as the first similarity, and the fusion feature similarity of the sample objects with different face identification information is used as the second similarity. Finally, the first similarity and the second similarity are spliced ​​into feature fusion evaluation data, so as to characterize the fusion effect of face fusion features through feature fusion evaluation data, and indirectly reflect the face recognition accuracy. Therefore, the candidate model group with the largest feature fusion evaluation data is used as the target model group, and face recognition is performed on the target face image data through the target model group, so that a model group with better face recognition accuracy can be selected, thereby improving the face recognition accuracy.

[0165] See also Fig. 9 The embodiment of the present application also provides a face recognition device, which can implement the above face recognition method, and the device includes:

[0166] The sample acquisition module 901 is used to acquire a face image sample set; wherein the face image sample set includes: face image sample data containing sample objects and face identification information of the face image sample data, and the face identification information is used to identify whether the sample objects contained in the face image sample data are the same or different;

[0167] The data input module 902 is used to input the face image sample data into a preset candidate model group; wherein the candidate model group includes at least two different selected face recognition models;

[0168] A feature extraction module 903 is used to extract features from the face image sample data through each selected face recognition model to obtain at least two different preliminary face features;

[0169] A fusion module 904 is used to fuse at least two different preliminary facial features to obtain a facial fusion feature of each candidate model group;

[0170] The feature evaluation module 905 is used to perform feature fusion evaluation on the face fusion feature according to the face identification information to obtain feature fusion evaluation data of the candidate model group; wherein the feature fusion evaluation data represents the recognition accuracy of the face fusion feature;

[0171] A screening module 906 is used to screen out a target model group from the candidate model group according to the feature fusion evaluation data;

[0172] The face recognition module 907 is used to perform face recognition on the target face image data through the target model group.

[0173] The specific implementation of the face recognition device is basically the same as the specific implementation of the above-mentioned face recognition method, and will not be repeated here.

[0174] The embodiment of the present application also provides a computer device, the computer device includes a memory and a processor, the memory stores a computer program, and the processor implements the above-mentioned face recognition method when executing the computer program. The computer device can be any intelligent terminal including a tablet computer, a car computer, etc.

[0175] See also Fig.10 , Fig.10 The hardware structure of a computer device according to another embodiment is shown, and the computer device includes:

[0176] The processor 1001 may be implemented by a general-purpose CPU (Central Processing Unit), a microprocessor, an application-specific integrated circuit (Application Specific Integrated Circuit, ASIC), or one or more integrated circuits, and is used to execute relevant programs to implement the technical solutions provided in the embodiments of the present application;

[0177] The memory 1002 can be implemented in the form of a read-only memory (ROM), a static storage device, a dynamic storage device, or a random access memory (RAM). The memory 1002 can store an operating system and other application programs. When the technical solution provided in the embodiment of this specification is implemented by software or firmware, the relevant program code is stored in the memory 1002, and the processor 1001 calls and executes the face recognition method of the embodiment of this application;

[0178] Input / output interface 1003, used to implement information input and output;

[0179] The communication interface 1004 is used to realize the communication interaction between the device and other devices. The communication can be realized through a wired manner (such as USB, network cable, etc.) or a wireless manner (such as mobile network, WIFI, Bluetooth, etc.);

[0180] A bus 1005 , which transmits information between various components of the device (e.g., the processor 1001 , the memory 1002 , the input / output interface 1003 , and the communication interface 1004 );

[0181] The processor 1001 , the memory 1002 , the input / output interface 1003 and the communication interface 1004 are connected to each other in communication within the device via the bus 1005 .

[0182] An embodiment of the present application also provides a computer-readable storage medium, which stores a computer program. When the computer program is executed by a processor, the above-mentioned face recognition method is implemented.

[0183] The memory, as a non-transient computer-readable storage medium, can be used to store non-transient software programs and non-transient computer executable programs. In addition, the memory may include a high-speed random access memory, and may also include a non-transient memory, such as at least one disk storage device, a flash memory device, or other non-transient solid-state storage device. In some embodiments, the memory may optionally include a memory remotely disposed relative to the processor, and these remote memories may be connected to the processor via a network. Examples of the above-mentioned network include, but are not limited to, the Internet, an intranet, a local area network, a mobile communication network, and combinations thereof.

[0184] The face recognition method and apparatus, computer equipment and storage medium provided in the embodiments of the present application form a candidate model group by combining at least two different selected face recognition models, and extracting preliminary face features from face image sample data through each selected face recognition model, so that at least two different selected face recognition models can obtain at least two different preliminary face features, and the face fusion features output by each candidate model group are obtained by fusing at least two different preliminary face features. In order to evaluate the face feature fusion effect of the candidate model group, that is, to analyze the face recognition accuracy of each candidate model group, the face fusion features are evaluated through face identification information to obtain feature fusion evaluation data that characterizes the face recognition accuracy of the candidate model group. Therefore, a target model group with better recognition accuracy is screened out from the candidate model group through the feature fusion evaluation data, so that face recognition is performed on the target face image data through the target model group to improve the accuracy of face recognition.

[0185] The embodiments described in the embodiments of the present application are intended to more clearly illustrate the technical solutions of the embodiments of the present application and do not constitute a limitation on the technical solutions provided in the embodiments of the present application. Those skilled in the art will appreciate that with the evolution of technology and the emergence of new application scenarios, the technical solutions provided in the embodiments of the present application are also applicable to similar technical problems.

[0186] Those skilled in the art will appreciate that the technical solutions shown in the figures do not constitute a limitation on the embodiments of the present application, and may include more or fewer steps than shown in the figures, or a combination of certain steps, or different steps.

[0187] The device embodiments described above are merely illustrative, and the units described as separate components may or may not be physically separated, that is, they may be located in one place or distributed on multiple network units. Some or all of the modules may be selected according to actual needs to achieve the purpose of the solution of this embodiment.

[0188] Those skilled in the art will appreciate that all or some of the steps in the methods disclosed above, and the functional modules / units in the systems and devices may be implemented as software, firmware, hardware, or a suitable combination thereof.

[0189] The terms "first", "second", "third", "fourth", etc. (if any) in the specification of the present application and the above-mentioned drawings are used to distinguish similar objects, and are not necessarily used to describe a specific order or sequence. It should be understood that the data used in this way can be interchangeable where appropriate, so that the embodiments of the present application described herein can be implemented in an order other than those illustrated or described herein. In addition, the terms "including" and "having" and any of their variations are intended to cover non-exclusive inclusions, for example, a process, method, system, product or device comprising a series of steps or units is not necessarily limited to those steps or units clearly listed, but may include other steps or units that are not clearly listed or inherent to these processes, methods, products or devices.

[0190] It should be understood that in the present application, "at least one (item)" means one or more, and "plurality" means two or more. "And / or" is used to describe the association relationship of associated objects, indicating that three relationships may exist. For example, "A and / or B" can mean: only A exists, only B exists, and A and B exist at the same time, where A and B can be singular or plural. The character " / " generally indicates that the objects associated before and after are in an "or" relationship. "At least one of the following" or similar expressions refers to any combination of these items, including any combination of single or plural items. For example, at least one of a, b or c can mean: a, b, c, "a and b", "a and c", "b and c", or "a and b and c", where a, b, c can be single or multiple.

[0191] In the several embodiments provided in the present application, it should be understood that the disclosed devices and methods can be implemented in other ways. For example, the device embodiments described above are only schematic. For example, the division of the above units is only a logical function division. There may be other division methods in actual implementation, such as multiple units or components can be combined or integrated into another system, or some features can be ignored or not executed. Another point is that the mutual coupling or direct coupling or communication connection shown or discussed can be through some interfaces, indirect coupling or communication connection of devices or units, which can be electrical, mechanical or other forms.

[0192] The units described above as separate components may or may not be physically separated, and the components shown as units may or may not be physical units, that is, they may be located in one place or distributed on multiple network units. Some or all of the units may be selected according to actual needs to achieve the purpose of the solution of this embodiment.

[0193] In addition, each functional unit in each embodiment of the present application may be integrated into one processing unit, or each unit may exist physically separately, or two or more units may be integrated into one unit. The above-mentioned integrated unit may be implemented in the form of hardware or in the form of software functional units.

[0194] If the integrated unit is implemented in the form of a software functional unit and sold or used as an independent product, it can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of the present application, or the part that contributes to the prior art, or all or part of the technical solution can be embodied in the form of a software product, and the computer software product is stored in a storage medium, including multiple instructions to enable a computer device (which can be a personal computer, server, or network device, etc.) to execute all or part of the steps of the methods of various embodiments of the present application. The aforementioned storage medium includes: U disk, mobile hard disk, read-only memory (Read-Only Memory, referred to as ROM), random access memory (Random Access Memory, referred to as RAM), disk or optical disk and other media that can store programs.

[0195] The preferred embodiments of the present invention are described above with reference to the accompanying drawings, but the scope of the rights of the present invention is not limited thereto. Any modification, equivalent substitution and improvement made by a person skilled in the art without departing from the scope and essence of the present invention should be within the scope of the rights of the present invention.

Claims

1. A face recognition method, It is characterized in that The method comprises: Acquire a face image sample set; wherein the face image sample set includes: face image sample data containing sample objects and face identification information of the face image sample data, wherein the face identification information is used to identify whether the sample objects contained in the face image sample data are the same or different; By inputting the face image sample data into a preset candidate model group; wherein the candidate model group includes at least two different selected face recognition models; Performing feature extraction on the face image sample data by using each of the selected face recognition models to obtain at least two different preliminary face features; Fusing at least two different preliminary facial features to obtain a facial fusion feature of each candidate model group; Performing feature fusion evaluation on the face fusion feature according to the face identification information to obtain feature fusion evaluation data of the candidate model group; wherein the feature fusion evaluation data represents the recognition accuracy of the face fusion feature; Screening out a target model group from the candidate model group according to the feature fusion evaluation data; Face recognition is performed on the target face image data through the target model group.

2. The method according to claim 1, It is characterized in that The step of fusing at least two different preliminary facial features to obtain facial fusion features of each candidate model group includes: Filtering target dimension information from preset candidate dimension information; Performing dimensionality reduction processing on each of the preliminary facial features according to the target dimension information to obtain a low-dimensional facial feature; At least two different low-dimensional facial features are fused to obtain the facial fusion feature.

3. The method according to claim 2, It is characterized in that The step of fusing at least two different low-dimensional facial features to obtain the facial fusion feature includes: Performing attention calculation on each of the low-dimensional facial features to obtain an attention weight coefficient; wherein the attention weight coefficient represents the importance of the low-dimensional facial feature; At least two different low-dimensional facial features are fused according to the attention weight coefficient to obtain the facial fusion feature.

4. The method according to any one of claims 1 to 3, It is characterized in that The step of performing feature fusion evaluation on the face fusion features according to the face identification information to obtain feature fusion evaluation data of the candidate model group includes: Calculating the similarity of any two face fusion features to obtain fusion feature similarity; Obtaining the fusion feature similarity of the sample objects identified by the face identification information being the same, to obtain a first similarity; Obtaining the fusion feature similarity that the face identification information identifies that the sample objects are different, to obtain a second similarity; The first similarity and the second similarity are concatenated to obtain the feature fusion evaluation data of the candidate model group.

5. The method according to any one of claims 1 to 3, It is characterized in that The step of performing feature fusion evaluation on the face fusion features according to the face identification information to obtain feature fusion evaluation data of the candidate model group includes: Acquire a sample object of the face fusion feature according to the face identification information to obtain face verification information; Performing face recognition on the face fusion feature to obtain face recognition information; wherein the face recognition information represents the sample object matched by the face fusion feature; The face recognition information and the face verification information are subjected to loss calculation to obtain feature fusion evaluation data of the candidate model group.

6. The method according to any one of claims 1 to 3, It is characterized in that Before inputting the face image sample data into a preset candidate model group, the method further includes: Constructing the candidate model group specifically includes: Extracting features from the facial image sample data using a preset candidate facial recognition model to obtain at least two different candidate facial features; Calculating the similarity of any two different candidate facial features to obtain the candidate feature similarity; Selecting at least two different selected face recognition models from the candidate face recognition models according to the candidate feature similarity; At least two different selected face recognition models are combined to obtain the candidate model group.

7. The method according to claim 6, It is characterized in that The step of selecting at least two different selected face recognition models from the candidate face recognition models according to the candidate feature similarity comprises: Screening the candidate face recognition models according to the candidate feature similarities to obtain a preliminary face recognition model; Obtaining model structure information of each of the preliminary face recognition models; At least two different selected face recognition models are screened out from the preliminary face recognition models according to the model structure information.

8. A face recognition device, It is characterized in that The device comprises: A sample acquisition module, used to acquire a face image sample set; wherein the face image sample set includes: face image sample data containing sample objects and face identification information of the face image sample data, wherein the face identification information is used to identify whether the sample objects contained in the face image sample data are the same or different; A data input module, used for inputting the face image sample data into a preset candidate model group; wherein the candidate model group includes at least two different selected face recognition models; A feature extraction module, used for performing feature extraction on the face image sample data through each of the selected face recognition models to obtain at least two different preliminary face features; A fusion module, used for fusing at least two different preliminary facial features to obtain a facial fusion feature of each candidate model group; A feature evaluation module, used to perform feature fusion evaluation on the face fusion feature according to the face identification information to obtain feature fusion evaluation data of the candidate model group; wherein the feature fusion evaluation data represents the recognition accuracy of the face fusion feature; A screening module, used for screening out a target model group from the candidate model group according to the feature fusion evaluation data; The face recognition module is used to perform face recognition on the target face image data through the target model group.

9. A computer device, It is characterized in that The computer device includes a memory and a processor, the memory stores a computer program, and the processor implements the face recognition method according to any one of claims 1 to 7 when executing the computer program.

10. A computer-readable storage medium storing a computer program. It is characterized in that When the computer program is executed by a processor, the face recognition method according to any one of claims 1 to 7 is implemented.

Images