Facial recognition methods, devices, computer-readable storage media, and electronic devices

Abstract

This application discloses a face recognition method, apparatus, computer-readable storage medium, and electronic device, relating to the field of artificial intelligence technology. The method includes: acquiring facial features and a reference feature set corresponding to a target face image, the reference feature set including at least one reference facial feature; extracting target similarity information between the facial features and each reference facial feature, and reference similarity information between the reference facial features; constructing a similarity matrix representing the association between the target similarity information and the reference similarity information; performing information confidence analysis based on the similarity matrix to obtain the confidence level of the match between the facial features and the target reference facial feature among at least one reference facial feature; determining the reference facial feature that matches the facial feature from the target reference facial features based on the confidence level; and determining the recognition result of the target face image based on the matched reference facial feature. This application improves the accuracy of face recognition.

Description

Technical Field

[0001] This application relates to the field of artificial intelligence technology, specifically to a face recognition method, device, computer-readable storage medium, and electronic device. Background Technology

[0002] Face recognition is a technology that identifies objects based on facial images. During face recognition, facial features in the facial image are acquired, and then a large-scale similarity matching and recognition of preset facial features is performed.

[0003] Currently, related technologies obtain recognition results by matching the most similar preset facial features and making simple decisions on the similarity scores corresponding to these preset facial features. However, this approach fails to perceive changes in the overall similarity score distribution caused by the quality issues of the facial image itself, resulting in a low overall accuracy of facial image recognition. Summary of the Invention

[0004] This application provides a face recognition method and related apparatus, which can effectively reduce the computational complexity of face recognition while improving the face recognition effect.

[0005] To address the aforementioned technical problems, this application provides the following technical solutions:

[0006] According to one embodiment of this application, a face recognition method is provided. The method includes: acquiring facial features and a reference feature set corresponding to a target face image, wherein the reference feature set includes at least one reference facial feature; extracting target similarity information between the facial features and each of the reference facial features, and reference similarity information between the reference facial features; constructing a similarity matrix characterizing the association between the target similarity information and the reference similarity information; performing information confidence analysis based on the similarity matrix to obtain a confidence level that the facial features match a target reference facial feature among at least one reference facial feature; determining a reference facial feature that matches the facial features from the target reference facial features based on the confidence level; and determining the recognition result of the target face image based on the matched reference facial features.

[0007] According to one embodiment of this application, a face recognition device includes: an acquisition module for acquiring facial features and a reference feature set corresponding to a target face image, the reference feature set including at least one reference face feature; an extraction module for extracting target similarity information between the facial features and each of the reference face features, and reference similarity information between the reference face features; a construction module for constructing a similarity matrix characterizing the association between the target similarity information and the reference similarity information; an analysis module for performing information confidence analysis based on the similarity matrix to obtain a confidence level of matching between the facial features and a target reference face feature among at least one reference face feature; a matching module for determining a reference face feature that matches the facial features from the target reference face features based on the confidence level; and a recognition module for determining the recognition result of the target face image based on the matched reference face features.

[0008] In some embodiments of this application, the acquisition module includes: a feature library acquisition unit, used to acquire facial features corresponding to a target facial image and a recognition feature library, wherein the recognition feature library includes at least one preset facial feature; a similarity calculation unit, used to calculate the similarity between the facial feature and each preset facial feature in the recognition feature library; and a feature set acquisition unit, used to acquire preset facial features ranked high in similarity from the recognition feature library as reference facial features to form the reference feature set.

[0009] In some embodiments of this application, the extraction module includes: a first score calculation unit, used to calculate the similarity score between the facial feature and each of the reference facial features to obtain the target similarity information; and a second score calculation unit, used to calculate the similarity score between the reference facial features to obtain the reference similarity information.

[0010] In some embodiments of this application, the extraction module includes: a first feature extraction unit, used to extract feature difference information between the facial features and each of the reference facial features to obtain the target similarity information; and a second feature extraction unit, used to extract feature difference information between the reference facial features to obtain the reference similarity information.

[0011] In some embodiments of this application, the construction module includes: a similarity acquisition unit, used to acquire the similarity between the facial feature and each of the reference facial features; a sorting unit, used to sort the facial feature and each of the reference facial features according to the similarity; and an adjacency matrix construction unit, used to construct an adjacency matrix between the target similarity information and the reference similarity information according to the sorting, thereby forming the similarity matrix.

[0012] In some embodiments of this application, the analysis module includes: a fusion encoding unit, used to perform fusion encoding processing on the similarity matrix to obtain fusion features; and a confidence analysis unit, used to perform confidence analysis based on the fusion features to obtain the confidence level of the facial features matching a target reference facial feature among at least one reference facial features.

[0013] In some embodiments of this application, the confidence analysis unit includes: a fully connected processing subunit, used to perform fully connected operations on the fused features to obtain fully connected features; and a classification processing subunit, used to perform classification processing based on the fully connected features to obtain the confidence level of the facial features matching a target reference facial feature among at least one reference facial features.

[0014] In some embodiments of this application, the target reference facial feature includes a reference facial feature with the highest similarity to the facial feature; the classification processing subunit is used to: perform classification processing based on the fully connected feature to obtain the confidence level of the reference facial feature with the highest similarity matching the facial feature.

[0015] In some embodiments of this application, the target reference facial features include at least two reference facial features with the highest similarity to the facial features; the classification processing subunit is used to: perform classification processing based on the fully connected features to obtain the confidence scores of the at least two reference facial features with the highest similarity matching the facial features.

[0016] In some embodiments of this application, the fusion coding unit is used to: perform convolution processing on the similarity matrix to generate the fusion feature.

[0017] In some embodiments of this application, the fusion encoding unit is used to: perform self-attention encoding on the similarity matrix to generate the fusion feature.

[0018] In some embodiments of this application, the matching module includes: a matching feature determination unit, configured to determine reference facial features whose confidence level meets the recognition conditions from the target reference facial features; and a matching feature determination unit, configured to use the reference facial features whose confidence level meets the recognition conditions as reference facial features that match the facial features.

[0019] According to another embodiment of this application, a computer-readable storage medium is provided, on which a computer program is stored, which, when executed by a computer's processor, causes the computer to perform the methods described in the embodiments of this application.

[0020] According to another embodiment of this application, an electronic device includes: a memory storing a computer program; and a processor reading the computer program stored in the memory to execute the methods described in the embodiments of this application.

[0021] According to another embodiment of this application, a computer program product or computer program includes computer instructions stored in a computer-readable storage medium. A processor of a computer device reads the computer instructions from the computer-readable storage medium and executes the computer instructions, causing the computer device to perform the methods provided in the various optional implementations described in the embodiments of this application.

[0022] In this embodiment, facial features and a reference feature set corresponding to the target facial image are obtained, the reference feature set including at least one reference facial feature; target similarity information between the facial features and each reference facial feature, and reference similarity information between the reference facial features are extracted; a similarity matrix representing the association between the target similarity information and the reference similarity information is constructed; information confidence analysis is performed based on the similarity matrix to obtain the confidence level of the match between the facial features and the target reference facial feature among at least one reference facial feature; reference facial features that match the facial features are determined from the target reference facial features according to the confidence level; and the recognition result of the target facial image is determined based on the matched reference facial features.

[0023] In this way, not only is the target similarity information of the facial features of the target face image and each reference facial feature utilized, but also the reference similarity information between the reference facial features is combined. Based on the similarity matrix that represents the relationship between the target similarity information and the reference similarity information, information confidence analysis is performed. This can assess the errors that exist when matching the facial features with each reference facial feature. For example, if the quality of the target face image itself is problematic, it may lead to the identification of a mismatched reference facial feature as a match, resulting in an error in the overall similarity distribution. Finally, the confidence of matching the facial feature with at least one target reference facial feature can be accurately assessed. Based on the confidence, the reference facial features that match the facial feature can be determined for face recognition, effectively improving the accuracy of face recognition. Attached Figure Description

[0024] To more clearly illustrate the technical solutions in the embodiments of this application, the accompanying drawings used in the description of the embodiments will be briefly introduced below. Obviously, the accompanying drawings described below are only some embodiments of this application. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0025] Figure 1A schematic diagram of a system to which embodiments of this application can be applied is shown.

[0026] Figure 2 A flowchart of a face recognition method according to an embodiment of this application is shown.

[0027] Figure 3 A flowchart of a feature set acquisition method according to an embodiment of this application is shown.

[0028] Figure 4 A flowchart illustrating the acquisition of similarity information according to an embodiment of this application is shown.

[0029] Figure 5 A flowchart of confidence analysis according to an embodiment of this application is shown.

[0030] Figure 6 A schematic diagram of a similarity matrix according to an embodiment of this application is shown.

[0031] Figure 7 A flowchart of confidence analysis according to an embodiment of this application is shown.

[0032] Figure 8 A flowchart of confidence analysis according to an embodiment of this application is shown.

[0033] Figure 9 A block diagram of a face recognition device according to another embodiment of this application is shown.

[0034] Figure 10 A block diagram of an electronic device according to an embodiment of this application is shown. Detailed Implementation

[0035] The technical solutions of the embodiments of this application will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of this application, and not all embodiments. Based on the embodiments of this application, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of this application.

[0036] In the following description, specific embodiments of this application will be illustrated with reference to steps and symbols performed by one or more computers, unless otherwise stated. Therefore, these steps and operations will be referred to several times as being performed by a computer, and computer execution as referred to herein includes operations by a computer processing unit representing electronic signals of data in a structured format. This operation transforms the data or maintains it at a location in the computer's memory system, which can be reconfigured or otherwise alter the operation of the computer in a manner well known to those skilled in the art. The data structure maintained by the data is the physical location of the memory, which has specific characteristics defined by the data format. However, the principles of this application are described in the foregoing text, which is not intended to be limiting, and those skilled in the art will understand that many of the steps and operations described below can also be implemented in hardware.

[0037] Figure 1 A schematic diagram of a system 100 to which embodiments of this application can be applied is shown. For example... Figure 1 As shown, system 100 may include server 101 and terminal 102. Server 101 and terminal 102 can be connected directly or indirectly via wireless communication, and this application does not impose any special restrictions on this connection.

[0038] Server 101 can be a standalone physical server, 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 communication, middleware services, domain name services, security services, CDN, and big data and artificial intelligence platforms.

[0039] In one implementation of this example, server 101 is a cloud server that can provide artificial intelligence cloud services, such as facial recognition AI cloud services. AI cloud services are generally also known as AIaaS (AI as a Service). This is a mainstream service model for AI platforms. Specifically, AIaaS platforms break down several common AI services and provide them as independent or packaged services in the cloud. This service model is similar to opening an AI-themed marketplace: all developers can access and use one or more AI services provided by the platform through API interfaces. Some experienced developers can also use the AI ​​framework and AI infrastructure provided by the platform to deploy and maintain their own dedicated cloud AI services.

[0040] In one implementation, server 101 can be a node in a blockchain system, which can record facial recognition data in the form of a blockchain. The blockchain includes a series of blocks that are sequentially generated. Once a new block is added to the blockchain, it will not be removed. The blocks record the data submitted by the nodes in the blockchain system.

[0041] Terminal 102 can be any device, including but not limited to mobile phones, computers, smart interactive devices, smart home appliances, in-vehicle terminals, VR / AR devices, smartwatches, merchant terminals, and computers, etc.

[0042] In one embodiment of this example, server 101 can acquire facial features and a reference feature set corresponding to a target facial image, the reference feature set including at least one reference facial feature; extract target similarity information between the facial features and each reference facial feature, as well as reference similarity information between the reference facial features; construct a similarity matrix representing the association between the target similarity information and the reference similarity information; perform information confidence analysis based on the similarity matrix to obtain the confidence level of the match between the facial features and the target reference facial feature among at least one reference facial feature; determine the reference facial feature that matches the facial feature from the target reference facial features according to the confidence level; and determine the recognition result of the target facial image based on the matched reference facial feature. The target facial image may be collected by terminal 102 and uploaded to server 101.

[0043] The embodiments of the facial recognition method in this application can be applied to any scenario that requires facial recognition, such as facial recognition payment, facial recognition for public transportation, and facial recognition for access control.

[0044] Taking facial recognition payment as an example, in related technologies, facial recognition terminals (e.g.) Figure 1 The terminal 102 shown acquires the target face image corresponding to the face scan and uploads it to the server. The server extracts the facial features corresponding to the target face image; then, it performs similarity matching between these facial features and each preset facial feature in the recognition feature library to obtain a similarity score for each preset facial feature; finally, it simply selects the associated information (e.g., identity information) of a preset facial feature as the recognition result based on the similarity score and performs associated payment. For example, typically, the associated information (e.g., identity information) of the preset facial feature with the highest similarity score is directly selected for associated payment.

[0045] In this situation, if the target facial image has problems, the technology cannot perceive the changes in the overall similarity score distribution caused by the quality issues of the facial image itself, leading to errors in target facial image recognition and incorrect payment processing. For example, if the target facial image has certain problems, it may incorrectly result in a facial feature having the highest similarity to a certain preset facial feature. However, it is highly likely that the preset facial feature with the second highest similarity score is the actual corresponding feature.

[0046] The face recognition method based on this application not only utilizes the target similarity information between the facial features of the target face image and each reference face feature, but also combines the reference similarity information between the reference face features. Based on a similarity matrix representing the relationship between the target similarity information and the reference similarity information, information confidence analysis is performed. This allows for the assessment of errors in the similarity matching between the face feature and each reference face feature. For example, if the quality of the target face image itself is problematic, it may lead to the misidentification of a non-matching reference face feature as a match, resulting in an error in the overall similarity distribution. Ultimately, the confidence of the match between the face feature and at least one target reference face feature is accurately assessed. Based on the confidence, the reference face feature that matches the face feature can be determined for face recognition (for example, based on the confidence, the preset face feature with the second-highest similarity score can be reliably identified as the true corresponding feature), effectively improving the accuracy of face recognition and thus the accuracy of face-scanning payment.

[0047] The following is a detailed description of embodiments of the face recognition method provided in this application.

[0048] Figure 2 A flowchart illustrating a face recognition method according to an embodiment of this application is shown schematically. The entity performing this face recognition method can be any terminal, such as... Figure 1 The server 101 or terminal 102 shown are examples of such servers.

[0049] like Figure 2 As shown, the face recognition method may include steps S210 to S260.

[0050] Step S210: Obtain the facial features and reference feature set corresponding to the target face image. The reference feature set includes at least one reference facial feature.

[0051] Step S220: Extract the target similarity information between the facial features and each reference facial feature, as well as the reference similarity information between the reference facial features;

[0052] Step S230: Construct a similarity matrix that represents the relationship between target similarity information and reference similarity information;

[0053] Step S240: Based on the similarity matrix, perform information confidence analysis to obtain the confidence level of the facial feature matching the target reference facial feature in at least one reference facial feature;

[0054] Step S250: Determine the reference facial features that match the facial features from the target reference facial features based on the confidence level;

[0055] Step S260: Determine the recognition result of the target face image based on the matched reference facial features.

[0056] The following describes the specific steps involved in facial recognition.

[0057] In step S210, the facial features and reference feature set corresponding to the target face image are obtained, and the reference feature set includes at least one reference facial feature.

[0058] In this example implementation, the target face image is the face image to be identified, such as the face image of the object to be identified captured by the terminal. Facial features can be extracted from the target face image through facial feature point detection and feature extraction; these facial features can be feature vectors. The reference feature set includes at least one reference facial feature, which is a pre-collected set of facial features, and each reference facial feature can correspond to an object's identity information. By determining the reference facial feature that matches the facial feature from the reference feature set, the target face image can be identified, and the identity of the object corresponding to the target face image can be determined.

[0059] In one embodiment, see Figure 3 Step S310: Obtain the facial features and reference feature set corresponding to the target face image, including:

[0060] Step S310: Obtain the facial features and recognition feature library corresponding to the target face image. The recognition feature library includes at least one preset facial feature. Step S320: Calculate the similarity between the facial features and each preset facial feature in the recognition feature library. Step S330: Obtain the preset facial features ranked higher than a predetermined number in similarity from the recognition feature library as reference facial features to form a reference feature set.

[0061] The recognition feature library includes at least one preset facial feature, which is a pre-collected facial feature. Each preset facial feature can correspond to an object's identity information.

[0062] By calculating the similarity between facial features and each preset facial feature in the recognition feature library, a predetermined number of preset facial features with the highest similarity to the facial features (i.e. preset facial features with a similarity ranking above a predetermined number) are selected from the recognition feature library as reference facial features to form a reference feature set. Further matching analysis can be performed based on some of the reference facial features with the highest similarity.

[0063] In step S220, target similarity information between facial features and each reference facial feature, as well as reference similarity information between reference facial features, are extracted.

[0064] In this example implementation, target similarity information can describe the similarity between two features. The similarity information can be a similarity score (the higher the similarity score, the more similar the two features are) or feature difference information (the smaller the feature difference, the more similar the two features are).

[0065] For facial features, target similarity information can be extracted between the facial feature and each reference facial feature. For example, if there are N reference facial features, then N target similarity information points will be extracted. The target similarity information between the facial feature and each reference facial feature can be used to preliminarily evaluate the degree of matching between the facial feature and the reference facial features.

[0066] Extracting the reference similarity information between every two reference facial features can further describe the association information between reference facial features in the reference feature set, and provide in-depth assistance to the target similarity information between facial features and each reference facial feature, so as to further evaluate the matching degree between facial features and reference facial features.

[0067] In one embodiment, see Figure 4 Step S220: Extract the target similarity information between the facial features and each reference facial feature, as well as the reference similarity information between the reference facial features, including:

[0068] Step S410: Calculate the similarity score between the facial feature and each reference facial feature to obtain target similarity information; Step S420: Calculate the similarity score between the reference facial features to obtain reference similarity information.

[0069] Similarity scores can be calculated using methods such as Euclidean distance or cosine similarity. In one example, for facial feature *a* and a reference facial feature *b*, the similarity score can be calculated using the formula cos(a, b) = a * b / (norm(a) * norm(b)) where cos(a, b) is the similarity score and norm represents normalization. Similarly, for reference facial feature *b* and reference facial feature *c*, the similarity score can be calculated using the formula cos(b, c) = b * c / (norm(b) * norm(c)) where cos(b, c) is the similarity score and norm represents normalization. It's also possible to obtain the highest similarity score (e.g., 1) between a facial feature and itself, and also the highest similarity score (e.g., 1) between a reference facial feature and itself.

[0070] In one embodiment, extracting target similarity information between facial features and each reference facial feature, as well as reference similarity information between reference facial features, includes:

[0071] Extract the feature difference information between the facial features and each reference facial feature to obtain the target similarity information; extract the feature difference information between the reference facial features to obtain the reference similarity information.

[0072] Feature difference information can inversely represent the similarity between two features; that is, the smaller the feature difference, the more similar the two features are. Feature difference information can be extracted by calculating the feature difference between two features. Facial features and each reference facial feature can be normalized into normalized feature values. Then, the difference between these normalized feature values ​​is calculated to obtain the feature difference information. It can also be understood that the feature difference information between a facial feature and itself is 0, and the feature difference information between a reference facial feature and itself is also 0.

[0073] In step S230, a similarity matrix is ​​constructed to represent the relationship between the target similarity information and the reference similarity information.

[0074] In this example implementation, each facial feature corresponds to a target similarity information, and every two reference facial features correspond to a reference similarity information. By using all target similarity information and reference similarity information as matrix elements and combining them in a specific order (e.g., the similarity between facial features and each reference facial feature) to construct a matrix, a similarity matrix representing the association between target similarity information and reference similarity information can be formed.

[0075] In one embodiment, step S230, constructing a similarity matrix representing the association between target similarity information and reference similarity information, includes:

[0076] Obtain the similarity between the facial feature and each reference facial feature; sort the facial feature and each reference facial feature according to the similarity; construct an adjacency matrix between the target similarity information and the reference similarity information based on the sorting, thus forming the similarity matrix.

[0077] The facial features are sorted according to their similarity to each reference facial feature. These features can be sorted in descending order of similarity score.

[0078] For example, see Figure 6 There is one facial feature (query) and N reference facial features (top1-topN). The facial feature (query) is ranked first, the reference facial feature with the highest similarity to the facial feature (top1) is ranked second, and the reference facial feature with the lowest similarity to the facial feature (topN) is ranked last.

[0079] An adjacency matrix is ​​a relational matrix obtained by sorting facial features by similarity, assigning each facial feature to a row and a column of a matrix, and filling in the relationships between features (i.e., target similarity information or reference similarity information).

[0080] For example, see Figure 6 If query is the first one, then query corresponds to the 1st row and 1st column of the matrix, and topN corresponds to the N+1th row and N+1th column of the adjacency matrix. That is, the dimension of the adjacency matrix is ​​N+1 (N+1 rows, N+1 columns).

[0081] When filling in the relationships between features, the similarity information between the query and itself can be 1, the target similarity information between the query and top1 is S1, and the reference similarity information between top1 and topN is top1-topN. Based on the correspondence between the query and top1-topN, and the correspondence between top1-topN themselves, the corresponding similarity information is filled in to obtain the following: Figure 5 The adjacency matrix (i.e., similarity matrix) shown has 1s on the diagonal.

[0082] In this similarity matrix, the similarity information between adjacent columns in the same row is closely related, and similarly, the similarity information between adjacent rows in the same column is closely related. Adjacent similarity information has a certain correlation. For example, comparing the target similarity information of query and top1 and top2 in the first row can often show whether top1 is a matching reference facial feature (i.e., whether it is the facial feature of the object corresponding to the pre-collected target facial image). Normally, if top1 is a matching reference facial feature, there will be a gap in the target similarity information between top1 and top2 and query. For example, if top1 is a matching reference facial feature of query, then the comparison of the first two rows in the third column will also be very close, and so on.

[0083] In this way, the similarity information between all features (including facial features and each reference facial feature) corresponds to a position in the matrix, accurately depicting the relationship between each feature (including facial features and each reference facial feature). The similarity matrix has complete similarity information corresponding to each feature (including facial features and each reference facial feature). For example, if a reference facial feature has an abnormally high similarity to a facial feature due to noise or other unexpected reasons, it can be effectively reflected in the similarity matrix. The similarity matrix is ​​very robust, and accurate similarity association information between features can be extracted based on this similarity matrix.

[0084] In step S240, information confidence analysis is performed based on the similarity matrix to obtain the confidence level of the facial feature matching the target reference facial feature among at least one reference facial feature.

[0085] In this example implementation, information confidence analysis based on the similarity matrix can be performed by extracting the similarity correlation information between features in the similarity matrix, and then conducting confidence analysis based on the extracted similarity correlation information to obtain the confidence score of the match between the target reference facial feature and the facial feature. When extracting the similarity correlation information between features in the similarity matrix, the similarity matrix can be fused and encoded to generate fused features representing the similarity correlation information between features, and the confidence score of the match between the facial feature and at least one target reference facial feature can be accurately performed based on the fused features.

[0086] In one embodiment, see Figure 5 Step S240: Based on the similarity matrix, perform information confidence analysis to obtain the confidence level of the facial feature matching the target reference facial feature in at least one reference facial feature, including:

[0087] Step S510: Perform fusion encoding on the similarity matrix to obtain fusion features; Step S520: Perform confidence analysis based on the fusion features to obtain the confidence of the facial features matching the target reference facial features in at least one reference facial feature.

[0088] Fusion coding can be used to extract fused features that accurately capture the similarity relationships between representational features.

[0089] Confidence analysis can be performed based on fully connected networks and classifiers to analyze matching probability. The matching probability is used as confidence score. Based on the fusion features, confidence analysis can accurately obtain the confidence score of a facial feature matching a target reference facial feature among at least one reference facial feature.

[0090] In one embodiment, the similarity matrix is ​​fused and encoded to obtain fused features, including: performing convolution processing on the similarity matrix to generate fused features.

[0091] In this embodiment, the similarity matrix is ​​fused and encoded based on convolutional processing. This involves performing multi-layer convolution and pooling on the similarity matrix using a convolutional network, resulting in a feature map with a sufficiently large receptive field for fusion. This convolutional processing approach efficiently and accurately obtains the fused features. (See also...) Figure 7 and Figure 8 In one example, a convolutional network can be designed consisting of convolutional layer 1 + pooling layer 1 + convolutional layer 2 + pooling layer 1. The similarity matrix is ​​processed by two convolutions and two pooling operations to obtain the final feature map as the fused feature. The kernel size during convolution can be set according to requirements, such as 2x2. Padding operations can be performed on the similarity matrix to keep the size of the feature map obtained from the convolution process unchanged; for example, the output feature map size can be N / 4 * N / 4.

[0092] In one embodiment, the similarity matrix is ​​fused and encoded to obtain fused features, including: performing self-attention encoding on the similarity matrix to generate fused features.

[0093] In this embodiment, the similarity matrix is ​​fused and encoded based on self-attention encoding. The similarity matrix can be fused and encoded using a self-attention mechanism encoder (SA Encoder) to obtain a self-attention encoded sequence that effectively represents the contextual association information between similarity information as a fusion feature.

[0094] In one embodiment, confidence analysis is performed based on fused features to obtain the confidence level of a facial feature matching a target reference facial feature among at least one reference facial feature, including:

[0095] The fused features are processed by a fully connected operation to obtain fully connected features; classification is performed based on the fully connected features to obtain the confidence score of the facial features matching the target reference facial features in at least one reference facial feature.

[0096] After obtaining the fusion features, a fully connected operation is performed on the fusion features. Based on the fusion features, global fully connected features with similarity information can be extracted. Finally, a classification layer can be used to perform classification processing based on the fully connected features, and the matching probability of the face feature matching the target reference face feature among at least one reference face feature is used as the confidence level.

[0097] In one embodiment, the target reference facial feature includes a reference facial feature with the highest similarity to the facial feature; classification processing is performed based on the fully connected feature to obtain the confidence level of the facial feature matching the target reference facial feature among at least one reference facial feature, including: classification processing is performed based on the fully connected feature to obtain the confidence level of the reference facial feature with the highest similarity matching the facial feature.

[0098] The reference feature set includes target reference facial features, which are the identified reference facial features with the highest similarity to the given facial features. See also... Figure 7 The fully connected features are classified based on the classification layer, and the reference facial feature with the highest similarity is output (e.g., Figure 6 The matching probability (P(top1)) of the reference facial feature shown in the figure is used as the confidence level. In this way, only the reference facial feature with the highest similarity to the facial feature is selected, and all similarity information is combined for in-depth evaluation.

[0099] In one embodiment, the target reference facial features include at least two reference facial features with the highest similarity to the facial features; classification processing based on fully connected features to obtain the confidence level of the facial features matching the target reference facial features among at least one reference facial feature includes: classification processing based on fully connected features to obtain the confidence level of the matching between the at least two reference facial features with the highest similarity and the facial features.

[0100] The reference feature set includes target reference facial features, which consist of at least two identified reference facial features with the highest similarity to the target facial features. See also... Figure 8 The fully connected features are classified based on the classification layer, and the output is at least two reference facial features with the highest similarity (e.g., Figure 6The matching probability (P(top1-topn)) between the reference facial features (top1 to topN) shown in the figure and the facial features is used as the confidence level. In this way, only some reference facial features with the highest similarity to the facial features are selected. By combining all similarity information, a deep evaluation can be performed to determine whether, for example, reference facial feature top2, etc., truly matches the facial features.

[0101] In step S250, a reference facial feature that matches the facial feature is determined from the target reference facial features based on the confidence level.

[0102] In this example implementation, when the reference facial feature with the highest similarity is output (e.g., Figure 6 When the matching probability (P(top1)) of the reference facial feature shown in the figure is used as the confidence level, the reference facial feature with the highest similarity among the target reference facial features can be determined as the reference facial feature that matches the facial feature when the confidence level is higher than a predetermined threshold. When at least two reference facial features with the highest similarity are output (e.g., Figure 6 When the matching probability (P(top1-topn)) of the reference facial features shown in the figure is used as the confidence level, the reference facial feature with the highest confidence level can be determined as the reference facial feature that matches the facial feature.

[0103] In one embodiment, determining a reference facial feature that matches the facial feature from the target reference facial features based on confidence level includes: determining a reference facial feature whose confidence level meets the recognition criteria from the target reference facial features; and using the reference facial feature whose confidence level meets the recognition criteria as the reference facial feature that matches the facial feature.

[0104] When outputting the reference facial feature with the highest similarity (e.g. Figure 6 When the matching probability (P(top1)) of the reference facial feature shown in the diagram is used as the confidence level, the recognition condition can be that the confidence level is higher than a predetermined threshold. In this case, when the confidence level is higher than the predetermined threshold, the reference facial feature with the highest similarity among the target reference facial features can be determined as the reference facial feature that matches the facial feature. When outputting at least two reference facial features with the highest similarity (e.g., Figure 6 When the matching probability (P(top1-topn)) between the reference facial features (top1 to topN) shown in the figure and the facial features is used as the confidence level, the identification condition can be the highest confidence level. In this case, the reference facial feature with the highest confidence level can be determined as the reference facial feature that matches the facial feature.

[0105] In step S260, the recognition result of the target face image is determined based on the matched reference face features.

[0106] In this example implementation, each reference facial feature can correspond to an object identity information. By determining the object identity information corresponding to the reference facial feature that matches the facial feature from the reference feature set, the object identity information of the object corresponding to the target facial image can be determined, and the recognition result can be obtained.

[0107] In this way, based on steps S210 to S260, when performing face recognition, not only is the target similarity information of the facial features of the target face image and each reference face feature utilized, but also the reference similarity information between the reference face features is combined. Based on the similarity matrix that represents the relationship between the target similarity information and the reference similarity information, information confidence analysis is performed. This can assess the error in the similarity matching between the face feature and each reference face feature. For example, if the quality of the target face image itself is problematic, it may lead to the identification of a mismatched reference face feature as a match, resulting in an error in the overall similarity distribution. Finally, the confidence of the face feature matching the target reference face feature among at least one reference face feature is accurately assessed. Based on the confidence, the reference face feature that matches the face feature can be determined for face recognition, effectively improving the accuracy of face recognition.

[0108] To facilitate better implementation of the face recognition method provided in this application, this application also provides a face recognition device based on the above-described face recognition method. The meanings of the terms used are the same as in the face recognition method described above, and specific implementation details can be found in the descriptions within the method embodiments. Figure 9 A block diagram of a face recognition device according to an embodiment of this application is shown.

[0109] like Figure 9 As shown, the face recognition device 600 may include an acquisition module 610, an extraction module 620, a construction module 630, an analysis module 640, a matching module 650, and a recognition module 660.

[0110] The acquisition module 610 can be used to acquire facial features and a reference feature set corresponding to the target facial image, wherein the reference feature set includes at least one reference facial feature; the extraction module 620 can be used to extract target similarity information between the facial features and each of the reference facial features, as well as reference similarity information between the reference facial features; the construction module 630 can be used to construct a similarity matrix representing the association between the target similarity information and the reference similarity information; the analysis module 640 can be used to perform information confidence analysis based on the similarity matrix to obtain the confidence level of the match between the facial features and the target reference facial features among at least one reference facial features; the matching module 650 can be used to determine the reference facial features that match the facial features from the target reference facial features according to the confidence level; and the recognition module 660 can be used to determine the recognition result of the target facial image based on the matched reference facial features.

[0111] In some embodiments of this application, the acquisition module 610 includes: a feature library acquisition unit, used to acquire facial features corresponding to a target facial image and a recognition feature library, wherein the recognition feature library includes at least one preset facial feature; a similarity calculation unit, used to calculate the similarity between the facial feature and each preset facial feature in the recognition feature library; and a feature set acquisition unit, used to acquire preset facial features ranked higher than a predetermined number in similarity from the recognition feature library as reference facial features to form the reference feature set.

[0112] In some embodiments of this application, the extraction module 620 includes: a first score calculation unit, used to calculate the similarity score between the facial feature and each of the reference facial features to obtain the target similarity information; and a second score calculation unit, used to calculate the similarity score between the reference facial features to obtain the reference similarity information.

[0113] In some embodiments of this application, the extraction module 620 includes: a first feature extraction unit, used to extract feature difference information between the facial features and each of the reference facial features to obtain the target similarity information; and a second feature extraction unit, used to extract feature difference information between the reference facial features to obtain the reference similarity information.

[0114] In some embodiments of this application, the construction module 630 includes: a similarity acquisition unit, used to acquire the similarity between the facial feature and each of the reference facial features; a sorting unit, used to sort the facial feature and each of the reference facial features according to the similarity; and an adjacency matrix construction unit, used to construct an adjacency matrix between the target similarity information and the reference similarity information according to the sorting, thereby forming the similarity matrix.

[0115] In some embodiments of this application, the analysis module 640 includes: a fusion encoding unit, used to perform fusion encoding processing on the similarity matrix to obtain fusion features; and a confidence analysis unit, used to perform confidence analysis based on the fusion features to obtain the confidence level of the facial features matching a target reference facial feature among at least one reference facial features.

[0116] In some embodiments of this application, the confidence analysis unit includes: a fully connected processing subunit, used to perform fully connected operations on the fused features to obtain fully connected features; and a classification processing subunit, used to perform classification processing based on the fully connected features to obtain the confidence level of the facial features matching a target reference facial feature among at least one reference facial features.

[0117] In some embodiments of this application, the target reference facial feature includes a reference facial feature with the highest similarity to the facial feature; the classification processing subunit is used to: perform classification processing based on the fully connected feature to obtain the confidence level of the reference facial feature with the highest similarity matching the facial feature.

[0118] In some embodiments of this application, the target reference facial features include at least two reference facial features with the highest similarity to the facial features; the classification processing subunit is used to: perform classification processing based on the fully connected features to obtain the confidence scores of the at least two reference facial features with the highest similarity matching the facial features.

[0119] In some embodiments of this application, the fusion coding unit is used to: perform convolution processing on the similarity matrix to generate the fusion feature.

[0120] In some embodiments of this application, the fusion encoding unit is used to: perform self-attention encoding on the similarity matrix to generate the fusion feature.

[0121] In some embodiments of this application, the matching module 650 includes: a matching feature determination unit, configured to determine reference facial features whose confidence level meets the recognition conditions from the target reference facial features; and a matching feature determination unit, configured to use the reference facial features whose confidence level meets the recognition conditions as reference facial features that match the facial features.

[0122] In this way, based on the face recognition device 600, when performing face recognition, it not only utilizes the target similarity information between the facial features of the target face image and each reference face feature, but also combines the reference similarity information between the reference face features. Based on the similarity matrix that represents the relationship between the target similarity information and the reference similarity information, information confidence analysis is performed. This can assess the errors that exist when matching the similarity between the face feature and each reference face feature. For example, if the quality of the target face image itself is problematic, it may lead to the identification of a mismatched reference face feature as a match, resulting in an error in the overall similarity distribution. Finally, the confidence of the face feature matching the target reference face feature among at least one reference face feature is accurately assessed. Based on the confidence, the reference face feature that matches the face feature can be determined for face recognition, effectively improving the accuracy of face recognition.

[0123] It should be noted that although several modules or units for the device used to perform actions have been mentioned in the detailed description above, this division is not mandatory. In fact, according to the embodiments of this application, the features and functions of two or more modules or units described above can be embodied in one module or unit. Conversely, the features and functions of one module or unit described above can be further divided and embodied by multiple modules or units.

[0124] Furthermore, embodiments of this application also provide an electronic device, which can be a terminal or a server, such as... Figure 10 As shown, it illustrates a structural schematic diagram of the electronic device involved in the embodiments of this application, specifically:

[0125] The electronic device may include components such as a processor 701 with one or more processing cores, a memory 702 with one or more computer-readable storage media, a power supply 703, and an input unit 704. Those skilled in the art will understand that... Figure 10 The electronic device structure shown does not constitute a limitation on the electronic device and may include more or fewer components than shown, or combine certain components, or have different component arrangements. Wherein:

[0126] The processor 701 is the control center of the electronic device. It connects to various parts of the computer device via various interfaces and lines. By running or executing software programs and / or modules stored in the memory 702, and by calling data stored in the memory 702, it performs various functions of the computer device and processes data, thereby performing overall detection of the electronic device. Optionally, the processor 701 may include one or more processing cores; preferably, the processor 701 may integrate an application processor and a modem processor, wherein the application processor mainly handles the operating system, user page, and application programs, and the modem processor mainly handles wireless communication. It is understood that the aforementioned modem processor may not be integrated into the processor 701.

[0127] The memory 702 can be used to store software programs and modules. The processor 701 executes various functional applications and data processing by running the software programs and modules stored in the memory 702. The memory 702 may mainly include a program storage area and a data storage area. The program storage area may store the operating system, application programs required for at least one function (such as sound playback function, image playback function, etc.), etc.; the data storage area may store data created according to the use of the computer device, etc. In addition, the memory 702 may include high-speed random access memory, and may also include non-volatile memory, such as at least one disk storage device, flash memory device, or other volatile solid-state storage device. Accordingly, the memory 702 may also include a memory controller to provide the processor 701 with access to the memory 702.

[0128] The electronic device also includes a power supply 703 that supplies power to the various components. Preferably, the power supply 703 can be logically connected to the processor 701 through a power management system, thereby enabling functions such as charging, discharging, and power consumption management through the power management system. The power supply 703 may also include one or more DC or AC power supplies, recharging systems, power fault detection circuits, power converters or inverters, power status indicators, and other arbitrary components.

[0129] The electronic device may also include an input unit 704, which can be used to receive input digital or character information and generate keyboard, mouse, joystick, optical or trackball signal inputs related to user settings and function control.

[0130] Although not shown, the electronic device may also include a display unit, etc., which will not be described in detail here. Specifically, in this embodiment, the processor 701 in the electronic device loads the executable files corresponding to the processes of one or more computer programs into the memory 702 according to the following instructions, and the processor 701 runs the computer programs stored in the memory 702 to realize various functions, such as the processor 701 executing:

[0131] The process involves: acquiring facial features and a reference feature set corresponding to a target facial image, wherein the reference feature set includes at least one reference facial feature; extracting target similarity information between the facial features and each of the reference facial features, as well as reference similarity information between the reference facial features; constructing a similarity matrix representing the association between the target similarity information and the reference similarity information; performing information confidence analysis based on the similarity matrix to obtain the confidence level of the match between the facial features and the target reference facial features among at least one reference facial feature; determining reference facial features that match the facial features from the target reference facial features based on the confidence level; and determining the recognition result of the target facial image based on the matched reference facial features.

[0132] In one embodiment, when acquiring the facial features and reference feature set corresponding to the target facial image, the processor 701 may perform the following: acquiring the facial features and recognition feature library corresponding to the target facial image, wherein the recognition feature library includes at least one preset facial feature; calculating the similarity between the facial feature and each preset facial feature in the recognition feature library; and acquiring preset facial features ranked higher than a predetermined similarity level from the recognition feature library as reference facial features to form the reference feature set.

[0133] In one embodiment, when extracting the target similarity information between the facial feature and each of the reference facial features, and the reference similarity information between the reference facial features, the processor 701 may perform the following: calculating the similarity score between the facial feature and each of the reference facial features to obtain the target similarity information; and calculating the similarity score between the reference facial features to obtain the reference similarity information.

[0134] In one embodiment, when extracting the target similarity information between the facial features and each of the reference facial features, and the reference similarity information between the reference facial features, the processor 701 may perform the following: extracting the feature difference information between the facial features and each of the reference facial features to obtain the target similarity information; and extracting the feature difference information between the reference facial features to obtain the reference similarity information.

[0135] In one embodiment, when constructing the similarity matrix representing the association between the target similarity information and the reference similarity information, the processor 701 may perform the following: obtaining the similarity between the facial feature and each of the reference facial features; sorting the facial feature and each of the reference facial features according to the similarity; and constructing an adjacency matrix between the target similarity information and the reference similarity information based on the sorting, thereby forming the similarity matrix.

[0136] In one embodiment, when performing information confidence analysis based on the similarity matrix to obtain the confidence level of the facial feature matching the target reference facial feature among at least one reference facial feature, the processor 701 may execute: performing fusion encoding processing on the similarity matrix to obtain fused features; and performing confidence analysis based on the fused features to obtain the confidence level of the facial feature matching the target reference facial feature among at least one reference facial feature.

[0137] In one embodiment, when performing confidence analysis based on the fused features to obtain the confidence level of the face feature matching the target reference face feature among at least one reference face feature, the processor 701 may execute: performing a fully connected operation on the fused features to obtain a fully connected feature; and performing classification processing based on the fully connected feature to obtain the confidence level of the face feature matching the target reference face feature among at least one reference face feature.

[0138] In one embodiment, the target reference facial feature includes a reference facial feature with the highest similarity to the facial feature; when performing classification processing based on the fully connected feature to obtain the confidence level that the facial feature matches the target reference facial feature among at least one reference facial feature, the processor 701 may execute: performing classification processing based on the fully connected feature to obtain the confidence level that the reference facial feature with the highest similarity matches the facial feature.

[0139] In one embodiment, the target reference facial feature includes at least two reference facial features with the highest similarity to the facial feature; when performing classification processing based on the fully connected features to obtain the confidence level of the facial feature matching the target reference facial feature among at least one reference facial feature, the processor 701 may execute: performing classification processing based on the fully connected features to obtain the confidence level of the at least two reference facial features with the highest similarity matching the facial feature.

[0140] In one embodiment, when performing fusion encoding processing on the similarity matrix to obtain fused features, the processor 701 may perform convolution processing on the similarity matrix to generate the fused features.

[0141] In one embodiment, when performing fusion encoding processing on the similarity matrix to obtain fused features, the processor 701 may perform self-attention encoding processing on the similarity matrix to generate the fused features.

[0142] In one embodiment, when determining a reference facial feature that matches the target reference facial feature from the target reference facial features based on the confidence level, the processor 701 may execute: determining a reference facial feature whose confidence level meets the recognition conditions from the target reference facial features; and using the reference facial feature whose confidence level meets the recognition conditions as the reference facial feature that matches the facial feature.

[0143] Those skilled in the art will understand that all or part of the steps in the various methods of the above embodiments can be performed by a computer program, or by a computer program controlling related hardware. The computer program can be stored in a computer-readable storage medium and loaded and executed by a processor.

[0144] Therefore, embodiments of this application also provide a computer-readable storage medium storing a computer program that can be loaded by a processor to perform the steps in any of the methods provided in embodiments of this application.

[0145] The computer-readable storage medium may include: read-only memory (ROM), random access memory (RAM), disk or optical disk, etc.

[0146] Since the computer program stored in the computer-readable storage medium can execute the steps in any of the methods provided in the embodiments of this application, the beneficial effects that the methods provided in the embodiments of this application can achieve can be realized. For details, please refer to the previous embodiments, which will not be repeated here.

[0147] According to one aspect of this application, a computer program product or computer program is provided, comprising computer instructions stored in a computer-readable storage medium. A processor of a computer device reads the computer instructions from the computer-readable storage medium and executes the computer instructions, causing the computer device to perform the methods provided in the various optional implementations of the above embodiments of this application.

[0148] Other embodiments of this application will readily occur to those skilled in the art upon consideration of the specification and practice of the embodiments disclosed herein. This application is intended to cover any variations, uses, or adaptations of this application that follow the general principles of this application and include common knowledge or customary techniques in the art not disclosed herein.

[0149] It should be understood that this application is not limited to the embodiments described above and shown in the accompanying drawings, but various modifications and changes can be made without departing from its scope.

Claims

1. A face recognition method characterized by, include: Obtain the facial features and reference feature set corresponding to the target face image, wherein the reference feature set includes at least one reference facial feature; The reference feature set also includes target reference facial features, which include at least one reference facial feature that has the highest similarity to the determined facial features; Extract the target similarity information between the facial features and each of the reference facial features, as well as the reference similarity information between the reference facial features; Construct a similarity matrix that represents the association between the target similarity information and the reference similarity information; Based on the similarity matrix, information confidence analysis is performed to obtain the confidence level of the match between the facial feature and the target reference facial feature in at least one reference facial feature; Based on the confidence level, a reference facial feature matching the target reference facial feature is determined from the target reference facial features; The recognition result of the target face image is determined based on the matched reference facial features.

2. The method of claim 1, wherein, The step of obtaining the facial features and reference feature set corresponding to the target facial image includes: Obtain the facial features and recognition feature library corresponding to the target face image, wherein the recognition feature library includes at least one preset facial feature; Calculate the similarity between the facial feature and each preset facial feature in the recognition feature library; Preset facial features with similarity rankings above a predetermined level are obtained from the recognition feature library as reference facial features to form the reference feature set.

3. The method of claim 1, wherein, The step of extracting the target similarity information between the facial features and each of the reference facial features, and the reference similarity information between the reference facial features, includes: Calculate the similarity score between the facial feature and each of the reference facial features to obtain the target similarity information; Calculate the similarity scores between the reference facial features to obtain the reference similarity information.

4. The method of claim 1, wherein, The step of extracting target similarity information between the facial features and each of the reference facial features, and reference similarity information between the reference facial features, includes: Extract the feature difference information between the facial features and each of the reference facial features to obtain the target similarity information; The feature difference information between the reference facial features is extracted to obtain the reference similarity information.

5. The method of claim 1, wherein, The construction of a similarity matrix representing the association between the target similarity information and the reference similarity information includes: Obtain the similarity between the facial feature and each of the reference facial features; The facial features are sorted with each of the reference facial features according to the similarity score. Based on the sorting, an adjacency matrix is ​​constructed between the target similarity information and the reference similarity information to form the similarity matrix.

6. The method of claim 1, wherein, The step of performing information confidence analysis based on the similarity matrix to obtain the confidence level of the match between the facial feature and at least one target reference facial feature includes: The similarity matrix is ​​fused and encoded to obtain fused features; Based on the fusion features, a confidence analysis is performed to obtain the confidence level of the match between the facial features and the target reference facial features in at least one reference facial feature.

7. The method according to claim 6, characterized in that, The confidence analysis based on the fused features, to obtain the confidence level of the match between the facial features and at least one target reference facial feature, includes: The fused features are processed by a fully connected operation to obtain fully connected features; Based on the fully connected features, classification processing is performed to obtain the confidence level of the facial features matching the target reference facial features among at least one reference facial features.

8. The method according to claim 7, characterized in that, The target reference facial feature includes the reference facial feature with the highest similarity to the facial feature; The classification process based on the fully connected features, to obtain the confidence score of the match between the facial features and at least one target reference facial feature, includes: Based on the fully connected features, classification processing is performed to obtain the confidence level of the reference facial feature with the highest similarity matching the facial feature.

9. The method according to claim 7, characterized in that, The target reference facial features include at least two reference facial features that have the highest similarity to the facial features; The classification process based on the fully connected features, to obtain the confidence score of the match between the facial features and at least one target reference facial feature, includes: Based on the fully connected features, classification processing is performed to obtain the confidence scores of at least two reference facial features with the highest similarity that match the facial features.

10. The method according to claim 6, characterized in that, The process of fusing and encoding the similarity matrix to obtain fused features includes: The similarity matrix is ​​convolved to generate the fused features.

11. The method according to claim 6, characterized in that, The process of fusing and encoding the similarity matrix to obtain fused features includes: The similarity matrix is ​​subjected to self-attention encoding to generate the fused features.

12. The method according to claim 1, characterized in that, The step of determining reference facial features that match the target reference facial features from the target reference facial features based on the confidence level includes: From the target reference facial features, determine the reference facial features whose confidence level meets the recognition criteria; Reference facial features that meet the recognition criteria with a confidence level are used as reference facial features that match the stated facial features.

13. A facial recognition device, characterized in that, include: The acquisition module is used to acquire the facial features and reference feature set corresponding to the target face image, wherein the reference feature set includes at least one reference facial feature; The reference feature set also includes target reference facial features, which include at least one reference facial feature that has the highest similarity to the determined facial features; An extraction module is used to extract target similarity information between the facial features and each of the reference facial features, as well as reference similarity information between the reference facial features; A construction module is used to construct a similarity matrix that represents the association between the target similarity information and the reference similarity information; The analysis module is used to perform information confidence analysis based on the similarity matrix to obtain the confidence level of the facial feature matching the target reference facial feature among at least one reference facial feature; A matching module is used to determine, based on the confidence level, a reference facial feature that matches the facial feature from the target reference facial features; The recognition module is used to determine the recognition result of the target face image based on the matched reference facial features.

14. The apparatus according to claim 13, characterized in that, The acquisition module includes: The feature library acquisition unit is used to acquire facial features and a recognition feature library corresponding to the target face image, wherein the recognition feature library includes at least one preset facial feature; A similarity calculation unit is used to calculate the similarity between the facial features and each preset facial feature in the recognition feature library; The feature set acquisition unit is used to acquire preset facial features with similarity rankings above a predetermined number from the recognition feature library as reference facial features to form the reference feature set.

15. The apparatus according to claim 13, characterized in that, The extraction module includes: The first score calculation unit is used to calculate the similarity score between the facial feature and each of the reference facial features to obtain the target similarity information; The second score calculation unit is used to calculate the similarity score between the reference facial features to obtain the reference similarity information.

16. The apparatus according to claim 13, characterized in that, The extraction module includes: The first feature extraction unit is used to extract feature difference information between the facial features and each of the reference facial features to obtain the target similarity information; The second feature extraction unit is used to extract feature difference information between the reference facial features to obtain the reference similarity information.

17. The apparatus according to claim 13, characterized in that, The building module includes: A similarity acquisition unit is used to acquire the similarity between the facial feature and each of the reference facial features; A sorting unit is used to sort the facial features and each of the reference facial features according to the similarity. The adjacency matrix construction unit is used to construct an adjacency matrix between the target similarity information and the reference similarity information according to the sorting, thereby forming the similarity matrix.

18. The apparatus according to claim 13, characterized in that, The analysis module includes: A fusion encoding unit is used to perform fusion encoding processing on the similarity matrix to obtain fusion features; The confidence analysis unit is used to perform confidence analysis based on the fused features to obtain the confidence level of the match between the facial features and the target reference facial features in at least one reference facial feature.

19. The apparatus according to claim 18, characterized in that, The confidence analysis unit includes: A fully connected processing subunit is used to perform fully connected operations on the fused features to obtain fully connected features; The classification processing subunit is used to perform classification processing based on the fully connected features to obtain the confidence level of the facial features matching the target reference facial features among at least one reference facial features.

20. The apparatus according to claim 19, characterized in that, The target reference facial feature includes the reference facial feature with the highest similarity to the facial feature; The classification processing subunit is used to: perform classification processing based on the fully connected features to obtain the confidence level of the highest similarity reference facial feature matching the facial feature.

21. The apparatus according to claim 19, characterized in that, The target reference facial features include at least two reference facial features that have the highest similarity to the facial features; The classification processing subunit is used to: perform classification processing based on the fully connected features to obtain the confidence scores of at least two reference facial features with the highest similarity matching the facial features.

22. The apparatus according to claim 18, characterized in that, The fusion encoding unit is used to: perform convolution processing on the similarity matrix to generate the fusion feature.

23. The apparatus according to claim 18, characterized in that, The fusion encoding unit is used to: perform self-attention encoding on the similarity matrix to generate the fusion feature.

24. The apparatus according to claim 13, characterized in that, The matching module includes: A matching feature determination unit is used to determine reference facial features with confidence levels that meet the recognition conditions from the target reference facial features; The matching feature determination unit is used to take reference facial features with confidence levels that meet the recognition conditions as reference facial features that match the facial features.

25. A computer-readable storage medium, characterized in that, It stores a computer program that, when executed by the computer's processor, causes the computer to perform the method described in any one of claims 1 to 12.

26. An electronic device, characterized in that, include: Memory, which stores computer programs; A processor reads a computer program stored in memory to execute the method according to any one of claims 1 to 12.

27. A computer program product, characterized in that, The computer program product includes computer instructions stored in a computer-readable storage medium; the processor of the computer device reads the computer instructions from the computer-readable storage medium and executes the computer instructions, causing the computer device to perform the method described in any one of claims 1 to 12.

Images