WO2021068613A1 - 面部识别方法、装置、设备及计算机可读存储介质 - Google Patents

面部识别方法、装置、设备及计算机可读存储介质 Download PDF

Info

Publication number
WO2021068613A1
WO2021068613A1 PCT/CN2020/106023 CN2020106023W WO2021068613A1 WO 2021068613 A1 WO2021068613 A1 WO 2021068613A1 CN 2020106023 W CN2020106023 W CN 2020106023W WO 2021068613 A1 WO2021068613 A1 WO 2021068613A1
Authority
WO
WIPO (PCT)
Prior art keywords
facial
frame
image
feature
facial feature
Prior art date
Application number
PCT/CN2020/106023
Other languages
English (en)
French (fr)
Inventor
卢宁
徐国强
Original Assignee
深圳壹账通智能科技有限公司
Priority date (The priority date is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the date listed.)
Filing date
Publication date
Application filed by 深圳壹账通智能科技有限公司 filed Critical 深圳壹账通智能科技有限公司
Publication of WO2021068613A1 publication Critical patent/WO2021068613A1/zh

Links

Classifications

    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
    • G06V40/00—Recognition of biometric, human-related or animal-related patterns in image or video data
    • G06V40/10—Human or animal bodies, e.g. vehicle occupants or pedestrians; Body parts, e.g. hands
    • G06V40/16—Human faces, e.g. facial parts, sketches or expressions
    • G06V40/168—Feature extraction; Face representation
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F18/00—Pattern recognition
    • G06F18/20—Analysing
    • G06F18/25—Fusion techniques
    • G06F18/253—Fusion techniques of extracted features
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F21/00—Security arrangements for protecting computers, components thereof, programs or data against unauthorised activity
    • G06F21/30—Authentication, i.e. establishing the identity or authorisation of security principals
    • G06F21/31—User authentication
    • G06F21/32—User authentication using biometric data, e.g. fingerprints, iris scans or voiceprints
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06Q—INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES; SYSTEMS OR METHODS SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES, NOT OTHERWISE PROVIDED FOR
    • G06Q40/00—Finance; Insurance; Tax strategies; Processing of corporate or income taxes
    • G06Q40/03—Credit; Loans; Processing thereof

Definitions

  • This application relates to the technical field of biometrics, and in particular to a facial recognition method, device, device, and computer-readable storage medium.
  • the current artificial intelligence technology mainly recognizes the customer's facial movements and analyzes the customer's fraud risk based on the facial movements.
  • the traditional facial movement recognition algorithm mainly focuses on image processing, which means that the video is decomposed into single-frame images, and then Facial action recognition for each frame of image.
  • the main purpose of this application is to provide a facial recognition method, device, equipment, and computer-readable storage medium, aiming to improve the accuracy of facial recognition.
  • the present application provides a facial recognition method.
  • the facial recognition method includes the following steps:
  • the facial recognition model includes a facial feature extraction layer, a facial feature relationship processing layer, a feature fusion layer, and a facial recognition layer;
  • the facial features corresponding to each frame of image are fused to obtain the target facial feature
  • a facial recognition result of the user in the video data is determined according to the target facial feature.
  • the present application also provides a facial recognition device, which includes:
  • An acquisition module for acquiring video data to be recognized and a facial recognition model where the facial recognition model includes a facial feature extraction layer, a facial feature relationship processing layer, a feature fusion layer, and a facial recognition layer;
  • the feature extraction module is configured to perform feature extraction on each frame of image in the video data based on the facial feature extraction layer to obtain the facial feature corresponding to each frame of image;
  • the feature relationship determination module is configured to determine the facial feature relationship between the facial features corresponding to each frame of image based on the facial feature relationship processing layer and according to the video time point corresponding to each frame of image;
  • the feature fusion module is configured to fuse the facial features corresponding to each frame of image based on the feature fusion layer and according to the facial feature relationship corresponding to each frame of image to obtain the target facial feature;
  • the facial recognition module is configured to determine the facial recognition result of the user in the video data according to the target facial feature based on the facial recognition layer.
  • the present application also provides a computer device that includes a processor, a memory, and a computer program stored on the memory and executable by the processor, wherein the computer program is When the processor executes, the following steps are implemented:
  • the facial recognition model includes a facial feature extraction layer, a facial feature relationship processing layer, a feature fusion layer, and a facial recognition layer;
  • the facial features corresponding to each frame of image are fused to obtain the target facial feature
  • a facial recognition result of the user in the video data is determined according to the target facial feature.
  • the present application also provides a computer-readable storage medium having a computer program stored on the computer-readable storage medium, and when the computer program is executed by a processor, the following steps are implemented:
  • the facial recognition model includes a facial feature extraction layer, a facial feature relationship processing layer, a feature fusion layer, and a facial recognition layer;
  • the facial features corresponding to each frame of image are fused to obtain the target facial feature
  • a facial recognition result of the user in the video data is determined according to the target facial feature.
  • This application provides a facial recognition method, device, equipment, and computer-readable storage medium.
  • This application extracts the facial features corresponding to each frame of the video to determine the facial features corresponding to each frame of the image and the corresponding facial feature of each frame of the image. Facial feature relationship between facial features. Based on the facial feature relationship, facial features can be fused to obtain accurate facial features of the user. Based on the facial features, facial recognition of the user can effectively improve the accuracy of facial recognition. Effectively reduce the risk of fraud.
  • FIG. 1 is a schematic flowchart of a facial recognition method provided by an embodiment of this application
  • FIG. 2 is a schematic diagram of a scene in which the facial recognition method provided by this embodiment is implemented
  • FIG. 3 is a schematic flowchart of another face recognition method provided by an embodiment of the application.
  • FIG. 4 is a schematic block diagram of a facial recognition device provided by an embodiment of the application.
  • FIG. 5 is a schematic block diagram of another facial recognition device provided by an embodiment of the application.
  • FIG. 6 is a schematic block diagram of the structure of a computer device related to an embodiment of this application.
  • the embodiments of the present application provide a facial recognition method, device, equipment, and computer-readable storage medium.
  • the facial recognition method can be applied to a server, and the server can be a single server or a server cluster composed of multiple servers.
  • FIG. 1 is a schematic flowchart of a facial recognition method provided by an embodiment of the application.
  • the face recognition method includes steps S101 to S105.
  • Step S101 Obtain video data to be recognized and a facial recognition model, where the facial recognition model includes a facial feature extraction layer, a facial feature relationship processing layer, a feature fusion layer, and a facial recognition layer.
  • the server may obtain the video data to be recognized in real time or at predetermined intervals, and the video data to be recognized includes the facial information of the user.
  • the user can record video data containing the user's facial information through the terminal device, and upload the video data to the server, the server stores the video data, and the server writes the video data that needs to be reviewed into the video data review In the queue, it is convenient for the server to subsequently obtain a piece of video data from the video data review queue periodically or in real time as the video data to be identified for review, and determine whether the user applying for a loan or a credit card is suspected of fraud.
  • the server can also perform facial recognition on the video data uploaded by the terminal device in real time.
  • the terminal device receives a credit card application instruction triggered by the user , Based on the credit card application instruction to display the information entry page.
  • the terminal receives the customer information input based on the information input page and detects the customer information confirmation instruction, the terminal displays the video recording page. After the video is recorded, the terminal receives the recorded credit card application video, and when it monitors the credit card confirmation application instruction, generates a credit card application request carrying the customer information and the credit card application video, and sends the credit card application request to the server.
  • the server When the server receives the credit card application request sent by the terminal device, it obtains the credit card application video from the credit card application request, uses the credit card application video as the target video that needs to recognize facial actions, and obtains the facial recognition model stored locally.
  • the user information includes but is not limited to user name, ID card information, mobile phone number, education information, permanent residence address, company name, company address, social security information, and provident fund information.
  • the server or server cluster stores a facial recognition model.
  • the facial recognition model includes a facial feature extraction layer, a facial feature relationship processing layer, a feature fusion layer, and a facial recognition layer.
  • the facial feature extraction layer is used to extract facial features from the target video.
  • the facial feature relationship processing layer is used to determine the facial feature relationship between the facial features of each frame of image, and the feature fusion layer is used to fuse the facial features based on the facial feature relationship between the facial features of each frame of image.
  • the facial recognition layer is used to recognize the user's facial actions from the video and obtain the facial recognition results.
  • the terminal device can be an electronic device such as a mobile phone, a tablet computer, a notebook computer, a desktop computer, a personal digital assistant, and a wearable device.
  • the facial recognition model is obtained based on training.
  • the specific process of training is to collect a large amount of video data with facial actions, and use the video data with facial actions as model sample data, and then use the model sample data
  • the facial action recognition layer in the facial recognition model is iteratively trained until the model converges.
  • the facial recognition model is designed based on a deep neural network, and the model parameters to be trained of the facial recognition model are model parameters of the facial action recognition layer.
  • Step S102 Perform feature extraction on each frame of image in the video data based on the facial feature extraction layer to obtain the facial feature corresponding to each frame of image.
  • the server performs feature extraction on each frame of image in the video data to be recognized to obtain the corresponding facial feature of each frame of image, that is, from each frame of image in the video data to be recognized through the facial feature extraction layer in the facial recognition model Extract the facial features corresponding to each frame of image.
  • a frame of image as an example, if 128 facial key points are extracted from a frame of image, the extracted 128 facial key points are taken as the facial features corresponding to this frame of image.
  • the above-mentioned preset number can be set based on actual conditions, which is not specifically limited in this application.
  • the facial feature is a feature matrix composed of 128 or 512 facial key points.
  • facial feature extraction algorithms include, but are not limited to, facial feature extraction algorithms based on Gabor filtering, facial feature extraction algorithms based on local binarization, facial feature extraction algorithms based on deep neural networks, and facial feature extraction algorithms based on geometric features.
  • the application does not make specific restrictions on this.
  • Step 103 Based on the facial feature relationship processing layer, determine the facial feature relationship between the facial features corresponding to each frame of image according to the video time point corresponding to each frame of image.
  • the facial feature relationship processing layer in the facial recognition model determines the facial feature relationship between the facial features corresponding to each frame of image according to the corresponding video time point of each frame of image .
  • the video time point is the time point of each frame of image in the video data to be recognized, and is used to indicate the time sequence of each frame of image.
  • the facial features corresponding to each frame of image are sorted to obtain a facial feature queue; according to the facial feature queue, the facial feature relationship between the facial features corresponding to each frame of image is determined . It should be noted that the smaller the video moment, the higher the ranking of facial features, and the larger the video moment, the lower the ranking of facial features.
  • the method of determining the facial feature relationship based on the facial feature queue is specifically: sequentially selecting a facial feature from the facial feature queue as the target facial feature, and obtaining the facial features adjacent to the target facial feature from the facial feature queue and Facial features that are not adjacent to the target facial feature, thereby obtaining the facial feature relationship between the target facial feature and each of the remaining facial features, that is, the adjacent relationship and the non-adjacent relationship.
  • the facial feature queue is [A, B, C, D]
  • the facial feature relationship between facial feature A and facial feature B is adjacent
  • the facial feature relationship between facial feature A and facial feature C It is a non-adjacent relationship
  • the facial feature relationship between facial feature A and facial feature D is non-adjacent relationship
  • the facial feature relationship between facial feature B and facial feature C is adjacent relationship
  • facial feature B and facial feature D The facial feature relationship between is non-adjacent relationship
  • the facial feature relationship between facial feature C and facial feature D is adjacent relationship.
  • Step 104 Based on the feature fusion layer, according to the facial feature relationship corresponding to each frame of image, the facial features corresponding to each frame of image are fused to obtain the target facial feature.
  • the facial features corresponding to each frame of image are fused to obtain the target facial feature.
  • the facial features corresponding to each frame of image are merged to obtain accurate facial features, which can improve the accuracy of facial recognition.
  • Step 105 Based on the facial recognition layer, determine the user's facial recognition result in the video data according to the target facial feature.
  • the facial recognition layer in the facial recognition model determines the user's facial recognition result in the video data based on the target facial features. In one embodiment, it is determined whether the preset facial feature set includes the target facial feature. If the preset facial feature set includes the target facial feature, the facial recognition result is failed, and the user's fraud suspicion is high. Assuming that the facial feature set does not contain the target facial feature, the facial recognition result is passed, and the user's fraud suspicion is low or there is no fraud suspicion. It should be noted that the aforementioned facial feature set can be set based on actual conditions, for example, corresponding facial features such as lips licking, cheeks rising, mouth drooping, and pupil dilation can be set, which is not specifically limited in this application.
  • the user's facial motion in the video data is determined according to the target facial features, and the user's facial recognition result is determined according to the facial motion.
  • the method for determining the facial action is specifically: acquiring a pre-stored mapping relationship table between facial features and facial actions, querying the mapping relationship table, and using the facial action corresponding to the target facial feature as the user's facial action in the video data. It should be noted that the mapping relationship table between facial features and facial actions can be set based on actual conditions, which is not specifically limited in this application.
  • the specific method for determining the user's facial recognition result based on facial motion is: determining whether the facial motion is located in the preset facial motion set, if the facial motion is located in the preset facial motion set, the facial recognition result is not passed, and the user’s facial recognition result is not passed.
  • the fraud suspicion is high. If the facial action is not in the preset facial action set, the facial recognition result is passed, and the user's fraud suspicion is low or there is no suspicion of fraud.
  • the facial action in the preset facial action set is that the facial recognition result is through the corresponding facial action, wherein the behavior result represented by the facial action in the preset facial action set can be set based on actual conditions, for example, Set facial actions such as licking lips, raising cheeks, drooping corners of the mouth, and dilated pupils, which are not specifically limited in this application.
  • the method of determining the user's facial recognition result based on facial motions may also be: after determining the user's facial motions, input the determined facial motions to the determination model of the facial recognition results, which is determined by the facial recognition results.
  • the determination model outputs the facial recognition result corresponding to the facial action, and the facial recognition result includes the facial recognition passed and the facial recognition failed.
  • the deterministic model of facial recognition is obtained through training, and the specific training process is: acquiring a large number of model sample data labeled with facial recognition results and facial actions, and based on neural network, convolutional neural network and/or cyclic convolutional neural network Design a facial recognition model, and then train the designed deterministic model of facial recognition results through the model sample data until the deterministic model of facial recognition results converges.
  • risk reminder information is sent to the terminal, and an information security verification operation is performed.
  • the specific information security verification operation is: when the terminal receives the risk reminder message, it displays the information security verification page, where the information security verification page displays the risk reminder message and the trigger control of the SMS verification code; when the user detects the SMS verification code
  • the SMS verification code acquisition request is sent to the server; the receiving server obtains the SMS verification code returned by the SMS verification code acquisition request based on the SMS verification code, and obtains the SMS verification code entered by the user on the information security verification page. If the user enters If the SMS verification code is consistent with the received SMS verification code, the information security verification is passed. On the contrary, if the SMS verification code entered by the user is inconsistent with the received SMS verification code, the information security verification is not passed.
  • FIG. 2 is a schematic diagram of a scene in which the facial recognition method provided by this embodiment is implemented.
  • the terminal device is used to receive the user's credit card application instruction, enter user information and record the credit card application video
  • the server is used to receive the request generated by the receiving terminal carrying the user information and the credit card application video, and perform facial recognition on the credit card application video.
  • the facial recognition method provided by the foregoing embodiment determines the facial feature relationship between the facial features corresponding to each frame of images and the facial features corresponding to each frame of images by extracting the facial features corresponding to each frame of the image in the video, based on the facial features Feature relationship fusion of facial features can obtain the user's accurate facial features, and then based on the facial features, facial recognition of the user can effectively improve the accuracy of facial recognition and effectively reduce the risk of fraud.
  • FIG. 3 is a schematic flowchart of a facial recognition method provided by an embodiment of the application.
  • the face recognition method includes steps S201 to S207.
  • Step S201 Obtain video data to be recognized and a facial recognition model, where the facial recognition model includes a facial feature extraction layer, a facial feature relationship processing layer, a feature fusion layer, and a facial recognition layer.
  • the server may obtain the video data to be recognized in real time or at predetermined intervals, and the video data to be recognized includes the facial information of the user.
  • the user can record video data containing the user's facial information through the terminal device, and upload the video data to the server, the server stores the video data, and the server writes the video data that needs to be reviewed into the video data review In the queue, it is convenient for the server to subsequently obtain a piece of video data from the video data review queue periodically or in real time as the video data to be identified for review, and determine whether the user applying for a loan or a credit card is suspected of fraud.
  • the server or server cluster stores a facial recognition model.
  • the facial recognition model includes a facial feature extraction layer, a facial feature relationship processing layer, and a facial action recognition layer.
  • the facial feature extraction layer is used to extract facial features from the target video.
  • the feature relationship processing layer is used to determine the facial feature relationship between the facial features of each frame of image, and the facial action recognition layer is used to recognize the user's facial actions from the video.
  • the terminal device can be an electronic device such as a mobile phone, a tablet computer, a notebook computer, a desktop computer, a personal digital assistant, and a wearable device.
  • Step S202 Perform feature extraction on each frame of image in the video data based on the facial feature extraction layer to obtain a respective facial feature corresponding to each frame of image.
  • the server performs feature extraction on each frame of image in the video data to be recognized to obtain the corresponding facial feature of each frame of image, that is, from each frame of image in the video data to be recognized through the facial feature extraction layer in the facial recognition model Extract the facial features corresponding to each frame of image.
  • a frame of image as an example, if 128 facial key points are extracted from a frame of image, the extracted 128 facial key points are taken as the facial features corresponding to this frame of image.
  • Step S203 Based on the facial feature relationship processing layer, determine the facial feature relationship between the facial features corresponding to each frame of image according to the respective video time point of each frame of image.
  • the facial feature relationship processing layer in the facial recognition model determines the facial feature relationship between the facial features corresponding to each frame of image according to the corresponding video time point of each frame of image .
  • the video time point is the time point of each frame of image in the video data to be recognized, and is used to indicate the time sequence of each frame of image.
  • Step S204 Based on the feature fusion layer, according to the facial feature relationship corresponding to each frame of image, sort the facial features corresponding to each frame of image to obtain a target facial feature queue.
  • the facial features corresponding to each frame of image are facial feature 1, facial feature 2, facial feature 3, and facial feature 4, and the facial feature relationship is that facial feature 1 is adjacent to facial feature 3, and is adjacent to facial feature 1 and facial feature.
  • Facial feature 2 is not adjacent; Facial feature 3 is adjacent to Facial Feature 1 and Facial Feature 4, but not adjacent to Facial Feature 2; Facial feature 2 is adjacent to Facial Feature 4, but not adjacent to Facial Feature 1 and Facial Feature 3; Facial feature 4 is adjacent to facial feature 3 and facial feature 2, but not adjacent to facial feature 1.
  • the target facial feature queue obtained is [1, 3, 4, 2].
  • Step S205 Perform a pairwise comparison of the facial features in the target facial feature queue to obtain a comparison result of every two facial features in the target facial feature queue.
  • the facial features in the target facial feature queue are compared in pairs to obtain the comparison result of every two facial features in the target facial feature queue.
  • the comparison result includes the same facial features and different facial features. For example, compare facial feature 1 with facial feature 2, facial feature 3, and facial feature 4 in the target facial feature queue [1, 3, 4, 2], and compare facial feature 2 with facial feature 3 and facial feature respectively
  • the feature 4 is compared, and the facial feature 3 is compared with the facial feature 4.
  • the comparison it is found that only the facial feature 1 is the same as the facial feature 3, and the rest are different.
  • Step S206 According to the comparison result of every two of the facial features, the facial features corresponding to each frame of image are merged to obtain the target facial feature.
  • the corresponding facial features of each frame of image are merged to obtain the target facial features.
  • the facial features in the target facial feature queue are deleted, where each facial feature in the target facial feature queue after the deletion processing is different; it will be deleted.
  • Each facial feature in the processed target facial feature queue is fused to obtain the target facial feature.
  • the target facial feature queue is facial feature 1, facial feature 2, facial feature 3, and facial feature 4 in [1, 3, 4, 2].
  • the comparison result between facial feature 1 and facial feature 3 is that facial feature 1 is the same as facial feature 3, and the rest are all If they are not the same, the target facial feature queue after deletion processing is [1, 4, 2].
  • the specific method of facial feature fusion is: select a facial feature from the target facial feature queue after the deletion process as the first facial feature to be fused, and then select from the target facial feature queue after the deletion process One of the remaining facial features is used as the second facial feature to be fused; the facial key points in the first facial feature are compared with the facial key points in the second facial feature to obtain different facial key points, and the second face Different facial key points in the features are merged into the first facial feature to update the first facial feature; then one of the remaining facial features is selected from the target facial feature queue after the deletion process, and the same method is used with the updated facial feature.
  • the first facial feature is fused, and after several repetitions, the last facial feature in the target facial feature queue after the deletion processing is fused with the first facial feature obtained by each fusion update to obtain the target facial feature.
  • the method of fusing different facial key points in the second facial feature to the first facial feature is specifically: obtaining different facial key points from the second facial feature to form a first feature matrix, and from the first face
  • the partial features acquire the facial key points corresponding to the different facial key points to form a second feature matrix; add the first feature matrix to the second feature matrix to obtain the fusion feature matrix, and then combine the face keys in the fusion feature matrix
  • the set of points and the remaining facial key points in the first facial feature is used as the updated first facial feature.
  • Step S207 Based on the facial recognition layer, determine the user's facial recognition result in the video data according to the target facial feature.
  • the facial recognition layer in the facial recognition model determines the user's facial recognition result in the video data based on the target facial features. In one embodiment, it is determined whether the preset facial feature set includes the target facial feature. If the preset facial feature set includes the target facial feature, the facial recognition result is failed, and the user's fraud suspicion is high. Assuming that the facial feature set does not contain the target facial feature, the facial recognition result is passed, and the user's fraud suspicion is low or there is no fraud suspicion. It should be noted that the aforementioned facial feature set can be set based on actual conditions, for example, corresponding facial features such as lips licking, cheeks rising, mouth drooping, and pupil dilation can be set, which is not specifically limited in this application.
  • the facial recognition method provided by the foregoing embodiment determines the facial feature relationship between the facial feature corresponding to each frame of image and the facial feature corresponding to each frame of image by extracting the facial features corresponding to each frame of image in the video.
  • the feature relationship can sort the corresponding facial features of each frame of image to obtain the target facial feature queue, and then according to the comparison result of every two facial features in the target facial feature queue, the corresponding facial features of each frame of image are merged, and then Based on the fused facial features, the user can be accurately facially recognized, which can effectively improve the accuracy of facial recognition.
  • FIG. 4 is a schematic block diagram of a facial recognition device provided by an embodiment of the application.
  • the facial recognition device 300 includes: an acquisition module 301, a feature extraction module 302, a feature relationship determination module 303, a feature fusion module 304, and a facial recognition module 305.
  • the acquiring module 301 is configured to acquire the video data to be recognized and the facial recognition model, where the facial recognition model includes a facial feature extraction layer, a facial feature relationship processing layer, a feature fusion layer, and a facial recognition layer;
  • the feature extraction module 302 is configured to perform feature extraction on each frame of image in the video data based on the facial feature extraction layer to obtain the facial feature corresponding to each frame of image;
  • the feature relationship determination module 303 is configured to determine, based on the facial feature relationship processing layer, the facial feature relationship between the facial features corresponding to each frame of image according to the respective video time point of each frame of image;
  • the feature fusion module 304 is configured to fuse the facial features corresponding to each frame of images according to the facial feature relationship corresponding to each frame of the image based on the feature fusion layer to obtain the target facial feature;
  • the facial recognition module 305 is configured to determine the facial recognition result of the user in the video data according to the target facial feature based on the facial recognition layer.
  • the feature extraction module 302 is further configured to split the to-be-recognized video data into each frame of image based on the facial feature extraction layer, and extract a preset number from each frame of image. Facial key points; according to extracting a preset number of facial key points from each frame of image, determine the corresponding facial features of each frame of image.
  • the feature relationship determination module 303 is further configured to sort the facial features corresponding to each frame of images according to the respective video moments of each frame of the image based on the facial feature relationship processing layer to obtain Facial feature queue; according to the facial feature queue, the facial feature relationship between the facial features corresponding to each frame of image is determined.
  • the facial recognition module 305 is further configured to determine, based on the facial recognition layer, the facial motion of the user in the video data according to the target facial feature, and determine the facial motion according to the facial motion. The user's facial recognition result.
  • the facial recognition device 300 further includes:
  • a sending module configured to send risk reminder information to the terminal when the facial recognition result is a preset facial recognition result
  • the security verification module is used to perform information security verification operations.
  • FIG. 5 is a schematic block diagram of another facial recognition device provided by an embodiment of the application.
  • the facial recognition device 400 includes: an acquisition module 401, a feature extraction module 402, a feature relationship determination module 403, a ranking module 404, a feature comparison module 405, a feature fusion module 406, and a facial recognition module 407.
  • the acquiring module 401 is configured to acquire video data to be recognized and a facial recognition model, where the facial recognition model includes a facial feature extraction layer, a facial feature relationship processing layer, a feature fusion layer, and a facial recognition layer.
  • the feature extraction module 402 is configured to perform feature extraction on each frame of image in the video data based on the facial feature extraction layer to obtain a facial feature corresponding to each frame of image.
  • the feature relationship determination module 403 is configured to determine the facial feature relationship between the facial features corresponding to each frame of image based on the facial feature relationship processing layer and according to the video time point corresponding to each frame of image.
  • the sorting module 404 is configured to sort the facial features corresponding to each frame of images according to the facial feature relationship corresponding to each frame of the image based on the feature fusion layer to obtain a target facial feature queue.
  • the feature comparison module 405 is configured to compare the facial features in the target facial feature queue pair by pair to obtain a comparison result of every two facial features in the target facial feature queue.
  • the feature fusion module 406 is configured to fuse the facial features corresponding to each frame of image according to the comparison result of every two of the facial features to obtain the target facial feature.
  • the facial recognition module 407 is configured to determine the facial recognition result of the user in the video data according to the target facial feature based on the facial recognition layer.
  • the feature fusion module 406 is further configured to delete facial features in the target facial feature queue according to the comparison result of every two facial features, where the deleted facial features Each of the facial features in the target facial feature queue is different; each of the facial features in the target facial feature queue after the deletion processing is fused to obtain the target facial feature.
  • the apparatus provided in the foregoing embodiment may be implemented in the form of a computer program, and the computer program may run on the computer device as shown in FIG. 6.
  • FIG. 6 is a schematic block diagram of the structure of a computer device provided by an embodiment of the application.
  • the computer device may be a server.
  • the computer device includes a processor, a memory, and a network interface connected through a system bus, where the memory may include a non-volatile or volatile storage medium and an internal memory.
  • Non-volatile or volatile storage media can store operating systems and computer programs.
  • the computer program includes program instructions, and when the program instructions are executed, the processor can execute any facial recognition method.
  • the processor is used to provide computing and control capabilities and support the operation of the entire computer equipment.
  • the internal memory provides an environment for the operation of a computer program in a non-volatile or volatile storage medium.
  • the processor can execute any facial recognition method.
  • the network interface is used for network communication, such as sending assigned tasks.
  • the network interface is used for network communication, such as sending assigned tasks.
  • FIG. 6 is only a block diagram of part of the structure related to the solution of the present application, and does not constitute a limitation on the computer device to which the solution of the present application is applied.
  • the specific computer device may Including more or fewer parts than shown in the figure, or combining some parts, or having a different arrangement of parts.
  • the processor may be a central processing unit (Central Processing Unit, CPU), and the processor may also be other general-purpose processors, digital signal processors (Digital Signal Processors, DSP), and application specific integrated circuits (Application Specific Integrated Circuits). Circuit, ASIC), Field Programmable Gate Array (Field-Programmable Gate Array, FPGA) or other programmable logic devices, discrete gates or transistor logic devices, discrete hardware components, etc.
  • the general-purpose processor may be a microprocessor or the processor may also be any conventional processor.
  • the processor is used to run a computer program stored in a memory to implement the following steps:
  • the facial recognition model includes a facial feature extraction layer, a facial feature relationship processing layer, a feature fusion layer, and a facial recognition layer;
  • the facial features corresponding to each frame of image are fused to obtain the target facial feature
  • a facial recognition result of the user in the video data is determined according to the target facial feature.
  • the processor when the processor implements feature extraction of each frame of image in the video data based on the facial feature extraction layer to obtain the facial feature corresponding to each frame of image, it is used to implement:
  • the corresponding facial feature of each frame of image is determined.
  • the processor when the processor implements the facial feature relationship processing layer based on the facial feature relationship processing layer, when determining the facial feature relationship between the facial features corresponding to each frame of image according to the video time point corresponding to each frame of image, use To achieve:
  • the facial feature relationship between the facial features corresponding to each frame of image is determined.
  • the processor implements the fusion of the facial features corresponding to each frame of image based on the feature fusion layer and the facial feature relationship corresponding to each frame of image to obtain the target facial feature.
  • the facial features corresponding to each frame of image are merged to obtain the target facial features.
  • the processor implements the fusion of the facial features corresponding to each frame of image according to the comparison result of every two of the facial features to obtain the target facial feature, which is used to achieve:
  • the facial features in the target facial feature queue are deleted, wherein each of the facial features in the target facial feature queue after the deletion processing is not the same;
  • Fusion of each of the facial features in the target facial feature queue after the deletion processing is performed to obtain the target facial feature.
  • the processor when used to determine the user's facial recognition result in the video data based on the facial recognition layer and according to the target facial feature, it is configured to achieve:
  • the facial motion of the user in the video data is determined according to the target facial feature, and the facial recognition result of the user is determined according to the facial motion.
  • the processor is further configured to implement the following after determining the user's facial recognition result in the video data based on the facial recognition layer and according to the target facial features:
  • risk reminder information is sent to the terminal, and an information security verification operation is performed.
  • the embodiments of the present application also provide a computer-readable storage medium, the computer-readable storage medium may be volatile or non-volatile, a computer program is stored on the computer-readable storage medium, and the computer
  • the program includes program instructions, and the method implemented when the program instructions are executed can refer to the various embodiments of the facial recognition method of the present application.
  • the computer-readable storage medium may be the internal storage unit of the computer device described in the foregoing embodiment, for example, the hard disk or memory of the computer device.
  • the computer-readable storage medium may also be an external storage device of the computer device, such as a plug-in hard disk, a smart memory card (Smart Media Card, SMC), and a secure digital (Secure Digital) equipped on the computer device. Digital, SD) card, flash memory card (Flash Card) and so on.

Landscapes

  • Engineering & Computer Science (AREA)
  • Theoretical Computer Science (AREA)
  • Physics & Mathematics (AREA)
  • General Physics & Mathematics (AREA)
  • Business, Economics & Management (AREA)
  • Finance (AREA)
  • Computer Security & Cryptography (AREA)
  • Oral & Maxillofacial Surgery (AREA)
  • Health & Medical Sciences (AREA)
  • General Engineering & Computer Science (AREA)
  • Data Mining & Analysis (AREA)
  • Computer Vision & Pattern Recognition (AREA)
  • Accounting & Taxation (AREA)
  • Strategic Management (AREA)
  • Marketing (AREA)
  • Artificial Intelligence (AREA)
  • Bioinformatics & Cheminformatics (AREA)
  • Bioinformatics & Computational Biology (AREA)
  • General Business, Economics & Management (AREA)
  • Evolutionary Biology (AREA)
  • Evolutionary Computation (AREA)
  • Technology Law (AREA)
  • Life Sciences & Earth Sciences (AREA)
  • Economics (AREA)
  • General Health & Medical Sciences (AREA)
  • Human Computer Interaction (AREA)
  • Multimedia (AREA)
  • Development Economics (AREA)
  • Computer Hardware Design (AREA)
  • Software Systems (AREA)
  • Image Analysis (AREA)

Abstract

一种面部识别方法、装置、设备及计算机可读存储介质，包括：获取待识别的视频数据和面部识别模型，面部识别模型包括面部特征提取层、面部特征关系处理层、特征融合层和面部识别层；基于面部特征提取层对视频数据中的每帧图像进行特征提取，得到每帧图像各自对应的面部特征；基于面部特征关系处理层，根据每帧图像各自对应的视频时刻点，确定每帧图像各自对应的面部特征之间的面部特征关系；基于特征融合层，根据每帧图像各自对应的面部特征关系，对每帧图像各自对应的面部特征进行融合，得到目标面部特征；基于面部识别层，根据目标面部特征确定视频数据中用户的面部识别结果。

Description

面部识别方法、装置、设备及计算机可读存储介质
本申请要求于2019年10月12日提交中国专利局、申请号为CN201910969394.4、名称为“面部识别方法、装置、设备及计算机可读存储介质”的中国专利申请的优先权，其全部内容通过引用结合在本申请中。
技术领域
本申请涉及生物识别的技术领域，尤其涉及一种面部识别方法、装置、设备及计算机可读存储介质。
 背景技术
在互联网金融领域，许多金融业务可以通过网络端的人工智能技术远程处理业务，这时往往需要利用人工智能技术来识别证客户真实信息，以降低欺诈风险。目前的人工智能技术主要是通过识别客户的面部动作，并基于面部动作来分析客户的欺诈风险，而传统的面部动作识别算法主要以图像处理为主，即将视频分解为单帧的图像，再对每帧图像的面部动作识别。
然而人的面部动作通常是一个连续性的过程，发明人意识到仅通过对每帧图像进行面部动作识别，而不进行其余处理，只能得到一帧图像中的面部动作，根据得到的一帧图像中的面部动作来单独识别，无法保证识别的面部动作的准确性，进而无法准确的确定面部识别结果。因此，如何提高面部识别的准确性是目前亟待解决的问题。
 技术解决方案
本申请的主要目的在于提供一种面部识别方法、装置、设备及计算机可读存储介质，旨在提高面部识别的准确性。
第一方面，本申请提供一种面部识别方法，所述面部识别方法包括以下步骤：
获取待识别的视频数据和面部识别模型，其中，所述面部识别模型包括面部特征提取层、面部特征关系处理层、特征融合层和面部识别层；
基于所述面部特征提取层对所述视频数据中的每帧图像进行特征提取，得到每帧图像各自对应的面部特征；
基于所述面部特征关系处理层，根据每帧图像各自对应的视频时刻点，确定每帧图像各自对应的面部特征之间的面部特征关系；
基于所述特征融合层，根据每帧图像各自对应的所述面部特征关系，对每帧图像各自对应的所述面部特征进行融合，得到目标面部特征；
基于所述面部识别层，根据所述目标面部特征确定所述视频数据中用户的面部识别结果。
第二方面，本申请还提供一种面部识别装置，所述面部识别装置包括：
获取模块，用于获取待识别的视频数据和面部识别模型，其中，所述面部识别模型包括面部特征提取层、面部特征关系处理层、特征融合层和面部识别层；
特征提取模块，用于基于所述面部特征提取层对所述视频数据中的每帧图像进行特征提取，得到每帧图像各自对应的面部特征；
特征关系确定模块，用于基于所述面部特征关系处理层，根据每帧图像各自对应的视频时刻点，确定每帧图像各自对应的面部特征之间的面部特征关系；
特征融合模块，用于基于所述特征融合层，根据每帧图像各自对应的所述面部特征关系，对每帧图像各自对应的所述面部特征进行融合，得到目标面部特征；
面部识别模块，用于基于所述面部识别层，根据所述目标面部特征确定所述视频数据中用户的面部识别结果。
第三方面，本申请还提供一种计算机设备，所述计算机设备包括处理器、存储器、以及存储在所述存储器上并可被所述处理器执行的计算机程序，其中所述计算机程序被所述处理器执行时，实现如下步骤：
获取待识别的视频数据和面部识别模型，其中，所述面部识别模型包括面部特征提取层、面部特征关系处理层、特征融合层和面部识别层；
基于所述面部特征提取层对所述视频数据中的每帧图像进行特征提取，得到每帧图像各自对应的面部特征；
基于所述面部特征关系处理层，根据每帧图像各自对应的视频时刻点，确定每帧图像各自对应的面部特征之间的面部特征关系；
基于所述特征融合层，根据每帧图像各自对应的所述面部特征关系，对每帧图像各自对应的所述面部特征进行融合，得到目标面部特征；
基于所述面部识别层，根据所述目标面部特征确定所述视频数据中用户的面部识别结果。
第四方面，本申请还提供一种计算机可读存储介质，所述计算机可读存储介质上存储有计算机程序，其中所述计算机程序被处理器执行时，实现如下步骤：
获取待识别的视频数据和面部识别模型，其中，所述面部识别模型包括面部特征提取层、面部特征关系处理层、特征融合层和面部识别层；
基于所述面部特征提取层对所述视频数据中的每帧图像进行特征提取，得到每帧图像各自对应的面部特征；
基于所述面部特征关系处理层，根据每帧图像各自对应的视频时刻点，确定每帧图像各自对应的面部特征之间的面部特征关系；
基于所述特征融合层，根据每帧图像各自对应的所述面部特征关系，对每帧图像各自对应的所述面部特征进行融合，得到目标面部特征；
基于所述面部识别层，根据所述目标面部特征确定所述视频数据中用户的面部识别结果。
本申请提供一种面部识别方法、装置、设备及计算机可读存储介质，本申请通过提取视频中每帧图像各自对应的面部特征，确定每帧图像各自对应的面部特征和每帧图像各自对应的面部特征之间的面部特征关系，基于该面部特征关系对面部特征进行融合，可以得到准确的用户的面部特征，再基于面部特征，对用户进行面部识别，可以有效的提高面部识别的准确性，有效的降低欺诈风险。
 附图说明
为了更清楚地说明本申请实施例技术方案，下面将对实施例描述中所需要使用的附图作简单地介绍，显而易见地，下面描述中的附图是本申请的一些实施例，对于本领域普通技术人员来讲，在不付出创造性劳动的前提下，还可以根据这些附图获得其他的附图。
图1为本申请实施例提供的一种面部识别方法的流程示意图；
图2为实施本实施例提供的面部识别方法的一场景示意图；
图3为本申请实施例提供的另一种面部识别方法的流程示意图；
图4为本申请实施例提供的一种面部识别装置的示意性框图；
图5为本申请实施例提供的另一种面部识别装置的示意性框图；
图6为本申请一实施例涉及的计算机设备的结构示意框图。
本申请目的实现、功能特点及优点将结合实施例，参照附图做进一步说明。
 本发明的实施方式
下面将结合本申请实施例中的附图，对本申请实施例中的技术方案进行清楚、完整地描述，显然，所描述的实施例是本申请一部分实施例，而不是全部的实施例。基于本申请中的实施例，本领域普通技术人员在没有做出创造性劳动前提下所获得的所有其他实施例，都属于本申请保护的范围。
附图中所示的流程图仅是示例说明，不是必须包括所有的内容和操作/步骤，也不是必须按所描述的顺序执行。例如，有的操作/步骤还可以分解、组合或部分合并，因此实际执行的顺序有可能根据实际情况改变。
本申请实施例提供一种面部识别方法、装置、设备及计算机可读存储介质。其中，该该面部识别方法可应用于服务器中，该服务器可以为单台的服务器，也可以为由多台服务器组成的服务器集群。
下面结合附图，对本申请的一些实施方式作详细说明。在不冲突的情况下，下述的实施例及实施例中的特征可以相互组合。
请参照图1，图1为本申请的实施例提供的一种面部识别方法的流程示意图。
如图1所示，该面部识别方法包括步骤S101至步骤S105。
步骤S101、获取待识别的视频数据和面部识别模型，其中，所述面部识别模型包括面部特征提取层、面部特征关系处理层、特征融合层和面部识别层。
服务器可以实时或以间隔预设时间获取待识别的视频数据，该待识别的视频数据包括用户的面部信息。其中，用户在申请贷款或信用卡时，可以通过终端设备录制包含用户面部信息的视频数据，并将该视频数据上传至服务器，服务器存储该视频数据，服务器将需要审核的视频数据写入视频数据审核队列中，便于服务器后续定时或实时的从视频数据审核队列中获取一个视频数据作为待识别的视频数据进行审核，确定申请贷款或信用卡的用户是否存在欺诈嫌疑。
服务器也可以实时对终端设备上传的视频数据进行面部识别，以信用卡申请场景为例：当用户需要申请信用卡时，可以通过终端设备远程申请信用卡，具体地，终端设备接收用户触发的信用卡申请指令时，基于该信用卡申请指令显示信息录入页面。当终端接收到基于该信息录入页面输入的客户信息，并检测到客户信息确认指令时，终端显示视频录制页面。当视频录制完毕后，终端接收该录制的信用卡申请视频，并在监测到信用卡确认申请指令时，生成携带有该客户信息和信用卡申请视频的信用卡申请请求，并且将该信用卡申请请求发送至服务器。
当服务器接收到终端设备发送的信用卡申请请求时，从该信用卡申请请求中获取信用卡申请视频，并将该信用卡申请视频作为需要识别面部动作的目标视频，且获取存储在本地的面部识别模型。其中，该用户信息包括但不限于用户姓名、身份证信息、手机号码、学历信息、常住地址、公司名称、公司地址、社保信息和公积金信息。
其中，服务器或服务器集群存储有面部识别模型，该面部识别模型包括面部特征提取层、面部特征关系处理层、特征融合层和面部识别层，该面部特征提取层用于从目标视频中提取面部特征，该面部特征关系处理层用于确定每帧图像的面部特征之间的面部特征关系，该特征融合层用于基于每帧图像的面部特征之间的面部特征关系，对面部特征进行融合，该面部识别层用于从视频中识别用户的面部动作，并得到面部识别结果。其中，该终端设备可以为手机、平板电脑、笔记本电脑、台式电脑、个人数字助理和穿戴式设备等电子设备。
具体地，该面部识别模型是基于训练得到的，训练的具体过程为：收集大量的标注有面部动作的视频数据，并将标注有面部动作的视频数据作为模型样本数据，然后利用该模型样本数据对面部识别模型中的面部动作识别层进行迭代训练，直到模型收敛。其中，该面部识别模型是基于深度神经网络设计，且该面部识别模型的待训练模型参数为面部动作识别层的模型参数。
步骤S102、基于所述面部特征提取层对所述视频数据中的每帧图像进行特征提取，得到每帧图像各自对应的面部特征。
服务器对该待识别的视频数据中的每帧图像进行特征提取，得到每帧图像各自对应的面部特征，即通过面部识别模型中的面部特征提取层从该待识别的视频数据中的每帧图像提取出每帧图像各自对应的面部特征。
具体地，将该待识别的视频数据拆分为每一帧图像，并从每帧图像中分别提取预设数量的面部关键点；根据从每帧图像中分别提取预设数量的面部关键点，确定每帧图像各自对应的面部特征。以一帧图像为例，从一帧图像中提取到128个面部关键点，则将提取到的128个面部关键点作为这一帧图像对应的面部特征。需要说明的是，上述预设数量可基于实际情况进行设置，本申请对此不作具体限定。可选地，面部关键点的数量为128个或512个，则面部特征为由128个或512个面部关键点组成的特征矩阵。其中，面部特征提取算法包括但不限于基于Gabor滤波的面部特征提取算法、基于局部二值化的面部特征提取算法和基于深度神经网络的面部特征提取算法和基于几何特征的面部特征提取算法，本申请对此不作具体限定。
步骤103、基于所述面部特征关系处理层，根据每帧图像各自对应的视频时刻点，确定每帧图像各自对应的面部特征之间的面部特征关系。
在获取到每帧图像各自对应的面部特征之后，通过面部识别模型中的面部特征关系处理层根据每帧图像各自对应的视频时刻点，确定每帧图像各自对应的面部特征之间的面部特征关系。需要说明的是，该视频时刻点为每帧图像在待识别的视频数据中的时刻点，用于表示每帧图像的时间顺序。
具体地，按照每帧图像的视频时刻点，对每帧图像各自对应的面部特征进行排序，得到面部特征队列；根据该面部特征队列，确定每帧图像各自对应的面部特征之间的面部特征关系。需要说明的是，视频时刻点越小，则面部特征的排序越靠前，视频时刻点越大，则面部特征的排序越靠后。
其中，基于面部特征队列确定面部特征关系的方式具体为：依次从该面部特征队列中选择一个面部特征作为目标面部特征，并从该面部特征队列中获取与该目标面部特征相邻的面部特征和不与该目标面部特征相邻的面部特征，从而得到该目标面部特征与其余每个面部特征之间的面部特征关系，即相邻关系和非相邻关系。
例如，面部特征队列为[A、B、C、D]，则可以得到面部特征A与面部特征B之间的面部特征关系为相邻关系，面部特征A与面部特征C之间的面部特征关系为非相邻关系，面部特征A与面部特征D之间的面部特征关系为非相邻关系，面部特征B与面部特征C之间的面部特征关系为相邻关系，面部特征B与面部特征D之间的面部特征关系为非相邻关系，面部特征C与面部特征D之间的面部特征关系为相邻关系。
步骤104、基于所述特征融合层，根据每帧图像各自对应的所述面部特征关系，对每帧图像各自对应的所述面部特征进行融合，得到目标面部特征。
在确定每帧图像各自对应的面部特征和面部特征关系之后，基于特征融合层，根据每帧图像各自对应的面部特征关系，对每帧图像各自对应的面部特征进行融合，得到目标面部特征。通过每帧图像各自对应的面部特征关系，对每帧图像各自对应的面部特征进行融合，可以得到准确的面部特征，可以提高面部识别的准确性。
步骤105、基于所述面部识别层，根据所述目标面部特征确定所述视频数据中用户的面部识别结果。
通过该面部识别模型中的面部识别层基于目标面部特征确定视频数据中用户的面部识别结果。在一实施例中，确定预设的面部特征集合是否包含该目标面部特征，如果预设的面部特征集合包含该目标面部特征，则面部识别结果为未通过，用户的欺诈嫌疑较高，如果预设的面部特征集合不包含该目标面部特征，则面部识别结果为通过，用户的欺诈嫌疑较低或没有欺诈嫌疑。需要说明的是，上述面部特征集合可基于实际情况进行设置，例如可设置舔嘴唇、面颊上扬，嘴角下垂和瞳孔放大等对应的面部特征，本申请对此不作具体限定。
在一实施例中，基于面部识别层，根据目标面部特征确定视频数据中用户的面部动作，并根据面部动作，确定用户的面部识别结果。其中，面部动作的确定方式具体为：获取预存的面部特征与面部动作之间的映射关系表，查询该映射关系表，将该目标面部特征对应的面部动作作为视频数据中用户的面部动作。需要说明的是，面部特征与面部动作之间的映射关系表可基于实际情况进行设置，本申请对此不作具体限定。
其中，基于面部动作，确定用户的面部识别结果的具体方式为：确定该面部动作是否位于预设面部动作集合，如果该面部动作位于预设面部动作集合，则面部识别结果为未通过，用户的欺诈嫌疑较高，如果该面部动作不位于该预设面部动作集合，则面部识别结果为通过，用户的欺诈嫌疑较低或没有欺诈嫌疑。需要说明的是，上述预设面部动作集合中的面部动作为面部识别结果为通过对应的面部动作，其中，该预设面部动作集合的面部动作表示的行为结果可基于实际情况进行设置，例如可设置舔嘴唇、面颊上扬，嘴角下垂和瞳孔放大等面部动作，本申请对此不作具体限定。
在另一实施例中，基于面部动作，确定用户的面部识别结果的方式还可以为：在确定用户的面部动作之后，将确定的面部动作输入至面部识别结果的确定模型，由面部识别结果的确定模型输出该面部动作对应的面部识别结果，该面部识别结果包括面部识别通过和面部识别未通过。其中，该面部识别的确定模型通过训练得到，具体的训练过程为：获取大量标注有面部识别结果和面部动作的模型样本数据，并基于神经网络、卷积神经网络和/或循环卷积神经网络设计面部识别模型，再通过该模型样本数据对设计的面部识别结果的确定模型进行训练，直到面部识别结果的确定模型收敛。
在一实施例中，在面部识别结果为预设面部识别结果时，向终端发送风险提醒信息，并执行信息安全验证操作。其中，终端接收到风险提醒信息时，可以提醒用户存在风险，需要验证信息安全。信息安全验证操作具体为：终端在接收到风险提醒信息时，显示信息安全验证页面，其中，信息安全验证页面显示有风险提醒信息和短信验证码的触发控件；当检测到用户对该短信验证码的触发控件的触控操作时，向服务器发送短信验证码获取请求；接收服务器基于该短信验证码获取请求返回的短信验证码，并获取用户在信息安全验证页面输入的短信验证码，如果用户输入的短信验证码与接收到的短信验证码一致，则通过信息安全验证，反之如果用户输入的短信验证码与接收到的短信验证码不一致，则未通过信息安全验证。
请参照图2，图2为实施本实施例提供的面部识别方法的一场景示意图。如图2所示，包括终端设备和服务器。该终端设备用于接收用户申请信用卡指令、录入用户信息和录制信用卡申请视频，服务器用于接收接收终端生成的携带有用户信息和信用卡申请视频的请求，并对信用卡申请视频进行面部识别。
上述实施例提供的面部识别方法，通过提取视频中每帧图像各自对应的面部特征，确定每帧图像各自对应的面部特征和每帧图像各自对应的面部特征之间的面部特征关系，基于该面部特征关系对面部特征进行融合，可以得到用户准确的面部特征，再基于面部特征，对用户进行面部识别，可以有效的提高面部识别的准确性，有效的降低欺诈风险。
请参照图3，图3为本申请的实施例提供的一种面部识别方法的流程示意图。
如图3所示，该面部识别方法包括步骤S201至步骤S207。
步骤S201、获取待识别的视频数据和面部识别模型，其中，所述面部识别模型包括面部特征提取层、面部特征关系处理层、特征融合层和面部识别层。
服务器可以实时或以间隔预设时间获取待识别的视频数据，该待识别的视频数据包括用户的面部信息。其中，用户在申请贷款或信用卡时，可以通过终端设备录制包含用户面部信息的视频数据，并将该视频数据上传至服务器，服务器存储该视频数据，服务器将需要审核的视频数据写入视频数据审核队列中，便于服务器后续定时或实时的从视频数据审核队列中获取一个视频数据作为待识别的视频数据进行审核，确定申请贷款或信用卡的用户是否存在欺诈嫌疑。
其中，服务器或服务器集群存储有面部识别模型，该面部识别模型包括面部特征提取层、面部特征关系处理层和面部动作识别层，该面部特征提取层用于从目标视频中提取面部特征，该面部特征关系处理层用于确定每帧图像的面部特征之间的面部特征关系，该面部动作识别层用于从视频中识别用户的面部动作。其中，该终端设备可以为手机、平板电脑、笔记本电脑、台式电脑、个人数字助理和穿戴式设备等电子设备。
步骤S202、基于所述面部特征提取层对所述视频数据中的每帧图像进行特征提取，得到每帧图像各自对应的面部特征。
服务器对该待识别的视频数据中的每帧图像进行特征提取，得到每帧图像各自对应的面部特征，即通过面部识别模型中的面部特征提取层从该待识别的视频数据中的每帧图像提取出每帧图像各自对应的面部特征。
具体地，将该待识别的视频数据拆分为每一帧图像，并从每帧图像中分别提取预设数量的面部关键点；根据从每帧图像中分别提取预设数量的面部关键点，确定每帧图像各自对应的面部特征。以一帧图像为例，从一帧图像中提取到128个面部关键点，则将提取到的128个面部关键点作为这一帧图像对应的面部特征。
步骤S203、基于所述面部特征关系处理层，根据每帧图像各自对应的视频时刻点，确定每帧图像各自对应的面部特征之间的面部特征关系。
在获取到每帧图像各自对应的面部特征之后，通过面部识别模型中的面部特征关系处理层根据每帧图像各自对应的视频时刻点，确定每帧图像各自对应的面部特征之间的面部特征关系。需要说明的是，该视频时刻点为每帧图像在待识别的视频数据中的时刻点，用于表示每帧图像的时间顺序。
步骤S204、基于所述特征融合层，根据每帧图像各自对应的所述面部特征关系，对每帧图像各自对应的面部特征进行排序，得到目标面部特征队列。
在获取到每帧图像各自对应的面部特征和面部特征关系之后，基于特征融合层，根据每帧图像各自对应的面部特征关系，对每帧图像各自对应的面部特征进行排序，得到目标面部特征队列。例如，每帧图像各自对应的面部特征分别为面部特征1、面部特征2、面部特征3和面部特征4，且面部特征关系分别为面部特征1与面部特征3相邻，与面部特征1和面部特征2不相邻；面部特征3与面部特征1和面部特征4相邻，与面部特征2不相邻；面部特征2与面部特征4相邻，与面部特征1和面部特征3不相邻；面部特征4与面部特征3和面部特征2相邻，与面部特征1不相邻，通过上述面部特征关系，得到的目标面部特征队列为[1、3、4、2]。
步骤S205、将所述目标面部特征队列中的所述面部特征进行两两比较，得到所述目标面部特征队列中每两个所述面部特征的比较结果。
在得到目标面部特征队列之后，将目标面部特征队列中的面部特征进行两两比较，得到目标面部特征队列中每两个面部特征的比较结果。其中，比较结果包括面部特征相同和面部特征不相同。例如，将目标面部特征队列为[1、3、4、2]中的面部特征1分别与面部特征2、面部特征3和面部特征4进行比较，以及将面部特征2分别与面部特征3和面部特征4进行比较，以及将面部特征3与面部特征4进行比较，通过比较发现，仅面部特征1与面部特征3相同，其余均不相同。
步骤S206、根据每两个所述面部特征的比较结果，对每帧图像各自对应的所述面部特征进行融合，得到目标面部特征。
在得到每两个面部特征的比较结果之后，根据每两个面部特征的比较结果，对每帧图像各自对应的面部特征进行融合，得到目标面部特征。具体地，根据每两个面部特征的比较结果，对目标面部特征队列中的面部特征进行删除处理，其中，经过删除处理后的目标面部特征队列中的每个面部特征均不相同；将经过删除处理后的目标面部特征队列中的每个面部特征进行融合，得到目标面部特征。例如，目标面部特征队列为[1、3、4、2]中的面部特征1、面部特征2、面部特征3和面部特征4之间的比较结果为面部特征1与面部特征3相同，其余均不相同，则经过删除处理后的目标面部特征队列为[1、4、2]。
其中，面部特征的融合的具体方式为：从经过删除处理后的目标面部特征队列中选择一个面部特征，作为待融合的第一面部特征，再从经过删除处理后的目标面部特征队列中选择一个其余的面部特征，作为待融合的第二面部特征；将第一面部特征中的面部关键点与第二面部特征中的面部关键点进行比较，得到不同的面部关键点，将第二面部特征中不同的面部关键点融合至第一面部特征，以更新第一面部特征；再从经过删除处理后的目标面部特征队列中选择一个其余的面部特征，按照同样的方式与更新后的第一面部特征进行融合，重复几次之后，经过删除处理后的目标面部特征队列中的最后一个面部特征与每次融合更新得到第一面部特征进行融合，即可得到目标面部特征。
其中，将第二面部特征中不同的面部关键点融合至第一面部特征的方式具体为：从第二面部特征中获取不同的面部关键点，以形成第一特征矩阵，并从第一面部特征获取所述不同的面部关键点对应的面部关键点，以形成第二特征矩阵；将第一特征矩阵累加到第二特征矩阵，从而得到融合特征矩阵，再将融合特征矩阵中的面部关键点与第一面部特征中的其余面部关键点的集合作为更新的第一面部特征。
步骤S207，基于所述面部识别层，根据所述目标面部特征确定所述视频数据中用户的面部识别结果。
通过该面部识别模型中的面部识别层基于目标面部特征确定视频数据中用户的面部识别结果。在一实施例中，确定预设的面部特征集合是否包含该目标面部特征，如果预设的面部特征集合包含该目标面部特征，则面部识别结果为未通过，用户的欺诈嫌疑较高，如果预设的面部特征集合不包含该目标面部特征，则面部识别结果为通过，用户的欺诈嫌疑较低或没有欺诈嫌疑。需要说明的是，上述面部特征集合可基于实际情况进行设置，例如可设置舔嘴唇、面颊上扬，嘴角下垂和瞳孔放大等对应的面部特征，本申请对此不作具体限定。
上述实施例提供的面部识别方法，通过提取视频中每帧图像各自对应的面部特征，确定每帧图像各自对应的面部特征和每帧图像各自对应的面部特征之间的面部特征关系，根据该面部特征关系可以对每帧图像各自对应的面部特征进行排序，得到目标面部特征队列，再根据目标面部特征队列中每两个面部特征的比较结果，对每帧图像各自对应的面部特征进行融合，然后基于融合后的面部特征，可以准确地对用户进行面部识别，能够有效的提高面部识别的准确性。
请参照图4，图4为本申请实施例提供的一种面部识别装置的示意性框图。
如图4所示，该面部识别装置300，包括：获取模块301、特征提取模块302、特征关系确定模块303、特征融合模块304和面部识别模块305。
获取模块301，用于获取待识别的视频数据和面部识别模型，其中，所述面部识别模型包括面部特征提取层、面部特征关系处理层、特征融合层和面部识别层；
特征提取模块302，用于基于所述面部特征提取层对所述视频数据中的每帧图像进行特征提取，得到每帧图像各自对应的面部特征；
特征关系确定模块303，用于基于所述面部特征关系处理层，根据每帧图像各自对应的视频时刻点，确定每帧图像各自对应的面部特征之间的面部特征关系；
特征融合模块304，用于基于所述特征融合层，根据每帧图像各自对应的所述面部特征关系，对每帧图像各自对应的所述面部特征进行融合，得到目标面部特征；
面部识别模块305，用于基于所述面部识别层，根据所述目标面部特征确定所述视频数据中用户的面部识别结果。
在一实施例中，所述特征提取模块302，还用于基于所述面部特征提取层将所述待识别的视频数据拆分为每一帧图像，并从每帧图像中分别提取预设数量的面部关键点；根据从每帧图像中分别提取预设数量的面部关键点，确定每帧图像各自对应的面部特征。
在一实施例中，所述特征关系确定模块303，还用于基于所述面部特征关系处理层，按照每帧图像各自对应的视频时刻点，对每帧图像各自对应的面部特征进行排序，得到面部特征队列；根据所述面部特征队列，确定每帧图像各自对应的面部特征之间的面部特征关系。
在一实施例中，所述面部识别模块305，还用于基于所述面部识别层，根据所述目标面部特征确定所述视频数据中用户的面部动作，并根据所述面部动作，确定所述用户的面部识别结果。
在一实施例中，所述面部识别装置300还包括：
发送模块，用于在所述面部识别结果为预设面部识别结果时，向终端发送风险提醒信息；
安全验证模块，用于执行信息安全验证操作。
请参照图5，图5为本申请实施例提供的另一种面部识别装置的示意性框图。
如图5所示，该面部识别装置400，包括：获取模块401、特征提取模块402、特征关系确定模块403、排序模块404、特征比较模块405、特征融合模块406和面部识别模块407。
获取模块401，用于获取待识别的视频数据和面部识别模型，其中，所述面部识别模型包括面部特征提取层、面部特征关系处理层、特征融合层和面部识别层。
特征提取模块402，用于基于所述面部特征提取层对所述视频数据中的每帧图像进行特征提取，得到每帧图像各自对应的面部特征。
特征关系确定模块403，用于基于所述面部特征关系处理层，根据每帧图像各自对应的视频时刻点，确定每帧图像各自对应的面部特征之间的面部特征关系。
排序模块404，用于基于所述特征融合层，根据每帧图像各自对应的所述面部特征关系，对每帧图像各自对应的面部特征进行排序，得到目标面部特征队列。
特征比较模块405，用于将所述目标面部特征队列中的所述面部特征进行两两比较，得到所述目标面部特征队列中每两个所述面部特征的比较结果。
特征融合模块406，用于根据每两个所述面部特征的比较结果，对每帧图像各自对应的所述面部特征进行融合，得到目标面部特征。
面部识别模块407，用于基于所述面部识别层，根据所述目标面部特征确定所述视频数据中用户的面部识别结果。
在一实施例中，所述特征融合模块406，还用于根据每两个所述面部特征的比较结果，对所述目标面部特征队列中的面部特征进行删除处理，其中，经过删除处理后的所述目标面部特征队列中的每个所述面部特征均不相同；将经过删除处理后的所述目标面部特征队列中的每个所述面部特征进行融合，得到目标面部特征。
需要说明的是，所属领域的技术人员可以清楚地了解到，为了描述的方便和简洁，上述描述的装置和各模块及单元的具体工作过程，可以参考前述面部识别方法实施例中的对应过程，在此不再赘述。
上述实施例提供的装置可以实现为一种计算机程序的形式，该计算机程序可以在如图6所示的计算机设备上运行。
请参阅图6，图6为本申请实施例提供的一种计算机设备的结构示意性框图。该计算机设备可以为服务器。
如图6所示，该计算机设备包括通过系统总线连接的处理器、存储器和网络接口，其中，存储器可以包括非易失性或易失性存储介质和内存储器。
非易失性或易失性存储介质可存储操作系统和计算机程序。该计算机程序包括程序指令，该程序指令被执行时，可使得处理器执行任意一种面部识别方法。
处理器用于提供计算和控制能力，支撑整个计算机设备的运行。
内存储器为非易失性或易失性存储介质中的计算机程序的运行提供环境，该计算机程序被处理器执行时，可使得处理器执行任意一种面部识别方法。
该网络接口用于进行网络通信，如发送分配的任务等。本领域技术人员可以理解，图6中示出的结构，仅仅是与本申请方案相关的部分结构的框图，并不构成对本申请方案所应用于其上的计算机设备的限定，具体的计算机设备可以包括比图中所示更多或更少的部件，或者组合某些部件，或者具有不同的部件布置。
应当理解的是，处理器可以是中央处理单元 (Central Processing Unit，CPU)，该处理器还可以是其他通用处理器、数字信号处理器 (Digital Signal Processor，DSP)、专用集成电路 (Application Specific Integrated Circuit，ASIC)、现场可编程门阵列 (Field-Programmable Gate Array，FPGA) 或者其他可编程逻辑器件、分立门或者晶体管逻辑器件、分立硬件组件等。其中，通用处理器可以是微处理器或者该处理器也可以是任何常规的处理器等。
其中，在一个实施例中，所述处理器用于运行存储在存储器中的计算机程序，以实现如下步骤：
获取待识别的视频数据和面部识别模型，其中，所述面部识别模型包括面部特征提取层、面部特征关系处理层、特征融合层和面部识别层；
基于所述面部特征提取层对所述视频数据中的每帧图像进行特征提取，得到每帧图像各自对应的面部特征；
基于所述面部特征关系处理层，根据每帧图像各自对应的视频时刻点，确定每帧图像各自对应的面部特征之间的面部特征关系；
基于所述特征融合层，根据每帧图像各自对应的所述面部特征关系，对每帧图像各自对应的所述面部特征进行融合，得到目标面部特征；
基于所述面部识别层，根据所述目标面部特征确定所述视频数据中用户的面部识别结果。
在一个实施例中，所述处理器在实现基于所述面部特征提取层对所述视频数据中的每帧图像进行特征提取，得到每帧图像各自对应的面部特征时，用于实现：
基于所述面部特征提取层将所述待识别的视频数据拆分为每一帧图像，并从每帧图像中分别提取预设数量的面部关键点；
根据从每帧图像中分别提取预设数量的面部关键点，确定每帧图像各自对应的面部特征。
在一个实施例中，所述处理器在实现基于所述面部特征关系处理层，根据每帧图像各自对应的视频时刻点，确定每帧图像各自对应的面部特征之间的面部特征关系时，用于实现：
基于所述面部特征关系处理层，按照每帧图像各自对应的视频时刻点，对每帧图像各自对应的面部特征进行排序，得到面部特征队列；
根据所述面部特征队列，确定每帧图像各自对应的面部特征之间的面部特征关系。
在一个实施例中，所述处理器在实现基于所述特征融合层，根据每帧图像各自对应的所述面部特征关系，对每帧图像各自对应的所述面部特征进行融合，得到目标面部特征时，用于实现：
基于所述特征融合层，根据每帧图像各自对应的所述面部特征关系，对每帧图像各自对应的面部特征进行排序，得到目标面部特征队列；
将所述目标面部特征队列中的所述面部特征进行两两比较，得到所述目标面部特征队列中每两个所述面部特征的比较结果；
根据每两个所述面部特征的比较结果，对每帧图像各自对应的所述面部特征进行融合，得到目标面部特征。
在一个实施例中，所述处理器在实现根据每两个所述面部特征的比较结果，对每帧图像各自对应的所述面部特征进行融合，得到目标面部特征，用于实现：
根据每两个所述面部特征的比较结果，对所述目标面部特征队列中的面部特征进行删除处理，其中，经过删除处理后的所述目标面部特征队列中的每个所述面部特征均不相同；
将经过删除处理后的所述目标面部特征队列中的每个所述面部特征进行融合，得到目标面部特征。
在一个实施例中，所述处理器在实现基于所述面部识别层，根据所述目标面部特征确定所述视频数据中用户的面部识别结果时，用于实现：
基于所述面部识别层，根据所述目标面部特征确定所述视频数据中用户的面部动作，并根据所述面部动作，确定所述用户的面部识别结果。
在一个实施例中，所述处理器在实现基于所述面部识别层，根据所述目标面部特征确定所述视频数据中用户的面部识别结果之后，还用于实现：
在所述面部识别结果为预设面部识别结果时，向终端发送风险提醒信息，并执行信息安全验证操作。
需要说明的是，所属领域的技术人员可以清楚地了解到，为了描述的方便和简洁，上述描述的计算机设备的具体工作过程，可以参考前述面部识别方法实施例中的对应过程，在此不再赘述。
本申请实施例还提供一种计算机可读存储介质，所述计算机可读存储介质可以是易失性，也可以是非易失性，所述计算机可读存储介质上存储有计算机程序，所述计算机程序中包括程序指令，所述程序指令被执行时所实现的方法可参照本申请面部识别方法的各个实施例。
其中，所述计算机可读存储介质可以是前述实施例所述的计算机设备的内部存储单元，例如所述计算机设备的硬盘或内存。所述计算机可读存储介质也可以是所述计算机设备的外部存储设备，例如所述计算机设备上配备的插接式硬盘，智能存储卡(Smart Media Card，SMC)，安全数字(Secure Digital，SD)卡，闪存卡(Flash Card)等。
应当理解，在此本申请说明书中所使用的术语仅仅是出于描述特定实施例的目的而并不意在限制本申请。如在本申请说明书和所附权利要求书中所使用的那样，除非上下文清楚地指明其它情况，否则单数形式的“一”、“一个”及“该”意在包括复数形式。
还应当理解，在本申请说明书和所附权利要求书中使用的术语“和/ 或”是指相关联列出的项中的一个或多个的任何组合以及所有可能组合，并且包括这些组合。需要说明的是，在本文中，术语“包括”、“包含”或者其任何其他变体意在涵盖非排他性的包含，从而使得包括一系列要素的过程、方法、物品或者系统不仅包括那些要素，而且还包括没有明确列出的其他要素，或者是还包括为这种过程、方法、物品或者系统所固有的要素。在没有更多限制的情况下，由语句“包括一个……”限定的要素，并不排除在包括该要素的过程、方法、物品或者系统中还存在另外的相同要素。
上述本申请实施例序号仅仅为了描述，不代表实施例的优劣。以上所述，仅为本申请的具体实施方式，但本申请的保护范围并不局限于此，任何熟悉本技术领域的技术人员在本申请揭露的技术范围内，可轻易想到各种等效的修改或替换，这些修改或替换都应涵盖在本申请的保护范围之内。因此，本申请的保护范围应以权利要求的保护范围为准。

Claims (20)

  1. 一种面部识别方法，其中，包括：
    获取待识别的视频数据和面部识别模型，其中，所述面部识别模型包括面部特征提取层、面部特征关系处理层、特征融合层和面部识别层；
    基于所述面部特征提取层对所述视频数据中的每帧图像进行特征提取，得到每帧图像各自对应的面部特征；
    基于所述面部特征关系处理层，根据每帧图像各自对应的视频时刻点，确定每帧图像各自对应的面部特征之间的面部特征关系；
    基于所述特征融合层，根据每帧图像各自对应的所述面部特征关系，对每帧图像各自对应的所述面部特征进行融合，得到目标面部特征；
    基于所述面部识别层，根据所述目标面部特征确定所述视频数据中用户的面部识别结果。
  2. 根据权利要求1所述的面部识别方法，其中，所述基于所述面部特征提取层对所述视频数据中的每帧图像进行特征提取，得到每帧图像各自对应的面部特征，包括：
    基于所述面部特征提取层将所述待识别的视频数据拆分为每一帧图像，并从每帧图像中分别提取预设数量的面部关键点；
    根据从每帧图像中分别提取预设数量的面部关键点，确定每帧图像各自对应的面部特征。
  3. 根据权利要求1所述的面部识别方法，其中，所述基于所述面部特征关系处理层，根据每帧图像各自对应的视频时刻点，确定每帧图像各自对应的面部特征之间的面部特征关系，包括：
    基于所述面部特征关系处理层，按照每帧图像各自对应的视频时刻点，对每帧图像各自对应的面部特征进行排序，得到面部特征队列；
    根据所述面部特征队列，确定每帧图像各自对应的面部特征之间的面部特征关系。
  4. 根据权利要求1至3中任一项所述的面部识别方法，其中，所述基于所述特征融合层，根据每帧图像各自对应的所述面部特征关系，对每帧图像各自对应的所述面部特征进行融合，得到目标面部特征，包括：
    基于所述特征融合层，根据每帧图像各自对应的所述面部特征关系，对每帧图像各自对应的面部特征进行排序，得到目标面部特征队列；
    将所述目标面部特征队列中的所述面部特征进行两两比较，得到所述目标面部特征队列中每两个所述面部特征的比较结果；
    根据每两个所述面部特征的比较结果，对每帧图像各自对应的所述面部特征进行融合，得到目标面部特征。
  5. 根据权利要求4所述的面部识别方法，其中，所述根据每两个所述面部特征的比较结果，对每帧图像各自对应的所述面部特征进行融合，得到目标面部特征，包括：
    根据每两个所述面部特征的比较结果，对所述目标面部特征队列中的面部特征进行删除处理，其中，经过删除处理后的所述目标面部特征队列中的每个所述面部特征均不相同；
    将经过删除处理后的所述目标面部特征队列中的每个所述面部特征进行融合，得到目标面部特征。
  6. 根据权利要求1至3中任一项所述的面部识别方法，其中，所述基于所述面部识别层，根据所述目标面部特征确定所述视频数据中用户的面部识别结果，包括：
    基于所述面部识别层，根据所述目标面部特征确定所述视频数据中用户的面部动作，并根据所述面部动作，确定所述用户的面部识别结果。
  7. 根据权利要求1至3中任一项所述的面部识别方法，其中，所述基于所述面部识别层，根据所述目标面部特征确定所述视频数据中用户的面部识别结果之后，还包括：
    在所述面部识别结果为预设面部识别结果时，向终端发送风险提醒信息，并执行信息安全验证操作。
  8. 一种面部识别装置，其中，所述面部识别装置包括：
    获取模块，用于获取待识别的视频数据和面部识别模型，其中，所述面部识别模型包括面部特征提取层、面部特征关系处理层、特征融合层和面部识别层；
    特征提取模块，用于基于所述面部特征提取层对所述视频数据中的每帧图像进行特征提取，得到每帧图像各自对应的面部特征；
    特征关系确定模块，用于基于所述面部特征关系处理层，根据每帧图像各自对应的视频时刻点，确定每帧图像各自对应的面部特征之间的面部特征关系；
    特征融合模块，用于基于所述特征融合层，根据每帧图像各自对应的所述面部特征关系，对每帧图像各自对应的所述面部特征进行融合，得到目标面部特征；
    面部识别模块，用于基于所述面部识别层，根据所述目标面部特征确定所述视频数据中用户的面部识别结果。
  9. 一种计算机设备，其中，所述计算机设备包括处理器、存储器、以及存储在所述存储器上并可被所述处理器执行的计算机程序，其中所述计算机程序被所述处理器执行时，实现如下步骤：
    获取待识别的视频数据和面部识别模型，其中，所述面部识别模型包括面部特征提取层、面部特征关系处理层、特征融合层和面部识别层；
    基于所述面部特征提取层对所述视频数据中的每帧图像进行特征提取，得到每帧图像各自对应的面部特征；
    基于所述面部特征关系处理层，根据每帧图像各自对应的视频时刻点，确定每帧图像各自对应的面部特征之间的面部特征关系；
    基于所述特征融合层，根据每帧图像各自对应的所述面部特征关系，对每帧图像各自对应的所述面部特征进行融合，得到目标面部特征；
    基于所述面部识别层，根据所述目标面部特征确定所述视频数据中用户的面部识别结果。
  10. 根据权利要求9所述的计算机设备，其中，所述基于所述面部特征提取层对所述视频数据中的每帧图像进行特征提取，得到每帧图像各自对应的面部特征，包括：
    基于所述面部特征提取层将所述待识别的视频数据拆分为每一帧图像，并从每帧图像中分别提取预设数量的面部关键点；
    根据从每帧图像中分别提取预设数量的面部关键点，确定每帧图像各自对应的面部特征。
  11. 根据权利要求9所述的计算机设备，其中，所述基于所述面部特征关系处理层，根据每帧图像各自对应的视频时刻点，确定每帧图像各自对应的面部特征之间的面部特征关系，包括：
    基于所述面部特征关系处理层，按照每帧图像各自对应的视频时刻点，对每帧图像各自对应的面部特征进行排序，得到面部特征队列；
    根据所述面部特征队列，确定每帧图像各自对应的面部特征之间的面部特征关系。
  12. 根据权利要求9至11中任一项所述的计算机设备，其中，所述基于所述特征融合层，根据每帧图像各自对应的所述面部特征关系，对每帧图像各自对应的所述面部特征进行融合，得到目标面部特征，包括：
    基于所述特征融合层，根据每帧图像各自对应的所述面部特征关系，对每帧图像各自对应的面部特征进行排序，得到目标面部特征队列；
    将所述目标面部特征队列中的所述面部特征进行两两比较，得到所述目标面部特征队列中每两个所述面部特征的比较结果；
    根据每两个所述面部特征的比较结果，对每帧图像各自对应的所述面部特征进行融合，得到目标面部特征。
  13. 根据权利要求12所述的计算机设备，其中，所述根据每两个所述面部特征的比较结果，对每帧图像各自对应的所述面部特征进行融合，得到目标面部特征，包括：
    根据每两个所述面部特征的比较结果，对所述目标面部特征队列中的面部特征进行删除处理，其中，经过删除处理后的所述目标面部特征队列中的每个所述面部特征均不相同；
    将经过删除处理后的所述目标面部特征队列中的每个所述面部特征进行融合，得到目标面部特征。
  14. 根据权利要求9至11中任一项所述的计算机设备，其中，所述基于所述面部识别层，根据所述目标面部特征确定所述视频数据中用户的面部识别结果，包括：
    基于所述面部识别层，根据所述目标面部特征确定所述视频数据中用户的面部动作，并根据所述面部动作，确定所述用户的面部识别结果。
  15. 根据权利要求9至11中任一项所述的计算机设备，其中，所述基于所述面部识别层，根据所述目标面部特征确定所述视频数据中用户的面部识别结果之后，所述计算机程序被所述处理器执行时还实现如下步骤：
    在所述面部识别结果为预设面部识别结果时，向终端发送风险提醒信息，并执行信息安全验证操作。
  16. 一种计算机可读存储介质，其中，所述计算机可读存储介质上存储有计算机程序，其中所述计算机程序被处理器执行时，实现如下步骤：
    获取待识别的视频数据和面部识别模型，其中，所述面部识别模型包括面部特征提取层、面部特征关系处理层、特征融合层和面部识别层；
    基于所述面部特征提取层对所述视频数据中的每帧图像进行特征提取，得到每帧图像各自对应的面部特征；
    基于所述面部特征关系处理层，根据每帧图像各自对应的视频时刻点，确定每帧图像各自对应的面部特征之间的面部特征关系；
    基于所述特征融合层，根据每帧图像各自对应的所述面部特征关系，对每帧图像各自对应的所述面部特征进行融合，得到目标面部特征；
    基于所述面部识别层，根据所述目标面部特征确定所述视频数据中用户的面部识别结果。
  17. 根据权利要求16所述的计算机可读存储介质，其中，所述基于所述面部特征提取层对所述视频数据中的每帧图像进行特征提取，得到每帧图像各自对应的面部特征，包括：
    基于所述面部特征提取层将所述待识别的视频数据拆分为每一帧图像，并从每帧图像中分别提取预设数量的面部关键点；
    根据从每帧图像中分别提取预设数量的面部关键点，确定每帧图像各自对应的面部特征。
  18. 根据权利要求16所述的计算机可读存储介质，其中，所述基于所述面部特征关系处理层，根据每帧图像各自对应的视频时刻点，确定每帧图像各自对应的面部特征之间的面部特征关系，包括：
    基于所述面部特征关系处理层，按照每帧图像各自对应的视频时刻点，对每帧图像各自对应的面部特征进行排序，得到面部特征队列；
    根据所述面部特征队列，确定每帧图像各自对应的面部特征之间的面部特征关系。
  19. 根据权利要求16至18中任一项所述的计算机可读存储介质，其中，所述基于所述特征融合层，根据每帧图像各自对应的所述面部特征关系，对每帧图像各自对应的所述面部特征进行融合，得到目标面部特征，包括：
    基于所述特征融合层，根据每帧图像各自对应的所述面部特征关系，对每帧图像各自对应的面部特征进行排序，得到目标面部特征队列；
    将所述目标面部特征队列中的所述面部特征进行两两比较，得到所述目标面部特征队列中每两个所述面部特征的比较结果；
    根据每两个所述面部特征的比较结果，对每帧图像各自对应的所述面部特征进行融合，得到目标面部特征。
  20. 根据权利要求19所述的计算机可读存储介质，其中，所述根据每两个所述面部特征的比较结果，对每帧图像各自对应的所述面部特征进行融合，得到目标面部特征，包括：
    根据每两个所述面部特征的比较结果，对所述目标面部特征队列中的面部特征进行删除处理，其中，经过删除处理后的所述目标面部特征队列中的每个所述面部特征均不相同；
    将经过删除处理后的所述目标面部特征队列中的每个所述面部特征进行融合，得到目标面部特征。
PCT/CN2020/106023 2019-10-12 2020-07-31 面部识别方法、装置、设备及计算机可读存储介质 WO2021068613A1 (zh)

Applications Claiming Priority (2)

Application Number Priority Date Filing Date Title
CN201910969394.4 2019-10-12
CN201910969394.4A CN110874570A (zh) 2019-10-12 2019-10-12 面部识别方法、装置、设备及计算机可读存储介质

Publications (1)

Publication Number Publication Date
WO2021068613A1 true WO2021068613A1 (zh) 2021-04-15

Family

ID=69717773

Family Applications (1)

Application Number Title Priority Date Filing Date
PCT/CN2020/106023 WO2021068613A1 (zh) 2019-10-12 2020-07-31 面部识别方法、装置、设备及计算机可读存储介质

Country Status (2)

Country Link
CN (1) CN110874570A (zh)
WO (1) WO2021068613A1 (zh)

Families Citing this family (3)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN110874570A (zh) * 2019-10-12 2020-03-10 深圳壹账通智能科技有限公司 面部识别方法、装置、设备及计算机可读存储介质
CN113762156B (zh) * 2021-09-08 2023-10-24 北京优酷科技有限公司 观影数据处理方法、装置及存储介质
CN115272923B (zh) * 2022-07-22 2023-04-21 华中科技大学同济医学院附属协和医院 一种基于大数据平台的智能识别方法和系统

Citations (6)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
US20180060649A1 (en) * 2016-08-30 2018-03-01 Irida Labs S.A. Fast, embedded, hybrid video face recognition system
CN108229330A (zh) * 2017-12-07 2018-06-29 深圳市商汤科技有限公司 人脸融合识别方法及装置、电子设备和存储介质
CN108446601A (zh) * 2018-02-27 2018-08-24 东南大学 一种基于动静特征融合的人脸识别方法
CN109325422A (zh) * 2018-08-28 2019-02-12 深圳壹账通智能科技有限公司 表情识别方法、装置、终端及计算机可读存储介质
CN109886110A (zh) * 2019-01-17 2019-06-14 深圳壹账通智能科技有限公司 微表情评分方法、装置、计算机设备及存储介质
CN110874570A (zh) * 2019-10-12 2020-03-10 深圳壹账通智能科技有限公司 面部识别方法、装置、设备及计算机可读存储介质

Patent Citations (6)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
US20180060649A1 (en) * 2016-08-30 2018-03-01 Irida Labs S.A. Fast, embedded, hybrid video face recognition system
CN108229330A (zh) * 2017-12-07 2018-06-29 深圳市商汤科技有限公司 人脸融合识别方法及装置、电子设备和存储介质
CN108446601A (zh) * 2018-02-27 2018-08-24 东南大学 一种基于动静特征融合的人脸识别方法
CN109325422A (zh) * 2018-08-28 2019-02-12 深圳壹账通智能科技有限公司 表情识别方法、装置、终端及计算机可读存储介质
CN109886110A (zh) * 2019-01-17 2019-06-14 深圳壹账通智能科技有限公司 微表情评分方法、装置、计算机设备及存储介质
CN110874570A (zh) * 2019-10-12 2020-03-10 深圳壹账通智能科技有限公司 面部识别方法、装置、设备及计算机可读存储介质

Non-Patent Citations (1)

* Cited by examiner, † Cited by third party
Title
MAHASSENI BEHROOZ; LAM MICHAEL; TODOROVIC SINISA: "Unsupervised Video Summarization with Adversarial LSTM Networks", 2017 IEEE CONFERENCE ON COMPUTER VISION AND PATTERN RECOGNITION (CVPR), IEEE COMPUTER SOCIETY, US, 21 July 2017 (2017-07-21), US, pages 2982 - 2991, XP033249644, ISSN: 1063-6919, DOI: 10.1109/CVPR.2017.318 *

Also Published As

Publication number Publication date
CN110874570A (zh) 2020-03-10

Similar Documents

Publication Publication Date Title
WO2020211388A1 (zh) 基于预测模型的行为预测方法、装置、设备及存储介质
CN105874474B (zh) 用于面部表示的系统和方法
US20210295114A1 (en) Method and apparatus for extracting structured data from image, and device
US10936672B2 (en) Automatic document negotiation
US11670415B2 (en) Data driven analysis, modeling, and semi-supervised machine learning for qualitative and quantitative determinations
WO2022105118A1 (zh) 基于图像的健康状态识别方法、装置、设备及存储介质
US10069891B2 (en) Channel accessible single function micro service data collection process for light analytics
CN109345553B (zh) 一种手掌及其关键点检测方法、装置和终端设备
WO2021068613A1 (zh) 面部识别方法、装置、设备及计算机可读存储介质
CN110321845B (zh) 一种从视频中提取表情包的方法、装置及电子设备
US10191920B1 (en) Graphical image retrieval based on emotional state of a user of a computing device
WO2020029466A1 (zh) 图像处理方法和装置
US10755344B2 (en) System framework processor for channel contacts
WO2021128846A1 (zh) 电子文件的控制方法、装置、计算机设备及存储介质
CN112634017A (zh) 远程开卡激活方法、装置、电子设备及计算机存储介质
US9747175B2 (en) System for aggregation and transformation of real-time data
CN107656959A (zh) 一种留言方法、装置及留言设备
US10664457B2 (en) System for real-time data structuring and storage
CN111985467A (zh) 聊天记录截图处理方法、装置、计算机设备和存储介质
WO2022227191A1 (zh) 非主动活体检测方法、装置、电子设备及存储介质
WO2022105120A1 (zh) 图片文字检测方法、装置、计算机设备及存储介质
US20210103717A1 (en) Method and system for verifying image identification
KR20220150060A (ko) 사용자 정보에 기반하여 업체 매칭 서비스를 제공하고 이에 대한 보안 서비스를 제공하는 플랫폼
CN109933969B (zh) 验证码识别方法、装置、电子设备及可读存储介质
US20220253602A1 (en) Systems and methods for increasing accuracy in categorizing characters in text string

Legal Events

Date Code Title Description
121 Ep: the epo has been informed by wipo that ep was designated in this application

Ref document number: 20873792

Country of ref document: EP

Kind code of ref document: A1

NENP Non-entry into the national phase

Ref country code: DE

32PN Ep: public notification in the ep bulletin as address of the adressee cannot be established

Free format text: NOTING OF LOSS OF RIGHTS PURSUANT TO RULE 112(1) EPC (EPO FORM 1205A DATED 19.08.2022)

122 Ep: pct application non-entry in european phase

Ref document number: 20873792

Country of ref document: EP

Kind code of ref document: A1