WO2020223937A1 - 一种人脸识别方法、人脸识别装置和计算机可读存储介质 - Google Patents

一种人脸识别方法、人脸识别装置和计算机可读存储介质 Download PDF

Info

Publication number
WO2020223937A1
WO2020223937A1 PCT/CN2019/086062 CN2019086062W WO2020223937A1 WO 2020223937 A1 WO2020223937 A1 WO 2020223937A1 CN 2019086062 W CN2019086062 W CN 2019086062W WO 2020223937 A1 WO2020223937 A1 WO 2020223937A1
Authority
WO
WIPO (PCT)
Prior art keywords
feature data
target object
face
face picture
feature
Prior art date
Application number
PCT/CN2019/086062
Other languages
English (en)
French (fr)
Inventor
吴勇辉
范文文
方宏俊
Original Assignee
深圳市汇顶科技股份有限公司
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 深圳市汇顶科技股份有限公司 filed Critical 深圳市汇顶科技股份有限公司
Priority to CN201980000669.1A priority Critical patent/CN110268419A/zh
Priority to PCT/CN2019/086062 priority patent/WO2020223937A1/zh
Publication of WO2020223937A1 publication Critical patent/WO2020223937A1/zh

Links

Images

Classifications

    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F16/00—Information retrieval; Database structures therefor; File system structures therefor
    • G06F16/50—Information retrieval; Database structures therefor; File system structures therefor of still image data
    • G06F16/58—Retrieval characterised by using metadata, e.g. metadata not derived from the content or metadata generated manually
    • G06F16/583—Retrieval characterised by using metadata, e.g. metadata not derived from the content or metadata generated manually using metadata automatically derived from the content
    • 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

Definitions

  • This application relates to the field of face recognition technology, and in particular to a face recognition method, a face recognition device and a computer-readable storage medium.
  • Face recognition is a kind of biometric recognition technology based on human facial feature information.
  • a series of related technologies that use a video camera or camera to collect images or video streams containing faces, and automatically detect and track faces in the images, and then perform facial recognition on the detected faces, usually also called face recognition and facial recognition .
  • 3D face recognition Based on the actual application of face recognition technology, it is necessary to register face first to obtain face image data. In order to increase the accuracy of face recognition, 3D face recognition technology is now used, and 3D face registration is used during registration. In practical applications, 3D face recognition collects and processes face data in real time, and compares it with a private database (ie, face feature database) registered in the module to determine whether the data is the same person’s data in the database. In order to decide whether to authorize actions such as unlocking.
  • a private database ie, face feature database
  • the purpose of some embodiments of this application is to provide a face recognition method, a face recognition device, and a computer-readable storage medium, so that the feature database can be updated with actual changes of the user, and the recognition result is more accurate and reliable.
  • the embodiment of the present application provides a face recognition method, which is applied to a face recognition device, and the method includes: collecting a face picture of a target object, obtaining feature data of the face picture of the target object; comparing the face of the target object The feature data of the picture and the pre-stored feature database; when the difference between the feature data contained in the feature database and the feature data of the target face picture belongs to the first feature data within the first preset range, the comparison result is determined to be recognized as passed; use The feature data of the face picture of the target object updates the first feature data.
  • the embodiment of the present application also provides a face recognition device, including: an acquisition module for acquiring a face picture of a target object; an acquisition module for acquiring feature data of the face picture of the target object; a comparison module , Used to compare the feature data of the target object face picture with a pre-stored feature database; the comparison result confirmation module, used to include the difference between the feature data of the target object face picture and the feature data in the feature database When it belongs to the first feature data within the first preset range, it is determined that the comparison result is recognized as passing; the self-learning module is used to identify the difference between the feature data of the target face image and the first feature data When within the second preset range, update the first feature data by using feature data of the target object face picture, wherein the second preset range is smaller than the first preset range.
  • An embodiment of the present application also provides a face recognition device, including: at least one processor; and, a memory communicatively connected with the at least one processor; wherein the memory stores the memory that can be used by the at least one processor.
  • the executed instructions are executed by the at least one processor, so that the at least one processor can execute the aforementioned face recognition method.
  • the embodiment of the present application also provides a computer-readable storage medium that stores a computer program, and when the computer program is executed by a processor, the above-mentioned face recognition method is implemented.
  • the embodiment of this application uses the feature data obtained in the process of face recognition to perform feature data self-learning, and updates the feature database when the conditions are met, so that the face data in the feature database will change over time.
  • the face changes and updates are more in line with the user’s current face situation, which helps improve the recognition accuracy.
  • the face data collected in the existing recognition process is used for feature data learning, no additional collection steps will be added. Only feature learning steps will not increase the system complexity too much, and the extension time can be almost ignored. , It is easy to maintain the existing recognition speed.
  • the method further includes: determining the difference between the feature data of the target object face picture Whether the difference between the feature data and the first feature data falls within a second preset range; if it does, execute the step of updating the first feature data with the feature data of the target object face picture; wherein ,
  • the second preset range is included in the first preset range.
  • the first preset range is less than or equal to a first threshold
  • the second preset range is greater than or equal to a second threshold and less than or equal to a first threshold, wherein the second threshold is less than the first threshold.
  • a threshold This embodiment can clarify one way of establishing two preset ranges.
  • the feature data in the feature database is stored in the form of a template; the using the feature data of the target object face picture to update the first feature data specifically includes: if the first feature data belongs to the user corresponding to If the number of templates is less than the first preset threshold, the feature data of the target face picture is stored as a new template of the user to which the first feature data belongs; if the number of templates corresponding to the user to which the first feature data belongs is greater than If it is equal to the first preset threshold, the feature data is used to replace one of the templates of the first user, or the feature data is merged into one of the templates of the user to which the first feature data belongs.
  • This embodiment clarifies several methods for updating.
  • a template in the feature database is from a face image collected at one time.
  • the collected face picture of the target object includes: floodlight image and/or structured light image;
  • the acquiring feature data of the face picture of the target object includes: if the collected face picture of the target object If it is a floodlight image, the feature data of the face picture of the target object is acquired according to the floodlight image; if the acquired face picture of the target object is a structured light image, the feature data is acquired according to the structured light image The feature data of the face picture of the target object; if the collected face picture of the target object includes a floodlight image and a structured light image, the target object face picture is acquired according to the floodlight image and the structured light image The characteristic data.
  • This embodiment clarifies several basis for obtaining characteristic data.
  • infrared light sources are used. This embodiment clearly collects light sources, reduces environmental interference, and solves the problem of insufficient lighting at night.
  • the method further includes: performing a process based on the structured light image of the target object. 3D face anti-counterfeiting; after the 3D face anti-counterfeiting is passed, the step of updating the first characteristic data using the characteristic data of the target object face picture is executed.
  • the method before comparing the feature data of the target object's face picture with a pre-stored feature database, the method further includes: performing 3D face anti-counterfeiting according to the structured light image of the target object; and executing after the 3D face anti-counterfeiting passes The step of comparing the acquired feature data with the pre-stored feature database.
  • This embodiment clearly also includes 3D anti-counterfeiting, and defines several different positions of the 3D anti-counterfeiting step.
  • performing 3D face anti-counterfeiting based on the structured light image of the target object specifically includes: performing 3D reconstruction on the structured light image to obtain a reconstructed image; confirming whether it is from a real person according to the reconstructed image; The 3D face security passed.
  • This embodiment clarifies the specific process of 3D anti-counterfeiting.
  • the collecting a face picture of the target object includes: collecting a picture; performing face detection on the picture; when a face is detected, using the picture as the face picture of the target object.
  • This embodiment clarifies the specific process of collecting face pictures.
  • the feature database does not contain the difference between the feature data of the target object face picture and the feature data belonging to the first preset range
  • the step of acquiring the face picture of the target object is executed again.
  • Fig. 1 is a flowchart of a face recognition method according to the first embodiment of the present application
  • Fig. 2 is a flowchart of a face recognition method according to a second embodiment of the present application
  • FIG. 3 is a schematic diagram of the principle in the face recognition method according to the second embodiment of the present application.
  • Fig. 4 is a flowchart of a face recognition method according to a third embodiment of the present application.
  • Fig. 5 is a schematic structural diagram of a face recognition device in a fourth embodiment of the present application.
  • Fig. 6 is a schematic structural diagram of a face recognition device according to a fifth embodiment of the present application.
  • the first embodiment of the present application relates to a face recognition method.
  • This embodiment can be applied to a face recognition device.
  • the user needs to stand in front of the camera module of the smart door lock when opening the smart door lock. After the camera module collects the face image, analyze , Process the picture, extract the feature data of the face, and compare it with the feature data in the database. If the comparison passes, the door lock will be opened.
  • FIG. 1 The specific process of the face recognition method in this embodiment is shown in FIG. 1.
  • Step 101 Collect a face picture of a target object.
  • this step specifically includes: collecting a picture, performing face detection on the picture, and using the picture as the collected face picture of the target object when the face is detected.
  • the camera device of the smart door lock is used to collect face pictures.
  • the camera device starts shooting, and after collecting a picture, it first performs face detection. Specifically, it can be 2D face detection to determine whether there is a human face. If there is a human face, proceed to the following steps. If there is no human face, the shooting may be wrong. Incomplete faces or faces that are not clear enough are captured. Then return to recollect the picture.
  • the deep learning method can be used for 2D face detection, and the detection network for face detection is trained through the deep learning method in advance.
  • the face detection network first detects whether there are faces in the 2D image , If there is a face, draw the position of the face frame, that is, extract the face image, remove the redundant background, etc.
  • a manually labeled face database can be used for training, and the labeled content can include feature contours such as eyes, nose, and mouth, so that the trained network has the ability to detect faces.
  • infrared light sources when collecting floodlight images, can be used to reduce environmental interference and solve problems such as insufficient lighting at night. Even if the user uses a smart door lock at night, facial features can be accurately identified.
  • Step 102 Obtain feature data of the face picture of the target object.
  • the characteristic data are specifically the aspect ratio of the eyes, the distance between the two eyes, the curve length and curvature of the eyebrows, and the aspect ratio of the mouth. , The radian of the chin, etc.
  • the above-mentioned characteristic data can be identified from the flooding image, and the characteristic data value can be obtained through measurement.
  • the feature data can be extracted only for this part of the face image, and there is no need to extract the feature data of the entire collected face image, which not only reduces
  • the processing volume of image data can eliminate background interference and increase the accuracy of extracted feature data.
  • the deep learning method can be used for 2D face recognition.
  • the recognition network for face recognition is trained through the deep learning method in advance, and the obtained face image with the redundant background part removed is sent to the recognition network for facial features Extraction to obtain characteristic data.
  • Step 103 Compare the feature data of the face image of the target object with the pre-stored feature database; if the comparison passes, step 104 is executed; if the comparison fails, return to step 101.
  • the first preset range may be less than the first threshold, and the value of the first threshold may be set by a technician based on experience, such as 30%.
  • the feature database is the facial features of each user obtained through registration, and may include: facial photos of each user, three-dimensional facial data, or features extracted from photos/three-dimensional information.
  • the feature data in the feature database is stored in the form of templates, and the feature data of each template comes from a frame of pictures collected, that is, when registering, every frame of pictures that meets the registration requirements is collected from the The extracted feature data will be saved as a template.
  • One user can correspond to multiple templates. For example, one user can correspond to 8 templates.
  • the comparison compare the acquired feature data with each template one by one.
  • the comparison can be performed by calculating the similarity between the feature vectors according to the feature vectors detected by face recognition, which will not be repeated here.
  • Step 104 Determine whether the difference between the feature data of the target object's face picture and the first feature data is within a second preset range; if yes, go to step 105; if not, end the face recognition process in this embodiment .
  • Step 105 Use the feature data of the face picture of the target object to perform self-learning of the feature database.
  • the feature data of the target face picture is used to update the first feature data, wherein the second preset The set range is included in the first preset range.
  • the second preset range may be greater than or equal to the second threshold and less than or equal to the first threshold, wherein the second threshold is less than the first threshold, for example, the first threshold is 30%, and the second The threshold is 10%.
  • the difference value exceeds 30%, the comparison is considered unsuccessful. Therefore, when the comparison is successful, the difference between the feature data obtained in step 102 and a template in the database must be less than 30%.
  • a lower limit of 10% can be set, that is, when the difference is greater than 10% and less than 30%, the feature data is used to update. If the difference is less than 10%, the feature data is considered too similar, that is, the feature data is not updated.
  • only the first preset range may be set, such as less than or equal to 30%, and after the difference is less than or equal to 30%, the feature data is directly updated.
  • the first preset range can also be set to be less than or equal to 30%. Greater than or equal to 1%, will not be listed here.
  • the specific method of updating can be to add a new template, that is, to store the feature data passed by the comparison as a new template.
  • the feature data fusion method can be used to fuse the newly collected feature data into one of the user’s templates, or the template generated by the newly collected feature data can be used to replace the original template
  • the information that has passed the comparison can be fed back, and the timing of the feedback information can be fed back after the comparison is passed, or after the update is completed, which will not be listed here.
  • this embodiment uses the feature data obtained in the face recognition process to perform feature data self-learning, and updates the feature database when the conditions are met, so that the face data in the feature database will be lost over time.
  • the face changes and updates, which is more in line with the user's current face situation, which helps improve the recognition accuracy.
  • the face data collected in the existing recognition process is used for feature data learning, no additional collection steps will be added. Only feature learning steps will not increase the system complexity too much, and the extension time can be almost ignored. , It is easy to maintain the existing recognition speed.
  • the update frequency is effectively controlled and not too frequent.
  • the second embodiment of the present application relates to a face recognition method.
  • This embodiment is roughly the same as the first embodiment.
  • the main difference is that the feature data in the first embodiment comes from the floodlight image, while the feature data in this embodiment comes from the combination of the floodlight image and the structured light image, because the structured light image has 3D information, so you can get richer 3D information and improve the accuracy of recognition.
  • Step 201 Collect floodlight images and structured light images of the target object.
  • a structured light image in addition to the floodlight image corresponding to the human face, a structured light image can also be acquired.
  • the camera device in this embodiment may be a 3D module.
  • structured light can be projected onto the face through a projector built into the 3D module, and then collected by the camera in the 3D module. Obtain the corresponding structured light image.
  • the set of projected light rays with a known spatial direction is called structured light, such as speckle
  • the image obtained by projecting structured light is called structured light image.
  • the structured light image can also be coded fringes, sinusoidal fringes, and so on.
  • Step 202 Obtain feature data of the floodlight image and structured light image of the target object.
  • the feature data extracted from the floodlight image of the human face in this step is similar to the first embodiment, and will not be repeated here.
  • 3D reconstruction is performed on the structured light image, and the feature data is extracted from the reconstructed image.
  • the data form of the reconstructed map obtained through 3D reconstruction may include a depth map or a three-dimensional point cloud, and in one example, it may also be a combination of the two.
  • the feature data of the reconstructed image is calculated to obtain the feature data of the face.
  • Step 203 Compare the feature data of the face image of the target object with the pre-stored feature database; if the comparison passes, step 204 is executed; if the comparison fails, return to step 201.
  • the comparison not only the 2D feature data from the floodlight image is used for comparison, but also the 3D feature data from the structured light image can be used for comparison.
  • the 2D feature data can be compared first, and the 3D feature data can be compared after the 2D feature data has passed the comparison.
  • the floodlight image can be acquired first, and the 2D feature data can be acquired and compared.
  • the structured light image can be collected, and then the 3D feature data can be acquired and compared with the 3D feature data.
  • the order of comparison is not limited here.
  • Step 204 to step 205 are similar to step 104 to step 105 in the first embodiment, and will not be repeated here.
  • FIG. 3 The structure and working principle of this embodiment can be shown in Figure 3, where the person 4 sends the collected information to the controller (or processor) 2 through the human-computer interaction device 3 (such as a touch screen), and the controller can be an AP ( Application Processor), the controller 2 sends a collection command to the camera module 1. After the camera module 1 receives the command, it projects structured light onto the face of the person 4. After reflection, the camera module 1 collects the picture and sends it to the controller 2 performs processing, and the controller 2 is specifically used to implement functions such as face detection, recognition, 3D reconstruction, and data fusion.
  • the controller or processor 2 through the human-computer interaction device 3 (such as a touch screen), and the controller can be an AP ( Application Processor)
  • the controller 2 sends a collection command to the camera module 1. After the camera module 1 receives the command, it projects structured light onto the face of the person 4. After reflection, the camera module 1 collects the picture and sends it to the controller 2 performs processing, and the controller 2 is specifically used to implement functions such
  • the basis for defining characteristic data in this embodiment can be a combination of floodlight image and structured light image. Two-dimensional information is acquired through floodlight image, and three-dimensional information is acquired through structured light image. Therefore, the combination of the two makes the information more abundant. Make the recognition result more accurate and credible.
  • the feature data obtained through the floodlight image and the structured light image are taken as an example in this embodiment, in practical applications, the feature data can be obtained only through the structured light image, which will not be repeated here.
  • the third embodiment of the present application relates to a face recognition method.
  • This embodiment is a further improvement over the second embodiment.
  • the main improvement lies in the addition of a new 3D anti-counterfeiting process using structured light images to avoid as much as possible the identification system from being attacked by images, videos or 3D avatars, and further guarantee The face recognition method is safe and reliable.
  • Step 401 and step 402 are similar to step 201 and step 202 in the second embodiment, and will not be repeated here.
  • Step 403 Compare the feature data of the face image of the target object with the pre-stored feature database; if the comparison passes, step 404 is executed; if the comparison fails, step 405 is executed.
  • Step 404 check whether the 3D anti-counterfeiting is passed; if yes, continue to perform step 406; if not, perform step 405.
  • 3D anti-counterfeiting is mainly used to detect whether the source of the collected image is a real person. If the source is a photo, image or 3D model, it should be excluded as much as possible, otherwise it will affect the credibility of the recognition result. More specifically, 3D anti-counterfeiting can be performed through structured light graphics. The specific steps include: 3D reconstruction of the structured light image to obtain a reconstructed image; confirm whether it is from a real person according to the reconstructed image; if it is confirmed to be from a real person, determine the 3D face anti-counterfeiting by.
  • the 3D reconstruction process of the structured light image can be specifically as follows: calculate the three-dimensional coordinates of the object corresponding to the structured light image according to the parameters of the imaging device, and the parameters of the imaging device include: internal parameters (such as camera focal length, principal point position, etc.) and External parameters (rotation and translation relationship between camera and projector). More specifically, the system pre-stores the pre-stored image of the camera device (which can be a speckle image), matches the collected image with the pre-stored image, obtains the parallax, and calculates the three-dimensional coordinates of the face according to the parallax, internal parameters, and external parameters. . Afterwards, the feature data of the face is extracted according to the calculated three-dimensional coordinates.
  • the pre-stored image of the camera device which can be a speckle image
  • the system pre-stores the pre-stored image of the camera device (which can be a speckle image)
  • matches the collected image with the pre-stored image obtains the parallax
  • this step may not be repeated.
  • the reconstructed image (3D image) generated by the conversion can be used to determine whether the collected face is a real human face or a photo. Since the photo is a two-dimensional object, if you use a photo As a collection object, it is impossible to obtain a 3D image with a normal stereoscopic effect. Therefore, in one example, it can be determined whether the collection object is a real person or a photo based on the 3D image generated by the conversion. If it is confirmed that it is from a real person, the above 3D face anti-counterfeiting is determined by.
  • 3D anti-counterfeiting can also be performed in other ways, which will not be listed here.
  • this embodiment performs 3D face anti-counterfeiting based on the structured light image after the comparison is passed, and then enters self-learning when the anti-counterfeiting passes.
  • 3D face anti-counterfeiting can also be performed first, and after the anti-counterfeiting is passed, the characteristic data is compared.
  • the execution position of 3D anti-counterfeiting is not limited here.
  • Step 405 Detect whether the number of retries exceeds the limit; if yes, end the face recognition method in this embodiment; if not, return to step 401.
  • a second preset threshold can be set for the total number of collections. This step specifically compares the total number of collections with the second preset threshold. If the total number of collections is less than the second preset threshold, then it is considered that there is no overrun, and you can again Collect and continue to retry, but if the total number of episodes is greater than or equal to the second preset threshold, then it is considered to have exceeded the limit, and there is no need to try again, it is considered that the recognition has failed, and the process is exited.
  • Steps 406 to 407 in this embodiment are similar to steps 204 to 205 in the second embodiment, and will not be repeated here.
  • 3D anti-counterfeiting is added to the recognition process, and several different positions of the 3D anti-counterfeiting step are limited, as far as possible to avoid the recognition system being attacked by images, videos or 3D avatars, and further ensure the safety and reliability of the face recognition method .
  • a second preset threshold can be added in this step to monitor the number of retries when an error occurs. When the number of retries does not exceed the limit, the face image is re-collected. If the number of retries exceeds the limit, the recognition is considered as a failure.
  • the fourth embodiment of the present application relates to a face recognition device.
  • FIG. 5 The schematic diagram of the device in this embodiment is shown in FIG. 5, and specifically includes:
  • the collection module is used to collect the face picture of the target object
  • An acquisition module for acquiring feature data of the face picture of the target object
  • the comparison module is used to compare the feature data of the target face picture with the pre-stored feature database
  • the comparison result confirmation module is used for determining that the comparison result is recognized as passing when the difference between the feature data of the target face image and the feature data included in the feature database belongs to the first feature data within the first preset range;
  • the self-learning module is used to update the first feature data with the feature data of the face picture of the target object.
  • the method further includes: a processing module, configured to confirm that the difference between the feature data contained in the feature database and the feature data of the target face picture belongs to the first feature data within the first preset range after the comparison result confirmation module confirms , Determining whether the difference between the feature data of the target object face picture and the first feature data falls within a second preset range.
  • a processing module configured to confirm that the difference between the feature data contained in the feature database and the feature data of the target face picture belongs to the first feature data within the first preset range after the comparison result confirmation module confirms , Determining whether the difference between the feature data of the target object face picture and the first feature data falls within a second preset range.
  • the self-learning module is specifically configured to use the feature data of the target face picture to update when the processing module determines that the difference between the feature data of the target face picture and the first feature data is within the second preset range The first feature data.
  • the second preset range is included in the first preset range.
  • the first preset range is less than or equal to the first threshold
  • the second preset range is greater than or equal to the second threshold and less than or equal to the first threshold, wherein the second threshold is less than the The first threshold.
  • the feature data in the feature database is stored in the form of a template; the self-learning module specifically includes:
  • the first update submodule is configured to store the feature data of the face picture of the target object as the user to which the first feature data belongs when the number of templates corresponding to the user to which the first feature data belongs is less than a first preset threshold New template.
  • the second update submodule is configured to replace one of the templates of the first user with the characteristic data when the number of templates corresponding to the user to which the first characteristic data belongs is greater than or equal to the first preset threshold, or The feature data is fused into one of the templates of the user to which the first feature data belongs.
  • a template in the feature database comes from a face image collected at one time.
  • the collected face image of the target object includes: floodlight image and/or structured light image; correspondingly, the acquisition module specifically includes:
  • the first acquisition sub-module is configured to acquire the characteristic data of the face picture of the target object according to the flood light image when the collected face picture of the target object is a flood light image.
  • the second acquisition submodule is configured to acquire the characteristic data of the target object's face picture according to the structured light image when the collected face picture of the target object is a structured light image.
  • the third acquisition sub-module is used to acquire the information of the target object’s face picture according to the flood light image and the structured light image when the collected face picture of the target object includes flood light image and structured light image The characteristic data.
  • an infrared light source is used when the flood image of the face picture of the target object is collected.
  • the face recognition device further includes a 3D anti-counterfeiting module, which is used to update the first image by using the feature data of the target object's face picture when the collected face picture of the target object includes a structured light image. Before the feature data, 3D face anti-counterfeiting is performed according to the structured light image of the target object.
  • the self-learning module is used to update the first feature data using the feature data of the target face picture after the 3D anti-counterfeiting module passes the anti-counterfeiting.
  • the face recognition device further includes a 3D anti-counterfeiting module, which is used to perform a 3D image based on the structured light image of the target object before comparing the feature data of the target object’s face picture with a pre-stored feature database. Face anti-counterfeiting.
  • the comparison module is used to compare the acquired feature data with the pre-stored feature database after the 3D anti-counterfeiting module passes anti-counterfeiting.
  • the 3D anti-counterfeiting module specifically includes:
  • the reconstruction sub-module is used to perform 3D reconstruction on the structured light image to obtain a reconstructed image.
  • the confirmation sub-module is used to confirm whether it is from a real person according to the reconstructed image.
  • the anti-counterfeiting result confirmation sub-module is used to confirm that the 3D face has passed the anti-counterfeiting when confirming that it is from a real person.
  • the acquisition module includes:
  • Collection sub-module used to collect pictures
  • the detection sub-module is used to perform face detection on the picture
  • the processing sub-module is configured to use the picture as the face picture of the target object when a human face is detected.
  • the face recognition device further includes: a number of acquisitions judging module, which is used for not including in the feature database the difference between the feature data of the target object face picture and the first feature within the first preset range Data, determine whether the total number of acquisitions exceeds the second preset threshold, and trigger the acquisition module when the total number of acquisitions does not exceed the second preset threshold.
  • a number of acquisitions judging module which is used for not including in the feature database the difference between the feature data of the target object face picture and the first feature within the first preset range Data, determine whether the total number of acquisitions exceeds the second preset threshold, and trigger the acquisition module when the total number of acquisitions does not exceed the second preset threshold.
  • this embodiment is an example of a device corresponding to the first embodiment, and this embodiment can be implemented in cooperation with the first embodiment.
  • the related technical details mentioned in the first embodiment are still valid in this embodiment, and in order to reduce repetition, they will not be repeated here.
  • the related technical details mentioned in this embodiment can also be applied in the first embodiment.
  • modules involved in this embodiment are all logical modules.
  • a logical unit can be a physical unit, a part of a physical unit, or multiple physical units. The combination of units is realized.
  • this embodiment does not introduce a unit that is not closely related to solving the technical problem proposed by the present invention, but this does not indicate that there are no other units in this embodiment.
  • the fifth embodiment of the present invention relates to a face recognition device, as shown in FIG. 6, including:
  • At least one processor and a memory connected in communication with the at least one processor; wherein the memory stores instructions executable by at least one processor, and the instructions are executed by at least one processor, so that at least one processor can execute as described above Any one of the face recognition methods in the first embodiment to the third embodiment.
  • the memory and the processor are connected in a bus manner, and the bus may include any number of interconnected buses and bridges, and the bus connects one or more processors and various circuits of the memory together.
  • the bus can also connect various other circuits such as peripheral devices, voltage regulators, power management circuits, etc., which are all well-known in the art, and therefore, no further description will be given herein.
  • the bus interface provides an interface between the bus and the transceiver.
  • the transceiver may be one element or multiple elements, such as multiple receivers and transmitters, providing a unit for communicating with various other devices on the transmission medium.
  • the data processed by the processor is transmitted on the wireless medium through the antenna, and further, the antenna also receives the data and transmits the data to the processor.
  • the processor is responsible for managing the bus and general processing, and can also provide various functions, including timing, peripheral interfaces, voltage regulation, power management, and other control functions.
  • the memory can be used to store data used by the processor when performing operations.
  • the seventh embodiment of the present invention relates to a computer-readable storage medium storing a computer program.
  • the computer program is executed by the processor, the above method embodiment is realized.
  • the program is stored in a storage medium and includes several instructions to enable a device ( It may be a single-chip microcomputer, a chip, etc.) or a processor (processor) to execute all or part of the steps of the methods in the embodiments of the present application.
  • the aforementioned storage media include: U disk, mobile hard disk, read-only memory (ROM, Read-Only Memory), random access memory (RAM, Random Access Memory), magnetic disk or optical disk and other media that can store program code .

Abstract

一种人脸识别方法、人脸识别装置和计算机可读存储介质。该人脸识别方法，应用于人脸识别装置，包括：采集目标对象的人脸图片(101)；获取目标对象的人脸图片的特征数据(102)；比对目标对象人脸图片的特征数据和预存的特征数据库(103)；在特征数据库中包含与目标对象人脸图片的特征数据的差异性属于第一预设范围内的第一特征数据时，确定比对结果为识别通过；利用目标对象人脸图片的特征数据更新第一特征数据(105)。采用上述方案使得特征数据库可以随用户实际变化而更新，识别结果更为准确可靠。

Description

一种人脸识别方法、人脸识别装置和计算机可读存储介质 技术领域
本申请涉及人脸识别技术领域，特别涉及一种人脸识别方法、人脸识别装置和计算机可读存储介质。
背景技术
人脸识别，是基于人的脸部特征信息进行身份识别的一种生物识别技术。用摄像机或摄像头采集含有人脸的图像或视频流，并自动在图像中检测和跟踪人脸，进而对检测到的人脸进行脸部识别的一系列相关技术，通常也叫做人像识别、面部识别。
基于人脸识别技术的实际应用，需要先进行人脸注册，获取人脸图片数据，为增加人脸识别的准确性，现采用3D人脸识别技术，注册时对应地采用3D人脸注册。实际应用中，3D人脸识别通过实时采集人脸数据并处理，与通过注册在模块内部的私有数据库(即人脸特征数据库)进行比对，判定该数据是否与数据库内是同一人的数据，从而决定是否授权通过解锁等行为。
发明人发现现有技术至少存在以下问题：在现有人脸识别过程中，使用的数据库为注册时收集的数据，识别时仍会遇到虽然是本人，但无法识别的情况，如减肥或增肥后，又如随生长变化较多的小孩子。
发明内容
本申请部分实施例的目的在于提供一种人脸识别方法、人脸识别装置和计算机可读存储介质，使得特征数据库可以随用户实际变化而更新，识别结果更为准确可靠。
本申请实施例提供了一种人脸识别方法，应用于人脸识别装置，所述方法包括：采集目标对象的人脸图片，获取目标对象的人脸图片的特征数据；比对目标对象人脸图片的特征数据和预存的特征数据库；在特征数据库中包含与目标对象人脸图片的特征数据的差异性属于第一预设范围内的第一特征数据时，确定比对结果为识别通过；利用目标对象人脸图片的特征数据更新第一特征数据。
本申请实施例还提供了一种人脸识别装置，包括：采集模块，用于采集目标对象的人脸图片；获取模块，用于获取所述目标对象的人脸图片的特征数据；比对模块，用于比对所述目标对象人脸图片的特征数据和预存的特征数据库；比对结果确认模块，用于在所述特征数据库中包含与所述目标对象人脸图片的特征数据的差异性属于第一预设范围内的第一特征数据时，确定比对结果为识别通过；自学习模块，用于在所述目标对象人脸图片的特征数据与所述第一特征数据的差异性属于第二预设范围内时，利用所述目标对象人脸图片的特征数据更新所述第一特征数据，其中，所述第二预设范围小于所述第一预设范围。
本申请实施例还提供了一种人脸识别装置，包括：至少一个处理器；以及，与所述至少一个处理器通信连接的存储器；其中，所述存储器存储有可被所述至少一个处理器执行的指令，所述指令被所述至少一个处理器执行，以使 所述至少一个处理器能够执行如上述的人脸识别方法。
本申请实施例还提供了一种计算机可读存储介质，存储有计算机程序，所述计算机程序被处理器执行时实现如上述的人脸识别方法。
本申请实施例相对于现有技术而言，利用人脸识别过程中获得的特征数据，进行特征数据自学习，在符合条件时更新特征数据库，使得特征数据库中的人脸数据会随时间推移中人脸变化而更新，更符合用户当下的人脸情况，有利于提高识别准确性。同时，由于特征数据学习时利用现有识别过程中采集到的人脸数据，也就不会增加额外的采集步骤，仅增加特征学习的步骤，不会过于增加系统复杂度，延长时间几乎可以忽略，便于保持现有的识别速度。
例如，在判定所述特征数据库中包含与所述目标对象人脸图片的特征数据的差异性属于第一预设范围内的第一特征数据之后，还包括：判断所述目标对象人脸图片的特征数据与所述第一特征数据的差异性是否属于第二预设范围内；若属于，则执行所述利用所述目标对象人脸图片的特征数据更新所述第一特征数据的步骤；其中，所述第二预设范围包含于所述第一预设范围。本实施例明确可以设立两个范围，分别用于确定识别是否通过和是否需要更新，且在更小范围内才进行更新，保证更新效果的同时，尽量减少更新次数。
例子，所述第一预设范围为小于或等于第一阈值，所述第二预设范围为大于或等于第二阈值且小于或等于第一阈值，其中，所述第二阈值小于所述第一阈值。本实施例可以明确两个预设范围的一种设立方式。
例如，所述特征数据库中的特征数据以模板的形式存储；所述利用所述目标对象人脸图片的特征数据更新所述第一特征数据，具体包括：若所述第一特征数据所属用户对应的模板数小于第一预设阈值，则将所述目标对象人脸图 片的特征数据存为所述第一特征数据所属用户的新模板；若所述第一特征数据所属用户对应的模板数大于或等于所述第一预设阈值，则利用所述特征数据替换所述第一用户的模板之一，或将所述特征数据融合入所述第一特征数据所属用户的模板之一。本实施例明确更新的几种方法。
例如，所述特征数据库中的一个模板来自一次采集到的人脸图片。
例如，所述采集的目标对象的人脸图片包括：泛光图像和/或结构光图像；所述获取所述目标对象的人脸图片的特征数据，包括：若采集的目标对象的人脸图片为泛光图像，则根据所述泛光图像获取所述目标对象人脸图片的所述特征数据；若采集的目标对象的人脸图片为结构光图像，则根据所述结构光图像获取所述目标对象人脸图片的所述特征数据；若采集的目标对象的人脸图片包括泛光图像和结构光图像，则根据所述泛光图像和所述结构光图像获取所述目标对象人脸图片的所述特征数据。本实施例明确获得特征数据的几种依据。
例如，在采集人脸图片的泛光图像时，采用红外光源。本实施例明确采集光源，减少环境干扰，解决夜晚照明不足的问题。
例如，若采集的目标对象的人脸图片包括结构光图像，则所述利用所述目标对象人脸图片的特征数据更新所述第一特征数据之前，还包括：根据目标对象的结构光图像进行3D人脸防伪；在3D人脸防伪通过后，执行所述利用所述目标对象人脸图片的特征数据更新所述第一特征数据的步骤。
例如，所述比对所述目标对象人脸图片的特征数据和预存的特征数据库之前，还包括：根据所述目标对象的结构光图像进行3D人脸防伪；在3D人脸防伪通过后，执行所述比对所获取到的特征数据和预存的特征数据库的步骤。本实施例明确还包括3D防伪，且限定了3D防伪步骤的几种不同位置。
例如，所述根据所述目标对象的结构光图像进行3D人脸防伪，具体包括：对结构光图进行3D重建，获得重建图；根据重建图确认是否来自真人；若确认来自真人，则确定所述3D人脸防伪通过。本实施例明确3D防伪的具体过程。
例如，所述采集目标对象的人脸图片，包括：采集图片；对所述图片进行人脸检测；在检测到人脸时，将所述图片作为所述目标对象的人脸图片。本实施例明确采集人脸图片的具体过程。
例如，所述比对所获取到的特征数据和预存的特征数据库之后，还包括：在所述特征数据库中未包含与所述目标对象人脸图片的特征数据的差异性属于第一预设范围内的第一特征数据，且总采集次数不超过第二预设阈值时，则重新执行所述采集目标对象的人脸图片的步骤。本实施例明确错误重试的次数有上限。
附图说明
一个或多个实施例通过与之对应的附图中的图片进行示例性说明，这些示例性说明并不构成对实施例的限定，附图中具有相同参考数字标号的元件表示为类似的元件，除非有特别申明，附图中的图不构成比例限制。
图1是根据本申请第一实施例中的人脸识别方法的流程图；
图2是根据本申请第二实施例中的人脸识别方法的流程图；
图3是根据本申请第二实施例中的人脸识别方法中的原理示意图；
图4是根据本申请第三实施例中的人脸识别方法的流程图；
图5是根据本申请第四实施例中的人脸识别装置的结构示意图；
图6是根据本申请第五实施例中的人脸识别装置的结构示意图。
具体实施方式
为了使本申请的目的、技术方案及优点更加清楚明白，以下结合附图及实施例，对本申请部分实施例进行进一步详细说明。应当理解，此处所描述的具体实施例仅仅用以解释本申请，并不用于限定本申请。
本申请第一实施例涉及一种人脸识别方法。
本实施方式可以应用于一种人脸识别装置，以智能门锁为例，用户在开启智能门锁时，需要站立于智能门锁的摄像模组前，摄像模组采集人脸图片后，解析、处理图片，提取人脸的特征数据，并和数据库中的特征数据进行比对，如果比对通过，则开启门锁。
本实施例中人脸识别方法的具体流程如图1所示。
步骤101，采集目标对象的人脸图片。
具体的说，本步骤具体包括：采集图片，对图片进行人脸检测，在检测到人脸时，将图片作为采集到的目标对象的人脸图片。以智能门锁为例进行说明，本实施方式中通过智能门锁自带的摄像设备采集人脸图片，在一个例子中，摄像设备启动拍摄，采集到一张图片后，先进行人脸检测，具体可以为2D人脸检测，判断是否存在人脸，如果存在人脸，则进行后续步骤，如果不存在人脸，可能拍摄出错，拍摄到不完全的人脸或拍摄到不够清晰的人脸，则返回重新采集图片。
在一个例子中，可以采用深度学习法进行2D人脸检测，预先通过深度学习方法训练用于人脸检测的检测网络，在实际应用时，首先通过人脸检测网 络检测2D图片中是否存在人脸，若存在人脸，则画出人脸框的位置，也就是提取出人脸图像，去除多余的背景等。在训练人脸检测网络时，可以采用人工标注的人脸数据库进行训练，标注的内容可以包含眼睛、鼻子、嘴巴等特征轮廓，从而使得经过训练的网络具备检测人脸的能力。
在一个例子中，在采集泛光图像时，可以采用红外光源，减少环境干扰，解决夜晚照明不足等问题，即使用户在夜晚使用智能门锁，也可以准确地被识别出人脸特征。
步骤102，获取目标对象的人脸图片的特征数据。
具体的说，以采集的目标对象的人脸图片为泛光图像为例进行说明，特征数据具体如眼睛的长宽比、两个眼睛的间距、眉毛的曲线长度和弧度、嘴巴的长宽比、下巴的弧度等，上述特征数据可以从泛光图像中识别出，并通过测量获得特征数据值。
在一个例子中，在步骤101确定出去除多余背景部分的人脸图像后，可以仅对这部分人脸图像进行特征数据的提取，无需对采集到的整个人脸图片进行特征数据提取，不仅减少图像数据的处理量，更能排除背景干扰，增加提取出的特征数据的准确性。
在一个例子中，可以利用深度学习法进行2D人脸识别，预先通过深度学习方法训练用于人脸识别的识别网络，将获取的去除多余背景部分的人脸图像送至识别网络进行人脸特征的提取，获得特征数据。
步骤103，比对目标对象人脸图片的特征数据和预存的特征数据库；若比对通过，则执行步骤104；若比对不通过，则返回执行步骤101。
具体的说，在特征数据库中包含与目标对象人脸图片的特征数据的差异 性属于第一预设范围内的第一特征数据时，确定比对结果为识别通过，即比对通过。其中，第一预设范围可以是小于第一阈值，第一阈值的值可以由技术人员根据经验设置，如30％。
具体的说，特征数据库为通过注册时得到各用户的人脸特征，可以包括：各用户的面部照片、面部三维数据或根据照片/三维信息提取的特征。在一个例子中，特征数据库中的特征数据以模板的形式存储，每个模板的特征数据来自采集到的一帧图片，也就是说，注册时，每采集到一帧符合注册要求的图片，从中提取的特征数据就会存为一个模板。一个用户可以对应有多个模板，如可以设置一个用户对应有8个模板。
在比对时，将获取的特征数据和各个模板进行逐一对比，在与一帧模板比对时，确定获取的特征数据和各个模板的差异性，如果差异性较大，如差异性大于等于上限值A，则认为比对不成功，继续下一帧比对，如果任意一帧比对成功即认为属于同一人。
继续说明，在比对时，特征数据库中如果没有一帧可以比对成功，那么就认为采集到的特征数据不属于该特征数据库，即比对不通过。
在一个例子中，在比对过程中可以根据人脸识别检测的特征向量，通过计算特征向量之间的相似度进行比对，在此不再赘述。
步骤104，判断目标对象人脸图片的特征数据与第一特征数据的差异性是否属于第二预设范围内；若是，则执行步骤105；若否，则结束本实施方式中的人脸识别流程。
步骤105，利用目标对象人脸图片的特征数据进行特征数据库的自学习。
具体的说，在特征数据与特征数据库中第一用户的特征数据的差异性属 于第二预设范围内时，利用目标对象人脸图片的特征数据更新上述第一特征数据，其中，第二预设范围包含于第一预设范围。
在一个例子中，第二预设范围可以为大于或等于第二阈值且小于或等于第一阈值，其中，所述第二阈值小于所述第一阈值，如第一阈值为30％，第二阈值为10％。由于差异值超过30％时，则认为比对不成功，所以在比对成功时，步骤102中获取的特征数据必然与数据库中某一模板的差异性小于30％，同时为了避免特征数据更新过于频繁，可以设置下限10％，也就是说，在差异性大于10％且小于30％时，才利用该特征数据进行更新，如果差异性小于10％，则认为特征数据过于相似，即不更新。
在一个例子中，可以仅设置第一预设范围，如小于或等于30％，在差异性满足小于或等于30％后，直接进行特征数据的更新。
在另一个例子中，第一预设范围也可以设置为小于或等于30％。大于或等于1％，在此不再一一列举。
在一个例子中，更新的具体方法可以是新增模板，也就是说，将比对通过的特征数据储存为新模板，模板越多，涵盖的特征数据必然更丰富，但比对耗时也将更大。进一步说，可以为用户设置模板数上限，如设置某个用户最多可以有25个模板，在该用户已有模板不足25个时，可以直接通过新增模板保存新收集到的特征数据，如果该用户已有模板大于或等于25时，可以采用特征数据融合的方式，将新收集到的特征数据融合入该用户的模板之一，也可以利用新收集到的特征数据生成的模板替换原有模板之一，替换模板时可以有多种机制，一种可以是替换生成时间最早的模板，一种可以是替换差异性最大的模板，实际应用中也可以设置其他替换机制，在此不再一一列举。
在本实施方式中，先判定是否比对通过，在比对通过后再进行更新。在一个例子中，比对通过可以反馈识别通过的信息，反馈信息的时机可以在比对通过后反馈，也可以在更新完成后反馈，在此不再一一列举。
本实施例相对于现有技术而言，利用人脸识别过程中获得的特征数据，进行特征数据自学习，在符合条件时更新特征数据库，使得特征数据库中的人脸数据会随时间推移中人脸变化而更新，更符合用户当下的人脸情况，有利于提高识别准确性。同时，由于特征数据学习时利用现有识别过程中采集到的人脸数据，也就不会增加额外的采集步骤，仅增加特征学习的步骤，不会过于增加系统复杂度，延长时间几乎可以忽略，便于保持现有的识别速度。另外，由于需要新特征数据和数据库中的特征数据的差异性在某一区间内才更新，所以更新频率得到有效控制，不会过于频繁。
本申请第二实施例涉及一种人脸识别方法。本实施方式与第一实施方式大致相同，主要区别在于：第一实施方式中特征数据来自泛光图像，而本实施方式中特征数据来自泛光图像和结构光图像的组合，由于结构光图像具有3D信息，所以可以获得更为丰富的3D信息，提高识别的准确度。
本实施方式中人脸识别方法的流程图如图2所示，具体如下：
步骤201，采集目标对象的泛光图像和结构光图像。
具体的说，本步骤中除了采集对应人脸的泛光图像，还可以采集结构光图像。在一个例子中，本实施方式中的摄像设备可以是3D模组，具体可以通过3D模组中内置的投射器，将结构光投射到人脸上，再通过3D模组中的摄像头进行采集，获得对应的结构光图像。其中，已知空间方向的投影光线的集合 称为结构光，如散斑，通过投射结构光获得的图像称为结构光图像。在一个例子中，结构光图像还可以是编码条纹、正弦条纹等。
步骤202，获取目标对象的泛光图像和结构光图像的特征数据。
具体的说，本步骤中对人脸的泛光图像提取特征数据和第一实施方式相类似，在此不再赘述。
而本步骤中对人脸的结构光图像提取特征数据时，对结构光图像进行3D重建，从重建图中提取特征数据。具体的说，通过3D重建所获得的重建图的数据形式可以包括：深度图或三维点云，在一个例子中，也可以是两者的组合。之后对重建图进行特征数据的计算，从而获得人脸的特征数据。
步骤203，比对目标对象人脸图片的特征数据和预存的特征数据库；若比对通过，则执行步骤204；若比对不通过，则返回执行步骤201。
具体的说，在比对时，不仅利用来自泛光图像的2D特征数据比对，还可以利用来自结构光图像的3D特征数据比对。
在一个例子中，可以先进行2D特征数据的比对，在2D特征数据比对通过后再进行3D特征数据的比对。此外，综合上述步骤201至步骤203，可以先采集泛光图像，进行2D特征数据获取并比对，在比对通过后，再采集结构光图像，继而进行3D特征数据获取并比对3D特征数据，同时，在比对多种类型的特征数据时，比对先后顺序在此不做限定。
步骤204至步骤205与第一实施方式中的步骤104至步骤105相类似，在此不再赘述。
本实施方式的结构和工作原理可以如图3所示，其中的人4通过人机交互设备3(例如触摸屏)向控制器(或处理器)2发采集信息，其中的控制器可 以是AP(Application Processor的简称)，控制器2向摄像模组1发送采集指令，摄像模组1接受命令后，投射结构光至人4的面部，经反射后由摄像模组1采集图片，送至控制器2进行处理，控制器2具体用于实现人脸检测、识别、3D重建和数据融合等功能。
可见，本实施方式中明确特征数据的依据可以是泛光图像和结构光图像的组合，通过泛光图像获取二维信息，通过结构光图像获取三维信息，所以两者结合，信息更为丰富，使得识别结果更为准确可信。
本实施方式中虽然以通过泛光图像和结构光图像共同获得特征数据为例，在实际应用中，特征数据可以仅通过结构光图像获取，在此不再赘述。
本申请第三实施例涉及一种人脸识别方法。本实施方式是在第二实施方式上做了进一步改进，主要改进之处在于：新增利用结构光图像进行3D防伪的过程，尽可能避免识别系统被图像、视频或3D头像等攻击，进一步保障人脸识别方法的安全可靠。
本实施方式中人脸识别方法的流程图如图4所示，具体如下：
步骤401和步骤402与第二实施方式中的步骤201和步骤202相类似，在此不再赘述。
步骤403，比对目标对象人脸图片的特征数据和预存的特征数据库；若比对通过，则执行步骤404；若比对不通过，则执行步骤405。
步骤404，检测3D防伪是否通过；若是，则继续执行步骤406；若否，则执行步骤405。
具体的说，3D防伪主要用于检测采集图片的来源是否为真人，如果来源 为照片、影像或3D模型等，要尽量排除，否则将影响识别结果的可信度。更具体的说，3D防伪可以具体通过结构光图形进行，具体步骤包括：对结构光图进行3D重建，获得重建图；根据重建图确认是否来自真人；若确认来自真人，则确定3D人脸防伪通过。
其中，对结构光图像进行3D重建的过程可以具体如下：根据摄像设备的参数计算出结构光图像所对应物体的三维坐标，摄像设备的参数包括：内参(如相机焦距、主点位置等)和外参(摄像头和投射器间的旋转和平移关系)。更具体的说，系统预存了摄像设备的预存图(可以是散斑图)，将采集到的图片和预存图进行匹配，获得视差，根据视差、内参、外参共同计算出人脸的三维坐标。之后，根据计算出的三维坐标提取出人脸的特征数据。
在一个例子中，如果对结构光图已经进行了3D重建，本步骤就可以不重复执行。
更具体的说，通过摄像设备采集结构光图像时，可以根据转换生成的重建图(3D图像)来判断被采集的人脸是真人人脸还是照片，由于照片是二维物体，所以如果用照片作为采集对象，就无法获得有正常立体效果的3D图像，所以在一个例子中，可以根据转换生成的3D图像就可以确定采集对象是真人还是照片，如果确认来自真人，则确定上述3D人脸防伪通过。
此外，在实际应用中，还可以通过其他方式进行3D防伪，在此不再一一列举。
继续说明，本实施方式在比对通过之后，根据结构光图像进行3D人脸防伪，之后在防伪通过时，进入自学习。实际应用中，也可以先进行3D人脸防伪，在防伪通过后，对特征数据进行比对，在此不对3D防伪的执行位置进 行限定。
步骤405，检测重试次数是否超限；若是，则结束本实施方式中的人脸识别方法，若否，则返回执行步骤401。
具体的说，可以为总采集次数设置第二预设阈值，本步骤具体比较总采集次数和第二预设阈值，如果总采集次数小于第二预设阈值，那么就认为没有超限，可以再次采集，继续重试，但如果总集次数大于或等于第二预设阈值，那么就认为已经超限，无需再试，认为识别失败，退出流程。
本实施方式中的步骤406至步骤407与第二实施方式中的步骤204至步骤205相类似，在此不再赘述。
可见，本实施方式中在识别过程中加入3D防伪，且限定了3D防伪步骤的几种不同位置，尽可能避免识别系统被图像、视频或3D头像等攻击，进一步保障人脸识别方法的安全可靠。另外，本步骤可以增设第二预设阈值，用于监控出错时的重试次数，在重试次数没有超限时，重新采集人脸图片，如果重试次数超限，那么就认为识别失败。
本申请第四实施例涉及一种人脸识别装置。
本实施方式中的装置示意图如图5所示，具体包括：
采集模块，用于采集目标对象的人脸图片；
获取模块，用于获取目标对象的人脸图片的特征数据；
比对模块，用于比对目标对象人脸图片的特征数据和预存的特征数据库；
比对结果确认模块，用于在特征数据库中包含与目标对象人脸图片的特征数据的差异性属于第一预设范围内的第一特征数据时，确定比对结果为识别通过；
自学习模块，用于利用目标对象人脸图片的特征数据更新第一特征数据。
在一个例子中，还包括：处理模块，用于在比对结果确认模块确认在特征数据库中包含与目标对象人脸图片的特征数据的差异性属于第一预设范围内的第一特征数据之后，判断所述目标对象人脸图片的特征数据与所述第一特征数据的差异性是否属于第二预设范围内。
对应的，自学习模块具体用于在处理模块判定目标对象人脸图片的特征数据与所述第一特征数据的差异性属于第二预设范围内时，利用目标对象人脸图片的特征数据更新第一特征数据。其中，所述第二预设范围包含于所述第一预设范围。
在一个例子中，第一预设范围为小于或等于第一阈值，所述第二预设范围为大于或等于第二阈值且小于或等于第一阈值，其中，所述第二阈值小于所述第一阈值。
在一个例子中，特征数据库中的特征数据以模板的形式存储；自学习模块，具体包括：
第一更新子模块，用于在所述第一特征数据所属用户对应的模板数小于第一预设阈值时，将所述目标对象人脸图片的特征数据存为所述第一特征数据所属用户的新模板。
第二更新子模块，用于在所述第一特征数据所属用户对应的模板数大 于或等于所述第一预设阈值时，利用所述特征数据替换所述第一用户的模板之一，或将所述特征数据融合入所述第一特征数据所属用户的模板之一。
在一个例子中，特征数据库中的一个模板来自一次采集到的人脸图片。
在一个例子中，所述采集的目标对象的人脸图片包括：泛光图像和/或结构光图像；对应的，所述获取模块具体包括：
第一获取子模块，用于在采集的所述目标对象的人脸图片为泛光图像时，根据所述泛光图像获取所述目标对象人脸图片的所述特征数据。
第二获取子模块，用于在采集的所述目标对象的人脸图片为结构光图像时，根据所述结构光图像获取所述目标对象人脸图片的所述特征数据。
第三获取子模块，用于在采集的所述目标对象的人脸图片包括泛光图像和结构光图像时，根据所述泛光图像和所述结构光图像获取所述目标对象人脸图片的所述特征数据。
在一个例子中，在采集所述目标对象的人脸图片的泛光图像时，采用红外光源。
在一个例子中，人脸识别装置还包括3D防伪模块，用于在采集的目标对象的人脸图片包括结构光图像，则所述利用所述目标对象人脸图片的特征数据更新所述第一特征数据之前，根据所述目标对象的结构光图像进行3D人脸防伪。
对应的，自学习模块用于在3D防伪模块防伪通过后，利用所述目标对象人脸图片的特征数据更新所述第一特征数据。
在另一个例子中，人脸识别装置还包括3D防伪模块，用于在比对所述 目标对象人脸图片的特征数据和预存的特征数据库之前，根据所述目标对象的结构光图像进行3D人脸防伪。
对应的，比对模块用于在3D防伪模块防伪通过后，比对所获取到的特征数据和预存的特征数据库。
在一个例子中，3D防伪模块，具体包括：
重建子模块，用于对所述结构光图像进行3D重建，获得重建图。
确认子模块，用于根据所述重建图确认是否来自真人。
防伪结果确认子模块，用于在确认来自真人时，确定所述3D人脸防伪通过。
在一个例子中，采集模块，包括：
采集子模块，用于采集图片；
检测子模块，用于对所述图片进行人脸检测；
处理子模块，用于在检测到人脸时，将所述图片作为所述目标对象的人脸图片。
在一个例子中，人脸识别装置还包括：采集次数判断模块，用于在特征数据库中未包含与所述目标对象人脸图片的特征数据的差异性属于第一预设范围内的第一特征数据，判断总采集次数是否超过第二预设阈值，并在总采集次数不超过第二预设阈值时，触发采集模块。
不难发现，本实施方式为与第一实施方式相对应的装置实施例，本实施方式可与第一实施方式互相配合实施。第一实施方式中提到的相关技术细节在本实施方式中依然有效，为了减少重复，这里不再赘述。相应地，本实施方式中提到的相关技术细节也可应用在第一实施方式中。
值得一提的是，本实施方式中所涉及到的各模块均为逻辑模块，在实际应用中，一个逻辑单元可以是一个物理单元，也可以是一个物理单元的一部分，还可以以多个物理单元的组合实现。此外，为了突出本发明的创新部分，本实施方式中并没有将与解决本发明所提出的技术问题关系不太密切的单元引入，但这并不表明本实施方式中不存在其它的单元。
本发明第五实施方式涉及一种人脸识别装置，如图6所示，包括：
至少一个处理器；以及，与至少一个处理器通信连接的存储器；其中，存储器存储有可被至少一个处理器执行的指令，指令被至少一个处理器执行，以使至少一个处理器能够执行如上述第一实施方式至第三实施方式中任意一个人脸识别方法。
其中，存储器和处理器采用总线方式连接，总线可以包括任意数量的互联的总线和桥，总线将一个或多个处理器和存储器的各种电路连接在一起。总线还可以将诸如外围设备、稳压器和功率管理电路等之类的各种其他电路连接在一起，这些都是本领域所公知的，因此，本文不再对其进行进一步描述。总线接口在总线和收发机之间提供接口。收发机可以是一个元件，也可以是多个元件，比如多个接收器和发送器，提供用于在传输介质上与各种其他装置通信的单元。经处理器处理的数据通过天线在无线介质上进行传输，进一步，天线还接收数据并将数据传送给处理器。
其中，处理器负责管理总线和通常的处理，还可以提供各种功能，包括定时，外围接口，电压调节、电源管理以及其他控制功能。而存储器可以被用于存储处理器在执行操作时所使用的数据。
本发明第七实施方式涉及一种计算机可读存储介质，存储有计算机程序。计算机程序被处理器执行时实现上述方法实施例。
即，本领域技术人员可以理解，实现上述实施例方法中的全部或部分步骤是可以通过程序来指令相关的硬件来完成，该程序存储在一个存储介质中，包括若干指令用以使得一个设备(可以是单片机，芯片等)或处理器(processor)执行本申请各个实施例方法的全部或部分步骤。而前述的存储介质包括：U盘、移动硬盘、只读存储器(ROM，Read-Only Memory)、随机存取存储器(RAM，Random Access Memory)、磁碟或者光盘等各种可以存储程序代码的介质。
本领域的普通技术人员可以理解，上述各实施例是实现本申请的具体实施例，而在实际应用中，可以在形式上和细节上对其作各种改变，而不偏离本申请的精神和范围。

Claims (23)

  1. 一种人脸识别方法，其特征在于，应用于人脸识别装置，所述方法包括：
    采集目标对象的人脸图片，获取所述目标对象的人脸图片的特征数据；
    比对所述目标对象人脸图片的特征数据和预存的特征数据库；
    在所述特征数据库中包含与所述目标对象人脸图片的特征数据的差异性属于第一预设范围内的第一特征数据时，确定比对结果为识别通过；
    利用所述目标对象人脸图片的特征数据更新所述第一特征数据。
  2. 如权利要求1所述的方法，其特征在于，在判定所述特征数据库中包含与所述目标对象人脸图片的特征数据的差异性属于第一预设范围内的第一特征数据之后，还包括：
    判断所述目标对象人脸图片的特征数据与所述第一特征数据的差异性是否属于第二预设范围内；
    若属于，则执行所述利用所述目标对象人脸图片的特征数据更新所述第一特征数据的步骤；
    其中，所述第二预设范围包含于所述第一预设范围。
  3. 如权利要求2所述的方法，其特征在于，所述第一预设范围为小于或等于第一阈值，所述第二预设范围为大于或等于第二阈值且小于或等于第一阈值，其中，所述第二阈值小于所述第一阈值。
  4. 如权利要求1所述的方法，其特征在于，所述特征数据库中的特征数据以模板的形式存储；
    所述利用所述目标对象人脸图片的特征数据更新所述第一特征数据包括：
    若所述第一特征数据所属用户对应的模板数小于第一预设阈值，则将所述目标对象人脸图片的特征数据存为所述第一特征数据所属用户的新模板；
    若所述第一特征数据所属用户对应的模板数大于或等于所述第一预设阈值，则利用所述特征数据替换所述第一用户的模板之一，或将所述特征数据融合入所述第一特征数据所属用户的模板之一。
  5. 如权利要求4所述的方法，其特征在于，所述特征数据库中的一个模板来自一次采集到的人脸图片。
  6. 如权利要求1所述的方法，其特征在于，所述采集的目标对象的人脸图片包括：泛光图像和/或结构光图像；
    所述获取所述目标对象的人脸图片的特征数据，包括：
    若采集的所述目标对象的人脸图片为泛光图像，则根据所述泛光图像获取所述目标对象人脸图片的所述特征数据；
    若采集的所述目标对象的人脸图片为结构光图像，则根据所述结构光图像获取所述目标对象人脸图片的所述特征数据；
    若采集的所述目标对象的人脸图片包括泛光图像和结构光图像，则根据所述泛光图像和所述结构光图像获取所述目标对象人脸图片的所述特征数据。
  7. 如权利要求6所述的方法，其特征在于，在采集所述目标对象的人脸图片的泛光图像时，采用红外光源。
  8. 如权利要求1所述的方法，其特征在于，若采集的目标对象的人脸图片包括结构光图像，则所述利用所述目标对象人脸图片的特征数据更新所述第一特征数据之前，还包括：
    根据所述目标对象的结构光图像进行3D人脸防伪；
    在3D人脸防伪通过后，执行所述利用所述目标对象人脸图片的特征数据更新所述第一特征数据的步骤。
  9. 如权利要求1所述的方法，其特征在于，所述比对所述目标对象人脸图片的特征数据和预存的特征数据库之前，还包括：
    根据所述目标对象的结构光图像进行3D人脸防伪；
    在3D人脸防伪通过后，执行所述比对所获取到的特征数据和预存的特征数据库的步骤。
  10. 如权利要求8或9所述的方法，其特征在于，所述根据所述目标对象的结构光图像进行3D人脸防伪包括：
    对所述结构光图像进行3D重建，获得重建图；
    根据所述重建图确认是否来自真人；
    若确认来自真人，则确定所述3D人脸防伪通过。
  11. 如权利要求1所述的方法，其特征在于，所述采集目标对象的人脸图片，包括：
    采集图片；
    对所述图片进行人脸检测；
    在检测到人脸时，将所述图片作为所述目标对象的人脸图片。
  12. 如权利要求1所述的方法，其特征在于，所述比对所获取到的特征数据和预存的特征数据库之后，还包括：
    在所述特征数据库中未包含与所述目标对象人脸图片的特征数据的差异性属于第一预设范围内的第一特征数据，且总采集次数不超过第二预设阈值时，则重新执行所述目标对象的采集人脸图片的步骤。
  13. 一种人脸识别装置，其特征在于，包括：
    采集模块，用于采集目标对象的人脸图片；
    获取模块，用于获取所述目标对象的人脸图片的特征数据；
    比对模块，用于比对所述目标对象人脸图片的特征数据和预存的特征数据库；
    比对结果确认模块，用于在所述特征数据库中包含与所述目标对象人脸图片的特征数据的差异性属于第一预设范围内的第一特征数据时，确定比对结果为识别通过；
    自学习模块，用于利用所述目标对象人脸图片的特征数据更新所述第一特征数据。
  14. 如权利要求13所述的装置，其特征在于，还包括：
    处理模块，用于在比对结果确认模块确认在特征数据库中包含与目标对象人脸图片的特征数据的差异性属于第一预设范围内的第一特征数据之后，判断所述目标对象人脸图片的特征数据与所述第一特征数据的差异性是否属于第二预设范围内；
    所述自学习模块用于在处理模块判定目标对象人脸图片的特征数据与所述第一特征数据的差异性属于第二预设范围内时，利用目标对象人脸图片的特征数据更新第一特征数据，其中，所述第二预设范围包含于所述第一预设范围。
  15. 如权利要求14所述的装置，其特征在于，第一预设范围为小于或等于第一阈值，所述第二预设范围为大于或等于第二阈值且小于或等于第一阈值，其中，所述第二阈值小于所述第一阈值。
  16. 如权利要求13所述的装置，其特征在于，所述特征数据库中的特征数据以模板的形式存储；所述自学习模块包括：
    第一更新子模块，用于在所述第一特征数据所属用户对应的模板数小于第一预设阈值时，将所述目标对象人脸图片的特征数据存为所述第一特征数据所属用户的新模板；
    第二更新子模块，用于在所述第一特征数据所属用户对应的模板数大于或等于所述第一预设阈值时，利用所述特征数据替换所述第一用户的模板之一，或将所述特征数据融合入所述第一特征数据所属用户的模板之一。
  17. 如权利要求16所述的装置，其特征在于，所述特征数据库中的一个模板来自一次采集到的人脸图片。
  18. 如权利要求13所述的装置，其特征在于，所述采集的目标对象的人脸图片包括：泛光图像和/或结构光图像；所述获取模块包括：
    第一获取子模块，用于在采集的所述目标对象的人脸图片为泛光图像时，根据所述泛光图像获取所述目标对象人脸图片的所述特征数据；
    第二获取子模块，用于在采集的所述目标对象的人脸图片为结构光图像时，根据所述结构光图像获取所述目标对象人脸图片的所述特征数据；
    第三获取子模块，用于在采集的所述目标对象的人脸图片包括泛光图像和结构光图像时，根据所述泛光图像和所述结构光图像获取所述目标对象人脸图片的所述特征数据。
  19. 如权利要求13所述的装置，其特征在于，还包括：3D防伪模块，用于在采集的目标对象的人脸图片包括结构光图像，则所述利用所述目标对象人 脸图片的特征数据更新所述第一特征数据之前，根据所述目标对象的结构光图像进行3D人脸防伪；
    所述自学习模块，用于在3D防伪模块防伪通过后，利用所述目标对象人脸图片的特征数据更新所述第一特征数据。
  20. 如权利要求13所述的装置，其特征在于，还包括：3D防伪模块，用于在比对所述目标对象人脸图片的特征数据和预存的特征数据库之前，根据所述目标对象的结构光图像进行3D人脸防伪；
    所述比对模块，用于在3D防伪模块防伪通过后，比对所获取到的特征数据和预存的特征数据库。
  21. 如权利要求19或20所述的装置，其特征在于，所述3D防伪模块包括：
    重建子模块，用于对所述结构光图像进行3D重建，获得重建图；
    确认子模块，用于根据所述重建图确认是否来自真人；
    防伪结果确认子模块，用于在确认来自真人时，确定所述3D人脸防伪通过。
  22. 一种人脸识别装置，其特征在于，包括：
    至少一个处理器；以及，
    与所述至少一个处理器通信连接的存储器；其中，
    所述存储器存储有可被所述至少一个处理器执行的指令，所述指令被所述至少一个处理器执行，以使所述至少一个处理器能够执行如权利要求1至12中任一项所述的人脸识别方法。
  23. 一种计算机可读存储介质，存储有计算机程序，其特征在于，所述计算机程序被处理器执行时实现如权利要求1至12中任一项所述的人脸识别方法。
PCT/CN2019/086062 2019-05-08 2019-05-08 一种人脸识别方法、人脸识别装置和计算机可读存储介质 WO2020223937A1 (zh)

Priority Applications (2)

Application Number Priority Date Filing Date Title
CN201980000669.1A CN110268419A (zh) 2019-05-08 2019-05-08 一种人脸识别方法、人脸识别装置和计算机可读存储介质
PCT/CN2019/086062 WO2020223937A1 (zh) 2019-05-08 2019-05-08 一种人脸识别方法、人脸识别装置和计算机可读存储介质

Applications Claiming Priority (1)

Application Number Priority Date Filing Date Title
PCT/CN2019/086062 WO2020223937A1 (zh) 2019-05-08 2019-05-08 一种人脸识别方法、人脸识别装置和计算机可读存储介质

Publications (1)

Publication Number Publication Date
WO2020223937A1 true WO2020223937A1 (zh) 2020-11-12

Family

ID=67912951

Family Applications (1)

Application Number Title Priority Date Filing Date
PCT/CN2019/086062 WO2020223937A1 (zh) 2019-05-08 2019-05-08 一种人脸识别方法、人脸识别装置和计算机可读存储介质

Country Status (2)

Country Link
CN (1) CN110268419A (zh)
WO (1) WO2020223937A1 (zh)

Cited By (3)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN113837006A (zh) * 2021-08-20 2021-12-24 奥比中光科技集团股份有限公司 一种人脸识别方法、装置、存储介质及电子设备
CN114743254A (zh) * 2022-06-13 2022-07-12 泽景(西安)汽车电子有限责任公司 脸部认证方法、装置、终端设备及存储介质
CN115641234A (zh) * 2022-10-19 2023-01-24 广州友好教育科技有限公司 一种基于大数据的远程教育系统

Families Citing this family (9)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN111325139B (zh) * 2020-02-18 2023-08-04 浙江大华技术股份有限公司 一种唇语识别方法及装置
CN111837133A (zh) * 2020-03-25 2020-10-27 深圳市汇顶科技股份有限公司 数据采集装置、人脸识别装置、设备、方法及存储介质
CN111814702A (zh) * 2020-07-13 2020-10-23 安徽兰臣信息科技有限公司 一种基于成年人脸和儿童照特征空间映射关系的儿童人脸识别方法
CN112325793B (zh) * 2020-09-30 2022-02-18 张成林 防伪检测方法、装置及多目相机
CN112270210B (zh) * 2020-10-09 2024-03-01 珠海格力电器股份有限公司 数据处理、操作指令识别方法、装置、设备和介质
CN114387635A (zh) * 2020-10-20 2022-04-22 杭州海康威视数字技术股份有限公司 更新生物特征库的方法、装置及电子设备
CN113158786A (zh) * 2021-03-11 2021-07-23 光控特斯联(上海)信息科技有限公司 人脸识别数据处理方法、装置、计算机设备和存储介质
CN113295635A (zh) * 2021-05-27 2021-08-24 河北先河环保科技股份有限公司 一种基于动态更新数据集的水质污染报警方法
CN115798023B (zh) * 2023-02-13 2023-04-18 成都睿瞳科技有限责任公司 一种人脸识别认证方法、装置、存储介质及处理器

Citations (4)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
US20130094768A1 (en) * 2011-10-14 2013-04-18 Cywee Group Limited Face-Tracking Method with High Accuracy
CN106295596A (zh) * 2016-08-17 2017-01-04 深圳市金立通信设备有限公司 一种基于人脸识别的解锁方法及终端
CN109117817A (zh) * 2018-08-28 2019-01-01 摩佰尔（天津）大数据科技有限公司 人脸识别的方法及装置
CN109344745A (zh) * 2018-09-15 2019-02-15 孙建军 一种基于监控系统的人脸图像识别方法

Family Cites Families (8)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN101303724A (zh) * 2007-05-10 2008-11-12 中国银联股份有限公司 一种认证授权方法及系统
CN102622588B (zh) * 2012-03-08 2013-10-09 无锡中科奥森科技有限公司 双验证人脸防伪方法及装置
JP5902661B2 (ja) * 2013-09-30 2016-04-13 株式会社東芝 認証装置、認証システムおよび認証方法
CN106778653A (zh) * 2016-12-27 2017-05-31 北京光年无限科技有限公司 面向智能机器人的基于人脸识别样本库的交互方法和装置
WO2019027504A1 (en) * 2017-08-01 2019-02-07 Apple Inc. METHOD FOR UPDATING MODELS USED IN FACIAL RECOGNITION
CN107590434A (zh) * 2017-08-09 2018-01-16 广东欧珀移动通信有限公司 识别模型更新方法、装置和终端设备
CN108446387A (zh) * 2018-03-22 2018-08-24 百度在线网络技术（北京）有限公司 用于更新人脸注册库的方法和装置
CN108596061A (zh) * 2018-04-12 2018-09-28 Oppo广东移动通信有限公司 人脸识别方法、装置及移动终端、存储介质

Patent Citations (4)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
US20130094768A1 (en) * 2011-10-14 2013-04-18 Cywee Group Limited Face-Tracking Method with High Accuracy
CN106295596A (zh) * 2016-08-17 2017-01-04 深圳市金立通信设备有限公司 一种基于人脸识别的解锁方法及终端
CN109117817A (zh) * 2018-08-28 2019-01-01 摩佰尔（天津）大数据科技有限公司 人脸识别的方法及装置
CN109344745A (zh) * 2018-09-15 2019-02-15 孙建军 一种基于监控系统的人脸图像识别方法

Cited By (4)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN113837006A (zh) * 2021-08-20 2021-12-24 奥比中光科技集团股份有限公司 一种人脸识别方法、装置、存储介质及电子设备
CN114743254A (zh) * 2022-06-13 2022-07-12 泽景(西安)汽车电子有限责任公司 脸部认证方法、装置、终端设备及存储介质
CN115641234A (zh) * 2022-10-19 2023-01-24 广州友好教育科技有限公司 一种基于大数据的远程教育系统
CN115641234B (zh) * 2022-10-19 2024-04-26 北京尚睿通教育科技股份有限公司 一种基于大数据的远程教育系统

Also Published As

Publication number Publication date
CN110268419A (zh) 2019-09-20

Similar Documents

Publication Publication Date Title
WO2020223937A1 (zh) 一种人脸识别方法、人脸识别装置和计算机可读存储介质
US11238270B2 (en) 3D face identity authentication method and apparatus
CN107609383B (zh) 3d人脸身份认证方法与装置
CN107633165B (zh) 3d人脸身份认证方法与装置
US11188788B2 (en) System and method to determine a timing update for an image recognition model
US9985963B2 (en) Method and system for authenticating liveness face, and computer program product thereof
US20180165511A1 (en) Three-dimensional facial recognition method and system
JP5923723B2 (ja) 人物属性推定システム、人物属性推定装置、及び人物属性推定方法
CA3152812A1 (en) Facial recognition method and apparatus
US20210342990A1 (en) Image coordinate system transformation method and apparatus, device, and storage medium
CN110309782A (zh) 一种基于红外与可见光双目系统的活体人脸检测方法
CN105654048A (zh) 一种多视角人脸比对方法
JP2016001447A (ja) 画像認識システム、画像認識装置、画像認識方法、およびコンピュータプログラム
CN107292300A (zh) 一种人脸识别设备及方法
CN102262727A (zh) 客户采集终端人脸图像质量实时监控方法
JP2008146356A (ja) 視線方向推定装置及び視線方向推定方法
CN110569785B (zh) 一种融合跟踪技术的人脸识别方法
CN105022999A (zh) 一种人码伴随实时采集系统
CN109858433B (zh) 一种基于三维人脸模型识别二维人脸图片的方法及装置
CN103907122A (zh) 用于生物计量类型的访问控制系统的欺诈检测
CN113327348A (zh) 一种联网型3d人脸智能锁
CN113792581A (zh) 一种活体检测方法、装置及电子设备
US11048926B2 (en) Adaptive hand tracking and gesture recognition using face-shoulder feature coordinate transforms
WO2020215229A1 (zh) 一种人脸注册方法、人脸注册装置、服务器和可存储介质
US11763601B2 (en) Techniques for detecting a three-dimensional face in facial recognition

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: 19928277

Country of ref document: EP

Kind code of ref document: A1

NENP Non-entry into the national phase

Ref country code: DE

122 Ep: pct application non-entry in european phase

Ref document number: 19928277

Country of ref document: EP

Kind code of ref document: A1