WO2021036436A1

WO2021036436A1 - Facial recognition method and apparatus

Info

Publication number: WO2021036436A1
Application number: PCT/CN2020/096992
Authority: WO
Inventors: 韩雨; 杭欣
Original assignee: 苏宁易购集团股份有限公司; 苏宁云计算有限公司
Priority date: 2019-08-26
Filing date: 2020-06-19
Publication date: 2021-03-04
Also published as: CN110688901A; CA3152812A1

Abstract

Disclosed are a facial recognition method and apparatus. The method comprises: acquiring a recognition request and an image to be recognized; according to the image to be recognized, acquiring a facial feature of the image to be recognized; according to the recognition request, acquiring, from a comparison library, all features of a reference image for comparison, wherein each ID in the comparison library comprises reference images of different postures in a plurality of different scenarios; and according to the facial feature of the image to be recognized, and all the features of the reference image for comparison, determining whether the image to be recognized matches the reference image for comparison. According to the present invention, by using the comparison library of reference images, of the same ID, of different postures in a plurality of different scenarios as a matching standard, the recognition accuracy is improved, the robustness of an algorithm is enhanced, and a better adaptability to the expression, image quality, etc. of an image to be recognized is provided.

Description

Face recognition method and device

Technical field

The present invention relates to the technical field of computer vision, in particular to a face recognition method and device.

Background technique

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 human faces, and automatically detect and track human faces in the images, and then recognize the detected human faces, usually also called face recognition and facial recognition. With the development of technology, face recognition technology has become one of the hottest applications of artificial intelligence, such as swiping face to board plane, swiping face to get toilet paper, swiping face to pay, swiping face to check attendance, swiping face to recognize pedestrians running red lights, and so on.

Face recognition usually has the following three application modes:

1. The 1:1 mode, also known as the identity verification mode, is essentially a process in which the computer quickly compares the current face with the portrait database and finds whether it matches. It can be simply understood as proving that you are you. "Face-swiping" boarding, ticket checking, and payment all belong to the 1:1 verification;

2. The 1 to N mode is to find the current user's face data (ie the image to be recognized) in a massive portrait database and perform matching. People who commit abductions and expose red lights are all classified as 1:N face recognition, that is, one target is found out of N faces;

3. The M-to-N mode is a process of facial recognition of all people in the scene through the computer and comparison with the portrait database. It is a dynamic face comparison, which can be fully applied to a variety of scenarios, such as public security, welcoming, and Robot applications, etc.

However, with the increase in the number of comparisons in the 1:1 mode and the increase in the number of registered base IDs in the 1:N mode, the probability of matching failures and matching errors will increase significantly, mainly due to changes in facial features Caused by big differences. Among them, the factors that affect the accuracy of face recognition include:

(1) The facial features are not fixed and will change significantly with age, expression, posture, makeup, etc.;

(2) The incomplete face caused by the occlusion of accessories, glasses, masks or other objects will also increase the difficulty of recognition;

(3) The collection environment of the training set, such as lighting, equipment parameters, etc., will limit the performance of the algorithm model on the test set of other scenarios;

(4) Different people may adopt similar makeup methods and styles, wear similar accessories, etc., which increase the similarity of different people.

Therefore, there is an urgent need to propose a new face recognition method that has solved many of the above problems

Summary of the invention

In order to solve the problems of the prior art, the embodiments of the present invention provide a face recognition method and device to overcome the face recognition caused by factors such as the increase in the number of comparisons and the increase in the number of registered base IDs in the prior art. Problems such as low accuracy.

In order to solve one or more of the above technical problems, the technical solution adopted by the present invention is:

On the one hand, a face recognition method is provided, and the method includes the following steps:

Obtain the recognition request and the image to be recognized;

Acquiring the facial features of the image to be recognized according to the image to be recognized;

Acquiring all the features of the reference image to be compared from the comparison library according to the recognition request, each ID in the comparison library includes multiple reference images with different poses in different scenarios;

According to the facial features of the image to be identified and all features of the reference image to be compared, it is determined whether the image to be identified matches the reference image to be compared.

Further, the acquiring the facial features of the image to be recognized according to the image to be recognized includes:

Performing face frame detection and face key point detection on the image to be recognized, and acquiring a face image and key point positions corresponding to the image to be recognized;

Normalize the face image according to the position of the key point, and obtain a processed face image;

Perform feature extraction on the processed face image, and obtain the face feature corresponding to the image to be recognized.

Further, when the identification request is a 1:1 identification request, the identification request includes the ID of the reference image to be compared;

The obtaining all the features of the reference image to be compared from the comparison library according to the recognition request includes:

Obtain all features of all reference images under the ID of the reference image to be compared from the comparison library according to the ID of the reference image to be compared.

Further, the judging whether the image to be identified matches the reference image to be compared according to the facial features of the image to be identified and all the features of the reference image to be compared includes:

All the features of each reference image under the ID of the reference image to be compared are compared with the facial features of the image to be recognized, and if the similarity exceeds a first preset threshold, it is determined that the reference image is The image to be recognized is successfully matched;

Acquire the number of reference images that are successfully matched with the image to be recognized, and if the number exceeds half of the total number of reference images, determine that the ID of the image to be recognized and the reference image to be compared are successfully matched .

Further, when the recognition request is a 1 to N recognition request, the reference images to be compared are all reference images under all IDs in the comparison library;

Obtain all features of all reference images under all IDs from the comparison library according to the identification request.

Calculating and acquiring the similarity between the image to be recognized and each ID according to all the features of each reference image under each ID and the facial features of the image to be recognized;

Determine whether the similarity of the ID with the highest similarity value meets the second preset threshold, if it is satisfied, determine that the image to be recognized matches the ID with the highest similarity value, otherwise determine that the image to be recognized is an unregistered image .

In another aspect, a face recognition device is provided, and the device includes:

The data acquisition module is used to acquire the recognition request and the image to be recognized;

The first feature acquisition module is configured to acquire the facial features of the image to be identified according to the image to be identified;

The second feature acquisition module is configured to acquire all the features of the reference image to be compared from the comparison library according to the recognition request, and each ID in the comparison library includes references of different poses in multiple different scenarios image;

The image recognition module is configured to determine whether the image to be recognized matches the reference image to be compared based on the facial features of the image to be recognized and all the features of the reference image to be compared.

Further, the first feature acquisition module includes:

An image detection unit, configured to perform face frame detection and face key point detection on the to-be-recognized image, and obtain a face image and key point positions corresponding to the to-be-recognized image;

A normalization processing unit, configured to perform normalization processing on the face image according to the position of the key point to obtain a processed face image;

The feature extraction unit is configured to perform feature extraction on the processed face image, and obtain the face feature corresponding to the image to be recognized.

The second feature acquisition module is specifically used for:

Further, the image recognition module includes:

The first calculation unit is configured to calculate each reference image and the image to be recognized according to all the features of each reference image under the ID of the reference image to be compared and the facial features of the image to be recognized The similarity;

A first comparison unit, configured to compare the similarity with a first preset threshold, and if the similarity is greater than the first preset threshold, determine that the reference image matches the image to be recognized successfully;

The second comparison unit is configured to obtain the number of reference images that are successfully matched with the image to be identified, and if the number exceeds half of the total number of the reference images, determine that the image to be identified is compared with the image to be identified The ID of the reference image matches successfully.

The second feature acquisition module is also used for:

Further, the image recognition module further includes:

The second calculation unit is configured to calculate and obtain the similarity between the image to be recognized and each ID according to all the features of each reference image under each ID and the facial features of the image to be recognized;

The third comparison unit is used to determine whether the similarity of the ID with the highest similarity value meets the second preset threshold, and if so, it is determined that the image to be recognized matches the ID with the highest similarity value, otherwise it is determined that the The image to be recognized is an unregistered image.

The beneficial effects brought about by the technical solutions provided by the embodiments of the present invention are:

1. The face recognition method and device provided by the embodiments of the present invention use a comparison library of reference images of different poses in multiple different scenarios with the same ID as the matching standard, which improves the accuracy of recognition and enhances the robustness of the algorithm It has good adaptability to the facial expression and picture quality of the recognized image;

2. The face recognition method and device provided by the embodiments of the present invention adopt the scheme of detecting the face frame and the key points of the face together, which can not only accurately locate the face position, but also reduce the steps and time in the recognition process. , Improve the efficiency of recognition;

3. The face recognition method and device provided by the embodiments of the present invention select the ID with the highest similarity by first calculating the similarity between each reference image under each ID and the image to be recognized in a 1:N recognition scenario , And then determine whether the similarity meets the second preset threshold to match the ID matching the image to be recognized, which improves the anti-attack ability of the algorithm.

Description of the drawings

In order to explain the technical solutions in the embodiments of the present invention more clearly, the following will briefly introduce the drawings needed in the description of the embodiments. Obviously, the drawings in the following description are only some embodiments of the present invention. For those of ordinary skill in the art, other drawings can be obtained from these drawings without creative work.

Fig. 1 is a flowchart showing a face recognition method according to an exemplary embodiment;

Fig. 2 is a flow chart showing obtaining facial features of an image to be recognized according to an image to be recognized according to an exemplary embodiment;

Fig. 3 shows according to an exemplary embodiment in the 1:1 mode, according to the facial features of the image to be recognized and all the features of the reference image to be compared, it is determined whether the image to be recognized matches the reference image to be compared Flow chart

Fig. 4 shows according to an exemplary embodiment in the 1 to N mode, according to the facial features of the image to be recognized and all the features of the reference image to be compared, it is determined whether the image to be recognized matches the reference image to be compared Flow chart

Fig. 5 is a schematic structural diagram of a face recognition device according to an exemplary embodiment.

detailed description

In order to make the objectives, technical solutions and advantages of the present invention clearer, the following will clearly and completely describe the technical solutions in the embodiments of the present invention with reference to the accompanying drawings in the embodiments of the present invention. Obviously, the described embodiments are only A part of the embodiments of the present invention, but not all the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those of ordinary skill in the art without creative work shall fall within the protection scope of the present invention.

Facial features are the most suitable type of biological features for identification. Compared with fingerprints, iris and other features, they have the advantages of simple collection, low cost, and easy identification. The use of human face for identity recognition has a wide range of applications in many scenarios such as face-swiping login, face-swiping credit investigation, and security verification. On the one hand, face recognition reduces manual operations and can save costs. On the other hand, it facilitates identity verification and improves user experience. The basic process of face recognition is to extract features from the face image to be recognized, and then compare with the face features registered in the database.

Fig. 1 is a flowchart of a face recognition method according to an exemplary embodiment. Referring to Fig. 1, the method includes the following steps:

S1: Obtain the recognition request and the image to be recognized.

Specifically, first, the image to be recognized is collected from the photo, video or camera, and the corresponding recognition request is obtained. Generally, face recognition includes three modes: 1:1 mode, 1:N mode, and M:N mode. It should be noted here that the face recognition method provided by the embodiment of the present invention is mainly applicable to 1:1 recognition scenarios and 1:N recognition scenarios. Therefore, the recognition request in the embodiment of the present invention mainly includes a 1:1 recognition request or a 1:1 recognition request. N identifies the request. Among them, 1:1 matching mainly solves the problem of determining whether the image to be recognized and the reference image belong to the same person, while 1:N matching mainly solves the problem of determining which person the image to be recognized belongs to.

S2: Acquire the facial features of the image to be recognized according to the image to be recognized.

Specifically, before performing face recognition on the image to be recognized, the facial features of the image to be recognized need to be extracted from the image to be recognized. As a better example, a convolutional neural network may be used in the embodiment of the present invention to extract the facial features of the image to be recognized. The facial features of the image (that is, the depth features of the image to be recognized).

S3: Acquire all the features of the reference image to be compared from the comparison library according to the recognition request, and each ID in the comparison library includes multiple reference images with different poses in different scenarios.

Specifically, the identification request in the embodiment of the present invention mainly includes a 1:1 identification request or a 1:N identification request. When performing specific recognition, a method of comparing the facial features of the image to be recognized with the features of the reference image to be compared is adopted. Since different recognition requests use different reference images to be compared, it is also necessary to obtain all the features of the corresponding reference images to be compared from the comparison library according to the recognition request.

Each ID in the comparison library includes multiple reference images with different poses in different scenarios to ensure the diversity of the comparison library, thereby increasing the probability of successful recognition. Among them, in the embodiment of the present invention, the reference image maintained under each ID in the comparison library mainly includes the following three scenarios:

1. Authorized ID photo that can be used for personal identity authentication

The ID photos uploaded by the company or group members are authorized to be used for internal personnel authentication, and can be used as a reference image for a class of scenes.

2. Authorized face-swiping login registration photo

The face-scanning login function of social software will ask for the user's authorization and use the avatar for personal identity authentication. During the software registration process, the user is asked to nod or shake his head. At this time, the camera will check the action, and automatically collect the right and left deflection facial image and store it under the corresponding ID as a reference image.

3. Collect photos of internal security

In internal offices, terminal equipment used for security monitoring will accumulate a certain amount of security face images. For security reasons, these photos can also be used as reference images for a class of scenes. When the security collection photos are stored under the corresponding ID as the reference image, the method of clustering by itself and then matching with the comparison library is generally adopted. The specific process is as follows:

First of all, check and proofread the image quality and orientation of the camera that collects security photos to ensure that photos with appropriate angles and lighting can be collected. Then register the qualified security photos into a temporary base library. If the similarity between the newly collected security photos and the most matching registered photo in the temporary base library exceeds the threshold, then the collected photos will be classified under the corresponding ID, otherwise Register as a new ID. Every time a new photo is added under the ID, the registration photo is updated once. The total similarity between the updated registration photo and other photos under the ID should be the largest. This registration photo is considered to be the most representative of this ID. Take the obtained registered photo as the photo to be identified, and use the logic of 1 to N to match the most approximate ID in the comparison library as the ID of the security photo.

It should also be noted here that when the reference image is collected, it must meet the following quality screening standards: meet the multi-dimensional thresholds of sharpness, angle and facial features, and the selected reference image should take into account the face and small deflection. The profile of the face is used to increase the probability of matching successfully with the image to be recognized in a variety of poses.

It should also be noted here that if the reference image in the comparison library is collected through a 1:1 port, when registering to the comparison library, enter a video of the registrant, and use the quality algorithm to obtain the best quality from the video. Feature extraction of several photos is registered and stored. If it is collected through a 1:N port, after collecting the photos to be recognized, compare them with the existing reference images in the comparison library. If the similarity of the closest reference image is also lower than the set threshold, then Considering that there is no reference image corresponding to the ID in the base library, track the person in the image to be recognized, collect a video, use a quality algorithm to select several pictures with the best quality from the video, and create a new ID to add to the registration library.

S4: Determine whether the image to be identified matches the reference image to be compared according to the facial features of the image to be identified and all features of the reference image to be compared.

Specifically, the similarity between the image to be identified and the reference image to be compared is calculated based on the facial features of the image to be identified and all the features of the reference image to be compared, and then the image to be identified and the reference image to be compared are judged based on the similarity. Whether the reference image matches.

Fig. 2 is a flow chart showing the acquisition of facial features of the image to be recognized according to the image to be recognized according to an exemplary embodiment. Referring to Fig. 2, as a preferred embodiment, in the embodiment of the present invention, the Obtaining the facial features of the image to be recognized according to the image to be recognized includes:

S2.1: Perform face frame detection and face key point detection on the image to be recognized, and obtain the face image and key point positions corresponding to the image to be recognized.

Specifically, performing face frame detection on the image to be recognized refers to detecting and locating the face in the image to be recognized, returning high-precision face frame coordinates, and intercepting the face image in the image to be recognized according to the coordinates. The key points of the face are detected on the image to be recognized, and the key areas of the face are located, including the key points of the eyes, nose, and mouth. Wherein, the detected face images in the image to be recognized include one or more, and the user can decide whether to recognize a single face frame or a multi-person face frame according to the actual application scenario, which is not limited in the embodiment of the present invention.

S2.2: Perform normalization processing on the face image according to the position of the key point, and obtain a processed face image.

Specifically, the normalization process is performed on the face image, the purpose is to make the photos of the same person taken under different imaging conditions (such as light intensity, direction, distance, posture, etc.) have consistency, so as to facilitate subsequent facial features extraction . Face normalization includes two aspects: one is geometric normalization, and the other is gray normalization. Geometric normalization is also called position calibration. It will help correct size differences and angle tilts caused by changes in imaging distance and face pose, and can solve the problems of face scale changes and face rotation. Specifically, it includes three links: normalization of face scale, flat face rotation correction (tilted head), and deep face rotation correction (face twisting). For some highly demanding deep face rotation correction, a 3D model of the face can be used. Gray normalization is used to compensate the face image obtained under different light intensities and light source directions, so as to reduce the change of the image signal caused by the change of light alone. It should be noted here that, in order to facilitate subsequent use of models (such as convolutional neural networks) to extract facial features, in the embodiment of the present invention, the facial image needs to be adjusted to a size suitable for the input model.

S2.3: Perform feature extraction on the processed face image, and obtain the face feature corresponding to the image to be recognized.

Specifically, as an example, in the embodiment of the present invention, a pre-trained convolutional neural network is used to perform feature extraction on the face image processed by the above steps to obtain the face feature corresponding to the image to be recognized.

As a preferred embodiment, in the embodiment of the present invention, when the identification request is a 1:1 identification request, the identification request includes the ID of the reference image to be compared;

Specifically, when the face recognition mode is a 1:1 mode, the acquired recognition request is a 1:1 recognition request. At this time, it is necessary to recognize whether the image to be recognized and the reference image belong to the same person. Therefore, the recognition request includes the recognition request. The ID of the compared reference image. At this time, according to the ID of the reference image to be compared, all the features of all the reference images under the ID are obtained from the comparison library.

Fig. 3 is a flow chart showing whether the image to be recognized matches the reference image to be compared according to the facial features of the image to be recognized and all the features of the reference image to be compared according to an exemplary embodiment, refer to Fig. 3 As shown, as a preferred embodiment, in the embodiment of the present invention, the judgment is made based on the facial features of the image to be recognized and all the features of the reference image to be compared. Whether the reference image to be compared matches or not includes:

S401: Calculate the similarity between each reference image and the image to be recognized according to all the features of each reference image under the ID of the reference image to be compared and the facial features of the image to be recognized.

Specifically, when the face recognition mode is a 1:1 mode, each reference image under the ID of the reference image to be compared needs to be compared with the image to be recognized. During specific implementation, the similarity between each reference image and the image to be recognized can be calculated based on all the features of each reference image under the ID of the reference image to be compared and the facial features of the image to be recognized, and the image can be determined based on the similarity. Comparison.

S402: Compare the similarity with a first preset threshold, and if the similarity is greater than the first preset threshold, determine that the reference image matches the image to be recognized successfully.

Specifically, the first preset threshold may be set according to actual requirements, and the specific value of the first preset threshold is not limited here. When the similarity between the reference image and the image to be recognized exceeds (that is, greater than) the first preset threshold, it is determined that the reference image is successfully matched with the image to be recognized.

S403: Acquire the number of reference images that are successfully matched with the image to be recognized, and if the number exceeds half of the total number of reference images, determine the ID of the image to be recognized and the reference image to be compared The match is successful.

Specifically, when the number of reference images successfully matched with the image to be recognized under the ID of the reference image to be compared exceeds the total number of reference images participating in the matching (here refers to the total number of reference images under the ID of the reference image to be compared ), it is finally determined that the ID matching of the image to be identified and the reference image to be compared is successful; otherwise, it is determined that the ID matching of the image to be identified and the reference image to be compared is unsuccessful. This setting can avoid the low similarity between the reference image with the same ID and the single image to be recognized due to large differences in scenes, occlusions, etc., and it can also shield the reference image with different IDs and the image to be recognized due to accident The problem of high similarity caused by factors can improve the accuracy of recognition.

As a preferred embodiment, in the embodiment of the present invention, when the recognition request is a 1:N recognition request, the reference images to be compared are all reference images under all IDs in the comparison library ；

The acquiring all the features of the reference image to be compared from the comparison library according to the recognition request includes:

Specifically, when the face recognition mode is a 1:N mode, the obtained recognition request is a 1:N recognition request. At this time, it is necessary to recognize which person the image to be recognized belongs to. Therefore, the reference image to be compared Compare all reference images under all IDs in the library. At this time, all features of all reference images under all IDs need to be obtained from the comparison library.

Fig. 4 shows according to an exemplary embodiment in the 1 to N mode, according to the facial features of the image to be recognized and all the features of the reference image to be compared, it is determined whether the image to be recognized matches the reference image to be compared 4, as a preferred embodiment, in an embodiment of the present invention, according to the facial features of the image to be recognized and all features of the reference image to be compared, Determining whether the image to be recognized matches the reference image to be compared includes:

S501: Calculate and obtain the similarity between the image to be recognized and each ID according to all the features of each reference image under each ID and the facial features of the image to be recognized.

Specifically, when the face recognition mode is a 1:N mode, all reference images under all IDs in the comparison library need to be compared with the image to be recognized. During specific implementation, the similarity between each reference image and the image to be recognized can be calculated based on all the features of each reference image under each ID and the facial features of the image to be recognized, and the image comparison can be performed based on the similarity.

S502: Determine whether the similarity of the ID with the highest similarity value meets the second preset threshold, if so, determine that the image to be recognized matches the ID with the highest similarity value; otherwise, determine that the image to be recognized is unidentified. Register the image.

Specifically, it is possible to sort all the reference images according to the similarity value first, select the ID corresponding to the reference image with the highest similarity of the image to be recognized, and then determine whether the similarity value meets the second preset threshold, if If it is satisfied, it is determined that the image to be recognized matches the ID with the highest similarity value; otherwise, it is determined that the image to be recognized is an unregistered image. It should be noted here that in the case of multiple reference images with multiple scenes for each ID, the probability that the image to be recognized matches the reference image under its own ID as the maximum similarity is improved, avoiding a single pending image. The problem of low similarity between the reference image under its ID and the image to be recognized due to accidental factors in the recognition image Similarly, the second preset threshold can be set according to actual needs. The specific value of the second preset threshold is not limited here, and the second preset threshold is set to shield the attack of unregistered ID photos.

As a preferred embodiment, in the embodiment of the present invention, the image to be recognized that is unsuccessfully matched with the reference image to be compared is registered with the comparison library, a new ID is generated, and the The image to be recognized is stored in the new ID as a reference image under the new ID.

Fig. 5 is a schematic structural diagram of a face recognition device according to an exemplary embodiment. Referring to Fig. 5, the device includes:

As a preferred implementation manner, in an embodiment of the present invention, the first feature acquisition module includes:

The second feature acquisition module is specifically used for:

As a preferred implementation manner, in an embodiment of the present invention, the image recognition module includes:

The second feature acquisition module is also used for:

As a preferred implementation manner, in an embodiment of the present invention, the image recognition module further includes:

In summary, the beneficial effects brought about by the technical solutions provided by the embodiments of the present invention are:

It should be noted that when the face recognition device provided in the above embodiment triggers the face recognition service, only the division of the above functional modules is used as an example for illustration. In practical applications, the above functions can be allocated to different functions according to needs. Module completion, that is, the internal structure of the device is divided into different functional modules to complete all or part of the functions described above. In addition, the face recognition device provided in the above embodiment and the face recognition method embodiment belong to the same concept, that is, the device is based on the face recognition method. For the specific implementation process, please refer to the method embodiment, which will not be repeated here.

A person of ordinary skill in the art can understand that all or part of the steps in the above embodiments can be implemented by hardware, or by a program to instruct relevant hardware. The program can be stored in a computer-readable storage medium. The storage medium mentioned can be a read-only memory, a magnetic disk or an optical disk, etc.

The above are only the preferred embodiments of the present invention and are not intended to limit the present invention. Any modification, equivalent replacement, improvement, etc. made within the spirit and principle of the present invention shall be included in the protection of the present invention. Within range.

Claims

A face recognition method, characterized in that the method includes the following steps:

Obtain the recognition request and the image to be recognized;

Acquiring the facial features of the image to be recognized according to the image to be recognized;

Acquiring all the features of the reference image to be compared from the comparison library according to the recognition request, each ID in the comparison library includes multiple reference images with different poses in different scenarios;

According to the facial features of the image to be identified and all features of the reference image to be compared, it is determined whether the image to be identified matches the reference image to be compared.
The face recognition method according to claim 1, wherein the acquiring the facial features of the image to be recognized according to the image to be recognized comprises:

Performing face frame detection and face key point detection on the image to be recognized, and acquiring a face image and key point positions corresponding to the image to be recognized;

Performing normalization processing on the face image according to the position of the key point to obtain a processed face image;

Perform feature extraction on the processed face image, and obtain the face feature corresponding to the image to be recognized.
The face recognition method according to claim 1 or 2, wherein when the recognition request is a 1:1 recognition request, the recognition request includes the ID of the reference image to be compared;

The acquiring all the features of the reference image to be compared from the comparison library according to the recognition request includes:

Obtain all features of all reference images under the ID of the reference image to be compared from the comparison library according to the ID of the reference image to be compared.
The face recognition method according to claim 3, characterized in that, the judgment is made based on the facial features of the image to be recognized and all the features of the reference image to be compared to determine the difference between the image to be recognized and the Whether the reference image to be compared matches includes:

Calculating the similarity between each reference image and the image to be recognized according to all the features of each reference image under the ID of the reference image to be compared and the facial features of the image to be recognized;

Comparing the similarity with a first preset threshold, and if the similarity is greater than the first preset threshold, determining that the reference image is successfully matched with the image to be recognized;

Acquire the number of reference images that are successfully matched with the image to be recognized, and if the number exceeds half of the total number of reference images, determine that the ID of the image to be recognized and the reference image to be compared are successfully matched .
The face recognition method according to claim 1 or 2, characterized in that, when the recognition request is a 1:N recognition request, the reference image to be compared is all IDs in the comparison library All reference images;

The acquiring all the features of the reference image to be compared from the comparison library according to the recognition request includes:

Obtain all features of all reference images under all IDs from the comparison library according to the identification request.
The face recognition method according to claim 5, wherein the judgment is made based on the facial features of the image to be recognized and all the features of the reference image to be compared. Whether the reference image to be compared matches includes:

Calculating and acquiring the similarity between the image to be recognized and each ID according to all the features of each reference image under each ID and the facial features of the image to be recognized;

Determine whether the similarity of the ID with the highest similarity value meets the second preset threshold, if it is satisfied, determine that the image to be recognized matches the ID with the highest similarity value, otherwise determine that the image to be recognized is an unregistered image .
A face recognition device based on the method of any one of claims 1 to 6, wherein the device comprises:

The data acquisition module is used to acquire the recognition request and the image to be recognized;

The first feature acquisition module is configured to acquire the facial features of the image to be identified according to the image to be identified;

The second feature acquisition module is configured to acquire all the features of the reference image to be compared from the comparison library according to the recognition request, and each ID in the comparison library includes references of different poses in multiple different scenarios image;

The image recognition module is configured to determine whether the image to be recognized matches the reference image to be compared based on the facial features of the image to be recognized and all the features of the reference image to be compared.
The face recognition device according to claim 7, wherein the first feature acquisition module comprises:

An image detection unit, configured to perform face frame detection and face key point detection on the to-be-recognized image, and obtain a face image and key point positions corresponding to the to-be-recognized image;

A normalization processing unit, configured to perform normalization processing on the face image according to the position of the key point to obtain a processed face image;

The feature extraction unit is configured to perform feature extraction on the processed face image, and obtain the face feature corresponding to the image to be recognized.
The face recognition device according to claim 7 or 8, wherein the image recognition module comprises:

The first calculation unit is configured to calculate each reference image and the image to be recognized according to all the features of each reference image under the ID of the reference image to be compared and the facial features of the image to be recognized The similarity;

A first comparison unit, configured to compare the similarity with a first preset threshold, and if the similarity is greater than the first preset threshold, determine that the reference image matches the image to be recognized successfully;

The second comparison unit is configured to obtain the number of reference images that are successfully matched with the image to be identified, and if the number exceeds half of the total number of the reference images, determine that the image to be identified is compared with the image to be identified The ID of the reference image matches successfully.
The face recognition device according to claim 7 or 8, wherein the image recognition module further comprises:

The second calculation unit is configured to calculate and obtain the similarity between the image to be recognized and each ID according to all the features of each reference image under each ID and the facial features of the image to be recognized;

The third comparison unit is used to determine whether the similarity of the ID with the highest similarity value meets the second preset threshold, and if so, it is determined that the image to be recognized matches the ID with the highest similarity value, otherwise it is determined that the The image to be recognized is an unregistered image.