GB/T 41772-2022Information technology—Biometrics—Technical requirements for face recognition system (English PDF)

3.1 Face recognition

The automatic identification of a natural person based on that individual's facial features. [Source: GB/T 38671-2020, 3.1.2, modified]

3.2 Face recognition system

A system that confirms a specific natural person or the identity of a specific natural person through face recognition.

3.3 User

< Face recognition system> Any natural person or organization that interacts with the face recognition system in any way.

Note: When discussing users related to face recognition systems, special terms will be used. Such as. face feature collection subject, face data subject, face recognition system Owner, face recognition system operator, etc. In this document, it is necessary to judge the specific object of "user" according to the context. [Source: GB/T 5271.37-2021, 3.7.20, modified]

3.4 Face feature item

facefeature Values or markers extracted from face samples for comparison.

3.5 Face sample

The analog representation or digital representation of the facial feature characteristics before the extraction of facial feature items.

3.6 Face reference

Stored face samples or face feature items that belong to the user and are used as face comparison objects.

Note: The selection and use of the face reference type is related to the application. For specific regulations, see GB/T 35273 and GB/T 40660.

3.7 Face reference identifier

Pointer to a face reference data record in the face reference database.

3.8 Face probe

A collection of face samples or face feature items that are input into the algorithm and compared with the face reference.

Note: The selection and use of the face probe type is related to the application. For specific regulations, see GB/T 35273 and GB/T 40660.

3.9 Face verification

Compare the generated face probe with the given stored face reference (1.1 comparison) to confirm whether the user is the claimed A face recognition application method for personal identity. [Source: GB/T 38671-2020, 3.1.4, modified]

3.10 Face recognition

Compare the generated face probe with all the face references in the specified range stored (1.N comparison) to confirm the user A face recognition application of identity. [Source: GB/T 38671-2020, 3.1.5, modified]

3.11 Sample quality

A measure of the fitness of a face sample that satisfies a specified condition for a face recognition application.

Note: The specified conditions for face recognition applications involve several aspects, such as clarity, face size, etc.

3.12 Quality judgment

The process of checking whether the quality of the collected samples meets the specified conditions of the face recognition application.

3.13 Similarity score

Alignment score that increases with increasing similarity. [Source: GB/T 5271.37-2021, 3.3.35]

3.14 False acceptance rate

In face verification, the ratio of the number of false acceptances to the sum of the number of correct rejections and the number of false acceptances.

Note: False acceptance rate is sometimes also referred to as false match rate, false recognition rate, false recognition rate, false positive rate, etc., and the appropriate term is selected according to the specific application, expressed as a percentage.

3.15 False rejection rate

In face verification, the ratio of the number of false rejections to the sum of the number of correct acceptances and the number of false rejections.

Note: False rejection rate is sometimes called rejection rate, false negative rate, etc., and the appropriate term is selected according to the specific application, expressed as a percentage.

3.16 False acceptance rate

In face recognition, the ratio of the number of false acceptances to the sum of the number of correct rejections and the number of false acceptances.

Note: False acceptance rate is sometimes also called false alarm rate, false alarm rate, false alarm rate, etc., expressed as a percentage.

5.1 System Architecture

The specific composition of the face recognition system is as follows.

a) View acquisition subsystem. used for the acquisition of face images or videos, including face acquisition equipment and the face acquisition process any subprocess.

b) View analysis subsystem. used for face image and/or video processing, including face detection, quality judgment, feature extraction, face Tracking, attribute detection, liveness detection, etc.

c) Storage subsystem. used for the storage of face registration data and real-time collected data, including.

1) Face registration database. used for storage of registration data;

2) Real-time collection database. used for storage of collected data.

d) Comparison subsystem. the face probe is compared with one or more face references to obtain a similarity score, and the similarity score is passed to Input to the face recognition decision-making subsystem, including two modes.

1) Face verification mode. compare the face probe with the specified face reference (1.1 comparison), and output a similarity score;

2) Face recognition mode. compare the face probe with some or all of the face references (1.N comparison), and output multiple similar degree score, and sorted according to the similarity score.

e) Decision-making subsystem. According to one or more similarity scores, it provides decision-making results for face recognition, including two modes.

1) Face verification mode. when the similarity score exceeds the specified threshold, the face reference and the face probe are matched;

2) Face recognition mode. When the similarity score exceeds the specified threshold, the corresponding face reference constitutes a match with the face probe. potential candidates.

f) Management subsystem. manage the overall strategy, execution and application of the face recognition system, including but not limited to.

1) Set thresholds, such as sample quality thresholds, similarity thresholds, liveness detection thresholds, etc.;

2) Log management. log generation, query and export, etc.;

3) Rights management. set the operation rights of different roles, etc.;

4) Interface configuration. configure the view acquisition subsystem of the face recognition system, etc.;

5) User management. store or delete user's face reference and other registration information;

6) Other management. control the working environment and storage of non-biometric data, provide feedback to the user during or after view collection Information feed, interactive management with face recognition applications, etc.

g) Application open interface. the interface between the face recognition system and the face recognition application, including the face registration interface, face verification interface, Face recognition interface, live detection interface, etc.

5.2 Business Process

The business process of the face recognition system includes face registration, face recognition, face update and face logout, etc. a) Face Registration.

1) Start the face registration process;

2) According to the face registration strategy, collect user data, such as face images, videos and/or non-biometric data;

3) The view analysis subsystem analyzes the collected views, such as quality judgment and liveness detection, etc.;

4) Store the user's data record in the face registration database;

5) End the registration process and record the log. b) Face recognition.

1) Start the face recognition process;

2) Collect face images or videos;

3) Compare the face probe with one or more face references;

4) According to the system strategy and similarity score, provide decision-making results for face recognition;

5) Transmit the decision result to the face recognition application;

6) End the identification process and record the log. c) Face update.

1) Start the face update process;

2) According to the face update strategy, use the face reference that passed this comparison to replace or partially replace the previous face reference;

3) End the update process and record the log. d) Face cancellation.

1) Start the face logout process;

2) According to the face cancellation policy, delete or anonymize all data associated with the user to be canceled in the face registration database;

3) End the logout process and record the log.

6.1 View Collection

The face recognition system should have the function of face collection and output face images or videos, and should meet the following requirements.

a) Automatically judge the location of the collection object, and automatically adjust the collection equipment according to the height and distance of the collection object;

b) When the collection objects are people with different physical functions, functions such as voice prompts and/or font size adjustments are provided.

6.2 View Analysis

The face recognition system should have the view analysis function and meet the following requirements.

a) It should be able to perform face detection;

b) It should be able to extract face feature items;

c) The quality of the sample should be judged and the judgment result should be given, and the face sample that fails the quality judgment can be prompted to re-collect;

d) It is advisable to carry out biopsy test on the samples and give the test results. The biopsy test should comply with the provisions of Appendix A. ISO /IEC 30107-2 The liveness detection data record format is given;

e) According to application requirements, face targets can be tracked;

f) Face attributes can be detected, such as gender, age, hairstyle, whether to wear a mask, whether to wear glasses, etc.

Note: In practical applications, whether to use face samples that have not passed the quality judgment may depend on application requirements and system strategies.

6.3 Storage

6.3.1 The face recognition system should have the data storage function. When the face reference is stored in the face recognition system, the following requirements should be met.

a) Support face reference storage to the face registration database, and return the face reference identifier, where the face registration complies with 6.3.4 Regulation;

b) Each user's face reference corresponds to a unique face reference identifier;

c) Support the deletion operation of the face reference identifier, after deletion, the identifier and the corresponding face reference become invalid;

d) Support the face reference identifier query operation, and confirm whether the face reference identifier and the corresponding face reference are valid.

6.3.2 When the face reference is stored outside the face recognition system (such as a user token), the face recognition system should be able to obtain and use the face refer to.

6.3.3 For non-essential purposes, the face recognition system should not store real-time collection data. When the face recognition system has real-time data collection and storage function, the specific requirements are as follows.

a) It should support the storage of face probes to the real-time acquisition database;

b) The query and deletion of real-time collected data should be supported.

6.3.4 The face recognition system should support multiple registration methods and meet the following requirements.

a) shall support on-site registration or off-site registration;

b) It should support batch registration of face samples;

c) It should meet the needs of a single user to register multiple face samples;

d) It should support face registration of different image sources, such as real-time collected photos or stored photos, etc.;

e) It can support face samples collected by one or more collection devices;

f) During the registration process, it should be able to properly interact with the user, such as reminding the user to cooperate, prompting successful registration, etc.

6.4 Comparison

The face recognition system should have a face comparison function and meet the following requirements.

a) The obtained face feature items should be compared with the face feature items in the face registration database, and the similarity score should be calculated;

b) The similarity score should be compared with the threshold, and the face comparison results with similarity scores higher than the threshold should be saved, which can support the comparison result. Results are retrieved and exported;

c) It should have the function of judging and handling abnormal situations.

Note 1. The similarity score is generally a real number ranging from 0 to 100.The lower the score, the less similar the facial features are, and the higher the score, the more similar the facial features.

Note 2. Abnormal situations include but are not limited to comparison failure and export failure.

6.5 Decision making

The face recognition system should have the face recognition decision-making function and meet the following requirements. a) Face verification.

1) If the similarity score compared with the face reference is greater than the specified threshold, it should be judged that the face probe matches the face reference, If it is lower than the specified threshold, it should be judged as a mismatch;

2) The similarity score and decision results should be output to the management subsystem. b) Face recognition.

1) If the similarity scores compared with all face references are less than the specified threshold, it should be judged that the face probe does not match any face reference matching;

2) If the similarity score compared with multiple face references is greater than the specified threshold, the face probe with the highest similarity score should be selected. Large face reference matching;

3) If it is necessary to output possible candidates, the length of the candidate list should be determined first, and then according to the similarity between the face probe and the face reference Degree score, sort output candidate list;

4) The similarity score and decision results should be output to the management subsystem.

c) Before determining the result of the face recognition decision, the face recognition system should allow the user to try the number of times set in the decision policy.

6.6 Management

The face recognition system should have the face recognition management function and meet the following requirements. a) Threshold setting.

1) Similarity threshold. the threshold can be initially set, and the user is allowed to adjust the threshold according to actual needs;

2) Quality Judgment Threshold. The threshold can be initially set, and the user is allowed to adjust the threshold according to actual needs.

Note: The supplier provides the mapping relationship between similarity threshold, quality judgment threshold and recognition performance indicators for user reference.

b) Log management. It should support authorized users to query and export logs.

c) Rights management. It should support the configuration of user operation rights.

d) Interface configuration. The application opening interface, protocol interface, hardware interface, etc. should be configured.

e) User management. User information can be added, modified, deleted, inquired, stopped/enabled, etc.

f) Other management. the final decision can be made on the decision result and/or similarity score output by the decision-making subsystem, etc.

6.7 Application Open Interface

The face recognition system should have an open application interface, and provide services for face recognition applications through this interface, and the algorithm and design of the service Equipment, etc. shall be registered in accordance with the provisions of GB/T 28826.2. Services provided include but are not limited to.

A.1 Types of prosthetic attacks

A.1.1 Overview

Types of prosthetic attacks include, but are not limited to, two-dimensional prosthetic attacks and three-dimensional prosthetic attacks.

A.1.2 Two-dimensional prosthetic attack

Two-dimensional prosthetic attack types include but are not limited to the following.

a) Two-dimensional static paper image attack, the factors to be considered are as follows.

- Sample materials. printing paper, matte photo paper, high gloss photo paper, suede photo paper, matte powder paper, coated paper, etc.;

- Sample quality. resolution, clarity, size, angle, lighting conditions, completeness, etc.;

- Presentation methods. distance, angle, movement, bending, folding, etc.;

- Cropping method. Whether to cut out the eyes, nose, mouth, etc. of the image.

b) For two-dimensional static electronic image attacks, the following factors need to be considered.

- Device type. mobile terminal, microcomputer, etc.;

- Device display performance. resolution, brightness, contrast, etc.;

- Sample quality. resolution, clarity, size, angle, lighting conditions, completeness, etc.;

- Presentation mode, distance, angle, movement, etc.

c) For two-dimensional dynamic image attacks, the following factors need to be considered.

- Two-dimensional dynamic image types. recorded video, synthesized video, etc.;

- Device type. mobile terminal, microcomputer, etc.;

- Device display performance. resolution, brightness, contrast, etc.;

- Two-dimensional dynamic image quality. resolution, clarity, frame rate, etc.;

- Presentation mode. distance, angle, movement, etc.

A.1.3 3D prosthetic attack

Three-dimensional prosthetic attack types include but are not limited to the following.

a) For 3D mask attack, the following factors need to be considered.

- Mask material. plastic mask, three-dimensional paper mask, silicone mask, etc.;

- Presentation method. distance, angle, movement, etc.;

- Light conditions. normal light, strong light, low light, backlight, etc.;

- Cutting method. whether to cut out the eyes, nose, mouth, etc. of the mask.

b) 3D head model attack, the following factors need to be considered.

- Head mold material. foam, resin, full-color sandstone, quartz sand, etc.;

- Presentation method. distance, angle, movement, etc.;

- Light conditions. normal light, strong light, low light, backlight, etc.

A.2 Liveness detection performance requirements

Liveness detection performance metrics include LDFAR and LDFRR.