Automatic authentification for MES system using facial recognition

• US 10,579,785 B2
• Filed: 09/29/2017
• Issued: 03/03/2020
• Est. Priority Date: 09/29/2017
• Status: Active Grant

• Current Assignee: General Electric Company
• Original Assignee: General Electric Company
• Inventors: Beacham, Jimmie, Jia, Tao, Tu, Jilin, Tu, Peter, Wang, Zhiguang
• Primary Examiner(s): Hirl, Joseph P
• Assistant Examiner(s): Gyorfi, Thomas A
• Application Number:
  US15/721,114
• Publication Number:
  US 20190102528A1
• Time in Patent Office: 886 Days
• CPC Class Codes: G06F 21/32 using biometric data, e.g. fingerprints, iris scans or voiceprints
• CPC Class Codes: G06V 10/82 using neural networks
• CPC Class Codes: G06V 40/166 using acquisition arrangements
• CPC Class Codes: G06V 40/171 Local features and components; Facial parts eye characteristics G06V40/18; Occluding parts, e.g. glasses; Geometrical relationships
• CPC Class Codes: G06V 40/172 Classification, e.g. identification
• CPC Class Codes: H04L 9/3231 Biological data, e.g. fingerprint, voice or retina network architectures or network communication protocols for supporting authentication of entities using biometrical features in a packet data network H04L63/0861

Abstract

An authentication system includes a processor, a non-transitory computer readable medium, and one or more programs stored on the computer readable medium, where the processor, under control of the programs implements at least one neural network trained to produce first feature vectors from facial features extracted from a population of first facial images and, after training, configured to produce a second feature vector from facial features extracted from a second facial image, a discriminative classifier trained to identify closely matching ones of the first feature vectors and configured to identify whether at least one first feature vector and the second feature vector meet a correlation threshold. The authentication system may also include an access interface configured to allow access if the correlation threshold is met.

Claims

1. An authentication system comprising:

one or more processors, a non-transitory computer readable medium, and one or more programs stored on the computer readable medium, wherein the one or more processors, under control of the programs implements;

at least one neural network trained to produce a first feature vector based on facial features extracted from a first facial image depicting a known person;

the at least one neural network configured to generate a set of augmented feature vectors associated with the known person, wherein each augmented feature vector in the set is based in part on the facial features extracted from the first facial image and also includes semantic variables corresponding to one or more external elements separate and distinct from the first facial image;

the at least one neural network, after training, configured to produce a candidate feature vector based on facial features extracted from a second facial image depicting a person to be identified; and

a discriminative classifier trained to compare the candidate feature vector to the first feature vector and the set of augmented feature vectors and to identify the person in the second facial image as the known person in the first facial image responsive to the candidate feature vector meeting a correlation threshold with the first feature vector or any of the augmented feature vectors in the set;

wherein the authentication system further comprises an access interface configured to allow the person in the second facial image access responsive to the correlation threshold being met.

2. The authentication system of claim 1, wherein the at least one neural network comprises a generative adversarial network.

3. The authentication system of claim 1, wherein the at least one neural network comprises a combination of at least one generative adversarial network and a Siamese neural network.

4. The authentication system of claim 1, wherein the discriminative classifier comprises a second neural network.

5. The authentication system of claim 1, wherein the at least one neural network is configured to utilize the set of augmented feature vectors to identify user worn equipment required for access.

6. The authentication system of claim 1, further comprising a data collection station including:

a camera for acquiring the first facial image;

a face detector for detecting a face of the known person in the first facial image; and

a feature extractor for producing the facial features extracted from the first facial image.

7. The authentication system of claim 1, further comprising:

an access control station including;

a camera for acquiring the second facial image;

a face detector for detecting a face of the person in the second facial image;

a feature extractor for producing the facial features extracted from the second facial image; and

a user interface for inputting additional credentials if the correlation threshold is not met;

the authentication system further comprising a credentials database with information for identifying a valid user for comparison to the additional credentials.

8. The authentication system of claim 1, wherein responsive to the candidate feature vector meeting the correlation threshold with a first augmented feature vector in the set of augmented feature vectors, then the one or more processors identify the person in the second facial image as the known person in the first facial image and determine that the one or more external elements associated with the semantic variables of the first augmented feature vector are present in the second facial image.

9. The authentication system of claim 1, wherein the semantic variables in a first augmented feature vector in the set correspond to at least one different external element than the semantic variables in a second augmented feature vector in the set.

10. The authentication system of claim 1, wherein the one or more external elements include user worn equipment.

11. The authentication system of claim 1, wherein the one or more external elements include one or more of glasses, face shields, goggles, beard guards or hoods.

12. An authentication method comprising:

training at least one neural network to produce a first feature vector based on facial features extracted from a first facial image depicting a known person;

using the at least one neural network to generate a set of augmented feature vectors with the known person, wherein each augmented feature vector in the set is based in part on the facial features extracted from the first facial image and also includes semantic variables corresponding to one or more external elements separate distinct from the first facial image;

using the at least one neural network to produce a candidate feature vector based on facial features extracted from a second facial image depicting a person to be identified;

training a discriminative classifier to compare the candidate feature vector to the first feature vector and the set of augmented feature vectors and to identify the person in the second facial image as the known person in the first facial image responsive to the candidate feature vector meeting a correlation threshold with the first feature vector or any of the augmented feature vectors in the set; and

allowing the person in the second facial image access responsive to the correlation threshold being met.

13. The authentication method of claim 12, wherein the at least one neural network comprises a generative adversarial network.

14. The authentication method of claim 12, wherein the at least one neural network comprises a combination of at least one generative adversarial network and a Siamese neural network.

15. The authentication method of claim 12, wherein the discriminative classifier comprises a second neural network.

16. The authentication method of claim 12, further comprising utilizing the set of augmented feature vectors to identify user worn equipment required for access.

17. The authentication method of claim 12, further comprising:

acquiring the first facial image using a camera;

detecting a face of the known person in the first facial image; and

using a feature extractor to produce the facial features extracted from the first facial image.

18. The authentication method of claim 12, further comprising:

acquiring the second facial image using a camera;

detecting a face of the person in the second facial image;

using a feature extractor to produce the facial features extracted from the second facial image;

inputting additional credentials if the correlation threshold is not met; and

allowing access if the additional credentials are valid.

19. The authentication method of claim 12, further comprising storing the first feature vector and the set of augmented feature vectors in a database, wherein the discriminative classifier accesses the database to compare the candidate feature vector to the first feature vector and the set of augmented feature vectors.

20. An authentication system comprising:

a database management system including one or more processors, a non-transitory computer readable medium, and one or more programs stored on the computer readable medium, wherein the one or more processors, under control of the programs implements;

at least one neural network trained to produce a first feature vector based on facial features extracted from a first facial image depicting a known person;

the at least one neural network configured to generate a set of augmented feature vectors associated with the known person, wherein each augmented feature vector in the set is based in part on the facial features extracted from the first facial image and also includes semantic variables corresponding to one or more external elements separate and distinct from the first facial image;

the at least one neural network, after training, configured to produce a candidate feature vector based on facial features extracted from a second facial image depicting a person to be identified; and

a discriminative classifier trained to compare the candidate feature vector to the first feature vector and the set of augmented feature vectors and to identify the person in the second facial image as the known person in the first facial image responsive to the candidate feature vector meeting a correlation threshold with the first feature vector or any of the augmented feature vectors in the set; and

an access control station including;

a camera for acquiring the second facial image;

a face detector for detecting a face of the person in the second facial image;

a feature extractor for producing the facial features extracted from the second facial image; and

an access interface, controlled by the access control station for providing the person in the second facial image access to equipment responsive to the correlation threshold being met.