Face cover-compatible biometrics and processes for generating and using same

1. A method, comprising:

receiving an uncovered face image of a subject;

generating a first fixed-size representation (FXR) based on the uncovered face image;

generating a covered face image of the subject based on the uncovered face image by at least:

generating a head pose estimation of the subject based on the uncovered face image; and fitting a 3-D virtual face covering to the uncovered face image of the subject based on the head pose estimation, wherein the covered face image of the subject comprises an artificial representation of a face covering, the face covering being a face mask;

generating a second FXR based on the covered face image; and

creating a record of the first FXR and the second FXR in association with the subject.

2. The method of claim 1 , further comprising:

receiving a second face image of a second subject;

determining a mask probability based on the second face image;

generating a third FXR based on the second face image;

retrieving the first FXR and the second FXR;

generating:

a first similarity score by comparing the third FXR to the first FXR; and

a second similarity score by comparing the third FXR to the second FXR;

calibrating the first similarity score and the second similarity score based on the mask probability to generate a third similarity score;

determining that the third similarity score satisfies a predetermined threshold; and

verifying that the second subject and the subject are the same based on the determination.

3. The method of claim 2 , further comprising providing the second subject access to a digital environment or restricted area based on the verification.

4. The method of claim 1 , further comprising:

receiving a second face image of a second subject;

determining a mask probability based on the second face image;

generating a third FXR based on the second face image;

generating a plurality of similarity scores based on comparisons between the third FXR and a plurality of paired FXRs, wherein a particular pair of the plurality of paired FXRs comprises the first FXR and the second FXR;

calibrating each of the plurality of similarity scores based on the mask probability to generate a plurality of calibrated similarity scores;

determining a top-ranked calibrated similarity score from the plurality of calibrated similarity scores, wherein the top-ranked calibrated similarity score was derived from the comparison between the third FXR and the particular pair of the plurality of paired FXRs; and

determining that the second subject and the subject are the same based on a determination that the top-ranked calibrated similarity score meets a predetermined threshold.

5. The method of claim 4 , further comprising transmitting an alert to a computing device based on the determination that the second subject and the subject are the same.

6. The method of claim 4 , wherein each of the pairs of the plurality of paired FXRs comprises a covered face FXR and an uncovered face FXR.

7. A method, comprising:

receiving an uncovered face image of a subject;

generating a first fixed-size representation (FXR) based on the uncovered face image;

generating a covered face image of the subject based on the uncovered face image by at least:

detecting, via a face detection algorithm, a face, a nose, and a mouth of the subject within the uncovered face image;

generating a cropped face image by cropping the face from the uncovered face image according to a normalized coordinate set;

aligning, via an image warping technique, the cropped face image according to the normalized coordinate set; and

fitting a 3-D virtual face covering to the cropped face image such that the 3-D virtual face covering covers the nose and the mouth of the subject;

generating a second FXR based on the covered face image; and

creating a record of the first FXR and the second FXR in association with the subject.

8. The method of claim 7 , further comprising detecting, via the face detection algorithm, a pair of eyes within the uncovered face image, wherein:

generating a covered face image of the subject further comprises generating a second cropped face image by aligning, via the image warping technique, the pair of eyes in the cropped face image according to a second normalized coordinate set; and

the step of fitting the 3-D virtual face covering is performed on the second cropped face image.

9. The method of claim 8 , wherein the second normalized coordinate set comprises a left eye center coordinate and a right eye center coordinate.

10. The method of claim 9 , wherein the second normalized coordinate set comprises a nose coordinate.

11. The method of claim 10 , wherein the second normalized coordinate set comprises a first corner mouth coordinate and a second corner mouth coordinate.

12. A method for biometric verification, comprising:

receiving an image of a subject, wherein the image comprises a face;

determining a mask probability based on the image;

generating a first FXR based on the image;

retrieving a second FXR and a third FXR;

generating:

a first similarity score by comparing the first FXR and the second FXR; and

a second similarity score by comparing the first FXR and the third FXR;

calibrating the first similarity score and the second similarity score based on the mask probability to generate a third similarity score, wherein calibrating the first similarity score and the second similarity score comprises computing a log-likelihood ratio, and wherein the third similarity score comprises the log-likelihood ratio;

determining that the third similarity score satisfies a predetermined threshold; and

verifying that the second subject and the subject are the same based on the determination.

13. The method of claim 12 , further comprising:

detecting the face in the image; and

generating a cropped facial image of the face, wherein:

determining the mask probability comprises applying a trained machine learning model to the cropped facial image to generate a posterior probability that the cropped facial image includes a face covering over the face; and

the mask probability comprises the posterior probability.

14. The method of claim 13 , wherein the trained machine learning model is a convolutional neural network.

15. The method of claim 14 , wherein:

the convolutional neural network was trained using a training dataset comprising a plurality of cropped facial images; and

the plurality of cropped facial images comprises:

a first subset comprising cropped covered facial images; and

a second subset comprising cropped uncovered facial images, wherein the second subset excludes the first subset.

16. The method of claim 12 , wherein:

calibrating the first similarity score and the second similarity score comprises computing a direct posterior probability estimation; and

the third similarity score comprises the direct posterior probability estimation.

17. A system for biometric verification, comprising a server having a processor configured to:

receive an image of a subject, wherein the image comprises a face;

determine a mask probability based on the image;

generate a first FXR based on the image;

retrieve a second FXR and a third FXR;

generate:

a first similarity score by comparing the first FXR and the second FXR; and

a second similarity score by comparing the first FXR and the third FXR;

calibrate the first similarity score and the second similarity score based on the mask probability to generate a third similarity score, wherein calibrating the first similarity score and the second similarity score comprises computing a log-likelihood ratio, and wherein the third similarity score comprises the log-likelihood ratio;

determine that the third similarity score satisfies a predetermined threshold; and

verify that the second subject and the subject are the same based on the determination.

18. The system of claim 17 , wherein the processor is further configured to:

detect the face in the image; and

generate a cropped facial image of the face, wherein:

determining the mask probability comprises applying a trained machine learning model to the cropped facial image to generate a posterior probability that the cropped facial image includes a face covering over the face; and

the mask probability comprises the posterior probability.

19. The system of claim 18 , wherein the trained machine learning model is a convolutional neural network.

20. The system of claim 19 , wherein:

the convolutional neural network was trained using a training dataset comprising a plurality of cropped facial images; and

the plurality of cropped facial images comprises:

a first subset comprising cropped covered facial images; and

a second subset comprising cropped uncovered facial images, wherein the second subset excludes the first subset.