Methods and devices for face anti-spoofing

1. A computer-implemented method for face anti-spoofing, comprising:

receiving, by a server and from a user device, a plurality of images of a user face, wherein each of the plurality of images corresponds to an end expression of a facial expression performed by a user in response to each of a plurality of interaction challenges, wherein the facial expression comprises at least a start expression and the corresponding end expression;

concatenating the plurality of images into a concatenated image;

processing the concatenated image using a convolutional neural network (CNN) classifier to obtain a confidence score for the concatenated image; and

determining liveness of the user face based on a comparison of the confidence score and a threshold value.

2. The computer-implemented method of claim 1 , wherein each interaction challenge of the plurality of interaction challenges is randomly selected from an interaction challenge library.

3. The computer-implemented method of claim 2 , further comprising:

building up the interaction challenge library.

4. The computer-implemented method of claim 2 , wherein the interaction challenge library comprises interaction challenges of at least eye blink, mouth open, mouth close, head pitch down, head pitch up, head roll left, head roll right, head yaw left, and head yaw right.

5. The computer-implemented method of claim 2 , further comprising:

randomly selecting three interaction challenges from the interaction challenge library to form the plurality of interaction challenges.

6. The computer-implemented method of claim 1 , further comprising:

training the CNN classifier with a training dataset having a size of N×V×H×W, wherein N is a number of images to be trained in one iteration, V indicates that ⅓(V) images are concatenated in each of the images that are used as input in the training dataset, H is a height of each of the images, and W is a width of each of the images.

7. A non-transitory computer-readable medium storing one or more instructions executable by a computer system to perform operations for face anti-spoofing, comprising:

receiving, by a server and from a user device, a plurality of images of a user face, wherein each of the plurality of images corresponds to an end expression of a facial expression performed by a user in response to each of a plurality of interaction challenges, wherein the facial expression comprises at least a start expression and the corresponding end expression;

concatenating the plurality of images into a concatenated image;

processing the concatenated image using a convolutional neural network (CNN) classifier to obtain a confidence score for the concatenated image; and

determining liveness of the user face based on a comparison of the confidence score and a threshold value.

8. The non-transitory computer-readable medium of claim 7 , wherein each interaction challenge of the plurality of interaction challenges is randomly selected from an interaction challenge library.

9. The non-transitory computer-readable medium of claim 8 , further comprising operations for:

building up the interaction challenge library.

10. The non-transitory computer-readable medium of claim 8 , wherein the interaction challenge library comprises interaction challenges of at least eye blink, mouth open, mouth close, head pitch down, head pitch up, head roll left, head roll right, head yaw left, and head yaw right.

11. The non-transitory computer-readable medium of claim 8 , further comprising operations for:

randomly selecting three interaction challenges from the interaction challenge library to form the plurality of interaction challenges.

12. The non-transitory computer-readable medium of claim 7 , further comprising operations for:

training the CNN classifier with a training dataset having a size of N×V×H×W, wherein N is a number of images to be trained in one iteration, V indicates that ⅓(V) images are concatenated in each of the images that are used as input in the training dataset, H is a height of each of the images, and W is a width of each of the images.

13. A computer-implemented system for face anti-spoofing, comprising:

one or more computers; and

one or more computer memory devices interoperably coupled with the one or more computers and having tangible, non-transitory, machine-readable media storing one or more instructions that, when executed by the one or more computers, cause the one or more computers to perform one or more operations comprising:

receiving, by a server and from a user device, a plurality of images of a user face, wherein each of the plurality of images corresponds to an end expression of a facial expression performed by a user in response to each of a plurality of interaction challenges, wherein the facial expression comprises at least a start expression and the corresponding end expression;

concatenating the plurality of images into a concatenated image;

processing the concatenated image using a convolutional neural network (CNN) classifier to obtain a confidence score for the concatenated image; and

determining liveness of the user face based on a comparison of the confidence score and a threshold value.

14. The computer-implemented system of claim 13 , wherein each interaction challenge of the plurality of interaction challenges is randomly selected from an interaction challenge library.

15. The computer-implemented system of claim 14 , further comprising operations for:

building up the interaction challenge library, wherein the interaction challenge library comprises interaction challenges of at least eye blink, mouth open, mouth close, head pitch down, head pitch up, head roll left, head roll right, head yaw left, and head yaw right.

16. The computer-implemented system of claim 14 , further comprising operations for:

randomly selecting three interaction challenges from the interaction challenge library to form the plurality of interaction challenges.

17. The computer-implemented system of claim 13 , further comprising operations for:

training the CNN classifier with a training dataset having a size of N×V×H×W, wherein N is a number of images to be trained in one iteration, V indicates that ⅓(V) images are concatenated in each of the images that are used as input in the training dataset, H is a height of each of the images, and W is a width of each of the images.