Determining user authenticity with face liveness detection

1. A computer-implemented method, comprising:

training a first deep learning model to classify general facial images, wherein the general facial images comprise non-facial regions, and wherein the first deep learning model is trained to extract image structure features of non-facial regions of the general facial images;

extracting cropped facial images from the general facial images;

training a second deep learning model based on the cropped facial images; and

performing face liveness detection based on the trained first deep learning model and the trained second deep learning model.

2. The computer-implemented method of claim 1 , wherein the general facial images are classified into at least live facial images and non-live facial images, wherein the live facial images are positive samples and the non-live facial images are negative samples.

3. The computer-implemented method of claim 2 , wherein the first deep learning model is a classification model and the general facial images are used as inputs of the first deep learning model, and wherein training the first deep learning model improves classification accuracy with respect to the general facial images.

4. The computer-implemented method of claim 1 , wherein a particular general facial image comprises a first label indicating whether the particular general facial image corresponding to the first label is a live facial image, and wherein the training of the first deep learning model comprises:

inputting the particular general facial image to the first deep learning model to generate a first prediction result, based on an image structure feature of the particular general facial image, of whether the particular general facial image is a live facial image; and

adjusting the first deep learning model based on the first prediction result and the first label.

5. The computer-implemented method of claim 1 , wherein a particular cropped facial image comprises a second label, and the second label indicates whether the particular cropped facial image corresponding to the second label is a live facial image and the training of the second deep learning model based on the cropped facial images comprises:

obtaining the particular cropped facial image;

inputting the particular cropped facial image to the second deep learning model to generate a second prediction result, based on an image material feature of the particular cropped facial image, of whether particular cropped facial image is a live facial image; and

adjusting the second deep learning model based on the second prediction result and the second label.

6. The computer-implemented method of claim 1 , wherein the first deep learning model and the second deep learning model are based on a convolutional neural network, and wherein a convolution kernel of a convolution unit in a convolutional neural network of the first deep learning model is greater than a convolution kernel of a convolution unit in a convolutional neural network of the second deep learning model.

7. The computer-implemented method of claim 1 , wherein the face liveness detection comprises:

obtaining a general facial image;

inputting the general facial image into the trained first deep learning model to obtain corresponding first prediction data;

obtaining a cropped facial image from the general facial image;

inputting the cropped facial image into the trained second deep learning model to obtain corresponding second prediction data; and

making a joint decision based on the first prediction data and the second prediction data to obtain a face liveness detection result.

8. A non-transitory, computer-readable medium storing one or more instructions executable by a computer system to perform operations comprising:

training a first deep learning model to classify general facial images, wherein the general facial images comprise non-facial regions, and wherein the first deep learning model is trained to extract image structure features of non-facial regions of the general facial images;

extracting cropped facial images from the general facial images;

training a second deep learning model based on the cropped facial images; and

performing face liveness detection based on the trained first deep learning model and the trained second deep learning model.

9. The non-transitory, computer-readable medium of claim 8 , wherein the general facial images are classified into at least live facial images and non-live facial images, wherein the live facial images are positive samples and the non-live facial images are negative samples.

10. The non-transitory, computer-readable medium of claim 9 , wherein the first deep learning model is a classification model and the general facial images are used as inputs of the first deep learning model, and wherein training the first deep learning model improves classification accuracy with respect to the general facial images.

11. The non-transitory, computer-readable medium of claim 8 , wherein a particular general facial image comprises a first label indicating whether the particular general facial image corresponding to the first label is a live facial image, and wherein the training of the first deep learning model comprises:

inputting the particular general facial image to the first deep learning model to generate a first prediction result, based on an image structure feature of the particular general facial image, of whether the particular general facial image is a live facial image; and

adjusting the first deep learning model based on the first prediction result and the first label.

12. The non-transitory, computer-readable medium of claim 8 , wherein a particular cropped facial image comprises a second label, and the second label indicates whether the particular cropped facial image corresponding to the second label is a live facial image and the training of the second deep learning model based on the cropped facial images comprises:

obtaining the particular cropped facial image;

inputting the particular cropped facial image to the second deep learning model to generate a second prediction result, based on an image material feature of the particular cropped facial image, of whether particular cropped facial image is a live facial image; and

adjusting the second deep learning model based on the second prediction result and the second label.

13. The non-transitory, computer-readable medium of claim 8 , wherein the first deep learning model and the second deep learning model are based on a convolutional neural network, and wherein a convolution kernel of a convolution unit in a convolutional neural network of the first deep learning model is greater than a convolution kernel of a convolution unit in a convolutional neural network of the second deep learning model.

14. The non-transitory, computer-readable medium of claim 8 , wherein the face liveness detection comprises:

obtaining a general facial image;

inputting the general facial image into the trained first deep learning model to obtain corresponding first prediction data;

obtaining a cropped facial image from the general facial image;

inputting the cropped facial image into the trained second deep learning model to obtain corresponding second prediction data; and

making a joint decision based on the first prediction data and the second prediction data to obtain a face liveness detection result.

15. A computer-implemented system, 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, perform one or more operations comprising:

training a first deep learning model to classify general facial images, wherein the general facial images comprise non-facial regions, and wherein the first deep learning model is trained to extract image structure features of non-facial regions of the general facial images;

extracting cropped facial images from the general facial images;

training a second deep learning model based on the cropped facial images; and

performing face liveness detection based on the trained first deep learning model and the trained second deep learning model.

16. The computer-implemented system of claim 15 , wherein the general facial images are classified into at least live facial images and non-live facial images, wherein the live facial images are positive samples and the non-live facial images are negative samples.

17. The computer-implemented system of claim 16 , wherein the first deep learning model is a classification model and the general facial images are used as inputs of the first deep learning model, and wherein training the first deep learning model improves classification accuracy with respect to the general facial images.

18. The computer-implemented system of claim 15 , wherein a particular general facial image comprises a first label indicating whether the particular general facial image corresponding to the first label is a live facial image, and wherein the training of the first deep learning model comprises:

inputting the particular general facial image to the first deep learning model to generate a first prediction result, based on an image structure feature of the particular general facial image, of whether the particular general facial image is a live facial image; and

adjusting the first deep learning model based on the first prediction result and the first label.

19. The computer-implemented system of claim 15 , wherein a particular cropped facial image comprises a second label, and the second label indicates whether the particular cropped facial image corresponding to the second label is a live facial image and the training of the second deep learning model based on the cropped facial images comprises:

obtaining the particular cropped facial image;

inputting the particular cropped facial image to the second deep learning model to generate a second prediction result, based on an image material feature of the particular cropped facial image, of whether particular cropped facial image is a live facial image; and

adjusting the second deep learning model based on the second prediction result and the second label.

20. The computer-implemented system of claim 15 , wherein the first deep learning model and the second deep learning model are based on a convolutional neural network, and wherein a convolution kernel of a convolution unit in a convolutional neural network of the first deep learning model is greater than a convolution kernel of a convolution unit in a convolutional neural network of the second deep learning model.