Systems and methods for using machine learning for image-based spoof detection

1. A method comprising:

obtaining an input-data set comprising a plurality of images captured of a biometric-authentication subject by a camera system, the plurality of images comprising a first image captured by a first camera in the camera system and a second image captured by a second camera in the camera system;

inputting the input-data set into a trained machine-learning module, wherein the machine-learning module comprises a plurality of neural networks, the plurality of neural networks comprising at least one data-analysis network and a conclusion network, the at least one data-analysis network comprising a first image-processing network and a second image-processing network;

processing the input-data set using the machine-learning module to obtain, from the machine-learning module, a spoof-detection result for the biometric-authentication subject the processing comprising:

processing the first image using the first image-processing network, wherein the first image-processing network thereafter comprises a first plurality of post-processing nodes, and wherein the first image-processing network is configured to generate a first-image-processing-network spoof-detection result based on the first image;

processing the second image using the second image-processing network, wherein the second image-processing network thereafter comprises a second plurality of post-processing nodes, and wherein the second image-processing network is configured to generate a second-image-processing-network spoof-detection result based on the second image;

inputting the first and second pluralities of post-processing nodes into the conclusion network;

processing at least the first and second pluralities of post-processing nodes using the conclusion network to obtain, from the conclusion network, a conclusion-network spoof-detection result; and

obtaining the spoof-detection result for the biometric-authentication subject based:

at least in part on the first-image-processing-network spoof-detection result,

at least in part on the second-image processing-network spoof-detection result, and

at least in part on the conclusion-network spoof-detection result; and

outputting the spoof-detection result for the biometric-authentication subject.

2. The method of claim 1 , wherein the first camera and second camera are both configured to capture images in a first light spectrum.

3. The method of claim 1 , wherein:

the first camera is configured to capture images in a first light spectrum; and

the second camera is configured to capture images in a second light spectrum, the second light spectrum being different than the first light spectrum.

4. The method of claim 3 , wherein the first light spectrum and the second light spectrum overlap.

5. The method of claim 3 , wherein the first spectrum and the second light spectrum do not overlap.

6. The method of claim 1 , wherein:

the camera system comprises at least one illumination source configured to emit structured light; and

the plurality of images in the input-data set comprises at least one image captured with the biometric-authentication subject under illumination of structured light from the at least one illumination source.

7. The method of claim 1 , wherein:

the first image is captured under a first illumination condition; and

the second image is captured under a second illumination condition, the second illumination condition being different than the first illumination condition.

8. The method of claim 1 , wherein:

the camera system comprises at least one illumination source configured to emit light having a first polarization orientation;

the first camera is configured to capture images using a second polarization orientation, the second polarization orientation being different than the first polarization orientation; and

the plurality of images in the input-data set comprises at least one image captured by the first camera of the biometric-authentication subject using the second polarization orientation with the biometric-authentication subject under illumination of light having the first polarization orientation from the at least one illumination source.

9. The method of claim 1 , wherein:

the first image is captured using a first focal configuration; and

the second image is captured using a second focal configuration, the second focal configuration being different than the first focal configuration.

10. The method of claim 1 , wherein:

the first camera is a right-side camera configured to capture a right-camera view of the biometric-authentication subject;

the first image comprises a right-camera image of the biometric-authentication subject captured by the right-side camera;

the second camera is a left-side camera configured to capture a left-camera view of the biometric-authentication subject; and

the second image comprises a left-camera image of the biometric-authentication subject captured by the left-side camera.

11. A system comprising:

a trained machine-learning module;

a processor; and

data storage containing instructions executable by the processor for causing the system to carry out a set of functions, wherein the set of functions comprises:

obtaining an input-data set comprising a plurality of images captured of a biometric-authentication subject by a camera system, the camera system comprising a right-side camera configured to capture a right-camera view of the biometric-authentication subject and a left-side camera configured to capture a left-camera view of the biometric-authentication subject, and the plurality of images comprising a right-camera image of the biometric-authentication subject captured by the right-side camera and a left-camera image of the biometric-authentication subject captured by the left-side camera;

inputting the input-data set into the trained machine-learning module, wherein the machine-learning module comprises a plurality of neural networks, the plurality of neural networks comprising at least one data-analysis network and a conclusion network, the at least one data-analysis network comprising a first image-processing network, a second image-processing network, and a disparity-data-processing network;

generating disparity data based on the right-camera image and the left-camera image;

processing the input-data set and disparity data using the machine-learning module to obtain, from the machine-learning module, a spoof-detection result for the biometric-authentication subject, the processing comprising:

processing the right-camera image using the first image-processing network, wherein the first image-processing network thereafter comprises a first plurality of post-processing nodes;

processing the left-camera image using the second image-processing network, wherein the second image-processing network thereafter comprises a second plurality of post-processing nodes;

processing the disparity data using the disparity-data-processing network, wherein the disparity-data-processing network thereafter comprises a third plurality of post-processing nodes;

inputting the first, second, and third pluralities of post-processing nodes into the conclusion network;

processing at least the first, second, and third pluralities of post-processing nodes using the conclusion network to obtain, from the conclusion network, the conclusion-network spoof-detection result; and

obtaining the spoof-detection result for the biometric-authentication subject based at least in part on the conclusion-network spoof-detection result; and

outputting the spoof-detection result for the biometric-authentication subject.

12. The method of claim 11 , wherein:

the first image-processing network is configured to generate a first-image-processing-network spoof-detection result based on the right-camera image;

the second image-processing network is configured to generate a second-image-processing-network spoof-detection result based on the left-camera image;

the disparity-data-processing network is configured to generate a disparity-data-processing-network spoof-detection result based on the disparity data; and

obtaining the spoof-detection result for the biometric-authentication subject based at least in part on the conclusion-network spoof-detection result comprises:

obtaining the spoof-detection result for the biometric-authentication subject based:

at least in part on the first-image-processing-network spoof-detection result,

at least in part on the second-image-processing-network spoof-detection result,

at least in part on the disparity-data-processing-network spoof-detection result, and

at least in part on the conclusion-network spoof-detection result.

13. The method of claim 11 , wherein the first image-processing network, the second image-processing network, and the disparity-data-processing network are trained independently of one another.

14. A non-transitory computer-readable medium containing instructions that, when executed by a processor, cause a computer system to carry out a set of functions, wherein the set of functions comprises:

obtaining an input-data set comprising a plurality of images captured of a biometric-authentication subject by a camera system, the plurality of images comprising a first image captured by a first camera in the camera system and a second image captured by a second camera in the camera system;

inputting the input-data set into a trained machine-learning module, wherein the machine-learning module comprises a plurality of neural networks, the plurality of neural networks comprising at least one data-analysis network and a conclusion network, the at least one data-analysis network comprising a first image-processing network and a second image-processing network;

processing the input-data set using the machine-learning module to obtain, from the machine-learning module, a spoof-detection result for the biometric-authentication subject, the processing comprising:

processing the first image using the first image-processing network, wherein the first image-processing network thereafter comprises a first plurality of post-processing nodes, and wherein the first image-processing network is configured to generate a first-image-processing-network spoof-detection result based on the first image;

processing the second image using the second image-processing network, wherein the second image-processing network thereafter comprises a second plurality of post-processing nodes, and wherein the second image-processing network is configured to generate a second-image-processing-network spoof-detection result based on the second image;

inputting the first and second pluralities of post-processing nodes into the conclusion network;

processing at least the first and second pluralities of post-processing nodes using the conclusion network to obtain, from the conclusion network, a conclusion-network spoof-detection result; and

obtaining the spoof-detection result for the biometric-authentication subject based:

at least in part on the first-image-processing-network spoof-detection result,

at least in part on the second-image processing-network spoof-detection result, and

at least in part on the conclusion-network spoof-detection result; and

outputting the spoof-detection result for the biometric-authentication subject.