Spoof detection by correlating images captured using front and back cameras of a mobile device

1. A computer-implemented method comprising:

obtaining a first image captured using a first camera disposed on a first side of a mobile device, the first image including representation of reflections visible on corneas of a subject;

obtaining a second image captured using a second camera disposed on a second side of the mobile device, the second side being on the opposite side of the first side;

pre-processing the first image and the second image to generate a third image and a fourth image, respectively, wherein a first field of view represented in the third image at least partially overlaps with a second field of view of the fourth image;

determining, based on the third and fourth images, that a scene represented in the first field of view is substantially same as a scene represented in the second field of view; and

in response to determining that the scene represented in the first field of view is substantially same as the scene represented in the second field of view, identifying the subject as a live person.

2. The computer-implemented method of claim 1 , wherein pre-processing the first image comprises:

extracting a portion of the first image that includes the reflections visible on the corneas; and

performing distortion corrections to compensate for catadioptric reflection properties of the corneas.

3. The computer-implemented method of claim 1 , wherein pre-processing the first image and the second image comprises scaling at least one of the first image or the second image.

4. The computer-implemented method of claim 1 , wherein pre-processing the first image comprises removing the reflection of the mobile device from the representation of reflections visible on the corneas of a subject.

5. The computer-implemented method of claim 1 , wherein determining that the scene represented in the first field of view is substantially same as a scene represented in the second field of view comprises:

providing the third image and the fourth image as an input to a machine learning model trained using a corpus of image pairs wherein,

one image in each pair is an image captured using a back camera of a mobile device, and

another image in the same pair is an image captured using the front camera of the same mobile device of either the live person or a spoof representation of the live person; and

determining, based on an output of the machine learning model in response to the input, that the scene represented in the first field of view is substantially same as a scene represented in the second field of view.

6. The computer-implemented method of claim 1 , further comprising initiating an authentication process for the subject in response to identifying the subject as the live person.

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

obtaining a fifth image captured using the first camera, the fifth image including representation of reflections visible on the corneas of another subject;

obtaining a sixth image captured using the second camera;

pre-processing the fifth image and the sixth image to generate a seventh image and an eighth image, respectively, wherein a third field of view represented in the seventh image at least partially overlaps with a fourth field of view of the eighth image;

determining, based on the fifth and sixth images, that a difference between a scene represented in the third field of view and a scene represented in the fourth field of view satisfies a threshold condition; and

in response to determining that the difference between the scene represented in the third field of view and the scene represented in the fourth field of view satisfies the threshold condition, identifying the another subject as a spoof representation of the live person.

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

obtaining a first image captured using a first camera disposed on a first side of a mobile device, the first image including representation of reflections visible on corneas of a subject;

obtaining a second image captured using a second camera disposed on a second side of the mobile device, the second side being on the opposite side of the first side;

pre-processing the first image and the second image to generate a third image and a fourth image, respectively, wherein a first field of view represented in the third image at least partially overlaps with a second field of view of the fourth image;

determining, based on the third and fourth images, that a scene represented in the first field of view is substantially same as a scene represented in the second field of view; and

in response to determining that the scene represented in the first field of view is substantially same as the scene represented in the second field of view, identifying the subject as a live person.

9. The non-transitory, computer-readable medium of claim 8 , wherein pre-processing the first image comprises:

extracting a portion of the first image that includes the reflections visible on the corneas; and

performing distortion corrections to compensate for catadioptric reflection properties of the corneas.

10. The non-transitory, computer-readable medium of claim 8 , wherein pre-processing the first image and the second image comprises scaling at least one of the first image or the second image.

11. The non-transitory, computer-readable medium of claim 8 , wherein pre-processing the first image comprises removing the reflection of the mobile device from the representation of reflections visible on the corneas of a subject.

12. The non-transitory, computer-readable medium of claim 8 , wherein determining that the scene represented in the first field of view is substantially same as a scene represented in the second field of view comprises:

providing the third image and the fourth image as an input to a machine learning model trained using a corpus of image pairs wherein,

one image in each pair is an image captured using a back camera of a mobile device, and

another image in the same pair is an image captured using the front camera of the same mobile device of either the live person or a spoof representation of the live person; and

determining, based on an output of the machine learning model in response to the input, that the scene represented in the first field of view is substantially same as a scene represented in the second field of view.

13. The non-transitory, computer-readable medium of claim 8 , further comprising initiating an authentication process for the subject in response to identifying the subject as the live person.

14. The non-transitory, computer-readable medium of claim 8 , further comprising:

obtaining a fifth image captured using the first camera, the fifth image including representation of reflections visible on the corneas of another subject;

obtaining a sixth image captured using the second camera;

pre-processing the fifth image and the sixth image to generate a seventh image and an eighth image, respectively, wherein a third field of view represented in the seventh image at least partially overlaps with a fourth field of view of the eighth image;

determining, based on the fifth and sixth images, that a difference between a scene represented in the third field of view and a scene represented in the fourth field of view satisfies a threshold condition; and

in response to determining that the difference between the scene represented in the third field of view and the scene represented in the fourth field of view satisfies the threshold condition, identifying the another subject as a spoof representation of the live person.

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 operations comprising:

obtaining a first image captured using a first camera disposed on a first side of a mobile device, the first image including representation of reflections visible on corneas of a subject,

obtaining a second image captured using a second camera disposed on a second side of the mobile device, the second side being on the opposite side of the first side,

pre-processing the first image and the second image to generate a third image and a fourth image, respectively, wherein a first field of view represented in the third image at least partially overlaps with a second field of view of the fourth image,

determining, based on the third and fourth images, that a scene represented in the first field of view is substantially same as a scene represented in the second field of view, and

in response to determining that the scene represented in the first field of view is substantially same as the scene represented in the second field of view, identifying the subject as a live person.

16. The computer-implemented system of claim 15 , wherein pre-processing the first image comprises:

extracting a portion of the first image that includes the reflections visible on the corneas; and

performing distortion corrections to compensate for catadioptric reflection properties of the corneas.

17. The computer-implemented system of claim 15 , wherein pre-processing the first image and the second image comprises scaling at least one of the first image or the second image.

18. The computer-implemented system of claim 15 , wherein pre-processing the first image comprises removing the reflection of the mobile device from the representation of reflections visible on the corneas of a subject.

19. The computer-implemented system of claim 15 , wherein determining that the scene represented in the first field of view is substantially same as a scene represented in the second field of view comprises:

providing the third image and the fourth image as an input to a machine learning model trained using a corpus of image pairs wherein,

one image in each pair is an image captured using a back camera of a mobile device, and

another image in the same pair is an image captured using the front camera of the same mobile device of either the live person or a spoof representation of the live person; and

determining, based on an output of the machine learning model in response to the input, that the scene represented in the first field of view is substantially same as a scene represented in the second field of view.

20. The computer-implemented system of claim 15 , further comprising:

obtaining a fifth image captured using the first camera, the fifth image including representation of reflections visible on the corneas of another subject;

obtaining a sixth image captured using the second camera;

pre-processing the fifth image and the sixth image to generate a seventh image and an eighth image, respectively, wherein a third field of view represented in the seventh image at least partially overlaps with a fourth field of view of the eighth image;

determining, based on the fifth and sixth images, that a difference between a scene represented in the third field of view and a scene represented in the fourth field of view satisfies a threshold condition;

in response to determining that the difference between the scene represented in the third field of view and the scene represented in the fourth field of view satisfies the threshold condition, identifying the another subject as a spoof representation of the live person; and initiating an authentication process for the subject in response to identifying the subject as the live person.