Evaluating the security of a facial recognition system using light projections
Embodiments of the invention are directed to systems, methods, and devices for testing the security of a facial recognition system (FRS). A target image depicting a target person enrolled in the FRS and a tester image depicting a tester may be obtained. A plurality of transformed images may be generated from an image of the target person or the tester image. A processed tester image (e.g., one that is likely to cause the FRS to misclassify) may be identified using the plurality of transformed images, the tester image, and the target image. Data representing a light pattern can be generated using the processed tester image and the light pattern can be projected onto a second person. Another image may be captured of the second person with the light pattern as projected. This image may be provided to the FRS and a remedial action may be performed based on the corresponding output.
1. A method for determining accuracy of a facial recognition system, comprising:
obtaining, by a computing device, a target image depicting a target person enrolled in the facial recognition system and a tester image depicting a tester;
generating, by the computing device, a plurality of transformed images based at least in part on applying, by the computing device, a plurality of transformations to at least one of: the tester image or the target image;
generating, by the computing device, a processed tester image based at least in part on the plurality of transformed images, the tester image, and the target image;
generating, by the computing device, data representing a light pattern based at least in part on the processed tester image;
providing, by the computing device, the data representing the light pattern to a projection device, the projection device being configured to project the light pattern onto the tester;
receiving, by the computing device, a subsequent tester image, the subsequent tester image depicting a second person with the light pattern projected on the second person;
providing, by the computing device, the subsequent tester image as input to the facial recognition system; and
performing, by the computing device, one or more remedial actions based at least in part on output generated by the facial recognition system.
2. The method of claim 1 , wherein the plurality of transformations are sampled from a set of predefined transformations.
3. The method of claim 1 , wherein the method further comprises generating a representational image based at least in part on the plurality of transformed images.
4. The method of claim 3 , wherein identifying the processed tester image comprises:
generating a masked tester image based at least in part on the tester image and a binary mask; and
generating a masked representational image based at least in part on the representational image and the binary mask.
5. The method of claim 4 , wherein identifying the processed tester image comprises:
calculating a modified masked tester image and a modified masked representational image is based at least in part on a matrix of random noise values; and
storing modifications to the masked tester image as an image.
6. The method of claim 5 , wherein identifying the processed tester image further comprises:
obtaining first output based at least in part on providing the modified masked tester image as input to a neural network function, wherein the first output identifies the modified masked tester image as being of a first subject; and
obtaining second output based at least in part on providing the modified masked representational image as input to the neural network function, wherein the second output identifies the modified masked representational image as being of a second subject.
7. The method of claim 6 , wherein identifying the processed tester image further comprises:
calculating a first distance between the first output and a target output corresponding to the target image; and
calculating a second distance between the second output and the target output corresponding to the target image.
8. The method of claim 7 , wherein identifying the processed tester image further comprises generating a combined distance based at least in part on the first distance, the second distance, and a fusion function.
9. The method of claim 8 , wherein identifying the processed tester image further comprises calculating a gradient based at least in part on the combined distance and a noise image.
10. The method of claim 9 , wherein identifying the processed tester image further comprises;
determining the gradient is less than a predetermined threshold value; and
calculating the processed tester image based at least in part on the masked tester image as most recently modified.
11. A computing device, comprising:
one or more processors;
a memory comprising computer-executable instructions that, when executed by the one or more processors, causes the computing device to:
obtain a target image depicting a target person enrolled in a facial recognition system and a tester image depicting a tester;
generate a plurality of transformed images based at least in part on applying a plurality of transformations to at least one of: the tester image or the target image;
generate a processed tester image based at least in part on the plurality of transformed images, the tester image, and the target image;
generate data representing a light pattern based at least in part on the processed tester image;
provide the data representing the light pattern to a projection device, the projection device being configured to project the light pattern onto the tester;
receive a subsequent tester image, the subsequent tester image depicting a second person with the light pattern projected on the second person;
provide the subsequent tester image as input to the facial recognition system; and
perform one or more remedial actions based at least in part on output generated by the facial recognition system.
12. The computing device of claim 11 , wherein the output generated by the facial recognition system indicates the subsequent tester image was incorrectly identified as being an image of the target person enrolled in the facial recognition system.
13. The computing device of claim 11 , wherein providing the subsequent tester image as input to the facial recognition system comprises:
calculating a transformation function for modifying the subsequent tester image based at least in part on the light pattern, the tester image, a first function associated with camera capture, a second function associated with projection by the projecting device, and a regression technique; and
applying the transformation function to the subsequent tester image prior to providing the subsequent tester image as input to the facial recognition system.
14. The computing device of claim 11 , wherein the plurality of transformed images comprise a plurality of transformed target images and wherein the computing device is further configured to generate a representational target image based at least in part on the target image and the plurality of transformed target images.
15. The computing device of claim 14 , wherein identifying the processed tester image is further based at least in part on the representational target image.
16. The computing device of claim 11 , wherein the processed tester image is identified based at least in part on minimizing a difference between the tester image with the light pattern projected thereon and the target image.
17. The computing device of claim 11 , wherein the processed tester image is identified based at least in part on maximizing a difference between the tester image with the light pattern projected thereon and the target image.
18. The computing device of claim 17 , wherein the computing device is further configured to modify the data representing the light pattern according to a set of predefined parameters associated with the projection device.
19. The computing device of claim 18 , wherein the tester image depicts a first face of the tester and the target image depicts a second face of the target person.
20. The computing device of claim 11 , further comprising modifying colors of the data representing the light pattern to adjust for at least one of: projection or eventual camera capture.