Privacy preserving face-based authentication
Example systems and methods for biometric authentication that can bridge fuzzy extractors with deep learning and achieve the goals of preserving privacy and providing recoverability from zero are disclosed. Embeddings comprising a face or speaker embedding in a non-Hamming distance space can be processed to create a personal reliable bit map and a reliable locality-sensitive hash (LSH) for mapping the non-Hamming distance space to a Hamming distance space. A fuzzy extractor can be applied to create metadata that can be stored on a computing device. A secret can be recovered from the metadata and can be used for identification.
1. A system for biometric-based user identification, comprising:
a client device comprising a processor; and
an authentication application executed by the processor, wherein the application causes the client device to at least:
capture biometric data from the client device for a generation of an embedding;
generate a locality-sensitive hash (LSH) from the embedding;
determine that a set of measurements of the LSH meet a distance threshold;
generate a personal mask to identify a plurality of reliable bits based at least in part on the set of measurements of the locality sensitive hash meeting the threshold;
generate a secret that includes a cryptographic key, the cryptographic key being configured for authentication of the client device with an authentication service;
generate a masked locality-sensitive hash using a reusable fuzzy extractor based at least in part on the secret; and
send metadata to the authorization service, the metadata comprising the masked locality-sensitive hash.
2. The system of claim 1 , wherein the biometric data is captured using at least one of a camera or a microphone of the client device.
3. The system of claim 1 , wherein the embedding is in a non-hamming distance space and the embedding is generated based at least in part on a plurality of facial scans of a user.
4. The system of claim 1 , wherein the secret is generated using a key generator executed by the client device.
5. The system of claim 1 , wherein generating the masked locality-sensitive hash further comprises generating a set of lockers that include the secret.
6. The system of claim 5 , wherein sending the metadata to the authorization service further comprising sending the set of lockers and the personal mask to the authentication service.
7. The system of claim 1 , wherein determining that the set of measurements of the locality sensitive hash meet the distance threshold further causes the client device to at least:
eliminate a noisy LSH bit that shows a variation across at least two measurements of the set of measurements.
8. A method for biometric-based user identification, comprising:
capturing, by a client device, biometric data for a generation of an embedding;
generating, by the client device, a locality-sensitive hash (LSH) from the embedding;
determine that a set of measurements of the locality sensitive hash meet a distance threshold;
generate a personal mask to identify a plurality of reliable bits based at least in part on the set of measurements of the locality sensitive hash meeting the threshold;
generate a secret that includes a cryptographic key, the cryptographic key being configured for authentication of the client device with an authentication service;
generate a masked locality-sensitive hash using a reusable fuzzy extractor based at least in part on the secret; and
send metadata to the authorization service, the metadata comprising the masked locality-sensitive hash.
9. The method of claim 8 , wherein the biometric data is captured using at least one of a camera or a microphone of the client device.
10. The method of claim 8 , wherein the embedding is in a non-hamming distance space.
11. The method of claim 8 , wherein the secret is generated using a key generator executed by the client device.
12. The method of claim 8 , wherein generating the masked locality-sensitive hash further comprises generating a set of lockers that include the secret.
13. The method of claim 12 , wherein sending the metadata to the authorization service further comprising sending the set of lockers and the personal mask to the authentication service.
14. The method of claim 8 , wherein determining that the set of measurements of the locality sensitive hash meet the distance threshold further causes the client device to at least:
eliminate a noisy LSH bit that shows a variation across at least two measurements of the set of measurements.
15. A non-transitory computer-readable medium embodying a program executable for biometric-based user identification on a client device, wherein the program, when executed, causes the client device to at least:
capture biometric data from the client device for a generation of an embedding;
generate a locality-sensitive hash (LSH) from the embedding;
determine that a set of measurements of the LSH meet a distance threshold;
generate a personal mask to identify a plurality of reliable bits based at least in part on the set of measurements of the locality sensitive hash meeting the threshold;
generate a secret that includes a cryptographic key, the cryptographic key being configured for authentication of the client device with an authentication service;
generate a masked locality-sensitive hash using a reusable fuzzy extractor based at least in part on the secret; and
send metadata to the authorization service, the metadata comprising the masked locality-sensitive hash.
16. The non-transitory computer-readable medium of claim 15 , wherein the biometric data is captured using at least one of a camera or a microphone of the client device.
17. The non-transitory computer-readable medium of claim 15 , wherein the embedding is in a non-hamming distance space.
18. The non-transitory computer-readable medium of claim 15 , wherein the secret is generated using a key generator executed by the client device.
19. The non-transitory computer-readable medium of claim 15 , wherein generating the masked locality-sensitive hash further comprises generating a set of lockers that include the secret.
20. The non-transitory computer-readable medium of claim 15 , wherein sending the metadata to the authorization service further comprising sending the set of lockers and the personal mask to the authentication service.