Systems and processes for lossy biometric representations
Systems and processes for improved processing of biometric data may include a hash controller including a processor, a server, and a registry. The hash controller can receive biometric information, such as a biometric scan, and apply an EGH transformation to convert the biometric information into an irreversible, unlinkable, and revocable EgHash. The EGH transformation can include blending biometric information with non-biometric information and permuting the biometric representation for additional security. The permuted biometric representation can be projected based on a randomly generated matrix and the output permuted to obtain an EgHash. The resultant EgHash can be lossy such that the EGH transform causes an irreversible loss of biometric information between the original biometric information and the EgHash. The EgHash can be compared and retrieved at speed and scale by the processor to support operations including, but not limited to, verification, identification, and database deduplication.
1. A process for transforming biometric data comprising:
receiving a biometric representation of a subject, wherein the biometric representation was encoded from a biometric sample via a neural network;
concatenating the biometric representation with a key to generate a concatenated biometric representation;
projecting the concatenated biometric representation based on one or more transformation parameters to generate an anonymized vector representation;
storing the projected anonymized vector representation in a hash registry;
receiving a second biometric representation associated with the subject;
generating a second projected anonymized vector representation based on the second biometric representation;
performing a 1:1 comparison between the stored projected anonymized vector representation and the second anonymized vector representation to generate a similarity score;
determining that the similarity score satisfies a similarity threshold; and
in response to determining the similarity score satisfies the similarity threshold, confirming a positive verification of an identity of the subject.
2. The process of claim 1 , wherein a dimension of the anonymized vector representation is less than a dimension of the concatenated biometric representation.
3. The process of claim 1 , wherein a vector dimension of the anonymized vector representation is equal to a vector dimension of the concatenated biometric representation.
4. The process of claim 1 , wherein the key is a unique key received from a user.
5. The process of claim 1 , wherein the key is a common key associated with a particular organization and is received with the biometric representation.
6. The process of claim 1 , further comprising:
generating a random seed; and
generating the key and the set of transformation parameters based on the random seed.
7. The process of claim 1 , further comprising:
prior to the projecting, permuting the concatenated biometric representation with the biometric representation and the key to generate a permuted, combined representation, wherein the projecting is performed using the permuted, combined representation; and
prior to the storing, permuting the anonymized vector representation to generate a permuted vector representation, wherein the permuted anonymized vector representation is stored in the hash registry.
8. The process of claim 1 , further comprising:
receiving a third biometric representation associated with the subject;
generating a third projected anonymized vector representation based on the third biometric representation;
performing a 1:N comparison between the third projected anonymized vector representation and a plurality of projected anonymized vector representations stored in the hash registry to generate a plurality of second similarity scores;
determining that a particular second similarity score of the plurality of second similarity scores satisfies a second similarity threshold; and
in response to determining the particular second similarity score satisfies the second similarity threshold, confirming a positive identification of the subject.
9. A system for transforming biometric data comprising:
a hash registry;
at least one processor; and
a non-transitory, machine-readable memory device comprising instructions stored thereon that, when executed by the at least one processor, cause the at least one processor to:
receive a biometric representation of a subject, wherein the biometric representation was encoded from a biometric sample via a neural network;
concatenate the biometric representation with a key to generate a concatenated biometric representation;
project the concatenated biometric representation based on one or more transformation parameters to generate an anonymized vector representation;
store the projected anonymized vector representation in the hash registry;
receive a second biometric representation associated with the subject;
generate a second projected anonymized vector representation based on the second biometric representation;
perform a 1:N comparison between the second projected anonymized vector representation and a plurality of projected anonymized vector representations stored in the hash registry to generate a plurality of similarity scores;
determine that a particular similarity score of the plurality of similarity scores satisfies a similarity threshold; and
in response to determining the similarity score satisfies the similarity threshold, confirm a positive identification of the subject.
10. The system of claim 9 , wherein a dimension of the anonymized vector representation is less than a dimension of the concatenated biometric representation.
11. The system of claim 9 , wherein a vector dimension of the anonymized vector representation is equal to a vector dimension of the concatenated biometric representation.
12. The system of claim 9 , wherein the key is a common key associated with a particular organization and is received with the biometric representation.
13. The system of claim 9 , wherein the instructions, when executed by the at least one processor, cause the at least one processor to:
generate a random seed; and
generate the key and the set of transformation parameters based on the random seed.
14. The system of claim 9 , wherein the instructions, when executed by the at least one processor, cause the at least one processor to:
prior to the projecting, permute the concatenated biometric representation with the biometric representation and the key to generate a permuted, combined representation, wherein the projecting is performed using the permuted, combined representation; and
prior to the storing, permute the anonymized vector representation to generate a permuted vector representation, wherein the permuted anonymized vector representation is stored in the hash registry.
15. The system of claim 9 , wherein the instructions, when executed by the at least one processor, cause the at least one processor to:
receive a third biometric representation associated with the subject;
generate a third projected anonymized vector representation based on the third biometric representation;
perform a 1:1 comparison between the stored projected anonymized vector representation and the third anonymized vector representation to generate a second similarity score;
determine that the second similarity score satisfies a second similarity threshold; and
in response to determining the second similarity score satisfies the second similarity threshold, confirm a positive verification of an identity of the subject.
16. A non-transitory, computer-readable medium comprising instructions that, when executed by a computer, cause the computer to:
receive a biometric representation of a subject, wherein the biometric representation was encoded from a biometric sample via a neural network;
concatenate the biometric representation with a key to generate a concatenated biometric representation;
project the concatenated biometric representation based on one or more transformation parameters to generate an anonymized vector representation;
store the projected anonymized vector representation in a hash registry;
receive a second biometric representation associated with the subject;
generate a second projected anonymized vector representation based on the second biometric representation;
perform a 1:N comparison between the second projected anonymized vector representation and a plurality of projected anonymized vector representations stored in the hash registry to generate a plurality of similarity scores;
determine that a particular similarity score of the plurality of similarity scores satisfies a similarity threshold; and
in response to determining the similarity score satisfies the similarity threshold, confirm a positive identification of the subject.
17. The non-transitory, computer-readable medium of claim 16 , wherein a dimension of the anonymized vector representation is less than a dimension of the concatenated biometric representation.
18. The non-transitory, computer-readable medium of claim 16 , wherein a vector dimension of the anonymized vector representation is equal to a vector dimension of the concatenated biometric representation.
19. The non-transitory, computer-readable medium of claim 16 , wherein the key is a common key associated with a particular organization and is received with the biometric representation.