Privacy protection-based user recognition methods, apparatuses, and devices

1. A computer-implemented method for privacy-protected user recognition, comprising:

collecting data using a terminal device and extracting a facial feature of a first user based on the collected data;

obtaining a first ciphertext feature comprising:

performing a homomorphic encryption on the facial feature to obtain a first high-dimensional feature; and

performing dimension reduction on the facial feature and performing the homomorphic encryption to obtain a first low-dimensional feature;

determining a candidate ciphertext feature from a predetermined ciphertext feature set based on the first ciphertext feature and a predetermined graph structure index, wherein the predetermined ciphertext feature set comprises a plurality of second ciphertext features obtained by performing the homomorphic encryption of a plurality of biometric features of multiple second users, and wherein the predetermined graph structure index is generated based on similarity among at least some of the plurality of second ciphertext features in the predetermined ciphertext feature set; and

determining a recognition result for the first user based on the candidate ciphertext feature, wherein the predetermined ciphertext feature set comprises a high-dimensional feature subset generated by performing the homomorphic encryption on the at least some of the plurality of biometric features of the multiple second users, and wherein determining a recognition result for the first user based on the candidate ciphertext feature comprises:

determining, from the high-dimensional feature subset, ciphertext features corresponding to the candidate ciphertext feature as comparison features; and

determining the recognition result for the first user by comparing the first high-dimensional feature with the comparison features.

2. The computer-implemented method according to claim 1 , wherein the predetermined ciphertext feature set comprises:

a low-dimensional feature subset generated by performing dimension reduction on the at least some of the plurality of biometric features of the multiple second users to get second low-dimensional features and performing the homomorphic encryption on the second low-dimensional features,

wherein the predetermined graph structure index is generated based on the low-dimensional feature subset, and

wherein determining a candidate ciphertext feature from the predetermined ciphertext feature set based on the first ciphertext feature and the predetermined graph structure index comprises determining the candidate ciphertext feature from the low-dimensional feature subset based on the first low-dimensional feature and the predetermined graph structure index.

3. The computer-implemented method according to claim 1 , wherein determining the recognition result for the first user by comparing the first high-dimensional feature with the comparison features comprises:

determining whether a comparison feature matches the first high-dimensional feature by comparing the first high-dimensional feature with the comparison features; and

in response to determining that the comparison feature matches the first high-dimensional feature, determining, as the recognition result for the first user, that the first user is one of the multiple second users that corresponds to the comparison feature.

4. The computer-implemented method according to claim 1 , wherein the predetermined graph structure index comprises one or more graph nodes representing at least some of the plurality of second ciphertext features in the predetermined ciphertext feature set and one or more edges generated between the one or more graph nodes; and

wherein determining the candidate ciphertext feature from the predetermined ciphertext feature set based on the first ciphertext feature and the predetermined graph structure index comprises determining, from the predetermined graph structure index and through index query, one or more ciphertext features close to the first ciphertext feature as the candidate ciphertext feature.

5. The computer-implemented method according to claim 4 , wherein the predetermined graph structure index is generated based on a hierarchical navigable small world (HNSW) algorithm.

6. The computer-implemented method according to claim 1 , wherein the method further comprises, before performing the homomorphic encryption on the facial feature of the first user, pre-receiving an encryption key used to obtain the first ciphertext feature through the homomorphic encryption, and

wherein performing the homomorphic encryption on the facial feature of the first user comprises performing the homomorphic encryption on the facial feature of the first user by using the encryption key.

7. The computer-implemented method according to claim 1 , wherein the terminal device is offline or has poor network connectivity.

8. The computer-implemented method according to claim 1 , wherein the method further comprises, before determining the candidate ciphertext feature from the predetermined ciphertext feature set based on the first ciphertext feature and the predetermined graph structure index:

pre-receiving the predetermined ciphertext feature set sent by a cloud-side device, wherein privacy comprises the plurality of biometric features of the multiple second users, and privacy protection is achieved by performing the homomorphic encryption on the plurality of biometric features of the multiple second users to obtain the predetermined ciphertext feature set.

9. 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 for privacy-protected user recognition:

collecting data using a terminal device and extracting a facial feature of a first user based on the collected data;

obtaining a first ciphertext feature, comprising:

performing a homomorphic encryption on the facial feature to obtain a first high-dimensional feature; and

performing dimension reduction on the facial feature and performing the homomorphic encryption to obtain a first low-dimensional feature;

determining a candidate ciphertext feature from a predetermined ciphertext feature set based on the first ciphertext feature and a predetermined graph structure index, wherein the predetermined ciphertext feature set comprises a plurality of second ciphertext features obtained by performing the homomorphic encryption of a plurality of biometric features of multiple second users, and wherein the predetermined graph structure index is generated based on similarity among at least some of the plurality of second ciphertext features in the predetermined ciphertext feature set; and

determining a recognition result for the first user based on the candidate ciphertext feature, wherein the predetermined ciphertext feature set comprises a high-dimensional feature subset generated by performing the homomorphic encryption on the at least some of the plurality of biometric features of the multiple second users, and wherein determining a recognition result for the first user based on the candidate ciphertext feature comprises:

determining, from the high-dimensional feature subset, ciphertext features corresponding to the candidate ciphertext feature as comparison features; and

determining the recognition result for the first user by comparing the first high-dimensional feature with the comparison features.

10. The computer-implemented system according to claim 9 , wherein the predetermined ciphertext feature set comprises:

a low-dimensional feature subset generated by performing dimension reduction on the at least some of the plurality of biometric features of the multiple second users to get second low- dimensional features and performing the homomorphic encryption on the second low-dimensional features,

wherein the predetermined graph structure index is generated based on the low- dimensional feature subset, and

wherein determining a candidate ciphertext feature from the predetermined ciphertext feature set based on the first ciphertext feature and the predetermined graph structure index comprises determining the candidate ciphertext feature from the low-dimensional feature subset based on the first low-dimensional feature and the predetermined graph structure index.

11. The computer-implemented system according to claim 9 , wherein determining the recognition result for the first user by comparing the first high-dimensional feature with the comparison features comprises:

determining whether a comparison feature matches the first high-dimensional feature by comparing the first high-dimensional feature with the comparison features; and

in response to determining that the comparison feature matches the first high-dimensional feature, determining, as the recognition result for the first user, that the first user is one of the multiple second users that corresponds to the comparison feature.

12. The computer-implemented system according to claim 9 , wherein the predetermined graph structure index comprises one or more graph nodes representing at least some of the plurality of second ciphertext features in the predetermined ciphertext feature set and one or more edges generated between the one or more graph nodes; and

wherein determining the candidate ciphertext feature from the predetermined ciphertext feature set based on the first ciphertext feature and the predetermined graph structure index comprises determining, from the predetermined graph structure index and through index query, one or more ciphertext features close to the first ciphertext feature as the candidate ciphertext feature.

13. The computer-implemented system according to claim 12 , wherein the predetermined graph structure index is generated based on a hierarchical navigable small world (HNSW) algorithm.

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

collecting data using a terminal device and extracting a facial feature of a first user based on the collected data;

obtaining a first ciphertext feature, comprising:

performing a homomorphic encryption on the facial feature to obtain a first high- dimensional feature; and

performing dimension reduction on the facial feature and performing the homomorphic encryption to obtain a first low-dimensional feature;

determining a candidate ciphertext feature from a predetermined ciphertext feature set based on the first ciphertext feature and a predetermined graph structure index, wherein the predetermined ciphertext feature set comprises a plurality of second ciphertext features obtained by performing the homomorphic encryption of a plurality of biometric features of multiple second users, and wherein the predetermined graph structure index is generated based on similarity among at least some of the plurality of second ciphertext features in the predetermined ciphertext feature set; and

determining a recognition result for the first user based on the candidate ciphertext feature, wherein the predetermined ciphertext feature set comprises a high-dimensional feature subset generated by performing the homomorphic encryption on the at least some of the plurality of biometric features of the multiple second users, and wherein determining a recognition result for the first user based on the candidate ciphertext feature comprises:

determining, from the high-dimensional feature subset, ciphertext features corresponding to the candidate ciphertext feature as comparison features; and

determining the recognition result for the first user by comparing the first high-dimensional feature with the comparison features.