Face retrieval method and apparatus, server, and storage medium

Patent metadata

Field	Value
Publication number	US-11449541-B2
Application number	US-202016988178-A
Country	US
Kind code	B2
Filing date	Aug 7, 2020
Priority date	May 4, 2018
Publication date	Sep 20, 2022
Grant date	Sep 20, 2022

Abstract

A face retrieval method is applied to a face retrieval system and performed by a computing device, the face retrieval system including a retrieval device cluster, the retrieval device cluster including at least one node. The method includes acquiring a face image, parsing the face image to obtain a first facial feature, and generating a first retrieval instruction according to the first facial feature, the first retrieval instruction carrying the first facial feature. The method further includes selecting a first node from the retrieval device cluster according to a load balancing rule, the first node including a first retrieval server. Finally, the method includes transmitting the first retrieval instruction to the first retrieval server, to trigger the first retrieval server to execute the first retrieval instruction to retrieve the first facial feature, to obtain a first retrieval result.

First claim

The invention claimed is:

1. A face retrieval method, applied to a face retrieval system, the face retrieval system comprising a retrieval device cluster, the retrieval device cluster comprising at least one node, the method being performed by a computing device and comprising:
acquiring a face image, and parsing the face image to obtain a first facial feature;
generating, by processing circuitry of the computing device, a first retrieval instruction according to the first facial feature, the first retrieval instruction carrying the first facial feature;
selecting, by the processing circuitry of the computing device, a first node from the retrieval device cluster according to a load balancing rule, the first node comprising a first retrieval server; and
transmitting the first retrieval instruction to the first retrieval server, wherein the first retrieval instruction causes the first retrieval server to obtain a first retrieval result by executing the first retrieval instruction.

2. The method according to claim 1 , wherein the retrieval device cluster further comprises a first root server, and the first node belongs to the first root server; and
the transmitting comprises:
transmitting the first retrieval instruction to the first root server; and
controlling the first root server to distribute the first retrieval instruction to the first retrieval server, wherein the first retrieval instruction causes the first retrieval server to obtain the first retrieval result by executing the first retrieval instruction.

3. The method according to claim 2 , wherein the first node further comprises a second retrieval server; and
after the transmitting the first retrieval instruction to the first root server, the method further comprises:
controlling the first root server to distribute the first retrieval instruction to the second retrieval server, wherein the first retrieval instruction causes the second retrieval server to obtain a second retrieval result by executing the first retrieval instruction.

4. The method according to claim 1 , wherein the selecting comprises:
obtaining a load value of each node in the retrieval device cluster; and
selecting, from the retrieval device cluster, the first node having a load value that is less than a threshold.

5. The method according to claim 3 , further comprising:
obtaining a second retrieval instruction, the second retrieval instruction carrying a second facial feature;
selecting a second node from the retrieval device cluster according to the load balancing rule, the second node comprising a third retrieval server;
allocating the second retrieval instruction to the third retrieval server, wherein the second retrieval instruction causes the third retrieval server to obtain a third retrieval result by executing the second retrieval instruction; and
returning at least one of the first retrieval result, the second retrieval result, and the third retrieval result to a terminal for output.

6. A face retrieval method, applied to a face retrieval system, the face retrieval system comprising a retrieval device cluster, the retrieval device cluster comprising at least a first node, the first node comprising a first retrieval server, the method being performed by the first retrieval server and comprising:
receiving a first retrieval instruction transmitted by a scheduling server, the first retrieval instruction carrying a first facial feature, the first retrieval instruction being generated by the scheduling server according to the first facial feature after acquiring a face image and parsing the face image to obtain the first facial feature, and the first retrieval instruction being transmitted after the scheduling server selects the first node from the retrieval device cluster according to a load balancing rule;
executing, by processing circuitry of the first retrieval server, the first retrieval instruction to obtain a first retrieval result; and
returning the first retrieval result to the scheduling server.

7. The method according to claim 6 , wherein the executing comprises:
matching the first facial feature with each reference feature in a database, the database comprising at least one reference feature and user data corresponding to each reference feature;
obtaining, from the database, first user data corresponding to a reference feature in the database matching the first facial feature in a case that the reference feature matching the first facial feature is found among the at least one reference feature in the database; and
generating the first retrieval result, the first retrieval result comprising the first user data.

8. The method according to claim 7 , wherein the database further comprises a label attribute of each reference feature; and
the matching comprises:
determining a target label attribute of the first facial feature;
extracting, from the database, a reference feature set corresponding to the target label attribute; and
matching the first facial feature with each reference feature in the extracted reference feature set.

9. The method according to claim 7 , wherein the database further comprises at least one compressed feature obtained after dimension reduction and/or precision compression is performed on each reference feature; and
the matching comprises:
performing the dimension reduction and/or the precision compression on the first facial feature to generate a processed first facial feature; and
identifying, in the database, a compressed feature matching the processed first facial feature.

10. The method according to claim 6 , wherein, before the executing the first retrieval instruction, the method further comprises:
allocating continuous storage space in an internal memory according to dimensions of the first facial feature; and
loading the first facial feature into the allocated continuous storage space for continuous storage.

11. The method according to claim 6 , wherein, after the receiving the first retrieval instruction transmitted by the scheduling server, the method further comprises:
combining, in a case that a second retrieval instruction transmitted by the scheduling server is received, the first retrieval instruction and the second retrieval instruction into a batch; and
performing face retrieval in batches.

12. A face retrieval apparatus, applied to a face retrieval system, the face retrieval system comprising a retrieval device cluster, the retrieval device cluster comprising at least one node, the apparatus comprising:
processing circuitry configured to
acquire a face image, parse the face image to obtain a first facial feature, and generate a first retrieval instruction including the first facial feature;
select a first node from the retrieval device cluster according to a load balancing rule, the first node comprising a first retrieval server; and
transmit the first retrieval instruction to the first retrieval server, wherein the first retrieval instruction causes the first retrieval server to obtain a first retrieval result by executing the first retrieval instruction.

13. A scheduling server, comprising:
a processor, configured to implement one or more instructions; and
a non-transitory computer storage medium, storing one or more instructions, the one or more instructions causing the processor to perform the face retrieval method according to claim 1 .

14. A retrieval server, comprising:
a processor, configured to implement one or more instructions; and
a non-transitory computer storage medium, storing one or more instructions, the one or more instructions causing the processor to perform the face retrieval met

Assignees

• Tencent Tech Shenzhen Co Ltd

Inventors

• Chen Zhibo
• Jiang Nan
• Wang Chuannan
• Shi Kaihong

Classifications

• Y02D10/00

  Energy efficient computing, e.g. low power processors, power management or thermal management · CPC title

• G06F9/505

  considering the load · CPC title

• G06F16/436

  using biological or physiological data of a human being, e.g. blood pressure, facial expression, gestures · CPC title

• G06V40/168

  Feature extraction; Face representation · CPC title

• G06V40/164

  using holistic features · CPC title