Method and apparatus for determining identity identifier of face in face image, and terminal

• US 10,402,627 B2
• Filed: 06/30/2017
• Issued: 09/03/2019
• Est. Priority Date: 01/04/2015
• Status: Active Grant

• Current Assignee: Huawei Technologies Co., Ltd. (Huawei Investment & Holding Co., Ltd.)
• Original Assignee: Huawei Technologies Co., Ltd. (Huawei Investment & Holding Co., Ltd.)
• Inventors: Ju, Wenqi, Li, Wei, Xu, Chunjing
• Primary Examiner(s): Grant, II, Jerome
• Application Number:
  US15/639,220
• Publication Number:
  US 20170300744A1
• Time in Patent Office: Days
• Field of Search: 382118
• CPC Class Codes: G06F 18/24155 Bayesian classification
• CPC Class Codes: G06V 10/764 using classification, e.g. of video objects
• CPC Class Codes: G06V 40/168 Feature extraction; Face representation
• CPC Class Codes: G06V 40/172 Classification, e.g. identification

Abstract

The present invention provides a method and an apparatus for determining an identity identifier of a face in a face image, and a terminal. The method includes: obtaining an original feature vector of a face image; selecting k candidate vectors from a face image database; selecting a matching vector of the original feature vector from the k candidate vectors; and determining, an identity identifier that is of the matching vector. In embodiments of the present invention, a face image database stores a medium-level feature vector formed by means of mutual interaction between a low-level face feature vector and autocorrelation and cross-correlation submatrices in a joint Bayesian probability matrix. The medium-level feature vector includes information about mutual interaction between the face feature vector and the autocorrelation and cross-correlation submatrices in the joint Bayesian probability matrix, so that efficiency and accuracy of facial recognition can be improved.

Claims

1. A method for determining an identity identifier of a face in a face image, comprising:

obtaining an original feature vector of a face image, comprising;

tracking locations of the face in different video images to obtain N consecutive frames of face images; and

obtaining the original feature vector from a t^th frame of face image of the N frames of face images, wherein 1≤

t<

N;

selecting k candidate vectors from a face image database according to the original feature vector, wherein a vector v* in the face image database comprises components [v·

A,v·

B·

v^T], a feature extraction manner for v is the same as that for the original feature vector, A represents a cross-correlation submatrix in a joint Bayesian probability matrix, B represents an autocorrelation submatrix in the joint Bayesian probability matrix, and k is a positive integer;

selecting a matching vector of the original feature vector from the k candidate vectors, wherein the matching vector of the original feature vector is a candidate vector of the k candidate vectors that has a shortest cosine distance from a vector [v1,1], and v1 represents the original feature vector;

determining, according to the matching vector of the original feature vector, an identity identifier that is of the matching vector and that is recorded in the face image database as an identity identifier of a face in the face image;

storing, into a cache database, the matching vector of the original feature vector and the identity identifier that is of the matching vector and that is recorded in the face image database;

obtaining an original feature vector of a (t+1)^th frame of face image of the N frames of face images, wherein the (t+1)^th frame of face image is a next frame of face image of the t^th frame of face image;

selecting a target vector from the cache database according to the original feature vector of the (t+1)^th frame of face image, wherein the cache database comprises the matching vector of the original feature vector of the t^th frame of face image; and

when a target distance between the original feature vector of the (t+1)^th frame of face image and the target vector is less than a preset threshold, determining an identity identifier that is of the target vector and that is recorded in the cache database as an identity identifier of a face in the (t+1)^th frame of face image.

2. The method according to claim 1, wherein the selecting a target vector from the cache database according to the original feature vector of the (t+1)^th frame of face image comprises:

selecting the target vector from the cache database according to a formula s=[v2,1]·

(v*_c)^T+v2·

B·

v2^T, wherein v2 represents the original feature vector of the (t+1)^th frame of face image, v*_c represents a vector in the cache database, and s represents a target distance between v2 and v*_c.

3. The method according to claim 1, wherein the selecting k candidate vectors from a face image database according to the original feature vector comprises:

selecting the k candidate vectors from the face image database according to a formula s*=∥

[v1,0]−

v*∥

₂ by using a kd-tree algorithm, wherein s* represents a Euclidean distance between [v1,0] and v*.

4. The method according to claim 1, wherein the selecting a matching vector of the original feature vector from the k candidate vectors comprises:

selecting the matching vector of the original feature vector from the k candidate vectors according to a formula s**=−

((s*_i)²−

(∥

v*_i∥

₂)²−

(∥

v1∥

₂)²)+c·

v1·

B·

v1^T, wherein v*_i represents an i^th candidate vector of the k candidate vectors, s** represents a cosine distance between [v1,1] and v*_i, s*_i=∥

[v1,0]−

v*_i∥

₂, s*_i represents a Euclidean distance between [v1,0] and v*_i, 1≤

i≤

k, and c is a constant.

5. The method according to claim 1, wherein the method is executed by a server;

and

the obtaining an original feature vector of a face image comprises;

obtaining, by the server, a vector v′

from a terminal, wherein v′

=[v1,v1·

B·

v1^T]; and

extracting, by the server, the original feature vector from the vector v′

.

6. A method for determining an identity identifier of a face in a face image, comprising:

obtaining an original feature vector of a face image;

selecting a target vector from a face image database according to a formula s=[v2,1]·

(v*_c)^T+v2·

B·

v2^T, wherein v2 represents the original feature vector, v*_c represents a vector in the face image database, v*_c comprises components [v·

A,v·

B·

v^T], a feature extraction manner for v is the same as that for the original feature vector, A represents a cross-correlation submatrix in a joint Bayesian probability matrix, B represents an autocorrelation submatrix in the joint Bayesian probability matrix, s represents a target distance between v2 and v*_c, and the target vector is a vector that is in the face image database and that has a smallest target distance from v2; and

when a target distance between the original feature vector and the target vector is less than a preset threshold, determining an identity identifier that is of the target vector and that is recorded in the face image database as an identity identifier of a face in the face image.

7. An apparatus for determining an identity identifier of a face in a face image, comprising:

a memory to store instructions; and

a processor to execute the instructions to cause the apparatus to;

obtain an original feature vector of a face image, comprising track locations of the face in different video images to obtain N consecutive frames of face images, and obtain the original feature vector from a t^th frame of face image of the N frames of face images, wherein 1≤

t<

N;

select k candidate vectors from a face image database according to the original feature vector, wherein a vector v in the face image database comprises components [v·

A,v·

B·

v^T], a feature extraction manner for v is the same as that for the original feature vector, A represents a cross-correlation submatrix in a joint Bayesian probability matrix, B represents an autocorrelation submatrix in the joint Bayesian probability matrix, and k is a positive integer;

select a matching vector of the original feature vector from the k candidate vectors, wherein the matching vector of the original feature vector is a candidate vector of the k candidate vectors that has a shortest cosine distance from a vector [v1,1], and v1 represents the original feature vector;

determine, according to the matching vector of the original feature vector, an identity identifier that is of the matching vector and that is recorded in the face image database as an identity identifier of a face in the face image;

store, into a cache database, the matching vector of the original feature vector and the identity identifier that is of the matching vector and that is recorded in the face image database;

obtain an original feature vector of a (t+1)^th frame of face image of the N frames of face images, wherein the (t+1)^th frame of face image is a next frame of face image of the t^th frame of face image;

select a target vector from the cache database according to the original feature vector of the (t+1)^th frame of face image, wherein the cache database comprises the matching vector of the original feature vector of the t^th frame of face image; and

when a target distance between the original feature vector of the (t+1)^th frame of face image and the target vector is less than a preset threshold, determine an identity identifier that is of the target vector and that is recorded in the cache database as an identity identifier of a face in the (t+1)^th frame of face image.

8. The apparatus according to claim 7, wherein the processor further executes the instructions to cause the apparatus to select the target vector from the cache database according to a formula s=[v2,1]·

(v*_c)^T+v2·

B·

v2^T, wherein v2 represents the original feature vector of the (t+1)^th frame of face image, v*_c represents a vector in the cache database, and s represents a target distance between v2 and v*_c.

9. The apparatus according to claim 7, wherein the processor further executes the instructions to cause the apparatus to select the k candidate vectors from the face image database according to a formula s*=∥

[v1,0]−

v*∥

₂ by using a kd-tree algorithm, wherein S represents a Euclidean distance between [v1,0] and v*.

10. The apparatus according to claim 7, wherein the processor further executes the instructions to cause the apparatus to select the matching vector of the original feature vector from the k candidate vectors according to a formula) s**=−

((s*_i)²−

(∥

v*_i∥

₂)²−

(∥

v1∥

₂)²)+c·

v1·

B·

v1^T, wherein v*_i represents an i^th candidate vector of the k candidate vectors, s** represents a cosine distance between [v1,1] and v*_i, s*_i=∥

[v1,0]−

v*_i∥

₂, s*_i represents a Euclidean distance between [v1,0] and v*_i, 1≤

i≤

k, and c is a constant.

11. The apparatus according to claim 7, wherein the apparatus is a server;

and

the processor further executes the instructions to cause the apparatus to obtain a vector v′

from a terminal, wherein v′

=[v1,v1·

B·

v1^T]; and

extract the original feature vector from the vector v′

.

12. A terminal, comprising:

a camera, configured to collect a face image;

a processor, configured to obtain an original feature vector v1 of the face image, and generate a vector v′

according to the original feature vector v1, wherein v40 =[v1,v1·

B·

v1^T] and B represents an autocorrelation submatrix in a joint Bayesian probability matrix; and

a transmitter, configured to send the vector v′

to a server, wherein the vector v40 is used by the server to recognize an identity identifier of a face in the face image.