WO2017070923A1

WO2017070923A1 - Human face recognition method and apparatus

Info

Publication number: WO2017070923A1
Application number: PCT/CN2015/093340
Authority: WO
Inventors: 杨奇; 陈书楷
Original assignee: 厦门中控生物识别信息技术有限公司
Priority date: 2015-10-30
Filing date: 2015-10-30
Publication date: 2017-05-04
Also published as: CN105518717A; CN105518717B

Abstract

A human face recognition method and apparatus for solving the technical problem that the robustness of existing human face recognition techniques is not good enough. The method comprises: pre-processing a human face image by means of linear spatial filtering; performing direction encoding on the pre-processed human face image, and generating a gradient direction vector f, wherein a direction partition index of the gradient direction vector f is: dindex = θ × bin/2π, and a direction weight is: wei ＝ (1 + index - dindex)³, where θ is the gradient direction, bin is the number of direction partitions, and index is an integer obtained after dindex is rounded off; extracting gradient local auto-correlation (glac) features according to the gradient direction vector; and utilizing a chi-squared test to calculate a distance between glac features to be matched, and recognizing the similarity.

Description

Face recognition method and device

Technical field

The present invention relates to the field of biometrics, and in particular, to a face recognition method and apparatus.

Background technique

In recent years, face recognition technology has made great progress and has begun to be widely used. There are many kinds of face recognition technologies, including, for example, a face recognition technology based on the gradient local auto-correlation feature (Gradient Local Auto-Correlation feature, English abbreviation: glac).

However, there are still some problems in the face recognition research that are not well solved. For example, the face recognition process will be affected by factors such as uneven illumination, face occlusion, and face posture changes, resulting in noise and reduced recognition. Precision.

Practice has found that existing face recognition techniques such as face recognition based on glac features are not robust enough.

Summary of the invention

Embodiments of the present invention provide a face recognition method and apparatus to solve the technical problem that the existing face recognition technology is not robust enough.

A first aspect of the present invention provides a face recognition method, including:

Preprocessing the face image with linear spatial filtering;

The pre-processed face image is direction-encoded to generate a gradient direction vector f, and the direction partition index of the gradient direction vector f is:

The direction weight is: wei=(1+index-dindex) ³ , where θ is the gradient direction, bin is the number of direction partitions, and index is the rounding of dindex;

Extracting a gradient local autocorrelation property glac feature according to the gradient direction vector;

The chi-square test is used to calculate the distance between the glac features to be matched, and the similarity is identified.

A second aspect of the present invention provides a face recognition device, including:

a preprocessing module for preprocessing the face image by using linear spatial filtering;

a direction coding module, configured to perform direction coding on the pre-processed face image to generate a gradient direction vector f, where the direction partition index of the gradient direction vector f is:

a feature extraction module, configured to extract a gradient local autocorrelation property glac feature according to the gradient direction vector;

A calculation module is configured to calculate a distance between the glac features to be matched by using a chi-square test to identify the similarity.

A third aspect of the present invention provides a terminal device, including: a processor and a memory; the memory is configured to store a program; the processor is configured to execute a program in the memory, so that the terminal device performs the first method according to the present invention. Aspects provide a face recognition method.

A fourth aspect of the present invention provides a storage medium storing one or more programs, the one or more programs including instructions that, when executed by a terminal device including one or more processors, cause the terminal The device performs the face recognition method as provided by the first aspect of the present invention.

As can be seen from the above, in some feasible embodiments of the present invention, an improved glac feature-based face recognition technical solution is provided, in which the direction of the pre-processed face image is performed. The code generates a gradient direction vector, and defines the direction weight of the gradient direction vector f as wei=(1+index-dindex) ³ , and extracts the glac feature, and calculates the distance between the glac features to be matched by using the chi-square test to realize the person. Face recognition improves the ability to resist noise interference and can effectively improve the robustness of face recognition.

DRAWINGS

In order to more clearly illustrate the technical solutions of the embodiments of the present invention, the drawings used in the embodiments and the prior art description will be briefly described below. Obviously, the drawings in the following description are only some implementations of the present invention. For example, other drawings may be obtained from those skilled in the art without any inventive effort.

1 is a schematic flowchart of a face recognition method according to an embodiment of the present invention;

Figure 2 (a) is a schematic diagram of a gradient direction vector f;

2(b) is a schematic diagram of a second-order adjacent position autocorrelation mode;

Figure 2(c) is

versus

Schematic diagram of the autocorrelation legend;

Figure 3 (a) is a schematic diagram showing the results of the FAR and FRR evaluation indicators of Experiment (1);

Figure 3 (b) is a schematic diagram of the results of the FAR and FRR evaluation indicators of Experiment (2);

Figure 3 (c) is a schematic diagram showing the results of the FAR and FRR evaluation indicators of Experiment (3);

Figure 3 (d) is a schematic diagram of the results of the FAR and FRR evaluation indicators of Experiment (4);

Figure 3(e) is a schematic diagram showing the results of the FAR and FRR evaluation indicators of Experiment (5);

Figure 3 (f) is a schematic diagram of the results of the FAR and FRR evaluation indicators of Experiment (6);

4 is a schematic structural diagram of a face recognition device according to an embodiment of the present invention;

FIG. 5 is a schematic structural diagram of a terminal device according to an embodiment of the present invention.

Detailed ways

The technical solutions in the embodiments of the present invention are clearly and completely described in the following with reference to the accompanying drawings in the embodiments of the present invention. It is an embodiment of the invention, but not all of the embodiments. All other embodiments obtained by those skilled in the art based on the embodiments of the present invention without creative efforts shall fall within the scope of the present invention.

The technical solution of the embodiment of the present invention is applicable to various application scenarios, for example, applicable to the Internet financial core, face recognition, face attribute recognition, face beautification, face cartoon drawing, face tracking, lip language recognition, living body Identify and other scenes. The method of the embodiment of the invention can be implemented by various terminal devices such as an attendance machine, an access control system, a computer, a mobile phone, a notebook computer and the like.

Referring to FIG. 1, a first embodiment of the present invention provides a face recognition method, which may include:

101. Preprocessing the face image by linear spatial filtering;

102. Perform direction coding on the pre-processed face image to generate a gradient direction vector f. The direction partition index of the gradient direction vector f is:

103. Extract a gradient local autocorrelation property glac feature according to the gradient direction vector;

104. Calculate the distance between the glac features to be matched by using a chi-square test to identify the similarity.

It can be seen that, in some feasible embodiments of the present invention, an improved glac feature-based face recognition technical solution is provided, in which a gradient direction is generated by directionally encoding a pre-processed face image. Vector, and define the direction weight of the gradient direction vector f as wei=(1+index-dindex) ³ , and extract the glac feature, and calculate the distance between the glac features to be matched by the chi-square test to realize face recognition and improve The ability to resist noise interference can effectively improve the robustness and recognition accuracy of face recognition.

It should be noted that the face recognition method provided by the embodiment of the present invention is a face recognition method based on the glac feature. The glac feature describes the autocorrelation feature of the local image gradient in space and direction, and it can sparsely express the gradient of the pixel according to the amplitude and direction of each pixel on the image. She has 2nd-order autocorrelation statistics. It is a continuation of the extension of the 1st order histogram with displacement and no additivity.

The face recognition method based on glac feature can include three processes: one, image preprocessing; second, extracting glac features; third, using chi-square test to calculate the distance between glac features to be matched, and obtaining similarity scores, identifying similarities Sex, complete face recognition. The second step of extracting the glac feature can be further divided into two steps: 1. Direction coding, 2. Using the gradient direction vector and the gradient magnitude, calculating the 1st and 2nd order autocorrelation gradient values in the local neighborhood, Get the glac feature. This is described in further detail below.

First, image preprocessing:

In the example of the present invention, the step of pre-processing the face image may further include two processes:

The first pre-processing; specifically, Gaussian difference filtering (Difference of Gaussian, English abbreviation: DoG) can be used for the first pre-processing of the face image. Gaussian filters are a class of linear smoothing filters that select weights based on the shape of a Gaussian function. Gaussian difference filtering is very effective for suppressing noise that obeys a normal distribution. By performing DoG processing on the face image, the influence of uneven illumination can be reduced.

The second pre-processing; specifically, the linear pre-processing of the first pre-processed face image may be performed by linear spatial filtering. Linear spatial filtering belongs to smoothing filtering. The signal is to remove the glitch of a wave or a part above a certain frequency, which is simply low-pass filtering; the response to the image is noise reduction and image blurring (due to high-frequency response details, Therefore, the details are removed to obtain a blurred outline. Since the edges of the image are generally at a high frequency portion, such smoothing filtering causes edge blurring, that is, the image and the background are not distinct but blurred. The output of the linear spatial filter is included in the filter mask The simple average value of the pixels in the neighborhood of the film, that is, the value of a certain pixel point is replaced by the average gray value of several pixels in the neighborhood, and the high frequency portion of the modified point is made smaller by the method of averaging.

Linear spatial filtering is applied to the prewitt operator. The prewitt operator is an edge detection of a first-order differential operator. The gray-scale difference between the upper and lower adjacent points of the pixel is used to reach the extreme detection edge at the edge, and some pseudo-edges are removed, which has a smooth effect on noise. In the embodiment of the present invention, the linear spatial filtering uses a presitt operator as shown below:

The presitt operator is in the x direction

In the y direction

In the example of the present invention, linear spatial filtering can effectively enhance the ability of anti-noise interference by adopting the above presitt operator.

Second, extract glac features

1, direction coding:

In the embodiment of the present invention, it is assumed that the pre-processed face image is I, r(x, y) is an arbitrary pixel point, and I is a function of r(x, y), then in the x and y directions,

The gradient is:

The amplitude is:

The direction is:

It can be seen that the face image signal I is a two-dimensional discrete function, and the image gradient g is actually a derivative of the two-dimensional discrete function I. The gradient g, the magnitude n, and the direction θ are all functions of r(x, y). It should be noted that in the function of the direction θ, arctan represents the arctangent in the inverse trigonometric function.

Then, by encoding the direction θ, a gradient orientation vector (G-O vector) can be generated. In this paper, the gradient direction vector is represented by f. The calculation of the gradient direction vector f is divided into two parts: a direction partition index and a direction vector. specific:

The direction partition index is:

The direction weight is: wei=(1+index-dindex) ³ ;(5)

Where θ is the gradient direction, bin is the number of direction partitions, and index is the rounding of dindex.

Please refer to FIG. 2(a), which is a schematic diagram of the gradient direction vector f in an example.

It should be noted that in the conventional glac feature extraction method, the calculation formula of the method weight wei is: wei=1+index-dindex. In the embodiment of the present invention, noise interference can be further suppressed by improving the calculation method of wei.

2, the calculation of glac features

The glac feature is calculated by calculating the 1st and 2nd order autocorrelation gradient values according to the gradient direction vector f and the amplitude n, and concatenating the 1st and 2nd order autocorrelation gradient values to obtain the glac feature. It should be noted that the gradient direction vectors f of the 1st order and the 2nd order are respectively recorded as orders in the order of

with

The theoretical calculation formula for the glac feature is:

Where a ₁ is the displacement vector relative to the reference pixel point r, n(r) is the magnitude of the gradient relative to the reference pixel point r, f _d is the dth gradient direction vector; d can be taken as d ₀ , d _{1 ,} etc.; R _N=0 (d ₀ ) is a first-order autocorrelation gradient value, and R _N=1 (d ₀ , d ₁ , a ₁ ) is a second-order autocorrelation gradient value. 0 ^th order represents a first-order calculation mode, and 1 ^th order represents a second-order calculation mode.

It should be noted that, in the embodiment of the present invention, the actual calculation formula is not adopted in the actual calculation. In the embodiment of the present invention, the specific steps of calculating the first-order and second-order autocorrelation gradient values include:

2.1. Set the center pixel point (x ₀ , y ₀ ) of the face image width and height as the reference pixel point, and calculate the spatial position weight w _{p of} each pixel point (x, y). The calculation formula is as follows:

Wherein the value of the coefficient gamma is preferably 0.035;

And in the 2*2 field at the pixel (x, y), the weights are voted according to the three-dimensional voting (Trilinear voting) to calculate the 1st order and 2nd autocorrelation gradient values.

2.2, the complete weight calculation formula is: weight=w _p ·mag·bias; (9)

Among them, bias is the position offset of the pixel in the 2*2 neighborhood. The complete weight calculations for the 1st and 2nd orders use the above formula (9).

Taking the second-order full weight calculation as an example, the mag in equation (9) is equal to that in equation (7).

In the complete weight calculation of the 1st order, mag is equal to n(r) in the formula (6).

2.3, f _d in practical applications, reflected in two aspects:

On the one hand, in the calculation of the autocorrelation gradient value, the direction weight wei is applied;

On the other hand, on the dimensional mapping of the glac feature, the direction partition index index is used.

The variables that calculate the dimension map include the spatial position of the pixel, the number of direction partitions, and the number of image partitions. Therefore, the numerical calculation of the glac feature is:

Glac _v =weight·wei;(10)

This formula (10) is applicable to both 1st and 2nd order and 3rd order calculations.

In the dimension calculation, the formula for calculating the 1st dimension is: dim ₁ = bin·nw·nh; (11)

The formula for calculating the second-order dimension is: dim ₂ = bin·nw·nh+model·bin·bin·nw·nh; (12)

Where nw and nh are equal fractions on the width and height of the image, respectively; model is a 2nd-order adjacent position autocorrelation mode, such as shown in Figure 2(b); Figure 2(c) shows

versus

Autocorrelation legend.

So far, the glac feature is calculated, and the glac feature is used for the numerical values and dimensional representations calculated by the above formulas (10)-(11).

It should be noted that the above formulas (6)-(7) are numerical theoretical formulas of glac features, which can be derived from equations (4), (5), (8)-(12), and equations (4), (5). ), (8)-(10) are the actual calculation formulas (6)(7), and the formulas (11), (12) are the dimensional calculations of the glac features. Among them, in formula (6)

It can be calculated by equations (4) and (5), where n(r) is the mag of the first order. In the embodiment of the present invention, the formulas (4), (5), and (8)-(12) can be actually used for calculation.

2.4. In the following, the calculation steps of the glac feature in the embodiment of the present invention are described with respect to the improvement of the prior art.

(1) After experiments on the infrared face database, in the embodiment of the present invention, it is preferable to set the coefficient before the center point (x ₀ , y ₀ ) of the image width and height from 0.5 to 0.55, that is, the horizontal and vertical of the center point. The coordinates are located 0.55 times the width and height of the image, respectively. Improvements can improve the robustness of the face recognition method.

(2) For equations (7) and (8):

The formula (7) in the prior art is:

In the embodiment of the present invention, min(n(r)+n(r+a ₁ )) in the formula (7) is changed.

Based on such improvements, the amplitude variation can be made more stable, avoiding large fluctuations due to noise.

Correspondingly, in the formula (8), for the second-order calculation, the mag in the prior art is equal to min(n(r)+n(r+a ₁ )), and is improved to mag equal to

For the 1st order calculation, still make mag equal to n(r). Based on such improvements, the amplitude variation can be made more stable, avoiding large fluctuations due to noise.

In the embodiment of the present invention, the formula (7) Changed

The 1st and 2st stepwise direction vector values are mapped from the plane to the circular surface, and have convex and concave properties, which can reduce the influence of noise.

(3) For the calculation of the spatial position weight in formula (8):

The calculation formula of the spatial position weight calculation method in the prior art is:

Among them, the value of gamma generally takes 0, indicating that the position of the center point (x ₀ , y ₀ ) does not affect the weight of formula (9).

In the embodiment of the present invention, the calculation formula of w _p is changed to be as shown in formula (8), and wherein the value of gamma is preferably 0.035, such an improvement can make the pixel point have a discriminative weight distribution due to different spatial positions. .

(4) Numerical calculation of the glac feature in equation (10):

In the embodiment of the present invention, the corresponding weight is added, for example, the weight of the first order is 0.1, the weight of the second order is 0.5, and the weight of the third order is 0.4, and these weights can be reflected in the parameter weight. In the 4 models of the 2nd order, the variation pattern of the displacement vector a ₁ relative to the reference pixel point r is represented by a matrix: [01; 11; 1-1], and only one model of the 3rd order is currently added: [1] , 0, 1], the 3rd order autocorrelation gradient value calculation method is similar to the 2nd order and will not be described in detail.

Among them, [01;11;1-1] is written as a standard matrix form:

(5) In the embodiment of the present invention, for the image segmentation, the glac feature is extracted separately, and different weights are also added to the segment. For example, a real number less than one may be multiplied in front of each chunk value, and the sum of such real numbers multiplied by each chunk is equal to one.

3, feature matching

In the embodiment of the present invention, the distance between the glac features to be matched is calculated by using the chi-square test to identify the similarity, thereby implementing face recognition.

It should be noted that, in a preferred embodiment of the present invention, when the distance between the glac features to be matched is calculated by the chi-square test, the first-order autocorrelation gradient values are removed, and weights are added to each image segment. Also, it is preferably the same as the block weight used to calculate the glac feature.

The embodiment of the invention improves the anti-noise interference ability by the feature matching method, and can effectively improve the robustness and recognition accuracy of the face recognition.

4. Experimental results

The embodiment of the present invention also performs a simulation experiment on the provided face recognition method, and uses FAR (False Acceptance Rate) and FRR (False Rejection Rate) indicators to evaluate the performance of the face recognition algorithm. In order to improve the face recognition method based on the glac feature at one time.

The experimental results are as follows:

(1) When using linear spatial filtering, the filter is replaced by the prewitt operator. The result of the FAR and FRR evaluation indicators obtained on the face-std database containing 29,800 face images is shown in Fig. 3(a).

(2) After improving the direction coding algorithm, the results of the FAR and FRR evaluation indicators obtained on the face-std database are shown in Fig. 3(b).

(3) After improving the glac feature calculation method, the results of the FAR and FRR evaluation indicators obtained on the face-std database are shown in Fig. 3(c).

(4) After adding the 3rd-order autocorrelation gradient, the results of FAR and FRR evaluation indicators obtained on the faceenrll-std database with 38225 face images are shown in Fig. 3(d).

(5) In the case where the rest of the same, the faceenrll-std face database, using DoG The results of the obtained FAR and FRR evaluation indicators are shown in Fig. 3(e).

(6) in the formula (7)

Changed

After that, the facenrll-std face database is processed by DoG, and the obtained FAR and FRR evaluation index results are shown in Fig. 3(f).

It can be seen from the above that the method of the embodiment of the present invention has improved the image pre-processing, extracting the glac feature and the feature matching, and can effectively reduce the FAR and FRR indicators of the face recognition algorithm, so that when the face is recognized, the method is more Great.

In summary, in some feasible embodiments of the present invention, an improved glac feature-based face recognition technical solution is provided, in which a gradient direction vector is generated by directionally encoding a preprocessed face image. And define the direction weight of the gradient direction vector f as wei=(1+index-dindex) ³ , and extract the glac feature, and calculate the distance between the glac features to be matched by the chi-square test to realize face recognition, and improve the face recognition. The ability to resist noise interference can effectively improve the robustness of face recognition.

In order to better implement the above solution of the embodiments of the present invention, related devices for cooperating to implement the above solutions are also provided below.

Referring to FIG. 4, a second embodiment of the present invention provides a face recognition device 400, which may include:

The preprocessing module 410 is configured to perform preprocessing on the face image by using linear spatial filtering;

The direction coding module 420 is configured to perform direction coding on the pre-processed face image to generate a gradient direction vector f, where the direction partition index of the gradient direction vector f is:

a feature extraction module 430, configured to extract a gradient local autocorrelation property glac feature according to the gradient direction vector;

The calculating module 440 is configured to calculate a distance between the glac features to be matched by using a chi-square test to identify the similarity.

In some embodiments of the present invention, the pre-processing module 410 may include:

a first pre-processing unit for performing a first pre-processing of the face image by using Gaussian difference filtering DoG;

a second pre-processing unit, configured to perform a second pre-processing on the first pre-processed face image by using linear spatial filtering;

Among them, linear spatial filtering uses the following presitt operator:

In the x direction

In the x direction

In some embodiments of the present invention, the direction encoding module 420 is specifically configured to:

Let the pre-processed face image be I, r(x, y) be any pixel point, then in the x and y directions,

The gradient is:

The amplitude is:

The direction is:

The direction θ is encoded to generate a gradient direction vector f.

In some embodiments of the present invention, the feature extraction module 430 is specifically configured to calculate 1st order and 2nd order autocorrelation gradient values according to the gradient direction vector f and the amplitude n, and connect in series to obtain a glac feature.

In some embodiments of the invention, the calculation module 440 uses the chi-square test to calculate the distance between the glac features to be matched, removes the 1st order autocorrelation gradient values, and adds weights to each image segmentation.

The face recognition device of the embodiment of the present invention may be, for example, an attendance machine, an access control system, or the like.

The function of each function module of the face recognition device in the embodiment of the present invention may be specifically implemented according to the method in the foregoing method embodiment. For the specific implementation process, reference may be made to the related description in the foregoing method embodiment, and details are not described herein again. .

Referring to FIG. 5, an embodiment of the present invention further provides a terminal device 500.

The terminal device 500 can be a micro-processing computer device, and can include: a processor 510 and a memory 520; the memory 520 is configured to store a program; the processor 510 is configured to execute the program in the memory, such that The terminal device 500 performs some or all of the steps of the face recognition method provided in the above method embodiment.

Optionally, the terminal device may further include a communication interface 530 and a bus 540. The processor 510, the memory 520, and the communication interface 530 communicate with each other through the bus 540. The communication interface 530 is configured to receive and send data. .

Specifically, the processor 510 can perform the following steps:

Preprocessing the face image with linear spatial filtering;

The terminal device may specifically be an attendance machine, an access control system, a computer, a mobile phone, a notebook computer, and the like.

It can be seen that, in some feasible embodiments of the present invention, an improved glac feature-based face recognition technical solution is provided, in which a gradient direction is generated by directionally encoding a pre-processed face image. Vector, and define the direction weight of the gradient direction vector f as wei=(1+index-dindex) ³ , and extract the glac feature, and calculate the distance between the glac features to be matched by the chi-square test to realize face recognition and improve The ability to resist noise interference can effectively improve the robustness of face recognition.

Embodiments of the present invention also provide a storage medium storing one or more programs, the one or more programs including instructions that, when executed by a terminal device including one or more processors, cause the terminal The apparatus performs some or all of the steps of the face recognition method as provided in the above method embodiments.

In the above embodiments, the descriptions of the various embodiments are different, and the parts that are not described in detail in a certain embodiment can be referred to the related description of other embodiments.

It should be noted that, for the foregoing method embodiments, for the sake of brevity, they are all described as a series of action combinations, but those skilled in the art should understand that the present invention is not limited by the described action sequence, because In accordance with the present invention, certain steps may be performed in other sequences or concurrently. In addition, those skilled in the art should also understand that the embodiments described in the specification are all preferred embodiments, and the actions and modules involved are not necessarily required by the present invention.

A person skilled in the art may understand that all or part of the various steps of the foregoing embodiments may be performed by a program to instruct related hardware. The program may be stored in a computer readable storage medium, and the storage medium may include: ROM, RAM, disk or CD.

The method and device for recognizing a face provided by the embodiments of the present invention are described in detail above. The principles and embodiments of the present invention are described in the following. The description of the above embodiments is only for helping to understand the present invention. The method and its core idea; at the same time, those skilled in the art, according to the idea of the present invention, there will be changes in the specific embodiments and application scope. In summary, the contents of this specification should not be construed as Limitations of the invention.

Claims

A face recognition method, comprising:

Preprocessing the face image with linear spatial filtering;

The pre-processed face image is direction-encoded to generate a gradient direction vector f, and the direction partition index of the gradient direction vector f is:
The direction weight is: wei=(1+index-dindex) 3 , where θ is the gradient direction, bin is the number of direction partitions, and index is the rounding of dindex;

Extracting a gradient local autocorrelation property glac feature according to the gradient direction vector;

The chi-square test is used to calculate the distance between the glac features to be matched, and the similarity is identified.
The method according to claim 1, wherein the step of preprocessing the face image by using linear spatial filtering comprises:

The Gaussian differential filtering DoG is used to perform the first preprocessing of the face image;

The second pre-processed face image is preprocessed by linear spatial filtering;

Among them, linear spatial filtering uses the following presitt operator:

In the x direction
In the y direction
The method according to claim 1, wherein the direction encoding the pre-processed face image to generate a gradient direction vector comprises:

Let the pre-processed face image be I, r(x, y) be any pixel point, then in the x and y directions,

The gradient is:

The amplitude is:

The direction is:

The direction θ is encoded to generate a gradient direction vector f.
The method according to claim 3, wherein said gradient direction vector f The steps of extracting the gradient local autocorrelation property glac feature include:

According to the gradient direction vector f and the amplitude n, the first-order and second-order autocorrelation gradient values are calculated and connected in series to obtain a glac feature.
The method according to claim 4, wherein the calculating the first-order and second-order autocorrelation gradient values according to the gradient direction vector f and the amplitude n and connecting in series to obtain the glac feature comprises:

The central pixel point of the face image width and height is (x 0 , y 0 ), and (x 0 , y 0 ) is used as the reference pixel point, and the spatial position weight w p of each pixel point (x, y) is calculated as :

The full position weight is: weight=w p ·mag·bias,

Where gamma is the preset coefficient, bias is the position offset of the pixel point (x, y) in the 2*2 neighborhood, mag is equal to n(r) in the first-order calculation, and mag is equal to the second-order calculation
a 1 is a displacement vector with respect to a reference pixel point;

Then, the value of the glac feature is: glac v = weight·wei;

The first order dimension of the glac feature is: dim 1 = bin·nw·nh;

The second-order dimension of the glac feature is: dim 2 = bin·nw·nh+model·bin·bin·nw·nh;

Where nw and nh are equal fractions on the width and height of the image, respectively, and model is a 2nd-order adjacent position autocorrelation mode.
The method of claim 4 wherein said step of calculating a distance between glac features to be matched using a chi-square test comprises:

When the chi-square test is used to calculate the distance between the glac features to be matched, the first-order autocorrelation gradient values are removed, and weights are added to each image block.
A face recognition device, comprising:

a preprocessing module for preprocessing the face image by using linear spatial filtering;

a direction coding module, configured to perform direction coding on the pre-processed face image to generate a gradient direction vector f, where the direction partition index of the gradient direction vector f is:
The direction weight is: wei=(1+index-dindex) 3 , where θ is the gradient direction, bin is the number of direction partitions, and index is the rounding of dindex;

a feature extraction module, configured to extract a gradient local autocorrelation property glac feature according to the gradient direction vector;

A calculation module is configured to calculate a distance between the glac features to be matched by using a chi-square test to identify the similarity.
The apparatus according to claim 7, wherein the preprocessing module comprises:

a first pre-processing unit for performing a first pre-processing of the face image by using Gaussian difference filtering DoG;

a second pre-processing unit, configured to perform a second pre-processing on the first pre-processed face image by using linear spatial filtering;

Among them, linear spatial filtering uses the following presitt operator:

In the x direction
In the x direction
The device of claim 7 wherein:

The direction coding module is specifically configured to:

Let the pre-processed face image be I, r(x, y) be any pixel point, then in the x and y directions,

The gradient is:

The amplitude is:

The direction is:

The direction θ is encoded to generate a gradient direction vector f.
The device of claim 7 wherein:

The feature extraction module is specifically configured to calculate the first-order and second-order autocorrelation gradient values according to the gradient direction vector f and the amplitude n, and connect in series to obtain a glac feature.
The device of claim 10 wherein:

When the calculation module calculates the distance between the glac features to be matched by using the chi-square test, the first order is removed. Autocorrelate the gradient values and add weights to each image block.
A terminal device, comprising: a processor and a memory; the memory is for storing a program; the processor is configured to execute a program in the memory, so that the terminal device performs any of claims 1 to 6 A face recognition method as described.
A storage medium storing one or more programs, the instructions including instructions that, when executed by a terminal device including one or more processors, cause the terminal device to perform as claimed in claim 1. The face recognition method according to any one of the six.