AU2021101197A4

AU2021101197A4 - Facial recognition system based on neural network

Info

Publication number: AU2021101197A4
Application number: AU2021101197A
Authority: AU
Inventors: Tao Dong; Xiaomei Gong; Zhuoxian ZHANG
Original assignee: Southwest University
Current assignee: Southwest University
Priority date: 2021-03-07
Filing date: 2021-03-07
Publication date: 2021-05-06
Anticipated expiration: 2029-03-07

Abstract

The invention provides a facial recognition system based on neural network, including: shell, camera, display and control module. Before using facial recognition system, users need to input multiple user photos of different angles into the central control module. The central control module recognizes different photos of the same user and obtains face static feature data through a built-in algorithm. When using the facial recognition system, the camera collects face video information and transmits it to the central control module. The central control module selects the most suitable face image and compensates the face image by distance and brightness. In the process of comparison, the face image is divided into several regions and compared one by one with the idea of calculus. The invention comprehensively obtains human facial features through early data recognition, speeds up recognition speed during face recognition, and increases recognition accuracy through calculus and weight distribution. 1/1 2 3 1 ___________ ___________________ 4 7 / Fig.1

Description

1/1

2 3

1 ___________ ___________________ 4

/ 7 Fig.1

A facial recognition system based on neural network

1. Technical Field

The present invention relates to the field of face recognition technology. In particular, it involves a

facial recognition system based on neural network.

2. Background

With the development of computer and network technology in the 21st century, the hidden danger

of information security is becoming more and more prominent. Countries pay more and more attention

to social public security. Meanwhile, information identification and detection have shown

unprecedented importance. Its application field is wide, which can include almost every field of the

society. Nowadays, identification methods such as number, magnetic card and password are mainly

used in life, which are easy to lose, easy to forge, easy to forget and so on. With the continuous

development of technology, traditional identification methods have been increasingly challenged. And

the reliability has been greatly reduced. Therefore, new information identification and detection

technologies are bound to emerge. Increasingly, people are turning their attention to biological signs,

which are determined by human DNA. And each person has his or her own unique biological signs.

Biometrics can be broadly divided into two categories. One is the physical signs of objects, such

as fingerprints, iris, face, body odor and so on. The other is behavioral signs, such as writing, pace,

inertia and so on, all of which can be identified by modem computer image processing technology.

Compared to other human physiological characteristics, face has the advantages of easy collection,

non-contact, static and so on. And it is easy to be accepted by the public.

Face recognition is one of the hot research directions in the field of computer vision, and it has a

wide range of applications. For example, it plays an important role in security, monitoring,

entertainment and other fields.

According to a survey, 90% of people get to know each other and get basic information about each

other through observing facial features. This is the so-called first impression. While external conditions

such as age, expression, light exposure, and a person's facial features change dramatically, but humans

can still recognize it. This phenomenon indicates that a lot of characteristic information exists in the

face of a person. By presenting features of a person's face, a person can be judged.

At present, the quality of devices with face recognition function is uneven, and some devices have

low efficiency and slow recognition process.

3. The Invention Content

Therefore, the invention provides a facial recognition system based on neural network to

overcome the problem of low efficiency of face recognition in the prior art.

To achieve the above purposes, the invention provides a facial recognition system based on a

neural network, including the following aspects:

Shell: The shell is used to protect its interior;

Camera: The camera is arranged on the upper part of the shell to scan the face in the detection

area;

Display: The display is arranged on the upper part of the shell. The display is located on one side

of the camera device and connected with the camera device to display the picture detected by the

camera device in real time and feedback the scanned face state to the user. The user adjusts the position

of his or her face according to the feedback information;

Central control module: The central control module is arranged inside the shell and connects to the

camera device, which transmits the scanned face information to the central control module;

Before using facial recognition system, users need to input multiple user photos of different angles

into the central control module. The central control module recognizes different photos of the same

user and obtains face static feature data through a built-in algorithm. Face static feature data is used as

standard image data sample. Each image data sample can generate the corresponding user's face image.

The data samples of multiple users are combined to generate data sample matrix group KO.

The camera collects face video information and delivers the video information to the central

control module. The central control module analyzes each frame of the video to obtain multiple frames

of face images to be recognized. Select the most suitable face image k.

The camera has the infrared detection function, which can detect the distance L between the

personnel and the camera device and transmit the detection results to the central control module. The

central control module compensates the face image k to according to the distance L.

The central control module projects the face image k' onto the plane and divides the face into n

regions in pixels. The central control module detects the brightness of all areas within the face image k,

adjusts the brightness and generates a new face image k.

The central control module compares the face image k with the user face image stored in the

matrix group KO and selects the two face images with the highest similarity for accurate comparison.

The central control module divides the face image k into eye image, nose image, mouth image and the

whole face image. The whole face image is divided into n regions. Eye image is divided into n2 regions.

Nose image is divided into n3 regions. The mouth image is divided into n4 regions. The central control

module compares all images of a single region with the corresponding parts of the selected two face

images, judges whether the single region is consistent, and calculates the anastomosis rate in each

partition image. Each partition image has different weight. The central control module calculates

whether the face image k passes by combining the anastomosis rate and weight in each partition

image.

When the face recognition system carries out face recognition, the camera collects face video

information and transmits the video information to the central control module, which analyzes each

frame of video to obtain multiple face images to be recognized; selects the most suitable face image k;

the face image is M1 in length and N1 in width

Camera has infrared detection function, which can detect distance L between personnel and

cameras and transfer the results to the central control module. The central control module compensates

the face image k to according to the distance L. The length of the compensated image is M, the width

is N, where M = M1 x L x m', N = N1 x L x n', m' is the length compensation parameter of the face

image, n' is the width compensation parameter of the face image.

The central control module projects the face image k' to the plane, establishes a rectangular

coordinate system with the tip of the nose as the origin, takes the transverse direction of the face as the

X axis, takes the longitudinal direction of the face as the Y axis, and divides the face into n regions in

pixels.

The central control module is equipped with a brightness matrix CO and a brightness adjustment

parameter matrix DO. For the brightness matrix CO, CO(C1,C2,C3,C4), where C1 is the first preset

brightness parameter, C2 is the second preset brightness parameter,C3 is the third preset brightness

parameter, C4 is the fourth preset brightness parameter, and the brightness parameters increase in order.

For the brightness adjustment parameter matrix DO, DO (D1,D2,D3,D4), where D1 is the first preset

brightness adjustment parameter,D2 is the second preset brightness adjustment parameterD3 is the

third preset brightness adjustment parameter, and D4 is the fourth preset brightness adjustment

parameter.

The central control module detects the brightness Ci of region i of the face image k'. Compare Ci with the internal parameters of the CO matrix:

When Ci C1, the central control module judges that the brightness of region i is insufficient and

selects D1 from the DO matrix as the brightness adjustment parameter.

When Cl < Ci C2, the central control module judges that the brightness of region i is

insufficient and selects D2 as the brightness adjustment parameter from the DO matrix.

When C2 < Ci C3, the central control module determines that the brightness of region i is in the

standard state.

When C3 < Ci C4, the central control module judges that the brightness of region i is too high

and selects D3 from the DO matrix as the brightness adjustment parameter.

When Ci >C4, the central control module judges that the region i of brightness is too high and

selects D4 from the DO matrix as the brightness adjustment parameter.

When the center control module judges that the brightness of region i is not in the standard state,

the center control module adjusts the brightness of region i to Ci'. When the center control module

judges that the brightness of region i is insufficient, Ci'=Ci+ (C3-Ci)xDj, j=1,2. When the middle

control module judges that the brightness of region i is too high, Ci'=Ci-(Ci-C2)xDr, r=3,4.

When the central control module adjusts the brightness of region i to Ci', the central control

module compares Ci' with the parameters in the CO matrix. When C2 < Ci' C3, the central control

module determines that the brightness of region i is in the standard state. When Ci' is not C2-C3,

repeat the above operation until C2 < Ci' C3.

The central control module adjusts all areas in face image k' and generates a new face image k.

The regional brightness of face image k is within the range of C2C3. The central control module

generates face image matrix group HO according to k.

matrix group KO.Select the two face images with the highest similarity for accurate comparison. The

two face images correspond to the data sample matrix group Ku and the data sample matrix group Kv.

Furtherly, the central control module projects the face image ku in the matrix group Ku to the

plane establishes a rectangular coordinate system with the tip of the nose as the origin, takes the

transverse direction of the face as the X axis, takes the longitudinal direction of the face as the Y axis,

and divides the face into n regions in pixels.

Ku(Kul,Ku2,Ku3......Kun), for Kul (aul,bul ),aul is the first endpoint where the u face contour in the Kul region intersects with the region, and bul is the second endpoint where the u face contour intersects with the region in Kul. Line segment kul is generated by connecting two points.

For Ku2 (au2,bu2) , au2 is the first endpoint where the u face contour in the Ku2 region intersects

with the region, and bu2 is the second endpoint where the u face contour intersects with the region in

the Ku2 region. Line segment ku2 is generated by connecting two points.

For Ku3 (au3,bu3) ,au3 is the first endpoint where the u face contour in the Ku3 region intersects

with the region, and bu3 is the second endpoint where the u face contour intersects with the region in

the Ku3 region. Line segment ku3 is generated by connecting two points.

For Kun (aun,bun) ,aun is the first endpoint where the u face contour in the Kun region intersects

with the region, and bun is the second endpoint where the u face contour intersects with the region in

the Kun region. Line segment kun is generated by connecting two points.

Furtherly, the central control module extracts data from face image k" and generates matrix kO (k1,

k2, k3, k4), where kl is the overall face image of face image k, k2 is the eye image of face image k,k3

is the nose image of face image k, and k4 is the mouth image of face image k.

For generating matrix groups of face images HO (H,H2,H3,......Hn), where H1-Hn in the region

of the ki, Ha-Hb in the region of the k2, Hc-Hd in the region of the k3 and He-Hf in the region of the

k4.

For Hi (p l,q ), pl is the first endpoint where the face contour of H intersects with the region in

Hi region, and ql is the second endpoint where the face contour of H intersects with the region in HI

region. Line segment hi is generated by connecting two points.

For H2 (p2,q2) , p2 is the first endpoint where the face contour of H intersects with the region in

H2 region, and q2 is the second endpoint where the face contour of H intersects with the region in H2

region. Line segment h2 is generated by connecting two points.

For H3 (p3,q3) , p3 is the first endpoint where the face contour of H intersects with the region in

H3 region, and q3 is the second endpoint where the face contour of H intersects with the region in H3

region. Line segment h3 is generated by connecting two points.

For Hn (pn,qn) , pn is the first endpoint where the face contour of H intersects with the region in

Hn region, and qn is the second endpoint where the face contour of H intersects with the region in Hn

region. Line segment hn is generated by connecting two points.

Furtherly, the central control module compares HO with Ku and generates a comparison matrix

MO (Ml,M2,M3..Mn). Ml(hl,kul) is defined as follows:

z-(hl, kul) M1(hl,kul) I (h1, kul) II (hi, kul)

/(h1,kul) represents the angle 01 of line segment hi and line segment kul in the preset

Cartesian coordinate system. I (hi, kul) represents the vertical distance al between line hl and line

kul in the preset Cartesian coordinate system. II (h,kul) represents the horizontal distance pl between line hl and line kul in the preset Cartesian coordinate system.

The central control module is provided with a formula for calculating the contrast parameter P and

the judgment value Q of a single region as follows: 3 Q=y012 + Ea13 + T1

y is the calculated weight of 01 against the judgment value Q. e is the calculation weight of al on the judgment value Q. r is the calculated weight of p1 on the judgment value Q. When Q < P, it is judged that the H 1 region is consistent with the Kul region image.

When Q DP, it is judged that the image of H1 region is not consistent with Kul region image.

Furtherly, the central control module compares all regions in the matrix group HO with the regions

in the matrix group Ku. ki region is divided into n regions, and there are m regions where the images

match. k2 region is divided into n2 regions, and there are m2 regions where the images match. k3

region is divided into n3 regions, and there are m3 regions where the images match. k4 region is

divided into n4 regions, and there are m4 regions where the images match.

The central control module calculates the regional anastomosis rate. The anastomosis rate of k

region is Sl=m. nl The anastomosis rate of k2 region is S 2 mn2 . The anastomosis rate ofk3 region is

The anastomosis rate of k4 region is S4= -.

The central control module calculates the coincidence rate Sz between the face image k" and the

face image ku:

Sz= VS1x z1+ S2 x z2+ S3 x z3+ S4 x z4

Where zl is the weight parameter of Si to Sz, z2 is the weight parameter of S2 to Sz, z3 is the

weight parameter of S3 to Sz, z4 is the weight parameter of S4 to Sz.

The central control module also has a parameter S of the coincidence rate between the face image

k" and the face image ku. The central control module compares Sz with S as follows:

When Sz > S, the central control module judges that face image k" and face image ku are the same

user's face.

When Sz S, the central control module judges that face image k" and face image ku are different

faces.

When the center control module judges that face image k" and face image ku are the same user

face. According to the preset algorithm, the central control module makes the face image k" obtain the

static feature data of the face. And face static feature data is incorporated into ku data samples as

standard images.

When the center control module judges that the face image k" and the face image ku are different

faces. It projects the face image kv in matrix group Kv to the plane, and repeats the above operation to

recognize the face image kv.

When the center control module judges that the face image k", the face image ku and the face

image kv are different faces. The center control module judges that the face can't pass.

Compared with the prior art, the invention has a beneficial effect in the process of using the facial

recognition system. First of all, when multiple photos from different angles of users are input into the

central control module, the central control module will identify different photos of the same user. Face

static feature data is obtained by built-in algorithm, which is taken as standard image data sample. Each

image data sample can generate the corresponding user's face image. The data samples of multiple

users were combined to generate the data sample matrix group KO. Through the data recognition in the

early stage, the human facial features can be obtained comprehensively and accelerate the speed of face

recognition.

Furtherly, the camera collects face video information and delivers the video information to the

central control module. The central control module analyzes each frame of the video to obtain multiple

frames of face images to be recognized. Select the most suitable face image k. The camera has the

infrared detection function, which can detect the distance L between the personnel and the camera

device, transmit the detection results to the central control module. The central control module

compensates the face image k to k' according to the distance L. The central control module projects the

face image k' onto the plane and divides the face into n regions in pixels. The central control module

detects the brightness of all areas within the face image k', adjusts the brightness and generates a new

face image k". By selecting and compensating the collected information, the recognition speed is further accelerated. Meanwhile, infrared detection devices can perform biometric identification.

Preventing people from using photos or sculptures to identify.

Furtherly, the central control module compares the face image k with the user face image stored

in the matrix group KO. The two face images with the highest similarity were selected for accurate

comparison. The central control module divides the face image k into eye image, nose image, mouth

image and the whole face image. The whole face image is divided into n regions. Eye image is divided

into n2 regions. Nose image is divided into n3 regions. The mouth image is divided into n4 regions.

The central control module will compare all the single area images with the corresponding parts of the

two face images. Judge whether a single region is consistent and calculate the coincidence rate in each

partition image. Each partition has a different weight. The central control module calculates whether

the face image k passes by combining the anastomosis rate and weight in each partition image. By

means of calculus and weight distribution, the accuracy of recognition is increased.

Furtherly, when face recognition is passed, the recognition image is incorporated into the data

sample. Timely supplement the samples to increase the number of samples. The identification ability of

the identification system is strengthened and the identification speed is further accelerated.

4. Instruction With Pictures

Fig. 1 is a structural schematic diagram of a facial recognition system based on neural network.

5. Specific Implementation Mode

In order to make the purpose and advantages of the invention more clearly understood, the present

invention will be further described in combination with examples. The specific embodiments described

herein are used only to explain the invention ,but it is not intended to define the invention.

The preferred embodiment of the invention is described with reference to the attached drawings.

Technicians in the field shall understand that these embodiments are used only to explain the technical

principle of the invention. It is not limiting the scope of protection of the invention.

It should be noted that, in the description of the invention, the terms "up", "down", "left", "right",

"inside", "out" and so on, indicating directions or position relations are based on the directions or

position relations shown in the attached drawings. This is just for the sake of description, rather than

indicating or suggesting that the device or element must have or be constructed and operated in a

particular orientation. And therefore it cannot be construed as a limitation of the present invention.

It is further noted that in the description of the invention, the terms "installation", "connection" and "connection" are to be understood broadly, unless it is otherwise expressly prescribed and qualified.

For example, it can be a fixed connection, can also be a detachable connection, or an integral

connection. It can be a mechanical connection or an electrical connection. It can be directly connected,

it can be indirectly connected through an intermediary, it can be the internal connection of two

components. For the technical personnel in the field, the specific meaning of the above terms in the

invention may be understood according to the specific situation.

The structural schematic diagram of the facial recognition system based on the neural network is

shown in Fig. 1. The facial recognition system based on neural network comprises the shell 1, the

camera device 2, the display screen 3 and the central control module 4. Shell 1 is used to protect its

internally loaded components. The camera device 2 is arranged on the upper part of the shell 1 to scan

the face in the detection area. The display 3 is arranged on the upper part of the shell. The display is

located on one side of the camera device 2 and connected with the camera device 2 to display the

picture detected by the camera device in real time and feedback the scanned face state to the user.

Users adjust the position of their face based on the feedback. The central control module 4 is arranged

inside the shell 1 and connected with the camera device 2. The camera device 2 transmits the scanned

face information to the central control module 4.

into the central control module 4. The central control module 4 recognizes different photos of the same

The camera 2 collects face video information and delivers the video information to the central

control module 4. The central control module 4 analyzes each frame of the video to obtain multiple

frames of face images to be recognized. Select the most suitable face image k.

The camera 2 has the infrared detection function, which can detect the distance L between the

personnel and the camera device 2. The detection results are transmitted to central control module 4.

The central control module 4 compensates the face image k to k' according to the distance L.

The central control module 4 projects the face image k' onto the plane and divides the face into n

regions in pixels. The central control module 4 detects the brightness of all areas within the face image

k', adjusts the brightness and generates a new face image k".

The central control module 4 compares the face image k with the user face image stored in the

The central control module 4 divides the face image k into eye image, nose image, mouth image and

the whole face image. The whole face image is divided into n regions. Eye image is divided into n2

regions. Nose image is divided into n3 regions. The mouth image is divided into n4 regions. The

central control module 4 compares all images of a single region with the corresponding parts of the

selected two face images, judges whether the single region is consistent, and calculates the anastomosis

rate in each partition image. Each partition image has different weight. The central control module 4

combines the anastomosis rate and weight in the divided images to calculate whether the face image k

passes. When face recognition passes, the recognition image is incorporated into the data sample.

Specifically, when the face recognition system carries out face recognition, the camera device 2

collects face video information and transmits the video information to the central control module 4,

which analyzes each frame of video to obtain multiple face images to be recognized. Selects the most

suitable face image k, the face image is M1 in length and N1 in width.

The camera device 2 has infrared detection function ,which can detect distance L between

personnel and the camera device 2. The results will be transmitted to central control module 4. The

central control module 4 compensates the face image k to according to the distance L. The length of

the compensated image is M, the width is N, where M = M1 x L x m', N = N1 x L x n'. m' is the

length compensation parameter of the face image. n' is the width compensation parameter of the face

image.

The central control module 4 projects the face image k' to the plane, establishes a rectangular

pixels.

Specifically, the central control module 4 is equipped with a brightness matrix CO and a brightness

adjustment parameter matrix DO.

For the brightness matrix CO, CO(C1,C2,C3,C4), where C1 is the first preset brightness parameter,

C2 is the second preset brightness parameterC3 is the third preset brightness parameter,C4 is the fourth

preset brightness parameter, and the brightness parameters increase in order.

For the brightness adjustment parameter matrix DO, DO (D1,D2,D3,D4), where D1 is the first preset brightness adjustment parameter, D2 is the second preset brightness adjustment parameter, D3 is the third preset brightness adjustment parameter, and D4 is the fourth preset brightness adjustment parameter.

The central control module detects the brightness Ci of region i of the face image k'. Compare Ci

with the internal parameters of the CO matrix:

When CiGC1, the central control module 4 judges that the brightness of region i is insufficient

and selects D1 from the DO matrix as the brightness adjustment parameter.

When Cl < Ci C2, the central control module 4 judges that the brightness of region i is

When C2 < CiC3, the central control module 4 judges that the brightness of region i is in the

standard state.

When C3 < Ci <C4, the central control module 4 judges that the brightness of region i is too high

and selects D3 from the DO matrix as the brightness adjustment parameter.

When Ci >C4, the central control module 4 judges that the region i of brightness is too high and

selects D4 from the DO matrix as the brightness adjustment parameter.

When the center control module 4 judges that the brightness of region i is not in the standard state,

the center control module 4 will adjust the brightness of region i to Ci'. When the center control

module 4 judges that the brightness of region i is insufficient, Ci' =Ci+ (C3-Ci)xDj, j=1,2. When the

middle control module 4 judges that the brightness of region i is too high, Ci'=Ci-(Ci- C2)xDr, r=3,4.

When the center control module 4 adjusts the brightness of region i to Ci', the central control

module 4 judges that the brightness of region i is in the standard state. When Ci' is not between C2 and

C3, repeat the above operation until C2 < Ci'GC3.

The central control module 4 adjusts all areas in the face image k' and generates a new face image

k". The regional brightness of the face image k" is within the range of C2-C3. The central control

module 4 generates face image matrix group HO according to k".

Specifically, the central control module 4 compares the face image k" with the face image of the

user stored in the matrix group KO. The two face images with the highest similarity were selected for

accurate comparison. The two face images correspond to the data sample matrix group Ku and the data

sample matrix group Kv.

The central control module 4 projects the face image ku in the matrix group Ku to the plane,

establishes a rectangular coordinate system with the tip of the nose as the origin, takes the transverse

direction of the face as the X axis, takes the longitudinal direction of the face as the Y axis, and divides

the face into n regions in pixels.

Ku(Kul,Ku2,Ku3......Kun), for Kul (aul,bul), aul is the first endpoint where the u face contour

in the Kul region intersects with the region, and bul is the second endpoint where the u face contour

intersects with the region in Kul. Line segment kul is generated by connecting two points.

the Ku2 region. Line segment ku2 is generated by connecting two points.

For Ku3 (au3,bu3) , au3 is the first endpoint where the u face contour in the Ku3 region intersects

the Ku3 region. Line segment ku3 is generated by connecting two points.

For Kun (aun,bun) , aun is the first endpoint where the u face contour in the Kun region intersects

the Kun region. Line segment kun is generated by connecting two points.

Specifically, the central control module 4 extracts data from face image k and generates matrix

kO (kl, k2, k3, k4), where kl is the overall face image of the face image k, k2 is the eye image of the

face image k, k3 is the nose image of the face image k and k4 is the mouth image of the face image.

For generating matrix groups of face images HO (H,H2,H3,......Hn), H1-Hn in the region of the

kl, Ha-Hb in the region of the k2, Hc-Hd in the region of the k3 and He-Hf in the region of the k4.

For Hi (p l,ql) , p 1is the first endpoint where the face contour of H intersects with the region in

Hi region, and q1 is the second endpoint where the face contour of H intersects with the region in H

region. Line segment hl is generated by connecting two points.

region. Line segment h2 is generated by connecting two points.

region. Line segment h3 is generated by connecting two points.

region. Line segment hn is generated by connecting two points.

Specifically, the central control module 4 compares HO with Ku and generates a comparison

matrix MO (M1,M2,M3..Mn) . M1(hl,kul) is defined as follows:

L (hi, kul) M1(h1, ku) =1 (h1, kul) ||(h1, kul)

t(h1,ku1) represents the angle 81 of line segment hi and line segment kul in the preset

Cartesian coordinate system. I (h1, kul) represents the vertical distance al between line h and line

kul in the preset Cartesian coordinate system. II (h,kul) represents the horizontal distance 31

between line hi and line kul in the preset Cartesian coordinate system.

The central control module 4 is provided with a formula for calculating the contrast parameter P

and the judgment value Q of a single region:

Q=Fyo12 + eaI3 + p13

Where y is the calculated weight of 81 to the judgment value Q. e is the calculation weight of

a1 on the judgment value Q. f is the calculated weight of P1 on the judgment value Q. When Q < P, it is judged that the H1 region is consistent with the Kul region image.

Specifically, the central control module 4 compares all regions in the matrix group HO with the

regions in the matrix group Ku. k1 region is divided into n regions, and there are m regions where the

images match. k2 region is divided into n2 regions, and there are m2 regions where the images match.

k3 region is divided into n3 regions, and there are m3 regions where the images match. k4 region is

divided into n4 regions, and there are m4 regions where the images match.

The central control module 4 calculates the regional anastomosis rate. The anastomosis rate of k1

region is Sl=, The anastomosis rate of k2 region is S2= ,The anastomosis rate of k3 region is n n2

S3= , The anastomosis rate of k4 region is S4=. n3 n4

The central control module 4 calculates the coincidence rate Sz between the face image k and the

face image ku:

Sz= v'S1 x z1 + S2 x z2 + S3 x z3 + S4 x z4

weight parameter of S3 to Sz, z4 is the weight parameter of S4 to Sz.

The central control module 4 is also provided with a coincidence rate parameter S of the face

image k and the face image ku. The central control module 4 compares Sz with S as follows:

When Sz > S, the central control module 4 judges that face image k and face image ku are the

same user's face.

When Sz S, the central control module 4 judges that face image k and face image ku are

different faces.

Specifically, When the center control module 4 judges that face image k and face image ku are

the same user face. According to the preset algorithm, the central control module 4 makes the face

image k obtain the static feature data of the face. And face static feature data is incorporated into ku

data samples as standard images.

When the center control module 4 judges that the face image k and the face image ku are

different faces. It projects the face image kv in matrix group Kv to the plane, and repeats the above

operation to recognize the face image kv.

Specifically, When the center control module 4 judges that the face image k , the face image ku

and the face image kv are different faces. The center control module 4 judges that the face can't pass.

Thus, the technical solution of the invention has been described in combination with the preferred

embodiment which is shown in the attached drawings. However, it is easy for technicians in this field

to understand that the protection scope of the invention is obviously not limited to these specific ways.

Without deviating from the principle of the invention, the technical personnel in the field may make

equivalent changes or substitutions to the relevant technical features. The technical solutions resulting

from such modifications or substitutions shall fall within the scope of protection of the invention.

Claims

The claims defining the invention are as follows:

1. A facial recognition system based on neural network is characterized by:

Shell: The shell is used to protect its interior.

area.

of his or her face according to the feedback information.

camera device, which transmits the scanned face information to the central control module.

of face images to be recognized. Select the most suitable face image k.

personnel and the camera device. Transmit the detection results to the central control module. The

adjusts the brightness and generates a new face image k.

Nose image is divided into n3 regions. The mouth image is divided into n4 regions. The central control module compares all images of a single region with the corresponding parts of the selected two face images, judges whether the single region is consistent, and calculates the anastomosis rate in each partition image. Each partition image has different weight. The central control module calculates whether the face image k passes by combining the anastomosis rate and weight in each partition image.

When face recognition passes, the recognition image is incorporated into the data sample.

2. According to claim 1, the facial recognition system based on neural network has the following

features:

When facial recognition system is carried out, the camera device collects the video information of

the face and transmits the video information to the central control module. The central control module

analyzes each frame of the video and obtains multiple frames of face images to be recognized. Select

the most suitable face image k. The face image is M1 in length and N1 in width.

Camera has infrared detection function which can detect distance L between personnel and

cameras. It transfers the results to the central control module. The central control module compensates

image, n' is the width compensation parameter of the face image.

pixels.

3. According to the facial recognition system based on neural network mentioned in Claim 2, it

has the following features:

brightness parameter, C2 is the second preset brightness parameter, C3 is the third preset brightness

brightness adjustment parameter, D2 is the second preset brightness adjustment parameter, D3 is the

parameter.

The central control module detects the brightness Ci of region i of the face image k. Compare Ci

with the internal parameters of the CO matrix:

selects D1 from the DO matrix as the brightness adjustment parameter.

When C1 < Ci C2, the central control module judges that the brightness of region i is

When C2 < Ci C3, the central control module judges that the brightness of region i is in the

standard state.

and selects D3 from the DO matrix as the brightness adjustment parameter.

selects D4 from the DO matrix as the brightness adjustment parameter.

module judges that the brightness of region i is in the standard state. When Ci' is not C2-C3, repeat the

above operation until C2 < Ci' <C3.

generates face image matrix group HO according to k.

4. According to claim 3, the facial recognition system based on neural network has the following

features:

matrix group KO. Select the two face images with the highest similarity for accurate comparison. The

The central control module projects the face image ku in the matrix group Ku to the plane,

establishes a rectangular coordinate system with the tip of the nose as the origin, takes the transverse direction of the face as the X axis, takes the longitudinal direction of the face as the Y axis, and divides the face into n regions in pixels.

Ku(Kul, Ku2, Ku3......Kun), for Kul (aul, bul), aul is the first endpoint where the u face

contour in the Kul region intersects with the region, and bul is the second endpoint where the u face

contour intersects with the region in Kul. Line segment kul is generated by connecting two points.

For Ku2 (au2, bu2), au2 is the first endpoint where the u face contour in the Ku2 region intersects

the Ku2 region. Line segment ku2 is generated by connecting two points.

For Ku3 (au3, bu3), au3 is the first endpoint where the u face contour in the Ku3 region intersects

the Ku3 region. Line segment ku3 is generated by connecting two points.

For Kun (aun, bun), aun is the first endpoint where the u face contour in the Kun region intersects

the Kun region. Line segment kun is generated by connecting two points.

5. According to the facial recognition system based on neural network mentioned in Claim 4, it

has the following features:

The central control module extracts data from face image k and generates matrix kO (ki, k2, k3,

k4), where kl is the overall face image of the face image k, k2 is the eye image of the face image k, k3

is the nose image of the face image k and k4 is the mouth image of the face image.

For generating matrix groups of face images HO (Hi, H2, H3,......Hn), where H1-Hn are in the

region of the kl, Ha-Hb are in the region of the k2, Hc-Hd are in the region of thek3 and He-Hf are

in the region of the k4.

For Hi (p1, ql) , p 1 is the first endpoint where the face contour of H intersects with the region in

region. Line segment hl is generated by connecting two points.

For H2 (p2, q2) , p2 is the first endpoint where the face contour of H intersects with the region in

region. Line segment h2 is generated by connecting two points.

For H3 (p3, q3) , p3 is the first endpoint where the face contour of H intersects with the region in

H3 region, and q3 is the second endpoint where the face contour of H intersects with the region in H3 region. Line segment h3 is generated by connecting two points.

For Hn (pn, qn) , pn is the first endpoint where the face contour of H intersects with the region in

region. Line segment hn is generated by connecting two points.

6. According to claim 5, the facial recognition system based on neural network has the following

features:

The central control module compares HO with Ku and generates a comparison matrix MO

(M1,M2,M3..Mn) . M1(hl,kul) is defined as follows:

z (h1, kul) Mi(h,kul) = (hi, kul) ||(hi, kul)

L(hi,kui) represents the angle 81 of line segment hi and line segment kul in the preset

kul in the preset Cartesian coordinate system. II (h,ku1) represents the horizontal distance 31

between line hi and line kul in the preset Cartesian coordinate system.

the judgment value Q of a single region: 3 Q= y12 + ea13 + lp1

y is the calculated weight of 01 against the judgment value Q. e is the calculation weight of al

on the judgment value Q. f is the calculated weight of P1 on the judgment value Q.

When Q < P, it is judged that the H1 region is consistent with the Kul region image.

When Q DP, it is judged that the image of H region is not consistent with the Kul region image.

7. According to the facial recognition system based on neural network mentioned in Claim 6, it

has the following features:

The central control module compares all regions in the matrix group HO with the regions in the

matrix group Ku. ki region is divided into n regions, and there arem regions where the images match.

k2 region is divided into n2 regions, and there are m2 regions where the images match. k3 region is

divided into n3 regions, and there are m3 regions where the images match. k4 region is divided into n4

regions, and there are m4 regions where the images match.

The central control module calculates the regional anastomosis rate. The anastomosis rate of k region is Sl=m. The anastomosis rate of k2 region is S2= Theanastomosisrateofk3regionis n n

S3=-.The anastomosis rate of k4 region is S4= -. n3 n4

The central control module calculates the coincidence rate Sz between the face image " and the

face image ku:

Sz= VS1x z1+ S2 x z2+ S3 x z3+ S4 x z4

where z Iis the weight parameter of Si to Sz, z2 is the weight parameter of S2 to Sz, z3 is the weight

parameter of S3 to Sz, z4 is the weight parameter of S4 to Sz.

user's face.

faces.

8. According to claim 7, the facial recognition system based on neural network has the following

features:

face, the central control module obtains the face static feature data according to the preset algorithm.

The face static feature data is incorporated into the Ku data sample as a standard image.

recognize the face image kv.

9. According to claim 8, the facial recognition system based on neural network has the following

features: