CN113239885A - Face detection and recognition method and system - Google Patents

Face detection and recognition method and system Download PDF

Info

Publication number
CN113239885A
CN113239885A CN202110626119.XA CN202110626119A CN113239885A CN 113239885 A CN113239885 A CN 113239885A CN 202110626119 A CN202110626119 A CN 202110626119A CN 113239885 A CN113239885 A CN 113239885A
Authority
CN
China
Prior art keywords
face
central point
recognition
diagram
network
Prior art date
Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
Pending
Application number
CN202110626119.XA
Other languages
Chinese (zh)
Inventor
徐小丹
刘小扬
何学智
Current Assignee (The listed assignees may be inaccurate. Google has not performed a legal analysis and makes no representation or warranty as to the accuracy of the list.)
Newland Digital Technology Co ltd
Original Assignee
Newland Digital Technology Co ltd
Priority date (The priority date is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the date listed.)
Filing date
Publication date
Application filed by Newland Digital Technology Co ltd filed Critical Newland Digital Technology Co ltd
Priority to CN202110626119.XA priority Critical patent/CN113239885A/en
Publication of CN113239885A publication Critical patent/CN113239885A/en
Pending legal-status Critical Current

Links

Images

Classifications

    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
    • G06V40/00—Recognition of biometric, human-related or animal-related patterns in image or video data
    • G06V40/10—Human or animal bodies, e.g. vehicle occupants or pedestrians; Body parts, e.g. hands
    • G06V40/16—Human faces, e.g. facial parts, sketches or expressions
    • G06V40/168—Feature extraction; Face representation
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
    • G06N3/00—Computing arrangements based on biological models
    • G06N3/02—Neural networks
    • G06N3/04—Architecture, e.g. interconnection topology
    • G06N3/045—Combinations of networks
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
    • G06N3/00—Computing arrangements based on biological models
    • G06N3/02—Neural networks
    • G06N3/08—Learning methods
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
    • G06V40/00—Recognition of biometric, human-related or animal-related patterns in image or video data
    • G06V40/10—Human or animal bodies, e.g. vehicle occupants or pedestrians; Body parts, e.g. hands
    • G06V40/16—Human faces, e.g. facial parts, sketches or expressions
    • G06V40/172—Classification, e.g. identification

Landscapes

  • Engineering & Computer Science (AREA)
  • Theoretical Computer Science (AREA)
  • Physics & Mathematics (AREA)
  • Health & Medical Sciences (AREA)
  • General Health & Medical Sciences (AREA)
  • General Physics & Mathematics (AREA)
  • Oral & Maxillofacial Surgery (AREA)
  • Biomedical Technology (AREA)
  • Biophysics (AREA)
  • Evolutionary Computation (AREA)
  • Computational Linguistics (AREA)
  • Molecular Biology (AREA)
  • Computing Systems (AREA)
  • General Engineering & Computer Science (AREA)
  • Data Mining & Analysis (AREA)
  • Mathematical Physics (AREA)
  • Software Systems (AREA)
  • Artificial Intelligence (AREA)
  • Multimedia (AREA)
  • Life Sciences & Earth Sciences (AREA)
  • Human Computer Interaction (AREA)
  • Computer Vision & Pattern Recognition (AREA)
  • Image Analysis (AREA)

Abstract

The invention discloses a face detection and identification method, which comprises the following steps: s1: preprocessing the image marked with the face frame to generate a training sample; s2: constructing a face detection and recognition network, wherein the face detection and recognition network adopts a deep learning network and fuses network high-level features and low-level features; s3: inputting training samples into the constructed face detection and recognition network for training until the training loss value is smaller than a preset threshold value, and obtaining a deep learning network capable of outputting face detection and face recognition results; according to the invention, a face detection and recognition network is designed, the face detection is regarded as the face central point problem, the face central point detection and the face feature vector extraction are combined for learning, a face frame is obtained, a face feature vector corresponding to the face frame can be obtained, then the face feature vector comparison is carried out to obtain a face recognition result, and therefore the network outputs the face detection and face recognition results.

Description

Face detection and recognition method and system
Technical Field
The invention relates to the field of artificial intelligence, in particular to a face detection and recognition method and system.
Background
With the development and progress of science and technology, face recognition has very wide application in crime fighting, fraud prevention, public safety guarantee, wide improvement of customer experience of various industries and the like. Such as identifying criminal suspects, finding lost children, intelligent stores, face payments, etc. Face recognition is a process of recognizing or verifying the identity of a person using facial information, and generally comprises three steps, step 1: face detection, which is an indispensable step in that it can detect and locate faces in images or videos; step 2: aligning the detected human faces, and converting one human face into a string of vectors by utilizing a human face feature extraction technology; and step 3: and calculating face similarity of the obtained feature vectors to judge whether the two faces belong to the same person.
The existing face recognition method has the defect of consuming time because the three steps are executed in a serial connection mode, the two functions are realized by a face detection network and a face recognition network which are designed in a distributed mode, in the face recognition process, the time for extracting the feature vectors is in direct proportion to the number of detected face frames, and the more the number of faces is, the more the time for extracting the feature vectors is, and therefore, the more the face recognition method in the mode is.
Disclosure of Invention
In order to solve the defects in the prior art, the invention designs a face recognition method which can simultaneously carry out two tasks of face detection and face recognition, thereby improving the efficiency of face detection and recognition and saving computer resources.
The technical scheme of the invention is as follows:
a face detection and recognition method is characterized by comprising the following steps: the method comprises the following steps:
s1: preprocessing the image marked with the face frame to generate a training sample;
s2: constructing a face detection and recognition network, wherein the face detection and recognition network adopts a deep learning network and fuses network high-level features and low-level features;
s3: inputting training samples into the constructed face detection and recognition network for training until the training loss value is smaller than a preset threshold value, and obtaining a deep learning network capable of outputting face detection and face recognition results;
the deep learning network for face detection comprises the following steps:
s31: generating a face central point thermodynamic diagram, a face central point offset diagram and a face width and height diagram;
s32: executing a non-maximum value suppression algorithm on the face central point thermodynamic diagram, extracting peak value points, respectively calculating thermodynamic response values, selecting points with the thermodynamic response values larger than a threshold value as candidate face central points, extracting face central point offset values at corresponding positions of the face central point offset map, adding to obtain face central point positions, and finally extracting face width and height values at corresponding positions of the face width and height map to generate a face frame;
the deep learning network for face recognition comprises the following steps:
s33: when step S31 is executed, image feature vectors of the entire image are extracted at the same time;
s34: selecting a feature vector corresponding to the position of the face frame from image feature vectors as a face feature vector, and matching the face feature vector with each face feature vector stored in a database to obtain a face recognition result;
the training loss value is formed by superposing face central point thermodynamic diagram loss, face central point offset diagram loss, face width and height diagram loss and face recognition loss.
Preferably, let the ith individual face frame on the image be represented by two points at the top left and bottom right of the frame
Figure BDA0003101194710000021
The face center point of the face frame
Figure BDA0003101194710000022
Is shown as
Figure BDA0003101194710000023
Order to
Figure BDA0003101194710000024
Is shown as
Figure BDA0003101194710000025
At the center point of the faceAnd if the corresponding position on the thermodynamic diagram corresponds to the generated human face central point thermodynamic diagram, the response value of the corresponding generated human face central point thermodynamic diagram is represented as:
Figure BDA0003101194710000026
where N represents the number of face frames on the image, σcExpressing the standard deviation of the Gaussian function;
the loss of the face center point thermodynamic diagram is represented as:
Figure BDA0003101194710000027
wherein, alpha and beta are modulation coefficients;
Figure BDA0003101194710000028
and representing the heat value of the center point of the face obtained by network prediction.
Preferably, let the ith individual face frame on the image be represented by its two upper left and lower right points as:
Figure BDA0003101194710000029
let its width and height be expressed as:
Figure BDA00031011947100000210
the face width height loss is expressed as:
Figure BDA00031011947100000211
wherein,
Figure BDA00031011947100000212
representing the width and height positions of the human face obtained by network prediction;
let the face center point of the ith face frame on the image be represented as
Figure BDA00031011947100000213
Order to
Figure BDA00031011947100000214
The corresponding position on the face central point thermodynamic diagram is represented as
Figure BDA0003101194710000031
Let the offset of the center point of the face be expressed as
Figure BDA0003101194710000032
Then the face center point offset loss is expressed as:
Figure BDA0003101194710000033
wherein,
Figure BDA0003101194710000034
and n is the multiple of the down-sampling of the deep neural network.
Preferably, the target central point of the ith face frame on the face central point thermodynamic diagram on the image is set as
Figure BDA0003101194710000035
Extracting the corresponding characteristic vector on the image characteristic vector diagram
Figure BDA0003101194710000036
Maps it to a class distribution vector pi(k) L for corresponding label class labeli(k) And if so, the face recognition loss is expressed as:
Figure BDA0003101194710000037
wherein N is the number of face frames, and K is the number of categories; p is a radical ofi(k) Is the probability that the ith face box belongs to the kth id, Li(k) And labeling the ith face frame.
Preferably, the face detection and recognition network uses resnet34 or Googlenet as a backbone network.
A face detection and recognition system comprising:
the image preprocessing module is used for preprocessing the image marked with the face frame to generate a training sample;
the human face feature extraction module is used for generating a human face central point thermodynamic diagram, a human face central point offset diagram and a human face width and height diagram and extracting an image feature vector of the whole image;
the training loss calculation module is used for calculating the thermodynamic diagram loss of the face center point, the offset diagram loss of the face center point, the width and height diagram loss of the face and the face recognition loss, performing superposition calculation, finishing training when the training loss value is smaller than a preset threshold value, and obtaining a deep learning network capable of outputting the face detection and face recognition results;
the human face detection module is used for executing a non-maximum value suppression algorithm on a human face central point thermodynamic diagram, calculating thermal response values of the human face central point thermodynamic diagram after extracting peak values, selecting points with the thermal response values larger than a threshold value as candidate human face central points, extracting human face central point offset values at corresponding positions of the human face central point offset quantity diagram, adding the human face central point offset values to obtain human face central point positions, and finally extracting human face width and height values at corresponding positions of the human face width and height diagram to generate a human face frame;
and the face recognition module is used for selecting the feature vector corresponding to the face frame position from the image feature vectors as a face feature vector, and matching the face feature vector with each face feature vector stored in the database to obtain a face recognition result.
By adopting the technical scheme, compared with the prior art, the invention has the following beneficial effects:
according to the invention, a face detection and recognition network is designed, the face detection is regarded as the face central point problem, the face central point detection and the face feature vector extraction are combined for learning, a face frame is obtained, a face feature vector corresponding to the face frame can be obtained, then the face feature vector comparison is carried out to obtain a face recognition result, and therefore the network outputs the face detection and face recognition results. The face detection and the face recognition share one network, so that the inference time is reduced, the forward time is irrelevant to the number of faces in the picture to be detected, the face recognition efficiency is improved, and moreover, the multi-task learning can supervise the learning mutually, and the network performance is favorably improved; on the other hand, the problem that the face detection is regarded as the face central point is solved, and the technical difficulty that ambiguity is easily caused when a plurality of faces are in charge of identity information of the same face in the prior art is overcome.
Drawings
FIG. 1 is a flow chart of a face detection and recognition method of the present invention;
FIG. 2 is a flowchart of the overall operation of the face detection and recognition method of the present invention;
fig. 3 is a diagram of a face detection and recognition network according to the present invention.
Detailed Description
The following further describes embodiments of the present invention with reference to the drawings. It should be noted that the description of the embodiments is provided to help understanding of the present invention, but the present invention is not limited thereto. In addition, the technical features involved in the embodiments of the present invention described below may be combined with each other as long as they do not conflict with each other.
Referring to fig. 1, the face detection and recognition method of the present invention includes the following steps:
s1: preprocessing the image marked with the face frame to generate a training sample;
s2: constructing a face detection and recognition network, wherein the face detection and recognition network adopts a deep learning network and fuses network high-level features and low-level features;
referring to fig. 3, in the embodiment, a resnet34 network is used as a backbone network in the face detection and recognition network, and the network fuses high-level and low-level features of the network through multiple hopping connections, so that the features are more robust;
s3: inputting training samples into the constructed face detection and recognition network for training until the training loss value is smaller than a preset threshold value, and obtaining a deep learning network capable of outputting face detection and face recognition results;
referring to fig. 1 and 2, the performing of the face detection by the deep learning network includes the following steps:
s31: generating a face central point thermodynamic diagram, a face central point offset diagram and a face width and height diagram;
in the embodiment, the step length is set to be 4, C × H × W images are input, C represents the number of channels, H and W respectively represent the height and width of the images, and after the images pass through a resnet34 network, a characteristic diagram with the shape of C × H/4 × W/4 is finally obtained;
referring to fig. 2 and fig. 3, the network includes a part for implementing the face detection function, the first branch is used for predicting the face central point thermodynamic diagram and is composed of a convolution of 256 × 3 × 3 and a convolution of 1 × 1 × 1, and finally, a thermodynamic diagram of 1 × H/4 × W/4 is obtained
Figure BDA0003101194710000041
The second branch is used for predicting the offset of the center point of the face and consists of a convolution of 256 multiplied by 3 and a convolution of 2 multiplied by 1, and finally the offset prediction result of 2 multiplied by H/4 multiplied by W/4 is obtained
Figure BDA0003101194710000051
The 3 rd branch is used for predicting the width and height of the face and consists of a convolution of 256 multiplied by 3 and a convolution of 2 multiplied by 1, and finally, a 2 multiplied by H/4 multiplied by W/4 face width and height prediction result is obtained
Figure BDA0003101194710000052
S32: executing a non-maximum value suppression algorithm on the face central point thermodynamic diagram, extracting peak value points, respectively calculating thermodynamic response values, selecting points with the thermodynamic response values larger than a threshold value as candidate face central points, extracting face central point offset values at corresponding positions of the face central point offset map, adding to obtain face central point positions, and finally extracting face width and height values at corresponding positions of the face width and height map to generate a face frame;
referring to fig. 2 and 3, the performing of the face recognition by the deep learning network includes the following steps:
s33: when step S31 is executed, image feature vectors of the entire image are extracted at the same time;
s34: selecting a feature vector corresponding to the position of the face frame from image feature vectors as a face feature vector, and matching the face feature vector with each face feature vector stored in a database to obtain a face recognition result;
in the embodiment, the network consists of a convolution of 256 × 3 × 3, a convolution of 128 × 1 × 1 (used for obtaining the face feature vector), and a convolution layer of K × 1 × 1, where K represents the face ID number, i.e., the number of classified categories, and the corresponding face feature vector is used as the identification of the face ID.
Referring to fig. 2, in an embodiment of the present invention, a face detection and recognition method includes the following steps:
making a feature vector database:
taking the face image of each id in the database as network input, and extracting corresponding characteristic vectors
Figure BDA0003101194710000053
As an identifier of each id, a database face feature vector set E ═ E is obtainedj|j＝1，…，K}。
Face detection and feature extraction:
using 3 × 960 × 720 images as input, a face center thermodynamic diagram of 1 × 0240 × 1180, a face center displacement diagram of 2 × 240 × 180, a width and height diagram of 2 × 240 × 180 faces, and an image feature vector of 128 × 240 × 180 are obtained. Executing a non-maximum value suppression algorithm on the 1 x 240 x 180 face central point thermodynamic diagram, extracting a peak face central point, and obtaining a thermal response value larger than T1N candidate face center points
Figure BDA0003101194710000054
Then, the offset of the center point of the corresponding face is taken
Figure BDA0003101194710000055
And width and height of human face
Figure BDA0003101194710000056
Obtaining a face frame after calculation
Figure BDA0003101194710000057
Taking the image feature vector set corresponding to the face frame from the image feature vectors of 128 × 240 × 180
Figure BDA0003101194710000058
I.e. the face feature vector.
Face matching:
collecting face characteristic vector
Figure BDA0003101194710000061
And the face feature vector set E ═ { E in the databasejComparing if 1 | j ═ 1.. K |, if
Figure BDA0003101194710000062
And in databases
Figure BDA00031011947100000617
Highest similarity value, and the similarity value is greater than threshold value T2Then it is considered as
Figure BDA0003101194710000063
And
Figure BDA00031011947100000616
corresponding to the same person.
In this example, T is taken1＝0.8，T2＝0.6。
In the embodiment of the invention, the training loss value is formed by superposing face central point thermodynamic diagram loss, face central point offset diagram loss, face width and height diagram loss and face identification loss.
Further, let the ith individual face frame on the image be represented by two points at the top left and bottom right of the frame
Figure BDA0003101194710000064
The face center point of the face frame
Figure BDA0003101194710000065
Is shown as
Figure BDA0003101194710000066
Order to
Figure BDA0003101194710000067
Is shown as
Figure BDA0003101194710000068
And at the corresponding position on the face central point thermodynamic diagram, the response value of the corresponding generated face central point thermodynamic diagram is represented as:
Figure BDA0003101194710000069
where N represents the number of face frames on the image, σcExpressing the standard deviation of the Gaussian function;
the loss of the face center point thermodynamic diagram is represented as:
Figure BDA00031011947100000610
wherein α and β are modulation coefficients, and in this embodiment, are set to 1 and 2, respectively;
Figure BDA00031011947100000611
and representing the heat value of the center point of the face obtained by network prediction.
Further, let the ith individual face frame on the image be represented by its two upper left and lower right points as:
Figure BDA00031011947100000612
let its width and height be expressed as:
Figure BDA00031011947100000613
the face width height loss is expressed as:
Figure BDA00031011947100000614
wherein,
Figure BDA00031011947100000615
representing the width and height positions of the human face obtained by network prediction;
in this embodiment, the face center point of the ith face frame on the image is expressed as
Figure BDA0003101194710000071
Order to
Figure BDA0003101194710000072
The corresponding position on the face central point thermodynamic diagram is represented as
Figure BDA0003101194710000073
Let the offset of the center point of the face be expressed as
Figure BDA0003101194710000074
Then the face center point offset loss is expressed as:
Figure BDA0003101194710000075
wherein,
Figure BDA0003101194710000076
and n is the multiple of the down-sampling of the deep neural network.
In this embodiment, n is 4, and for the label box
Figure BDA0003101194710000077
Its width and height
Figure BDA0003101194710000078
Corresponding center point offset is
Figure BDA0003101194710000079
Further, the target central point of the ith face frame on the face central point thermodynamic diagram on the image is set as
Figure BDA00031011947100000710
Extracting the corresponding characteristic vector on the image characteristic vector diagram
Figure BDA00031011947100000711
Maps it to a class distribution vector pi(k) L for corresponding label class labeli(k) And if so, the face recognition loss is expressed as:
Figure BDA00031011947100000712
wherein N is the number of face frames, and K is the number of categories; p is a radical ofi(k) Is the probability that the ith face box belongs to the kth id, Li(k) And labeling the ith face frame.
In this embodiment, the K value is 10000, and when the ith face frame belongs to the 1 st id, L isi(k) (1, 0, 0, 0,. 0, 0, 0), 9999 of which are 0.
In the embodiment provided by the invention, the face detection and recognition network adopts the resnet34 or Googlenet as a backbone network.
The invention also provides a face detection and recognition system, comprising:
the image preprocessing module is used for preprocessing the image marked with the face frame to generate a training sample;
the human face feature extraction module is used for generating a human face central point thermodynamic diagram, a human face central point offset diagram and a human face width and height diagram and extracting an image feature vector of the whole image;
the training loss calculation module is used for calculating the thermodynamic diagram loss of the face center point, the offset diagram loss of the face center point, the width and height diagram loss of the face and the face recognition loss, performing superposition calculation, finishing training when the training loss value is smaller than a preset threshold value, and obtaining a deep learning network capable of outputting the face detection and face recognition results;
the human face detection module is used for executing a non-maximum value suppression algorithm on a human face central point thermodynamic diagram, calculating thermal response values of the human face central point thermodynamic diagram after extracting peak values, selecting points with the thermal response values larger than a threshold value as candidate human face central points, extracting human face central point offset values at corresponding positions of the human face central point offset quantity diagram, adding the human face central point offset values to obtain human face central point positions, and finally extracting human face width and height values at corresponding positions of the human face width and height diagram to generate a human face frame;
and the face recognition module is used for selecting the feature vector corresponding to the face frame position from the image feature vectors as a face feature vector, and matching the face feature vector with each face feature vector stored in the database to obtain a face recognition result.
The embodiments of the present invention have been described in detail with reference to the accompanying drawings, but the present invention is not limited to the described embodiments. It will be apparent to those skilled in the art that various changes, modifications, substitutions and alterations can be made in these embodiments without departing from the principles and spirit of the invention, and the scope of protection is still within the scope of the invention.

Claims (6)

1. A face detection and recognition method is characterized by comprising the following steps: the method comprises the following steps:
s1: preprocessing the image marked with the face frame to generate a training sample;
s2: constructing a face detection and recognition network, wherein the face detection and recognition network adopts a deep learning network and fuses network high-level features and low-level features;
s3: inputting training samples into the constructed face detection and recognition network for training until the training loss value is smaller than a preset threshold value, and obtaining a deep learning network capable of outputting face detection and face recognition results;
the deep learning network for face detection comprises the following steps:
s31: generating a face central point thermodynamic diagram, a face central point offset diagram and a face width and height diagram;
s32: executing a non-maximum value suppression algorithm on the face central point thermodynamic diagram, extracting peak value points, respectively calculating thermodynamic response values, selecting points with the thermodynamic response values larger than a threshold value as candidate face central points, extracting face central point offset values at corresponding positions of the face central point offset map, adding to obtain face central point positions, and finally extracting face width and height values at corresponding positions of the face width and height map to generate a face frame;
the deep learning network for face recognition comprises the following steps:
s33: when step S31 is executed, image feature vectors of the entire image are extracted at the same time;
s34: selecting a feature vector corresponding to the position of the face frame from image feature vectors as a face feature vector, and matching the face feature vector with each face feature vector stored in a database to obtain a face recognition result;
the training loss value is formed by superposing face central point thermodynamic diagram loss, face central point offset diagram loss, face width and height diagram loss and face recognition loss.
2. A face detection and recognition method as claimed in claim 1, wherein:
let the ith personal face frame on the image be represented by two points, upper left and lower right, as
Figure FDA0003101194700000011
The face center point of the face frame
Figure FDA0003101194700000012
Is shown as
Figure FDA0003101194700000013
Order to
Figure FDA0003101194700000014
Is shown as
Figure FDA0003101194700000015
And at the corresponding position on the face central point thermodynamic diagram, the response value of the corresponding generated face central point thermodynamic diagram is represented as:
Figure FDA0003101194700000016
where N represents the number of face frames on the image, σcExpressing the standard deviation of the Gaussian function;
the loss of the face center point thermodynamic diagram is represented as:
Figure FDA0003101194700000017
wherein, alpha and beta are modulation coefficients;
Figure FDA0003101194700000021
and representing the heat value of the center point of the face obtained by network prediction.
3. A face detection and recognition method as claimed in claim 2, wherein:
let the ith personal face frame on the image be represented by its two upper left and lower right points as:
Figure FDA0003101194700000022
let its width and height be expressed as:
Figure FDA0003101194700000023
the face width height loss is expressed as:
Figure FDA0003101194700000024
wherein,
Figure FDA0003101194700000025
representing the width and height positions of the human face obtained by network prediction;
let the face center point of the ith face frame on the image be represented as
Figure FDA0003101194700000026
Order to
Figure FDA0003101194700000027
The corresponding position on the face central point thermodynamic diagram is represented as
Figure FDA0003101194700000028
Let the offset of the center point of the face be expressed as
Figure FDA0003101194700000029
Then the face center point offset loss is expressed as:
Figure FDA00031011947000000210
wherein,
Figure FDA00031011947000000211
and n is the multiple of the down-sampling of the deep neural network.
4. A face detection and recognition method as claimed in claim 3, wherein:
the target central point of the ith face frame on the face central point thermodynamic diagram on the image is set as
Figure FDA00031011947000000212
Extracting the corresponding characteristic vector on the image characteristic vector diagram
Figure FDA00031011947000000213
Maps it to a class distribution vector pi(k) L for corresponding label class labeli(k) And if so, the face recognition loss is expressed as:
Figure FDA00031011947000000214
wherein N is the number of face frames, and K is the number of categories; p is a radical ofi(k) Is the probability that the ith face box belongs to the kth id, Li(k) Is the label of the ith detection frame.
5. A face detection and recognition method as claimed in any one of claims 1 to 4, wherein: the face detection and recognition network adopts resnet34 or Googlenet as a backbone network.
6. A face detection and recognition system, comprising:
the image preprocessing module is used for preprocessing the image marked with the face frame to generate a training sample; the human face feature extraction module is used for generating a human face central point thermodynamic diagram, a human face central point offset diagram and a human face width and height diagram and extracting an image feature vector of the whole image;
the training loss calculation module is used for calculating the thermodynamic diagram loss of the face center point, the offset diagram loss of the face center point, the width and height diagram loss of the face and the face recognition loss, performing superposition calculation, finishing training when the training loss value is smaller than a preset threshold value, and obtaining a deep learning network capable of outputting the face detection and face recognition results;
the human face detection module is used for executing a non-maximum value suppression algorithm on a human face central point thermodynamic diagram, calculating thermal response values of the human face central point thermodynamic diagram after extracting peak values, selecting points with the thermal response values larger than a threshold value as candidate human face central points, extracting human face central point offset values at corresponding positions of the human face central point offset quantity diagram, adding the human face central point offset values to obtain human face central point positions, and finally extracting human face width and height values at corresponding positions of the human face width and height diagram to generate a human face frame;
and the face recognition module is used for selecting the feature vector corresponding to the face frame position from the image feature vectors as a face feature vector, and matching the face feature vector with each face feature vector stored in the database to obtain a face recognition result.
CN202110626119.XA 2021-06-04 2021-06-04 Face detection and recognition method and system Pending CN113239885A (en)

Priority Applications (1)

Application Number Priority Date Filing Date Title
CN202110626119.XA CN113239885A (en) 2021-06-04 2021-06-04 Face detection and recognition method and system

Applications Claiming Priority (1)

Application Number Priority Date Filing Date Title
CN202110626119.XA CN113239885A (en) 2021-06-04 2021-06-04 Face detection and recognition method and system

Publications (1)

Publication Number Publication Date
CN113239885A true CN113239885A (en) 2021-08-10

Family

ID=77136745

Family Applications (1)

Application Number Title Priority Date Filing Date
CN202110626119.XA Pending CN113239885A (en) 2021-06-04 2021-06-04 Face detection and recognition method and system

Country Status (1)

Country Link
CN (1) CN113239885A (en)

Cited By (3)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN114462495A (en) * 2021-12-30 2022-05-10 浙江大华技术股份有限公司 Training method of face shielding detection model and related device
TWI779815B (en) * 2021-09-03 2022-10-01 瑞昱半導體股份有限公司 Face recognition network model with face alignment based on knowledge distillation
CN115565207A (en) * 2022-11-29 2023-01-03 武汉图科智能科技有限公司 Occlusion scene downlink person detection method with feature simulation fused

Citations (10)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN106503669A (en) * 2016-11-02 2017-03-15 重庆中科云丛科技有限公司 A kind of based on the training of multitask deep learning network, recognition methods and system
CN108268822A (en) * 2016-12-30 2018-07-10 深圳光启合众科技有限公司 Face identification method, device and robot
CN109086660A (en) * 2018-06-14 2018-12-25 深圳市博威创盛科技有限公司 Training method, equipment and the storage medium of multi-task learning depth network
CN109919097A (en) * 2019-03-08 2019-06-21 中国科学院自动化研究所 Face and key point combined detection system, method based on multi-task learning
WO2019128646A1 (en) * 2017-12-28 2019-07-04 深圳励飞科技有限公司 Face detection method, method and device for training parameters of convolutional neural network, and medium
CN110705357A (en) * 2019-09-02 2020-01-17 深圳中兴网信科技有限公司 Face recognition method and face recognition device
CN111160108A (en) * 2019-12-06 2020-05-15 华侨大学 Anchor-free face detection method and system
WO2020102988A1 (en) * 2018-11-20 2020-05-28 西安电子科技大学 Feature fusion and dense connection based infrared plane target detection method
CN112270310A (en) * 2020-11-24 2021-01-26 上海工程技术大学 Cross-camera pedestrian multi-target tracking method and device based on deep learning
CN112749626A (en) * 2020-12-10 2021-05-04 同济大学 DSP platform-oriented rapid face detection and recognition method

Patent Citations (10)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN106503669A (en) * 2016-11-02 2017-03-15 重庆中科云丛科技有限公司 A kind of based on the training of multitask deep learning network, recognition methods and system
CN108268822A (en) * 2016-12-30 2018-07-10 深圳光启合众科技有限公司 Face identification method, device and robot
WO2019128646A1 (en) * 2017-12-28 2019-07-04 深圳励飞科技有限公司 Face detection method, method and device for training parameters of convolutional neural network, and medium
CN109086660A (en) * 2018-06-14 2018-12-25 深圳市博威创盛科技有限公司 Training method, equipment and the storage medium of multi-task learning depth network
WO2020102988A1 (en) * 2018-11-20 2020-05-28 西安电子科技大学 Feature fusion and dense connection based infrared plane target detection method
CN109919097A (en) * 2019-03-08 2019-06-21 中国科学院自动化研究所 Face and key point combined detection system, method based on multi-task learning
CN110705357A (en) * 2019-09-02 2020-01-17 深圳中兴网信科技有限公司 Face recognition method and face recognition device
CN111160108A (en) * 2019-12-06 2020-05-15 华侨大学 Anchor-free face detection method and system
CN112270310A (en) * 2020-11-24 2021-01-26 上海工程技术大学 Cross-camera pedestrian multi-target tracking method and device based on deep learning
CN112749626A (en) * 2020-12-10 2021-05-04 同济大学 DSP platform-oriented rapid face detection and recognition method

Cited By (4)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
TWI779815B (en) * 2021-09-03 2022-10-01 瑞昱半導體股份有限公司 Face recognition network model with face alignment based on knowledge distillation
US11847821B2 (en) 2021-09-03 2023-12-19 Realtek Semiconductor Corp. Face recognition network model with face alignment based on knowledge distillation
CN114462495A (en) * 2021-12-30 2022-05-10 浙江大华技术股份有限公司 Training method of face shielding detection model and related device
CN115565207A (en) * 2022-11-29 2023-01-03 武汉图科智能科技有限公司 Occlusion scene downlink person detection method with feature simulation fused

Similar Documents

Publication Publication Date Title
CN111126360B (en) Cross-domain pedestrian re-identification method based on unsupervised combined multi-loss model
Ye et al. PurifyNet: A robust person re-identification model with noisy labels
CN111259786B (en) Pedestrian re-identification method based on synchronous enhancement of appearance and motion information of video
CN113239885A (en) Face detection and recognition method and system
CN109255289B (en) Cross-aging face recognition method based on unified generation model
CN110751018A (en) Group pedestrian re-identification method based on mixed attention mechanism
CN109214263A (en) A kind of face identification method based on feature multiplexing
CN110765841A (en) Group pedestrian re-identification system and terminal based on mixed attention mechanism
CN111274922A (en) Pedestrian re-identification method and system based on multi-level deep learning network
WO2019167784A1 (en) Position specifying device, position specifying method, and computer program
Xia et al. Face occlusion detection using deep convolutional neural networks
CN112036379A (en) Skeleton action identification method based on attention time pooling graph convolution
CN113379771A (en) Hierarchical human body analytic semantic segmentation method with edge constraint
CN112364791A (en) Pedestrian re-identification method and system based on generation of confrontation network
CN115984901A (en) Multi-mode-based graph convolution neural network pedestrian re-identification method
CN112906520A (en) Gesture coding-based action recognition method and device
CN112613474B (en) Pedestrian re-identification method and device
CN114387496A (en) Target detection method and electronic equipment
Tran et al. DynGeoNet: fusion network for micro-expression spotting
CN114120076B (en) Cross-view video gait recognition method based on gait motion estimation
Nguyen et al. Real-time Human Detection under Omni-dir ectional Camera based on CNN with Unified Detection and AGMM for Visual Surveillance
Galiyawala et al. Dsa-pr: discrete soft biometric attribute-based person retrieval in surveillance videos
Jiashu Performance analysis of facial recognition: A critical review through glass factor
Han et al. Hyperbolic face anti-spoofing
Sun et al. FastPR: One-stage Semantic Person Retrieval via Self-supervised Learning

Legal Events

Date Code Title Description
PB01 Publication
PB01 Publication
SE01 Entry into force of request for substantive examination
SE01 Entry into force of request for substantive examination
WD01 Invention patent application deemed withdrawn after publication

Application publication date: 20210810

WD01 Invention patent application deemed withdrawn after publication