CN109977781A - Method for detecting human face and device, readable storage medium storing program for executing - Google Patents
Method for detecting human face and device, readable storage medium storing program for executing Download PDFInfo
- Publication number
- CN109977781A CN109977781A CN201910142852.7A CN201910142852A CN109977781A CN 109977781 A CN109977781 A CN 109977781A CN 201910142852 A CN201910142852 A CN 201910142852A CN 109977781 A CN109977781 A CN 109977781A
- Authority
- CN
- China
- Prior art keywords
- face
- image
- convolutional neural
- facial image
- neural networks
- 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
Links
Classifications
-
- 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
- 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
-
- 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
- G06V40/174—Facial expression recognition
-
- 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
- G06V40/178—Human faces, e.g. facial parts, sketches or expressions estimating age from face image; using age information for improving recognition
-
- 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
- G06V40/179—Human faces, e.g. facial parts, sketches or expressions metadata assisted face recognition
Abstract
A kind of method for detecting human face and device, readable storage medium storing program for executing, the method for detecting human face include obtaining image to be detected;Recognition of face is carried out to the image to be detected, obtains facial image;Face character detection is carried out to the facial image, obtains face character;The face character includes following at least two classes: the face value score of the corresponding gender of head pose, face and age, the mood of face and face.Above scheme can satisfy a variety of demands of Face datection.
Description
Technical field
The invention belongs to technical field of face recognition, in particular to a kind of method for detecting human face and device, readable storage medium
Matter.
Background technique
The essence of Face datection and attributive classification is image object detection, image classification problem, but present face is examined
Survey method is general only to identify the face in image, is unable to get the more detailed information of face, can not disposably expire
The demand of a variety of Face datections of foot.
Summary of the invention
What the embodiment of the present invention solved is how to meet a variety of demands of Face datection.
In order to solve the above technical problems, the embodiment of the present invention provides a kind of method for detecting human face, method for detecting human face includes
Obtain image to be detected;Recognition of face is carried out to the image to be detected, obtains facial image;To the facial image into
Pedestrian's face detection of attribute, obtains face character;The face character includes following at least two classes: head pose, face are corresponding
The face value score of gender and age, the mood of face and face.
Optionally, described that recognition of face is carried out to the figure to be detected, comprising: to use the first convolutional neural networks mould
Type carries out recognition of face.
Optionally, described that face character detection is carried out to the facial image, comprising: to use with the face character one by one
Corresponding convolutional neural networks model carries out face character detection to the facial image.
Optionally, described to use following at least one side with the one-to-one convolutional neural networks model of the face character
Method carries out model training: simultaneously being filled with random pixel value, adjusts brightness of image, Random-fuzzy filter at random in hollow out random image region
Wave, randomised particulars filtering and random sharp filtering.
Optionally, described to use following at least two letter with the one-to-one convolutional neural networks model of the face character
Number weighting composition recombination losses function: the mean square error of the convolutional neural networks classification layer output distribution expectation and true value,
Layer output that the convolutional neural networks classification layer exports the variance of distribution, the convolutional neural networks are classified and the friendship being really grouped
Pitch entropy.
Optionally, after obtaining image to be detected, further includes: carry out school to the direction of the image to be detected
Just, so that correction after image in portrait direction be non-tilt state.
Optionally, described that recognition of face is carried out to the image to be detected, comprising: in the identification image to be detected
The position of face and quantity.
Optionally, described that recognition of face is carried out to the image to be detected, comprising: to generate in the image to be detected
Candidate frame, and merge height be overlapped part;The non-face candidate frame in image after removal merging;Identify face key point
Position obtains coordinate information, face confidence level and the key point coordinate of the facial image.
Optionally, described that face character detection is carried out to the facial image, comprising: to be rolled up using preset second multi output
Product neural network model detects the head pose in the facial image, obtains the Eulerian angles that face deflects in the facial image
Degree.
Optionally, described that face character detection is carried out to the facial image, comprising: to be rolled up using preset third multi output
Product neural network model detects the corresponding gender of face and age in the facial image, and obtained gender and age are tied
Fruit feeds back to the preset third multi output convolutional neural networks model simultaneously, so that the preset third multi output volume
Product neural network model extracts the face characteristic of different dimensions again.
Optionally, described that face character detection is carried out to the facial image, comprising: to use preset 4th shallow-layer convolution
Neural network model detects the mood of the face in the facial image.
Optionally, described that face character detection is carried out to the facial image, comprising: to use preset 5th shallow-layer convolution
Neural network model detects the face value score of the face in the facial image.
In order to solve the above technical problems, the embodiment of the invention also discloses a kind of human face detection device, human face detection device
Including acquiring unit, for obtaining image to be detected;Face identification unit, for carrying out face to the image to be detected
Identification, obtains facial image;Face character detection unit obtains people for carrying out face character detection to the facial image
Face attribute;The face character includes following at least two classes: the mood of head pose, face corresponding gender and age, face
And the face value score of face.
Optionally, the face identification unit, for carrying out recognition of face using the first convolution neural network model.
Optionally, the face character detection unit, for using and the one-to-one convolutional Neural of the face character
Network model carries out face character detection to the facial image.
Optionally, described to use following at least one side with the one-to-one convolutional neural networks model of the face character
Method carries out model training: simultaneously being filled with random pixel value, adjusts brightness of image, Random-fuzzy filter at random in hollow out random image region
Wave, randomised particulars filtering and random sharp filtering.
Optionally, described to use following at least two letter with the one-to-one convolutional neural networks model of the face character
Number weighting composition recombination losses function: the mean square error of the convolutional neural networks classification layer output distribution expectation and true value,
Layer output that the convolutional neural networks classification layer exports the variance of distribution, the convolutional neural networks are classified and the friendship being really grouped
Pitch entropy.
Optionally, the acquiring unit, is also used to: being corrected to the direction of the image to be detected, so that correction
Portrait direction in image afterwards is non-tilt state.
Optionally, the face identification unit, for identification position of face and quantity in the image to be detected.
Optionally, the face identification unit for generating the candidate frame in the image to be detected, and merges height
The part of coincidence;The non-face candidate frame in image after removal merging;It identifies face key point position, obtains the face figure
Coordinate information, face confidence level and the key point coordinate of picture.
Optionally, the face character detection unit, for using preset second multi output convolutional neural networks model
The head pose in the facial image is detected, the Euler angle that face deflects in the facial image is obtained.
Optionally, the face character detection unit, for using preset third multi output convolutional neural networks model
The corresponding gender of face and the age in the facial image are detected, and by obtained gender and age result while feeding back to institute
Preset third multi output convolutional neural networks model is stated, so that the preset third multi output convolutional neural networks model
The face characteristic of different dimensions is extracted again.
Optionally, the face character detection unit, for being examined using preset 4th shallow-layer convolutional neural networks model
Survey the mood of the face in the facial image.
Optionally, the face character detection unit, for being examined using preset 5th shallow-layer convolutional neural networks model
Survey the face value score of the face in the facial image.
The embodiment of the invention also discloses a kind of computer readable storage medium, computer readable storage medium is non-volatile
Property storage medium or non-transitory storage media, be stored thereon with computer instruction, the computer instruction executes above-mentioned when running
The step of any described method for detecting human face.
The embodiment of the invention also provides a kind of human face detection device, including memory and processor, on the memory
It is stored with the computer instruction that can be run on the processor, the processor executes above-mentioned when running the computer instruction
The step of any described method for detecting human face.
Compared with prior art, the technical solution of the embodiment of the present invention has the advantages that
Recognition of face is carried out to image to be detected, obtains facial image.Face character is carried out to the facial image again
Detection, obtains face character;The face character includes following at least two classes: the corresponding gender of head pose, face and age,
The mood of face and the face value score of face.By identification facial image and corresponding face character, face can satisfy
A variety of demands of detection.
Further, using with the one-to-one convolutional neural networks model of face character, to the facial image carry out people
Face detection of attribute.The accuracy that different convolutional neural networks model structures improves face character testing result is preset, simultaneously
So that the process of face character detection is decentralized, modularization, reduce algorithm iteration cost.
Detailed description of the invention
Fig. 1 is the flow chart of one of embodiment of the present invention method for detecting human face;
Fig. 2 is the structural schematic diagram of one of embodiment of the present invention human face detection device.
Specific embodiment
In the prior art, the essence of Face datection and attributive classification is image object detection, image classification problem, but is showed such as
Modern method for detecting human face is general only to identify the face in image, is unable to get the more detailed information of face, also not
The demand of the energy a variety of Face datections of all promising policy.
In the embodiment of the present invention, recognition of face is carried out to image to be detected, obtains facial image.Again to the face figure
As carrying out face character detection, face character is obtained;The face character includes following at least two classes: head pose, face pair
The face value score of the gender answered and age, the mood of face and face.Pass through identification facial image and corresponding face category
Property, it can satisfy a variety of demands of Face datection.
It is understandable to enable above-mentioned purpose of the invention, feature and beneficial effect to become apparent, with reference to the accompanying drawing to this
The specific embodiment of invention is described in detail.
The embodiment of the invention provides a kind of method for detecting human face, referring to Fig.1, carry out specifically below by way of specific steps
It is bright.
Step S101 obtains image to be detected.
In specific implementation, after obtaining image to be detected, the direction of image to be detected can be corrected,
So that correction after image in portrait direction be non-tilt state.The correction course of image direction is image to be detected in people
Pretreatment before face detection, the postrotational image that will acquire will scheme according to the angle of the portrait direction remedial frames in image
Image rotation goes to normotopia, i.e. non-tilt state, to improve the fault-tolerance of Face datection, guarantees subsequent recognition of face and face character
The precision and efficiency of detection.
In practical applications, method for detecting human face can be promoted comprising some by introducing the human face data of different ethnic groups
Detectability under specific ethnic group scene.
Step S102 carries out recognition of face to the image to be detected, obtains facial image.
In specific implementation, recognition of face can be carried out to image to be detected, identifies face in image to be detected
Position and quantity.
In specific implementation, recognition of face can be carried out to image to be detected using the first convolution neural network model.
In practical applications, the convolution kernel in the first convolution neural network model can be used as the " feature of the neural network
Description ".The weight of convolution kernel is continued to optimize in the feed-forward of Face datection, keeps convolutional neural networks model more flexible
Ground extracts the feature of facial image, while feature extraction also being made to become more careful and comprehensive.
In specific implementation, the process of recognition of face includes the candidate frame generated in the image to be detected, and is merged
The part that height is overlapped;The non-face candidate frame in image after removal merging;It identifies face key point position, obtains the people
Coordinate information, face confidence level and the key point coordinate of face image, to located the face area in the facial image of input
Domain, and each human face region is split reaches determining recognition of face main body, the removal background information and inspection unrelated with face
Look into the purpose of picture quality.Meanwhile passing through multitask (coordinate information, face confidence level and the key point coordinate of facial image)
Study reduces operation expense, improves recognition of face efficiency.
Step S103 carries out face character detection to the facial image, obtains face character.
In specific implementation, face character detection may include following at least two classes: head pose detection, face are corresponding
The face value score detection of gender and age detection, the detection of the mood of face and face.
In an embodiment of the present invention, the obtained facial image of step S102 is subjected to face character detection, to face
It is detected towards angle, age, gender, mood, face value score on personage head in image.For example, image warp to be detected
After crossing step 102~step S103 processing, corresponding facial image is obtained, and be known that corresponding people in facial image
Object quantity is one, and the head of the personage deflects to the right, is the higher middle-aged male of face value score, and front is without table this moment
Feelings.
In specific implementation, can using with the one-to-one convolutional neural networks model of the face character, to described
Facial image carries out face character detection.Image characteristics extraction is carried out using convolutional neural networks structure, in specific face character
Different network structures is designed in detection process to realize higher precision and efficiency.
Simultaneously as having preset different convolutional neural networks model structures improves the accurate of face character testing result
Property, while the process that face character is detected is decentralized, modularization, reduces algorithm iteration cost.
In specific implementation, preset second multi output convolutional neural networks model, preset third multi output convolution mind
It is equal through network model, preset 4th shallow-layer convolutional neural networks model and preset 5th shallow-layer convolutional neural networks model
Following at least one method can be used to carry out model training: simultaneously being filled with random pixel value, is random in hollow out random image region
Adjust brightness of image, Random-fuzzy filtering, randomised particulars filtering and random sharp filtering.
In practical applications, to preset second multi output convolutional neural networks model, preset third multi output volume
Product neural network model, preset 4th shallow-layer convolutional neural networks model and preset 5th shallow-layer convolutional neural networks mould
During the model training of type, for rich image training set, preferably to extract facial image feature, extensive model and prevent
Only model over-fitting can carry out data enhancing to data image.Can specifically include: rotation image, changes figure at clip image
As color difference, warp image feature, change picture size size, enhancing image noise (use gaussian noise, salt green pepper noise)
Deng.By increasing the data volume of model training, to improve the generalization ability of model;By increasing noise data, with lift scheme
Robustness.
In specific implementation, preset second multi output convolutional neural networks model, preset third multi output convolution mind
It is equal through network model, preset 4th shallow-layer convolutional neural networks model and preset 5th shallow-layer convolutional neural networks model
Following at least two recombination losses function can be used: the convolutional neural networks classification layer output distribution expectation and true value
Mean square error, the variance of convolutional neural networks classification layer output distribution, convolutional neural networks classification layer output with it is true
The cross entropy being grouped in fact.It is understood that according to the different demands of different user, above-mentioned three kinds of functions and right can be chosen
The weight answered constitutes recombination losses function, can also choose any two kinds of functions weighting composition recombination losses function, and the present invention exists
This is not repeated.
Face character detection may include head pose detection, the corresponding gender of face and age detection, the mood of face
The detection of the face value score of detection and face, is illustrated individually below.
Head pose refers to the direction on personage head, and popular understanding is the movement such as low new line, rotary head.Head pose inspection
Survey can be applied to dynamic In vivo detection, the prediction of personage's attention etc..Head appearance is indicated using Euler angle in the present invention
The degree of deflection of all directions in state.
It in specific implementation, can be using facial image described in preset second multi output convolutional neural networks model inspection
In head pose, obtain the Euler angle that face in the facial image deflects.Due to output be face three axis (x, y,
Z-axis) on angle, therefore the second multi output convolutional neural networks model be multi output structure, and then can simultaneously pass through more
Business (angle on three axis) study reduces operation expense, promotes recognition efficiency.
In specific implementation, using in facial image described in preset third multi output convolutional neural networks model inspection
The corresponding gender of face and age, and by obtained gender and age result while feeding back to the preset third multi output volume
Product neural network model, so that the preset third multi output convolutional neural networks model extracts the people of different dimensions again
Face feature.Operation expense is reduced by multitask (age, gender) study, feedback learning, promotes recognition efficiency.
In practical applications, the output layer of preset third multi output convolutional neural networks model can be with to age and gender
Classify simultaneously, wherein the forecast interval at age can be 10 to 80 years old.For example, gender and age in face character detection
Testing result is " male ", " 41 years old ".
It in specific implementation, can be using in facial image described in preset 4th shallow-layer convolutional neural networks model inspection
Face mood.Using shallow-layer convolutional neural networks model, the i.e. network structure of lightweight, compared to profound network mould
Type can promote detection efficiency.
In specific implementation, the mood of face may include following one of which: indignation, detest, it is frightened, happy, sad,
It is pleasantly surprised and poker-faced.For example, the mood testing result in face character detection is " poker-faced ".
Face value indicates attraction index of the face under Popular Aesthetics, is beaten in the present invention using Popular Aesthetics face
Divided data detects face value score.Specifically, totally 5 integers indicate face value score to 1-5, wherein 1 indicates that face value score is minimum,
5 indicate face value score highest.For example, the face value score testing result in face character detection is " 2.83 ".
In specific implementation, using the people in facial image described in preset 5th shallow-layer convolutional neural networks model inspection
The face value score of face.Using shallow-layer convolutional neural networks model, the i.e. network structure of lightweight, compared to profound network mould
Type can promote detection efficiency.
In conclusion carrying out recognition of face to image to be detected, facial image is obtained.The facial image is carried out again
Face character detection, obtains face character;The face character includes following at least two classes: the corresponding property of head pose, face
Other and age, the mood of face and face face value score.It, can be with by identification facial image and corresponding face character
Meet a variety of demands of Face datection.
Referring to Fig. 2, the embodiment of the invention also provides human face detection devices 20, comprising: acquiring unit 201, recognition of face
Unit 202 and face character detection unit 203;
Wherein, the acquiring unit 201, for obtaining image to be detected;
The face identification unit 202 obtains facial image for carrying out recognition of face to the image to be detected;
The face character detection unit 203 obtains face category for carrying out face character detection to the facial image
Property;The face character includes following at least two classes: the corresponding gender of head pose, face and age, face mood and
The face value score of face.
In specific implementation, face identification unit 202 can be used for carrying out face using the first convolution neural network model
Identification.
In specific implementation, face character detection unit 203 can be used for using one-to-one with the face character
Convolutional neural networks model carries out face character detection to the facial image.
In specific implementation, with the one-to-one convolutional neural networks model of the face character can using it is following at least
A kind of method progress model training: it is simultaneously filled with random pixel value, adjusts brightness of image at random, is random in hollow out random image region
Fuzzy filter, randomised particulars filtering and random sharp filtering.
In specific implementation, with the one-to-one convolutional neural networks model of the face character can using it is following at least
Two kinds of function weightings constitute recombination losses function: the convolutional neural networks classification layer output distribution expectation is square with true value
Error, the variance of convolutional neural networks classification layer output distribution, convolutional neural networks classification layer export and true point
The cross entropy of group.
In specific implementation, the acquiring unit 201, can be also used for: carry out to the direction of the image to be detected
It corrects, so that the portrait direction in the image after correction is non-tilt state.
In specific implementation, face identification unit 202 can be used for identifying the position of face in the image to be detected
And quantity.
In specific implementation, face identification unit 202 can be used for generating the candidate frame in the image to be detected,
And merge the part that height is overlapped;The non-face candidate frame in image after removal merging;It identifies face key point position, obtains
Coordinate information, face confidence level and the key point coordinate of the facial image.
In specific implementation, face character detection unit 203 can be used for using preset second multi output convolutional Neural
Network model detects the head pose in the facial image, obtains the Euler angle that face deflects in the facial image.
In specific implementation, face character detection unit 203 can be used for using preset third multi output convolutional Neural
Network model detects the corresponding gender of face and age in the facial image, and simultaneously by obtained gender and age result
The preset third multi output convolutional neural networks model is fed back to, so that the preset third multi output convolutional Neural
Network model extracts the face characteristic of different dimensions again.
In specific implementation, face character detection unit 203 can be used for using preset 4th shallow-layer convolutional Neural net
The mood of face in facial image described in network model inspection.
In specific implementation, face character detection unit 203 can be used for using preset 5th shallow-layer convolutional Neural net
The face value score of face in facial image described in network model inspection.
The embodiment of the invention also provides a kind of computer readable storage medium, computer readable storage medium is non-volatile
Property storage medium or non-transitory storage media, be stored thereon with computer instruction, the computer instruction executes this hair when running
The step of any described method for detecting human face provided in bright above-described embodiment.
The embodiment of the invention also provides a kind of human face detection device, including memory and processor, on the memory
It is stored with the computer instruction that can be run on the processor and executes sheet when the processor runs shown computer instruction
The step of any described method for detecting human face provided in invention above-described embodiment.
Those of ordinary skill in the art will appreciate that all or part of the steps in the various methods of above-described embodiment is can
It is completed with instructing relevant hardware by program, which can store in any computer readable storage medium storing program for executing, deposit
Storage media may include: ROM, RAM, disk or CD etc..
Although present disclosure is as above, present invention is not limited to this.Anyone skilled in the art are not departing from this
It in the spirit and scope of invention, can make various changes or modifications, therefore protection scope of the present invention should be with claim institute
Subject to the range of restriction.
Claims (26)
1. a kind of method for detecting human face characterized by comprising
Obtain image to be detected;
Recognition of face is carried out to the image to be detected, obtains facial image;
Face character detection is carried out to the facial image, obtains face character;The face character includes following at least two classes:
The face value score of the corresponding gender of head pose, face and age, the mood of face and face.
2. method for detecting human face as described in claim 1, which is characterized in that described to carry out face to the figure to be detected
Identification, comprising:
Recognition of face is carried out using the first convolution neural network model.
3. method for detecting human face as described in claim 1, which is characterized in that described to carry out face character to the facial image
Detection, comprising: use and the one-to-one convolutional neural networks model of the face character carry out face to the facial image
Detection of attribute.
4. method for detecting human face as claimed in claim 3, which is characterized in that described to be rolled up correspondingly with the face character
Product neural network model use following at least one method progress model training: hollow out random image region and with random pixel value
Filling adjusts brightness of image, Random-fuzzy filtering, randomised particulars filtering and random sharp filtering at random.
5. method for detecting human face as claimed in claim 3, which is characterized in that described to be rolled up correspondingly with the face character
Product neural network model constitutes recombination losses function using the weighting of following at least two function: the convolutional neural networks classification layer
Output distribution expectation and the mean square error of true value, variance, the convolution of convolutional neural networks classification layer output distribution
The cross entropy that neural network classification layer is exported and is really grouped.
6. method for detecting human face as described in claim 1, which is characterized in that after obtaining image to be detected, further includes:
The direction of the image to be detected is corrected, so that the portrait direction in the image after correction is non-tilt state.
7. method for detecting human face as described in claim 1, which is characterized in that described to carry out face to the image to be detected
Identification, comprising: the position of face and quantity in the identification image to be detected.
8. method for detecting human face as described in claim 1, which is characterized in that described to carry out face to the image to be detected
Identification, comprising:
The candidate frame in the image to be detected is generated, and merges the part that height is overlapped;In image after removal merging
Non-face candidate frame;It identifies face key point position, obtains coordinate information, face confidence level and the key of the facial image
Point coordinate.
9. method for detecting human face as described in claim 1, which is characterized in that described to carry out face character to the facial image
Detection, comprising: using the head pose in facial image described in preset second multi output convolutional neural networks model inspection, obtain
The Euler angle that face deflects into the facial image.
10. method for detecting human face as described in claim 1, which is characterized in that described to carry out face category to the facial image
Property detection, comprising: it is corresponding using the face in facial image described in preset third multi output convolutional neural networks model inspection
Gender and the age, and by obtained gender and age result while feeding back to the preset third multi output convolutional Neural net
Network model, so that the preset third multi output convolutional neural networks model extracts the face characteristic of different dimensions again.
11. method for detecting human face as described in claim 1, which is characterized in that described to carry out face category to the facial image
Property detection, comprising: using the mood of the face in facial image described in preset 4th shallow-layer convolutional neural networks model inspection.
12. method for detecting human face as described in claim 1, which is characterized in that described to carry out face category to the facial image
Property detection, comprising: using the face value of the face in facial image described in preset 5th shallow-layer convolutional neural networks model inspection
Score.
13. a kind of human face detection device characterized by comprising
Acquiring unit, for obtaining image to be detected;
Face identification unit obtains facial image for carrying out recognition of face to the image to be detected;
Face character detection unit obtains face character for carrying out face character detection to the facial image;The face
Attribute includes following at least two classes: the corresponding gender of head pose, face and the face value at age, the mood of face and face point
Number.
14. human face detection device as claimed in claim 13, which is characterized in that the face identification unit, for using the
One convolution neural network model carries out recognition of face.
15. human face detection device as claimed in claim 13, which is characterized in that the face character detection unit, for adopting
With with the one-to-one convolutional neural networks model of the face character, to the facial image carry out face character detection.
16. human face detection device as claimed in claim 15, which is characterized in that described one-to-one with the face character
Convolutional neural networks model uses following at least one method to carry out model training: hollow out random image region and with random pixel
Value filling adjusts brightness of image, Random-fuzzy filtering, randomised particulars filtering and random sharp filtering at random.
17. human face detection device as claimed in claim 15, which is characterized in that described one-to-one with the face character
Convolutional neural networks model constitutes recombination losses function using the weighting of following at least two function: the convolutional neural networks classification
Layer output distribution expectation and the mean square error of true value, the variance of convolutional neural networks classification layer output distribution, the volume
The cross entropy that product neural network classification layer is exported and is really grouped.
18. human face detection device as claimed in claim 13, which is characterized in that the acquiring unit is also used to: to it is described to
The direction of the image of detection is corrected, so that the portrait direction in the image after correction is non-tilt state.
19. human face detection device as claimed in claim 13, which is characterized in that the face identification unit, for identification institute
State the position of face and quantity in image to be detected.
20. human face detection device as claimed in claim 13, which is characterized in that the face identification unit, for generating
The candidate frame in image to be detected is stated, and merges the part that height is overlapped;The non-face candidate in image after removal merging
Frame;It identifies face key point position, obtains coordinate information, face confidence level and the key point coordinate of the facial image.
21. human face detection device as claimed in claim 13, which is characterized in that the face character detection unit, for adopting
Head pose in the facial image described in preset second multi output convolutional neural networks model inspection, obtains the face figure
The Euler angle of face deflection as in.
22. human face detection device as claimed in claim 13, which is characterized in that the face character detection unit, for adopting
The corresponding gender of face and age in the facial image described in preset third multi output convolutional neural networks model inspection, and
By obtained gender and age result while the preset third multi output convolutional neural networks model is fed back to, so that institute
State the face characteristic that preset third multi output convolutional neural networks model extracts different dimensions again.
23. human face detection device as claimed in claim 13, which is characterized in that the face character detection unit, for adopting
The mood of face in the facial image described in preset 4th shallow-layer convolutional neural networks model inspection.
24. human face detection device as claimed in claim 13, which is characterized in that the face character detection unit, for adopting
The face value score of face in the facial image described in preset 5th shallow-layer convolutional neural networks model inspection.
25. a kind of computer readable storage medium, computer readable storage medium is non-volatile memory medium or non-transient deposits
Storage media is stored thereon with computer instruction, which is characterized in that perform claim requires 1 to 12 when the computer instruction is run
Any one of described in method for detecting human face the step of.
26. a kind of human face detection device, including memory and processor, being stored on the memory can be on the processor
The computer instruction of operation, which is characterized in that perform claim requires 1 to 12 when the processor runs the computer instruction
The step of method for detecting human face described in one.
Priority Applications (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201910142852.7A CN109977781A (en) | 2019-02-26 | 2019-02-26 | Method for detecting human face and device, readable storage medium storing program for executing |
Applications Claiming Priority (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201910142852.7A CN109977781A (en) | 2019-02-26 | 2019-02-26 | Method for detecting human face and device, readable storage medium storing program for executing |
Publications (1)
Publication Number | Publication Date |
---|---|
CN109977781A true CN109977781A (en) | 2019-07-05 |
Family
ID=67077440
Family Applications (1)
Application Number | Title | Priority Date | Filing Date |
---|---|---|---|
CN201910142852.7A Pending CN109977781A (en) | 2019-02-26 | 2019-02-26 | Method for detecting human face and device, readable storage medium storing program for executing |
Country Status (1)
Country | Link |
---|---|
CN (1) | CN109977781A (en) |
Cited By (11)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN110598638A (en) * | 2019-09-12 | 2019-12-20 | Oppo广东移动通信有限公司 | Model training method, face gender prediction method, device and storage medium |
CN111160197A (en) * | 2019-12-23 | 2020-05-15 | 爱驰汽车有限公司 | Face detection method and device, electronic equipment and storage medium |
CN111539357A (en) * | 2020-04-28 | 2020-08-14 | 睿魔智能科技(深圳)有限公司 | Human body recognition model, training method and system thereof, storage medium and equipment |
CN111627086A (en) * | 2020-06-03 | 2020-09-04 | 上海商汤智能科技有限公司 | Head portrait display method and device, computer equipment and storage medium |
CN111862031A (en) * | 2020-07-15 | 2020-10-30 | 北京百度网讯科技有限公司 | Face synthetic image detection method and device, electronic equipment and storage medium |
CN111881746A (en) * | 2020-06-23 | 2020-11-03 | 安徽清新互联信息科技有限公司 | Face feature point positioning method and system based on information fusion |
WO2021047069A1 (en) * | 2019-09-11 | 2021-03-18 | 深圳传音控股股份有限公司 | Face recognition method and electronic terminal device |
CN112825119A (en) * | 2019-11-20 | 2021-05-21 | 北京眼神智能科技有限公司 | Face attribute judgment method and device, computer readable storage medium and equipment |
CN112926542A (en) * | 2021-04-09 | 2021-06-08 | 博众精工科技股份有限公司 | Performance detection method and device, electronic equipment and storage medium |
WO2021142975A1 (en) * | 2020-01-19 | 2021-07-22 | 深圳壹账通智能科技有限公司 | Artificial intelligence detection method and apparatus, terminal and computer-readable storage medium |
WO2022083504A1 (en) * | 2020-10-23 | 2022-04-28 | Huawei Technologies Co., Ltd. | Machine-learning model, methods and systems for removal of unwanted people from photographs |
Citations (10)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN105912990A (en) * | 2016-04-05 | 2016-08-31 | 深圳先进技术研究院 | Face detection method and face detection device |
CN106778627A (en) * | 2016-12-20 | 2017-05-31 | 北京奇虎科技有限公司 | Detect method, device and the mobile terminal of face face value |
CN107038422A (en) * | 2017-04-20 | 2017-08-11 | 杭州电子科技大学 | The fatigue state recognition method of deep learning is constrained based on space geometry |
CN107729838A (en) * | 2017-10-12 | 2018-02-23 | 中科视拓(北京)科技有限公司 | A kind of head pose evaluation method based on deep learning |
CN108154093A (en) * | 2017-12-13 | 2018-06-12 | 北京小米移动软件有限公司 | Face information recognition methods and device, electronic equipment, machine readable storage medium |
CN108229269A (en) * | 2016-12-31 | 2018-06-29 | 深圳市商汤科技有限公司 | Method for detecting human face, device and electronic equipment |
CN108898125A (en) * | 2018-07-10 | 2018-11-27 | 深圳市巨龙创视科技有限公司 | One kind being based on embedded human face identification and management system |
CN109271884A (en) * | 2018-08-29 | 2019-01-25 | 厦门理工学院 | Face character recognition methods, device, terminal device and storage medium |
CN109376713A (en) * | 2018-12-11 | 2019-02-22 | 天地伟业技术有限公司 | A kind of embedded human face detection device |
CN109376786A (en) * | 2018-10-31 | 2019-02-22 | 中国科学院深圳先进技术研究院 | A kind of image classification method, device, terminal device and readable storage medium storing program for executing |
-
2019
- 2019-02-26 CN CN201910142852.7A patent/CN109977781A/en active Pending
Patent Citations (10)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN105912990A (en) * | 2016-04-05 | 2016-08-31 | 深圳先进技术研究院 | Face detection method and face detection device |
CN106778627A (en) * | 2016-12-20 | 2017-05-31 | 北京奇虎科技有限公司 | Detect method, device and the mobile terminal of face face value |
CN108229269A (en) * | 2016-12-31 | 2018-06-29 | 深圳市商汤科技有限公司 | Method for detecting human face, device and electronic equipment |
CN107038422A (en) * | 2017-04-20 | 2017-08-11 | 杭州电子科技大学 | The fatigue state recognition method of deep learning is constrained based on space geometry |
CN107729838A (en) * | 2017-10-12 | 2018-02-23 | 中科视拓(北京)科技有限公司 | A kind of head pose evaluation method based on deep learning |
CN108154093A (en) * | 2017-12-13 | 2018-06-12 | 北京小米移动软件有限公司 | Face information recognition methods and device, electronic equipment, machine readable storage medium |
CN108898125A (en) * | 2018-07-10 | 2018-11-27 | 深圳市巨龙创视科技有限公司 | One kind being based on embedded human face identification and management system |
CN109271884A (en) * | 2018-08-29 | 2019-01-25 | 厦门理工学院 | Face character recognition methods, device, terminal device and storage medium |
CN109376786A (en) * | 2018-10-31 | 2019-02-22 | 中国科学院深圳先进技术研究院 | A kind of image classification method, device, terminal device and readable storage medium storing program for executing |
CN109376713A (en) * | 2018-12-11 | 2019-02-22 | 天地伟业技术有限公司 | A kind of embedded human face detection device |
Non-Patent Citations (2)
Title |
---|
产文涛: "基于卷积神经网络的人脸表情和性别识别", 《中国优秀硕士学位论文全文数据库信息科技辑》 * |
高彦杰: "《深度学习 核心技术、工具与案例解析》", 30 June 2018 * |
Cited By (14)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
WO2021047069A1 (en) * | 2019-09-11 | 2021-03-18 | 深圳传音控股股份有限公司 | Face recognition method and electronic terminal device |
CN110598638A (en) * | 2019-09-12 | 2019-12-20 | Oppo广东移动通信有限公司 | Model training method, face gender prediction method, device and storage medium |
CN112825119A (en) * | 2019-11-20 | 2021-05-21 | 北京眼神智能科技有限公司 | Face attribute judgment method and device, computer readable storage medium and equipment |
CN111160197A (en) * | 2019-12-23 | 2020-05-15 | 爱驰汽车有限公司 | Face detection method and device, electronic equipment and storage medium |
WO2021142975A1 (en) * | 2020-01-19 | 2021-07-22 | 深圳壹账通智能科技有限公司 | Artificial intelligence detection method and apparatus, terminal and computer-readable storage medium |
CN111539357A (en) * | 2020-04-28 | 2020-08-14 | 睿魔智能科技(深圳)有限公司 | Human body recognition model, training method and system thereof, storage medium and equipment |
CN111539357B (en) * | 2020-04-28 | 2024-01-23 | 睿魔智能科技(深圳)有限公司 | Human body recognition model, training method, training system, storage medium and training equipment thereof |
CN111627086A (en) * | 2020-06-03 | 2020-09-04 | 上海商汤智能科技有限公司 | Head portrait display method and device, computer equipment and storage medium |
CN111881746A (en) * | 2020-06-23 | 2020-11-03 | 安徽清新互联信息科技有限公司 | Face feature point positioning method and system based on information fusion |
CN111881746B (en) * | 2020-06-23 | 2024-04-02 | 安徽清新互联信息科技有限公司 | Face feature point positioning method and system based on information fusion |
CN111862031A (en) * | 2020-07-15 | 2020-10-30 | 北京百度网讯科技有限公司 | Face synthetic image detection method and device, electronic equipment and storage medium |
WO2022083504A1 (en) * | 2020-10-23 | 2022-04-28 | Huawei Technologies Co., Ltd. | Machine-learning model, methods and systems for removal of unwanted people from photographs |
US11676390B2 (en) | 2020-10-23 | 2023-06-13 | Huawei Technologies Co., Ltd. | Machine-learning model, methods and systems for removal of unwanted people from photographs |
CN112926542A (en) * | 2021-04-09 | 2021-06-08 | 博众精工科技股份有限公司 | Performance detection method and device, electronic equipment and storage medium |
Similar Documents
Publication | Publication Date | Title |
---|---|---|
CN109977781A (en) | Method for detecting human face and device, readable storage medium storing program for executing | |
Deng et al. | Learning to predict crisp boundaries | |
Nazeri et al. | Edgeconnect: Generative image inpainting with adversarial edge learning | |
CN106096538B (en) | Face identification method and device based on sequencing neural network model | |
Han et al. | Two-stage learning to predict human eye fixations via SDAEs | |
CN105825183B (en) | Facial expression recognizing method based on partial occlusion image | |
CN101714262B (en) | Method for reconstructing three-dimensional scene of single image | |
Jan et al. | Accurate facial parts localization and deep learning for 3D facial expression recognition | |
CN101751689A (en) | Three-dimensional facial reconstruction method | |
CN104794693B (en) | A kind of portrait optimization method of face key area automatic detection masking-out | |
CN111125403B (en) | Aided design drawing method and system based on artificial intelligence | |
CN110852326B (en) | Handwriting layout analysis and multi-style ancient book background fusion method | |
CN104268520A (en) | Human motion recognition method based on depth movement trail | |
CN109034012A (en) | First person gesture identification method based on dynamic image and video sequence | |
CN109978907A (en) | A kind of sitting posture of student detection method towards household scene | |
Zhang et al. | Multi-task convnet for blind face inpainting with application to face verification | |
CN110232390A (en) | Image characteristic extracting method under a kind of variation illumination | |
Chen et al. | Single depth image super-resolution using convolutional neural networks | |
Yao et al. | Contour-aware recurrent cross constraint network for salient object detection | |
Li et al. | Line drawing guided progressive inpainting of mural damages | |
Hu et al. | Resolution-aware 3D morphable model | |
Liu et al. | A3GAN: An attribute-aware attentive generative adversarial network for face aging | |
Ling et al. | A facial expression recognition system for smart learning based on YOLO and vision transformer | |
Ashtari et al. | Reference based sketch extraction via attention mechanism | |
Xia et al. | Lazy texture selection based on active 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 | ||
RJ01 | Rejection of invention patent application after publication |
Application publication date: 20190705 |
|
RJ01 | Rejection of invention patent application after publication |