CN109934058A - Face image processing process, device, electronic equipment, storage medium and program - Google Patents
Face image processing process, device, electronic equipment, storage medium and program Download PDFInfo
- Publication number
- CN109934058A CN109934058A CN201711352295.9A CN201711352295A CN109934058A CN 109934058 A CN109934058 A CN 109934058A CN 201711352295 A CN201711352295 A CN 201711352295A CN 109934058 A CN109934058 A CN 109934058A
- Authority
- CN
- China
- Prior art keywords
- key point
- face
- facial image
- neural network
- point
- 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.)
- Granted
Links
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/21—Design or setup of recognition systems or techniques; Extraction of features in feature space; Blind source separation
- G06F18/214—Generating training patterns; Bootstrap methods, e.g. bagging or boosting
- G06F18/2155—Generating training patterns; Bootstrap methods, e.g. bagging or boosting characterised by the incorporation of unlabelled data, e.g. multiple instance learning [MIL], semi-supervised techniques using expectation-maximisation [EM] or naïve labelling
-
- 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
Landscapes
- Engineering & Computer Science (AREA)
- Theoretical Computer Science (AREA)
- Physics & Mathematics (AREA)
- Data Mining & Analysis (AREA)
- Life Sciences & Earth Sciences (AREA)
- Artificial Intelligence (AREA)
- General Physics & Mathematics (AREA)
- General Engineering & Computer Science (AREA)
- Evolutionary Computation (AREA)
- Computer Vision & Pattern Recognition (AREA)
- Computational Linguistics (AREA)
- Bioinformatics & Computational Biology (AREA)
- Bioinformatics & Cheminformatics (AREA)
- Health & Medical Sciences (AREA)
- Biomedical Technology (AREA)
- Biophysics (AREA)
- Evolutionary Biology (AREA)
- General Health & Medical Sciences (AREA)
- Molecular Biology (AREA)
- Computing Systems (AREA)
- Mathematical Physics (AREA)
- Software Systems (AREA)
- Image Analysis (AREA)
- Processing Or Creating Images (AREA)
- Image Processing (AREA)
Abstract
The application embodiment discloses a kind of face image processing process, the method, apparatus for training neural network, electronic equipment, computer readable storage medium and computer program, and method therein, which specifically includes that, is input to facial image to be processed in neural network;The face key point information including at least one three-dimensional motionless key point of the facial image to be processed is extracted via the neural network.
Description
Technical field
This application involves computer vision techniques, more particularly, to a kind of face image processing process, face image processing
Device, the method for training neural network, the device for training neural network, electronic equipment, computer-readable storage medium
Matter and computer program.
Background technique
Face key point is a part indispensable in many applications such as recognition of face.Accurately determine face key
Point is especially accurately determined with big expression or there are the face key point of the face of wide-angle deflection, not only favorably
It is corrected in the relevant position to face, and is conducive to enhance the semantic information of face.
Summary of the invention
The application embodiment provides a kind of face image processing technical solution.
According to the application embodiment one aspect, a kind of face image processing process is provided, this method comprises:
Facial image to be processed is input in neural network;Include via what the neural network extracted the facial image to be processed
Face key point information including at least one three-dimensional motionless key point.
In one embodiment of the application, the face key point information that the neural network extracts includes: people to be processed
The face key point information being located at the position that is not blocked at position and/or is blocked in face image.
In the another embodiment of the application, the motionless key point of three-dimensional includes: eyelid center key point, nose lower edge
At least one of center key point, lip lip peak key point, lip center key point and chin center key point.
In the application a further embodiment, eyelid center key point includes: left eye upper eyelid center key point, a left side
At least one of eyelid center key point, right eye upper eyelid center key point and right eye eyelid center key point now.
In the application a further embodiment, lip center key point include: upper lip upper lip line center key point,
In upper lip lower lip line center key point, lower lip upper lip line center key point and lower lip lower lip line center key point extremely
It is one few.
In the application a further embodiment, lip lip peak key point includes: the left lip peak key point of lip and mouth
At least one of right lip peak key point of lip.
In the application a further embodiment, the face key point information that the neural network extracts includes: 68,
106,240,252 or 270 face key point informations.
In the application a further embodiment, the neural network is to utilize to be labeled with: motionless including at least one three-dimensional
The facial image sample of face key point markup information including key point, made of training.
In the application a further embodiment, the method also includes: obtain multiple facial image samples, wherein described
Facial image sample is labeled with: the face key point markup information including at least one three-dimensional motionless key point;Using institute
State neural network described in multiple facial image sample trainings.
In the application a further embodiment, the facial image sample is labeled with being located at not in facial image sample
The face key point markup information for being blocked at position and/or being blocked at position.
In the application a further embodiment, the face key point markup information of the facial image sample includes: 68,
106,240,252 or 270 face key point markup informations.
In the application a further embodiment, before the neural network described in the utilization facial image sample training, also
Including obtaining the face key point markup information using following step: determining at least one portion of face in facial image sample
The curve control point of position;According to the curve control point forming curves;Determine three as facial image sample on the curve
Tie up motionless key point;In at least side of the motionless key point of three-dimensional of facial image sample, the point on curve is chosen;According to described
The motionless key point of the three-dimensional of the point of selection and the facial image sample forms face key point markup information.
In the application a further embodiment, as the motionless pass of three-dimensional of facial image sample on the determination curve
Key point, which comprises determining that, meets pre- spacing at a distance from the motionless key point of three-dimensional marked on facial image sample on the curve
The motionless key point of three-dimensional from desired point, by the point as facial image sample.
It is described to utilize neural network packet described in the multiple facial image sample training in the application a further embodiment
It includes: facial image sample being inputted into the neural network, the people of the facial image sample is extracted via the neural network
Face key point information;It is calculated between the different key points at least one organ according to the face key point information extracted
First distance;Second between the different key points at least one organ is calculated according to the face key point markup information
Distance;It is exercised supervision study via first-loss function to the neural network based on the first distance and second distance.
In the application a further embodiment, the face key point information extracted according to calculates at least one
The first distance between different key points on organ includes: for the upper eyelid line key point of left eye by extracting and corresponding
Left eye palpebra inferior line key point formed any group of key point, calculate separately the distance between two key points in group,
To obtain at least one first distance based on left eye eyelid line;And/or it is crucial for the upper eyelid line of the right eye by extracting
Any group of key point that point is formed with the palpebra inferior line key point of corresponding right eye, calculates separately between two key points in group
Distance, to obtain at least one first distance based on right eye eyelid line;And/or for the upper lip upper lip by extracting
Any group of key point that line key point is formed with corresponding lower lip lower lip line key point calculates separately two key points in group
The distance between, to obtain at least one first distance based on outer lip line;And/or for the upper lip lower lip by extracting
Any group of key point that line key point is formed with corresponding lower lip upper lip line key point calculates separately two key points in group
The distance between, to obtain at least one first distance based on epipharynx line.
In the application a further embodiment, the first-loss function includes: for first based on left eye eyelid line
The first-loss function of distance, for the first distance based on right eye eyelid line first-loss function, for be based on outer lip line
First distance first-loss function and/or first-loss function for the first distance based on epipharynx line.
It is described to utilize neural network packet described in the multiple facial image sample training in the application a further embodiment
Include: at least one set of key point formed for the face key point face key point corresponding with mark by extracting determines
The distance between two key points in group, to obtain at least one drift distance;Based on the drift distance via the second damage
Function is lost to exercise supervision study to the neural network.
In terms of according to the application embodiment another one, a kind of method for training neural network is provided, it should
Method includes: to obtain multiple facial image samples, wherein the facial image sample is labeled with: three-dimensional not including at least one
Face key point markup information including dynamic key point;Utilize the multiple facial image sample training neural network.
In one embodiment of the application, the facial image sample be labeled in facial image sample be located at not by
Face key point markup information at blocking position and/or at the position that is blocked.
In the another embodiment of the application, the face key point markup information of the facial image sample includes: 68,
106,240,252 or 270 face key point markup informations.
In the application a further embodiment, the method utilizes the multiple facial image sample training nerve described
Before network, further includes: determine the curve control point at least one position of face in facial image sample;According to the curve
Control point forming curves;Determine the motionless key point of three-dimensional on the curve as facial image sample;In facial image sample
The motionless key point of three-dimensional at least side, choose the point on curve;According to the point of the selection and the facial image sample
This motionless key point of three-dimensional forms face key point markup information.
In the application a further embodiment, as the motionless pass of three-dimensional of facial image sample on the determination curve
Key point, which comprises determining that, meets pre- spacing at a distance from the motionless key point of three-dimensional marked on facial image sample on the curve
The motionless key point of three-dimensional from desired point, by the point as facial image sample.
It is described to include: using the multiple facial image sample training neural network in the application a further embodiment
Facial image sample is inputted into neural network, the face key point of the facial image sample is extracted via the neural network
Information;According to the face key point information extracted calculate between the different key points at least one organ first away from
From;The second distance between the different key points at least one organ is calculated according to the face key point markup information;Base
It exercises supervision study via first-loss function to the neural network in the first distance and second distance.
In the application a further embodiment, the face key point information extracted according to calculates at least one
The first distance between different key points on organ includes: for the upper eyelid line key point of left eye by extracting and corresponding
Left eye palpebra inferior line key point formed any group of key point, calculate separately the distance between two key points in group,
To obtain at least one first distance based on left eye eyelid line;And/or it is crucial for the upper eyelid line of the right eye by extracting
Any group of key point that point is formed with the palpebra inferior line key point of corresponding right eye, calculates separately between two key points in group
Distance, to obtain at least one first distance based on right eye eyelid line;And/or for the upper lip upper lip by extracting
Any group of key point that line key point is formed with corresponding lower lip lower lip line key point calculates separately two key points in group
The distance between, to obtain at least one first distance based on outer lip line;And/or for the upper lip lower lip by extracting
Any group of key point that line key point is formed with corresponding lower lip upper lip line key point calculates separately two key points in group
The distance between, to obtain at least one first distance based on epipharynx line.
In the application a further embodiment, the first-loss function includes: for first based on left eye eyelid line
The first-loss function of distance, for the first distance based on right eye eyelid line first-loss function, for be based on outer lip line
First distance first-loss function and/or first-loss function for the first distance based on epipharynx line.
It is described to utilize neural network packet described in the multiple facial image sample training in the application a further embodiment
Include: at least one set of key point formed for the face key point face key point corresponding with mark by extracting determines
The distance between two key points in group, to obtain at least one drift distance;Based on the drift distance via the second damage
Function is lost to exercise supervision study to the neural network.
According to the application embodiment wherein another aspect, a kind of face image processing device is provided, the device packet
It includes: input picture module, for facial image to be processed to be input in neural network;And crucial point module is obtained, it is used for
The face including at least one three-dimensional motionless key point of the facial image to be processed is extracted via the neural network
Key point information.
In one embodiment of the application, the face image processing device further include: for training the dress of neural network
It sets, the device for being used to train neural network includes: to obtain sample module, for obtaining multiple facial image samples, wherein
The facial image sample is labeled with: the face key point markup information including at least one three-dimensional motionless key point;Instruction
Practice module, for utilizing neural network described in the multiple facial image sample training.
It is described for training the device of neural network in the another embodiment of the application further include: labeling module, it is described
Labeling module is used for: determining the curve control point at least one position of face in facial image sample;According to the curve control
System point forming curves;Determine the motionless key point of three-dimensional on the curve as facial image sample;In facial image sample
At least side of three-dimensional motionless key point, chooses the point on curve;According to the point of the selection and the facial image sample
The motionless key point of three-dimensional form face key point markup information.
In the application a further embodiment, the training module is specifically used for: facial image sample is inputted the mind
Through network, the face key point information of the facial image sample is extracted via the neural network;It is extracted according to described
Face key point information calculate the first distance between the different key points at least one organ;It is crucial according to the face
Point markup information calculates the second distance between the different key points at least one organ;Based on the first distance and second
Distance exercises supervision study to the neural network via first-loss function.
In the application a further embodiment, the training module is specifically used for: for the face key point by extracting
At least one set of key point that face key point corresponding with mark is formed, determines the distance between two key points in group,
To obtain at least one drift distance;It is exercised supervision via the second loss function to the neural network based on the drift distance
Study.
According to the application embodiment wherein another aspect, provide it is a kind of for training the device of neural network, should
Device includes: to obtain sample module, for obtaining multiple facial image samples, wherein the facial image sample is labeled with: packet
Include the face key point markup information including at least one three-dimensional motionless key point;And training module, it is described more for utilizing
Neural network described in a facial image sample training.
In one embodiment of the application, described device further include: labeling module, the labeling module are used for: determining people
The curve control point at least one position of face in face image sample;According to the curve control point forming curves;Determine institute
State the motionless key point of three-dimensional on curve as facial image sample;Facial image sample the motionless key point of three-dimensional at least
The point on curve is chosen in side;It is formed according to the motionless key point of the three-dimensional of the point of the selection and the facial image sample
Face key point markup information.
In the another embodiment of the application, the training module is specifically used for: facial image sample is inputted the mind
Through network, the face key point information of the facial image sample is extracted via the neural network;It is extracted according to described
Face key point information calculate the first distance between the different key points at least one organ;It is crucial according to the face
Point markup information calculates the second distance between the different key points at least one organ;Based on the first distance and second
Distance exercises supervision study to the neural network via first-loss function.
In the application a further embodiment, the training module is specifically used for: for the face key point by extracting
At least one set of key point that face key point corresponding with mark is formed, determines the distance between two key points in group,
To obtain at least one drift distance;It is exercised supervision via the second loss function to the neural network based on the drift distance
Study.
According to another aspect of the application embodiment, a kind of electronic equipment is provided, comprising: memory, for depositing
Store up computer program;Processor, for executing the computer program stored in the memory, and the computer program is held
When row, the step in any face image processing process in the application is realized, or realize any for instructing in the application
Practice the step in the method for neural network.
According to another aspect of the application embodiment, a kind of computer readable storage medium is provided, is stored thereon with
Computer program when the computer program is executed by processor, is realized in any face image processing process in the application
Step, or realize the step in any method for training neural network in the application.
According to another aspect of the application embodiment, a kind of computer program, including computer instruction are provided, works as institute
When stating computer instruction and running in the processor of equipment, the step in any face image processing process in the application is realized
Suddenly, the step in any method for training neural network or in realization the application.
Method, apparatus, electronics based on face image processing process provided by the present application, for training neural network are set
Standby, computer readable storage medium and computer program, the application is by extracting facial image to be processed using neural network
In the motionless key point of three-dimensional, from regardless of whether face deflects, also specifically regardless of people's facial expression, it is three-dimensional not
Privileged site this semanteme on face represented by dynamic key point be it is fixed, will not change, therefore, the application can
The privileged site in face is accurately oriented with the face for variform, this is for image rendering, Face Changing processing, U.S. face
It is very important for the various applications such as processing, makeups processing, recognition of face, face state-detection and expression detection.By
This is it is found that the application is conducive to improve the accuracy for extracting face key point.
Below by drawings and embodiments, the technical solution of the application is described in further detail.
Detailed description of the invention
The attached drawing for constituting part of specification describes presently filed embodiment, and together with description for solving
Release the principle of the application.
The application can be more clearly understood according to following detailed description referring to attached drawing, in which:
Fig. 1 is the flow chart of one embodiment of face image processing process of the application;
Fig. 2 is the schematic diagram of the motionless key point of three-dimensional of the application;
Fig. 3 is the flow chart of an embodiment of the method for training neural network of the application;
Fig. 4 is the flow chart of a specific example of the method for training neural network of the application;
Fig. 5 is the flow chart of the mark facial image sample of the application;
Fig. 6 is the structural schematic diagram of one embodiment of face image processing device of the application;
Fig. 7 is the structural schematic diagram of an embodiment of the device for training neural network of the application;
Fig. 8 is the block diagram for realizing an example devices of the application embodiment.
Specific embodiment
Carry out the various exemplary embodiments of detailed description of the present invention now with reference to attached drawing.It should also be noted that unless in addition having
Body explanation, the unlimited system of component and the positioned opposite of step, numerical expression and the numerical value otherwise illustrated in these embodiments is originally
The range of invention.
Simultaneously, it should be appreciated that for ease of description, the size of various pieces shown in attached drawing is not according to reality
Proportionate relationship draw.
Be to the description only actually of at least one exemplary embodiment below it is illustrative, never as to the present invention
And its application or any restrictions used.
Technology, method and apparatus known to person of ordinary skill in the relevant may be not discussed in detail, but suitable
In the case of, the technology, method and apparatus should be considered as part of specification.
It should also be noted that similar label and letter indicate similar terms in following attached drawing, therefore, once a certain Xiang Yi
It is defined in a attached drawing, then in subsequent attached drawing does not need that it is further discussed.
The embodiment of the present invention can be applied to the electronic equipments such as terminal device, computer system, server, can with it is numerous
Other general or specialized computing system environments or configuration operate together.Suitable for electric with terminal device, computer system, server etc.
The example of well-known terminal device, computing system, environment and/or configuration that sub- equipment is used together includes but is not limited to:
Personal computer system, thin client, thick client computer, hand-held or laptop devices, is based on microprocessor at server computer system
System, set-top box, programmable consumer electronics, NetPC Network PC, little type Ji calculate machine Xi Tong ﹑ large computer system and
Distributed cloud computing technology environment, etc. including above-mentioned any system.
The electronic equipments such as terminal device, computer system, server can be in the department of computer science executed by computer system
It is described under the general context of system executable instruction (such as program module).In general, program module may include routine, program, mesh
Beacon course sequence, component, logic, data structure etc., they execute specific task or realize specific abstract data type.Meter
Calculation machine systems/servers can be implemented in distributed cloud computing environment, and in distributed cloud computing environment, task is by by logical
What the remote processing devices of communication network link executed.In distributed cloud computing environment, it includes storage that program module, which can be located at,
On the Local or Remote computing system storage medium of equipment.
Exemplary embodiment
Fig. 1 is the flow chart of the application method one embodiment.As shown in Figure 1, the embodiment method includes: step S100
And step S110.
S100, facial image to be processed is input in neural network.
In an optional example, picture or photo that the facial image to be processed in the application can be static for presentation
Equal images, or the video frame etc. in dynamic video is presented.Face in the facial image to be processed can be positive face,
It or can also be that there are big angles there are the side face of small angle deflection (for example, face deflection amplitude is less than predetermined angular)
The side face etc. of degree deflection (for example, face deflection amplitude is not less than predetermined angular).In addition, the facial image to be processed in the application
In face can be the face with big expression, for example, the positive face of big expression is had, for example, with big expression and there are small
The side face of angular deflection, for another example with big expression and there are side faces of wide-angle deflection etc..Big expression in the application is usual
Refer to the expression for making the form of at least one organ of face generate large change compared with its normal morphology.It is big in the application
Expression is sometimes referred to as complex expression.The application does not limit the deflection angle and table of the face in facial image to be processed
The specific forms such as feelings.
In an optional example, the neural network in the application, which is mainly used for extracting in facial image to be processed, includes
Face key point information including at least one three-dimensional motionless key point.The network structure of the neural network can be according to extraction people
The actual demand flexible design of face key point information, the embodiment of the present application are not intended to limit the specific network structure of the neural network.
For example, the neural network of the application can include but is not limited to convolutional layer, elu layers of linear R, pond layer and full articulamentum
Deng the number of plies that neural network is included is more, then network is deeper;For another example the network structure of the neural network of the application can be with
Using but be not limited to ALexNet, depth residual error network (Deep Residual Network, ResNet) or VGGnet (Visual
Geometry Group Network, visual geometric group network) etc. network structure used by neural networks.
It is closed since the neural network of the application both can be used for extracting face for the facial image to be processed based on positive face
Key point information can be used for extracting face key point information, therefore, the application for the facial image to be processed based on side face
Another Processing with Neural Network base can be utilized to avoid a to be processed facial image of the Processing with Neural Network based on positive face is utilized
In the facial image to be processed of side face, the video frame continuously played is handled by the switching of neural network, and is caused
Face key point there is a phenomenon where jump;It is mentioned from the video frame continuously played to which the application is conducive to raising neural network
Take the continuity of face key point.
S110, extracted via neural network facial image to be processed including at least one three-dimensional motionless key point
Face key point information.
In an optional example, the motionless key point of three-dimensional in the application refers to that face having in three dimensions is pre-
The point of attribute justice.Three-dimensional motionless key point can be projected in the point on two-dimensional surface by it to indicate.That is, the application
In the motionless key point of three-dimensional be point on two-dimensional surface, the point on the two-dimensional surface is that face having in three dimensions is pre-
The point of attribute justice, the subpoint on two-dimensional surface.Three-dimensional motionless key point is that have the face of preset semanteme crucial
Point, i.e., three-dimensional motionless key point can explicitly represent a privileged site on face, and no matter whether face deflects,
Also specifically regardless of people's facial expression, this semanteme of privileged site represented by three-dimensional motionless key point is fixed, no
It can change;That is, no matter whether face deflects, also regardless of people's facial expression, a three-dimensional is motionless
Key point always indicates a privileged site on face, can be with people although coordinate of the privileged site in three-dimensional coordinate
The deflection of face and the variation of expression and change, however, always show on face one of the privileged site fixes with certain
Semantic point.
In an optional example, the three-dimensional that the neural network in the application is extracted from facial image to be processed is motionless
Key point may include: crucial along center key point, lip lip peak key point, lip center under eyelid center key point, nose
Point and chin center key point in one or more.Eyelid center therein key point generally includes: left eye upper eyelid
Center key point, oculus sinisterlower lid center key point, right eye upper eyelid center key point and right eye eyelid center key point
In one or more.Lip center therein key point generally includes: upper lip upper lip line center key point (i.e. upper lip
Upper contour line center key point), upper lip lower lip line center key point (i.e. upper lip lower contour center key point), lower lip
Upper lip line center key point (i.e. contour line center key point on lower lip) and lower lip lower lip line center key point (descend mouth
One or more of lip lower contour center key point).Lip lip therein peak key point generally includes: closing at the left lip peak of lip
One or more in key point and the right lip peak key point of lip.
In an optional example, the neural network of the application can extract 12 three-dimensionals from facial image to be processed
Motionless key point, as shown in Fig. 2, 12 motionless key points of three-dimensional in the application are respectively as follows: right eye upper eyelid center key point
1, right eye eyelid center key point 2, left eye upper eyelid center key point 3, oculus sinisterlower lid center key point 4, nose lower edge
The right lip peak key point 6 of center key point 5, lip, upper lip upper lip line center key point 7, the left lip peak key point 8 of lip, upper mouth
Lip lower lip line center key point 9, lower lip upper lip line center key point 10, lower lip lower lip line center key point 11 and chin
Center key point 12.Face in Fig. 2 is although with the deflection of certain angle, and still, the motionless key point of above-mentioned 12 three-dimensionals is still
So there is its predetermined semanteme, for example, using the line of the left corners of the mouth and the right corners of the mouth in Fig. 2 as x coordinate axis, and by the left corners of the mouth and
In the case where the origin that the midpoint of the line of the right corners of the mouth intersects as x-axis with y-axis, upper lip upper lip line center key point 7, upper mouth
Lip lower lip line center key point 9, lower lip upper lip line center key point 10 and lower lip lower lip line center key point 11 are not
On the y axis.That is, upper lip upper lip line center key point 7 is not upper lip upper lip line in Fig. 2 (i.e. on upper lip
Contour line) central point in physical coordinates meaning, but face in three dimensions there is upper lip upper lip line central point
Semantic point, the point being projected on two-dimensional surface;Likewise, upper lip lower lip line center key point 9 is not upper in Fig. 2
Central point of the lip lower lip line (i.e. upper lip lower contour) in physical coordinates meaning, but the tool of face in three dimensions
There is the point of upper lip lower lip line central point semanteme, the point being projected on two-dimensional surface;Lower lip upper lip line center key point 10
It is not central point of the lower lip upper lip line (i.e. contour line on lower lip) in physical coordinates meaning in Fig. 2, but face
The point with lower lip upper lip line central point semanteme in three dimensions, the point being projected on two-dimensional surface;Under lower lip
Lip line center key point 11 is not lower lip lower lip line (i.e. lower lip lower contour) in Fig. 2 in physical coordinates meaning
Central point, but the point with lower lip lower lip line central point semanteme of face in three dimensions, are projected in two-dimensional surface
On point.
In an optional example, the face that the neural network in the application is extracted from facial image to be processed is crucial
Point information not only may include: the face key point information being located at unobstructed position in facial image to be processed, can be with
It include: the face key point information being located at the position that is blocked in facial image to be processed.Face to be processed in the application
The face key point information being located at the position that is blocked in image may be considered, and being located in facial image to be processed is hidden
Gear sets the face key point information obtained based on pivottable mode at place.The application is closed by extracting the face at the position that is blocked
Key point information is conducive to accurately delineating to the form of face, so that the face key point information for extracting neural network can
To be preferably suitable for image rendering, the Face Changing processing, U.S. face processing, makeups processing, recognition of face, face state-detection of face
And the application such as expression detection.
In an optional example, the face that the neural network in the application is extracted from facial image to be processed is crucial
Point information generally includes: (i.e. face key point is in facial image to be processed for face key point number and face key point coordinate
Coordinate system in coordinate), for example, neural network extracts 21-300 face key point respectively from facial image to be processed
Number and respective coordinate.Above-mentioned 21-300 face key point can be specially 68,106,240,252 or
270 etc..Certainly, the neural network in the application can also extract greater number or more from facial image to be processed
The face key point information of small number, for example, the neural network in the application only extracted from facial image to be processed it is above-mentioned
12 motionless key point informations of three-dimensional.The face key point information extracted from facial image to be processed is not limited from the application
Particular number.
The face key point information including three-dimensional motionless key point that the application obtains can be used for the image of face
The applications such as rendering, Face Changing processing, U.S. face processing, makeups processing, recognition of face, face state-detection and expression detection, this Shen
The concrete application range of the face key point information extracted via neural network is not limited please.
In an optional example, the application can use multiple facial image samples of sample data concentration to nerve net
Network is trained.The facial image sample that sample data is concentrated has face key point markup information, face key point mark letter
It should include: at least one three-dimensional motionless key point markup information in breath, for example, facial image sample is closed with 21-300 face
Key point markup information, this 21-300 face key point markup information can be specially 68,106,240,252 or
270 face key point markup informations include 12 motionless key points of three-dimensional in this 21-300 face key point markup information
Markup information.The application does not limit the particular number and face for the face key point markup information that facial image sample is had
The particular number for the motionless key point markup information of three-dimensional that key point markup information is included.
In an optional example, the face key point markup information that facial image sample is had includes the people being blocked
Face key point markup information, i.e. the application be not crucial only for the face being located at non-blocking position in facial image sample
Point is labeled, and can be also labeled for the face key point being located at the position that is blocked in facial image sample.One
Optional example, it is 106 that the application, which can be set in advance in the face total number of keypoints amount in facial image sample needed to mark,
A, regardless of the deflection direction of the face in facial image sample and expression, the application is directed to any facial image sample
106 face key point markup informations should all be marked out.In addition, the motionless key point of three-dimensional either in facial image sample,
Or other face key points in addition to three-dimensional motionless key point in facial image sample, in facial image sample
For the face key point at the position that is blocked, the application is that the perspective that the coordinate that it is marked is the face key point is sat
Mark.
In an optional example, the process of the face key point markup information in facial image sample is arranged such as in the application
Shown in Fig. 3.
In Fig. 3, S300, curve control point on face in facial image sample is determined, and according to curve controlled dot
At curve.
In an optional example, all face key points are usually to pass through a plurality of curve to be based on inserting in facial image sample
What the mode of value was formed, each curve is usually to pass through its curve control point to be formed;For example, in facial image sample
Left eye for, the application should obtain the multiple upper eyelid line traffic control points and multiple palpebra inferior line traffic control points of left eye, and using more
A upper eyelid line traffic control point forms the upper eyelid line of left eye, and the palpebra inferior line of left eye is formed using multiple palpebra inferior lines;For people
For right eye in face image sample, the application should obtain right eye multiple upper eyelid line traffic control points and multiple palpebra inferior line traffic controls
Point, and using the upper eyelid line of multiple upper eyelid line traffic control points formation right eye, utilize multiple palpebra inferior lines to form the lower eye of right eye
Eyelid line;For the nose in facial image sample, the application should obtain multiple nose lower edges control point, and utilize multiple
Nose lower edge control point formed nose it is following along;For the mouth in facial image sample, the application should be obtained
Multiple upper lip upper lip lines control point, multiple upper lip lower lip lines control point, multiple lower lip upper lip lines control point and multiple
Lower lip lower lip line control point, and upper lip upper lip line is formed using multiple upper lip upper lip lines control point, utilize multiple upper mouths
Lip lower lip line control point forms upper lip upper lip line, forms lower lip upper lip line, benefit using multiple lower lip upper lip lines control point
Lower lip upper lip line is formed with multiple lower lip lower lip lines control point;For the face mask in facial image sample, this
Application should obtain multiple face mask control points, and form face mask using multiple face mask control points.
In an optional example, the application pixel-by-pixel by the way of interpolation, can pass through using between curve control point
Corresponding multiple curve control points form corresponding curve.The application does not limit the specific reality by curve control point forming curves
Existing mode.
S310, the motionless key point of three-dimensional on curve as facial image sample is determined.
In an optional example, the application can determine on curve with marked on facial image sample three-dimensional it is motionless
The distance of key point meets the point of preset distance requirement, and will meet three of the point of preset distance requirement as facial image sample
Tie up motionless key point, optionally, the application can determine on curve with the motionless key of three-dimensional that is marked on facial image sample
Point is apart from nearest point, and the motionless key point of three-dimensional by the nearest point of the distance as facial image sample.
In an optional example, label has three-dimensional motionless key on the facial image sample in the application
Point, for example, marking eye under right eye upper eyelid center key point, right eye on facial image sample by way of manually marking
Along center key point, mouth under eyelid center key point, left eye upper eyelid center key point, oculus sinisterlower lid center key point, nose
The right lip peak key point of lip, upper lip upper lip line center key point, the left lip peak key point of lip, upper lip lower lip line center are crucial
Point, lower lip upper lip line center key point, lower lip lower lip line center key point and chin center key point, this 12 three-dimensionals
The coordinate of motionless key point.
In an optional example, the motionless key point of three-dimensional that handmarking goes out on facial image sample is usually corresponding
One curve.For any motionless key point of three-dimensional marked on facial image sample, the application can be in the three-dimensional
Near motionless key point, select from the corresponding curve of the motionless key point of the three-dimensional at 1 points, and according to marking
The coordinate of three-dimensional motionless key point and the coordinate of the point on curve calculate each point selected and the motionless key point of the three-dimensional it
Between distance, so as to select with the motionless key point of the three-dimensional apart from the smallest point;The application can use the distance most
The above-mentioned motionless key point of three-dimensional marked of small point substitution.
The application can also be using the motionless key point of three-dimensional of other modes setting facial image sample, for example, the application
The motionless key point of three-dimensional that directly handmarking can be gone out is as the motionless key point of three-dimensional of facial image sample.The application is not
The specific implementation of the motionless key point of three-dimensional of limitation setting facial image sample.
S320, the point on at least side of the motionless key point of three-dimensional of facial image sample, selection curve.
In an optional example, the application would generally choose respectively quantity phase from the left and right sides of three-dimensional motionless key point
Deng point.The application can choose the point on curve by way of interpolation;For example, if the curve be simple eye upper eyelid line or
Person's palpebra inferior line can be then inserted into (such as uniform insertion) except three-dimensional motionless key point on upper eyelid line or palpebra inferior line
9-14 (such as 10) a point except (such as upper eyelid center key point or palpebra inferior center key point);For another example if the song
Line is the upper lip upper lip line of mouth, then can be inserted on upper lip upper lip line (such as uniform insertion) except three-dimensional motionless key
13-18 except point (the right lip peak key point of such as lip, upper lip upper lip line center key point and the left lip peak key point of lip)
(such as 14) a point;For another example if the curve is the lower lip upper lip line or lower lip lower lip line of mouth, it can be in lip
Insertion (such as uniform insertion) is except three-dimensional motionless key point (such as lower lip upper lip line center key point on upper lip line or lower lip line
Or lower lip lower lip line key point) except 12-17 (such as 14) a point;For another example if the curve is face mask curve,
Can be then inserted on face mask curve (such as uniform insertion) except three-dimensional motionless key point (following Palestine and China's heart key point) it
Outer at least 32 points.
It should be strongly noted that the quantity for the point being inserted into the application for a curve can be according to actual needs
It determines, still, should ensure that for the quantity for the point that a curve is inserted into: the curve counterpart of the be fitted formation of point through being inserted into
The degree of error of the practice sites curve of face, for the error of the practice sites curve of the opposite face of curve formed by curve control point
1/5-the 1/10 of degree.It follows that the application is that facial image sample is formed by expressed by face key point markup information
Shape, can be closer to actual position shape, to be more advantageous to trained neural network.
S330, face key point mark is formed according to the motionless key point of three-dimensional of the point and facial image sample of above-mentioned selection
Infuse information.
In an optional example, the motionless key point of three-dimensional in point and curve being inserted into curve is in facial image
Coordinate information in sample is the coordinate information in corresponding face key point markup information, the point and song being inserted into curve
Serial number of the motionless key point of three-dimensional in curve in line can be converted into the key of the corresponding face in facial image sample
Number in point markup information.
The process of the method for being trained to neural network of the application is as shown in Figure 4.
In Fig. 4, S400, multiple facial image samples are obtained.
The facial image sample of the application is labeled with: the face key point including at least one three-dimensional motionless key point
Markup information.Facial image sample is labeled with: the face key point being located at the position that is not blocked in facial image sample
Face key point markup information at markup information and the position that is blocked in facial image sample.In facial image sample
21-300 face key point markup information can be labeled with, for example, can be labeled in facial image sample 68,106,
240,252 or 270 face key point markup informations.The mistake of face key point markup information is marked for facial image sample
Journey may refer to the above-mentioned description for Fig. 3.
S410, multiple facial image sample training neural networks using acquisition.
In an optional example, facial image sample is inputted neural network by the application, is extracted via neural network
The face key point information of facial image sample, and the face key point information and facial image extracted according to neural network
The face key point markup information marked in sample, using corresponding loss function, to adjust the parameter in neural network.This Shen
Detailed process is as follows states description in Fig. 5 by one for please being trained using facial image sample to neural network.
In an optional example, a specific example of the method for being trained to neural network of the application
Process is as shown in Figure 5.
In Fig. 5, facial image sample S500, is inputted into neural network, extracts facial image sample via neural network
Face key point information.
In an optional example, the facial image sample in the application has multiple face key point markup informations.People
The face key point markup information that face image sample is had includes: at least one three-dimensional motionless key point markup information.One
Optional example, facial image sample have 106 face key point markup informations, this 106 face key point markup informations
In include the motionless key point markup information of 12 three-dimensionals, for example, right eye upper eyelid center key point markup information, right eye eyelid
Center key point markup information, left eye upper eyelid center key point markup information, oculus sinisterlower lid center key point markup information,
Along center key point markup information, the right lip peak key point markup information of lip, upper lip upper lip line center key point mark under nose
Infuse information, the left lip peak key point markup information of lip, upper lip lower lip line center key point markup information, in lower lip upper lip line
Heart key point markup information, lower lip lower lip line center key point markup information and chin center key point markup information.Often
One key point markup information includes: the coordinate of key point number and key point in facial image sample.In addition, no matter people
With the presence or absence of the face key point being located at the position that is blocked in face image sample, (such as 106 faces close all face key points
Key point) all have markup information.
S510, calculated according to the face key point information that extracts between the different key points at least one organ the
One distance.
In an optional example, for the organ (such as left eye or right eye or mouth) in face, mind
Multiple face key points can be extracted through network, multiple face key points can form at least one set of face key point, the application
The distance between two face key points in each group can be calculated, to obtain at least one first distance.
One optional example, in the case where organ is simple eye (left eye or right eye), neural network can extract multiple
The usually upper bottom of upper eyelid line key point and multiple palpebra inferior line key points, upper eyelid line key point and palpebra inferior line key point
It sets relatively, the opposite upper eyelid line key point in upper and lower position and palpebra inferior line key point can form one group of eyelid line key point,
To which the application can form multiple groups eyelid line key point, by that can be counted according to the coordinate of two eyelid line key points in group
The distance between two eyelid line key points in the group are calculated, to obtain the first eyelid linear distance.
One optional example, in the case where organ is mouth, neural network can extract multiple upper lip upper lip lines and close
Key point and multiple lower lip lower lip line key points, upper lip upper lip line key point and lower lip lower lip line key point are usually up and down
Position is opposite, and the opposite upper lip upper lip line key point in upper and lower position and lower lip lower lip line key point can form one group of outer lip
Line key point passes through the seat according to two lip line key points in group so that the application can form multiple groups outer lip line key point
Mark can calculate the distance between two lip line key points in the group, to obtain the first outer lip linear distance.
One optional example, in the case where organ is mouth, neural network can extract multiple upper lip lower lip lines and close
Key point and multiple lower lip upper lip line key points, upper lip lower lip line key point and lower lip upper lip line key point are usually up and down
Position is opposite, and the opposite upper lip lower lip line key point in upper and lower position and lower lip upper lip line key point can form one group of epipharynx
Line key point passes through the seat according to two lip line key points in group so that the application can form multiple groups epipharynx line key point
Mark can calculate the distance between two lip line key points in the group, to obtain the first epipharynx linear distance.
S520, according to face key point markup information calculate between the different key points at least one organ second away from
From.
In an optional example, for an organ in face, multiple face key points are usually labeled with, are marked
The multiple face key points outpoured can form at least one set of face key point, and the application can calculate two people in each group
The distance between face key point, to obtain at least one second distance.
One optional example, in the case where organ is simple eye (left eye or right eye), usual emblem in facial image sample
It is marked with multiple upper eyelid line key points and multiple palpebra inferior line key points, upper eyelid line key point and palpebra inferior line key point are logical
Normal upper and lower position is opposite, and the opposite upper eyelid line key point in upper and lower position and palpebra inferior line key point can form one group of eyelid line
Key point passes through the mark according to two eyelid line key points in group so that the application can form multiple groups eyelid line key point
Note coordinate can calculate the distance between two eyelid line key points in the group, to obtain the second eyelid linear distance.
One optional example would generally be labeled with multiple upper mouths in facial image sample in the case where organ is mouth
Lip upper lip line key point and multiple lower lip lower lip line key points, upper lip upper lip line key point and lower lip lower lip line are crucial
The usual upper and lower position of point is opposite, and the opposite upper lip upper lip line key point in upper and lower position and lower lip lower lip line key point can be with shapes
At one group of outer lip line key point, so that the application can form multiple groups outer lip line key point, by according to two lip lines in group
The mark coordinate of key point can calculate the distance between two lip line key points in the group, to obtain the second outer lip line
Distance.
One optional example would generally be labeled with multiple upper mouths in facial image sample in the case where organ is mouth
Lip lower lip line key point and multiple lower lip upper lip line key points, upper lip lower lip line key point and lower lip upper lip line are crucial
The usual upper and lower position of point is opposite, and the opposite upper lip lower lip line key point in upper and lower position and lower lip upper lip line key point can be with shapes
At one group of epipharynx line key point, so that the application can form multiple groups epipharynx line key point, by according to two lip lines in group
The mark coordinate of key point can calculate the distance between two lip line key points in the group, to obtain the second epipharynx line
Distance.
S530, it is exercised supervision via first-loss function to neural network based on above-mentioned first distance and second distance
It practises.
In an optional example, the application is usually provided with four first-loss functions, this four first-loss functions
Left eye, right eye, outer lip line and epipharynx line are respectively corresponded, the first eyelid linear distance and second gone out based on left eye eyelid line computation
Eyelid linear distance is updated in a first-loss function, the first eyelid linear distance gone out based on right eye eyelid line computation and the
Two eyelid linear distances are updated in another loss function, outside the first outer lip linear distance and second gone out based on outer lip line computation
Lip linear distance is updated in another loss function, the first epipharynx linear distance and the second epipharynx line gone out based on epipharynx line computation
Distance is updated in remaining first-loss function.The application by using four first-loss functions to neural network into
Row supervised learning, the face key point for being conducive to improve the facial image that neural network is directed to big expression or wide-angle deflection mention
The accuracy taken.
In an optional example, the training process of the application can also include step S540.
Step S540, it is formed for the face key point face key point corresponding with mark that neural network extracts
Any group of key point calculates separately in the group for example, two face key points with identical number form one group of key point
The distance between two key points, to obtain at least one drift distance;And based on the drift distance via the second loss function
It exercises supervision study to neural network.The application is by exercising supervision study to neural network using the second loss function, favorably
In the accuracy for improving neural network extraction face key point.
Fig. 6 is the structural schematic diagram of one embodiment of the face image processing device of the application.As shown in fig. 6, the reality
The device for applying example includes: input picture module 600 and the crucial point module 610 of acquisition.Optionally, at the facial image of the embodiment
Reason device can also include: the device 620 for training neural network.
Input picture module 600 is for facial image to be processed to be input in neural network.Input picture module 600 has
The operation that body the executes such as above-mentioned description for the S100 in Fig. 1, is no longer described in detail herein.
It obtains crucial point module 610 and is used to extract the three-dimensional including at least one of facial image to be processed via neural network
Face key point information including motionless key point.Obtain operation that crucial point module 610 specifically executes such as above-mentioned in Fig. 1
S110 description, be no longer described in detail herein.
Device 620 for training neural network is used to utilize the nerve net in multiple facial image sample training the application
Network.For train neural network device 620 specific structure referring to following descriptions for Fig. 7, be no longer described in detail herein.
Fig. 7 is the structural schematic diagram of an embodiment of the device 620 for training neural network of the application.Such as Fig. 7 institute
Show, the device of the embodiment includes: to obtain sample module 621 and training module 622.Optionally, for training neural network
Device 620 can also include: labeling module 623.
Sample module 621 is obtained for obtaining multiple facial image samples.Obtain the facial image that sample module 621 obtains
Sample is labeled with: the face key point markup information including at least one three-dimensional motionless key point.
Training module 622 is used for the multiple facial image sample training neural networks obtained using sample module 621 is obtained.
In an optional example, training module 622 can be specifically used for inputting facial image sample into neural network, warp
The face key point information that facial image sample is extracted by neural network, according to the face key point information extracted calculate to
The first distance between different key points on a few organ, calculates at least one organ according to face key point markup information
On different key points between second distance;And first distance and second distance are based on via first-loss function to nerve net
Network exercises supervision study.
In an optional example, training module 622 can also be specifically used for, for by extracting face key point with
At least one set of key point that the corresponding face key point of mark is formed, determines the distance between two key points in group, with
Obtain at least one drift distance;It is exercised supervision study via the second loss function to neural network based on drift distance.
The operation that training module 622 specifically the executes such as above-mentioned description for each step in Fig. 5, herein no longer specifically
It is bright.
Labeling module 623 is used to determine the curve control point at least one position of face in facial image sample, according to
The curve control point forming curves determine the motionless key point of three-dimensional on the curve as facial image sample, in facial image
At least side (such as the left and right sides) of the motionless key point of three-dimensional of sample, chooses the point on curve, and according to the point of selection and
The motionless key point of three-dimensional of facial image sample forms face key point markup information.The operation that labeling module 623 specifically executes
Such as the above-mentioned description for each step in Fig. 3, no longer it is described in detail herein.
Example devices
Fig. 8 shows the example devices 800 for being adapted for carrying out the application, and equipment 800 can be the control configured in automobile
System/electronic system, mobile terminal (for example, intelligent mobile phone etc.), personal computer (PC, for example, desktop computer or
Notebook computer etc.), tablet computer and server etc..In Fig. 8, equipment 800 includes one or more processor, communication
Portion etc., one or more of processors can be with are as follows: one or more central processing unit (CPU) 801, and/or, one
Or multiple images processor (GPU) 813 etc., processor can be executable in read-only memory (ROM) 802 according to being stored in
Instruction or executed from the executable instruction that storage section 808 is loaded into random access storage device (RAM) 803 it is various appropriate
Movement and processing.Communication unit 812 can include but is not limited to network interface card, and the network interface card can include but is not limited to IB
(Infiniband) network interface card.Processor can be executed with communicating in read-only memory 802 and/or random access storage device 830 can
It executes instruction, be connected by bus 804 with communication unit 812 and communicated through communication unit 812 with other target devices, to complete
Corresponding steps in the application.
In an optional example, processor upon executing the commands, realizes any face image processing in the application
Step in method, or realize the step in any method for training neural network in the application.Processor institute is real
Existing step may refer to the associated description in above-mentioned each method embodiment, no longer be described in detail herein.
In addition, in RAM 803, various programs and data needed for device operation can also be stored with.CPU801,
ROM802 and RAM803 is connected with each other by bus 804.In the case where there is RAM803, ROM802 is optional module.
RAM803 stores executable instruction, or executable instruction is written into ROM802 at runtime, and executable instruction makes central processing
Unit 801 executes step included by above-mentioned method for segmenting objects.Input/output (I/O) interface 805 is also connected to bus 804.
Communication unit 812 can integrate setting, may be set to be with multiple submodule (for example, multiple IB network interface cards), and respectively and always
Line connection.
I/O interface 805 is connected to lower component: the importation 806 including keyboard, mouse etc.;It is penetrated including such as cathode
The output par, c 807 of spool (CRT), liquid crystal display (LCD) etc. and loudspeaker etc.;Storage section 808 including hard disk etc.;
And the communications portion 809 of the network interface card including LAN card, modem etc..Communications portion 809 via such as because
The network of spy's net executes communication process.Driver 810 is also connected to I/O interface 805 as needed.Detachable media 811, such as
Disk, CD, magneto-optic disk, semiconductor memory etc. are mounted on as needed on driver 810, in order to read from thereon
Computer program be installed in storage section 808 as needed.
It should be strongly noted that framework as shown in Figure 8 is only a kind of optional implementation, in concrete practice process
In, can the component count amount and type according to actual needs to above-mentioned Fig. 8 selected, deleted, increased or replaced;In different function
In component setting, can also be used it is separately positioned or integrally disposed and other implementations, for example, the separable setting of GPU and CPU, for another example
GPU, can be integrated on CPU by reason, the separable setting of communication unit, can also be integrally disposed on CPU or GPU etc..These are replaceable
Embodiment each fall within the protection scope of the application.
Particularly, it according to presently filed embodiment, may be implemented as calculating below with reference to the process of flow chart description
Machine software program is stored thereon with computer program for example, the application embodiment includes a kind of computer program product, should
When computer program is executed by processor, the step in any face image processing process in the application is realized, or realize
The step in any method for training neural network in the application.
In such an embodiment, which can be downloaded and be pacified from network by communications portion 809
Dress, and/or be mounted from detachable media 811.When the computer program is executed by central processing unit (CPU) 801, execute
Above-mentioned steps as described in this application.
The present processes and device, electronic equipment and computer-readable storage medium may be achieved in many ways
Matter.For example, can be realized by any combination of software, hardware, firmware or software, hardware, firmware the present processes and
Device, electronic equipment and computer readable storage medium.The said sequence of the step of for method merely to be illustrated,
The step of the present processes, is not limited to sequence described in detail above, unless specifically stated otherwise.In addition, some
In embodiment, the application can be also embodied as recording program in the recording medium, these programs include for realizing basis
The machine readable instructions of the present processes.Thus, the application also covers storage for executing the facial image according to the application
Processing method and/or recording medium (such as computer-readable storage medium for training the computer program of the method for neural network
Matter).
The description of the present application is given for the purpose of illustration and description, and is not exhaustively or by the application
It is limited to disclosed form.Many modifications and variations are obvious for the ordinary skill in the art.It selects and retouches
Embodiment is stated and be the principle and practical application in order to more preferably illustrate the application, and enables those skilled in the art
Understand that the application is suitable for the various embodiments with various modifications of special-purpose to design.
Claims (10)
1. a kind of face image processing process characterized by comprising
Facial image to be processed is input in neural network;
Via the neural network extract the facial image to be processed including at least one three-dimensional motionless key point
Face key point information.
2. face image processing process according to claim 1, which is characterized in that the face that the neural network extracts
Key point information includes: that the face being located at the position that is not blocked at position and/or is blocked in facial image to be processed closes
Key point information.
3. face image processing process according to any one of claim 1 to 2, which is characterized in that the three-dimensional is motionless
Key point include: under eyelid center key point, nose along center key point, lip lip peak key point, lip center key point with
And at least one of chin center key point.
4. face image processing process according to claim 3, which is characterized in that eyelid center key point includes:
Left eye upper eyelid center key point, oculus sinisterlower lid center key point, right eye upper eyelid center key point and right eye eyelid
At least one of center key point.
5. a kind of method for training neural network, which is characterized in that the described method includes:
Obtain multiple facial image samples, wherein the facial image sample is labeled with: including at least one three-dimensional motionless key
Face key point markup information including point;
Utilize the multiple facial image sample training neural network.
6. a kind of face image processing device characterized by comprising
Input picture module, for facial image to be processed to be input in neural network;
Obtain crucial point module, for extract the facial image to be processed via the neural network include at least one three
Tie up the face key point information including motionless key point.
7. a kind of for training the device of neural network, which is characterized in that described device includes:
Sample module is obtained, for obtaining multiple facial image samples, wherein the facial image sample is labeled with: including extremely
Face key point markup information including a few motionless key point of three-dimensional;
Training module, for utilizing neural network described in the multiple facial image sample training.
8. a kind of electronic equipment, comprising:
Memory, for storing computer program;
Processor, for executing the computer program stored in the memory, and the computer program is performed, and is realized
Step in face image processing process described in any one of the claims 1-4, or realize 5 institute of the claims
The step in the method for training neural network stated.
9. a kind of computer readable storage medium, is stored thereon with computer program, when which is executed by processor,
It realizes the step in face image processing process described in any one of the claims 1-4, or realizes that aforesaid right is wanted
The step in method described in asking 5 for training neural network.
10. a kind of computer program, including computer instruction, when the computer instruction is run in the processor of equipment,
It realizes the step in face image processing process described in any one of the claims 1-4, or realizes that aforesaid right is wanted
The step in method described in asking 5 for training neural network.
Priority Applications (2)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN202110982660.4A CN113688737A (en) | 2017-12-15 | 2017-12-15 | Face image processing method, face image processing device, electronic apparatus, storage medium, and program |
CN201711352295.9A CN109934058B (en) | 2017-12-15 | 2017-12-15 | Face image processing method, face image processing device, electronic apparatus, storage medium, and program |
Applications Claiming Priority (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201711352295.9A CN109934058B (en) | 2017-12-15 | 2017-12-15 | Face image processing method, face image processing device, electronic apparatus, storage medium, and program |
Related Child Applications (1)
Application Number | Title | Priority Date | Filing Date |
---|---|---|---|
CN202110982660.4A Division CN113688737A (en) | 2017-12-15 | 2017-12-15 | Face image processing method, face image processing device, electronic apparatus, storage medium, and program |
Publications (2)
Publication Number | Publication Date |
---|---|
CN109934058A true CN109934058A (en) | 2019-06-25 |
CN109934058B CN109934058B (en) | 2021-08-06 |
Family
ID=66980352
Family Applications (2)
Application Number | Title | Priority Date | Filing Date |
---|---|---|---|
CN202110982660.4A Pending CN113688737A (en) | 2017-12-15 | 2017-12-15 | Face image processing method, face image processing device, electronic apparatus, storage medium, and program |
CN201711352295.9A Active CN109934058B (en) | 2017-12-15 | 2017-12-15 | Face image processing method, face image processing device, electronic apparatus, storage medium, and program |
Family Applications Before (1)
Application Number | Title | Priority Date | Filing Date |
---|---|---|---|
CN202110982660.4A Pending CN113688737A (en) | 2017-12-15 | 2017-12-15 | Face image processing method, face image processing device, electronic apparatus, storage medium, and program |
Country Status (1)
Country | Link |
---|---|
CN (2) | CN113688737A (en) |
Cited By (2)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN112069993A (en) * | 2020-09-04 | 2020-12-11 | 西安西图之光智能科技有限公司 | Dense face detection method and system based on facial features mask constraint and storage medium |
CN113689527A (en) * | 2020-05-15 | 2021-11-23 | 武汉Tcl集团工业研究院有限公司 | Training method of face conversion model and face image conversion method |
Citations (12)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN104778712A (en) * | 2015-04-27 | 2015-07-15 | 厦门美图之家科技有限公司 | Method and system for pasting image to human face based on affine transformation |
CN105205827A (en) * | 2015-10-16 | 2015-12-30 | 中科院成都信息技术股份有限公司 | Auxiliary feature point labeling method for statistical shape model |
CN105760836A (en) * | 2016-02-17 | 2016-07-13 | 厦门美图之家科技有限公司 | Multi-angle face alignment method based on deep learning and system thereof and photographing terminal |
CN105868769A (en) * | 2015-01-23 | 2016-08-17 | 阿里巴巴集团控股有限公司 | Method and device for positioning face key points in image |
CN106203376A (en) * | 2016-07-19 | 2016-12-07 | 北京旷视科技有限公司 | Face key point localization method and device |
CN106295533A (en) * | 2016-08-01 | 2017-01-04 | 厦门美图之家科技有限公司 | Optimization method, device and the camera terminal of a kind of image of autodyning |
CN106548179A (en) * | 2016-09-29 | 2017-03-29 | 北京市商汤科技开发有限公司 | The detection method of object and dress ornament key point, device and electronic equipment |
CN106951840A (en) * | 2017-03-09 | 2017-07-14 | 北京工业大学 | A kind of facial feature points detection method |
CN107016370A (en) * | 2017-04-10 | 2017-08-04 | 电子科技大学 | One kind is based on the enhanced partial occlusion face identification method of data |
CN107122705A (en) * | 2017-03-17 | 2017-09-01 | 中国科学院自动化研究所 | Face critical point detection method based on three-dimensional face model |
CN107239758A (en) * | 2017-05-24 | 2017-10-10 | 北京小米移动软件有限公司 | The method and device of face key point location |
CN107451965A (en) * | 2017-07-24 | 2017-12-08 | 深圳市智美达科技股份有限公司 | Distort face image correcting method, device, computer equipment and storage medium |
Family Cites Families (2)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN106203242B (en) * | 2015-05-07 | 2019-12-24 | 阿里巴巴集团控股有限公司 | Similar image identification method and equipment |
CN104899563B (en) * | 2015-05-29 | 2020-01-07 | 深圳大学 | Two-dimensional face key feature point positioning method and system |
-
2017
- 2017-12-15 CN CN202110982660.4A patent/CN113688737A/en active Pending
- 2017-12-15 CN CN201711352295.9A patent/CN109934058B/en active Active
Patent Citations (12)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN105868769A (en) * | 2015-01-23 | 2016-08-17 | 阿里巴巴集团控股有限公司 | Method and device for positioning face key points in image |
CN104778712A (en) * | 2015-04-27 | 2015-07-15 | 厦门美图之家科技有限公司 | Method and system for pasting image to human face based on affine transformation |
CN105205827A (en) * | 2015-10-16 | 2015-12-30 | 中科院成都信息技术股份有限公司 | Auxiliary feature point labeling method for statistical shape model |
CN105760836A (en) * | 2016-02-17 | 2016-07-13 | 厦门美图之家科技有限公司 | Multi-angle face alignment method based on deep learning and system thereof and photographing terminal |
CN106203376A (en) * | 2016-07-19 | 2016-12-07 | 北京旷视科技有限公司 | Face key point localization method and device |
CN106295533A (en) * | 2016-08-01 | 2017-01-04 | 厦门美图之家科技有限公司 | Optimization method, device and the camera terminal of a kind of image of autodyning |
CN106548179A (en) * | 2016-09-29 | 2017-03-29 | 北京市商汤科技开发有限公司 | The detection method of object and dress ornament key point, device and electronic equipment |
CN106951840A (en) * | 2017-03-09 | 2017-07-14 | 北京工业大学 | A kind of facial feature points detection method |
CN107122705A (en) * | 2017-03-17 | 2017-09-01 | 中国科学院自动化研究所 | Face critical point detection method based on three-dimensional face model |
CN107016370A (en) * | 2017-04-10 | 2017-08-04 | 电子科技大学 | One kind is based on the enhanced partial occlusion face identification method of data |
CN107239758A (en) * | 2017-05-24 | 2017-10-10 | 北京小米移动软件有限公司 | The method and device of face key point location |
CN107451965A (en) * | 2017-07-24 | 2017-12-08 | 深圳市智美达科技股份有限公司 | Distort face image correcting method, device, computer equipment and storage medium |
Cited By (3)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN113689527A (en) * | 2020-05-15 | 2021-11-23 | 武汉Tcl集团工业研究院有限公司 | Training method of face conversion model and face image conversion method |
CN112069993A (en) * | 2020-09-04 | 2020-12-11 | 西安西图之光智能科技有限公司 | Dense face detection method and system based on facial features mask constraint and storage medium |
CN112069993B (en) * | 2020-09-04 | 2024-02-13 | 西安西图之光智能科技有限公司 | Dense face detection method and system based on five-sense organ mask constraint and storage medium |
Also Published As
Publication number | Publication date |
---|---|
CN113688737A (en) | 2021-11-23 |
CN109934058B (en) | 2021-08-06 |
Similar Documents
Publication | Publication Date | Title |
---|---|---|
US11341769B2 (en) | Face pose analysis method, electronic device, and storage medium | |
KR102347336B1 (en) | Gaze point determination method and apparatus, electronic device and computer storage medium | |
CN110610453B (en) | Image processing method and device and computer readable storage medium | |
CN108229293A (en) | Face image processing process, device and electronic equipment | |
CN101339606B (en) | Human face critical organ contour characteristic points positioning and tracking method and device | |
CN108229470A (en) | character image processing method, device, equipment and storage medium | |
CN109325972A (en) | Processing method, device, equipment and the medium of laser radar sparse depth figure | |
CN109960974A (en) | Face critical point detection method, apparatus, electronic equipment and storage medium | |
CN108734078B (en) | Image processing method, image processing apparatus, electronic device, storage medium, and program | |
CN106097833B (en) | A kind of Educational toy external member and its digit recognition method | |
CN109598198A (en) | The method, apparatus of gesture moving direction, medium, program and equipment for identification | |
JP2009545813A (en) | Establish and use dominant lines of images | |
US10853651B2 (en) | Virtual reality interaction method, apparatus and system | |
CN108229301A (en) | Eyelid line detecting method, device and electronic equipment | |
US10401947B2 (en) | Method for simulating and controlling virtual sphere in a mobile device | |
US20210407125A1 (en) | Object recognition neural network for amodal center prediction | |
CN109934058A (en) | Face image processing process, device, electronic equipment, storage medium and program | |
US20170358273A1 (en) | Systems and methods for resolution adjustment of streamed video imaging | |
CN108288299A (en) | A kind of method and system of expression data driving three-dimensional image expression | |
CN113989831A (en) | Myopia prevention and control method, device, terminal equipment and storage medium | |
WO2022026603A1 (en) | Object recognition neural network training using multiple data sources | |
Thiengtham et al. | Improve template matching method in mobile augmented reality for thai alphabet learning | |
CN110308785A (en) | E-book implementation method based on AR technology | |
EP4318314A1 (en) | Image acquisition model training method and apparatus, image detection method and apparatus, and device | |
CN106406507A (en) | Image processing method and electronic equipment |
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 | ||
GR01 | Patent grant | ||
GR01 | Patent grant |