CN108268822A - Face identification method, device and robot - Google Patents
Face identification method, device and robot Download PDFInfo
- Publication number
- CN108268822A CN108268822A CN201611263999.4A CN201611263999A CN108268822A CN 108268822 A CN108268822 A CN 108268822A CN 201611263999 A CN201611263999 A CN 201611263999A CN 108268822 A CN108268822 A CN 108268822A
- Authority
- CN
- China
- Prior art keywords
- face
- image
- recognition
- pending
- full articulamentum
- 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
- G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
- G06V40/00—Recognition of biometric, human-related or animal-related patterns in image or video data
- G06V40/10—Human or animal bodies, e.g. vehicle occupants or pedestrians; Body parts, e.g. hands
- G06V40/16—Human faces, e.g. facial parts, sketches or expressions
- G06V40/168—Feature extraction; Face representation
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/04—Architecture, e.g. interconnection topology
- G06N3/045—Combinations of networks
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
- G06V40/00—Recognition of biometric, human-related or animal-related patterns in image or video data
- G06V40/10—Human or animal bodies, e.g. vehicle occupants or pedestrians; Body parts, e.g. hands
- G06V40/16—Human faces, e.g. facial parts, sketches or expressions
- G06V40/172—Classification, e.g. identification
Abstract
The invention discloses a kind of face identification method, device and robot, this method includes：Extract the characteristics of image in pending recognition of face image；Based on the characteristics of image in pending recognition of face image, whether pending recognition of face image is detected comprising human face region, obtains first as a result, wherein, whether the first result is used to indicate in pending recognition of face image comprising human face region；Based on the characteristics of image in pending recognition of face image, while whether pending recognition of face image is detected comprising human face region, the face region of pending recognition of face image is demarcated, obtains the second result, wherein, the second result is used to demarcate the face region in pending recognition of face image；First result and the second result are subjected to integration processing, obtain the recognition result of pending recognition of face image.The present invention is solved carries out the technical issues of recognition of face program is slow using convolutional neural networks.
Description
Technical field
The present invention relates to field of face identification, in particular to a kind of face identification method, device and robot.
Background technology
In the prior art, using the convolutional neural networks of depth, the work of Face datection and face calibration is carried out generally all
It is carried out separately.During the work time, the convolutional neural networks of two different depth of generally use, the volume of first depth
Product neural network detection human face region, then carries out cutting processing by the human face region detected, recycles another depth
Convolutional neural networks carry out face calibration.During Face datection is carried out, convolutional neural networks can be from image learning
To the feature of human face region, a separator is then recycled, distinguishes human face region and non-face region, so as to reach face inspection
The purpose of survey；Equally, face calibration is also to cut out in the human face region come, utilizes convolutional neural networks, study face area
The feature of face and position in domain achieve the purpose that calibration face, and the image of human face region needs to carry out twice in the process
Convolutional Neural extracts feature, and provincial characteristics is repeated study and calculating, the delay time for the calculating deepened significantly.
The problem of recognition of face program is slow is carried out using convolutional neural networks for above-mentioned, not yet proposes effective solution at present
Certainly scheme.
Invention content
An embodiment of the present invention provides a kind of face identification method, device and robot, at least to solve to utilize convolution god
The technical issues of recognition of face program is slow is carried out through network.
One side according to embodiments of the present invention provides a kind of face identification method, including：Extract pending face
Identify the characteristics of image in image；Based on the characteristics of image in the pending recognition of face image, the pending people is detected
Whether face identification image obtains first as a result, wherein, first result be used to indicate the pending people comprising human face region
Whether human face region is included in face identification image；Based on the characteristics of image in the pending recognition of face image, in detection institute
While pending recognition of face image is stated whether comprising human face region, the face area of the pending recognition of face image is demarcated
Domain obtains second as a result, wherein, second result is used to demarcate the face region in the pending recognition of face image；
First result and second result are subjected to integration processing, obtain the identification knot of the pending recognition of face image
Fruit.
Further, before the characteristics of image in extracting pending recognition of face image, the method further includes：It establishes
Convolutional neural networks, wherein, the convolutional neural networks include the first full articulamentum and the second full articulamentum；Treated described in detection into
Whether row recognition of face image includes comprising human face region：The pending recognition of face is detected using the described first full articulamentum
Whether image obtains first result comprising human face region；Demarcate the face region packet of the pending recognition of face image
It includes：The face region of the pending recognition of face image is demarcated using the described second full articulamentum, obtains second result.
Further, record has face characteristic in the described first full articulamentum, and institute is detected using the described first full articulamentum
State whether pending recognition of face image includes comprising human face region：Detect the characteristics of image of the pending recognition of face image
In whether include the face characteristic；If it is detected that the face is included in the characteristics of image of the pending recognition of face image
Feature, it is determined that the pending recognition of face image includes the human face region；If it is detected that the pending recognition of face
The face characteristic is not included in the characteristics of image of image, it is determined that the pending recognition of face image does not include the face
Region.
Further, record has grader in the described second full articulamentum, and being recorded in the grader has face feature, profit
The face region that the pending recognition of face image is demarcated with the described second full articulamentum includes：It, will by the grader
Characteristics of image in the pending recognition of face image is included into classification corresponding with the face feature respectively, wherein, each
Classification corresponds to an organic region in the face region.
Further, convolutional neural networks are established to include：First full articulamentum and the second full articulamentum are set；Obtain face
Feature and face feature；The described first full articulamentum is trained using the face characteristic, obtains first-loss value, and make
The described second full articulamentum is trained with the face feature, obtains the second penalty values；By the first-loss value and institute
The superposition of the second penalty values is stated, obtains the penalty values of the convolutional neural networks；Using the penalty values to the convolutional Neural net
Parameter in network is trained, the convolutional neural networks after being trained, wherein, the parameter includes the described first full articulamentum
With each coefficient in the cost function of the described second full articulamentum.
Other side according to embodiments of the present invention provides a kind of face identification device, including：Extraction unit is used
Characteristics of image in the pending recognition of face image of extraction；Detection unit, for being based on the pending recognition of face image
In characteristics of image, whether detect the pending recognition of face image comprising human face region, obtain first as a result, wherein, institute
The first result is stated whether to be used to indicate in the pending recognition of face image comprising human face region；Unit is demarcated, for being based on
Characteristics of image in the pending recognition of face image is detecting whether the pending recognition of face image includes face area
While domain, the face region of the pending recognition of face image is demarcated, obtains second as a result, wherein, second result
For demarcating the face region in the pending recognition of face image；Integral unit, by first result and described second
As a result integration processing is carried out, obtains the recognition result of the pending recognition of face image.
Further, described device further includes：Creation module, for establishing convolutional neural networks, wherein, the convolution god
Include the first full articulamentum and the second full articulamentum through network；The detection unit includes：Detection module, for utilizing described the
Whether one full articulamentum detects the pending recognition of face image comprising human face region, obtains first result；The mark
Order member includes：Demarcating module, for demarcating the face of the pending recognition of face image using the described second full articulamentum
Region obtains second result.
Further, record has face characteristic, the detection module in the first full articulamentum that the creation module is established
Including：Detection sub-module, it is whether special comprising the face in the characteristics of image of the pending recognition of face image for detecting
Sign；First determining module, it is special for including the face in the characteristics of image for detecting the pending recognition of face image
In the case of sign, determine that the pending recognition of face image includes the human face region；Second determining module, for detecting
Go out in the characteristics of image of the pending recognition of face image in the case of not including the face characteristic, determine described pending
Recognition of face image does not include the human face region.
Further, record has grader in the second full articulamentum that the creation module is established, and remembers in the grader
Record has face feature, and the demarcating module includes：Module is included into, for passing through the grader, the pending face is known
Characteristics of image in other image is included into classification corresponding with the face feature respectively, wherein, each classification corresponds to the face
An organic region in region.
Further, the creation module includes：Setup module, for setting the first full articulamentum and the second full connection
Layer；Acquisition module, for obtaining face characteristic and face feature；First training module, for using the face characteristic to institute
State the first full articulamentum to be trained, obtain first-loss value, and using the face feature to the described second full articulamentum into
Row training, obtains the second penalty values；Laminating module for the first-loss value and second penalty values to be superimposed, obtains
The penalty values of the convolutional neural networks；Second training module, for utilizing the penalty values in the convolutional neural networks
Parameter be trained, the convolutional neural networks after being trained, wherein, the parameter include the described first full articulamentum and institute
State each coefficient in the cost function of the second full articulamentum.
Another aspect according to embodiments of the present invention provides a kind of recognition of face machine people, including：Image collector
It puts, for acquiring pending recognition of face image；Face identification device.
In embodiments of the present invention, a feature extraction is carried out to pending recognition of face image, and is treated according to extraction
The characteristics of image for carrying out recognition of face image carries out the face calibration of recognition of face and face respectively, wherein, by detection treat into
Whether row recognition of face image obtains first result comprising human face region；By demarcating people in pending recognition of face image
The face region of face obtains second as a result, integrated to the first result and the second result that obtain, obtains pending face and knows
The recognition result of other image.Recognition of face and people are carried out by carrying out an image characteristics extraction to pending recognition of face image
The face calibration of face, without carrying out multiple image characteristics extraction to pending recognition of face image, can accelerate to pending
Recognition of face image carries out the speed of recognition of face and the face calibration of face, solves and carries out face using convolutional neural networks
The technical issues of recognizer is slow.
Description of the drawings
Attached drawing described herein is used to provide further understanding of the present invention, and forms the part of the application, this hair
Bright illustrative embodiments and their description do not constitute improper limitations of the present invention for explaining the present invention.In the accompanying drawings：
Fig. 1 is a kind of flow chart of face identification method according to embodiments of the present invention；
Fig. 2 is a kind of schematic diagram of face identification device according to embodiments of the present invention；
Fig. 3 is the schematic diagram of recognition of face machine people according to embodiments of the present invention a kind of.
Specific embodiment
In order to which those skilled in the art is made to more fully understand the present invention program, below in conjunction in the embodiment of the present invention
The technical solution in the embodiment of the present invention is clearly and completely described in attached drawing, it is clear that described embodiment is only
The embodiment of a part of the invention, instead of all the embodiments.Based on the embodiments of the present invention, ordinary skill people
Member's all other embodiments obtained without making creative work should all belong to the model that the present invention protects
It encloses.
It should be noted that term " first " in description and claims of this specification and above-mentioned attached drawing, "
Two " etc. be the object for distinguishing similar, and specific sequence or precedence are described without being used for.It should be appreciated that it uses in this way
Data can be interchanged in the appropriate case, so as to the embodiment of the present invention described herein can in addition to illustrating herein or
Sequence other than those of description is implemented.In addition, term " comprising " and " having " and their any deformation, it is intended that cover
Cover it is non-exclusive include, be not necessarily limited to for example, containing the process of series of steps or unit, method, system, product or equipment
Those steps or unit clearly listed, but may include not listing clearly or for these processes, method, product
Or the intrinsic other steps of equipment or unit.
According to embodiments of the present invention, a kind of face identification method embodiment is provided, it should be noted that in the stream of attached drawing
The step of journey illustrates can perform in the computer system of such as a group of computer-executable instructions, although also, flowing
Logical order is shown in journey figure, but in some cases, can be performed with the sequence being different from herein shown or described by
The step of.
Fig. 1 is a kind of flow chart of face identification method according to embodiments of the present invention, as shown in Figure 1, this method includes
Following steps：
Step S102 extracts the characteristics of image in pending recognition of face image；
Step S104, based on the characteristics of image in pending recognition of face image, detecting pending recognition of face image is
It is no to include human face region, first is obtained as a result, wherein, the first result is used to indicate in pending recognition of face image and whether includes
Human face region；
Step S106 based on the characteristics of image in pending recognition of face image, is detecting pending recognition of face image
While whether including human face region, the face region of pending recognition of face image is demarcated, obtains second as a result, wherein, the
Two results are used to demarcate the face region in pending recognition of face image；
First result and the second result are carried out integration processing, obtain the knowledge of pending recognition of face image by step S108
Other result.
In the above-described embodiments, to the pending recognition of face image for carrying out recognition of face is needed to carry out a characteristics of image
It extracts, and carries out the face mark of recognition of face and face respectively according to the characteristics of image of the pending recognition of face image of extraction
It is fixed, wherein, by whether detecting pending recognition of face image comprising human face region, obtain the first result；It is pending detecting
While whether recognition of face image includes human face region, by the face area that face is demarcated in pending recognition of face image
Domain obtains second as a result, integrated to the first result and the second result that obtain, obtains the knowledge of pending recognition of face image
Other result.The face mark for carrying out recognition of face and face by carrying out an image characteristics extraction to pending recognition of face image
It is fixed, without carrying out multiple image characteristics extraction to pending recognition of face image, can accelerate to pending recognition of face figure
As the speed of face calibration for carrying out recognition of face and face, solve slow using convolutional neural networks progress recognition of face program
The technical issues of.
Optionally, pending recognition of face image can be the image acquired in advance, extract and scheme in the image acquired in advance
As feature, and according to the pending recognition of face image of Image Feature Detection of extraction whether comprising human face region and calibration treat into
The face region of row recognition of face image.
Optionally, pending recognition of face image can be in the case where needing to carry out recognition of face and face calibration in real time
Image in the current detection environment of acquisition, the image for obtaining environment to be detected in real time by picture pick-up device are used as pending face
Identify image, wherein, picture pick-up device obtain pending recognition of face image can be intercepted according to Fixed Time Interval it is more
The sequencing of multiple times according to interception of interception is arranged in order by a image, and successively will interception according to putting in order
Image as pending recognition of face image, detect whether pending recognition of face image is treated comprising human face region and calibration
Carry out the face region of recognition of face image, if detection and calibration failure, according to put in order to the image of next cis-position into
Row identification and calibration；If detecting and demarcating successfully, stopping is identified and demarcates to the subsequent image that this puts in order,
In, detection and calibration unsuccessfully refer to detect pending recognition of face image not comprising human face region or can not demarcate pending people
The face region of face identification image or the pending recognition of face image of detection do not include human face region, and can not demarcate yet and treat
Carry out the face region of recognition of face image；It detects and demarcates and successfully refer to that detecting pending recognition of face image includes face area
Domain, and successfully demarcate the face region of pending recognition of face image.
An optional embodiment, before the characteristics of image in extracting pending recognition of face image, method further includes：
Convolutional neural networks are established, wherein, convolutional neural networks include the first full articulamentum and the second full articulamentum；Detect pending people
Whether face identification image includes comprising human face region：Detect whether pending recognition of face image includes using the first full articulamentum
Human face region obtains the first result；The face region for demarcating pending recognition of face image includes：Utilize the second full articulamentum mark
The face region of fixed pending recognition of face image, obtains the second result.
Specifically, it before the characteristics of image in extracting pending recognition of face image, needs first to establish convolutional Neural net
Network, then by the convolutional neural networks, the characteristics of image in the pending recognition of face image of extraction carries out recognition of face
It is demarcated with face, wherein, the neural network of the convolution of foundation includes the first full articulamentum and the second full articulamentum, complete using first
Articulamentum detects whether pending recognition of face image includes human face region；It is and pending being detected by the first full articulamentum
While whether recognition of face image is human face region, the face of pending recognition of face image is demarcated using the second full articulamentum
Face region, in addition, the operation due to convolutional neural networks can be run in computer program, it is complete by convolutional neural networks
Into the recognition of face of pending recognition of face image and face calibration can be run in computer environment, and for people
First full articulamentum of face identification and the second full articulamentum for face to be demarcated are carried out at the same time, and can accelerate convolutional neural networks
The speed demarcated to recognition of face and face.
As an optional embodiment, record has face characteristic in the first full articulamentum, is examined using the first full articulamentum
Survey whether pending recognition of face image includes comprising human face region：Detecting in the characteristics of image of pending recognition of face image is
It is no to include face characteristic；If it is detected that in the characteristics of image of pending recognition of face image include face characteristic, it is determined that treat into
Row recognition of face image includes human face region；If it is detected that it is special not include face in the characteristics of image of pending recognition of face image
Sign, it is determined that pending recognition of face image does not include human face region.
Specifically, the information for having face characteristic is recorded in the first full articulamentum of convolutional neural networks, it is complete using first
Whether articulamentum detects pending recognition of face image can be by detecting pending recognition of face image comprising human face region
Whether realized in characteristics of image comprising face characteristic, if it is detected that including people in the characteristics of image of pending recognition of face image
Face feature, it is determined that pending recognition of face image includes human face region；If it is detected that the image of pending recognition of face image
Face characteristic is not included in feature, it is determined that pending recognition of face image does not include human face region, by will entirely connect first
It connects the face characteristic information that is recorded in layer and whether detects the characteristics of image of pending recognition of face image comprising face characteristic, it can be with
Whether human face region is included in the pending recognition of face image of accurate judgement.
It is alternatively possible to by the image in the face characteristic recorded in the first full articulamentum and pending recognition of face image
Feature compares, if the characteristics of image in pending recognition of face image and the face characteristic phase recorded in the first full articulamentum
Together, it is determined that pending recognition of face image includes human face region, if characteristics of image in pending recognition of face image and the
The face characteristic recorded in one full articulamentum differs completely, it is determined that pending recognition of face image does not include human face region.
An optional embodiment, record has a grader in the second full articulamentum, and record has face feature in grader, profit
The face region that pending recognition of face image is demarcated with the second full articulamentum includes：By grader, pending face is known
Characteristics of image in other image is included into classification corresponding with face feature respectively, wherein, each classification is corresponded in face region
One organic region.
Specifically, record has grader, and recorded in grader in the second full articulamentum of convolutional neural networks
There is face feature, wherein, the face recorded in separator are characterized in the corresponding organ characteristic of each organic region in face region.
For example, face can be divided into five organic regions, i.e. brow region, eye according to eyebrow, eyes, ear, nose and face
Feature in brow region is recorded as eyebrow organ by eyeball region, ear region, nasal area and face region, grader
Feature；Feature in eye areas is recorded as eyes organ characteristic；Feature in nasal area is recorded as nose organ spy
Sign；Feature in face region is recorded as face organ characteristic.Grader is corresponded to according to organic region each in face region
Organ characteristic, the characteristics of image in pending recognition of face image is classified, it is special that characteristics of image is included into each organ
Corresponding classification is levied, wherein, each classification corresponds to an organic region in face region, and convolutional Neural can be according to each class
Characteristics of image in not demarcates pending recognition of face image, so as to complete the face to pending recognition of face image
The calibration in region.
Optionally, it can also include being not belonging to the classification of face feature in grader, if in pending recognition of face image
Characteristics of image be not belonging to face feature, then this feature is divided into the classification for being not belonging to face feature, and not in the category
Characteristics of image demarcated.
In an optional embodiment, establish convolutional neural networks and include：The first full articulamentum and second is set to connect entirely
Connect layer；Obtain face characteristic and face feature；Using face characteristic, to first, full articulamentum is trained, and obtains first-loss
Value, and using face feature, to second, full articulamentum is trained, and obtains the second penalty values；First-loss value and second are lost
Value superposition, obtains the penalty values of convolutional neural networks；The parameter in convolutional neural networks is trained using penalty values, is obtained
Convolutional neural networks after training, wherein, parameter includes each in the cost function of the first full articulamentum and the second full articulamentum
A coefficient.
Specifically, the process of convolutional neural networks is established, the connection entirely of setting first in convolutional neural networks can be included in
Layer and the second full articulamentum, and face characteristic and face feature are obtained, by face characteristic, to first, full articulamentum is trained,
Using trained result as first-loss value；It is complete to second by face feature while the first full articulamentum is trained
Articulamentum is trained, using trained result as the second penalty values, the first-loss value and the second penalty values that training is obtained
Superposition, the result after superposition is the penalty values of convolutional neural networks, using the penalty values to the parameter in convolutional neural networks
It is trained, the convolutional neural networks after training is obtained, wherein the parameter being trained can be the first full articulamentum and second
Coefficient in the cost function of full articulamentum.
An optional embodiment, the recognition result of pending recognition of face image include：The instruction of first result is pending
In recognition of face image the face region in pending recognition of face image is demarcated comprising human face region, the second result；Or first
As a result it indicates not including human face region in pending recognition of face image, the second result is not demarcated in pending recognition of face image
Face region.
It should be noted that cost function is also known as performance indicator, cost function, it is that regulation optimizes target to be achieved
Mathematic(al) representation, usually state variable, the scalar function for controlling variable, performance variable.
Optionally, training after convolutional neural networks be used directly for recognition of face and face calibration, by
Be implanted into computer the convolutional neural networks can be completed the recognition of face to pending recognition of face image and face calibration
Work.
It, can be by convolutional neural networks to the image in pending recognition of face image as an optional embodiment
Feature extraction carries out recognition of face and calibration face further according to the characteristics of image of extraction.Concrete scheme is as follows：
A convolutional neural networks are built, wherein, multiple convolutional layers are included in the convolutional neural networks
(convolutional layer), multiple pond layers (pooling layer) and full articulamentum, it is defeated in first convolutional layer
Enter the entire image to be identified of extraction and extract the characteristics of image in image.
A pond layer is connected after each convolutional layer and one is corrected linear unit ReLU (Rectified Linear
Units), wherein, pond layer is used to the characteristics of image that convolutional layer before obtains doing an aggregate statistics, can be by multiple numerical value
A numerical value is mapped as, achievees the purpose that reduce data volume；Linear unit is corrected, is mainly used for making the mapping data of convolutional layer to use up
Coefficient of discharge arranges, and is vision reflection of the processing to image closer to people, reaches better handling result.
Two full articulamentums point are set after the full articulamentum for handling and extracting the parts such as the convolutional layer of characteristics of image
Branch, the first branch is the first full articulamentum, and second point of branch is the second full articulamentum, wherein, the first branch is by two
Grader Softmax is accessed after a full connection, whether is human face region by the detection of classifier image；Second branch is also the same
By two full connections, 10 dimensional vectors are then exported, then calculate the using Euclid's loss layer EuclideanLoss
The distance of 10 dimensional vectors of five coordinate compositions of 10 dimensional vectors of two branches output and human face five-sense-organ central point.
Second is obtained using two vectorial distances that the second branch obtains as cost function during convolution god's network training
Second penalty values of branch；In addition, generation of first branch by the use of grader loss layer SoftmaxWithLoss as Face datection
Valency function obtains the first-loss value of Face datection, is then superimposed first-loss value and the second penalty values, obtains entire convolution
The penalty values of neural network recycle this penalty values to be trained network parameters, ultimately form trained treat
Carry out the exclusive convolutional Neural net of recognition of face image.
In entire convolutional neural networks, the obtained data of the last one convolutional layer for carrying out image characteristics extraction will be by
The full articulamentum of Liang Ge branches uses, and during network training, error back propagation can be by the error of Liang Ge branches simultaneously
It is propagated through and.It, can be by the last one above-mentioned convolution in order to avoid interfering with each other for Liang Ge branches caused by error back propagation
Two parts of layer copy is denoted as the copy data copy2 of the first copy data copy1 and second respectively, wherein, the first copy data are by the
One branch calls, and the second copy data are called by the second branch, and in error back propagation, the first of the first copy data compare
Value is calculated by the first branch, and the second fiducial value of the second copy data is calculated by the second branch, is respectively obtaining the first copy
After the fiducial value of data and the second copy data, two the first fiducial values are added with the second fiducial value obtain it is above-mentioned last
The fiducial value of a convolutional layer reaches entire convolutional neural networks training continuing travels back by the last one above-mentioned convolutional layer
Purpose.
After convolutional neural networks train, for a width picture by the calculating of a convolutional neural networks, by two
A branch judges whether the picture is human face region respectively, while the coordinate composition of face point in human face region can also be exported
Vector achievees the purpose that be carried out at the same time Face datection and the calibration of face point.
Entire convolutional neural networks calculate picture just with a convolutional neural networks, it is possible to reduce by multiple convolution
The problem of picture region characteristics of image generated during neural computing repeats to extract and compute repeatedly greatly reduces convolution god
Run time through network, improves calculating speed.
Fig. 2 is a kind of schematic diagram of face identification device according to embodiments of the present invention, as shown in Fig. 2, the device includes：
Extraction unit 21, for extracting the characteristics of image in pending recognition of face image；Detection unit 23, for being based on pending people
Whether characteristics of image in face identification image, detect pending recognition of face image comprising human face region, obtain first as a result, its
In, whether the first result is used to indicate in pending recognition of face image comprising human face region；Unit 25 is demarcated, for being based on treating
The characteristics of image in recognition of face image is carried out, while whether pending recognition of face image is detected comprising human face region,
The face region of pending recognition of face image is demarcated, obtains second as a result, wherein, the second result is used to demarcate pending face
Identify the face region in image；First result and the second result are carried out integration processing, obtain pending people by integral unit 27
Face identifies the recognition result of image.
In the above-described embodiments, by extraction unit, to need to carry out the pending recognition of face image of recognition of face into
Feature extraction of row, and pass through detection unit and calibration unit according to the feature of the pending recognition of face image of extraction difference
Recognition of face and the face calibration of face are carried out, wherein, by detection unit, detect whether pending recognition of face image includes
Human face region obtains the first result；By demarcating unit, the face region of face is demarcated in pending recognition of face image,
Second is obtained as a result, again integrate the first result obtained and the second result by integral unit, obtains pending face
Identify the recognition result of image.Recognition of face and face are carried out by carrying out a feature extraction to pending recognition of face image
Face calibration, without carrying out multiple feature extraction to pending recognition of face image, can accelerate to know pending face
Other image carries out the speed of recognition of face and the face calibration of face, solves and carries out recognition of face journey using convolutional neural networks
The technical issues of sequence is slow.
As an optional embodiment, device further includes：Creation module, for establishing convolutional neural networks, wherein, volume
Product neural network includes the first full articulamentum and the second full articulamentum；Detection unit includes：Detection module, for complete using first
Whether articulamentum detects pending recognition of face image comprising human face region, obtains the first result；Calibration unit includes：Calibration mold
Block for demarcating the face region of pending recognition of face image using the second full articulamentum, obtains the second result.
In an optional embodiment, record has face characteristic in the first full articulamentum that creation module is established, and detects
Module includes：Detection sub-module, for whether detecting in the characteristics of image of pending recognition of face image comprising face characteristic；The
One determining module, in the case of including face characteristic in the characteristics of image for detecting pending recognition of face image, really
Fixed pending recognition of face image includes human face region；Second determining module, for detecting pending recognition of face image
Characteristics of image in do not include face characteristic in the case of, determine pending recognition of face image do not include human face region.
An optional embodiment, creation module establish the second full articulamentum in record have grader, remember in grader
Record has face feature, and demarcating module includes：Module is included into, for passing through grader, by the figure in pending recognition of face image
As feature is included into classification corresponding with face feature respectively, wherein, each classification corresponds to an organic region in face region.
As an optional embodiment, creation module includes：Setup module, for setting the first full articulamentum and second
Full articulamentum；Acquisition module, for obtaining face characteristic and face feature；First training module, for using face characteristic pair
First full articulamentum is trained, and obtains first-loss value, and full articulamentum is trained to second using face feature, is obtained
Second penalty values；Laminating module for first-loss value and the second penalty values to be superimposed, obtains the loss of convolutional neural networks
Value；Second training module, for being trained using penalty values to the parameter in convolutional neural networks, the convolution after being trained
Neural network, wherein, parameter includes each coefficient in the cost function of the first full articulamentum and the second full articulamentum.
Fig. 3 is the schematic diagram of recognition of face machine people according to embodiments of the present invention a kind of, as shown in figure 3, the device packet
It includes：Image collecting device 31, for acquiring pending recognition of face image；Face identification device 32.
In the above-described embodiments, recognition of face machine people acquires pending recognition of face image by image collecting device,
Again by extraction unit, to the pending recognition of face image for carrying out recognition of face is needed to carry out a feature extraction, and pass through
Detection unit and calibration unit carry out recognition of face and face respectively according to the feature of the pending recognition of face image of extraction
Face are demarcated, wherein, by detection unit, whether pending recognition of face image is detected comprising human face region, obtains the first knot
Fruit；By demarcating unit, the face region of face is demarcated in pending recognition of face image, obtains second as a result, passing through again
Integral unit integrates the first result obtained and the second result, obtains the recognition result of pending recognition of face image.
The face that recognition of face and face are carried out by carrying out a feature extraction to pending recognition of face image are demarcated, without right
Pending recognition of face image carries out multiple feature extraction, can accelerate to carry out pending recognition of face image recognition of face and
The speed of the face calibration of face, solves and carries out the technical issues of recognition of face program is slow using convolutional neural networks.
The embodiments of the present invention are for illustration only, do not represent the quality of embodiment.
In the above embodiment of the present invention, all emphasize particularly on different fields to the description of each embodiment, do not have in some embodiment
The part of detailed description may refer to the associated description of other embodiment.
In several embodiments provided herein, it should be understood that disclosed technology contents can pass through others
Mode is realized.Wherein, the apparatus embodiments described above are merely exemplary, such as the division of the unit, Ke Yiwei
A kind of division of logic function, can there is an other dividing mode in actual implementation, for example, multiple units or component can combine or
Person is desirably integrated into another system or some features can be ignored or does not perform.Another point, shown or discussed is mutual
Between coupling, direct-coupling or communication connection can be INDIRECT COUPLING or communication link by some interfaces, unit or module
It connects, can be electrical or other forms.
The unit illustrated as separating component may or may not be physically separate, be shown as unit
The component shown may or may not be physical unit, you can be located at a place or can also be distributed to multiple
On unit.Some or all of unit therein can be selected according to the actual needs to realize the purpose of this embodiment scheme.
In addition, each functional unit in each embodiment of the present invention can be integrated in a processing unit, it can also
That each unit is individually physically present, can also two or more units integrate in a unit.Above-mentioned integrated list
The form that hardware had both may be used in member is realized, can also be realized in the form of SFU software functional unit.
If the integrated unit is realized in the form of SFU software functional unit and is independent product sale or uses
When, it can be stored in a computer read/write memory medium.Based on such understanding, technical scheme of the present invention is substantially
The part to contribute in other words to the prior art or all or part of the technical solution can be in the form of software products
It embodies, which is stored in a storage medium, is used including some instructions so that a computer
Equipment (can be personal computer, server or network equipment etc.) perform each embodiment the method for the present invention whole or
Part steps.And aforementioned storage medium includes：USB flash disk, read-only memory (ROM, Read-Only Memory), arbitrary access are deposited
Reservoir (RAM, Random Access Memory), mobile hard disk, magnetic disc or CD etc. are various can to store program code
Medium.
The above is only the preferred embodiment of the present invention, it is noted that for the ordinary skill people of the art
For member, various improvements and modifications may be made without departing from the principle of the present invention, these improvements and modifications also should
It is considered as protection scope of the present invention.
Claims (11)
1. a kind of face identification method, which is characterized in that including：
Extract the characteristics of image in pending recognition of face image；
Based on the characteristics of image in the pending recognition of face image, detect whether the pending recognition of face image includes
Human face region obtains first as a result, wherein, first result is used to indicate in the pending recognition of face image and whether wraps
Containing human face region；
Based on the characteristics of image in the pending recognition of face image, detecting whether the pending recognition of face image wraps
While containing human face region, demarcates the face region of the pending recognition of face image, obtains second as a result, wherein, described
Second result is used to demarcate the face region in the pending recognition of face image；
First result and second result are subjected to integration processing, obtain the identification of the pending recognition of face image
As a result.
2. according to the method described in claim 1, it is characterized in that,
Before the characteristics of image in extracting pending recognition of face image, the method further includes：Establish convolutional neural networks,
Wherein, the convolutional neural networks include the first full articulamentum and the second full articulamentum；
Detect whether the pending recognition of face image includes comprising human face region：Utilize the described first full articulamentum detection institute
Pending recognition of face image is stated whether comprising human face region, obtains first result；
The face region for demarcating the pending recognition of face image includes：Using treated described in the described second full articulamentum calibration into
The face region of row recognition of face image obtains second result.
3. according to the method described in claim 2, it is characterized in that, in the first full articulamentum record have face characteristic, profit
Detect whether the pending recognition of face image includes comprising human face region with the described first full articulamentum：
It whether detects in the characteristics of image of the pending recognition of face image comprising the face characteristic；
If it is detected that in the characteristics of image of the pending recognition of face image include the face characteristic, it is determined that it is described treat into
Row recognition of face image includes the human face region；
If it is detected that the face characteristic is not included in the characteristics of image of the pending recognition of face image, it is determined that described to treat
It carries out recognition of face image and does not include the human face region.
4. according to the method described in claim 2, it is characterized in that, in the second full articulamentum record have grader, it is described
Record has face feature in grader, and the face area of the pending recognition of face image is demarcated using the described second full articulamentum
Domain includes：
By the grader, the characteristics of image in the pending recognition of face image is included into and the face feature respectively
Corresponding classification, wherein, each classification corresponds to an organic region in the face region.
5. according to the method described in claim 2, include it is characterized in that, establishing convolutional neural networks：
First full articulamentum and the second full articulamentum are set；
Obtain face characteristic and face feature；
The described first full articulamentum is trained using the face characteristic, obtains first-loss value, and use the face
Feature is trained the described second full articulamentum, obtains the second penalty values；
The first-loss value and second penalty values are superimposed, obtain the penalty values of the convolutional neural networks；
The parameter in the convolutional neural networks is trained using the penalty values, the convolutional Neural net after being trained
Network, wherein, the parameter includes each coefficient in the cost function of the described first full articulamentum and the second full articulamentum.
6. a kind of face identification device, which is characterized in that including：
Extraction unit, for extracting the characteristics of image in pending recognition of face image；
Detection unit, for based on the characteristics of image in the pending recognition of face image, the detection pending face to be known
Whether other image obtains first as a result, wherein, first result be used to indicate the pending face know comprising human face region
Whether human face region is included in other image；
Unit is demarcated, for based on the characteristics of image in the pending recognition of face image, detecting the pending face
While identifying whether image includes human face region, the face region of the pending recognition of face image is demarcated, obtains second
As a result, wherein, second result is used to demarcate the face region in the pending recognition of face image；
First result and second result are carried out integration processing, obtain the pending recognition of face by integral unit
The recognition result of image.
7. device according to claim 6, which is characterized in that
Described device further includes：Creation module, for establishing convolutional neural networks, wherein, the convolutional neural networks include the
One full articulamentum and the second full articulamentum；
The detection unit includes：Detection module, for detecting the pending recognition of face using the described first full articulamentum
Whether image obtains first result comprising human face region；
The calibration unit includes：Demarcating module, for demarcating the pending recognition of face using the described second full articulamentum
The face region of image obtains second result.
8. device according to claim 7, which is characterized in that recorded in the first full articulamentum that the creation module is established
There is face characteristic, the detection module includes：
Detection sub-module, it is whether special comprising the face in the characteristics of image of the pending recognition of face image for detecting
Sign；
First determining module, it is special for including the face in the characteristics of image for detecting the pending recognition of face image
In the case of sign, determine that the pending recognition of face image includes the human face region；
Second determining module, for not including the face in the characteristics of image for detecting the pending recognition of face image
In the case of feature, determine that the pending recognition of face image does not include the human face region.
9. device according to claim 7, which is characterized in that recorded in the second full articulamentum that the creation module is established
There is grader, being recorded in the grader has face feature, and the demarcating module includes：
Module is included into, for passing through the grader, the characteristics of image in the pending recognition of face image is included into respectively
Classification corresponding with the face feature, wherein, each classification corresponds to an organic region in the face region.
10. device according to claim 7, which is characterized in that the creation module includes：
Setup module, for setting the first full articulamentum and the second full articulamentum；
Acquisition module, for obtaining face characteristic and face feature；
First training module for being trained using the face characteristic to the described first full articulamentum, obtains first-loss
Value, and the described second full articulamentum is trained using the face feature, obtain the second penalty values；
Laminating module for the first-loss value and second penalty values to be superimposed, obtains the convolutional neural networks
Penalty values；
Second training module for being trained using the penalty values to the parameter in the convolutional neural networks, is instructed
Convolutional neural networks after white silk, wherein, the parameter includes the cost of the described first full articulamentum and the second full articulamentum
Each coefficient in function.
11. a kind of robot, which is characterized in that including：
Image collecting device, for acquiring pending recognition of face image；Face in claim 6 to 10 described in any one
Identification device.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201611263999.4A CN108268822A (en) | 2016-12-30 | 2016-12-30 | Face identification method, device and robot |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201611263999.4A CN108268822A (en) | 2016-12-30 | 2016-12-30 | Face identification method, device and robot |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN108268822A true CN108268822A (en) | 2018-07-10 |
Family
ID=62755320
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201611263999.4A Pending CN108268822A (en) | 2016-12-30 | 2016-12-30 | Face identification method, device and robot |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN108268822A (en) |
Cited By (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN110186167A (en) * | 2019-05-31 | 2019-08-30 | 广东美的制冷设备有限公司 | Control method, device, air conditioner and the storage medium of air conditioner |
| CN110987189A (en) * | 2019-11-21 | 2020-04-10 | 北京都是科技有限公司 | Method, system and device for detecting temperature of target object |
| CN111898622A (en) * | 2019-05-05 | 2020-11-06 | 阿里巴巴集团控股有限公司 | Information processing method, information display method, model training method, information display system, model training system and equipment |
| CN113239885A (en) * | 2021-06-04 | 2021-08-10 | 新大陆数字技术股份有限公司 | Face detection and recognition method and system |
Citations (5)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN1967562A (en) * | 2005-11-15 | 2007-05-23 | 中华电信股份有限公司 | Facial identification method based on human facial features identification |
| CN105354565A (en) * | 2015-12-23 | 2016-02-24 | 北京市商汤科技开发有限公司 | Full convolution network based facial feature positioning and distinguishing method and system |
| CN105469041A (en) * | 2015-11-19 | 2016-04-06 | 上海交通大学 | Facial point detection system based on multi-task regularization and layer-by-layer supervision neural networ |
| US20160140436A1 (en) * | 2014-11-15 | 2016-05-19 | Beijing Kuangshi Technology Co., Ltd. | Face Detection Using Machine Learning |
| CN105912990A (en) * | 2016-04-05 | 2016-08-31 | 深圳先进技术研究院 | Face detection method and face detection device |
-
2016
- 2016-12-30 CN CN201611263999.4A patent/CN108268822A/en active Pending
Patent Citations (5)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN1967562A (en) * | 2005-11-15 | 2007-05-23 | 中华电信股份有限公司 | Facial identification method based on human facial features identification |
| US20160140436A1 (en) * | 2014-11-15 | 2016-05-19 | Beijing Kuangshi Technology Co., Ltd. | Face Detection Using Machine Learning |
| CN105469041A (en) * | 2015-11-19 | 2016-04-06 | 上海交通大学 | Facial point detection system based on multi-task regularization and layer-by-layer supervision neural networ |
| CN105354565A (en) * | 2015-12-23 | 2016-02-24 | 北京市商汤科技开发有限公司 | Full convolution network based facial feature positioning and distinguishing method and system |
| CN105912990A (en) * | 2016-04-05 | 2016-08-31 | 深圳先进技术研究院 | Face detection method and face detection device |
Non-Patent Citations (2)
| Title |
|---|
| YANG S, LUO P, LOY C C: "《From Facial Parts Responses to Face Detection: A Deep》", 《IEEE INTERNATIONAL CONFERENCE ON COMPUTER VISION》 * |
| 邵蔚元; 郭跃飞: "《多任务学习及卷积神经网络在人脸识别中的应用》", 《计算机工程与应用》 * |
Cited By (6)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN111898622A (en) * | 2019-05-05 | 2020-11-06 | 阿里巴巴集团控股有限公司 | Information processing method, information display method, model training method, information display system, model training system and equipment |
| CN111898622B (en) * | 2019-05-05 | 2022-07-15 | 阿里巴巴集团控股有限公司 | Information processing method, information display method, model training method, information display system, model training system and equipment |
| CN110186167A (en) * | 2019-05-31 | 2019-08-30 | 广东美的制冷设备有限公司 | Control method, device, air conditioner and the storage medium of air conditioner |
| CN110987189A (en) * | 2019-11-21 | 2020-04-10 | 北京都是科技有限公司 | Method, system and device for detecting temperature of target object |
| CN110987189B (en) * | 2019-11-21 | 2021-11-02 | 北京都是科技有限公司 | Method, system and device for detecting temperature of target object |
| CN113239885A (en) * | 2021-06-04 | 2021-08-10 | 新大陆数字技术股份有限公司 | Face detection and recognition method and system |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN107657249A (en) | Method, apparatus, storage medium and the processor that Analysis On Multi-scale Features pedestrian identifies again | |
| CN105976400B (en) | Method for tracking target and device based on neural network model | |
| CN111340008B (en) | Method and system for generation of counterpatch, training of detection model and defense of counterpatch | |
| CN108268822A (en) | Face identification method, device and robot | |
| CN106778684A (en) | deep neural network training method and face identification method | |
| CN106372662A (en) | Helmet wearing detection method and device, camera, and server | |
| CN110738101A (en) | Behavior recognition method and device and computer readable storage medium | |
| CN107862282A (en) | A kind of finger vena identification and safety certifying method and its terminal and system | |
| CN110059546A (en) | Vivo identification method, device, terminal and readable medium based on spectrum analysis | |
| CN109670441A (en) | A kind of realization safety cap wearing knows method for distinguishing, system, terminal and computer readable storage medium | |
| CN109359666A (en) | A kind of model recognizing method and processing terminal based on multiple features fusion neural network | |
| CN107194361A (en) | Two-dimentional pose detection method and device | |
| CN110287889A (en) | A kind of method and device of identification | |
| CN108256404A (en) | Pedestrian detection method and device | |
| CN109902660A (en) | A kind of expression recognition method and device | |
| CN107578034A (en) | information generating method and device | |
| CN107967442A (en) | A kind of finger vein identification method and system based on unsupervised learning and deep layer network | |
| CN108009481A (en) | A kind of training method and device of CNN models, face identification method and device | |
| CN110619316A (en) | Human body key point detection method and device and electronic equipment | |
| CN115880558B (en) | Farming behavior detection method and device, electronic equipment and storage medium | |
| CN111914665A (en) | Face shielding detection method, device, equipment and storage medium | |
| CN108614987A (en) | The method, apparatus and robot of data processing | |
| CN109117746A (en) | Hand detection method and machine readable storage medium | |
| CN110069983A (en) | Vivo identification method, device, terminal and readable medium based on display medium | |
| CN108268823A (en) | Target recognition methods and device again |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| PB01 | Publication | ||
| PB01 | Publication | ||
| TA01 | Transfer of patent application right | ||
| TA01 | Transfer of patent application right |
Effective date of registration: 20180822 Address after: 311100 1101, room 14, 1008, Longxiang street, Cang Qian street, Yuhang District, Hangzhou, Zhejiang. Applicant after: Hangzhou Institute of artificial intelligence Address before: 518000 Guangdong, Shenzhen, Nanshan District, Nanhai Road, West Guangxi Temple Road North Sunshine Huayi Building 1 15D-02F Applicant before: Shenzhen Guangqi Hezhong Technology Co., Ltd. Applicant before: Shenzhen Kuang-Chi Innovation Technology Co., Ltd. |
|
| 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 | ||
| RJ01 | Rejection of invention patent application after publication |
Application publication date: 20180710 |