CN106485235A - A kind of convolutional neural networks generation method, age recognition methods and relevant apparatus - Google Patents
A kind of convolutional neural networks generation method, age recognition methods and relevant apparatus Download PDFInfo
- Publication number
- CN106485235A CN106485235A CN201610925676.0A CN201610925676A CN106485235A CN 106485235 A CN106485235 A CN 106485235A CN 201610925676 A CN201610925676 A CN 201610925676A CN 106485235 A CN106485235 A CN 106485235A
- Authority
- CN
- China
- Prior art keywords
- age
- convolutional neural
- neural networks
- grader
- face
- 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
- 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
-
- 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
- G06N3/084—Backpropagation, e.g. using gradient descent
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
- G06V40/00—Recognition of biometric, human-related or animal-related patterns in image or video data
- G06V40/10—Human or animal bodies, e.g. vehicle occupants or pedestrians; Body parts, e.g. hands
- G06V40/16—Human faces, e.g. facial parts, sketches or expressions
- G06V40/178—Human faces, e.g. facial parts, sketches or expressions estimating age from face image; using age information for improving recognition
Landscapes
- Engineering & Computer Science (AREA)
- Theoretical Computer Science (AREA)
- Physics & Mathematics (AREA)
- General Health & Medical Sciences (AREA)
- Health & Medical Sciences (AREA)
- General Physics & Mathematics (AREA)
- Biomedical Technology (AREA)
- Mathematical Physics (AREA)
- Data Mining & Analysis (AREA)
- Evolutionary Computation (AREA)
- Biophysics (AREA)
- Molecular Biology (AREA)
- Computing Systems (AREA)
- General Engineering & Computer Science (AREA)
- Artificial Intelligence (AREA)
- Computational Linguistics (AREA)
- Software Systems (AREA)
- Life Sciences & Earth Sciences (AREA)
- Oral & Maxillofacial Surgery (AREA)
- Human Computer Interaction (AREA)
- Multimedia (AREA)
- Image Analysis (AREA)
- Image Processing (AREA)
Abstract
The invention discloses a kind of generation method for carrying out the convolutional neural networks of age identification to the face in image, age recognition methods, relevant apparatus and computing device, the generation method of the convolutional neural networks includes：First convolutional neural networks are trained, the first convolutional neural networks include multiple convolution groups, multiple full articulamentums and the first grader being sequentially connected；Complete for part in the first convolutional neural networks for training articulamentum and the first grader are replaced accordingly, is generated the second convolutional neural networks and simultaneously which is trained；Add new full articulamentum and grader to the second convolutional neural networks for training, to generate the 3rd convolutional neural networks and be trained；Add new full articulamentum and grader to the 3rd convolutional neural networks for training, to generate Volume Four product neutral net and be trained.Wherein, before being trained to above-mentioned each convolutional neural networks, the human face image information for training can be anticipated.
Description
Technical field
The present invention relates to technical field of image processing, more particularly to a kind of for carrying out age identification to the face in image
Convolutional neural networks generation method, age recognition methods, relevant apparatus and computing device.
Background technology
Face has contained bulk information, such as age, sex, ethnic group as one of important biological characteristic, on facial image
Deng.With going deep into further to facial image research in image processing techniques, particularly in terms of the identification of face age, with volume
Product neutral net (CNN：Convolutional Neural Network) based on the recognition methods of face age also gradually send out
Exhibition is got up, and all has important effect in numerous reality scenes.In 2014《Computer Science》On,
Karen Simonyan and Andrew Zisserman have delivered an entitled " Very Deep Convolutional
A kind of paper of Networks for Large-Scale Image Recognition ", it is proposed that the deeper convolution god of level
Through network model, i.e. VGG (Visual Geometry Group) model, the degree of accuracy for carrying out age identification based on the model have
Certain lifting.
However, for the facial image of different age group, between the amplitude of its change, gap is larger, the especially baby of 0～5 years old
Child, change are very huge, but the adult of 25～30 years old or so, then varying less, this phenomenon is brought to age identification
Large effect.In existing face age recognition methods, replace age true value using age distribution value, alleviate to a certain extent
The problems referred to above, but the result for getting is in a certain age range, and not accurate enough, once malfunction in interval critical value,
Then age estimate can fall into adjacent interval, bring larger error.And as the numerical value span at age is larger, even if taking
Above-mentioned VGG model carries out age identification, it is also desirable to which the distribution according to age bracket is taken multiple models to realize, complex, and
Versatility is relatively low.
Content of the invention
For this purpose, the present invention provides a kind of convolutional neural networks generation side for carrying out age identification to the face in image
Case, and propose the age identifying schemes based on the convolutional neural networks, is solved with trying hard to or at least alleviates and exist above
Problem.
According to an aspect of the present invention, a kind of convolutional Neural for carrying out age identification to the face in image is provided
Network generation method, is suitable to execute in computing device, and the method comprises the steps：First, according to the advance face for obtaining
First convolutional neural networks are trained being applied to identification face, people so as to the first convolutional neural networks by sets of image data
Face image data acquisition system includes multiple human face image information, and each facial image packet includes people in facial image and correspondence image
Age information, multiple convolution groups that the first convolutional neural networks include to be sequentially connected, the first full articulamentum, the second full connection
Layer, the 3rd full articulamentum and the first grader；According to default age threshold to each face in face image data set
Image information is processed, and to add new age threshold attribute for each human face image information, age threshold attribute is indicated
The age of corresponding people is greater than default age threshold and is also no more than default age threshold；By the first convolution for training
The 3rd full articulamentum and the first grader in neutral net replaces with the 4th full articulamentum and the second grader respectively, to generate
Second convolutional neural networks, and according to the addition of the face image data set of age threshold attribute to the second convolutional neural networks
It is trained, so that the age of people corresponding to the output indication face of the second grader of the second convolutional neural networks is above year
Age threshold value is also no greater than age threshold；After the first full articulamentum in the second convolutional neural networks for training, add
The 5th full articulamentum, the 6th full articulamentum and the 3rd grader being sequentially connected, to generate the 3rd convolutional neural networks, and selects
In face image data set, the age be not more than default age threshold human face image information to the 3rd convolutional neural networks
It is trained, so that the age of people corresponding to the output indication face of the 3rd grader of the 3rd convolutional neural networks is zero to institute
Which in default age threshold；After the first full articulamentum in the 3rd convolutional neural networks for training, add according to
Secondary connected 7th full articulamentum, eight convergent points articulamentum and the 4th grader, accumulate neutral net to generate Volume Four, and according to people
Face image data acquisition system is trained to Volume Four product neutral net, so that people corresponding to the output indication face of the 4th grader
Age.
Alternatively, generate in the convolutional neural networks for carrying out age identification to the face in image according to the present invention
In method, in face image data set, the facial image of each human face image information all keeps horizontal front and meets presetting chi
Very little, facial image corresponds to the integer that the age of people is between 0～100.
Alternatively, generate in the convolutional neural networks for carrying out age identification to the face in image according to the present invention
In method, in each convolution group of the first convolutional neural networks, all include at least one convolutional layer.
Alternatively, generate in the convolutional neural networks for carrying out age identification to the face in image according to the present invention
In method, after the first full articulamentum in the second convolutional neural networks for training, add the 5th for being sequentially connected complete
Articulamentum, the 6th full articulamentum and the 3rd grader, the step of to generate three convolutional neural networks before, also include step：
To in face image data set, the age is not more than age of human face image information of default age threshold carries out 0/1 coding,
0/1 coding includes to add 1 sum as encoding digit with default age threshold, and each is any one in numeral 0 and numeral 1,
From the beginning of first place, the difference that the number of times of 1 appearance of numeral subtracts 1 is the age.
Alternatively, generate in the convolutional neural networks for carrying out age identification to the face in image according to the present invention
In method, the 6th full articulamentum includes the full articulamentum of the son of multiple parallel connections, and the number of the full articulamentum of son is default age threshold
Plus 1 sum.
Alternatively, generate in the convolutional neural networks for carrying out age identification to the face in image according to the present invention
In method, after the first full articulamentum in the 3rd convolutional neural networks for training, add the 7th for being sequentially connected complete
Articulamentum, eight convergent points articulamentum and the 4th grader, before generating the step of Volume Four accumulates neutral net, also to include step：
Age to human face image information in face image data set carries out distributed code, and distributed code includes to be entered according to Gaussian Profile
The row age encodes.
Alternatively, generate in the convolutional neural networks for carrying out age identification to the face in image according to the present invention
In method, default age threshold is 12.
According to a further aspect of the invention, a kind of age recognition methods is provided, is suitable to execute in computing device, the party
Method is based on the Volume Four product for carrying out to the face in image in the convolutional neural networks generation method of age identification, training
Neutral net carries out age identification to the face in image, including step：Facial image to be identified is input to for training
Age identification is carried out in four convolutional neural networks；The good Volume Four of training of judgement accumulates the output of the second grader in neutral net
No more than default age threshold；If the output of the second grader is not more than default age threshold, for training is obtained
In four convolutional neural networks, the 3rd grader is output as the age of people corresponding to face；If the output of the second grader is more than pre-
If age threshold, then obtain train Volume Four product neutral net in the 4th grader be output as people corresponding to face
Age.
Alternatively, in the age recognition methods according to the present invention, also include to pre-process to obtain images to be recognized
Take facial image to be identified.
Alternatively, in the age recognition methods according to the present invention, images to be recognized is pre-processed and is waited to know to obtain
Others includes face image：Face datection is carried out to images to be recognized, obtains face location information；By face location information, will
Change to pre-set dimension after face cutting in images to be recognized；Face is calculated according to face key point information carries out Plane Rotation
Transformation matrix；The facial image under pre-set dimension is rotated into horizontal front to obtain face figure to be identified using transformation matrix
Picture.
According to a further aspect of the invention, a kind of convolution god for carrying out age identification to the face in image is provided
Through network generating means, be suitable to reside in computing device, the device include the first training module, attribute add module, first
Generation module, the second training module, the second generation module, the 3rd training module, the 3rd generation module and the 4th training module.Its
In, the first training module is suitable to, according to the face image data set for obtaining in advance, be trained the first convolutional neural networks
It is applied to identification face so as to the first convolutional neural networks, face image data set includes multiple human face image information, each
Human face image information includes the age information of people in facial image and correspondence image, and the first convolutional neural networks include to be sequentially connected
Multiple convolution groups, the first full articulamentum, the second full articulamentum, the 3rd full articulamentum and the first grader；Attribute add module
It is suitable to process each human face image information in face image data set according to default age threshold, to be every
Individual human face image information adds new age threshold attribute, and it is default that age threshold attribute indicates that the age of corresponding people is greater than
Age threshold is also no more than default age threshold；First generation module is suitable in the first convolutional neural networks that will be trained
The 3rd full articulamentum and the first grader replace with the 4th full articulamentum and the second grader respectively, with generate the second convolution god
Through network；Second training module is suitable to basis and with the addition of the face image data set of age threshold attribute to the second convolutional Neural
Network is trained, so that the age of people corresponding to the output indication face of the second grader of the second convolutional neural networks is high
Age threshold is also no greater than in age threshold；Second generation module is suitable in the second convolutional neural networks for training
After one full articulamentum, add the 5th full articulamentum, the 6th full articulamentum and the 3rd grader being sequentially connected, to generate the 3rd
Convolutional neural networks；3rd training module is suitably selected in face image data set, the age is not more than default age threshold
The human face image information of value is trained to the 3rd convolutional neural networks, so that the 3rd grader of the 3rd convolutional neural networks
The age of people corresponding to output indication face by zero to preset age threshold in which；3rd generation module is suitable in instruction
After the first full articulamentum in the 3rd convolutional neural networks that perfects, add the 7th full articulamentum, the eight convergent points being sequentially connected
Articulamentum and the 4th grader, accumulate neutral net to generate Volume Four；4th training module is suitable to according to face image data collection
Close and Volume Four product neutral net is trained, so that the age of people corresponding to the output indication face of the 4th grader.
According to a further aspect of the invention, a kind of age identifying device is provided, is suitable to reside in computing device, the dress
Put based on the Volume Four product for carrying out to the face in image in the convolutional neural networks generating means of age identification, training
Neutral net carries out age identification to the face in image, including identification module, judge module and acquisition module.Wherein, recognize
Module is suitable to be input to facial image to be identified in the Volume Four product neutral net for training carries out age identification；Judge module
It is suitable to judge that identification module carries out the output that the Volume Four for training after age identification accumulates the second grader in neutral net；Obtain
Delivery block is suitable to, when the output that judge module judges the second grader is not more than default age threshold, obtain identification module
Carry out the year that the 3rd grader in the Volume Four product neutral net for training after age identification is output as people corresponding to face
In age, when judge module judges the output of the second grader more than default age threshold, obtaining identification module carries out the age
In the Volume Four product neutral net for training after identification, the 4th grader is output as the age of people corresponding to face.
According to a further aspect of the invention, a kind of computing device is also provided, including according to the present invention for image
In face carry out the generating means of convolutional neural networks and the age identifying device of age identification.
The technical side that the convolutional neural networks for carrying out age identification to the face in image according to the present invention are generated
First convolutional neural networks are trained by case first, multiple convolution groups that the first convolutional neural networks include to be sequentially connected,
Multiple full articulamentums and the first grader, by complete for part in the first convolutional neural networks for training articulamentum and the first classification
Device is replaced accordingly, generates the second convolutional neural networks and which is trained, then to the second convolutional Neural for training
Network adds new full articulamentum and grader, to generate the 3rd convolutional neural networks and be trained, most trains backward
3rd convolutional neural networks add new full articulamentum and grader, to generate Volume Four product neutral net and be trained.?
In technique scheme, framework based on convolutional neural networks, and before each convolutional neural networks are generated in case training,
To for training face image data carry out such as age threshold segmentation, 0/1 coding and age distribution coding etc. process, aid in by
The model of step modification convolutional neural networks, ultimately forms a convolutional neural networks model that can accurately identify the age, without the need for dividing
Cut, succinct directly perceived, versatility is higher.Further, according to the age recognition methods of the present invention, facial image to be identified is input to
In the Volume Four product neutral net for training, the range of age and its concrete numerical value are judged according to the output of different classifications device, knot
Fruit accuracy has huge lifting.
Description of the drawings
In order to above-mentioned and related purpose is realized, some illustrative sides are described herein in conjunction with explained below and accompanying drawing
Face, indicate in terms of these can be to put into practice principles disclosed herein various modes, and all aspects and its equivalent aspect
It is intended to fall under in the range of theme required for protection.By being read in conjunction with the accompanying detailed description below, the disclosure above-mentioned
And other purposes, feature and advantage will be apparent from.Throughout the disclosure, identical reference generally refers to identical
Part or element.
Fig. 1 shows the schematic diagram of computing device according to an embodiment of the invention 100；
Fig. 2 shows the convolution god for carrying out age identification to the face in image according to an embodiment of the invention
Flow chart through network generation method 200；
Fig. 3 shows the convolution for carrying out age identification to the face in image according to another embodiment of the present invention
The flow chart of neutral net generation method 300；
Fig. 4 shows the structural representation of the first convolutional neural networks according to an embodiment of the invention；
Fig. 5 shows the structural representation of the second convolutional neural networks according to an embodiment of the invention；
Fig. 6 shows the structural representation of the 3rd convolutional neural networks according to an embodiment of the invention；
Fig. 7 shows that Volume Four according to an embodiment of the invention accumulates the structural representation of neutral net；
The flow chart that Fig. 8 shows age recognition methods 400 according to an embodiment of the invention；
The flow chart that Fig. 9 shows the age recognition methods 500 according to another embodiment of the present invention；
Figure 10 shows the volume for carrying out age identification to the face in image according to an embodiment of the invention
The schematic diagram of product neutral net generating means 600；
Figure 11 show according to still another embodiment of the invention for carrying out age identification to the face in image
The schematic diagram of convolutional neural networks generating means 700；
Figure 12 shows the schematic diagram of age identifying device 800 according to an embodiment of the invention；And
Figure 13 shows the schematic diagram of age identifying device 900 according to still another embodiment of the invention.
Specific embodiment
The exemplary embodiment of the disclosure is more fully described below with reference to accompanying drawings.Although the disclosure is shown in accompanying drawing
Exemplary embodiment, it being understood, however, that may be realized in various forms the disclosure and should not be by embodiments set forth here
Limited.Conversely, provide these embodiments to be able to be best understood from the disclosure, and can be by the scope of the present disclosure
Complete conveys to those skilled in the art.
Fig. 1 is the block diagram of Example Computing Device 100.In basic configuration 102, computing device 100 typically comprises and is
System memory 106 and one or more processor 104.Memory bus 108 can be used for storing in processor 104 and system
Communication between device 106.
Depending on desired configuration, processor 104 can be any kind of process, including but not limited to：Microprocessor
(μ P), microcontroller (μ C), digital information processor (DSP) or any combination of them.Processor 104 can be included such as
The cache of one or more rank of on-chip cache 110 and second level cache 112 etc, processor core
114 and register 116.The processor core 114 of example can include arithmetic and logical unit (ALU), floating-point unit (FPU),
Digital signal processing core (DSP core) or any combination of them.The Memory Controller 118 of example can be with processor
104 are used together, or in some implementations, Memory Controller 118 can be an interior section of processor 104.
Depending on desired configuration, system storage 106 can be any type of memory, including but not limited to：Easily
The property lost memory (RAM), nonvolatile memory (ROM, flash memory etc.) or any combination of them.System is stored
Device 106 can include operating system 120, one or more application 122 and routine data 124.In some embodiments,
Application 122 may be arranged to be operated using routine data 124 on an operating system.
Computing device 100 can also include to contribute to from various interface equipments (for example, output equipment 142, Peripheral Interface
144 and communication equipment 146) to basic configuration 102 via the communication of bus/interface controller 130 interface bus 140.Example
Output equipment 142 include GPU 148 and audio treatment unit 150.They can be configured to contribute to via
One or more A/V port 152 is communicated with the various external equipments of such as display or loudspeaker etc.Outside example
If interface 144 can include serial interface controller 154 and parallel interface controller 156, they can be configured to contribute to
Via one or more I/O port 158 and such as input equipment (for example, keyboard, mouse, pen, voice-input device, touch
Input equipment) or the external equipment of other peripheral hardwares (such as printer, scanner etc.) etc communicated.The communication of example sets
Standby 146 can include network controller 160, and which can be arranged to be easy to via one or more COM1 164 and
Other communications of computing device 162 by network communication link individual or multiple.
Network communication link can be an example of communication media.Communication media can be generally presented as in such as carrier wave
Or the computer-readable instruction in the modulated data signal of other transmission mechanisms etc, data structure, program module, and can
To include any information delivery media." modulated data signal " can be with such signal, and in its data set is many
Individual or it change can the mode of coding information in the signal carry out.Used as nonrestrictive example, communication media is permissible
Including the wire medium of such as cable network or private line network etc and such as sound, radio frequency (RF), microwave, infrared
Or other wireless mediums are in interior various wireless mediums (IR).Term computer-readable medium used herein can include to deposit
Both storage media and communication media.
Computing device 100 can be implemented as server, and these servers can be such as file server, database service
The servers such as device, apps server and WEB server.Computing device 100 can be implemented as small size portable (or move
Dynamic) part of electronic equipment, these electronic equipments can be such as cell phone, personal digital assistant (PDA), individual media
Player device, wireless network browsing apparatus, personal helmet, application specific equipment or any of the above work(can be included
The mixing apparatus of energy.Computing device 100 is also implemented as including the personal meter of desktop computer and notebook computer configuration
Calculation machine.In certain embodiments, computing device 100 be configured to execute according to the present invention for carrying out to the face in image
The generation method of the convolutional neural networks of age identification and age recognition methods.Application 122 is included according to the present invention for right
Face in image carries out the convolutional neural networks generating means 600 of age identification and age identifying device 700.
Fig. 2 shows the convolution god for carrying out age identification to the face in image according to an embodiment of the invention
Flow chart through network generation method 200.For carrying out the convolutional neural networks generation side of age identification to the face in image
Method 200 is suitable to execute in computing device (computing device 100 for example shown in Fig. 1).
As shown in Fig. 2 method 200 starts from step S210.In step S210, according to the advance face image data for obtaining
First convolutional neural networks are trained being applied to identification face, facial image number so as to the first convolutional neural networks by set
Include multiple human face image information according to set, each facial image packet includes the age letter of people in facial image and correspondence image
Breath is multiple convolution groups that the first convolutional neural networks include to be sequentially connected, the first full articulamentum, the second full articulamentum, the 3rd complete
Articulamentum and the first grader.Wherein, in face image data set, the facial image of each human face image information all keeps water
Straight and even face and meet pre-set dimension, the facial image corresponds to the integer that the age of people is between 0～100, the first convolutional Neural
All include at least one convolutional layer in each convolution group of network.
In the present embodiment, face image data collection is combined into the human face data collection of MsCeleb, and the human face data is concentrated
For human face image information, its facial image for including is the image after rotation processing, i.e., in advance by the facial image rotation
Switch to horizontal front, and facial image is Three Channel Color image, pre-set dimension is 224px × 224px.First convolution nerve net
Network is a kind of in the VGG model for adopting, and Fig. 4 shows the structure of the first convolutional neural networks according to an embodiment of the invention
Schematic diagram.As shown in figure 4, in the first convolutional neural networks, including five convolution groups, with the first convolution group as input, after
Face is sequentially connected the second convolution group, the 3rd convolution group, Volume Four product group, the 5th convolution group, the first full articulamentum, the second full connection
Layer, the 3rd full articulamentum and the first grader, wherein the first grader are output end, are the defeated of the first convolution group with facial image
Enter, the age of people is the first grader in correspondence image output carries out the training of the first convolutional neural networks.In fact, adjacent
Two convolution groups between all there is a down-sampling layer, and there is also under one between the 5th convolution group and the first full articulamentum
Sample level, according to the order from input to output end, each down-sampling layer is become the first down-sampling layer, the second down-sampling successively
Layer, the 3rd down-sampling layer, the 4th down-sampling layer and the 5th down-sampling layer.All include two in first convolution group and the second convolution group
Convolutional layer, all includes three convolutional layers in the 3rd convolution group, Volume Four product group and the 5th convolution group.With face image data set
In human face image information A as a example by illustrate, human face image information A includes the age of people in facial image A1 and correspondence image
Information A2, A2 are 7 years old.
In the first convolutional neural networks, first A1 is input to the first convolution group.A1 is Three Channel Color image, size
For 224px × 224px.Two convolutional layers in first convolution group all have 64 convolution kernels, each convolution in first convolutional layer
The number of parameters of core is 3 × 3 × 3, carries out convolution equivalent to the convolution kernel of 33 × 3 sizes in each passage, and step-length is 1.Warp
After crossing the convolution of first convolutional layer, according to (224-3)/1+1=222, the size of the image for now obtaining be 222px ×
222px, that is, obtain the characteristic pattern of 64 222px × 222px sizes.Due to triple channel being closed in first convolutional layer
Convolution is carried out together, and the therefore input of second convolutional layer is the single channel image of 64 222px × 222px.Second convolution
In layer, the number of parameters of each convolution kernel is 3 × 3, carries out convolution equivalent to the convolution kernel of 13 × 3 size, and step-length is 1.Then warp
After crossing the convolution of second convolutional layer, according to (222-3)/1+1=220,64 220px × 220px are now obtained big
Little characteristic pattern.Subsequently, the first down-sampling layer is entered, and down-sampling is also known as pond, is the principle using image local correlation,
Sub-sample is carried out to image, so as to reduce under data processing and retain useful information.Here, pond is using Maximum overlap pond
Changing, i.e., piecemeal being carried out to the characteristic pattern of 220px × 220px, the size of each block is 2 × 2, and step-length is 2, and counts each block
Maximum, used as the pixel value of image behind pond.According to the characteristic pattern of (220-2)/2+1=27, Chi Huahou a size of
110px × 110px, then after the first down-sampling layer, obtain the characteristic pattern of 64 110px × 110px.
Two convolutional layers in second convolution group all have 128 convolution kernels, and three convolutional layers in the 3rd convolution group all have
256 convolution kernels, three convolutional layers in Volume Four product group all have 512 convolution kernels, three convolutional layers in the 5th convolution group
Also all there are 512 convolution kernels, the wherein number of parameters of each convolution kernel is 3 × 3, enters equivalent to the convolution kernel of 13 × 3 size
Row convolution, step-length are 1.Second down-sampling layer, the 3rd down-sampling layer, the 4th down-sampling layer and the 5th down-sampling layer are all using maximum
Overlapping pool, the size for carrying out the block of piecemeal process to characteristic pattern are 2 × 2, and step-length is 2, according under the first convolution group and first
The processing procedure of sample level, finally gives the characteristic pattern that the 5th down-sampling layer is output as 512 2px × 2px.Wherein, entering
During row Maximum overlap pond, as the length of side of characteristic pattern is not necessarily 2 multiple, the process side for retaining edge is therefore taken
Method, will characteristic pattern the length of side be filled to 02 multiple, then carry out pond again.In fact, completing volume in each convolutional layer
After product is processed, in addition it is also necessary to adjust the output of the convolutional layer by activation primitive, it is to avoid the next layer of line for being output as last layer
Property combination and arbitrary function cannot be approached.Enter one using ReLU (Rectified Linear Unit) function as activation primitive
Step alleviates over-fitting problem, and its expression formula is conventional linear function for f (x)=max (0, w x+b), wherein w x+b.Per
Individual convolutional layer completes, to image, the characteristic pattern obtained after process of convolution, may be applicable to the adjustment of above-mentioned activation primitive, below will
Repeat no more.
Subsequently, the first full articulamentum is entered, and the neuron number of the first full articulamentum selects 4096, then the first full connection
Layer is output as the characteristic pattern of 4096 1px × 1px sizes.In actual treatment, it will usually first lead to above-mentioned 4096 characteristic patterns
The activation of ReLU function is crossed, then dropout process is carried out, dropout can be understood as model averagely, i.e., in forward direction in training process
During conduction, the activation value of certain neuron is allowed to quit work with certain Probability p, i.e., the activation value of the neuron is become with Probability p
For 0.Such as, the neuron of the first full articulamentum is 4096, if dropout ratio selects 0.4, then this layer of neuron warp
After crossing dropout, the value that wherein there are about 1638 neurons is set to 0, equivalent to the collaboration work by preventing some features
It is used for alleviating over-fitting, it is to avoid the appearance of a neuron depends on the phenomenon of another neuron.After dropout is processed
Characteristic pattern is input in the second full articulamentum, and the neuron number of the second full articulamentum is still 4096, then which is output as 4096
The characteristic pattern of 1px × 1px size is opened, and the activation of ReLU function equally can be first passed through to which, then carries out dropout process, and will place
Result after reason is input in the 3rd full articulamentum.Due to being that the age is identified, it is classification problem more than, and in this reality
Apply the integer that the age in example is between 0～100, therefore the neuron number of the 3rd full articulamentum is chosen as 101, then final
Three full articulamentum outputs are also 101, correspond to the probability at 0～100 this 101 ages respectively.Softmax selected by first grader
Grader, its are output as the maximum probability corresponding age, and the age should be the age A2 of facial image A1.Divide with regard to softmax
The content of class device, is ripe technological means, is not repeated herein.In order to train the first convolutional neural networks, according to input
The corresponding age A2 of facial image A1 be 7 years old this foreseen outcome, the output to the first grader is adjusted, by minimization
The method backpropagation of error is to adjust each parameter in the first convolutional neural networks.A large amount of in face image data set
Human face image information be trained after, obtain the first convolutional neural networks for training.
In step S220, according to default age threshold to each human face image information in face image data set
Processed, to add new age threshold attribute for each human face image information, age threshold attribute indicates corresponding people
Age be greater than default age threshold and be also no more than default age threshold.In the present embodiment, default age threshold
It is worth for 12, then in the face image data set age threshold attribute added by each human face image information, may indicate that corresponding people
Age be greater than 12 years old being also no more than 12 years old.
Subsequently, step S230 is entered, by the 3rd full connection in the first convolutional neural networks trained in step S210
Layer and the first grader replace with the 4th full articulamentum and the second grader respectively, to generate the second convolutional neural networks, and root
The second convolutional neural networks are trained according to the face image data set that with the addition of age threshold attribute, so as to the second convolution
Corresponding to the output indication face of the second grader of neutral net, the age of people is above age threshold and is also no greater than the age
Threshold value.In the present embodiment, the neuron of the 4th full articulamentum should be 2, and the result of its output is to be more than 12 years old and not at the age
Probability more than 12 years old, then correspond to the softmax grader of the second grader by corresponding for output probability higher value the range of age.
By taking human face image information A as an example, the age A2 in human face image information A is 7 years old, after age threshold attribute is added, it is known that year
Age A2 is not higher than age threshold, i.e., no more than 12 years old.Using facial image A1 as the first convolution group in the second convolutional neural networks
Input, the age be not more than the output as the second grader in the second convolutional neural networks in 12 years old, to the second convolution nerve net
Network carries out fine-tune training.Fig. 5 shows the structural representation of the second convolutional neural networks according to an embodiment of the invention
Figure.As shown in figure 5, the second convolutional neural networks and the first convolutional neural networks are on all four in hierarchical structure.
Next, in step S240, the second convolutional neural networks for training in obtaining step S230, trained
After the first full articulamentum in second convolutional neural networks, add the 5th full articulamentum, the 6th full articulamentum being sequentially connected
With the 3rd grader, to generate the 3rd convolutional neural networks, and select in face image data set, the age is not more than and is preset
The human face image information of age threshold the 3rd convolutional neural networks are trained, so as to the 3rd of the 3rd convolutional neural networks the
The age of people corresponding to the output indication face of grader by zero to preset age threshold in which.6th full articulamentum
Including the full articulamentum of multiple sons in parallel, the number of the full articulamentum of son adds 1 sum for default age threshold.Fig. 6 shows root
Structural representation according to the 3rd convolutional neural networks of one embodiment of the invention.As shown in fig. 6, the first full articulamentum it
Afterwards, with the addition of the 5th full articulamentum, the 6th full articulamentum and the 3rd grader being sequentially connected, respectively with the second full articulamentum,
4th full articulamentum and the second grader are in same level, so as to define respectively with the first grader as first point of output
Prop up and with the second grader as the second branch of output.In 5th full articulamentum, the number of neuron is 4096, and the 6th connects entirely
Connect layer to be made up of the full articulamentum of 13 sons, therefore this full articulamentum of 13 sons and the 4th full articulamentum are same level, select
Softmax grader is used as the 3rd grader.It should be noted that in the 6th full articulamentum the full articulamentum of 13 sons neuron
Number is 2, and the output result of each sub full articulamentum is numeral 1 and the corresponding probability of digital 0 difference, then as the 3rd classification
Size according to probable value is exported greater probability and is worth corresponding numeral by the softmax grader of device.
In the present embodiment, the human face image information the 3rd convolutional neural networks being trained, uses face figure
As in data acquisition system, the age be not more than the human face image information of 12 years old, that is, the age of the human face image information for using should be 0～12
Integer in year.In hands-on, the 3rd convolutional neural networks are trained using Ordinal loss as loss function.
With regard to Ordinal loss, i.e. order loss, typically processed by ordered logistic regress model, which is that one kind is directed to
The homing method of ordinal scale dependent variable data, can utilize predictive variable, and such as classifying type variable and numeric type variable, to orderly
The modeling returned by the response variable of classificatory scale type, related technology contents are here is omitted.
Finally, step S250, the first full connection in the 3rd convolutional neural networks trained by step S240 are entered
After layer, add the 7th full articulamentum, eight convergent points articulamentum and the 4th grader being sequentially connected, to generate Volume Four product nerve
Network, and Volume Four product neutral net is trained according to face image data set, refer to so as to the output of the 4th grader
Let others have a look at age of people corresponding to face.Fig. 7 shows that the structure of Volume Four according to an embodiment of the invention product neutral net is shown
It is intended to.As shown in fig. 7, after the first full articulamentum, with the addition of be sequentially connected the 7th full articulamentum, eight convergent points articulamentum and
4th grader, so as to define respectively with the second grader as export the first branch, with the 3rd grader as the of output
Two branches and with the 4th grader as export the 3rd branch.7th full articulamentum and the second full articulamentum, the 5th full articulamentum
For same level, neuron number is 4096, and eight convergent points articulamentum and the 4th full articulamentum, the 6th full articulamentum are same layer
Level, neuron number are 101, and the 4th grader and the second grader, the 3rd grader are same level, same selection
Softmax grader is used as the 4th grader.In hands-on, using Euclidean loss as loss function to Volume Four
Product neutral net is trained.With regard to Euclidean loss, i.e. Euclidean distance loss, its function expression are as follows：
Wherein, predicted valueLabel value y ∈ [- ∞ ,+∞], N are number of samples.With regard to Euclidean
The realization of loss has ripe technical method, is not repeated herein.
Fig. 3 shows the convolution for carrying out age identification to the face in image according to another embodiment of the present invention
The flow chart of neutral net generation method 300.As shown in figure 3, S310, S320, S330, S340 and S350 the step of method 300
Corresponded with S210, S220, S230, S240 and S250 the step of method in Fig. 2 200 respectively, be consistent, and in step
Step S360 is increased before S340, increased step S370 before step S350.
In step S360, in face image data set, the age be not more than default age threshold facial image
The age of information carries out 0/1 coding, and 0/1 coding includes to add 1 sum as encoding digit with default age threshold, and each is number
Any one in word 0 and numeral 1, from the beginning of first place, the difference that the number of times of 1 appearance of numeral subtracts 1 is the age.In the present embodiment
In, human face image information that the 3rd convolutional neural networks are trained, use in face image data set, the age not
Human face image information more than 12 years old, that is, the age of the human face image information for using should be the integer in 0～12 years old.With face figure
As the age A2 that as a example by information A, which includes is 7 years old, can be used to train the 3rd convolutional neural networks.First, age A2 is carried out
0/1 coding, the digit of 0/1 coding is 13, and from the beginning of first place, the difference that the number of times of 1 appearance of numeral subtracts 1 is 7, then illustrate to open from first place
Beginning sequentially to amount to has 8 continuous numerals 1, is therefore encoded to 1111111100000 within 7 years old corresponding 0/1.So as in step
In S340, age A2 is carried out the age 1111111100000 after 0/1 coding in obtaining step S360, by human face image information A
In facial image A1 as the input of the first convolution group, age in the 3rd convolutional neural networks generated in step S330
1111111100000 as in step S330 generate the 3rd convolutional neural networks in the 3rd grader output, with Ordinal
Loss is trained to the 3rd convolutional neural networks as loss function.
And in step S370, the age to human face image information in face image data set carries out distributed code, point
Cloth coding includes to carry out age coding according to Gaussian Profile.In the present embodiment, carry out distributed code to refer to one to the age
The numerical value at age, is indicated with a Gaussian Profile.And in face image data set human face image information age be 0～
Integer in 100, then to 0,1,2 ..., 99,100 this 101 integers all there is the general of a corresponding Gaussian distributed
Rate density function, the expression formula of the function are as follows：
X is the integer in 0～100
In formula, pxI it is x that () represents that age i obeys an average, and variance is δ2Gaussian Profile.
For example, human face image information B in face image data set, which is included in facial image B1 and correspondence image
Age information B2 of people, wherein age B2 are 28 years old.Now, the value of x is 28, and its corresponding probability density function is：
As can be seen from the above equation, the probability density function p obtained after distributed code is carried out to age B228For (i), work as i
When=28, p28I () obtains maximum p28(28).
As the output result of eight convergent points articulamentum is 0～100 years old this corresponding probability of 101 age numerical value, and conduct
Size according to probable value is exported the most probable value corresponding age, therefore in step by the softmax grader of the 4th grader
When being trained to Volume Four product neutral net in rapid S350, it is the Volume Four generated as step S340 by facial image B1
Result after the input of the first convolution group, age B2 carry out distributed code in step S370 in product neutral net is used as step
The output of eight convergent points articulamentum in the Volume Four product neutral net generated by S340, using Euclidean loss as loss function
Volume Four product neutral net is trained.
The flow chart that Fig. 8 shows age recognition methods 400 according to an embodiment of the invention.Age recognition methods
400 are suitable to execute in computing device (computing device 100 for example shown in Fig. 1), based on for carrying out to the face in image
In the convolutional neural networks generation method of age identification, the Volume Four that trains product neutral net carry out age identification.
As shown in figure 4, method 400 starts from step S410.In step S410, facial image to be identified is input to training
Age identification is carried out in good Volume Four product neutral net.In the present embodiment, are carried out to two facial images to be identified the age
Identification, the corresponding age of image one are 9 years old, and the corresponding age of image two is 43 years old, and image one and image two are input to training
Age identification is carried out in good Volume Four product neutral net.
Subsequently, step S420 is entered, judges in step S410 to carry out facial image to be identified after age identification, trains
Volume Four product neutral net in the output of the second grader whether be more than default age threshold.For image one, judge
The output of the second grader is not more than 12 years old, for image two, judges the output of the second grader more than 12 years old.
Next, in step S430, if the second grader is output as no more than default age threshold, obtaining step
3rd grader in the Volume Four product neutral net for facial image to be identified is carried out after age identification in rapid S410, training
It is output as the age of people corresponding to face.As corresponding second grader of image one is output as no more than default age threshold
Value, i.e., no more than 12 years old, then the age for obtaining the output of the 3rd grader is final result, is expressed as 1111111111000, from head
Position starts to have altogether 10 continuous numerals 1, thus identify that age be 9 years old, consistent with actual value.
And in step S440, if the second grader is output as more than default age threshold, obtaining step S410
In facial image to be identified is carried out after age identification, the 4th grader is output as in the Volume Four that trains product neutral net
The age of people corresponding to face.As corresponding second grader of image two is output as more than default age threshold, i.e., greatly
In 12 years old, then the age for obtaining the output of the 4th grader was final result, and 101 due to now eight convergent points articulamentum output are general
In rate value, the numerical value of the 44th is maximum, and therefore the age of the 4th grader output is 43 years old, consistent with actual value.
The flow chart that Fig. 8 shows the age recognition methods 500 according to another embodiment of the present invention.Age recognition methods
500 are suitable to execute in computing device (computing device 100 for example shown in Fig. 1), based on for carrying out to the face in image
In the convolutional neural networks generation method of age identification, the Volume Four that trains product neutral net carry out age identification.As Fig. 8 institute
Show, step S410, S420 the step of method 500 in S510, S520, S530 and S540 and method 400, S430 and S440 difference
Correspond, be consistent, and step S550 increased before step S510.
In step S550, images to be recognized is pre-processed to obtain facial image to be identified.This is because right
Before face in images to be recognized carries out age identification, need first to extract the facial image in images to be recognized.
First, Face datection is carried out to images to be recognized, obtains face location information；By face location information, by images to be recognized
In face cutting after change to pre-set dimension；Face is calculated according to face key point information carries out the conversion square of Plane Rotation
Battle array；The facial image under pre-set dimension is rotated into horizontal front to obtain facial image to be identified using transformation matrix.At this
In embodiment, to the advanced row Face datection of images to be recognized, i.e., first determine that scan image is carried out in a region, to each sector scanning
To position carry out feature extraction, then process of classifying is judging whether the position includes face.For there is human face region
Images to be recognized, will change to 224px × 224px size after face cutting.Due to above-mentioned face location typically refer to face and
Outline, then face rotational value need the line that two fixing points determine obtaining, select pupil as fixing point, by two
The line of pupil calculates an angle with facial image horizontal line, adopts affine transformation by the angle, obtains spin matrix,
To the image using after spin matrix, you can face is rotated to be interpupillary line and is in parallel relation with image level line, so as to
Obtain the facial image to be identified for keeping horizontal front.According to 2 points of line of pupil in face, this line and horizontal line is calculated
Angle to obtain angle A ngleValue of rotation, and can be entered using the get RotationMatrix2D function in OpenCV
The correlation computations of the row spin matrix, carry out face rotation using warpAffine, and concrete function is as follows：
RotateMatrix=cv2.getRotationMatrix2D (center=(Img.shape [1]/2,
Img.shape [0]/2), angle=AngleValue, scale=1)
RotImg=cv2.warpAffine (Img, RotateMatrix, (Img.shape [0] * 2, Img.shape [1] *
2))
Figure 10 shows the convolution for carrying out age identification to the face in image according to an embodiment of the invention
The schematic diagram of neutral net generating means 600.The device includes：First training module 610, attribute add module 620, first are given birth to
Become module 630, the second training module 640, the second generation module 650, the 3rd training module 660, the 3rd generation module 670 and
Four training modules 680.
First training mould 610 is suitable to, according to the face image data set for obtaining in advance, enter the first convolutional neural networks
Row training is applied to identification face so as to the first convolutional neural networks, and face image data set is comprising multiple facial image letters
Breath, each facial image packet include the age information of people in facial image and correspondence image, and the first convolutional neural networks include
Multiple convolution groups, the first full articulamentum, the second full articulamentum, the 3rd full articulamentum and the first grader being sequentially connected.Wherein,
The horizontal front of the facial image of each human face image information all holdings and pre-set dimension is met in face image data set, face
Image corresponds to the integer that the age of people is between 0～100, all includes at least one in each convolution group of the first convolutional neural networks
Convolutional layer.
Attribute add module 620 is suitable to according to default age threshold to each the face figure in face image data set
As information is processed, to add new age threshold attribute for each human face image information, it is right that age threshold attribute is indicated
The age of the people for answering is greater than default age threshold and is also no more than default age threshold.Wherein, default age threshold
For 12.
First generation module 630 is connected with the first training module 610, is suitable to train in the first training module 610
The 3rd full articulamentum and the first grader in first convolutional neural networks replaces with the 4th full articulamentum and the second classification respectively
Device, to generate the second convolutional neural networks.
Second training module 640 is connected with attribute add module 620 and the first generation module 630 respectively, is suitable to according to category
The face image data set of age threshold attribute is with the addition of in property add module 620, and the first generation module 630 is generated
Second convolutional neural networks are trained, so that corresponding to the output indication face of the second grader of the second convolutional neural networks
The age of people is above age threshold and is also no greater than age threshold.
Second generation module 650 is connected with the second training module 640, is suitable to obtain what the second training module 640 was trained
Second convolutional neural networks, after the first full articulamentum in the second convolutional neural networks for training, interpolation is sequentially connected
The 5th full articulamentum, the 6th full articulamentum and the 3rd grader, to generate the 3rd convolutional neural networks.Wherein, the 6th connects entirely
Connecing layer includes the full articulamentum of the son of multiple parallel connections, and the number of the full articulamentum of son adds 1 sum for default age threshold.
3rd training module 660 is connected with the second generation module 650, is suitably selected in face image data set, the age
Threeth convolutional neural networks of the human face image information of no more than default age threshold to generation in the second generation module 650
It is trained, so that the age of people corresponding to the output indication face of the 3rd grader of the 3rd convolutional neural networks is zero to institute
Which in default age threshold.
3rd generation module 670 is connected with the 3rd training module 660, is suitable to obtain what the 3rd training module 670 was trained
3rd convolutional neural networks, after the first full articulamentum in the 3rd convolutional neural networks for training, interpolation is sequentially connected
The 7th full articulamentum, eight convergent points articulamentum and the 4th grader, accumulate neutral net to generate Volume Four.
4th training module 680 is connected with the 3rd generation module 670, is suitable to according to face image data set to the three lives
The Volume Four product neutral net for becoming module 670 to be generated is trained, so that corresponding to the output indication face of the 4th grader
The age of people.
Figure 11 shows the schematic diagram of the age identifying device 700 according to another embodiment of this method.As shown in figure 11,
First training module 710 of device 700, attribute add module 720, the first generation module 730, the second training module 740, second
Generation module 750, the 3rd training module 760, the 3rd generation module 770 and the 4th training module 780, respectively with device in Figure 10
600 the first training module 610, attribute add module 620, the first generation module 630, the second training module 640, second are generated
Module 650, the 3rd training module 660, the 3rd generation module 670 and the 4th training module 680 are corresponded, be consistent, and
Increase newly with 760 connected first coding module 790 of the 3rd training module and encoded with the 4th training module 780 connected second
Module 792.
First coding module 791 is connected with the 3rd training module 760, be suitable to in face image data set, the age not
0/1 coding is carried out more than the age of the human face image information of default age threshold, 0/1 coding is included with default age threshold
Plus 1 sum for coding digit, each for numeral 0 and numeral 1 in any one, from the beginning of first place, numeral 1 occur number of times subtract
1 difference is the age.Further, the 3rd training module 760 be suitably selected in face image data set, the age no more than default
The human face image information of age threshold, the 3rd convolutional neural networks to generating in the second generation module 750 are trained, wherein
Age for training the human face image information of the 3rd convolutional neural networks carries out coded treatment by the first coding module 791.
Second coding module 792 is connected with the 4th training module 780, is suitable to facial image in face image data set
The age of information carries out distributed code, and distributed code includes to carry out age coding according to Gaussian Profile.Then the 4th training module 780
It is suitable to be trained according to the Volume Four product neutral net that face image data set generates the 3rd generation module 770, its
In for train Volume Four product neutral net human face image information coded treatment is carried out by the second coding module 792.
With regard to for carrying out concrete steps and the reality that the convolutional neural networks of age identification are generated to the face in image
Example is applied, detailed disclosure in the description based on Fig. 2-7, here is omitted.
Figure 12 shows the schematic diagram of age identifying device 800 according to an embodiment of the invention.Age identifying device
800 based on the Volume Four for carrying out to the face in image in the convolutional neural networks generating means of age identification, training
Product neutral net carries out age identification, and the device includes：Identification module 810, judge module 820 and acquisition module 830.Wherein,
Identification module 810 is connected with judge module 810, is suitable to facial image to be identified is input to the Volume Four product nerve net for training
Age identification is carried out in network.Judge module 820 be suitable to judge identification module 810 carry out age identification after the Volume Four for training
In product neutral net, whether the output of the second grader is more than default age threshold.Acquisition module 830 respectively with identification module
810 are connected with judge module 820, are suitable to be not more than the default age when the output that judge module 820 judges the second grader
During threshold value, the 3rd grader in the Volume Four product neutral net for training after identification module 810 carries out age identification is obtained
The age of people corresponding to face is output as, when the output of the second grader is more than default age threshold, obtains identification module
810 carry out the 4th grader in the Volume Four product neutral net for training after age identification is output as people corresponding to face
Age.
Figure 13 shows the schematic diagram of the age identifying device 900 according to another embodiment of the present invention.Age identification dress
900 are put based on the 4th for carrying out to the face in image in the convolutional neural networks generating means of age identification, training
Convolutional neural networks carry out age identification.As shown in figure 13, the identification module 910 of device 900, judge module 920 and acquisition mould
Block 930, identification module 810 respectively with device in Figure 12 800, judge module 820 and acquisition module 830 are corresponded, and are one
Cause, and increased pretreatment module 940 newly, be connected with identification module 910, be suitable to pre-process images to be recognized to obtain
The facial image to be identified obtained from pretreatment module 940 is input to training so as to identification module 910 by facial image to be identified
Age identification is carried out in good Volume Four product neutral net.Pretreatment module 940 is further adapted for entering pedestrian to images to be recognized
Face detection, obtains face location information；By the face location information, will turn after the face cutting in the images to be recognized
Shift to pre-set dimension；Face is calculated according to face key point information carries out the transformation matrix of Plane Rotation；Using the conversion square
Facial image under pre-set dimension is rotated into horizontal front to obtain facial image to be identified by battle array.
Concrete steps and embodiment with regard to age identification, detailed disclosure in the description based on Fig. 8-9, herein
Repeat no more.
In existing face age recognition methods, replace age true value using age distribution value, the result for getting is in
A certain age range, not accurate enough, even if taking VGG model to carry out age identification, it is also desirable to which the distribution according to age bracket is taken
Multiple models are realizing, complex, and versatility is relatively low.According to the present invention for carrying out age knowledge to the face in image
First convolutional neural networks are trained, the first convolution nerve net by technical scheme that other convolutional neural networks are generated first
Network includes multiple convolution groups, multiple full articulamentums and the first grader being sequentially connected, by the first convolutional Neural for training
The full articulamentum in part in network and the first grader are replaced accordingly, and generating the second convolutional neural networks is simultaneously carried out to which
Training, then add new full articulamentum and grader to the second convolutional neural networks for training, to generate the 3rd convolutional Neural
Network is simultaneously trained, and the 3rd convolutional neural networks for most training backward add new full articulamentum and grader, to generate
Volume Four product neutral net is simultaneously trained.In technique scheme, framework based on convolutional neural networks, and giving birth to
Become each convolutional neural networks in case before training, such as age threshold segmentation, 0/1 being carried out to the face image data for training and being compiled
Code and age distribution coding etc. are processed, and auxiliary progressively changes the model of convolutional neural networks, and ultimately forming one can accurately identify
The convolutional neural networks model at age, without the need for segmentation, succinct directly perceived, versatility is higher.Further, recognized according to the age of the present invention
Method, facial image to be identified is input in the Volume Four product neutral net for having trained, according to the output of different classifications device
To judge the range of age and its concrete numerical value, as a result accuracy has huge lifting.
A5. the method as any one of A1-4, the 6th full articulamentum include the full articulamentums of the son of multiple parallel connections,
The number of the full articulamentum of son adds 1 sum for default age threshold.A6. the method as any one of A1-5, described
After the first full articulamentum in the 3rd convolutional neural networks for training, add be sequentially connected the 7th full articulamentum, the 8th
Full articulamentum and the 4th grader, before generating the step of Volume Four accumulates neutral net, also to include step：To facial image number
Distributed code is carried out according to the age of human face image information in set, the distributed code includes age volume to be carried out according to Gaussian Profile
Code.A7. the method as any one of A1-6, the default age threshold are 12.B9. the method as described in B8, also wraps
Include and images to be recognized is pre-processed to obtain facial image to be identified.B10. the method as described in B9, described to be identified
Image is pre-processed to be included with obtaining facial image to be identified：Face datection is carried out to images to be recognized, obtains face location
Information；By the face location information, will change to pre-set dimension after the face cutting in the images to be recognized；According to people
Face key point information calculates face and carries out the transformation matrix of Plane Rotation；Using the transformation matrix by the face under pre-set dimension
Image rotation becomes horizontal front to obtain facial image to be identified.C14. the device as any one of C11-13, also includes
First coding module, is suitable to：To in the face image data set, the age be not more than default age threshold facial image
The age of information carries out 0/1 coding, and 0/1 coding includes to add 1 sum as encoding digit with default age threshold, each
For any one in numeral 0 and numeral 1, from the beginning of first place, the difference that the number of times of 1 appearance of numeral subtracts 1 is the age.C15. such as
Device any one of C11-14, the 6th full articulamentum include the full articulamentum of the son of multiple parallel connections, the full articulamentum of son
Number add 1 sum for default age threshold.C16. the device as any one of C11-15, also includes the second coding mould
Block, is suitable to：Age to human face image information in face image data set carries out distributed code, and the distributed code includes root
Age coding is carried out according to Gaussian Profile.C17. the device as any one of C11-16, the default age threshold are 12.
D19. the device as described in D18, also includes pretreatment module, is suitable to pre-process to obtain people to be identified images to be recognized
Face image.D20. the device as described in D19, the pretreatment module are further adapted for：Face datection is carried out to images to be recognized,
Obtain face location information；By the face location information, will change after the face cutting in the images to be recognized to pre-
If size；Face is calculated according to face key point information carries out the transformation matrix of Plane Rotation；Will be pre- using the transformation matrix
If the facial image under size is rotated into horizontal front to obtain facial image to be identified.
Claims (10)
1. a kind of convolutional neural networks generation method for carrying out age identification to the face in image, is suitable in computing device
Middle execution, methods described include step：
According to the advance face image data set for obtaining, the first convolutional neural networks are trained so as to first convolution
Neutral net is applied to identification face, and the face image data set includes multiple human face image information, each facial image
Information includes the age information of people in facial image and correspondence image, and it is many that first convolutional neural networks include to be sequentially connected
Individual convolution group, the first full articulamentum, the second full articulamentum, the 3rd full articulamentum and the first grader；
Each human face image information in the face image data set is processed according to default age threshold, so as to
Add new age threshold attribute for each human face image information, the age threshold attribute indicates that the age of corresponding people is big
Default age threshold is also no more than in default age threshold；
The 3rd full articulamentum in the first convolutional neural networks for training and the first grader are replaced with the 4th respectively connect entirely
Layer and the second grader is connect, so that the second convolutional neural networks are generated, and according to the face figure that with the addition of age threshold attribute
As data acquisition system is trained to the second convolutional neural networks, so that the second grader of second convolutional neural networks is defeated
Go out to indicate that the age of people corresponding to face is above the age threshold and is also no greater than the age threshold；
After the first full articulamentum in the second convolutional neural networks for training, add the 5th full connection being sequentially connected
Layer, the 6th full articulamentum and the 3rd grader, to generate the 3rd convolutional neural networks, and select the face image data set
In, the age be not more than default age threshold human face image information the 3rd convolutional neural networks are trained, with toilet
The age of people corresponding to the output indication face of the 3rd grader for stating the 3rd convolutional neural networks is by zero to default age threshold
Which in value；
After the first full articulamentum in the 3rd convolutional neural networks for training, add the 7th full connection being sequentially connected
Layer, eight convergent points articulamentum and the 4th grader, accumulate neutral net to generate Volume Four, and according to the face image data set
Volume Four product neutral net is trained, so that the year of people corresponding to the output indication face of the 4th grader
Age.
2. the method for claim 1, the facial image of each human face image information in the face image data set
All the horizontal front of holding and meet pre-set dimension, the facial image corresponds to the integer that the age of people is between 0～100.
3. method as claimed in claim 1 or 2, all includes at least one in each convolution group of first convolutional neural networks
Convolutional layer.
4. the method as any one of claim 1-3, described in the second convolutional neural networks for training
After one full articulamentum, add the 5th full articulamentum, the 6th full articulamentum and the 3rd grader being sequentially connected, to generate the 3rd
Before the step of convolutional neural networks, also include step：
To in the face image data set, the age is not more than age of human face image information of default age threshold carries out
0/1 coding, 0/1 coding include to add 1 sum as encoding digit with default age threshold, and each is numeral 0 and numeral 1
In any one, from the beginning of first place, numeral 1 occur number of times subtract 1 difference be the age.
5. a kind of age recognition methods, is suitable to execute in computing device, and methods described is based on any one of claim 1-4 institute
The Volume Four product neutral net for training that states carries out age identification to the face in image, including step：
Facial image to be identified is input in the Volume Four product neutral net for training carries out age identification；
In the Volume Four product neutral net trained described in judging, whether the output of the second grader is more than default age threshold；
If second grader is output as no more than default age threshold, the Volume Four product trained described in acquisition is refreshing
The age of people corresponding to face is output as through the 3rd grader in network；
If second grader is output as more than default age threshold, the Volume Four product for training is obtained neural
In network, the 4th grader is output as the age of people corresponding to face.
6. a kind of convolutional neural networks generating means for carrying out age identification to the face in image, are suitable to reside in calculating
In equipment, described device includes：
First training module, is suitable to, according to the face image data set for obtaining in advance, instruct the first convolutional neural networks
Practice and be applied to identification face so as to first convolutional neural networks, the face image data set includes multiple facial images
Information, each facial image packet include the age information of people in facial image and correspondence image, the first convolution nerve net
Network includes multiple convolution groups, the first full articulamentum, the second full articulamentum, the 3rd full articulamentum and the first classification being sequentially connected
Device；
Attribute add module, is suitable to according to default age threshold to each facial image in the face image data set
Information is processed, and to add new age threshold attribute for each human face image information, the age threshold attribute is indicated
The age of corresponding people is greater than default age threshold and is also no more than default age threshold；
First generation module, the 3rd full articulamentum being suitable in the first convolutional neural networks that will be trained and the first grader divide
Fourth full articulamentum and second grader are not replaced with, to generate the second convolutional neural networks；
Second training module, is suitable to according to the face image data set that with the addition of age threshold attribute to the second convolution god
It is trained through network, so that the year of people corresponding to the output indication face of the second grader of second convolutional neural networks
Age is above the age threshold and is also no greater than the age threshold；
Second generation module, is suitable to after the first full articulamentum in the second convolutional neural networks for training, and adds successively
Connected 5th full articulamentum, the 6th full articulamentum and the 3rd grader, to generate the 3rd convolutional neural networks；
3rd training module, is suitably selected in the face image data set, the age is not more than default age threshold
Human face image information is trained to the 3rd convolutional neural networks, so that the 3rd grader of the 3rd convolutional neural networks
The age of people corresponding to output indication face by zero to preset age threshold in which；
3rd generation module, is suitable to after the first full articulamentum in the 3rd convolutional neural networks for training, and adds successively
Connected 7th full articulamentum, eight convergent points articulamentum and the 4th grader, accumulate neutral net to generate Volume Four；
4th training module, is suitable to be trained Volume Four product neutral net according to the face image data set,
So that the age of people corresponding to the output indication face of the 4th grader.
7. device as claimed in claim 6, the facial image of each human face image information in the face image data set
All the horizontal front of holding and meet pre-set dimension, the facial image corresponds to the integer that the age of people is between 0～100.
8. device as claimed in claims 6 or 7, all includes at least one in each convolution group of first convolutional neural networks
Convolutional layer.
9. a kind of age identifying device, is suitable to reside in computing device, and described device is based on any one of claim 6-8 institute
The Volume Four product neutral net for training that states carries out age identification to the face in image, including：
Identification module, being suitable to be input to facial image to be identified in the Volume Four product neutral net for training carries out age knowledge
Not；
Judge module, be suitable to judge the identification module carry out age identification after the Volume Four product neutral net for training in the
Whether the output of two graders is more than default age threshold；
Acquisition module, is suitable to when the output that the judge module judges the second grader is not more than default age threshold,
Obtain the 3rd grader in the Volume Four product neutral net for training after the identification module carries out age identification to be output as
The age of people corresponding to face, when the judge module judges the output of the second grader more than default age threshold,
Obtain the 4th grader in the Volume Four product neutral net for training after the identification module carries out age identification to be output as
The age of people corresponding to face.
10. a kind of computing device, including：
The convolutional neural networks for carrying out age identification to the face in image as any one of claim 6-8 are given birth to
Become device；And
Age identifying device as claimed in claim 9.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201610925676.0A CN106485235B (en) | 2016-10-24 | 2016-10-24 | A kind of convolutional neural networks generation method, age recognition methods and relevant apparatus |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201610925676.0A CN106485235B (en) | 2016-10-24 | 2016-10-24 | A kind of convolutional neural networks generation method, age recognition methods and relevant apparatus |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN106485235A true CN106485235A (en) | 2017-03-08 |
| CN106485235B CN106485235B (en) | 2019-05-03 |
Family
ID=58270488
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201610925676.0A Active CN106485235B (en) | 2016-10-24 | 2016-10-24 | A kind of convolutional neural networks generation method, age recognition methods and relevant apparatus |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN106485235B (en) |
Cited By (25)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN106874964A (en) * | 2017-03-30 | 2017-06-20 | 李文谦 | A kind of foot type image size automatic prediction method and prediction meanss based on modified convolutional neural networks |
| CN106980830A (en) * | 2017-03-17 | 2017-07-25 | 中国人民解放军国防科学技术大学 | One kind is based on depth convolutional network from affiliation recognition methods and device |
| CN107122798A (en) * | 2017-04-17 | 2017-09-01 | 深圳市淘米科技有限公司 | Chin-up count detection method and device based on depth convolutional network |
| CN107194407A (en) * | 2017-05-18 | 2017-09-22 | 网易（杭州）网络有限公司 | A kind of method and apparatus of image understanding |
| CN107527024A (en) * | 2017-08-08 | 2017-12-29 | 北京小米移动软件有限公司 | Face face value appraisal procedure and device |
| CN107622261A (en) * | 2017-11-03 | 2018-01-23 | 北方工业大学 | Face age estimation method and device based on deep learning |
| CN108021863A (en) * | 2017-11-01 | 2018-05-11 | 平安科技（深圳）有限公司 | Electronic device, the character classification by age method based on image and storage medium |
| CN108052861A (en) * | 2017-11-08 | 2018-05-18 | 北京卓视智通科技有限责任公司 | A kind of nerve network system and the model recognizing method based on the nerve network system |
| CN108073910A (en) * | 2017-12-29 | 2018-05-25 | 百度在线网络技术（北京）有限公司 | For generating the method and apparatus of face characteristic |
| CN108197602A (en) * | 2018-01-30 | 2018-06-22 | 厦门美图之家科技有限公司 | A kind of convolutional neural networks generation method and expression recognition method |
| CN108596042A (en) * | 2018-03-29 | 2018-09-28 | 青岛海尔智能技术研发有限公司 | Enabling control method and system |
| CN109002884A (en) * | 2018-07-20 | 2018-12-14 | 郑州云海信息技术有限公司 | A kind of pond processing unit and pond processing method |
| CN109101938A (en) * | 2018-08-23 | 2018-12-28 | 武汉嫦娥医学抗衰机器人股份有限公司 | A kind of multi-tag age estimation method based on convolutional neural networks |
| CN109308481A (en) * | 2017-07-27 | 2019-02-05 | 丰田自动车株式会社 | Identification device |
| CN109522872A (en) * | 2018-12-04 | 2019-03-26 | 西安电子科技大学 | A kind of face identification method, device, computer equipment and storage medium |
| CN109711546A (en) * | 2018-12-21 | 2019-05-03 | 深圳市商汤科技有限公司 | Neural network training method and device, electronic equipment and storage medium |
| CN110185939A (en) * | 2019-05-16 | 2019-08-30 | 西北工业大学 | Gas pipe leakage recognition methods based on convolutional neural networks |
| CN110399813A (en) * | 2019-07-10 | 2019-11-01 | 深兰科技（上海）有限公司 | A kind of age recognition methods, device, electronic equipment and storage medium |
| CN110971901A (en) * | 2018-09-29 | 2020-04-07 | 杭州海康威视数字技术股份有限公司 | Convolutional neural network processing method and device |
| CN111096736A (en) * | 2018-10-26 | 2020-05-05 | 深圳市理邦精密仪器股份有限公司 | Electrocardiogram classification method, device and system based on active learning |
| CN111159450A (en) * | 2019-12-30 | 2020-05-15 | 上海眼控科技股份有限公司 | Picture classification method and device, computer equipment and storage medium |
| CN111325256A (en) * | 2020-02-13 | 2020-06-23 | 上海眼控科技股份有限公司 | Vehicle appearance detection method and device, computer equipment and storage medium |
| CN112052710A (en) * | 2019-06-06 | 2020-12-08 | 搜狗（杭州）智能科技有限公司 | Face age identification method and device |
| WO2021012383A1 (en) * | 2019-07-24 | 2021-01-28 | 初速度（苏州）科技有限公司 | Age prediction method and device for infrared image |
| CN112927801A (en) * | 2021-01-24 | 2021-06-08 | 武汉东湖大数据交易中心股份有限公司 | Health prediction model construction method and device based on dressing meteorological index |
Citations (8)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN103824054A (en) * | 2014-02-17 | 2014-05-28 | 北京旷视科技有限公司 | Cascaded depth neural network-based face attribute recognition method |
| CN104504376A (en) * | 2014-12-22 | 2015-04-08 | 厦门美图之家科技有限公司 | Age classification method and system for face images |
| CN104866829A (en) * | 2015-05-25 | 2015-08-26 | 苏州大学 | Cross-age face verify method based on characteristic learning |
| CN105095833A (en) * | 2014-05-08 | 2015-11-25 | 中国科学院声学研究所 | Network constructing method for human face identification, identification method and system |
| CN105279499A (en) * | 2015-10-30 | 2016-01-27 | 小米科技有限责任公司 | Age recognition method and device |
| CN105426963A (en) * | 2015-12-01 | 2016-03-23 | 北京天诚盛业科技有限公司 | Convolutional neural network Training method and apparatus for human face identification and application |
| CN105825191A (en) * | 2016-03-23 | 2016-08-03 | 厦门美图之家科技有限公司 | Face multi-attribute information-based gender recognition method and system and shooting terminal |
| CN105975916A (en) * | 2016-04-28 | 2016-09-28 | 西安电子科技大学 | Age estimation method based on multi-output convolution neural network and ordered regression |
-
2016
- 2016-10-24 CN CN201610925676.0A patent/CN106485235B/en active Active
Patent Citations (8)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN103824054A (en) * | 2014-02-17 | 2014-05-28 | 北京旷视科技有限公司 | Cascaded depth neural network-based face attribute recognition method |
| CN105095833A (en) * | 2014-05-08 | 2015-11-25 | 中国科学院声学研究所 | Network constructing method for human face identification, identification method and system |
| CN104504376A (en) * | 2014-12-22 | 2015-04-08 | 厦门美图之家科技有限公司 | Age classification method and system for face images |
| CN104866829A (en) * | 2015-05-25 | 2015-08-26 | 苏州大学 | Cross-age face verify method based on characteristic learning |
| CN105279499A (en) * | 2015-10-30 | 2016-01-27 | 小米科技有限责任公司 | Age recognition method and device |
| CN105426963A (en) * | 2015-12-01 | 2016-03-23 | 北京天诚盛业科技有限公司 | Convolutional neural network Training method and apparatus for human face identification and application |
| CN105825191A (en) * | 2016-03-23 | 2016-08-03 | 厦门美图之家科技有限公司 | Face multi-attribute information-based gender recognition method and system and shooting terminal |
| CN105975916A (en) * | 2016-04-28 | 2016-09-28 | 西安电子科技大学 | Age estimation method based on multi-output convolution neural network and ordered regression |
Non-Patent Citations (3)
| Title |
|---|
| GIL LEVI等: "《Age and Gender Classification using Convolutional Neural Networks》", 《2015 IEEE CONFERENCE ON COMPUTER VISION AND PATTERN RECOGNITION WORKSHOPS (CVPRW)》 * |
| 周志华等: "《神经网络集成》", 《计算机学报》 * |
| 方晟: "《基于人脸特征的性别判断和年龄估计方法研究》", 《中国优秀硕士论文全文数据库》 * |
Cited By (32)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN106980830A (en) * | 2017-03-17 | 2017-07-25 | 中国人民解放军国防科学技术大学 | One kind is based on depth convolutional network from affiliation recognition methods and device |
| CN106874964A (en) * | 2017-03-30 | 2017-06-20 | 李文谦 | A kind of foot type image size automatic prediction method and prediction meanss based on modified convolutional neural networks |
| CN106874964B (en) * | 2017-03-30 | 2023-11-03 | 李文谦 | Foot-type image size automatic prediction method and prediction device based on improved convolutional neural network |
| CN107122798A (en) * | 2017-04-17 | 2017-09-01 | 深圳市淘米科技有限公司 | Chin-up count detection method and device based on depth convolutional network |
| CN107194407A (en) * | 2017-05-18 | 2017-09-22 | 网易（杭州）网络有限公司 | A kind of method and apparatus of image understanding |
| CN109308481A (en) * | 2017-07-27 | 2019-02-05 | 丰田自动车株式会社 | Identification device |
| CN109308481B (en) * | 2017-07-27 | 2022-08-30 | 丰田自动车株式会社 | Identification device |
| CN107527024A (en) * | 2017-08-08 | 2017-12-29 | 北京小米移动软件有限公司 | Face face value appraisal procedure and device |
| CN108021863A (en) * | 2017-11-01 | 2018-05-11 | 平安科技（深圳）有限公司 | Electronic device, the character classification by age method based on image and storage medium |
| WO2019085338A1 (en) * | 2017-11-01 | 2019-05-09 | 平安科技（深圳）有限公司 | Electronic apparatus, image-based age classification method and system, and storage medium |
| CN107622261A (en) * | 2017-11-03 | 2018-01-23 | 北方工业大学 | Face age estimation method and device based on deep learning |
| CN108052861A (en) * | 2017-11-08 | 2018-05-18 | 北京卓视智通科技有限责任公司 | A kind of nerve network system and the model recognizing method based on the nerve network system |
| CN108073910A (en) * | 2017-12-29 | 2018-05-25 | 百度在线网络技术（北京）有限公司 | For generating the method and apparatus of face characteristic |
| CN108197602A (en) * | 2018-01-30 | 2018-06-22 | 厦门美图之家科技有限公司 | A kind of convolutional neural networks generation method and expression recognition method |
| CN108596042A (en) * | 2018-03-29 | 2018-09-28 | 青岛海尔智能技术研发有限公司 | Enabling control method and system |
| CN109002884A (en) * | 2018-07-20 | 2018-12-14 | 郑州云海信息技术有限公司 | A kind of pond processing unit and pond processing method |
| CN109101938B (en) * | 2018-08-23 | 2021-01-15 | 武汉嫦娥医学抗衰机器人股份有限公司 | Multi-label age estimation method based on convolutional neural network |
| CN109101938A (en) * | 2018-08-23 | 2018-12-28 | 武汉嫦娥医学抗衰机器人股份有限公司 | A kind of multi-tag age estimation method based on convolutional neural networks |
| CN110971901B (en) * | 2018-09-29 | 2022-08-02 | 杭州海康威视数字技术股份有限公司 | Processing method, device and equipment of convolutional neural network and storage medium |
| CN110971901A (en) * | 2018-09-29 | 2020-04-07 | 杭州海康威视数字技术股份有限公司 | Convolutional neural network processing method and device |
| CN111096736A (en) * | 2018-10-26 | 2020-05-05 | 深圳市理邦精密仪器股份有限公司 | Electrocardiogram classification method, device and system based on active learning |
| CN109522872A (en) * | 2018-12-04 | 2019-03-26 | 西安电子科技大学 | A kind of face identification method, device, computer equipment and storage medium |
| CN109711546A (en) * | 2018-12-21 | 2019-05-03 | 深圳市商汤科技有限公司 | Neural network training method and device, electronic equipment and storage medium |
| CN110185939A (en) * | 2019-05-16 | 2019-08-30 | 西北工业大学 | Gas pipe leakage recognition methods based on convolutional neural networks |
| CN112052710A (en) * | 2019-06-06 | 2020-12-08 | 搜狗（杭州）智能科技有限公司 | Face age identification method and device |
| CN110399813A (en) * | 2019-07-10 | 2019-11-01 | 深兰科技（上海）有限公司 | A kind of age recognition methods, device, electronic equipment and storage medium |
| WO2021012383A1 (en) * | 2019-07-24 | 2021-01-28 | 初速度（苏州）科技有限公司 | Age prediction method and device for infrared image |
| CN111159450A (en) * | 2019-12-30 | 2020-05-15 | 上海眼控科技股份有限公司 | Picture classification method and device, computer equipment and storage medium |
| CN111159450B (en) * | 2019-12-30 | 2024-11-01 | 上海眼控科技股份有限公司 | Picture classification method, device, computer equipment and storage medium |
| CN111325256A (en) * | 2020-02-13 | 2020-06-23 | 上海眼控科技股份有限公司 | Vehicle appearance detection method and device, computer equipment and storage medium |
| CN112927801A (en) * | 2021-01-24 | 2021-06-08 | 武汉东湖大数据交易中心股份有限公司 | Health prediction model construction method and device based on dressing meteorological index |
| CN112927801B (en) * | 2021-01-24 | 2024-05-14 | 武汉东湖大数据科技股份有限公司 | Construction method and device of health prediction model based on dressing meteorological index |
Also Published As
| Publication number | Publication date |
|---|---|
| CN106485235B (en) | 2019-05-03 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN106485235A (en) | A kind of convolutional neural networks generation method, age recognition methods and relevant apparatus | |
| CN107767408B (en) | Image processing method, processing device and processing equipment | |
| CN104217214B (en) | RGB D personage's Activity recognition methods based on configurable convolutional neural networks | |
| Wang et al. | Large-scale continuous gesture recognition using convolutional neural networks | |
| CN108596039A (en) | A kind of bimodal emotion recognition method and system based on 3D convolutional neural networks | |
| CN110096950A (en) | A kind of multiple features fusion Activity recognition method based on key frame | |
| CN108647591A (en) | Activity recognition method and system in a kind of video of view-based access control model-semantic feature | |
| CN107358626A (en) | A kind of method that confrontation network calculations parallax is generated using condition | |
| Gupta et al. | ALMNet: Adjacent layer driven multiscale features for salient object detection | |
| CN109558832A (en) | A kind of human body attitude detection method, device, equipment and storage medium | |
| CN107437096A (en) | Image classification method based on the efficient depth residual error network model of parameter | |
| Li et al. | LPSNet: a novel log path signature feature based hand gesture recognition framework | |
| CN108416327A (en) | A kind of object detection method, device, computer equipment and readable storage medium storing program for executing | |
| CN110399788A (en) | AU detection method, device, electronic equipment and the storage medium of image | |
| CN113128424B (en) | Method for identifying action of graph convolution neural network based on attention mechanism | |
| CN104361316B (en) | Dimension emotion recognition method based on multi-scale time sequence modeling | |
| CN110414498A (en) | A kind of natural scene text recognition method based on intersection attention mechanism | |
| CN111986180B (en) | Face forged video detection method based on multi-correlation frame attention mechanism | |
| CN106557173A (en) | Dynamic gesture identification method and device | |
| CN108304786A (en) | A kind of pedestrian detection method based on binaryzation convolutional neural networks | |
| CN117876824B (en) | Multi-modal crowd counting model training method, system, storage medium and equipment | |
| CN109034012A (en) | First person gesture identification method based on dynamic image and video sequence | |
| CN110008961A (en) | Text real-time identification method, device, computer equipment and storage medium | |
| Guo et al. | A deep reinforcement learning method for multimodal data fusion in action recognition | |
| Sahoo et al. | DeReFNet: Dual-stream Dense Residual Fusion Network for static hand gesture recognition |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| C06 | 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 |