CN109670437A - Age prediction model training method, face-image recognition methods and device - Google Patents
Age prediction model training method, face-image recognition methods and device Download PDFInfo
- Publication number
- CN109670437A CN109670437A CN201811532337.1A CN201811532337A CN109670437A CN 109670437 A CN109670437 A CN 109670437A CN 201811532337 A CN201811532337 A CN 201811532337A CN 109670437 A CN109670437 A CN 109670437A
- Authority
- CN
- China
- Prior art keywords
- image
- age
- training
- face
- current generation
- 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
- 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)
- Health & Medical Sciences (AREA)
- General Health & Medical Sciences (AREA)
- Oral & Maxillofacial Surgery (AREA)
- Human Computer Interaction (AREA)
- Physics & Mathematics (AREA)
- General Physics & Mathematics (AREA)
- Multimedia (AREA)
- Theoretical Computer Science (AREA)
- Image Analysis (AREA)
Abstract
This application involves a kind of age prediction model training method, face-image recognition methods and device, the age prediction model training method includes: to obtain current generation corresponding facial sample graph image set;For each facial sample image that facial sample image is concentrated, adjustment image quality parameter obtains parameter adjustment face-image, and chooses age mark value from mark range of corresponding age;The age mark value that face-image and selection are adjusted according to parameter treats training pattern and carries out the training of current generation, until meeting the training suspension condition of current generation;The training suspension condition of next stage is determined according to the training suspension condition of current generation;Using next stage as the current generation, returns to acquisition current generation corresponding facial sample graph image set and continue to train, until the training suspension condition of next stage meets training completion condition, obtain age prediction model.The age that scheme provided by the present application improves age prediction model estimates accuracy.
Description
Technical field
This application involves technical field of image processing, scheme more particularly to a kind of age prediction model training method, face
As recognition methods, device, computer readable storage medium and computer equipment.
Background technique
With the development of data processing technique, image processing techniques is also developed rapidly, so that image processing techniques
Application field it is quite extensive, such as the application field that traffic monitoring and target identification etc. are more can lead in target identification
The target object in image can be identified by crossing image procossing.
However, can determine that face-image is corresponding by the identification to face-image in current image processing techniques
Age, traditional to carry out age identification method by face-image, recognition result is highly susceptible to environmental factor when Image Acquisition
Influence, so as to cause by face-image carry out age identification accuracy it is lower.
Summary of the invention
Based on this, it is necessary to for the lower technical problem of the accuracy for carrying out age identification by face-image, provide
A kind of age prediction model training method, face-image recognition methods, device, computer readable storage medium and computer are set
It is standby.
A kind of age prediction model training method, comprising:
Obtain current generation corresponding facial sample graph image set;
For each facial sample image that the facial sample image is concentrated, adjustment image quality parameter obtains parameter tune
Whole face-image, and age mark value is chosen from mark range of corresponding age;
The age mark value that face-image and selection are adjusted according to the parameter treats training pattern and carries out the current generation
Training, until meeting the training suspension condition of current generation;
The training suspension condition of next stage is determined according to the training suspension condition of the current generation;
Using next stage as the current generation, returns to acquisitions current generation corresponding face sample graph image set and continue pair
It is described to be trained to training pattern, until the training suspension condition of next stage meets training completion condition, it is pre- to obtain the age
Estimate model.
A kind of age prediction model training device, described device include:
Sample image obtains module, for obtaining current generation corresponding facial sample graph image set;
Mass parameter adjusts module, each facial sample image for concentrating for the facial sample image, adjustment
Image quality parameter obtains parameter adjustment face-image, and chooses age mark value from mark range of corresponding age;
Model training module treats training for adjusting the age mark value of face-image and selection according to the parameter
Model carries out the training of current generation, until meeting the training suspension condition of current generation;
Stop condition determining module determines the training of next stage for the training suspension condition according to the current generation
Suspension condition;
Training loop module, for using next stage as the current generation, notifying the sample image obtains module to obtain
Current generation corresponding facial sample graph image set, to continue to be trained to described to training pattern, until the instruction of next stage
Practice suspension condition and meet training completion condition, obtains age prediction model.
A kind of computer equipment, including memory and processor are stored with computer program, the meter in the memory
When calculation machine program is executed by processor, so that the processor executes following steps:
Obtain current generation corresponding facial sample graph image set;
For each facial sample image that the facial sample image is concentrated, adjustment image quality parameter obtains parameter tune
Whole face-image, and age mark value is chosen from mark range of corresponding age;
The age mark value that face-image and selection are adjusted according to the parameter treats training pattern and carries out the current generation
Training, until meeting the training suspension condition of current generation;
The training suspension condition of next stage is determined according to the training suspension condition of the current generation;
Using next stage as the current generation, returns to acquisitions current generation corresponding face sample graph image set and continue pair
It is described to be trained to training pattern, until the training suspension condition of next stage meets training completion condition, it is pre- to obtain the age
Estimate model.
A kind of storage medium being stored with computer program, when the computer program is executed by processor, so that processing
Device executes following steps:
Obtain current generation corresponding facial sample graph image set;
For each facial sample image that the facial sample image is concentrated, adjustment image quality parameter obtains parameter tune
Whole face-image, and age mark value is chosen from mark range of corresponding age;
The age mark value that face-image and selection are adjusted according to the parameter treats training pattern and carries out the current generation
Training, until meeting the training suspension condition of current generation;
The training suspension condition of next stage is determined according to the training suspension condition of the current generation;
Using next stage as the current generation, returns to acquisitions current generation corresponding face sample graph image set and continue pair
It is described to be trained to training pattern, until the training suspension condition of next stage meets training completion condition, it is pre- to obtain the age
Estimate model.
Above-mentioned age prediction model training method, device, computer readable storage medium and computer equipment, to current rank
The corresponding facial sample image of section concentrates each facial sample image, and adjustment image quality parameter obtains parameter adjustment face figure
Picture, and age mark value is chosen from mark range of corresponding age, it is marked according to the age that parameter adjusts face-image and selection
The training of note value progress current generation.By adjusting image quality parameter during training, so that the model that training obtains
Fully consider influence of the environmental factor to picture quality, thus consider influence of the picture quality to model recognition result,
After training by multiple stages, the age for improving the age prediction model that training obtains estimates accuracy.
A kind of age prediction model training method, comprising:
Obtain face-image;
By the face-image age prediction model, each default age bracket difference of the age prediction model output is obtained
Corresponding probability value;
According to the corresponding probability value of each default age bracket, age desired value is determined;
Age discreet value with the corresponding age discreet value of the age desired value, as the face-image.
A kind of face-image identification device, described device include:
Face-image obtains module, for obtaining face-image;
Face-image identification module, for obtaining the age prediction model for the face-image age prediction model
The corresponding probability value of each default age bracket of output;
Age it is expected determining module, for determining age desired value according to the corresponding probability value of each default age bracket;
Age estimates module, is used for the corresponding age discreet value of the age desired value, as the face-image
Age discreet value.
A kind of computer equipment, including memory and processor are stored with computer program, the meter in the memory
When calculation machine program is executed by processor, so that the processor executes following steps:
Obtain face-image;
By the face-image age prediction model, each default age bracket difference of the age prediction model output is obtained
Corresponding probability value;
According to the corresponding probability value of each default age bracket, age desired value is determined;
Age discreet value with the corresponding age discreet value of the age desired value, as the face-image.
A kind of storage medium being stored with computer program, when the computer program is executed by processor, so that processing
Device executes following steps:
Obtain face-image;
By the face-image age prediction model, each default age bracket difference of the age prediction model output is obtained
Corresponding probability value;
According to the corresponding probability value of each default age bracket, age desired value is determined;
Age discreet value with the corresponding age discreet value of the age desired value, as the face-image.
Above-mentioned face-image recognition methods, device, computer readable storage medium and computer equipment, according in view of ring
The age prediction model of influence of the border factor to collected face-image carries out age identification to the face-image got,
Improve the accuracy of the corresponding probability value of each default age bracket.Further according to the corresponding probability value of each default age bracket
Age desired value is calculated, to be worth to obtain age discreet value according to age expectation, to further improve age discreet value
Accuracy.
Detailed description of the invention
Fig. 1 is the applied environment figure of age prediction model training method in one embodiment;
Fig. 2 is the flow diagram of age prediction model training method in one embodiment;
Fig. 3 is flow diagram the step of obtaining facial sample graph image set in one embodiment;
Fig. 4 is flow diagram the step of adjusting image quality parameter in one embodiment;
Fig. 5 is the schematic diagram marked at the age of one embodiment septum reset sample image;
Fig. 6 is the flow diagram of age prediction model training method in another embodiment;
Fig. 7 is the schematic diagram that image quality parameter is adjusted in one embodiment;
Fig. 8 is the flow diagram of one embodiment septum reset image-recognizing method;
Fig. 9 is to identify schematic diagram to the age of face-image in one embodiment;
Figure 10 is flow diagram the step of obtaining face-image in one embodiment;
Figure 11 is that customer's details show schematic diagram in one embodiment;
Figure 12 is the schematic diagram that customer's acess control result is shown in one embodiment;
Figure 13 is the structural block diagram of age prediction model training device in one embodiment;
Figure 14 is the structural block diagram of one embodiment septum reset pattern recognition device;
Figure 15 is the structural block diagram of computer equipment in one embodiment;
Figure 16 is the structural block diagram of computer equipment in another embodiment.
Specific embodiment
It is with reference to the accompanying drawings and embodiments, right in order to which the objects, technical solutions and advantages of the application are more clearly understood
The application is further elaborated.It should be appreciated that specific embodiment described herein is only used to explain the application, and
It is not used in restriction the application.
Fig. 1 is the applied environment figure of age prediction model training method in one embodiment.Referring to Fig.1, which estimates
Model training method is applied to age prediction model training system.The age prediction model training system includes 110 kimonos of terminal
Business device 120.Terminal 110 and server 120 pass through network connection.Terminal 110 specifically can be terminal console or mobile terminal, move
Dynamic terminal specifically can be at least one of mobile phone, tablet computer, laptop etc..Server 120 can use independent clothes
The server cluster of business device either multiple servers composition is realized.
As shown in Fig. 2, in one embodiment, providing a kind of age prediction model training method.The present embodiment is main
The server 120 being applied in above-mentioned Fig. 1 in this way comes for example, the terminal 110 in Fig. 1 can also be applied.Reference
Fig. 2, the age prediction model training method specifically comprise the following steps:
S202 obtains current generation corresponding facial sample graph image set.
Wherein, the database of storage sample image training set is provided in server, sample image training set includes each year
The corresponding facial sample image of age section.Current generation, corresponding facial sample graph image set was current model training stage institute
The set of the facial sample image needed.Facial sample image is to carry out the figure that Image Acquisition obtains to the face of target object
Picture, facial sample image can be face sample image.
Specifically, server receives the model training instruction of terminal transmission, according to model training instruction to storage sample
The database of training set of images accesses, and current generation corresponding facial sample image is extracted from database, is obtained current
Stage corresponding facial sample graph image set.
In one embodiment, server obtains current generation corresponding pre-set image quantity, according to pre-set image quantity
Facial sample image is screened from storage sample image training set, using the facial sample image that screens as current generation correspondence
Facial sample graph image set.
S204, for each facial sample image that facial sample image is concentrated, adjustment image quality parameter obtains parameter
Face-image is adjusted, and chooses age mark value from mark range of corresponding age.
Wherein, image quality parameter indicates the parameter of the picture quality of facial sample image, and image quality parameter specifically may be used
It can also include pixel quantity, resolution ratio and transparency etc. to include image blur and brightness of image.Facial sample graph image set
In each facial sample image it is all corresponding be labelled with age mark range, such as age mark range can be 2-4 years old, 2-3
Age mark value in the age mark range in year is 2 years old, 3 years old and 4 years old.
Specifically, server reads each facial sample image that facial sample image is concentrated, and obtains the face read
The image quality parameter of sample image is tuned up or is turned down to each parameter in the image quality parameter got, obtained
Parameter adjusts face-image.Server extracts age mark range corresponding with the facial sample image read, determines and extracts
The age mark value in age mark range arrived, chooses age mark value from determining age mark value.Server extracts
To age mark range it is corresponding with the facial sample image read, parameter adjusts face-image and the facial sample that reads
Image is corresponding, and the corresponding age mark range of same face sample image and parameter adjustment face-image are corresponding.
S206 adjusts the age mark value of face-image and selection according to parameter, treats training pattern and carries out the current generation
Training, until meeting the training suspension condition of current generation.
Wherein, the training suspension condition of current generation for the suspension current generation and enters the next stage item to be met
Part.
Specifically, server is with the age mark of same face sample image corresponding parameter adjustment face-image and selection
Value outputs and inputs the training of progress current generation respectively as to training pattern, until meeting in the training of current generation
Only condition.
In one embodiment, the training suspension condition of current generation can stop duration for training, and server is with current
Stage, corresponding facial sample image was concentrated, the corresponding parameter adjustment face-image of each face sample image and selection
Age mark value, carry out the training of current generation respectively as to outputting and inputting for training pattern, and count trained consuming
Duration stops duration equal to current generation corresponding training when the training counted on expends duration, then meets the training of current generation
Suspension condition stops the model training of current generation.
S208 determines the training suspension condition of next stage according to the training suspension condition of current generation.
Wherein, the training suspension condition of next stage depends on the training suspension condition of current generation.
Specifically, when training suspension condition is that training stops wheel number, when the current generation counted on corresponding exercise wheel
When number is equal to current generation corresponding training and stops wheel number, server stops current generation corresponding model training, according to default
Suspension condition adjustment mode, training corresponding to the current generation stop wheel number and are adjusted, and the training for obtaining next stage stops
Take turns number.
In one embodiment, corresponding when the current generation counted on when training suspension condition is that training stops duration
Training when expending duration and being equal to current generation corresponding training and stop duration, server stops current generation corresponding model instruction
Practice, according to default suspension condition adjustment mode, training corresponding to the current generation stops duration and is adjusted, and obtains next stage
Training stop duration.
For example, current generation corresponding training is 100 minutes a length of when stopping, then presetting suspension condition adjustment mode is decline
Adjustment 10% obtains 90 minutes then to the adjustment of decline in 100 minutes 10%, then stops for next stage corresponding training within 90 minutes
Duration.
S210 is returned using next stage as the current generation and is obtained current generation corresponding facial sample graph image set continuation
It treats training pattern to be trained, until the training suspension condition of next stage meets training completion condition, obtaining the age is estimated
Model.
Specifically, server is returned to S202 and is reacquired current generation corresponding face using next stage as the current generation
Portion's sample graph image set treats training pattern according to the facial sample graph image set reacquired and continues to train, instantly single order
When the corresponding trained suspension condition of section meets training completion condition, deconditioning is estimated using the model that training obtains as the age
Model.
In one embodiment, when training suspension condition is that training stops duration, stopped according to the training of current generation
Duration determines that the training of next stage stops duration, completes duration equal to default training when the training of next stage stops duration
When, then meet trained completion condition.It can be 0 that duration is completed in default training.
In the present embodiment, face sample image corresponding to the current generation concentrates each facial sample image, adjusts image
Mass parameter obtains parameter adjustment face-image, and chooses age mark value from mark range of corresponding age, according to parameter
Adjust the training of the age mark value progress current generation of face-image and selection.By adjusting image matter during training
Parameter is measured, so that the influence that the model that training obtains has fully considered environmental factor to picture quality, to consider image matter
The influence to model recognition result is measured, after the training by multiple stages, improves the age prediction model that training obtains
Age estimates accuracy.
As shown in figure 3, in one embodiment, S202 specifically includes the step of obtaining facial sample graph image set, the step
Specifically include the following contents:
S302 obtains the sample ratio of age of default sample size and each age group.
Wherein, presetting sample size is the quantity that facial sample image concentrates the facial sample image for including.Each age group
Sample ratio of age, the ratio of the quantity of the facial sample image of each age group is concentrated for facial sample image.
Specifically, server parses training sign on, leads to after receiving terminal and sending training sign on
Cross the sample ratio of age of default sample size and each age group that parsing is extracted in training sign on.
For example, each age group may include 0-12 years old, 13-65 years old and be greater than 65 years old three age bracket, 0-12 years old,
13-65 years old and greater than 65 years old three age bracket sample ratio of age be 1:3:1, then it represents that the facial sample of 0-12 years old age bracket
The facial sample image quantity that amount of images accounts for 1/5,13-65 years old age bracket of facial sample image collection septum reset sample size accounts for
The 3/5 of facial sample image collection septum reset sample size, the facial sample image quantity greater than 65 years old age bracket account for facial sample
The 1/5 of image set septum reset sample size.Each age group can also be divided with other age brackets.
S304 determines the corresponding image of each age group according to the sample proportion of default sample size and each age group
Sample size.
Specifically, after server obtains the sample proportion for presetting sample size and each age group, according to the sample of each age group
The corresponding sample accounting of this ratio-dependent each age group, according to the corresponding sample accounting of each age group and default sample size, really
Determine the corresponding image pattern quantity of each age group.
For example, if each age group include 0-12 years old, 13-65 years old and be greater than 65 years old three age bracket, 0-12 years old, 13-
65 years old and be 1:3:1 greater than the sample ratio of age of 65 years old three age bracket, it is determined that 0-12 year old, 13-65 years old and greater than 65 years old
The corresponding sample accounting of three age brackets is 1/5,3/5 and 1/5.If the year that default sample size is 1000,0-12 years old
The corresponding image pattern quantity=1000 × 1/5=200 of age section, corresponding image pattern quantity=1000 of the age bracket of 13-65
× 3/5=600, image pattern quantity=1000 × 1/5=200 corresponding greater than 65 years old age bracket.
S306 is concentrated from facial sample training according to the corresponding image pattern quantity of each age group and is screened facial sample
This image obtains current generation corresponding facial sample graph image set.
Specifically, server is concentrated from facial sample training and is sieved according to the corresponding image pattern quantity of each age group
Select the facial sample image of each age group of image pattern quantity Matching corresponding with each age group, the face obtained by screening
Portion's sample image forms current generation corresponding facial sample graph image set.
In one embodiment, server concentrates the age of each facial sample image to mark model according to facial sample training
Enclose, determine and be belonging respectively to the facial sample image of each age group, according to the corresponding image pattern quantity of each age group from point
Do not belong in the facial sample image of each age group, screen facial sample image, obtains current generation corresponding facial sample graph
Image set.
In the present embodiment, according to the sample ratio of age of default sample size and each age group, according to default sample size
With the sample proportion of each age group, determines the corresponding image pattern quantity of each age group, respectively corresponded according to each age group
Image pattern quantity screen facial sample image, guarantee obtains the facial sample image collection septum reset sample image corresponding age
Accurate distribution, further improve the accuracy of age prediction model got according to the training of facial sample image.
As shown in figure 4, in one embodiment, S204 specifically further includes the steps that adjusting image quality parameter, the step
Specifically include the following contents:
S402 reads each facial sample image that facial sample image is concentrated.
Specifically, server reads facial sample image after screening current generation corresponding facial sample graph image set
The each facial sample image concentrated.
In one embodiment, server is according to data localized quantity, from current generation corresponding facial sample graph
In image set, read with the facial sample image of data localized quantity Matching, using the facial sample image that reads as working as
The corresponding facial sample image of preceding round.
S404 obtains image quality parameter corresponding with the facial sample image read and age mark range.
Wherein, image quality parameter is the parameter for indicating picture quality.Image quality parameter may include brightness of image and
Image blur.It is the range of age marked to facial sample image that age, which marks range,.
Specifically, server parses the parameter of the facial sample image read after reading facial sample image, leads to
It crosses parsing and obtains image quality parameter from the parameter of the facial sample image read.Server extracts the facial sample read
The image identification of this image inquires age mark range corresponding with the image identification extracted.
S406 determines the corresponding parameter adjustment mode of each image quality parameter in the image quality parameter that gets.
Specifically, there are various image quality parameters, every kind of picture quality ginseng in the image quality parameter that server is read
The corresponding parameter adjustment mode of number is not also identical.Server determines each middle image quality parameter in the image quality parameter got
Affiliated parameter type, according to inquiry parameter adjustment mode corresponding with the parameter type determined.
S408 is adjusted image quality parameter according to determining parameter adjustment mode, obtains parameter adjustment face figure
Picture.
Specifically, when determining parameter adjustment mode is random adjustment, each image quality parameter that server determines divides
Not corresponding parameter area randomly selects image quality parameter, root in the corresponding parameter area of each image quality parameter
The facial sample image read is adjusted according to the image quality parameter of selection, obtains parameter adjustment face-image.
In one embodiment, when determining parameter adjustment mode is random adjustment, the determining each image matter of server
The corresponding parameter Candidate Set of parameter is measured, randomly selects image quality parameter from parameter Candidate Set.
In one embodiment, image quality parameter includes brightness of image and image blur.When brightness of image is corresponding
When parameter adjustment mode is random adjustment, server determines the parameter area of brightness of image, random in determining parameter area
Choose brightness of image.When the corresponding parameter adjustment mode of image blur be pixel adjustment mode, then at random generate sampling parameter,
Down-sampling or up-sampling treatment are carried out according to pixel of the sampling parameter generated at random to the facial sample image read, is obtained
Facial sample image after to sampling is adjusted the facial sample image after sampling according to the brightness of image randomly selected,
Obtain parameter adjustment face-image.
In one embodiment, one and the face sample graph are arranged for each facial sample image in server
As corresponding face-image set, multiple parameter adjustment faces corresponding with the face sample image are store in face-image set
Image, and the image quality parameter of each parameter adjustment face-image is different from.Server is from the facial sample graph read
As Selecting All Parameters adjust face-image in corresponding face-image set.
S410 chooses age mark value from age criterion range, as the corresponding age mark of parameter adjustment face-image
Note value.
Specifically, server determines the age mark within the scope of the corresponding age criterion of facial sample image read
Value, randomly selects an age mark value in determining age mark value, is adjusted using the age mark value of selection as parameter
The corresponding age mark value of face-image.
For example, being referred to Fig. 5, Fig. 5 is the schematic diagram marked at the age of one embodiment septum reset sample image.
The corresponding age mark range of facial sample image in Fig. 5 is 2-4 years old, then the age marks range 2-4 years old corresponding age mark
Note value includes 2 years old, 3 years old and 4 years old, and an age mark value is chosen from 2 years old, 3 years old and 4 years old when training.
In the present embodiment, the image quality parameter for reading facial sample image is adjusted, obtains parameter adjustment figure
Picture chooses age mark value as parameter from mark range of corresponding age and adjusts the corresponding age mark value of face-image,
Face-image and corresponding age mark value training pattern are adjusted according to parameter, guarantees that the model trained considers image parameter
Variation and change of age improve the accuracy for the model that training obtains.
In one embodiment, S410 specifically includes the following contents: determining each age mark value point in age criterion range
Not corresponding selection probability;Age mark value is chosen from each age mark value according to determining selection probability;With the year of selection
Age mark value is as the corresponding age mark value of parameter adjustment face-image.
For example, the corresponding age mark range of facial sample image is 2-4 year old, then age mark range 2-4 years old it is right
The age mark value answered includes 2 years old, 3 years old and 4 years old, and each age mark value is both provided with corresponding selection probability, if each age
When the corresponding selection probability of standard value is 1/3, server chooses an age mark from 2 years old, 3 years old and 4 years old at random
Value;If 2 years old selection probability value is 1/5,3 years old selection probability value is 3/5 and 4 years old selection probability value is 1/5,3 are chosen
The maximum probability in year.
As shown in fig. 6, in one embodiment, providing a kind of age prediction model training method, training suspension condition is
Target exercise wheel number;This method specifically includes the following contents:
S602 obtains current generation corresponding facial sample graph image set.
S604, for each facial sample image that the facial sample image is concentrated, adjustment image quality parameter is obtained
Parameter adjusts face-image, and chooses age mark value from mark range of corresponding age.
Specifically, server is concentrated from current generation corresponding facial sample image, is read according to data localized quantity
The corresponding facial sample image of current training round is taken, for each facial sample graph in the facial sample image that reads
Picture, adjustment image quality parameter obtains parameter adjustment face-image, and chooses age mark from mark range of corresponding age
Value.
For example, Fig. 7 is the schematic diagram for adjusting image quality parameter in one embodiment.Image quality parameter includes figure
As fuzziness and brightness of image, server carries out the image blur and brightness of image of the facial sample image read random
Adjustment obtains parameter adjustment face-image.Referring to Fig. 7, illustrate to adjust brightness of image or image blur at random in Fig. 7
Whole a variety of adjustment results.
S606 adjusts face-image as input and using the age mark value of selection as output using parameter, treats training
The training of model progress current generation.
Specifically, each page sample image corresponding parameter adjustment face-image concentrated for facial sample image and
The age mark value of selection, server adjust face-image as the input to training pattern and the year to choose using parameter
Age mark value is treated training pattern and is trained as the output to training pattern.
In one embodiment, server reads the corresponding facial sample image of current training round, with what is read
Each face sample image corresponding parameter adjustment face-image as input and using the age mark value of selection as exporting,
It treats training pattern and carries out the training for currently training round in the current generation, when according to the corresponding facial sample of current training round
After image training, then from current generation corresponding facial sample image concentration, read the corresponding face of next trained round
Sample image, which continues to treat training pattern, to be trained.
In one embodiment, characteristics of image in server extracting parameter adjustment face-image, will extract characteristics of image
The input layer to training pattern is inputted, is obtained to each output node in the output layer of training pattern according to the characteristics of image extracted
The training output valve of output constructs loss function according to the age mark value of the training output valve and selection got, according to damage
The parameter that mistake function treats training pattern is adjusted.
In one embodiment, server is obtained to each output node in the output layer of training pattern according to the figure extracted
As the training output valve that feature exports, the training output valve of each output node output is normalized, each output is obtained
The corresponding normalized value of node.Server is according to the age mark value structure of the corresponding normalized value of each output node and selection
Build loss function.
In one embodiment, it can be normalized by the training output valve that following formula export each output node
Processing:
Wherein, aiFor the corresponding normalized value of i-th of output node, ziFor the corresponding training output of i-th of output node
Value, e are natural constant, and k is the quantity of output node;
Loss function can be constructed by following formula:
Wherein, C indicates that loss function, i indicate i-th of output node, aiFor the corresponding normalization of i-th of output node
Value, yiIndicate the corresponding age mark value of i-th of output node.
S608, statistics current generation corresponding exercise wheel number.
Specifically, when server is according to the facial sample image collection training pattern of current generation, preset quantity is read every time
Facial sample image, complete to complete the one of the current generation according to the training process of sample image training pattern read
Wheel training.Server is when according to the facial sample image collection training pattern of current generation, the corresponding training of statistics current generation
Take turns number.
S610, when statistics exercise wheel number be more than or equal to the current generation target exercise wheel number, then stop the current generation
Training.
Specifically, server obtains the target exercise wheel number of current generation, by the exercise wheel number of statistics and current generation
Target exercise wheel number is compared, and when the exercise wheel number of statistics is more than or equal to the target exercise wheel number of current generation, is then stopped
The training of current generation;When the exercise wheel number of statistics is less than the target exercise wheel number of current generation, continue according to the current generation
The facial sample image that corresponding face sample image is concentrated is treated training pattern and is trained.
S612 determines the target of next stage according to the target exercise wheel number of default wheel number down ratio and current generation
Exercise wheel number.
Wherein, presetting wheel number down ratio is that next stage corresponding target exercise wheel number is corresponding relative to the current generation
Target exercise wheel number, the ratio of exercise wheel number reduction.
Specifically, default wheel number down ratio is multiplied by server with the target exercise wheel number of current generation, obtains
Wheel number is reduced, the target training data and reduction wheel number to the current generation do subtraction, obtain the target training of next stage
Take turns number.
For example, default wheel number down ratio is 20%, if current generation corresponding target exercise wheel number is 5000 wheels,
Then reduce wheel number=5000 × 20%=1000 wheel, target exercise wheel number=5000-1000=4000 wheel of next stage.
S614, when the target exercise wheel number of next stage is greater than 0, using a next stage as the current generation;Return to S602.
Specifically, server is after determining the target exercise wheel number of next stage, by the target exercise wheel number of next stage
It is compared with 0, when the target exercise wheel number of next stage is greater than 0, expression does not meet trained completion condition, with next stage
As the current generation, returns to S602 and reacquire current generation corresponding facial sample graph image set, according to the face reacquired
Portion's sample graph image set, which continues to treat training pattern, to be trained.
S616 then meets trained completion condition when the target exercise wheel number of next stage is equal to 0, and obtaining the age estimates
Model.
Specifically, when the target exercise wheel number of next stage is equal to 0, then meet trained completion condition, server stops
Model training, using current generation trained model as age prediction model.
In the present embodiment, the target exercise wheel number of next stage is controlled by default wheel number down ratio, improves training
Decrease speed in the process saves the time spent by model training.Each stage requires to choose facial sample graph again
Image set continues to be trained model according to the facial sample graph image set chosen again, avoids the model opposite that training obtains
The image that portion's sample image is concentrated excessively is fitted, to reduce the accuracy for the age prediction model that training obtains.
As shown in figure 8, in one embodiment, providing a kind of face-image recognition methods.The present embodiment is mainly with the party
The terminal 110 that method is applied in Fig. 1 is come for example, the server 120 in Fig. 1 can also be applied.Wherein, the age estimates mould
Type is obtained according to the training of age prediction model training method.Referring to Fig. 8, which specifically includes following
Step:
S802 obtains face-image.
Specifically, image capture device is installed in terminal, image capture device is called to carry out the face of target object
Image Acquisition, by calling image capture device to obtain face-image.
In one embodiment, terminal and image capture device pass through network connection.Image capture device is to target scene
In target object carry out Image Acquisition, acquired image is sent to terminal.Terminal receives what image capture device was sent
Image extracts face-image from acquired image.
In one embodiment, image library is provided in terminal, terminal obtains the image selection instruction of triggering, according to image
Selection instruction chooses face-image from image library.
Face-image is inputted age prediction model by S804, obtains each default age bracket point of age prediction model output
Not corresponding probability value.
Specifically, terminal extracts the characteristics of image of face-image after obtaining face-image, the characteristics of image that will be extracted
It is input to the input layer of age prediction model, obtains age prediction model according to each default year for the characteristics of image output extracted
The corresponding probability value of age section.
For example, Fig. 9 is to identify schematic diagram to the age of face-image in one embodiment.Face is schemed referring to Fig. 9
As input age prediction model, the corresponding probability value of the import and export of age prediction model each age group shown in Fig. 9, for example, 0 years old
Corresponding probability value be 0.01,1 years old corresponding probability value be 0.09 ..., 64 years old corresponding probability value be that 0.002 and 65 years old are right
The probability value answered is 0.001, wherein what is indicated within 65 years old may be greater than equal to 65 years old.
In one embodiment, S804 specifically further includes the following contents: extracting the characteristics of image in face-image;It will extract
To characteristics of image be input to the input layer in age prediction model;Obtain the right respectively with each default age bracket of output layer output
The output valve answered;The output valve got is normalized, the corresponding probability value of each default age bracket is obtained.
Specifically, terminal extracts characteristics of image from face-image, and the characteristics of image extracted is input to the age and is estimated
The each input node of input layer in model.Terminal obtain age prediction model in output layer each output node output with it is each pre-
If each probability value got is normalized in the corresponding output valve of age bracket, each default age bracket point is obtained
Not corresponding probability value.
In one embodiment, the corresponding output valve of each default age bracket is added by terminal, obtains output valve
Sum, obtain each default age bracket correspondence divided by sum of output valve with each default corresponding output valve of age bracket respectively
Probability value.
In one embodiment, normalizing can be carried out to the corresponding output valve of each default age bracket by following formula
Change processing:
Wherein, aiFor the corresponding normalized value of i-th of default age bracket, ziIt is defeated for the corresponding training of i-th of default age bracket
It is worth out, e is natural constant, and k is the quantity of default age bracket.
S806 determines age desired value according to the corresponding probability value of each default age bracket.
Specifically, after terminal obtains the corresponding probability value of each default age bracket, by each default age bracket respectively with it is right
The probability value answered is multiplied, and obtains the corresponding product of each default age bracket, sum to the corresponding product of each default age bracket
To age desired value.
In one embodiment, age desired value can be calculated according to the following formula:
Age desired value=∑ m × p (m)
Wherein, m indicates that age bracket, p (m) indicate the corresponding probability value of age bracket m.
S808, the age discreet value with the corresponding age discreet value of age desired value, as face-image.
Specifically, terminal determines age desired value, the integer in determining age desired value is extracted, with the integer extracted
Age discreet value as face-image.For example, age desired value be 23.24, rounding obtain 23, then 23 be face-image year
Age discreet value;Age desired value be 24.68, rounding obtain 24, then 24 be face-image age discreet value.
In a real-time example, after terminal determines age desired value, round up to determining age desired value, with
Age discreet value of the age desired value to round up as face-image.For example, age desired value is 23.24, it is right
23.24 are rounded up to obtain 23, then 23 be face-image age discreet value;Age desired value is 24.68, to 24.68
Rounded up to obtain 25, then 25 be face-image age discreet value.
In the present embodiment, according to consider influence of the environmental factor to collected face-image age prediction model,
Age identification is carried out to the face-image got, improves the accuracy of the corresponding probability value of each default age bracket.Again
Age desired value is calculated according to the corresponding probability value of each default age bracket, is estimated to be worth to obtain the age according to age expectation
Value, to further improve the accuracy of age discreet value.
As shown in Figure 10, in one embodiment, S802 specifically includes the step of obtaining face-image, which specifically wraps
Include the following contents:
S1002 acquires video image.
Specifically, terminal acquires equipment by network connection image, and image capture device carries out video figure to place scene
As acquisition, collected video image acquisition is sent to terminal.Terminal receives the video image that image capture device is sent.Figure
Scene can the application scenarios such as intelligent shop, automatic vending machine and intelligent robot as where acquisition equipment.
S1004 identifies the facial area in video image.
Specifically, terminal identifies the video frame in video image, the characteristics of image in video frame is extracted, to image
Feature is analyzed, it is determined whether there are facial characteristics, facial characteristics if it exists, it is determined that the facial area in video frame.
In one embodiment, S1004 is specific further include: collected video image is inputted facial identification model, is obtained
To face recognition result;In collected video image, the video frame comprising facial area is determined according to face recognition result;
Facial area in determining video frame is labeled.
Specifically, face recognition model is provided in terminal, face recognition model is used to scheme the face in video image
As being identified.Collected video image is inputted facial identification model by terminal, obtains the face of face recognition model output
Recognition result, face recognition result include face-image position.Terminal is according to face recognition result from collected video
The video frame including face-image is extracted in image, facial area is determined according to face recognition result in the video frame, to determination
Video frame in facial area add mark.Addition mark specifically can be addition can mark facial area have closed sides
The frame on boundary, can be box.
S1006 intercepts the facial area recognized from video image, obtains face-image.
Specifically, terminal is after recognizing the facial area in the video frame in video image, according to the face recognized
Region intercepts parts of images from video frame, using the parts of images that is truncated to as face-image.
In the present embodiment, by acquiring video image, identifies the facial area in video image, intercepted from video image
The facial area recognized extracts face-image by the identification to video image septum reset region and improves face-image
Obtain efficiency and accuracy.
It should be understood that each step in each flow chart is successively shown according to the instruction of arrow, but these steps
It is not that the inevitable sequence according to arrow instruction successively executes.Unless expressly stating otherwise herein, the execution of these steps is simultaneously
Not stringent sequence limitation, these steps can execute in other order.Moreover, at least part in each flow chart walks
It suddenly may include that perhaps these sub-steps of multiple stages or stage have been executed in synchronization to multiple sub-steps
At, but can execute at different times, the execution sequence in these sub-steps or stage, which is also not necessarily, successively to be carried out, and
It is that can be executed in turn or alternately at least part of the sub-step or stage of other steps or other steps.
In one embodiment, face-image recognition methods is applied in intelligent shop, and image capture device acquisition is each
The face-image of position customer, is transferred to terminal for face-image.Terminal identifies collected face-image, available
The face-image corresponding age can also extract shop corresponding with collected face-image from database and show up information.
For example, Figure 11 is that customer's details show schematic diagram in one embodiment.Referring to Fig.1 1, terminal is to face-image
Identified, obtain customer's details, customer's details are shown, customer's details include face-image, type be ordinary customer,
Face number FaceID is 41865, gender is male, the age 43, time of arrival is 2018-09-23 13:49:58, goes through for the first time
History visiting number is 318 times.
Figure 12 is the schematic diagram that customer's acess control result is shown in one embodiment.Referring to Fig.1 2, scheme according to face
The recognition result of picture can count the men and women's proportion visited in each age group in preset time period.Dark parts are male in figure
Property, light-colored part be that women, percentage identify each year section section and visits task proportion, for example, Figure 12 Lipid on Young-middle Male is visited
Number accounting is 27.3%, and young women visiting number accounting is 30.2%.
As shown in figure 13, in one embodiment, a kind of age prediction model training device 1300 is provided, the device is specific
Including the following contents: sample image obtains module 1302, mass parameter adjustment module 1304, model training module 1306, stops
Condition determining module 1308 and training loop module 1310.
Sample image obtains module 1302, for obtaining current generation corresponding facial sample graph image set.
Mass parameter adjusts module 1304, each facial sample image for concentrating for facial sample image, adjustment
Image quality parameter obtains parameter adjustment face-image, and chooses age mark value from mark range of corresponding age.
Model training module 1306 treats training for adjusting the age mark value of face-image and selection according to parameter
Model carries out the training of current generation, until meeting the training suspension condition of current generation.
Stop condition determining module 1308 determines the training of next stage for the training suspension condition according to the current generation
Suspension condition.
Training loop module 1310, for using next stage as the current generation, notice sample image to obtain module 1302
Current generation corresponding facial sample graph image set is obtained, is trained with continuing to treat training pattern, until the instruction of next stage
Practice suspension condition and meet training completion condition, obtains age prediction model.
In one embodiment, sample image obtains module 1302 and is also used to obtain default sample size and each age group
Sample ratio of age;According to the sample proportion of default sample size and each age group, the corresponding image of each age group is determined
Sample size;According to the corresponding image pattern quantity of each age group, is concentrated from facial sample training and screen facial sample graph
Picture obtains current generation corresponding facial sample graph image set.
In one embodiment, mass parameter adjustment module 1304 is also used to read every one side that facial sample image is concentrated
Portion's sample image;Obtain image quality parameter corresponding with the facial sample image read and age mark range;Determination obtains
The corresponding parameter adjustment mode of each image quality parameter in the image quality parameter got;According to determining parameter adjustment side
Formula is adjusted image quality parameter, obtains parameter adjustment face-image;Age mark value is chosen from age criterion range,
As the corresponding age mark value of parameter adjustment face-image.
In one embodiment, mass parameter adjustment module 1304 is also used to determine each age mark in age criterion range
It is worth corresponding selection probability;Age mark value is chosen from each age mark value according to determining selection probability;To choose
Age mark value as the corresponding age mark value of parameter adjustment face-image.
In one embodiment, training suspension condition is target exercise wheel number;Model training module 1306 is also used to join
Number adjustment face-image treats training pattern and carries out the current generation as input and using the age mark value of selection as output
Training;Count current generation corresponding exercise wheel number;When the exercise wheel number of statistics is more than or equal to the target training of current generation
Number is taken turns, then stops the training of current generation.
In one embodiment, stop condition determining module 1308 is also used to according to default wheel number down ratio and current rank
The target exercise wheel number of section, determines the target exercise wheel number of next stage.
In one embodiment, training loop module 1310 is also used to when the target exercise wheel number of next stage is greater than 0,
Using a next stage as the current generation, notice sample image obtains module 1302 and obtains current generation corresponding facial sample image
Collection, is trained with continuing to treat training pattern;When the target exercise wheel number of next stage is equal to 0, then meet trained completion
Condition obtains age prediction model.
In the present embodiment, face sample image corresponding to the current generation concentrates each facial sample image, adjusts image
Mass parameter obtains parameter adjustment face-image, and chooses age mark value from mark range of corresponding age, according to parameter
Adjust the training of the age mark value progress current generation of face-image and selection.By adjusting image matter during training
Parameter is measured, so that the influence that the model that training obtains has fully considered environmental factor to picture quality, to consider image matter
The influence to model recognition result is measured, after the training by multiple stages, improves the age prediction model that training obtains
Age estimates accuracy.
As shown in figure 14, in one embodiment, a kind of face-image identification device 1400 is provided, which specifically includes
The following contents: face-image obtains module 1402, face-image identification module 1404, age expectation determining module 1406 and age
Estimate module 1408.
Face-image obtains module 1402, for obtaining face-image.
Face-image identification module 1404, for obtaining the output of age prediction model for face-image age prediction model
The corresponding probability value of each default age bracket.
Age it is expected determining module 1406, for determining age period according to the corresponding probability value of each default age bracket
Prestige value.
Age estimates module 1408, for the age with the corresponding age discreet value of age desired value, as face-image
Discreet value.
In one embodiment, face-image obtains module 1402 and is also used to acquire video image;It identifies in video image
Facial area;The facial area recognized is intercepted from video image, obtains face-image.
In one embodiment, face-image obtains module 1402 and is also used to collected video image inputting face and knows
Other model, obtains face recognition result;In collected video image, determine to include facial area according to face recognition result
Video frame;Facial area in determining video frame is labeled.
In one embodiment, face-image identification module 1404 is also used to extract the characteristics of image in face-image;It will
The characteristics of image extracted is input to the input layer in age prediction model;Obtain dividing with each default age bracket for output layer output
Not corresponding output valve;The output valve got is normalized, the corresponding probability of each default age bracket is obtained
Value.
In the present embodiment, according to consider influence of the environmental factor to collected face-image age prediction model,
Age identification is carried out to the face-image got, improves the accuracy of the corresponding probability value of each default age bracket.Again
Age desired value is calculated according to the corresponding probability value of each default age bracket, is estimated to be worth to obtain the age according to age expectation
Value, to further improve the accuracy of age discreet value.
Figure 15 shows the internal structure chart of computer equipment in one embodiment.The computer equipment specifically can be figure
Server 120 in 1.As shown in figure 15, it includes being connected by system bus which, which includes the computer equipment,
Processor, memory and network interface.Wherein, memory includes non-volatile memory medium and built-in storage.The computer is set
Standby non-volatile memory medium is stored with operating system, can also be stored with computer program, and the computer program is by processor
When execution, processor may make to realize age prediction model training method.Computer program can also be stored in the built-in storage,
When the computer program is executed by processor, processor may make to execute age prediction model training method.
Figure 16 shows the internal structure chart of computer equipment in another embodiment.The computer equipment specifically can be
Terminal 140 in Fig. 1.As shown in figure 16, it includes being connected by system bus which, which includes the computer equipment,
Processor, memory, network interface, input unit and display screen.Wherein, memory includes non-volatile memory medium and memory
Reservoir.The non-volatile memory medium of the computer equipment is stored with operating system, can also be stored with computer program, the calculating
When machine program is executed by processor, processor may make to realize face-image recognition methods.It can also be stored in the built-in storage
Computer program when the computer program is executed by processor, may make processor to execute face-image recognition methods.Computer
The display screen of equipment can be liquid crystal display or electric ink display screen, and the input unit of computer equipment can be display
The touch layer covered on screen is also possible to the key being arranged on computer equipment shell, trace ball or Trackpad, can also be outer
Keyboard, Trackpad or mouse for connecing etc..
It will be understood by those skilled in the art that structure shown in Figure 15 and 16, only relevant to application scheme
The block diagram of part-structure does not constitute the restriction for the computer equipment being applied thereon to application scheme, specific to calculate
Machine equipment may include perhaps combining certain components or with different components than more or fewer components as shown in the figure
Arrangement.
In one embodiment, age prediction model training device provided by the present application can be implemented as a kind of computer journey
The form of sequence, computer program can be run in computer equipment as shown in figure 15.It can be deposited in the memory of computer equipment
Storage forms each program module of the age prediction model training device, for example, sample image shown in Figure 13 obtains module
1302, mass parameter adjustment module 1304, model training module 1306, stop condition determining module 1308 and training loop module
1310.The computer program that each program module is constituted makes processor execute each implementation of the application described in this specification
Step in the age prediction model training method of example.
For example, computer equipment shown in figure 15 can be by age prediction model training device as shown in fig. 13 that
Sample image obtains module 1302 and obtains current generation corresponding facial sample graph image set.Computer equipment can pass through mass parameter
Each facial sample image that adjustment module 1304 concentrates facial sample image, adjustment image quality parameter obtain parameter tune
Whole face-image, and age mark value is chosen from mark range of corresponding age.Computer equipment can pass through model training mould
Block 1306 adjusts the age mark value of face-image and selection according to parameter, treats the training that training pattern carries out the current generation,
Until meeting the training suspension condition of current generation.Computer equipment can be by stop condition determining module 1308 according to current rank
The training suspension condition of section determines the training suspension condition of next stage.Computer equipment can by training loop module 1310 with
Next stage as the current generation, notifies sample image to obtain module 1302 and obtain current generation corresponding facial sample image
Collection, is trained with continuing to treat training pattern, until the training suspension condition of next stage meets training completion condition, is obtained
Age prediction model.
In one embodiment, age prediction model training device provided by the present application can be implemented as a kind of computer journey
The form of sequence, computer program can be run in computer equipment as shown in figure 16.It can be deposited in the memory of computer equipment
Storage forms each program module of the face-image identification device, for example, face-image shown in Figure 14 obtains module 1402, face
Portion's picture recognition module 1404, age expectation determining module 1406 and age estimate module 1408.What each program module was constituted
Computer program executes processor in the face-image recognition methods of each embodiment of the application described in this specification
The step of.
For example, computer equipment shown in Figure 16 can pass through the face in face-image identification device as shown in figure 14
Image collection module 1402 obtains face-image.Computer equipment can be by face-image identification module 1404 by face-image year
Age prediction model obtains the corresponding probability value of each default age bracket of age prediction model output.Computer equipment can lead to
Crossing age period hopes determining module 1406 according to the corresponding probability value of each default age bracket, determines age desired value.Computer
Equipment can estimate module 1408 by the age with the corresponding age discreet value of age desired value, and the age as face-image is estimated
Value.
In one embodiment, a kind of computer equipment, including memory and processor are provided, memory is stored with meter
Calculation machine program, when computer program is executed by processor, so that processor executes the step of above-mentioned age prediction model training method
Suddenly.The step of age prediction model training method can be in the age prediction model training method of above-mentioned each embodiment herein
The step of.
In one embodiment, a kind of computer readable storage medium is provided, computer program, computer journey are stored with
When sequence is executed by processor, so that the step of processor executes above-mentioned age prediction model training method.The age estimates mould herein
The step of type training method, can be the step in the age prediction model training method of above-mentioned each embodiment.
In one embodiment, a kind of computer equipment, including memory and processor are provided, memory is stored with meter
Calculation machine program, when computer program is executed by processor, so that the step of processor executes above-mentioned face-image recognition methods.This
The step of locating face-image recognition methods can be the step in the face-image recognition methods of above-mentioned each embodiment.
In one embodiment, a kind of computer readable storage medium is provided, computer program, computer journey are stored with
When sequence is executed by processor, so that the step of processor executes above-mentioned face-image recognition methods.Face-image identification side herein
The step of method, can be the step in the face-image recognition methods of above-mentioned each embodiment.
Those of ordinary skill in the art will appreciate that realizing all or part of the process in above-described embodiment method, being can be with
Relevant hardware is instructed to complete by computer program, the program can be stored in a non-volatile computer and can be read
In storage medium, the program is when being executed, it may include such as the process of the embodiment of above-mentioned each method.Wherein, provided herein
Each embodiment used in any reference to memory, storage, database or other media, may each comprise non-volatile
And/or volatile memory.Nonvolatile memory may include that read-only memory (ROM), programming ROM (PROM), electricity can be compiled
Journey ROM (EPROM), electrically erasable ROM (EEPROM) or flash memory.Volatile memory may include random access memory
(RAM) or external cache.By way of illustration and not limitation, RAM is available in many forms, such as static state RAM
(SRAM), dynamic ram (DRAM), synchronous dram (SDRAM), double data rate sdram (DDRSDRAM), enhanced SDRAM
(ESDRAM), synchronization link (Synchlink) DRAM (SLDRAM), memory bus (Rambus) directly RAM (RDRAM), straight
Connect memory bus dynamic ram (DRDRAM) and memory bus dynamic ram (RDRAM) etc..
Each technical characteristic of above embodiments can be combined arbitrarily, for simplicity of description, not to above-described embodiment
In each technical characteristic it is all possible combination be all described, as long as however, the combination of these technical characteristics be not present lance
Shield all should be considered as described in this specification.
The several embodiments of the application above described embodiment only expresses, the description thereof is more specific and detailed, but simultaneously
The limitation to the application the scope of the patents therefore cannot be interpreted as.It should be pointed out that for those of ordinary skill in the art
For, without departing from the concept of this application, various modifications and improvements can be made, these belong to the guarantor of the application
Protect range.Therefore, the scope of protection shall be subject to the appended claims for the application patent.
Claims (15)
1. a kind of age prediction model training method, comprising:
Obtain current generation corresponding facial sample graph image set;
For each facial sample image that the facial sample image is concentrated, adjustment image quality parameter obtains parameter adjustment face
Portion's image, and age mark value is chosen from mark range of corresponding age;
The age mark value that face-image and selection are adjusted according to the parameter treats the instruction that training pattern carries out the current generation
Practice, until meeting the training suspension condition of current generation;
The training suspension condition of next stage is determined according to the training suspension condition of the current generation;
Using next stage as the current generation, returns to the acquisition current generation corresponding facial sample graph image set and continue to described
It is trained to training pattern, until the training suspension condition of next stage meets training completion condition, obtaining the age estimates mould
Type.
2. the method according to claim 1, wherein the acquisition current generation corresponding facial sample graph image set
Include:
Obtain the sample ratio of age of default sample size and each age group;
According to the sample proportion of the default sample size and each age group, the corresponding image sample of each age group is determined
This quantity;
According to the corresponding image pattern quantity of each age group, is concentrated from facial sample training and screen facial sample graph
Picture obtains current generation corresponding facial sample graph image set.
3. the method according to claim 1, wherein the every one side concentrated for the facial sample image
Portion's sample image, adjustment image quality parameter obtains parameter adjustment face-image, and chooses from mark range of corresponding age
Age mark value includes:
Read each facial sample image that the facial sample image is concentrated;
Obtain image quality parameter corresponding with the facial sample image read and age mark range;
Determine the corresponding parameter adjustment mode of each image quality parameter in the image quality parameter that gets;
Image quality parameter is adjusted according to determining parameter adjustment mode, obtains parameter adjustment face-image;
Age mark value is chosen from the age criterion range, as the corresponding age mark of parameter adjustment face-image
Value.
4. according to the method described in claim 3, it is characterized in that, described choose age mark from the age criterion range
Value, adjusting the corresponding age mark value of face-image as the parameter includes:
Determine the corresponding selection probability of each age mark value in the age criterion range;
Age mark value is chosen from each age mark value according to determining selection probability;
The corresponding age mark value of face-image is adjusted using the age mark value of selection as the parameter.
5. the method according to claim 1, wherein the training suspension condition is target exercise wheel number;It is described
The age mark value that face-image and selection are adjusted according to the parameter treats the training that training pattern carries out the current generation, directly
Include: to the training suspension condition for meeting the current generation
Face-image is adjusted as input and using the age mark value of selection as output using the parameter, treats training pattern
Carry out the training of current generation;
Count current generation corresponding exercise wheel number;
When statistics exercise wheel number be more than or equal to the current generation target exercise wheel number, then stop the training of current generation.
6. according to the method described in claim 5, it is characterized in that, described true according to the training suspension condition of the current generation
The training suspension condition for determining next stage includes:
According to the target exercise wheel number of default wheel number down ratio and the current generation, the target exercise wheel of next stage is determined
Number;
It is described using next stage as the current generation, return to acquisitions current generation corresponding face sample graph image set and continue pair
It is described to be trained to training pattern, until the training suspension condition of next stage meets training completion condition, it is pre- to obtain the age
Estimating model includes:
When the target exercise wheel number of next stage is greater than 0, using a next stage as the current generation, the current rank of acquisition is returned to
The corresponding facial sample graph image set of section continues to be trained to described to training pattern;
When the target exercise wheel number of next stage is equal to 0, then meets trained completion condition, obtain age prediction model.
7. a kind of face-image recognition methods, comprising:
Obtain face-image;
By the face-image age prediction model, each default age bracket for obtaining the age prediction model output is respectively corresponded
Probability value;
According to the corresponding probability value of each default age bracket, age desired value is determined;
Age discreet value with the corresponding age discreet value of the age desired value, as the face-image.
8. the method according to the description of claim 7 is characterized in that the acquisition face-image includes:
Acquire video image;
Identify the facial area in the video image;
The facial area recognized is intercepted from the video image, obtains face-image.
9. according to the method described in claim 8, it is characterized in that, the facial area packet identified in the video image
It includes:
Collected video image is inputted into facial identification model, obtains face recognition result;
In the collected video image, the video frame comprising facial area is determined according to face recognition result;
Facial area in determining video frame is labeled.
10. obtaining the method according to the description of claim 7 is characterized in that described by the face-image age prediction model
The corresponding probability value of each default age bracket of age prediction model output includes:
Extract the characteristics of image in the face-image;
The input layer characteristics of image extracted being input in age prediction model;
Obtain the output valve corresponding with each default age bracket of output layer output;
The output valve got is normalized, the corresponding probability value of each default age bracket is obtained.
11. the method according to the description of claim 7 is characterized in that the age prediction model is according to claim 1 in -6
What described in any item method training generated.
12. a kind of age prediction model training device, which is characterized in that described device includes:
Sample image obtains module, for obtaining current generation corresponding facial sample graph image set;
Mass parameter adjusts module, and each facial sample image for concentrating for the facial sample image adjusts image
Mass parameter obtains parameter adjustment face-image, and chooses age mark value from mark range of corresponding age;
Model training module treats training pattern for adjusting the age mark value of face-image and selection according to the parameter
The training of current generation is carried out, until meeting the training suspension condition of current generation;
Stop condition determining module determines that the training of next stage stops for the training suspension condition according to the current generation
Condition;
Training loop module, for using next stage as the current generation, notifying the sample image obtains module to obtain currently
Stage corresponding facial sample graph image set, to continue to be trained to described to training pattern, until in the training of next stage
Only condition meets training completion condition, obtains age prediction model.
13. a kind of face-image identification device, which is characterized in that described device includes:
Face-image obtains module, for obtaining face-image;
Face-image identification module, for obtaining the age prediction model output for the face-image age prediction model
The corresponding probability value of each default age bracket;
Age it is expected determining module, for determining age desired value according to the corresponding probability value of each default age bracket;
Age estimates module, for the age with the corresponding age discreet value of the age desired value, as the face-image
Discreet value.
14. a kind of computer readable storage medium is stored with computer program, when the computer program is executed by processor,
So that the processor is executed such as the step of any one of claims 1 to 11 the method.
15. a kind of computer equipment, including memory and processor, the memory is stored with computer program, the calculating
When machine program is executed by the processor, so that the processor is executed such as any one of claims 1 to 11 the method
Step.
Priority Applications (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201811532337.1A CN109670437B (en) | 2018-12-14 | 2018-12-14 | Age estimation model training method, facial image recognition method and device |
Applications Claiming Priority (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201811532337.1A CN109670437B (en) | 2018-12-14 | 2018-12-14 | Age estimation model training method, facial image recognition method and device |
Publications (2)
Publication Number | Publication Date |
---|---|
CN109670437A true CN109670437A (en) | 2019-04-23 |
CN109670437B CN109670437B (en) | 2021-05-07 |
Family
ID=66143908
Family Applications (1)
Application Number | Title | Priority Date | Filing Date |
---|---|---|---|
CN201811532337.1A Active CN109670437B (en) | 2018-12-14 | 2018-12-14 | Age estimation model training method, facial image recognition method and device |
Country Status (1)
Country | Link |
---|---|
CN (1) | CN109670437B (en) |
Cited By (9)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN110287942A (en) * | 2019-07-03 | 2019-09-27 | 成都旷视金智科技有限公司 | Training method, age estimation method and the corresponding device of age estimation model |
CN110378306A (en) * | 2019-07-25 | 2019-10-25 | 厦门美图之家科技有限公司 | Age prediction technique, device and image processing equipment |
CN110674397A (en) * | 2019-08-30 | 2020-01-10 | 北京百度网讯科技有限公司 | Method, device, equipment and readable medium for training age point prediction model |
CN110889457A (en) * | 2019-12-03 | 2020-03-17 | 深圳奇迹智慧网络有限公司 | Sample image classification training method and device, computer equipment and storage medium |
CN111144344A (en) * | 2019-12-30 | 2020-05-12 | 广州市百果园网络科技有限公司 | Method, device and equipment for determining age of person and storage medium |
CN111444945A (en) * | 2020-03-20 | 2020-07-24 | 北京每日优鲜电子商务有限公司 | Sample information filtering method and device, computer equipment and storage medium |
CN111931586A (en) * | 2020-07-14 | 2020-11-13 | 珠海市卓轩科技有限公司 | Face age identification method and device and storage medium |
CN112241761A (en) * | 2020-10-15 | 2021-01-19 | 北京字跳网络技术有限公司 | Model training method and device and electronic equipment |
CN112329693A (en) * | 2020-11-17 | 2021-02-05 | 汇纳科技股份有限公司 | Training method, recognition method, medium and device for gender and age recognition model |
Citations (4)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN102508907A (en) * | 2011-11-11 | 2012-06-20 | 北京航空航天大学 | Dynamic recommendation method based on training set optimization for recommendation system |
CN104463241A (en) * | 2014-10-31 | 2015-03-25 | 北京理工大学 | Vehicle type recognition method in intelligent transportation monitoring system |
CN107545245A (en) * | 2017-08-14 | 2018-01-05 | 中国科学院半导体研究所 | A kind of age estimation method and equipment |
CN107977633A (en) * | 2017-12-06 | 2018-05-01 | 平安科技(深圳)有限公司 | Age recognition methods, device and the storage medium of facial image |
-
2018
- 2018-12-14 CN CN201811532337.1A patent/CN109670437B/en active Active
Patent Citations (4)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN102508907A (en) * | 2011-11-11 | 2012-06-20 | 北京航空航天大学 | Dynamic recommendation method based on training set optimization for recommendation system |
CN104463241A (en) * | 2014-10-31 | 2015-03-25 | 北京理工大学 | Vehicle type recognition method in intelligent transportation monitoring system |
CN107545245A (en) * | 2017-08-14 | 2018-01-05 | 中国科学院半导体研究所 | A kind of age estimation method and equipment |
CN107977633A (en) * | 2017-12-06 | 2018-05-01 | 平安科技(深圳)有限公司 | Age recognition methods, device and the storage medium of facial image |
Non-Patent Citations (1)
Title |
---|
RASMUS ROTHE, RADU TIMOFTE, LUC VAN GOOL: "Deep Expectation of Real and Apparent Age from a Single Image Without Facial Landmarks", 《INTERNATIONAL JOURNAL OF COMPUTER VISION》 * |
Cited By (14)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN110287942B (en) * | 2019-07-03 | 2021-09-17 | 成都旷视金智科技有限公司 | Training method of age estimation model, age estimation method and corresponding device |
CN110287942A (en) * | 2019-07-03 | 2019-09-27 | 成都旷视金智科技有限公司 | Training method, age estimation method and the corresponding device of age estimation model |
CN110378306A (en) * | 2019-07-25 | 2019-10-25 | 厦门美图之家科技有限公司 | Age prediction technique, device and image processing equipment |
CN110674397A (en) * | 2019-08-30 | 2020-01-10 | 北京百度网讯科技有限公司 | Method, device, equipment and readable medium for training age point prediction model |
CN110889457A (en) * | 2019-12-03 | 2020-03-17 | 深圳奇迹智慧网络有限公司 | Sample image classification training method and device, computer equipment and storage medium |
CN110889457B (en) * | 2019-12-03 | 2022-08-19 | 深圳奇迹智慧网络有限公司 | Sample image classification training method and device, computer equipment and storage medium |
CN111144344A (en) * | 2019-12-30 | 2020-05-12 | 广州市百果园网络科技有限公司 | Method, device and equipment for determining age of person and storage medium |
CN111144344B (en) * | 2019-12-30 | 2023-09-22 | 广州市百果园网络科技有限公司 | Method, device, equipment and storage medium for determining person age |
CN111444945A (en) * | 2020-03-20 | 2020-07-24 | 北京每日优鲜电子商务有限公司 | Sample information filtering method and device, computer equipment and storage medium |
CN111931586A (en) * | 2020-07-14 | 2020-11-13 | 珠海市卓轩科技有限公司 | Face age identification method and device and storage medium |
CN112241761A (en) * | 2020-10-15 | 2021-01-19 | 北京字跳网络技术有限公司 | Model training method and device and electronic equipment |
CN112241761B (en) * | 2020-10-15 | 2024-03-26 | 北京字跳网络技术有限公司 | Model training method and device and electronic equipment |
CN112329693A (en) * | 2020-11-17 | 2021-02-05 | 汇纳科技股份有限公司 | Training method, recognition method, medium and device for gender and age recognition model |
CN112329693B (en) * | 2020-11-17 | 2024-01-19 | 汇纳科技股份有限公司 | Training method, identification method, medium and equipment for gender and age identification model |
Also Published As
Publication number | Publication date |
---|---|
CN109670437B (en) | 2021-05-07 |
Similar Documents
Publication | Publication Date | Title |
---|---|---|
CN109670437A (en) | Age prediction model training method, face-image recognition methods and device | |
WO2020098250A1 (en) | Character recognition method, server, and computer readable storage medium | |
CN109783730A (en) | Products Show method, apparatus, computer equipment and storage medium | |
CN109345374A (en) | Risk control method, device, computer equipment and storage medium | |
CN110163235A (en) | Training, image enchancing method, device and the storage medium of image enhancement model | |
CN111738243B (en) | Method, device and equipment for selecting face image and storage medium | |
CN111080628A (en) | Image tampering detection method and device, computer equipment and storage medium | |
CN108510402A (en) | Insurance kind information recommendation method, device, computer equipment and storage medium | |
CN109472213A (en) | Palm grain identification method, device, computer equipment and storage medium | |
CN109767320A (en) | Order checking method, device, computer equipment and storage medium | |
CN108921131B (en) | Method and device for generating face detection model and three-dimensional face image | |
CN112669342A (en) | Training method and device of image segmentation network, and image segmentation method and device | |
CN109886110A (en) | Micro- expression methods of marking, device, computer equipment and storage medium | |
Đukić et al. | A low-shot object counting network with iterative prototype adaptation | |
CN108304243A (en) | Interface creating method, device, computer equipment and storage medium | |
CN112668640A (en) | Text image quality evaluation method, device, equipment and medium | |
CN108182633A (en) | Loan data processing method, device, computer equipment and storage medium | |
CN110796039B (en) | Face flaw detection method and device, electronic equipment and storage medium | |
Rangrej et al. | Consistency driven sequential transformers attention model for partially observable scenes | |
CN113283388B (en) | Training method, device, equipment and storage medium of living body face detection model | |
CN113361567A (en) | Image processing method, image processing device, electronic equipment and storage medium | |
CN110162689A (en) | Information-pushing method, device, computer equipment and storage medium | |
CN113838076A (en) | Method and device for labeling object contour in target image and storage medium | |
CN114170481B (en) | Method, apparatus, storage medium, and program product for image processing | |
CN112509052B (en) | Method, device, computer equipment and storage medium for detecting macula fovea |
Legal Events
Date | Code | Title | Description |
---|---|---|---|
PB01 | Publication | ||
PB01 | Publication | ||
SE01 | Entry into force of request for substantive examination | ||
SE01 | Entry into force of request for substantive examination | ||
GR01 | Patent grant | ||
GR01 | Patent grant |