CN106651978A - Face image prediction method and system - Google Patents
Face image prediction method and system Download PDFInfo
- Publication number
- CN106651978A CN106651978A CN201610886084.2A CN201610886084A CN106651978A CN 106651978 A CN106651978 A CN 106651978A CN 201610886084 A CN201610886084 A CN 201610886084A CN 106651978 A CN106651978 A CN 106651978A
- Authority
- CN
- China
- Prior art keywords
- facial image
- face
- model
- forecast model
- node
- 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
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T11/00—2D [Two Dimensional] image generation
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/23—Clustering techniques
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
- G06V40/00—Recognition of biometric, human-related or animal-related patterns in image or video data
- G06V40/10—Human or animal bodies, e.g. vehicle occupants or pedestrians; Body parts, e.g. hands
- G06V40/16—Human faces, e.g. facial parts, sketches or expressions
- G06V40/168—Feature extraction; Face representation
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- 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
Landscapes
- Engineering & Computer Science (AREA)
- Theoretical Computer Science (AREA)
- General Physics & Mathematics (AREA)
- Physics & Mathematics (AREA)
- Data Mining & Analysis (AREA)
- Oral & Maxillofacial Surgery (AREA)
- Health & Medical Sciences (AREA)
- Multimedia (AREA)
- Human Computer Interaction (AREA)
- Computer Vision & Pattern Recognition (AREA)
- General Health & Medical Sciences (AREA)
- Bioinformatics & Computational Biology (AREA)
- General Engineering & Computer Science (AREA)
- Evolutionary Computation (AREA)
- Evolutionary Biology (AREA)
- Bioinformatics & Cheminformatics (AREA)
- Artificial Intelligence (AREA)
- Life Sciences & Earth Sciences (AREA)
- Processing Or Creating Images (AREA)
- Image Analysis (AREA)
Abstract
The invention discloses a face image prediction method and system. The face image prediction method comprises the steps of acquiring a face image to be predicted and a time point of face image prediction; extracting face attribute features from the face image; determining a corresponding face image prediction model by using the face attribute features; and inputting pixel points of the face image into the face image prediction model so as to acquire a predicted face image. By using the face image prediction method, the face image acquired by prediction can be enabled to be more correlated with the face image to be predicted, and the user experience is greatly improved.
Description
Technical field
The present invention relates to image processing field, and in particular to a kind of facial image Forecasting Methodology and system.
Background technology
With the continuous improvement and the continuous progress of science and technology of modern life level, people are for entertainment orientation, informationalized need
Ask also more and more diversified.The need of social management are aided in order to adapt to the growing material and cultural needs of people and high-tech
Will, predicted before a people by image processing techniques or the later equal demand of length is strong all the more, especially it is input into a use
Appearance of the photo prediction user in family in the corresponding time is widely used in terms of amusement, the such as 20 years old or 30 years old corresponding time or
20 Nian Hou etc., can be used for predicting the appearance when childhood of certain star or old age in the production of concrete application such as film, search for phase
Answer stand-in；The photo of user oneself corresponding time is and for example predicted in the application of mobile phone, increases the recreational of mobile phone application；
Certainly also there are corresponding demand, such as public safety field in other fields, the appearance to suspect after for many years is predicted,
The police are helped to solve a case etc..
Existing facial image Forecasting Methodology is generally based on human face segmentation technology, during prediction facial image, using three-dimensional
Method for reconstructing synthesizes the facial image of corresponding age bracket, and the three-dimensional reconstruction refers to be carried out after critical point detection to face, to input
Image carries out three-dimensional modeling, and face texture in image is attached on threedimensional model, and the key point to detecting carries out three-dimensional deformation,
Interpolation processing is carried out to portion three-dimensional picture pasting；Finally the wrinkle of the corresponding age bracket for previously generating is added to into the threedimensional model after processing
On, and be smoothed, so as to obtain the facial image of the corresponding age bracket rebuild, the wrinkle needs to collect a large amount of in advance
Data carry out model training, using the modeling that obtains of training out.However, because everyone face texture variations are deposited
In larger difference, existing method when the facial image of corresponding age bracket is synthesized, using the corresponding age bracket for previously generating
Wrinkle is added on the threedimensional model, and the wrinkle for previously generating is to the facial image that active user is input into and uncorrelated, adds
Plus the facial image that provides with user of the facial image after wrinkle often otherness is larger, especially add the Edge difference of wrinkle
Become apparent from, so that the facial image after synthesis seems stranger, the sense of reality is poor, and user experience is relatively low.
The content of the invention
The embodiment of the present invention provides a kind of facial image Forecasting Methodology and system, to improve the facial image that prediction is obtained
The sense of reality, lifts user experience.
For this purpose, the present invention provides following technical scheme：
A kind of facial image Forecasting Methodology, including：
The time point of acquisition facial image to be predicted and prediction facial image；
Face character feature is extracted from the facial image；
Determine corresponding facial image forecast model using the face character feature；
The pixel of the facial image is input into into the face forecast model, the facial image predicted is obtained.
Preferably, the facial image forecast model includes time following current model and/or time inverse flow model, the time
Following current model is used to predict the appearance situation that face is following that the time inverse flow model to be used to predict the past appearance of face；
Methods described also includes building facial image forecast model in the following manner：
A large amount of facial images are collected, time conversion database is built；
Face character feature is extracted in the facial image converted in database from the time；
To the time convert database in facial image carry out it is regular, obtain it is regular after facial image；
According to extract face character feature to it is described it is regular after facial image cluster, the face after being clustered
Image；
According to the facial image after cluster, face forecast model is built.
Preferably, the face character feature include it is following any one or more：Sex, expression, whether wear glasses,
Domain, occupation；
Extracting face character feature in the facial image converted in database from the time includes：
Face datection and facial modeling are carried out to facial image, the position of the local feature region of face in image is obtained
Put；
It is special according to the face character that each facial image is extracted in the position of each local feature region and the disaggregated model of training in advance
Levy.
Preferably, it is described to the time convert database in facial image carry out it is regular including：
To the time convert database in facial image in face coordinate and yardstick carry out it is regular.
Preferably, it is described according to the face character feature extracted to it is described it is regular after facial image carry out cluster and include：
(1) a kind of face character feature is selected as the root node of decision tree, according to selected face character feature
Value determines each bar side of the root node, and facial image is divided into into multiclass；
(2) using the face character feature extracted, calculate per remaining each face attributive character value in class facial image
Minimum variance；
(3) judge the minimum variance whether more than setting value；If it is, execution step (4)；Otherwise execution step (5)；
(4) using the node belonging to the corresponding class of the minimum variance as leaf node, do not continue to divide；Then perform
Step (6)；
(5) using the corresponding attributive character of minimum variance described in every class as every class facial image upper layer node, and root
The side of the upper layer node is obtained according to the value of the upper layer node, will continue to be divided into multiclass per class facial image；
(6) judge whether that also attributive character is not added in decision tree；If it has, execution step (2)；Otherwise, perform
Step (7)；
(7) the facial image quantity under each leaf node is counted, if the facial image quantity in leaf node is less than
The amount threshold of setting, then delete the leaf node and its brotgher of node, and by the leaf node and its people in the brotgher of node
Face image is added to the father node of the leaf node, and decision tree builds and completes.
Preferably, the facial image according to after cluster, building face forecast model includes：
For each leaf node in the decision tree, build to should leaf node face forecast model, specifically
Including：
Facial image in time conversion database is ranked up, the facial image of same people is successively arranged according to the age
Sequence；
Face forecast model is initialized using the data after sequence, obtains initialized face forecast model；
Incremental training is carried out to the initialized face forecast model, final face forecast model is obtained.
Preferably, it is described to determine that corresponding facial image forecast model includes using the face character feature：
According to the face character feature, the decision tree is traveled through, find corresponding leaf node；
Obtain the corresponding face forecast model of the leaf node.
Preferably, methods described also includes：
According to the face character feature, the facial image of the prediction is reduced, the face figure after being reduced
Picture.
Preferably, methods described also includes：
By the back of the body of the facial image after the facial image of the prediction or the reduction and the facial image to be predicted
Scape is synthesized, the facial image after being synthesized.
Preferably, methods described also includes：
Facial image after the facial image of the prediction or reduction or the facial image after synthesis are fed back to into user.
A kind of facial image forecasting system, including：
Receiver module, for obtaining the time point of facial image to be predicted and prediction facial image；
Characteristic extracting module, for extracting face character feature from the facial image；
Model selection module, for determining corresponding facial image forecast model using the face character feature；
Prediction module, by the pixel of the facial image face forecast model is input into, and obtains the face figure predicted
Picture.
Preferably, the facial image forecast model includes time following current model and/or time inverse flow model, the time
Following current model is used to predict the appearance situation that face is following that the time inverse flow model to be used to predict the past appearance of face；
The system also includes that forecast model builds module, and the forecast model builds module to be included：
Image collection unit, for collecting a large amount of facial images, builds time conversion database；
Feature extraction unit, for extracting face character feature in the facial image from time conversion database；
Regular unit, for the time convert database in facial image carry out it is regular, obtain it is regular after people
Face image；
Cluster cell, for according to extract face character feature to it is described it is regular after facial image cluster, obtain
Facial image to after cluster；
Model construction unit, for according to the facial image after cluster, building face forecast model.
Preferably, the face character feature include it is following any one or more：Sex, expression, whether wear glasses,
Domain, occupation；
The feature extraction unit includes：
Locator unit, for carrying out Face datection and facial modeling to facial image, obtains face in image
Local feature region position；
Subelement is extracted, for extracting each face figure according to the disaggregated model of the position of each local feature region and training in advance
The face character feature of picture.
Preferably, the regular unit, specifically for face in the facial image in time conversion database
Coordinate and yardstick carry out regular.
Preferably, the cluster cell specifically in the following manner to it is described it is regular after facial image cluster：
(1) a kind of face character feature is selected as the root node of decision tree, according to selected face character feature
Value determines each bar side of the root node, and facial image is divided into into multiclass；
(2) using the face character feature extracted, calculate per remaining each face attributive character value in class facial image
Minimum variance；
(3) judge the minimum variance whether more than setting value；If it is, execution step (4)；Otherwise execution step (5)；
(4) using the node belonging to the corresponding class of the minimum variance as leaf node, do not continue to divide；Then perform
Step (6)；
(5) using the corresponding attributive character of minimum variance described in every class as every class facial image upper layer node, and root
The side of the upper layer node is obtained according to the value of the upper layer node, will continue to be divided into multiclass per class facial image；
(6) judge whether that also attributive character is not added in decision tree；If it has, execution step (2)；Otherwise, perform
Step (7)；
(7) the facial image quantity under each leaf node is counted, if the facial image quantity in leaf node is less than
The amount threshold of setting, then delete the leaf node and its brotgher of node, and by the leaf node and its people in the brotgher of node
Face image is added to the father node of the leaf node, and decision tree builds and completes.
Preferably, the model construction unit, specifically for for each leaf node in the decision tree, it is right to build
Should leaf node face forecast model；The model construction unit is specifically included：
Sequence subelement, for being ranked up to facial image in time conversion database, the facial image of same people is pressed
Successively it is ranked up according to the age；
Initialization subelement, for being initialized to face forecast model using the data after sequence, is initialized
Face forecast model；
Incremental training subelement, for carrying out incremental training to the initialized face forecast model, obtains final
Face forecast model.
Preferably, the Model selection module includes：
Traversal Unit, for according to the face character feature, traveling through the decision tree, finds corresponding leaf node；
Model acquiring unit, for obtaining the corresponding face forecast model of the leaf node.
Preferably, the system also includes：
Recovery module, for according to the face character feature, reducing to the facial image of the prediction, is gone back
Facial image after original.
Preferably, the system also includes：
Synthesis module, for the facial image after the facial image of the prediction or the reduction is to be predicted with described
The background of facial image is synthesized, the facial image after being synthesized.
Preferably, the system also includes：
Feedback module, for by the facial image of the prediction or reduction after facial image or synthesis after face figure
As feeding back to user.
Facial image Forecasting Methodology provided in an embodiment of the present invention and system, build in advance facial image forecast model, institute
Stating facial image forecast model can include：Time following current model and/or time inverse flow model, wherein, the time following current mould
Type is used to predict the appearance situation that face is following that the time inverse flow model to be used to predict the past appearance of face.To be predicted
Facial image, extract the face character feature related to face appearance or face change, then the face character of utilization extraction
Feature determines corresponding face forecast model, then the pixel of the facial image is input into into the face forecast model, obtains
To the facial image of the time point so that the correlation of the facial image that obtains of prediction and facial image to be predicted compared with
Greatly, the sense of reality is stronger, and to user is a kind of sense is substituted into, and substantially increases Consumer's Experience.
Description of the drawings
In order to be illustrated more clearly that the embodiment of the present application or technical scheme of the prior art, below will be to institute in embodiment
The accompanying drawing that needs are used is briefly described, it should be apparent that, drawings in the following description are only described in the present invention
A little embodiments, for those of ordinary skill in the art, can be with according to these other accompanying drawings of accompanying drawings acquisition.
Fig. 1 is the flow chart that facial image forecast model is built in the embodiment of the present invention；
Fig. 2 is the flow chart clustered to facial image using decision tree in the embodiment of the present invention；
Fig. 3 is that the corresponding face forecast model of each leaf node builds flow chart in the embodiment of the present invention；
Fig. 4 is the flow chart of facial image Forecasting Methodology provided in an embodiment of the present invention；
Fig. 5 is a kind of structural representation of embodiment of the present invention facial image forecasting system；
Fig. 6 is the structural representation that forecast model builds module in the embodiment of the present invention；
Fig. 7 is another kind of structural representation of embodiment of the present invention facial image forecasting system.
Specific embodiment
In order that those skilled in the art more fully understand the scheme of the embodiment of the present invention, below in conjunction with the accompanying drawings and implement
Mode is described in further detail to the embodiment of the present invention.
Facial image Forecasting Methodology provided in an embodiment of the present invention and system, build in advance facial image forecast model, tool
When body builds, the face character feature that great amount of images is extracted is clustered using decision tree, according to falling decision tree after cluster
In each leaf node facial image attributive character build facial image forecast model.The facial image forecast model can be with
Including：Time following current model and/or time inverse flow model, wherein, the time following current model is used to predict the length that face is following
Phase situation, the time inverse flow model is used to predict the past appearance of face.To facial image to be predicted, face character is extracted
Feature, then obtains corresponding face forecast model using the face character feature extracted, using the face forecast model prediction
The facial image of specified time point.
The building process of facial image forecast model in the embodiment of the present invention is described in detail first below.
As shown in figure 1, being the flow chart that facial image forecast model is built in the embodiment of the present invention, comprise the following steps：
Step 101, collects a large amount of facial images, builds time conversion database.
Specifically, multiple facial images of same facial image in all ages and classes are collected, multiple not the same years of different faces
The facial image in age constitutes time conversion database.It should be noted that when time conversion database is built, people can be passed through
The facial image of large area disappearance is abandoned in the facial image that face detection tech will be collected, in order to avoid affect facial image forecast model
The accuracy of parameter.
Step 102, extracts face character feature in the facial image converted in database from the time.
First, Face datection and facial modeling are carried out to facial image, to obtain image in face local it is special
Levy position a little；Then, the people that each facial image is extracted according to the position of each local feature region and the disaggregated model of training in advance
Face attributive character.
Wherein, the Face datection is primarily to find face position in image, specific practice and prior art
It is identical, such as by collecting a large amount of images comprising face in advance, extract SIFT (Scale-invariant feature
Transform, Scale invariant features transform) feature, training face and non-face disaggregated model, using the disaggregated model pair
Facial image carries out Face datection in database.The facial modeling primarily on the basis of Face datection,
The local location of face, such as eyes, eyebrow, nose, face, face's outline are further determined that, it is main logical when specifically positioning
The position constraint crossed between the textural characteristics of face and each characteristic point is combined, and can such as adopt ASM (Active Shape
Model, active shape model) or AAM (Active Appreance Model, active appearance models) algorithm carry out face characteristic
Point location, obtains the local feature region position of face in image.
In embodiments of the present invention, the face character feature refers mainly to the attribute related to face appearance or face change
Feature, the face character feature of extraction can include it is following any one or more：Age, sex, human face expression, whether hyperphoria with fixed eyeballs
Mirror, region, occupation etc..The value of each attributive character is as follows：
Age：To fix the time as span, multiple intervals, such as with 5 years as span, the year of 0 to 99 years old will be divided into the age
Age can be divided into 20 intervals, when extracting age characteristics, directly give the age range belonging to facial image；
Sex：The male sex, women；
Expression：Human face expression can be greatly classified into pleasure, anger, sorrow, happiness；
Whether wear glasses：Be, it is no；
Region：Region can be divided according to province, prediction obtains the province in image belonging to face；
Occupation：Several different occupations can be divided, such as one kind is divided into infant, student, peasant, clerical workforce.
Can be in advance one disaggregated model of every kind of features training, using the disaggregated model to every kind of during concrete extraction
The attribute value of feature is predicted.The disaggregated model can be described using deep neural network, concrete extracting method with it is existing
There is technology identical, will not be described in detail herein.
Step 103, to the time convert database in facial image carry out it is regular, obtain it is regular after face figure
Picture.
It is described it is regular mainly include the regular of the coordinate to face in facial image and yardstick, such as according to prenasale centered on
Point, eyes line is x-axis, as y-axis, face coordinate and yardstick is carried out regular with through the prenasale straight line vertical with x-axis；Together
When for the presence facial image that blocks or wear glasses, predict unobstructed, glasses-free image using Image Smoothing Skill, will
Image after prediction replaces image original in database, the smoothing technique to be such as based on the Image Reconstruction of deep neural network,
Or the method for solving based on sparse expression.
Step 104, according to extract face character feature to it is described it is regular after facial image cluster, clustered
Facial image afterwards.
In embodiments of the present invention, it is possible to use decision tree to it is described it is regular after facial image cluster, decision tree
Node represent every kind of face character feature, each edge represents the concrete value of face character feature, and concrete cluster process will be
It is described in detail below.It is of course also possible to using other clustering methods more of the prior art to it is described it is regular after face figure
As being clustered, this embodiment of the present invention is not limited.
Step 105, according to the facial image after cluster, builds face forecast model.
After the completion of facial image cluster, complete decision tree is obtained, every facial image is according to its face character feature
Value condition, in falling the leaf node of decision tree, falls the shared same model of facial image of same leaf node, different leaves
The facial image of child node is respectively trained different face forecast models.The face forecast model specifically can be using depth god
The structure of Jing networks carries out model training, and using the method for incremental training during model training, concrete building process will later
Describe in detail.
As shown in Fig. 2 be the flow chart clustered to facial image using decision tree in the embodiment of the present invention, including with
Lower step：
Step 201, selects a kind of face character feature as the root node of decision tree, special according to selected face character
The value levied determines each bar side of the root node, and facial image is divided into into multiclass.
The node of decision tree represents every kind of attributive character, and each edge represents the concrete value of attributive character.Specifically, can be with
Using the stronger face character feature of distinction as the root node of decision tree, root section is obtained according to the value of the face character feature
Each bar side of point, so as to facial image is divided into into multiple classes.Such as, it is male or female according to the value of gender attribute feature, will
Facial image is divided into two classes, i.e. respectively man's face image and woman's face image.
Step 202, using the face character feature extracted, calculates per remaining each face attributive character in class facial image
The minimum variance of value.
Specifically, for every class facial image, first the remaining face character spy of its each facial image is calculated for everyone
The variance of value of collecting；Then for calculating the variance and (each of each remaining face character feature value per everyone in class
Attribute one variance of correspondence and)；The last variance for selecting these attribute values again and in a minimum variance and as in such
Attribute value minimum variance.
For example, using gender attribute feature as root node, then remaining face character feature is respectively the age, expresses one's feelings, is
It is no wear glasses, region and occupation.For everyone multiple different facial images, remaining each face of each image is calculated respectively
The variance of attributive character value, that is, obtain the variance of the remaining face character feature value of everyone different facial images.Example
Such as, correspondence Zhang San, there is 4 images, then calculate respectively this 4 images age, expression, whether wear glasses, region and it is professional this
The variance (for convenience, being referred to as property variance) of value in 5 attributive character；Equally, correspondence Li Si, there is 3 figures
Picture, then calculate respectively this 3 images age, expression, whether wear glasses, value in region and professional this 5 attributive character
Variance.Then, in for every class, proprietary every kind of face character feature, by the corresponding of the proprietary facial image for obtaining
The variance summation of attributive character value, that is, obtain in every class the variance of proprietary remaining each face attributive character value and.
It should be noted that considering that variance is larger, then the value deviation of corresponding face character feature is larger, i.e. feature
During extraction, the deviation of prediction is larger.Therefore, if being calculated someone property variance more than given threshold, should calculating
During the minimum variance of the affiliated class of people, the value of the corresponding face attributive character of the people can not be considered.
Whether step 203, judge the minimum variance more than setting value.If it is, execution step 204；Otherwise, perform
Step 205.
Step 204, using the node belonging to the corresponding class of the minimum variance as leaf node, does not continue to divide；So
Execution step 206 afterwards.
Step 205, using the corresponding attributive character of minimum variance described in every class as every class facial image upper layer node,
The corresponding side of every class upper layer node is obtained according to the value of every class upper layer node, will continue to be divided into multiclass per class facial image.
Step 206, judges whether that also attributive character is not added in decision tree；If it is, execution step 202；Otherwise,
Execution step 207.
Step 207, counts the facial image quantity under each leaf node, if the facial image quantity in leaf node
Less than the amount threshold of setting, then the leaf node and its brotgher of node are deleted, and by the leaf node and its brotgher of node
Facial image be added in the father node of the leaf node, decision tree build complete, while facial image cluster terminate.
As shown in figure 3, be the corresponding face forecast model building process of each leaf node in the embodiment of the present invention, including
Following steps：
Step 301, is ranked up to facial image in time conversion database, and the facial image of same people is according to age elder generation
After be ranked up.
Step 302, is initialized using the data after sequence to face forecast model, obtains initialized face prediction
Model.
For the training of time following current model, using minimum age interval facial image as input, such as using 5 years as one
Age range, using the facial image in next age range as output, carries out the training of model, obtains with short-term prediction weight
The time following current model of structure function；
For the training of time inverse flow model, using minimum age interval facial image as model output, by next year
Facial image in age is interval carries out the training of model as input, obtains having the time adverse current for pushing back recombination function in short-term
Model.
Step 303, to the initialized face forecast model incremental training is carried out, and obtains final face prediction mould
Type.
Increase the facial image of age range successively, time following current model in step 302 and time inverse flow model are carried out
Incremental training, it is specifically as described below：
For time following current model, first using the facial image in minimum age interval as input, prediction obtains next year
Age interval facial image, then the facial image of next age range that prediction is obtained, used as input, prediction obtains the 3rd year
Age interval facial image, minimizes the facial image of the 3rd age range that prediction is obtained and the people of true 3rd age range
The error of face image, training is updated to time following current model parameter, and the facial image of true 3rd age range is
Facial image in time conversion database；Increase the age range of facial image successively, increase one or more area every time at ages
Between, time following current model is trained, often increase an age range, time following current model is once updated, until
All age range increases terminate, and obtain final time following current model；
For time inverse flow model, the facial image after age range will be increased first as input, such as by the 3rd age
Interval facial image pushes back the facial image of the previous age range that reconstruct is obtained as input using time inverse flow model,
The input of the previous age range of reconstruct as time inverse flow model will be pushed back again, push back the face figure in reconstruct minimum age interval
Picture, minimum pushes back the mistake between the minimum age interval facial image of reconstruct and the facial image that true minimum age is interval
Difference, training is updated to time inverse flow model parameter, and the facial image in the true minimum age interval is to change number in the time
According to facial image in storehouse, successively the increase age range of facial image, increases every time one or more age ranges, inverse to the time
Flow model is trained, and often increases an age range, time inverse flow model is once updated, until all age ranges
Increase terminates, and obtains final time inverse flow model.
Using above-mentioned face forecast model, flow chart such as Fig. 4 of facial image Forecasting Methodology provided in an embodiment of the present invention
It is shown, comprise the following steps：
The time point of step 401, acquisition facial image to be predicted and prediction facial image.
Described image can be the image that user directly uploads, or the figure for directly being obtained by camera shooting
Picture.
The time point of the prediction facial image refers to that user wants to predict the time point of the facial image for obtaining, and specifically can make
Represented with the age, the age that such as current face's image prediction is obtained is 20 years old, facial image when user expects 30 years old.It is described
Time point can be by user input or selection.
Step 402, extracts face character feature from the facial image.
The face character feature can include it is following any one or more：Sex, expression, whether wear glasses, region,
Occupation.When the extracting method of face character feature may be referred to above facial image model construction, a large amount of facial images are extracted
The method of face character feature, i.e. first, to facial image Face datection and facial modeling are carried out, to obtain image
The position of the local feature region of middle face；Then, extracted according to the position of each local feature region and the disaggregated model of training in advance
The face character feature of each facial image.Certainly, if user provide image on not only have face, go back band have powerful connections and with
The unrelated additional information of face (such as earrings, glasses etc.), then need first to remove background and these irrelevant informations.
Step 403, using the face character feature corresponding facial image forecast model is determined.
Specifically, decision tree can be traveled through according to the face character feature, finds these face character features corresponding
Leaf node position, then obtains the corresponding face forecast model of the leaf node.
Step 404, by the pixel of the facial image face forecast model is input into, and obtains the face figure predicted
Picture.
If the time point is carried out pre- after the current time point for receiving facial image using time following current model
Survey, if the time point is before the time point of present image, image is carried out using time inverse flow model push back reconstruct.
If user input is after 20 years, then after the corresponding time point of current face's image, if user input is before 20 years, current
Before the corresponding time point of facial image, or user directly inputs current age and the age after prediction.Such as current face
The age of image is 30 years old, the corresponding facial image of random time point when user expects 30 to 99 years old, then fairing when using
Flow model is predicted to present image, the corresponding facial image of random time point when user expects 0 to 29 years old, then use
Time inverse flow model carries out pushing back reconstruct to present image.
During concrete prediction, can determine the affiliated age range of current face's image and use according to the pre- age range for dividing
Expect the age range of facial image in family；Then current face's image is predicted every time as the input of face forecast model
Or the facial image of one age range of reconstruct, subsequently prediction or the facial image for reconstructing are continued to predict or reconstructed as input
The facial image of next age range, until the conceivable facial image place age range of user；Finally will prediction or reconstruct
Facial image as ultimately generating facial image.
After the facial image predicted is obtained, the facial image can be fed directly to user.
It is previously noted that in the facial image to be predicted of user's offer in addition to comprising face information, it is also possible to can include
The additional information unrelated with face, therefore, for this kind of facial image, in addition it is also necessary to according to the face character feature extracted, to pre-
Whether the facial image of survey carries out the reduction of respective attributes feature, such as, express one's feelings, wear glasses.Specifically, it is possible to use side
Edge infomation detection goes out to receive the adjunct such as glasses in image, and the eye areas pixel is directly substituted into into the facial image of generation
On correspondence position；For expressive features, it is possible to use the deformation of the facial image facial zone that the control of face key point is generated, intend
Splice grafting receives the corresponding human face expression of image, and concrete grammar is same as the prior art, will not be described in detail herein.By reduction treatment, can
So that the facial image after reduction has more the sense of reality, the facial image after reduction is fed back to into user, can further improve use
Experience at family.
In addition, if also including background in the facial image to be predicted of user's offer, then the face that will be predicted is also needed to
Image is synthesized with the background of the facial image to be predicted, generates the figure with facial image same background to be predicted
Picture.Specifically, the facial image first to predicting carries out that yardstick is regular, i.e., according to the face orientation information of facial image to be predicted
The facial image of the rear prediction is zoomed in and out, the facial image after being scaled；Then by the facial image after scaling with
The background of facial image to be predicted enters row interpolation, and the facial image after being synthesized, concrete grammar is same as the prior art,
This is no longer described in detail.Processed by synthesis, the facial image picture after synthesis can be made more rich, the facial image after synthesis is anti-
Feed user, can further improve Consumer's Experience.
Facial image Forecasting Methodology provided in an embodiment of the present invention, builds in advance facial image forecast model, the face
Image prediction model can include：Time following current model and/or time inverse flow model, wherein, the time following current model is used for
Following appearance situation of prediction face, the time inverse flow model is used to predict the past appearance of face.To face to be predicted
Image, extracts the face character feature related to face appearance or face change, then true using the face character feature extracted
Fixed corresponding face forecast model, is then input into the face forecast model by the pixel of the facial image, obtains described
The facial image of time point, so that the facial image that prediction is obtained is larger with the correlation of facial image to be predicted, truly
Sense is relatively strong, and to user is a kind of sense is substituted into, and substantially increases Consumer's Experience.
Correspondingly, the embodiment of the present invention also provides a kind of facial image forecasting system, as shown in figure 5, being enforcement of the present invention
A kind of structural representation of example facial image forecasting system, including following module：
Receiver module 501, for obtaining the time point of facial image to be predicted and prediction facial image；
Characteristic extracting module 502, for extracting face character feature from the facial image；
Model selection module 503, for determining corresponding facial image forecast model using the face character feature；
Prediction module 504, by the pixel of the facial image face forecast model is input into, and obtains the face predicted
Image.
Above-mentioned facial image to be predicted and prediction facial image time point can user directly input, the people
Face attributive character can include it is following any one or more：Sex, expression, whether wear glasses, region, occupation.Correspondingly, it is special
To levy extraction module 502 first can carry out Face datection and facial modeling to facial image, to obtain image in face
The position of local feature region；Then, each face figure is extracted according to the position of each local feature region and the disaggregated model of training in advance
The face character feature of picture.
The decision tree that the Model selection module 503 builds when can be trained according to facial image forecast model is obtaining phase
The facial image forecast model answered, the module can specifically include following two units：
Traversal Unit, for according to the face character feature, traveling through the decision tree, finds corresponding leaf node；
Model acquiring unit, for obtaining the corresponding face forecast model of the leaf node.
The facial image forecasting system of the embodiment of the present invention carries out face using the facial image forecast model for building in advance
Image prediction, the facial image forecast model can include：Time following current model and/or time inverse flow model, wherein, it is described
Time following current model is used to predict the appearance situation that face is following that the time inverse flow model to be used to predict the past length of face
Phase.The facial image forecast model can be built module to build by corresponding forecast model, be belonged to corresponding to different faces
Property feature, needs to build corresponding facial image forecast model.In actual applications, the forecast model builds module and can lead to
The method of cluster is crossed building the face forecast model for various different face attributive character.And, the forecast model structure
Modeling block can be used as a part for present system, it is also possible to independently of present system, i.e., as an independent entity,
This embodiment of the present invention is not limited.
As shown in fig. 6, be the structural representation that forecast model builds module in the embodiment of the present invention, including following list
Unit：
Image collection unit 61, for collecting a large amount of facial images, builds time conversion database, specifically, Ke Yishou
Collect multiple facial images of same facial image in all ages and classes, when the facial image of multiple all ages and classes of different faces is constituted
Light converts database；
Feature extraction unit 62, it is special for extracting face character in the facial image from time conversion database
Levy；
Regular unit 63, for the time convert database in facial image carry out it is regular, obtain it is regular after
Facial image, specifically can be carried out regular to the coordinate of face in the facial image in time conversion database and yardstick；
Cluster cell 64, for according to extract face character feature to it is described it is regular after facial image cluster,
Facial image after being clustered；
Model construction unit 65, for according to the facial image after cluster, building face forecast model.
Features described above extraction unit 62 can include：Locator unit and extraction subelement, wherein, locator unit is used for
Face datection and facial modeling are carried out to facial image, the position of the local feature region of face in image is obtained；Extract
Subelement is used to extract the face character of each facial image according to the position of each local feature region and the disaggregated model of training in advance
Feature.
Above-mentioned cluster cell 64 specifically can be clustered according to mode shown in Fig. 2 to the facial image after regular.
Above-mentioned model construction unit 65 is needed for each leaf node in the decision tree, is built to should leaf section
The face forecast model of point；A kind of concrete structure of the model construction unit 65 can include following subelement：
Sequence subelement, for being ranked up to facial image in time conversion database, the facial image of same people is pressed
Successively it is ranked up according to the age；
Initialization subelement, for being initialized to face forecast model using the data after sequence, is initialized
Face forecast model；
Incremental training subelement, for carrying out incremental training to the initialized face forecast model, obtains final
Face forecast model.
As shown in fig. 7, in another embodiment of the present inventor's face image forecasting system, the system may also include：
Recovery module 701, for according to the face character feature, reducing to the facial image of the prediction, obtains
Facial image to after reduction.
Further, the system may also include：
Synthesis module 702, for facial image after the facial image of the prediction or the reduction to be treated into pre- with described
The background of the facial image of survey is synthesized, the facial image after being synthesized.
Further, the system may also include：
Feedback module 703, for by the facial image of the prediction or reduction after facial image or synthesis after people
Face image feeds back to user.
It should be noted that in actual applications, according in the image to be predicted that user provides whether band have powerful connections and
Whether some additional informations unrelated with face etc. are also included in image, above-mentioned recovery module 701 and synthesis module 702 can be with roots
Select according to needing.And, the facial image for finally giving can also be fed back to use by feedback module 703 in several ways
Family, such as, directly represent on screen or that image is stored in into corresponding file be medium, and this embodiment of the present invention is not done
Limit.
Facial image forecasting system provided in an embodiment of the present invention, builds in advance facial image forecast model, the face
Image prediction model can include：Time following current model and/or time inverse flow model, wherein, the time following current model is used for
Following appearance situation of prediction face, the time inverse flow model is used to predict the past appearance of face.To face to be predicted
Image, extracts the face character feature related to face appearance or face change, then true using the face character feature extracted
Fixed corresponding face forecast model, is then input into the face forecast model by the pixel of the facial image, obtains described
The facial image of time point, so that the facial image that prediction is obtained is larger with the correlation of facial image to be predicted, truly
Sense is relatively strong, and to user is a kind of sense is substituted into, and substantially increases Consumer's Experience.
Each embodiment in this specification is described by the way of progressive, identical similar portion between each embodiment
Divide mutually referring to what each embodiment was stressed is the difference with other embodiment.Especially for system reality
For applying example, because it is substantially similar to embodiment of the method, so describing fairly simple, related part is referring to embodiment of the method
Part explanation.System embodiment described above is only schematic, wherein described illustrate as separating component
Unit can be or may not be physically separate, can be as the part that unit shows or may not be
Physical location, you can be located at a place, or can also be distributed on multiple NEs.Can be according to the actual needs
Select some or all of module therein to realize the purpose of this embodiment scheme.Those of ordinary skill in the art are not paying
In the case of creative work, you can to understand and implement.
The embodiment of the present invention is described in detail above, specific embodiment used herein is carried out to the present invention
Illustrate, the explanation of above example is only intended to help and understands the method for the present invention and system；Simultaneously for the one of this area
As technical staff, according to the present invention thought, will change in specific embodiments and applications, to sum up institute
State, this specification content should not be construed as limiting the invention.
Claims (20)
1. a kind of facial image Forecasting Methodology, it is characterised in that include：
The time point of acquisition facial image to be predicted and prediction facial image；
Face character feature is extracted from the facial image；
Determine corresponding facial image forecast model using the face character feature；
The pixel of the facial image is input into into the face forecast model, the facial image predicted is obtained.
2. method according to claim 1, it is characterised in that the facial image forecast model includes time following current model
And/or time inverse flow model, the time following current model appearance situation following for predicting face, the time inverse flow model
For predicting the past appearance of face；
Methods described also includes building facial image forecast model in the following manner：
A large amount of facial images are collected, time conversion database is built；
Face character feature is extracted in the facial image converted in database from the time；
To the time convert database in facial image carry out it is regular, obtain it is regular after facial image；
According to extract face character feature to it is described it is regular after facial image cluster, the face figure after being clustered
Picture；
According to the facial image after cluster, face forecast model is built.
3. method according to claim 2, it is characterised in that the face character feature include it is following any one or it is many
Kind：Sex, expression, whether wear glasses, region, occupation；
Extracting face character feature in the facial image converted in database from the time includes：
Face datection and facial modeling are carried out to facial image, the position of the local feature region of face in image is obtained；
The face character feature of each facial image is extracted according to the position of each local feature region and the disaggregated model of training in advance.
4. method according to claim 2, it is characterised in that the facial image time converted in database
Carry out it is regular including：
To the time convert database in facial image in face coordinate and yardstick carry out it is regular.
5. method according to claim 2, it is characterised in that it is described according to the face character feature extracted to described regular
Facial image afterwards carries out cluster to be included：
(1) a kind of face character feature is selected as the root node of decision tree, according to the value of selected face character feature
Determine each bar side of the root node, and facial image is divided into into multiclass；
(2) using the face character feature extracted, calculate per remaining each face attributive character value in class facial image most
Little variance；
(3) judge the minimum variance whether more than setting value；If it is, execution step (4)；Otherwise execution step (5)；
(4) using the node belonging to the corresponding class of the minimum variance as leaf node, do not continue to divide；Then execution step
(6)；
(5) using the corresponding attributive character of minimum variance described in every class as every class facial image upper layer node, and according to institute
The value for stating upper layer node obtains the side of the upper layer node, will continue to be divided into multiclass per class facial image；
(6) judge whether that also attributive character is not added in decision tree；If it has, execution step (2)；Otherwise, execution step
(7)；
(7) the facial image quantity under each leaf node is counted, if the facial image quantity in leaf node is less than setting
Amount threshold, then delete the leaf node and its brotgher of node, and by the leaf node and its face figure in the brotgher of node
Father node as being added to the leaf node, decision tree builds and completes.
6. method according to claim 5, it is characterised in that the facial image according to after cluster, builds face pre-
Surveying model includes：
For each leaf node in the decision tree, build to should leaf node face forecast model, specifically include：
Facial image in time conversion database is ranked up, the facial image of same people is successively ranked up according to the age；
Face forecast model is initialized using the data after sequence, obtains initialized face forecast model；
Incremental training is carried out to the initialized face forecast model, final face forecast model is obtained.
7. method according to claim 5, it is characterised in that described to determine corresponding people using the face character feature
Face image forecast model includes：
According to the face character feature, the decision tree is traveled through, find corresponding leaf node；
Obtain the corresponding face forecast model of the leaf node.
8. the method according to any one of claim 1-7, it is characterised in that methods described also includes：
According to the face character feature, the facial image of the prediction is reduced, the facial image after being reduced.
9. method according to claim 8, it is characterised in that methods described also includes：
Facial image after the facial image of the prediction or the reduction is entered with the background of the facial image to be predicted
Row synthesis, the facial image after being synthesized.
10. method according to claim 9, it is characterised in that methods described also includes：
Facial image after the facial image of the prediction or reduction or the facial image after synthesis are fed back to into user.
11. a kind of facial image forecasting systems, it is characterised in that include：
Receiver module, for obtaining the time point of facial image to be predicted and prediction facial image；
Characteristic extracting module, for extracting face character feature from the facial image；
Model selection module, for determining corresponding facial image forecast model using the face character feature；
Prediction module, by the pixel of the facial image face forecast model is input into, and obtains the facial image predicted.
12. systems according to claim 11, it is characterised in that the facial image forecast model includes time following current mould
Type and/or time inverse flow model, the time following current model is used to predict the appearance situation that face is following, the time adverse current mould
Type is used to predict the past appearance of face；
The system also includes that forecast model builds module, and the forecast model builds module to be included：
Image collection unit, for collecting a large amount of facial images, builds time conversion database；
Feature extraction unit, for extracting face character feature in the facial image from time conversion database；
Regular unit, for the time convert database in facial image carry out it is regular, obtain it is regular after face figure
Picture；
Cluster cell, for according to extract face character feature to it is described it is regular after facial image cluster, gathered
Facial image after class；
Model construction unit, for according to the facial image after cluster, building face forecast model.
13. systems according to claim 12, it is characterised in that the face character feature include it is following any one or
It is various：Sex, expression, whether wear glasses, region, occupation；
The feature extraction unit includes：
Locator unit, for carrying out Face datection and facial modeling to facial image, obtains the office of face in image
The position of portion's characteristic point；
Subelement is extracted, for extracting each facial image according to the disaggregated model of the position of each local feature region and training in advance
Face character feature.
14. systems according to claim 12, it is characterised in that
The regular unit, enters specifically for the coordinate and yardstick to face in the facial image in time conversion database
Professional etiquette is whole.
15. systems according to claim 12, it is characterised in that the cluster cell is specifically in the following manner to institute
State it is regular after facial image clustered：
(1) a kind of face character feature is selected as the root node of decision tree, according to the value of selected face character feature
Determine each bar side of the root node, and facial image is divided into into multiclass；
(2) using the face character feature extracted, calculate per remaining each face attributive character value in class facial image most
Little variance；
(3) judge the minimum variance whether more than setting value；If it is, execution step (4)；Otherwise execution step (5)；
(4) using the node belonging to the corresponding class of the minimum variance as leaf node, do not continue to divide；Then execution step
(6)；
(5) using the corresponding attributive character of minimum variance described in every class as every class facial image upper layer node, and according to institute
The value for stating upper layer node obtains the side of the upper layer node, will continue to be divided into multiclass per class facial image；
(6) judge whether that also attributive character is not added in decision tree；If it has, execution step (2)；Otherwise, execution step
(7)；
(7) the facial image quantity under each leaf node is counted, if the facial image quantity in leaf node is less than setting
Amount threshold, then delete the leaf node and its brotgher of node, and by the leaf node and its face figure in the brotgher of node
Father node as being added to the leaf node, decision tree builds and completes.
16. systems according to claim 15, it is characterised in that
The model construction unit, specifically for for each leaf node in the decision tree, building to should leaf section
The face forecast model of point；The model construction unit is specifically included：
Sequence subelement, for being ranked up to facial image in time conversion database, the facial image of same people is according to year
Age is successively ranked up；
Initialization subelement, for initializing to face forecast model using the data after sequence, obtains initialized people
Face forecast model；
Incremental training subelement, for carrying out incremental training to the initialized face forecast model, obtains final face
Forecast model.
17. systems according to claim 15, it is characterised in that the Model selection module includes：
Traversal Unit, for according to the face character feature, traveling through the decision tree, finds corresponding leaf node；
Model acquiring unit, for obtaining the corresponding face forecast model of the leaf node.
18. systems according to any one of claim 11-17, it is characterised in that the system also includes：
Recovery module, for according to the face character feature, reducing to the facial image of the prediction, after being reduced
Facial image.
19. systems according to claim 18, it is characterised in that the system also includes：
Synthesis module, for by the facial image after the facial image of the prediction or the reduction and the face to be predicted
The background of image is synthesized, the facial image after being synthesized.
20. systems according to claim 19, it is characterised in that the system also includes：
Feedback module, for the facial image after the facial image of the prediction or reduction or the facial image after synthesis is anti-
Feed user.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201610886084.2A CN106651978B (en) | 2016-10-10 | 2016-10-10 | Face image prediction method and system |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201610886084.2A CN106651978B (en) | 2016-10-10 | 2016-10-10 | Face image prediction method and system |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN106651978A true CN106651978A (en) | 2017-05-10 |
| CN106651978B CN106651978B (en) | 2021-04-02 |
Family
ID=58854367
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201610886084.2A Active CN106651978B (en) | 2016-10-10 | 2016-10-10 | Face image prediction method and system |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN106651978B (en) |
Cited By (9)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN107194868A (en) * | 2017-05-19 | 2017-09-22 | 成都通甲优博科技有限责任公司 | A kind of Face image synthesis method and device |
| CN107563122A (en) * | 2017-09-20 | 2018-01-09 | 长沙学院 | The method of crime prediction of Recognition with Recurrent Neural Network is locally connected based on interleaving time sequence |
| CN107729886A (en) * | 2017-11-24 | 2018-02-23 | 北京小米移动软件有限公司 | The processing method and processing device of facial image |
| CN108171167A (en) * | 2017-12-28 | 2018-06-15 | 百度在线网络技术（北京）有限公司 | For exporting the method and apparatus of image |
| CN108520036A (en) * | 2018-03-30 | 2018-09-11 | 广东欧珀移动通信有限公司 | Choosing method, device, storage medium and the electronic equipment of image |
| CN110035271A (en) * | 2019-03-21 | 2019-07-19 | 北京字节跳动网络技术有限公司 | Fidelity image generation method, device and electronic equipment |
| CN111339991A (en) * | 2020-03-12 | 2020-06-26 | 北京爱笔科技有限公司 | Human body attribute identification method and device |
| CN111340932A (en) * | 2018-12-18 | 2020-06-26 | 富士通株式会社 | Image processing method and information processing apparatus |
| CN113808010A (en) * | 2021-09-24 | 2021-12-17 | 深圳万兴软件有限公司 | Cartoon portrait generating method, device, equipment and medium without attribute deviation |
Citations (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN1870047A (en) * | 2006-06-15 | 2006-11-29 | 西安交通大学 | Human face image age changing method based on average face and senile proportional image |
| CN101556701A (en) * | 2009-05-15 | 2009-10-14 | 陕西盛世辉煌智能科技有限公司 | Human face image age changing method based on average face and aging scale map |
| US20120288166A1 (en) * | 2011-05-13 | 2012-11-15 | Microsoft Corporation | Association and prediction in facial recognition |
| CN105787974A (en) * | 2014-12-24 | 2016-07-20 | 中国科学院苏州纳米技术与纳米仿生研究所 | Establishment method for establishing bionic human facial aging model |
-
2016
- 2016-10-10 CN CN201610886084.2A patent/CN106651978B/en active Active
Patent Citations (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN1870047A (en) * | 2006-06-15 | 2006-11-29 | 西安交通大学 | Human face image age changing method based on average face and senile proportional image |
| CN101556701A (en) * | 2009-05-15 | 2009-10-14 | 陕西盛世辉煌智能科技有限公司 | Human face image age changing method based on average face and aging scale map |
| US20120288166A1 (en) * | 2011-05-13 | 2012-11-15 | Microsoft Corporation | Association and prediction in facial recognition |
| CN105787974A (en) * | 2014-12-24 | 2016-07-20 | 中国科学院苏州纳米技术与纳米仿生研究所 | Establishment method for establishing bionic human facial aging model |
Non-Patent Citations (1)
| Title |
|---|
| JINLI SUO 等: "A Compositional and Dynamic Model for Face Aging", 《IEEE TRANSACTIONS ON PATTERN ANALYSIS AND MACHINE INTELLIGENCE》 * |
Cited By (13)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN107194868A (en) * | 2017-05-19 | 2017-09-22 | 成都通甲优博科技有限责任公司 | A kind of Face image synthesis method and device |
| CN107563122B (en) * | 2017-09-20 | 2020-05-19 | 长沙学院 | Crime prediction method based on interleaving time sequence local connection cyclic neural network |
| CN107563122A (en) * | 2017-09-20 | 2018-01-09 | 长沙学院 | The method of crime prediction of Recognition with Recurrent Neural Network is locally connected based on interleaving time sequence |
| CN107729886A (en) * | 2017-11-24 | 2018-02-23 | 北京小米移动软件有限公司 | The processing method and processing device of facial image |
| CN108171167A (en) * | 2017-12-28 | 2018-06-15 | 百度在线网络技术（北京）有限公司 | For exporting the method and apparatus of image |
| CN108520036A (en) * | 2018-03-30 | 2018-09-11 | 广东欧珀移动通信有限公司 | Choosing method, device, storage medium and the electronic equipment of image |
| CN108520036B (en) * | 2018-03-30 | 2020-08-14 | Oppo广东移动通信有限公司 | Image selection method and device, storage medium and electronic equipment |
| CN111340932A (en) * | 2018-12-18 | 2020-06-26 | 富士通株式会社 | Image processing method and information processing apparatus |
| CN110035271A (en) * | 2019-03-21 | 2019-07-19 | 北京字节跳动网络技术有限公司 | Fidelity image generation method, device and electronic equipment |
| CN110035271B (en) * | 2019-03-21 | 2020-06-02 | 北京字节跳动网络技术有限公司 | Fidelity image generation method and device and electronic equipment |
| CN111339991A (en) * | 2020-03-12 | 2020-06-26 | 北京爱笔科技有限公司 | Human body attribute identification method and device |
| CN113808010A (en) * | 2021-09-24 | 2021-12-17 | 深圳万兴软件有限公司 | Cartoon portrait generating method, device, equipment and medium without attribute deviation |
| CN113808010B (en) * | 2021-09-24 | 2023-08-11 | 深圳万兴软件有限公司 | Cartoon portrait generating method, device, equipment and medium without attribute deviation |
Also Published As
| Publication number | Publication date |
|---|---|
| CN106651978B (en) | 2021-04-02 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN106651978A (en) | Face image prediction method and system | |
| CN108510437B (en) | Virtual image generation method, device, equipment and readable storage medium | |
| CN105512289B (en) | Image search method based on deep learning and Hash | |
| KR20220066366A (en) | Predictive individual 3D body model | |
| CN102567716B (en) | Face synthetic system and implementation method | |
| CN109902565B (en) | Multi-feature fusion human behavior recognition method | |
| CN109815826A (en) | The generation method and device of face character model | |
| JP2021520567A (en) | Face matching method and device, storage medium | |
| CN111401339B (en) | Method and device for identifying age of person in face image and electronic equipment | |
| CN106407369A (en) | Photo management method and system based on deep learning face recognition | |
| CN109726619A (en) | A kind of convolutional neural networks face identification method and system based on parameter sharing | |
| CN114676853A (en) | Data processing method, device, equipment and medium | |
| CN110232331B (en) | Online face clustering method and system | |
| JP2020177620A (en) | Method of generating 3d facial model for avatar and related device | |
| CN107194344A (en) | The Human bodys' response method at adaptive bone center | |
| KR20210100592A (en) | Face recognition technology based on heuristic Gaussian cloud transformation | |
| CN111985332A (en) | Gait recognition method for improving loss function based on deep learning | |
| CN110598097B (en) | Hair style recommendation system, method, equipment and storage medium based on CNN | |
| CN113222181A (en) | Federated learning method facing k-means clustering algorithm | |
| CN113361387A (en) | Face image fusion method and device, storage medium and electronic equipment | |
| CN113139415A (en) | Video key frame extraction method, computer device and storage medium | |
| CN104881647B (en) | Information processing method, information processing system and information processing unit | |
| CN112836755B (en) | Sample image generation method and system based on deep learning | |
| CN109145140A (en) | One kind being based on the matched image search method of hand-drawn outline figure and system | |
| CN111626212B (en) | Method and device for identifying object in picture, storage medium and electronic device |
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 |