CN109886873A - A kind of simulated portrait generation method and device based on deep learning - Google Patents
A kind of simulated portrait generation method and device based on deep learning Download PDFInfo
- Publication number
- CN109886873A CN109886873A CN201910059491.XA CN201910059491A CN109886873A CN 109886873 A CN109886873 A CN 109886873A CN 201910059491 A CN201910059491 A CN 201910059491A CN 109886873 A CN109886873 A CN 109886873A
- Authority
- CN
- China
- Prior art keywords
- face
- facial
- module
- image
- portrait
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Pending
Links
Landscapes
- Image Analysis (AREA)
- Image Processing (AREA)
Abstract
The simulated portrait generation method based on deep learning that the invention discloses a kind of includes the following steps: S1: obtaining face spliced map or facial contour figure；S2: generating facial image, and the condition entry of confrontation network is generated using the face spliced map or facial contour figure as condition, complete facial image similar with the two respectively, more true to nature, more natural is generated for constraint network；S3: face identification and editor, victim recognize the complete facial image of generation, and propose partly or wholly suggestion for revision, and picture is edited or regenerated according to suggestion for revision, final output facial image.The invention also discloses the corresponding devices of this method.Simulated portrait generation method of the invention, improvement based on existing hand-drawing method and spliced simulated portrait method, it introduces condition and generates confrontation network algorithm and convolutional neural networks algorithm, so that the simulated portrait generated is more vivid natural, true to nature, vivid, practical function will be greatly promoted.
Description
Technical field
The invention belongs to intelligent technique of criminal investigation fields, raw more particularly, to a kind of simulated portrait based on deep learning
At method and device.
Background technique
The wide coverage for process of handling a case with all kinds of media the police and the generally raising of people's education level, crime are disliked
People is doubted by various information channel researchs police solving criminal cases process, counter-investigation ability is increasingly promoted.Even if being covered possessing range
Today of the high fingerprint of the monitoring network and accuracy of Gai Jiguang, DNA identification technology, suspect still can be by hiding
The modes such as monitoring, destruction monitor and control facility, erasing fingerprint, cleaning scene of a crime destroy clue and evidence, to police's investigation and solve a case
It has caused great difficulties.In this kind of case, victim with hatred in brain for leaving in terror to suspicion of crime
The deep impression of human face's feature is defeated by memory of the victim to suspect's appearance at investigation, the precious clue ordered to arrest
Out to the urgent need for becoming cracking of cases work on paper, simulated portrait technology exactly serves this important need.
Simulated portrait is just taken seriously in criminal investigation field since ancient times, and professional simulated portrait teacher passes through the friendship with victim
Stream, understands the facial characteristics of suspect, draws the simulation of suspect on paper by painting modes such as manual sketches
Portrait, and modified repeatedly according to the feedback of victim, until reaching similar to the greatest extent to the haunting impression of victim.This
Kind is still generally used by way of giving an oral account with Freehandhand-drawing until today, and is played during the detection of large amount of complex case
Great function.The advantages of this mode is more can subtly to show the face of suspect by constantly modifying
Feature and making to its more special feature such as special shape face, birthmark, scar etc. is targetedly portrayed；The disadvantage is that excessively
Rely on simulated portrait teacher technology also have higher requirements with experience, to the ability to express of victim, the time of drawing a picture is longer, miss by
Evil person's impressive time, it is difficult to large-scale popularization etc..
With being constantly progressive for computer hardware technique, public security department also cooperates with scientific research institution, has developed automatic
Simulated portrait software, such software have played the advantage that public security system possesses extensive face database, have established comprising various
The candidate library of the face characteristic material of various kinds, victim only need to select in candidate library similar with suspect shape of face,
They, are stitched together by the materials such as eyes, eyebrow, nose, mouth in software, can quickly generate a suspect
Simulated portrait.The advantages of this mode is that speed is fast, eliminate the reliance on simulated portrait teacher technology and experience, also avoid it is aggrieved
The inaccuracy that person causes information to convey in dictation process, it is easy to spread.But there is also some disadvantages for this automatic imitation portrait
And deficiency, the portrait being spliced into such as it often seem stiff, it is not vivid enough, using effect it is difficult to ensure that.
Summary of the invention
Aiming at the above defects or improvement requirements of the prior art, the present invention provides a kind of simulated portrait based on deep learning
Generation method and device quickly generate image, the natural, suspect with height resolvability it is intended that realizing
Simulated portrait, and this simulated portrait generation method intelligence degree based on deep learning is higher, effect is more preferable, is easier to
It promotes.
To achieve the goals above, according to one aspect of the present invention, a kind of simulated portrait based on deep learning is provided
Generation method includes the following steps:
S1: face spliced map or facial contour figure are obtained；
Face merging features mode or face are used to the memory characteristic selection of suspect's appearance by victim
Profile draws mode, and it is respectively face spliced map or people that the face characteristic connecting method or facial contour, which draw the output of mode,
Face profile diagram；
S2: facial image is generated；
The condition entry that confrontation network is generated using the face spliced map or facial contour figure as condition, for about
Beam network generates complete facial image similar with the two respectively, more true to nature, more natural；
S3: face identification and editor；
Victim recognizes the complete facial image of generation, and proposes partly or wholly suggestion for revision, if modification limit
In local feature, then can be done by portrait editor suitably modified；If global revision, then return step S1, until victim distinguishes
See clearly it is clear after, final output facial image.
Further, in step S2, it includes generator and arbiter that the condition, which generates confrontation network, comprising following step
It is rapid:
S21: the condition entry by the face spliced map or facial contour figure and noise together as the generator, life
At a facial image；
S22: the facial image of generation and real human face image are inputted into the arbiter, and input figure is sentenced by it
It seem real human face image, or the facial image generated by the generator.
Further, in step S2, the condition generates confrontation network training and includes the following steps:
S25: arbiter loss function value in the positive and negative samples of input, backpropagation loss function, to differentiation are sought
The gradient of model parameter carries out associated update in device, generator；
S26: promoting respective performance by the game of generator and arbiter, until facial image that generator generates and true
Probability value of the real image in arbiter it is stable near 0.5 when, terminate training.
Further, in step S22, use positive and negative samples as input during the differentiation of the arbiter；Wherein,
The negative sample is the face spliced map or face of the facial image that the generator generates and respective conditions input
Profile diagram；
The positive sample is the people that the characteristics of image in real human face image and corresponding shearing real human face is spliced
Face spliced map or facial contour figure.
Further, the face spliced map that the characteristics of image in real human face is spliced includes the following steps:
S221: it selects high quality face image data and forms training dataset；
S222: picture that the training data is concentrated is subjected to the cutting of face and is again spliced, it is corresponding to form every picture
Spliced map；
S223: to artwork data and corresponding stitching image data be adjusted brightness, contrast, minor translation, rotation,
Flip horizontal operation obtains the face spliced map that the characteristics of image in real human face is spliced.
Other side according to the invention provides a kind of simulated portrait generating means based on deep learning, for real
The existing simulated portrait generation method, the device include:
The facial feature database module and facial contour drafting module being independently arranged, wherein the face characteristic data
For library module for selecting face characteristic data to be spliced to form face spliced map, the facial contour drafting module is used for drawing human-face
Profile diagram；
Face generation module is used to generate confrontation network using the face spliced map or facial contour figure as condition
Condition entry generates facial image；
Portrait editor module, it is vivid, natural, true to nature to obtain for the facial image of generation to be recognized and modified
Facial image；And
Human-computer interaction module generates mould for realizing with facial feature database module, facial contour drafting module, face
The interactive operation of block and portrait editor module.
Further, the facial feature database module includes face characteristic data, face splicing submodule and data
Submodule is managed in depositary management；Wherein,
The face characteristic data include shape of face data, face data, individualized feature data and decoration data；
The data base administration submodule is used to take tree-shaped knot according to its generic, attribute to face characteristic data
Structure storage.
Further, the facial contour drafting module includes hardware adaptor and software submodules；Wherein,
The hardware adaptor includes hand drawing board and pressure sensi-tive pen；
The software submodules include painting canvas and memory interface.
Further, the face generation module includes human face segmentation device based on merging features and based on hand-drawn outline
Human face segmentation device, the two correspond respectively to facial feature database module and facial contour drafting module.
Further, the human-computer interaction module includes the software interface interacted on PC with user；Wherein,
The interactive interface is incorporated into the facial feature database module, facial contour drafting module, face and generates
Module, portrait editor module sub-interface button and be currently generated the effect picture of portrait；And
The interactive interface further include image is created, is saved, is loaded into, is exported, the function button of printing.
In general, through the invention it is contemplated above technical scheme is compared with the prior art, can obtain down and show
Beneficial effect:
1. the simulated portrait generation method of the invention based on deep learning, the technical solution of proposition is based on existing hand
The improvement for drawing method and spliced simulated portrait method, the condition that introduces generates confrontation network algorithm and convolutional neural networks are calculated
Method will greatly promote practical function so that the simulated portrait generated is more vivid natural, true to nature, vivid.
2. the simulated portrait generation method of the invention based on deep learning generates confrontation network algorithm and volume based on condition
To data scale and stronger dependence is distributed in the performance of product neural network algorithm even depth learning algorithm, algorithm model, passes through reality
The continuous accumulation of data and continuing to optimize for data distribution are trampled, the performance of algorithm model and the quality for generating simulated portrait will be continuous
It is promoted.
3. the simulated portrait generation method of the invention based on deep learning, relative to traditional hand design formula simulated portrait method
Have the advantages that formation speed is fast, victim can be caught to remember and retain most valuable moment, by quickly generating, fast velocity modulation
Whole, in excitation victim's memory details, reaches specific, careful, true to nature to the greatest extent.
4. the simulated portrait generation method of the invention based on deep learning, relative to traditional hand design formula simulated portrait method
Have the characteristics that easy to spread, software and hardware cost is not high, and not by technical staff (such as simulated portrait teacher) technical level height
It influences.
Detailed description of the invention
Fig. 1 is simulated portrait generation method flow diagram of the embodiment of the present invention based on deep learning；
Fig. 2 is that the condition in face of embodiment of the present invention generation module generates confrontation network diagram；
Fig. 3 is that the condition in face of embodiment of the present invention generation module generates the flow diagram for fighting network；
Fig. 4 is the structural schematic diagram of simulated portrait generating means of the embodiment of the present invention based on deep learning；
Fig. 5 is facial feature database of embodiment of the present invention module composed structure schematic diagram；
Fig. 6 is facial contour of embodiment of the present invention drafting module composed structure schematic diagram；
Fig. 7 is face of embodiment of the present invention generation module composed structure schematic diagram.
Specific embodiment
In order to make the objectives, technical solutions, and advantages of the present invention clearer, with reference to the accompanying drawings and embodiments, right
The present invention is further elaborated.It should be appreciated that described herein, specific examples are only used to explain the present invention, not
For limiting the present invention.As long as in addition, technical characteristic involved in the various embodiments of the present invention described below that
Not constituting conflict between this can be combined with each other.
As shown in Figure 1, a kind of simulated portrait generation method based on deep learning of the embodiment of the present invention, including walk as follows
It is rapid:
(1) face generating mode is selected
The embodiment of the present invention provides two kinds of mutually independent modes and generates simulated portrait, in actual use, technical staff
The memory characteristic selection of suspect's appearance is drawn using face merging features mode or facial contour by victim
Mode；The former is suitable for victim and remembers relatively clear, with deep impression to local feature situation, and the latter is suitable for only to face
Type, face length height have the case where vague memory；It can integrate using two methods in special circumstances.
(2) face generates
The output that face characteristic connecting method and facial contour draw mode is respectively face spliced map and facial contour figure,
They generate the condition entry of confrontation network respectively as the condition in face generation, generate for constraint network similar to them
, complete facial image more true to nature, more natural.
(3) face editor
The image that face generation module generates needs victim to recognize, proposes part and whole suggestion for revision, if
Modification be limited to local feature (such as the wing of nose is wide, and canthus is longer etc.), may be used portrait editor do it is suitably modified, it is if desired larger
Change (such as face profile needs to elongate, and forehead needs wider etc.), then can come back to facial feature database or facial contour is drawn,
Face of beginning one's life anew splices or repaints facial contour figure, repeats step (2), after victim's identification is clear, final output
Facial image.Simulated portrait generation method based on deep learning of the invention, the technical solution of proposition are based on existing hand
The improvement for drawing method and spliced simulated portrait method, the condition that introduces generates confrontation network algorithm and convolutional neural networks are calculated
Method will greatly promote practical function so that the simulated portrait generated is more vivid natural, true to nature, vivid.
Fig. 2 show face in step of the embodiment of the present invention (2) and generates the structure that used condition generates confrontation network
Schematic diagram.Two independent networks are contained, face spliced map and facial contour figure are respectively used to；Network structure is similar, with people
It is illustrated for the network that face spliced map uses.Condition used in the present invention generates confrontation network by one based on convolution mind
Generator and an arbiter composition based on convolutional neural networks through network；In training process, the target of generator is logical
It crosses condition entry (face spliced map) and noise inputs generates a face picture, it is desirable to which it is as similar to true picture as possible；
Negative sample is the picture and respective conditions input (face spliced map) that generator generates in the input of arbiter, and positive sample is true
Facial image and corresponding condition entry (the face spliced map that the characteristics of image in shearing real human face is spliced), target
Be differentiate in the case where the same or similar face spliced map is as condition entry, the facial image of input be it is true or
It is generated by generator；Arbiter loss function value in the small lot sample of input, backpropagation loss function are sought in training
Gradient to the model parameter in arbiter, generator simultaneously carries out associated update to parameter.By generator in optimization process and
The game of arbiter promotes respective performance, until the probability value of the facial image and true picture of generator generation in arbiter
It is stable near 0.5 when, illustrate that generator performance is optimal, terminate training.In use, victim is according to memory, by
Condition entry of the face spliced map that facial feature database module is spliced into as generator produces image, nature, forces
Genuine facial image.Simulated portrait generation method based on deep learning of the invention generates confrontation network algorithm based on condition
To data scale and stronger dependence is distributed with the performance of convolutional neural networks algorithm even depth learning algorithm, algorithm model, leads to
The continuous accumulation of practical data and continuing to optimize for data distribution are crossed, the performance of algorithm model and the quality for generating simulated portrait will
Constantly promoted.
Fig. 3 is that the condition in face of embodiment of the present invention generation module generates the flow diagram for fighting network.Such as Fig. 3 institute
Show, high quality face image data is selected in the extensive face image database that public security system is grasped and forms training data
Picture in data set is carried out the cutting of face and again spliced, forms the corresponding spliced map of every picture by collection, to artwork data and
Corresponding stitching image data are adjusted the data enhancement operations such as brightness, contrast, minor translation, rotation, flip horizontal, use
Confrontation network is generated in the condition that training can generate corresponding face according to stitching image；In training, condition generates confrontation network
Generator with spliced map and noise be input, export facial image corresponding to spliced map；Arbiter is with real human face image
Or generator generate facial image be input, export they be real human face image probability, with logarithmic function be loss letter
Number optimizes generator and arbiter parameter by small lot gradient descent method alternating iteration, promotes generator and arbiter performance；
The practical middle image for using victim to go out by the merging features in facial feature database fights the input of network as generation,
Complete facial image similar in feature selected by the features such as face, shape of face and victim can be obtained.Likewise, being based on hand-drawn outline
Human face segmentation device need for training data concentrate picture draw corresponding facial contour image, by facial contour in training
Image is input to generator and arbiter, it is practical in can be drawn by victim or facial contour image that portrait teacher draws it is raw
At complete facial image.Simulated portrait generation method based on deep learning of the invention simulates picture relative to traditional hand design formula
Image space method has the advantages that formation speed is fast, and victim can be caught to remember and retain most valuable moment, by quickly generating,
Quickly adjustment excites the details in victim's memory, reaches specific, careful, true to nature to the greatest extent.The present invention realizes as a result,
A kind of simulated portrait generation method based on deep learning can greatly promote the quality and efficiency of simulated portrait generation, at
The good assistant to handle a case for the police.
As shown in figure 4, providing a kind of simulated portrait generation dress based on deep learning in another embodiment of the invention
It sets, including facial feature database module 100, facial contour drafting module 200, face generation module 300, portrait editor module
400 and human-computer interaction module 500.
As shown in figure 5, facial feature database module 100 contains the face characteristic data 102 of database purchase, face
Splice submodule 103 and data base administration submodule 101.Wherein, face characteristic data 102 include type shape of face number abundant
According to face features picture materials such as 1021, face data 1022, individualized feature data 1023 and decoration datas 1024.Herein
Face data 1022 refer to: eyebrow, eyes, nose, mouth, ear etc.；Individualized feature data 1023 include wrinkle, tire
The features such as note, acne, scar, black mole；Decoration data 1024 includes cap, scarf, glasses, necklace, earrings, tatoos.
Data base administration submodule 101 is used to take tree-shaped face characteristic data 102 according to its generic, attribute
Structure stores, and classification, attribute, sub- attribute etc. need to be successively only selected according to the list defined in interface in user's search procedure, i.e.,
Corresponding face feature image can accurately be found.Data base administration submodule 101 simultaneously for system maintenance personnel provide insertion,
The operation interfaces such as editor, deletion, it is convenient that database is managed；User according to memory, selects in the database in use
With the most matched face feature of suspect, face spliced map is formed, is used for face generation module.
As shown in fig. 6, facial contour drafting module 200 includes hardware adaptor 201 and software submodules 202, wherein hardware
Peripheral hardware 201 includes the external devices such as hand drawing board 2011 and pressure sensi-tive pen 2012.Software submodules 202 include painting canvas 2021 and storage
Interface 2022 etc.；The facial contour drafting module 200 is used for the simple facial contour skeleton of hand drawn, including shape of face, face
Position and profile etc.；In use, victim or portrait teacher draw simple face wheel by pressure sensi-tive pen on hand drawing board
Exterior feature is used for face generation module.
Preferably, face generation module 300 includes human face segmentation device 301 based on merging features and based on hand-drawn outline
Human face segmentation device 302, the two correspond respectively to facial feature database module 100 and facial contour drafting module 200.Victim
Face spliced map is obtained using facial feature database module 100 or obtains face wheel using facial contour drafting module 200
Exterior feature figure, in addition noise data generates the input of confrontation network together as condition.It includes a base that the condition, which generates confrontation network,
In the generator of convolutional neural networks and an arbiter based on convolutional neural networks；In training process, the target of generator
It is that a face picture is generated by condition entry (face spliced map or facial contour figure) and noise inputs, it is desirable to which it is as far as possible
It is similar to true picture；Negative sample is the picture and respective conditions input (face splicing that generator generates in the input of arbiter
Figure), positive sample is real human face image and the corresponding condition entry (people that the characteristics of image in shearing real human face is spliced
Face spliced map), target is to differentiate the face of input in the case where the same or similar face spliced map is as condition entry
Image is true or is generated by generator；Arbiter loss function value in the small lot sample of input is sought in training,
Backpropagation loss function carries out associated update to the gradient of the model parameter in arbiter, generator and to parameter.By excellent
The game of generator and arbiter promotes respective performance during change, until facial image and true picture that generator generates exist
Probability value in arbiter it is stable near 0.5 when, illustrate that generator performance is optimal, terminate training.In use, aggrieved
Person is according to memory, condition entry of the face spliced map being spliced by facial feature database module as generator
Generate vivid, natural, true to nature facial image.
In addition, portrait editor module 400 is used to generate portrait generation module 300 according to victim the feedback of facial image
Carry out modification appropriate, such as cheek, chin shape, eyes size, wing of nose width, lip thickness.Portrait editor module 400 can
Seek to seek to cooperate with figure software of repairing more popular on Vehicles Collected from Market, call its interface, completes face and remold function.
Further, human-computer interaction module 500 is the software interface interacted on PC with user.Human-computer interaction interface
It contains into above-mentioned facial feature database module, facial contour drafting module, face generation module, portrait editor module
The button of sub-interface and the effect picture for being currently generated portrait；Human-computer interaction interface further includes being created, being saved to image, carried
The function button of operations such as enter, export, printing.
As it will be easily appreciated by one skilled in the art that the foregoing is merely illustrative of the preferred embodiments of the present invention, it is not used to
The limitation present invention, any modifications, equivalent substitutions and improvements made within the spirit and principles of the present invention should all include
Within protection scope of the present invention.
Claims (10)
1. a kind of simulated portrait generation method based on deep learning, which comprises the steps of:
S1: face spliced map or facial contour figure are obtained；
Face merging features mode or facial contour are used to the memory characteristic selection of suspect's appearance by victim
The output that drafting mode, the face characteristic connecting method or facial contour draw mode is respectively face spliced map or face wheel
Exterior feature figure；
S2: facial image is generated；
The condition entry that confrontation network is generated using the face spliced map or facial contour figure as condition, for constraining net
Network generates complete facial image similar with the two respectively, more true to nature, more natural；
S3: face identification and editor；
Victim recognizes the complete facial image of generation, and proposes partly or wholly suggestion for revision, if modification is limited to office
Portion's feature can then be done suitably modified by portrait editor；If global revision, then return step S1, until victim's identification is clear
After clear, final output facial image.
2. a kind of simulated portrait generation method based on deep learning according to claim 1, which is characterized in that step S2
In, it includes generator and arbiter that the condition, which generates confrontation network, comprising following steps:
S21: by the face spliced map or facial contour figure and noise together as the condition entry of the generator, one is generated
Open facial image；
S22: the facial image of generation and real human face image are inputted into the arbiter, and sentence input picture by it to be
Real human face image, or the facial image generated by the generator.
3. a kind of simulated portrait generation method based on deep learning according to claim 1 or 2, which is characterized in that step
In rapid S2, the condition generates confrontation network training and includes the following steps:
S25: seek input positive and negative samples in arbiter loss function value, the backpropagation loss function, with to arbiter,
The gradient of model parameter carries out associated update in generator；
S26: promoting respective performance by the game of generator and arbiter, until facial image and true figure that generator generates
As probability value in arbiter it is stable near 0.5 when, terminate training.
4. a kind of simulated portrait generation method based on deep learning according to any one of claim 1-3, feature
It is, in step S22, uses positive and negative samples as input during the differentiation of the arbiter；Wherein,
The negative sample is the face spliced map or facial contour of the facial image that the generator generates and respective conditions input
Figure；
The positive sample is that the face that the characteristics of image in real human face image and corresponding shearing real human face is spliced is spelled
Map interlinking or facial contour figure.
5. a kind of simulated portrait generation method based on deep learning, feature described in any one of -4 according to claim 1
It is, the face spliced map that the characteristics of image in real human face is spliced includes the following steps:
S221: it selects high quality face image data and forms training dataset；
S222: picture that the training data is concentrated is subjected to the cutting of face and is again spliced, the corresponding splicing of every picture is formed
Figure；
S223: brightness, contrast, minor translation, rotation, level are adjusted to artwork data and corresponding stitching image data
Turning operation obtains the face spliced map that the characteristics of image in real human face is spliced.
6. a kind of simulated portrait generating means based on deep learning, for realizing according to any one of claims 1 to 5
Simulated portrait generation method, which is characterized in that the device includes:
The facial feature database module (100) and facial contour drafting module (200) being independently arranged, wherein the face is special
Sign database module (100) is for selecting face characteristic data to be spliced to form face spliced map, the facial contour drafting module
(200) it is used for drawing human-face profile diagram；
Face generation module (300) is used to generate confrontation network using the face spliced map or facial contour figure as condition
Condition entry, generate facial image；
Portrait editor module (400), it is vivid, natural, true to nature to obtain for the facial image of generation to be recognized and modified
Facial image；And
Human-computer interaction module (500), for realizing with facial feature database module (100), facial contour drafting module
(200), the interactive operation of face generation module (300) and portrait editor module (400).
7. a kind of simulated portrait generating means based on deep learning according to claim 6, which is characterized in that the people
Face characteristic library module (100) includes face characteristic data (102), face splicing submodule (103) and data base administration
Module (101)；Wherein,
The face characteristic data (102) include shape of face data (1021), face data (1022), individualized feature data
(1023) and decoration data (1024)；
The data base administration submodule (101) is used to take face characteristic data (102) according to its generic, attribute
Tree storage.
8. a kind of simulated portrait generating means based on deep learning according to claim 6 or 7, which is characterized in that institute
Stating facial contour drafting module (200) includes hardware adaptor (201) and software submodules (202)；Wherein,
The hardware adaptor (201) includes hand drawing board (2011) and pressure sensi-tive pen (2012)；
The software submodules (202) include painting canvas (2021) and memory interface (2022).
9. a kind of simulated portrait generating means based on deep learning a method according to any one of claims 6-8, feature
It is, the face generation module (300) includes the human face segmentation device (301) based on merging features and the people based on hand-drawn outline
Face synthesizer (302), the two correspond respectively to facial feature database module (100) and facial contour drafting module (200).
10. a kind of simulated portrait generating means based on deep learning, feature according to any one of claim 6-9
It is, the human-computer interaction module (500) includes the software interface interacted on PC with user；Wherein,
The interactive interface is incorporated into the facial feature database module (100), facial contour drafting module (200), people
Face generation module (300), portrait editor module (400) sub-interface button and be currently generated the effect picture of portrait；And
The interactive interface further include image is created, is saved, is loaded into, is exported, the function button of printing.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201910059491.XA CN109886873A (en) | 2019-01-22 | 2019-01-22 | A kind of simulated portrait generation method and device based on deep learning |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201910059491.XA CN109886873A (en) | 2019-01-22 | 2019-01-22 | A kind of simulated portrait generation method and device based on deep learning |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN109886873A true CN109886873A (en) | 2019-06-14 |
Family
ID=66926530
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201910059491.XA Pending CN109886873A (en) | 2019-01-22 | 2019-01-22 | A kind of simulated portrait generation method and device based on deep learning |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN109886873A (en) |
Cited By (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN112132912A (en) * | 2019-06-25 | 2020-12-25 | 北京百度网讯科技有限公司 | Method and device for establishing face generation model and generating face image |
| CN112307234A (en) * | 2020-11-03 | 2021-02-02 | 厦门兆慧网络科技有限公司 | Face bottom library synthesis method, system, device and storage medium |
| CN113506220A (en) * | 2021-07-16 | 2021-10-15 | 厦门美图之家科技有限公司 | Human face posture editing method and system driven by 3D (three-dimensional) vertex and electronic equipment |
| CN114359034A (en) * | 2021-12-24 | 2022-04-15 | 北京航空航天大学 | Method and system for generating face picture based on hand drawing |
Citations (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN105976429A (en) * | 2016-05-03 | 2016-09-28 | 西安邮电大学 | Criminal investigation portrait computer aided system and method |
| CN107730458A (en) * | 2017-09-05 | 2018-02-23 | 北京飞搜科技有限公司 | A kind of fuzzy facial reconstruction method and system based on production confrontation network |
-
2019
- 2019-01-22 CN CN201910059491.XA patent/CN109886873A/en active Pending
Patent Citations (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN105976429A (en) * | 2016-05-03 | 2016-09-28 | 西安邮电大学 | Criminal investigation portrait computer aided system and method |
| CN107730458A (en) * | 2017-09-05 | 2018-02-23 | 北京飞搜科技有限公司 | A kind of fuzzy facial reconstruction method and system based on production confrontation network |
Non-Patent Citations (2)
| Title |
|---|
| 杨杰: "谈谈摹拟画像在刑事侦查中的实践与运用", 《万方数据库》 * |
| 程旭 等: "基于多损失的生成式对抗目标跟踪算法", 《东南大学学报（自然科学版）》 * |
Cited By (7)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN112132912A (en) * | 2019-06-25 | 2020-12-25 | 北京百度网讯科技有限公司 | Method and device for establishing face generation model and generating face image |
| CN112132912B (en) * | 2019-06-25 | 2024-02-13 | 北京百度网讯科技有限公司 | Method and device for establishing face generation model and generating face image |
| CN112307234A (en) * | 2020-11-03 | 2021-02-02 | 厦门兆慧网络科技有限公司 | Face bottom library synthesis method, system, device and storage medium |
| CN113506220A (en) * | 2021-07-16 | 2021-10-15 | 厦门美图之家科技有限公司 | Human face posture editing method and system driven by 3D (three-dimensional) vertex and electronic equipment |
| CN113506220B (en) * | 2021-07-16 | 2024-04-05 | 厦门美图之家科技有限公司 | Face gesture editing method and system driven by 3D vertex and electronic equipment |
| CN114359034A (en) * | 2021-12-24 | 2022-04-15 | 北京航空航天大学 | Method and system for generating face picture based on hand drawing |
| CN114359034B (en) * | 2021-12-24 | 2023-08-08 | 北京航空航天大学 | Face picture generation method and system based on hand drawing |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN109886873A (en) | A kind of simulated portrait generation method and device based on deep learning | |
| CN107491726A (en) | A kind of real-time expression recognition method based on multi-channel parallel convolutional neural networks | |
| US20210174072A1 (en) | Microexpression-based image recognition method and apparatus, and related device | |
| CN109325443A (en) | A kind of face character recognition methods based on the study of more example multi-tag depth migrations | |
| CN105787974B (en) | Bionic human face aging model method for building up | |
| CN110378985A (en) | A kind of animation drawing auxiliary creative method based on GAN | |
| CN108899050A (en) | Speech signal analysis subsystem based on multi-modal Emotion identification system | |
| Tan et al. | Expressing visual relationships via language | |
| CN108805088A (en) | Physiological signal analyzing subsystem based on multi-modal Emotion identification system | |
| CN104376333A (en) | Facial expression recognition method based on random forests | |
| CN110033029A (en) | A kind of emotion identification method and device based on multi-modal emotion model | |
| CN108492817A (en) | A kind of song data processing method and performance interactive system based on virtual idol | |
| CN111126280B (en) | Gesture recognition fusion-based aphasia patient auxiliary rehabilitation training system and method | |
| WO2020151273A1 (en) | System and method for automatically generating virtual reality scene that stimulates emotions | |
| CN109858392A (en) | One kind is for front and back facial image automatic identifying method of making up | |
| CN109886154A (en) | Most pedestrian's appearance attribute recognition methods according to collection joint training based on Inception V3 | |
| Chen et al. | CreativeBioMan: a brain-and body-wearable, computing-based, creative gaming system | |
| CN109034090A (en) | A kind of emotion recognition system and method based on limb action | |
| Zhai et al. | Asian female facial beauty prediction using deep neural networks via transfer learning and multi-channel feature fusion | |
| WO2021223042A1 (en) | Method for implementing machine intelligence similar to human intelligence | |
| Gao et al. | Wallpaper texture generation and style transfer based on multi-label semantics | |
| Zhu et al. | Application of intelligent image color technology in teaching Chinese painting color | |
| CN110309510B (en) | C-S and GRU-based painting and calligraphy observation method | |
| CN116883608A (en) | Multi-mode digital person social attribute control method and related device | |
| Zeng et al. | Controllable face aging |
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 | ||
| RJ01 | Rejection of invention patent application after publication | ||
| RJ01 | Rejection of invention patent application after publication |
Application publication date: 20190614 |