CN108960159A - A kind of thermal imaging face identification method based on generation confrontation network - Google Patents
A kind of thermal imaging face identification method based on generation confrontation network Download PDFInfo
- Publication number
- CN108960159A CN108960159A CN201810748262.4A CN201810748262A CN108960159A CN 108960159 A CN108960159 A CN 108960159A CN 201810748262 A CN201810748262 A CN 201810748262A CN 108960159 A CN108960159 A CN 108960159A
- Authority
- CN
- China
- Prior art keywords
- image
- loss function
- function
- face
- thermal imaging
- 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.)
- Withdrawn
Links
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
- G06V40/00—Recognition of biometric, human-related or animal-related patterns in image or video data
- G06V40/10—Human or animal bodies, e.g. vehicle occupants or pedestrians; Body parts, e.g. hands
- G06V40/16—Human faces, e.g. facial parts, sketches or expressions
- G06V40/172—Classification, e.g. identification
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/04—Architecture, e.g. interconnection topology
- G06N3/045—Combinations of networks
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/08—Learning methods
Landscapes
- Engineering & Computer Science (AREA)
- Theoretical Computer Science (AREA)
- Physics & Mathematics (AREA)
- General Health & Medical Sciences (AREA)
- Health & Medical Sciences (AREA)
- General Physics & Mathematics (AREA)
- Biomedical Technology (AREA)
- Mathematical Physics (AREA)
- Data Mining & Analysis (AREA)
- Evolutionary Computation (AREA)
- Biophysics (AREA)
- Molecular Biology (AREA)
- Computing Systems (AREA)
- General Engineering & Computer Science (AREA)
- Artificial Intelligence (AREA)
- Computational Linguistics (AREA)
- Software Systems (AREA)
- Life Sciences & Earth Sciences (AREA)
- Oral & Maxillofacial Surgery (AREA)
- Human Computer Interaction (AREA)
- Multimedia (AREA)
- Image Processing (AREA)
- Image Analysis (AREA)
Abstract
What is proposed in the present invention is a kind of based on the thermal imaging face identification method for generating confrontation network, its main contents includes generating confrontation model, loss function, training method, its process is, the graphic images X and the corresponding real human face image Y of the image of given input, X first passes through image generation module G and obtains image G (X), and then X, G (X) and Y pass through judgement of the resolution module progress image very with vacation together；Application conditions cost function and L1 norm loss function, application conditions cost function and identification loss function in resolution module on generation module, the two linear, additive obtains global loss function, under the training method of preset parameter, by minimizing global loss function, the result close to true picture is produced.The present invention can carry out the recognition of face of enclosed region under the conditions of insufficient light, provide the generation confrontation model based on depth convolutional network, improve thermal imaging or the facial image recognition effect of other active IR imagings.
Description
Technical field
The present invention relates to field of image enhancement, are known more particularly, to a kind of based on the thermal imaging face for generating confrontation network
Other method.
Background technique
Face recognition technology is one of biological identification technology, in safety monitoring, population control and data confidentiality etc.
There is very big application value in direction, is the research hotspot of area of pattern recognition.Recognition of face in infrared thermal imaging is face
One branch in identification field is hot spot of interest biological identification technology recent years, at the same be also one have it is very big
The project of application value.Since infrared emanation has the characteristics that anti-camouflage and the independent, anti-interference of light source, antifraud, because
This can make up the intrinsic problem of visible images recognition of face.
With the increasingly raising of infrared thermal imagery face recognition technology, practical application can be generated also more next more extensively
Bigger value.For example, 1) enterprise, house etc. be public or the safety management of private buildings object, access control system is improved comprehensively, letter
Easily, owner, visitor and stranger are easily and safely identified, the probability of happening of potential unsafe incidents is reduced；2) denser population
The flow statistical system in region quickly calculates overall flow of the people and formulates emergency evacuation scheme, while in sensitizing range such as machine
Field, customs etc. implement in place the tracking to individual；3) in criminal investigation field, pass through taking the photograph for more and more covering public domains
It is undoubtedly more efficient to chasing for suspect as head investment service；In addition, for example self-service in sensitive information security fields
Government services, financial assets service, network security etc., the recognition of face in infrared thermal imaging can also bring technical advantage.
So far, it is that this identification method has mostly based on visible light using most face identification systems itself to be difficult to gram
The defect of clothes, especially when the illumination of environment changes, the recognition effect of system be will be greatly reduced, and lead to recognition effect
The needs of reality system are unable to satisfy, therefore the identifying system of this classification can be only applied in specific environment.Moreover, face
It is a 3 d elastic body, it can change with the variation of the hair style of people, posture, makeup, expression, shade, light etc.；
Certain variations for face, such as expression, wrinkle, need by taking certain aptitude manner to be made up, and this is difficult
It is realized by certain technological means perfection, this various adverse effect all limits it in the hair of technical field of face recognition
Exhibition.
The invention proposes a kind of based on the thermal imaging face identification method for generating confrontation network, the heat of input given first
The image X and image corresponding real human face image Y, X first pass through image generation module G and obtain image G (X), then X,
G (X) and Y passes through resolution module D together and carries out judgement of the image very with vacation；On generation module application conditions cost function and
L1 norm loss function, application conditions cost function and identification loss function, the two linear, additive obtain in resolution module
Global loss function, by minimizing global loss function, is produced close to true figure under the training method of preset parameter
The result of picture.The present invention can carry out the recognition of face of enclosed region under the conditions of insufficient light, provide one and rolled up based on depth
The generation confrontation model of product network, improves thermal imaging or the facial image recognition effect of other active IR imagings.
Summary of the invention
Solving the problems, such as progress face recovery and identification in graphic images, the purpose of the present invention is to provide one
Kind is corresponded to based on the thermal imaging face identification method for generating confrontation network, the graphic images X and the image of input given first
Real human face image Y, X first passes through image generation module G and obtains image G (X), and then X, G (X) and Y are together by differentiating mould
Block D carries out judgement of the image very with vacation；On generation module application conditions cost function and L1 norm loss function, differentiating
Application conditions cost function and identification loss function, the two linear, additive obtain global loss function in module, join fixed
Under several training methods, by minimizing global loss function, the result close to true picture is produced.
To solve the above problems, the present invention provides a kind of thermal imaging face identification method based on generation confrontation network,
Main contents include:
(1) confrontation model is generated；
(2) loss function；
(3) training method.
Wherein, the generation confrontation model (one) gives the graphic images X and the corresponding true people of the image of input
Face image Y, X first pass through image generation module and obtain image G (X), and then G (X) and Y carries out condition cost by resolution module
The calculating of function obtains satisfactory image and people by minimizing the summation of condition cost function and local losses function
Face recognition result.
The generation module carries out convolution, Chi Huahe to input picture using the depth convolutional network of a pre-training
Nonlinear Mapping operation, the image after output transform, i.e.,Wherein, w and h respectively refers to image ruler
Very little width and height.
The resolution module, the depth convolutional network initialized using one to the image received to classifying,
It designs a global loss function to control the learning outcome of image pair, output image true and false differentiation result and face
Recognition result, i.e.,Wherein, w and h respectively refers to the width and height of picture size
Degree.
The condition cost function obeys G (X) in Y and generates image or true picture distribution PdataIn the case where, have:
As condition cost function.
The loss function, including local losses function and global loss function.
The local losses function uses L1 norm loss function and identification on generation module and resolution module respectively
Loss function specifically obeys G (X) in X and Y and generates image or true picture distribution PdataIn the case where, have:
As L1 norm loss function；
As identify loss function, DyFor indicator variable, work as yi=1, then face recognition result is that the image is subordinated to
Face Xi。
The global loss function, with the mode of linear, additive by condition cost functionLocal losses functionWithIt is applied to during the entire process of convolutional neural networks, specifically, obeys G (X) in X and Y and generate image or true figure
As distribution PdataIn the case where, have:
Wherein, λ1And λ2For weight coefficient.
The training method initializes simultaneously image using deep learning training frame Tensorflow platform one by one
Training, initial method uses Adam optimizer, after preset parameter, forward and backward training the number of iterations be 65 times or until
When convergence, deconditioning.
The preset parameter carries out assignment to the parameter in network, specifically, 1) weight coefficient λ1And λ2It all sets respectively
It is set to 100；2) learning rate of convolutional neural networks is set as 0.0002；3) batch processing is dimensioned to 1.
Detailed description of the invention
Fig. 1 is a kind of system flow chart based on the thermal imaging face identification method for generating confrontation network of the present invention.
Fig. 2 is a kind of effect contrast figure based on the thermal imaging face identification method for generating confrontation network of the present invention.
Specific embodiment
It should be noted that in the absence of conflict, the features in the embodiments and the embodiments of the present application can phase
It mutually combines, invention is further described in detail in the following with reference to the drawings and specific embodiments.
Fig. 1 is a kind of system flow chart based on the thermal imaging face identification method for generating confrontation network of the present invention.Mainly
Including generating confrontation model；Loss function；Training method.
Wherein, confrontation model is generated, the graphic images X and the image corresponding real human face image Y, X of input are given
It first passes through image generation module and obtains image G (X), then G (X) and Y carries out the meter of condition cost function by resolution module
It calculates, by minimizing the summation of condition cost function and local losses function, obtains satisfactory image and recognition of face knot
Fruit.
Generation module carries out convolution, pond and non-linear to input picture using the depth convolutional network of a pre-training
Mapping operations, the image after output transform, i.e.,Wherein, w and h respectively refers to the width of picture size
Degree and height.
Resolution module, the depth convolutional network initialized using one, to classifying, design one to the image received
A overall situation loss function controls the learning outcome of image pair, output image true and false differentiation result and recognition of face knot
Fruit, i.e., Wherein, w and h respectively refers to the width and height of picture size.
Condition cost function obeys G (X) in Y and generates image or true picture distribution PdataIn the case where, have:
As condition cost function.
Loss function, including local losses function and global loss function.
Local losses function uses L1 norm loss function and identification loss letter on generation module and resolution module respectively
Number specifically obeys G (X) in X and Y and generates image or true picture distribution PdataIn the case where, have:
As L1 norm loss function；
As identify loss function, DyFor indicator variable, work as yi=1, then face recognition result is that the image is subordinated to
Face Xi。
Global loss function, with the mode of linear, additive by condition cost functionLocal losses functionWith
It is applied to during the entire process of convolutional neural networks, specifically, obeys G (X) in X and Y and generate image or true picture distribution
PdataIn the case where, have:
Wherein, λ1And λ2For weight coefficient.
Training method initializes image one by one and trains, just using deep learning training frame Tensorflow platform
Beginning method uses Adam optimizer, and after preset parameter, forward and backward training the number of iterations is 65 times or until when convergence,
Deconditioning.
Preset parameter carries out assignment to the parameter in network, specifically, 1) weight coefficient λ1And λ2It is both configured to respectively
100；2) learning rate of convolutional neural networks is set as 0.0002；3) batch processing is dimensioned to 1.
Fig. 2 is that the present invention is a kind of based on the thermal imaging face identification method effect contrast figure for generating confrontation network, as schemed institute
Show, in the case where thermal imaging face figure (the 1st column) of given ultra-low resolution, identification that the method for the present invention (the 4th column) obtains
Effect is best, and clarity highest, detailed information is richer, also closer to actual face original image (the 5th column).
For those skilled in the art, the present invention is not limited to the details of above-described embodiment, without departing substantially from essence of the invention
In the case where mind and range, the present invention can be realized in other specific forms.In addition, those skilled in the art can be to this
Invention carries out various modification and variations without departing from the spirit and scope of the present invention, and these improvements and modifications also should be regarded as this hair
Bright protection scope.Therefore, it includes preferred embodiment and the institute for falling into the scope of the invention that the following claims are intended to be interpreted as
Have altered and modifies.
Claims (10)
1. a kind of based on the thermal imaging face identification method for generating confrontation network, which is characterized in that main includes generating confrontation mould
Type (one)；Loss function (two)；Training method (three).
2. based on generation confrontation model (one) described in claims 1, which is characterized in that the graphic images X of given input
And the image corresponding real human face image Y, X first pass through image generation module and obtain image G (X), then G (X) and Y is through excessive
It distinguishes that module carries out the calculating of condition cost function, by minimizing the summation of condition cost function and local losses function, obtains
Satisfactory image and face recognition result.
3. based on generation module described in claims 2, which is characterized in that use the depth convolutional network pair of a pre-training
Input picture carries out convolution, pond and Nonlinear Mapping operation, the image after output transform, i.e.,
Wherein, w and h respectively refers to the width and height of picture size.
4. based on resolution module described in claims 2, which is characterized in that the depth convolutional network pair initialized using one
The image received designs a global loss function and controls the learning outcome of image pair to classifying, output figure
As true and false differentiation result and face recognition result, i.e.,Wherein, w and h points
Do not refer to the width and height of picture size.
5. based on condition cost function described in claims 2, which is characterized in that in Y obedience G (X) generation image or really
Image distribution PdataIn the case where, have:
As condition cost function.
6. based on loss function described in claims 1 (two), which is characterized in that including local losses function and global loss
Function.
7. based on local losses function described in claims 6, which is characterized in that respectively on generation module and resolution module
Using L1 norm loss function and identification loss function, specifically, G (X) is obeyed in X and Y and generates image or true picture distribution
PdataIn the case where, have:
As L1 norm loss function；
As identify loss function, DyFor indicator variable, work as yi=1, then face recognition result is that the image is subordinated to face
Xi。
8. based on global loss function described in claims 6, which is characterized in that with the mode of linear, additive by condition cost
FunctionLocal losses functionWithIt is applied to during the entire process of convolutional neural networks, specifically, is taken in X and Y
Image is generated from G (X) or true picture is distributed PdataIn the case where, have:
Wherein, λ1And λ2For weight coefficient.
9. based on training method described in claims 1 (three), which is characterized in that use deep learning training frame
Tensorflow platform, initializes image and training one by one, and initial method uses Adam optimizer, preceding after preset parameter
To with reverse train the number of iterations be 65 times or until convergence when, deconditioning.
10. based on preset parameter described in claims 9, which is characterized in that carry out assignment to the parameter in network, specifically
It is 1) weight coefficient λ1And λ2It is both configured to 100 respectively；2) learning rate of convolutional neural networks is set as 0.0002；3) batch processing
It is dimensioned to 1.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201810748262.4A CN108960159A (en) | 2018-07-10 | 2018-07-10 | A kind of thermal imaging face identification method based on generation confrontation network |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201810748262.4A CN108960159A (en) | 2018-07-10 | 2018-07-10 | A kind of thermal imaging face identification method based on generation confrontation network |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN108960159A true CN108960159A (en) | 2018-12-07 |
Family
ID=64482345
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201810748262.4A Withdrawn CN108960159A (en) | 2018-07-10 | 2018-07-10 | A kind of thermal imaging face identification method based on generation confrontation network |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN108960159A (en) |
Cited By (7)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN110009003A (en) * | 2019-03-14 | 2019-07-12 | 北京旷视科技有限公司 | Training method, the device and system of image procossing and image comparison model |
| CN110309861A (en) * | 2019-06-10 | 2019-10-08 | 浙江大学 | A kind of multi-modal mankind's activity recognition methods based on generation confrontation network |
| CN111968111A (en) * | 2020-09-02 | 2020-11-20 | 广州海兆印丰信息科技有限公司 | Method and device for identifying visceral organs or artifacts of CT (computed tomography) image |
| CN111999731A (en) * | 2020-08-26 | 2020-11-27 | 合肥工业大学 | Electromagnetic backscattering imaging method based on perception generation countermeasure network |
| CN112084962A (en) * | 2020-09-11 | 2020-12-15 | 贵州大学 | Face privacy protection method based on generation type countermeasure network |
| US10970526B1 (en) | 2019-11-01 | 2021-04-06 | Industrial Technology Research Institute | Facial image reconstruction method and system |
| CN113052980A (en) * | 2021-04-27 | 2021-06-29 | 云南大学 | Virtual fitting method and system |
Citations (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN106951867A (en) * | 2017-03-22 | 2017-07-14 | 成都擎天树科技有限公司 | Face identification method, device, system and equipment based on convolutional neural networks |
| CN107154023A (en) * | 2017-05-17 | 2017-09-12 | 电子科技大学 | Face super-resolution reconstruction method based on generation confrontation network and sub-pix convolution |
-
2018
- 2018-07-10 CN CN201810748262.4A patent/CN108960159A/en not_active Withdrawn
Patent Citations (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN106951867A (en) * | 2017-03-22 | 2017-07-14 | 成都擎天树科技有限公司 | Face identification method, device, system and equipment based on convolutional neural networks |
| CN107154023A (en) * | 2017-05-17 | 2017-09-12 | 电子科技大学 | Face super-resolution reconstruction method based on generation confrontation network and sub-pix convolution |
Non-Patent Citations (1)
| Title |
|---|
| TENG ZHANG ET.AL: ""TV-GAN: Generative Adversarial Network Based Thermal to Visible Face Recognition"", 《ARXIV:1712.02514V1 [CS.CV]》 * |
Cited By (10)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN110009003A (en) * | 2019-03-14 | 2019-07-12 | 北京旷视科技有限公司 | Training method, the device and system of image procossing and image comparison model |
| CN110309861A (en) * | 2019-06-10 | 2019-10-08 | 浙江大学 | A kind of multi-modal mankind's activity recognition methods based on generation confrontation network |
| CN110309861B (en) * | 2019-06-10 | 2021-05-25 | 浙江大学 | Multi-modal human activity recognition method based on generation of confrontation network |
| US10970526B1 (en) | 2019-11-01 | 2021-04-06 | Industrial Technology Research Institute | Facial image reconstruction method and system |
| CN111999731A (en) * | 2020-08-26 | 2020-11-27 | 合肥工业大学 | Electromagnetic backscattering imaging method based on perception generation countermeasure network |
| CN111999731B (en) * | 2020-08-26 | 2022-03-22 | 合肥工业大学 | Electromagnetic backscattering imaging method based on perception generation countermeasure network |
| CN111968111A (en) * | 2020-09-02 | 2020-11-20 | 广州海兆印丰信息科技有限公司 | Method and device for identifying visceral organs or artifacts of CT (computed tomography) image |
| CN112084962A (en) * | 2020-09-11 | 2020-12-15 | 贵州大学 | Face privacy protection method based on generation type countermeasure network |
| CN112084962B (en) * | 2020-09-11 | 2021-05-25 | 贵州大学 | Face privacy protection method based on generation type countermeasure network |
| CN113052980A (en) * | 2021-04-27 | 2021-06-29 | 云南大学 | Virtual fitting method and system |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN108960159A (en) | A kind of thermal imaging face identification method based on generation confrontation network | |
| CN112766160B (en) | Face replacement method based on multi-stage attribute encoder and attention mechanism | |
| CN112418095B (en) | Facial expression recognition method and system combined with attention mechanism | |
| KR102629380B1 (en) | Method for Distinguishing a Real Three-Dimensional Object from a Two-Dimensional Spoof of the Real Object | |
| US20200097767A1 (en) | System and method for image de-identification | |
| JP6788264B2 (en) | Facial expression recognition method, facial expression recognition device, computer program and advertisement management system | |
| CN111444881A (en) | Fake face video detection method and device | |
| CN106803069A (en) | Crowd's level of happiness recognition methods based on deep learning | |
| CN105138954A (en) | Image automatic screening, query and identification system | |
| CN103902958A (en) | Method for face recognition | |
| US10062013B2 (en) | Method of image processing | |
| Müller et al. | Convolutional neural networks for semantic segmentation as a tool for multiclass face analysis in thermal infrared | |
| CN108594321A (en) | A kind of Weakly supervised object localization method based on data enhancing | |
| CN109492601A (en) | Face comparison method and device, computer-readable medium and electronic equipment | |
| Tvoroshenko et al. | Current state of development of applications for recognition of faces in the image and frames of video captures | |
| CN107369086A (en) | A kind of identity card stamp system and method | |
| Mui | A statistical multi-experts approach to image classification and segmentation | |
| WO2022246612A1 (en) | Liveness detection method, training method for liveness detection model, apparatus thereof, and system | |
| CN116994206A (en) | Indoor personnel counting method and system based on multi-metadata fusion | |
| Usmani et al. | Efficient deepfake detection using shallow vision transformer | |
| Zambare et al. | Deep Learning Model for Disease Identification of Cotton Plants | |
| Khalid et al. | Multistage deep neural network framework for people detection and localization using fusion of visible and thermal images | |
| Vignesh Baalaji et al. | Autonomous face mask detection using single shot multibox detector, and ResNet-50 with identity retrieval through face matching using deep siamese neural network | |
| Xu | An efficient crowd estimation method using convolutional neural network with thermal images | |
| Tushar et al. | Face Mask and Social Distancing Detection Using Faster Regional Convolutional Neural Network |
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 | ||
| WW01 | Invention patent application withdrawn after publication |
Application publication date: 20181207 |
|
| WW01 | Invention patent application withdrawn after publication |