CN108665432A - A kind of single image to the fog method based on generation confrontation network - Google Patents
A kind of single image to the fog method based on generation confrontation network Download PDFInfo
- Publication number
- CN108665432A CN108665432A CN201810480313.XA CN201810480313A CN108665432A CN 108665432 A CN108665432 A CN 108665432A CN 201810480313 A CN201810480313 A CN 201810480313A CN 108665432 A CN108665432 A CN 108665432A
- Authority
- CN
- China
- Prior art keywords
- image
- layer
- network model
- convolution
- fog
- 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- Withdrawn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
- 238000000034 method Methods 0.000 title claims abstract description 18
- 239000003595 mist Substances 0.000 claims abstract description 28
- 238000012549 training Methods 0.000 claims abstract description 21
- 238000012360 testing method Methods 0.000 claims abstract description 15
- 230000008030 elimination Effects 0.000 claims abstract description 12
- 238000003379 elimination reaction Methods 0.000 claims abstract description 12
- 238000004364 calculation method Methods 0.000 claims abstract description 10
- 238000012545 processing Methods 0.000 claims abstract description 4
- 238000005070 sampling Methods 0.000 claims description 17
- 230000004913 activation Effects 0.000 claims description 8
- 239000012141 concentrate Substances 0.000 claims description 3
- 238000010606 normalization Methods 0.000 claims description 3
- 238000010276 construction Methods 0.000 claims 1
- 238000013461 design Methods 0.000 abstract description 3
- 230000000694 effects Effects 0.000 abstract description 3
- 230000006870 function Effects 0.000 description 15
- 238000013527 convolutional neural network Methods 0.000 description 2
- 230000015556 catabolic process Effects 0.000 description 1
- 238000013135 deep learning Methods 0.000 description 1
- 230000007812 deficiency Effects 0.000 description 1
- 238000006731 degradation reaction Methods 0.000 description 1
- 230000006866 deterioration Effects 0.000 description 1
- 238000002474 experimental method Methods 0.000 description 1
- 239000002245 particle Substances 0.000 description 1
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T5/00—Image enhancement or restoration
- G06T5/73—Deblurring; Sharpening
-
- 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
Landscapes
- Engineering & Computer Science (AREA)
- Physics & Mathematics (AREA)
- Theoretical Computer Science (AREA)
- General Physics & Mathematics (AREA)
- Data Mining & Analysis (AREA)
- Molecular Biology (AREA)
- Biophysics (AREA)
- Computational Linguistics (AREA)
- Artificial Intelligence (AREA)
- Evolutionary Computation (AREA)
- General Health & Medical Sciences (AREA)
- Biomedical Technology (AREA)
- Computing Systems (AREA)
- General Engineering & Computer Science (AREA)
- Life Sciences & Earth Sciences (AREA)
- Mathematical Physics (AREA)
- Software Systems (AREA)
- Health & Medical Sciences (AREA)
- Image Analysis (AREA)
Abstract
The present invention provides a kind of based on the single image to the fog method for generating confrontation network, step 1, generator network model is built, by by adding the training sample set of mist processing to be input in generator network model, generates the preliminary mist elimination image of fog free images in Emulation Testing sample set；Step 2, build decision device network model, preliminary mist elimination image is input in the decision device network model, calculate cost function, step 2.1, if cost function calculation result is less than pre-set defogging threshold value, input picture is judged for the fog free images of test sample collection, and using the generator network model as optimal training pattern；Step 2.2, otherwise, then judge the preliminary mist elimination image that input picture generates for generator network model, trained using tensorflow and generate confrontation network, go to step 2；Step 3, foggy image collection is inputted into optimal training pattern, exports the image after defogging.The present invention has the advantages that design science, highly practical, easy to operate and defog effect are high.
Description
Technical field
The present invention relates to single image defogging technical field, specifically, relates to a kind of based on generating confrontation network
Single image to the fog method.
Background technology
Under haze weather, there are many airborne particulates in air.These particles not only absorb and scatter the reflection of scene
Light, but also some atmosphere lights are scattered into camera, the image deterioration for causing camera to obtain so that picture contrast is low,
Poor visibility, quality degradation.
Currently, image defogging algorithm can be mainly divided into three classes：The first kind is to be based on image enhancement, but increase based on image
Certain information characteristics of image can be lost by force.Second class is the image restoration based on physical model, the purpose of Image Restoration Algorithm
It is the clearly image naturally for obtaining that there is good visibility, while keeping good color restorability；Based on hazy condition
The worsening reason of hypograph establishes the physical model of atmospheric scattering, it is necessary first to physical parameter model is estimated, as atmosphere light is shone
Intensity and transmissivity (depth), it is then inverse to solve the physical model to obtain fog free images, but the image based on physical model is multiple
Original place reason is limited in scope；Third class is the image defogging algorithm based on deep learning, such as convolutional neural networks applied to image
Defogging.
In order to solve the above problems, people are seeking always a kind of ideal technical solution.
Invention content
The purpose of the present invention is in view of the deficiencies of the prior art, to provide a kind of design science, highly practical, operation letter
Just high based on the single image to the fog method for generating confrontation network with defog effect.
To achieve the goals above, the technical solution adopted in the present invention is：A kind of single width based on generation confrontation network
Image defogging method, obtain fog free images collection be used as test sample collection, to the fog free images collection utilize image processing software into
Row plus mist handle to obtain foggy image collection as training sample set, which further includes,
Step 1, generator network model is built, it will be by adding the training sample set of mist processing to be input to the generator net
In network model, generates and imitate the preliminary mist elimination image that the test sample concentrates fog free images；
Step 2, decision device network model is built, the preliminary mist elimination image is input to the decision device network model
In, cost function is calculated,
Step 2.1, if cost function calculation result is less than pre-set defogging threshold value, judge input picture for test
Fog free images in sample set, and using the generator network model as optimal training pattern；
Step 2.2, if cost function calculation result is more than pre-set defogging threshold value, input picture is judged to generate
The preliminary mist elimination image that device network model generates is trained using tensorflow and generates confrontation network, updates generator network mould
Type goes to step 2；
Step 3, training sample set is inputted into optimal training pattern, obtains the image after defogging.
Based on above-mentioned, the generator network model includes coder structure and decoder architecture,
The coder structure is eight layers of convolutional network structure, is arranged after every layer of convolutional network structure
BatchNormalization layers and prelu activation primitives；Every layer of convolutional network structure carry out down-sampling, split 2, eight layers
Convolution number is respectively 64-128-254-512-512-512-512-512；Convolution size is 4*4, and the size of input picture is
256*256*3 is exported wherein 3 indicate port number as one-dimensional vector；
The decoder architecture is eight layer network structures, and every layer includes 4*4 deconvolution, BatchNormalization successively
Layer and prelu activation primitives, every layer of convolutional network structure up-sampled, and convolution size is 4*4, eight layers of decoder architecture
Convolution number is set as：The final result of 512-512-512-512-512-254-128-64, every layer of decoder are itself convolution
As a result it is added with symmetrical coder structure convolutional layer, every layer of actual convolution number is decoder architecture and coder structure
Addition：512-1024-1024-1024-1024-512-254-128；The size for exporting image is 256*256*3.
Based on above-mentioned, the decision device network model includes four layers of down-sampling layer and a judgement layer, and input image size is
256*256*3, output result are one-dimensional；Wherein, the convolution kernel size of every layer of down-sampling layer is 4*4, stride 2, input picture
Often pass through one layer of down-sampling layer, length and wide size reduce half, every layer of down-sampling layer include successively convolution,
The number of Normalization layers of Batch and prelu activation primitives, the convolution of every layer of down-sampling layer is 64-128-254-
512；The judgement layer is one-dimensional vector, and convolution kernel size is 4*4, and stride 1, convolution number is 1.
Based on above-mentioned, the cost function loss calculation formula are：
Loss=lossGAN+λlossMSE
Wherein, λ indicates adjustable parameter, lossGANIt indicates to generate confrontation network cost function, lossMSEIndicate that image is square
Difference.
Based on above-mentioned, acquisition Middlebury Stereo Datasets and the figure bright and fogless in online download
As constituting fog free images collection as test sample collection；It is artificial to the fog free images collection using Adobe lightroom CC softwares
Add mist, obtains foggy image collection as training sample set.
The present invention has substantive distinguishing features outstanding and significant progress compared with the prior art, and specifically, the present invention carries
A kind of single image to the fog method based on generation confrontation network has been supplied, convolutional neural networks are applied and are generating confrontation network
On GAN, generator network model and decision device network model are built, adds the training sample set that mist is handled as input, obtains
Take fog free images collection as test sample collection as canonical reference；The present invention effectively improves picture quality, with design department
It learns, advantage highly practical, easy to operate and high defog effect.
Specific implementation mode
Below by specific implementation mode, technical scheme of the present invention will be described in further detail.
A kind of single image to the fog method based on generation confrontation network obtains fog free images collection as test sample first
Collection, the fog free images collection is carried out using image processing software plus mist handles to obtain foggy image collection as training sample set,
The single image to the fog method further includes,
Step 1, generator network model is built, it will be by adding the training sample set of mist processing to be input to the generator net
In network model, generates and imitate the preliminary mist elimination image that the test sample concentrates fog free images；
Step 2, decision device network model is built, the preliminary mist elimination image is input to the decision device network model
In, cost function is calculated,
Step 2.1, if cost function calculation result is less than pre-set defogging threshold value, judge input picture for test
The fog free images of sample set, and using the generator network model as optimal training pattern；
Step 2.2, if cost function calculation result is more than pre-set defogging threshold value, input picture is judged to generate
The preliminary mist elimination image that device network model generates is trained using tensorflow and generates confrontation network, updates generator network mould
Type goes to step 2；
Step 3, foggy image collection is inputted into optimal training pattern, exports the image after defogging.
The present invention gives a kind of specific implementation modes obtaining test sample collection and training sample set, obtain
Middlebury Stereo Datasets and the image bright and fogless in online download, constitute fog free images collection；Profit
Mist is manually added to the fog free images collection with Adobe lightroom CC softwares, obtains foggy image collection.Specifically,
The fog free images collection includes the various scenes in indoor and outdoor, due to being difficult to obtain Same Scene to have mist and nothing in reality
The image pair of mist, the existing image defogging algorithm based on study, foggy image is mostly by depth map via atmospheric scattering mould
It is artificial synthesized to be randomly provided parameter for type；Atmospherical scattering model formula is I=Jt+A (1-t)；Network inputs foggy image, it is defeated
Go out foggy image transmissivity, then calculates fog free images using backstepping.
In order to simplify the above process so that image defogging method of the invention can be directly over generation confrontation network
Fog free images are obtained, are artificially the fog free images using the dehaze functions of the lightroom CC softwares of Adobe companies
Collection carries out plus mist；Meanwhile in order to adapt to the mistiness degree under the conditions of different weather, the spy of study to different mistiness degree images
Fog free images have been assembled the mist that concentration is respectively 10,20,30,40,50,60,70,80,90,100, have obtained mist figure by sign
Image set, the input as generator network model；The image length and width of foggy image collection are cut into the size of 256*256, with
Adapt to decision device network structure.
Specifically, the generator network model includes coder structure and decoder architecture, the coder structure is
Eight layers of convolutional network structure, every layer of convolutional network structure are all carrying out down-sampling, split 2；After every layer of convolutional network structure
There are BatchNormalization layers and prelu activation primitives, improves the non-linear and generalization ability of network；Convolution size is 4*
4, the size of input picture 256*256*3, wherein 3 indicate port number, final output one-dimensional vector；Eight layers of convolutional network structure
Convolution number is respectively 64-128-254-512-512-512-512-512, can regard every layer of convolution as and all carry out down-sampling,
Split is 2.
The decoder architecture be eight layer network structures, every layer successively include deconvolution, BatchNormalization and
Prelu is constantly up-sampled, the image of the 256*256*3 identical as input image size of final output one, and convolution size is 4*4.
The decoder architecture is equivalent to the inverse process of the coder structure, and therefore, generator network model is symmetrical knot
Structure.To keep the structures such as bottom, the final result of every layer of the decoder architecture is itself convolution results and symmetrical encoder
Convolutional layer is added.Eight layers of convolution number of the decoder architecture, which are arranged, is：512-512-512-512-512-254-128-64, and
Every layer of actual convolution number is that the decoder architecture is added with the coder structure：512-1024-1024-1024-
1024-512-254-128。
Specifically, the decision device network model includes four layers of down-sampling layer and a judgement layer, the volume of every layer of down-sampling layer
Product core size is 4*4, stride 2, and the number of layer convolution is 64-128-254-512, input picture often pass through one layer it is described under adopt
Sample layer, long and wide size reduce half；Every layer of down-sampling layer successively include convolution, Normalization layers of Batch and
Prelu activation primitives；Input image size is 256*256*3, and the judgement layer is one-dimensional vector, and convolution kernel size is 4*4, step
Width is 1, and convolution number is 1, and output result is one-dimensional.
Specifically, cost function loss consists of two parts.A part makes a living into confrontation network cost function lossGAN, one
Part is image mean-squared deviation lossMSE, the cost function loss calculation formula are：
Loss=lossGAN+λlossMSE
Wherein, λ indicates adjustable parameter, lossGANIt indicates to generate confrontation network cost function, lossMSEIndicate that image is square
Difference.
Specifically, the decision device network model and cost function：The foggy image collection is inputted into the decision device net
Network model calculates cost function；By multiple contrast and experiment, following optimized parameter is obtained, maximum iteration is set as
2000000 times, learning rate 0.0002, λ 100；Then it is trained under tensorflow frames, obtains trained network mould
Shape parameter.
Finally it should be noted that：The above embodiments are merely illustrative of the technical scheme of the present invention and are not intended to be limiting thereof；To the greatest extent
The present invention is described in detail with reference to preferred embodiments for pipe, those of ordinary skills in the art should understand that：Still
It can modify to the specific implementation mode of the present invention or equivalent replacement is carried out to some technical characteristics；Without departing from this hair
The spirit of bright technical solution should all cover within the scope of the technical scheme claimed by the invention.
Claims (5)
1. a kind of based on the single image to the fog method for generating confrontation network, acquisition fog free images collection is right as test sample collection
The fog free images collection is carried out using image processing software plus mist handles to obtain foggy image collection as training sample set, feature
It is：The single image to the fog method further includes,
Step 1, generator network model is built, it will be by adding the training sample set of mist processing to be input to the generator network mould
In type, generates and imitate the preliminary mist elimination image that the test sample concentrates fog free images；
Step 2, decision device network model is built, the preliminary mist elimination image is input in the decision device network model, is counted
Cost function is calculated,
Step 2.1, if cost function calculation result is less than pre-set defogging threshold value, judge input picture for test sample
The fog free images of concentration, and using the generator network model as optimal training pattern；
Step 2.2, if cost function calculation result is more than pre-set defogging threshold value, judge input picture for generator net
The preliminary mist elimination image that network model generates is trained using tensorflow and generates confrontation network, update generator network model, turned
Step 2；
Step 3, training sample set is inputted into optimal training pattern, obtains the image after defogging.
2. according to claim 1 based on the single image to the fog method for generating confrontation network, it is characterised in that：The life
Network model of growing up to be a useful person includes coder structure and decoder architecture,
The coder structure is eight layers of convolutional network structure, is arranged after every layer of convolutional network structure
BatchNormalization layers and prelu activation primitives；Every layer of convolutional network structure carry out down-sampling, split 2, eight layers
Convolution number is respectively 64-128-254-512-512-512-512-512；Convolution size is 4*4, and the size of input picture is
256*256*3 is exported wherein 3 indicate port number as one-dimensional vector；
The decoder architecture be eight layer network structures, every layer successively include 4*4 deconvolution, BatchNormalization layers and
Prelu activation primitives, every layer of convolutional network structure are being up-sampled, and convolution size is 4*4, eight layers of convolution of decoder architecture
Number is set as：The final result of 512-512-512-512-512-254-128-64, every layer of decoder are itself convolution results
It is added with symmetrical coder structure convolutional layer, every layer of actual convolution number is the phase of decoder architecture and coder structure
Add：512-1024-1024-1024-1024-512-254-128；The size for exporting image is 256*256*3.
3. according to claim 2 based on the single image to the fog method for generating confrontation network, it is characterised in that：It is described to sentence
Certainly device network model includes four layers of down-sampling layer and a judgement layer, and input image size 256*256*3, output result is one
Dimension；Wherein, the convolution kernel size of every layer of down-sampling layer be 4*4, stride 2, input picture often pass through one layer of down-sampling layer,
It is grown and wide size reduces half, every layer of down-sampling layer include successively convolution, Normalization layers of Batch and
The number of prelu activation primitives, the convolution of every layer of down-sampling layer is 64-128-254-512；The judgement layer is one-dimensional vector,
Convolution kernel size is 4*4, and stride 1, convolution number is 1.
4. according to claim 3 based on the single image to the fog method for generating confrontation network, it is characterised in that：The generation
Valence function loss calculation formula are：
Loss=lossGAN+λlossMSE
Wherein, λ indicates adjustable parameter, lossGANIt indicates to generate confrontation network cost function, lossMSEIndicate image mean-squared deviation.
5. according to claim 4 based on the single image to the fog method for generating confrontation network, it is characterised in that：It obtains
Middlebury Stereo Datasets and the image construction fog free images collection conduct bright and fogless in online download
Test sample collection；Mist is manually added to the fog free images collection using Adobe lightroom CC softwares, obtains foggy image collection
As training sample set.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201810480313.XA CN108665432A (en) | 2018-05-18 | 2018-05-18 | A kind of single image to the fog method based on generation confrontation network |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201810480313.XA CN108665432A (en) | 2018-05-18 | 2018-05-18 | A kind of single image to the fog method based on generation confrontation network |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN108665432A true CN108665432A (en) | 2018-10-16 |
Family
ID=63776912
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201810480313.XA Withdrawn CN108665432A (en) | 2018-05-18 | 2018-05-18 | A kind of single image to the fog method based on generation confrontation network |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN108665432A (en) |
Cited By (19)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN109389569A (en) * | 2018-10-26 | 2019-02-26 | 大象智能科技(南京)有限公司 | Based on the real-time defogging method of monitor video for improving DehazeNet |
| CN109472818A (en) * | 2018-10-17 | 2019-03-15 | 天津大学 | A kind of image defogging method based on deep neural network |
| CN109509156A (en) * | 2018-10-31 | 2019-03-22 | 聚时科技(上海)有限公司 | A kind of image defogging processing method based on generation confrontation model |
| CN109816610A (en) * | 2019-01-31 | 2019-05-28 | 上海工程技术大学 | A kind of image defogging system |
| CN109903232A (en) * | 2018-12-20 | 2019-06-18 | 江南大学 | A kind of image defogging method based on convolutional neural networks |
| CN110245677A (en) * | 2019-04-24 | 2019-09-17 | 广东工业大学 | A kind of image descriptor dimension reduction method based on convolution self-encoding encoder |
| CN110288550A (en) * | 2019-06-28 | 2019-09-27 | 中国人民解放***箭军工程大学 | The single image defogging method of confrontation network is generated based on priori knowledge guiding conditions |
| CN110390336A (en) * | 2019-06-05 | 2019-10-29 | 广东工业大学 | A method of improving Feature Points Matching precision |
| CN110517203A (en) * | 2019-08-30 | 2019-11-29 | 山东工商学院 | A kind of defogging method rebuild based on reference picture |
| CN110544213A (en) * | 2019-08-06 | 2019-12-06 | 天津大学 | Image defogging method based on global and local feature fusion |
| CN110766640A (en) * | 2019-11-05 | 2020-02-07 | 中山大学 | Image defogging method based on depth semantic segmentation |
| CN110837825A (en) * | 2019-10-17 | 2020-02-25 | 天津大学 | Meter identification and reading system based on embedded type |
| CN111179202A (en) * | 2019-12-31 | 2020-05-19 | 内蒙古工业大学 | Single image defogging enhancement method and system based on generation countermeasure network |
| CN111709888A (en) * | 2020-06-03 | 2020-09-25 | 中科九度(北京)空间信息技术有限责任公司 | Aerial image defogging method based on improved generation countermeasure network |
| CN111724310A (en) * | 2019-03-21 | 2020-09-29 | 马上消费金融股份有限公司 | Training method of image restoration model, image restoration method and device |
| CN111861936A (en) * | 2020-07-29 | 2020-10-30 | 北京字节跳动网络技术有限公司 | Image defogging method and device, electronic equipment and computer readable storage medium |
| WO2020233709A1 (en) * | 2019-05-22 | 2020-11-26 | 华为技术有限公司 | Model compression method, and device |
| CN109410144B (en) * | 2018-10-31 | 2020-11-27 | 聚时科技(上海)有限公司 | End-to-end image defogging processing method based on deep learning |
| CN112950521A (en) * | 2021-04-27 | 2021-06-11 | 上海海事大学 | Image defogging method and generator network |
Citations (5)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN106127702A (en) * | 2016-06-17 | 2016-11-16 | 兰州理工大学 | A kind of image mist elimination algorithm based on degree of depth study |
| CN106910175A (en) * | 2017-02-28 | 2017-06-30 | 武汉大学 | A kind of single image defogging algorithm based on deep learning |
| CN107729852A (en) * | 2017-10-24 | 2018-02-23 | 贵阳宏益房地产开发有限公司 | Facial information register method, apparatus and system |
| CN107798669A (en) * | 2017-12-08 | 2018-03-13 | 北京小米移动软件有限公司 | Image defogging method, device and computer-readable recording medium |
| CN107968962A (en) * | 2017-12-12 | 2018-04-27 | 华中科技大学 | A kind of video generation method of the non-conterminous image of two frames based on deep learning |
-
2018
- 2018-05-18 CN CN201810480313.XA patent/CN108665432A/en not_active Withdrawn
Patent Citations (5)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN106127702A (en) * | 2016-06-17 | 2016-11-16 | 兰州理工大学 | A kind of image mist elimination algorithm based on degree of depth study |
| CN106910175A (en) * | 2017-02-28 | 2017-06-30 | 武汉大学 | A kind of single image defogging algorithm based on deep learning |
| CN107729852A (en) * | 2017-10-24 | 2018-02-23 | 贵阳宏益房地产开发有限公司 | Facial information register method, apparatus and system |
| CN107798669A (en) * | 2017-12-08 | 2018-03-13 | 北京小米移动软件有限公司 | Image defogging method, device and computer-readable recording medium |
| CN107968962A (en) * | 2017-12-12 | 2018-04-27 | 华中科技大学 | A kind of video generation method of the non-conterminous image of two frames based on deep learning |
Non-Patent Citations (1)
| Title |
|---|
| 贾绪仲 等: ""一种基于条件生成对抗网络的去雾方法"", 《信息与电脑》 * |
Cited By (29)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN109472818A (en) * | 2018-10-17 | 2019-03-15 | 天津大学 | A kind of image defogging method based on deep neural network |
| CN109472818B (en) * | 2018-10-17 | 2021-07-02 | 天津大学 | Image defogging method based on deep neural network |
| CN109389569A (en) * | 2018-10-26 | 2019-02-26 | 大象智能科技(南京)有限公司 | Based on the real-time defogging method of monitor video for improving DehazeNet |
| CN109509156A (en) * | 2018-10-31 | 2019-03-22 | 聚时科技(上海)有限公司 | A kind of image defogging processing method based on generation confrontation model |
| CN109509156B (en) * | 2018-10-31 | 2021-02-05 | 聚时科技(上海)有限公司 | Image defogging processing method based on generation countermeasure model |
| CN109410144B (en) * | 2018-10-31 | 2020-11-27 | 聚时科技(上海)有限公司 | End-to-end image defogging processing method based on deep learning |
| CN109903232A (en) * | 2018-12-20 | 2019-06-18 | 江南大学 | A kind of image defogging method based on convolutional neural networks |
| CN109816610A (en) * | 2019-01-31 | 2019-05-28 | 上海工程技术大学 | A kind of image defogging system |
| CN111724310A (en) * | 2019-03-21 | 2020-09-29 | 马上消费金融股份有限公司 | Training method of image restoration model, image restoration method and device |
| CN111724310B (en) * | 2019-03-21 | 2023-08-15 | 马上消费金融股份有限公司 | Training method of image restoration model, image restoration method and device |
| CN110245677A (en) * | 2019-04-24 | 2019-09-17 | 广东工业大学 | A kind of image descriptor dimension reduction method based on convolution self-encoding encoder |
| WO2020233709A1 (en) * | 2019-05-22 | 2020-11-26 | 华为技术有限公司 | Model compression method, and device |
| CN110390336A (en) * | 2019-06-05 | 2019-10-29 | 广东工业大学 | A method of improving Feature Points Matching precision |
| CN110288550A (en) * | 2019-06-28 | 2019-09-27 | 中国人民解放***箭军工程大学 | The single image defogging method of confrontation network is generated based on priori knowledge guiding conditions |
| CN110288550B (en) * | 2019-06-28 | 2020-04-24 | 中国人民解放***箭军工程大学 | Single-image defogging method for generating countermeasure network based on priori knowledge guiding condition |
| CN110544213A (en) * | 2019-08-06 | 2019-12-06 | 天津大学 | Image defogging method based on global and local feature fusion |
| CN110544213B (en) * | 2019-08-06 | 2023-06-13 | 天津大学 | Image defogging method based on global and local feature fusion |
| CN110517203B (en) * | 2019-08-30 | 2023-06-23 | 山东工商学院 | Defogging method based on reference image reconstruction |
| CN110517203A (en) * | 2019-08-30 | 2019-11-29 | 山东工商学院 | A kind of defogging method rebuild based on reference picture |
| CN110837825A (en) * | 2019-10-17 | 2020-02-25 | 天津大学 | Meter identification and reading system based on embedded type |
| CN110766640A (en) * | 2019-11-05 | 2020-02-07 | 中山大学 | Image defogging method based on depth semantic segmentation |
| CN111179202B (en) * | 2019-12-31 | 2022-09-02 | 内蒙古工业大学 | Single image defogging enhancement method and system based on generation countermeasure network |
| CN111179202A (en) * | 2019-12-31 | 2020-05-19 | 内蒙古工业大学 | Single image defogging enhancement method and system based on generation countermeasure network |
| CN111709888A (en) * | 2020-06-03 | 2020-09-25 | 中科九度(北京)空间信息技术有限责任公司 | Aerial image defogging method based on improved generation countermeasure network |
| CN111709888B (en) * | 2020-06-03 | 2023-12-08 | 中科九度(北京)空间信息技术有限责任公司 | Aerial image defogging method based on improved generation countermeasure network |
| CN111861936B (en) * | 2020-07-29 | 2023-03-24 | 抖音视界有限公司 | Image defogging method and device, electronic equipment and computer readable storage medium |
| CN111861936A (en) * | 2020-07-29 | 2020-10-30 | 北京字节跳动网络技术有限公司 | Image defogging method and device, electronic equipment and computer readable storage medium |
| CN112950521A (en) * | 2021-04-27 | 2021-06-11 | 上海海事大学 | Image defogging method and generator network |
| CN112950521B (en) * | 2021-04-27 | 2024-03-01 | 上海海事大学 | Image defogging method and generator network |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN108665432A (en) | A kind of single image to the fog method based on generation confrontation network | |
| CN105574827B (en) | A kind of method, apparatus of image defogging | |
| CN106910175B (en) | Single image defogging algorithm based on deep learning | |
| CN109255831A (en) | The method that single-view face three-dimensional reconstruction and texture based on multi-task learning generate | |
| Turk | Generating textures on arbitrary surfaces using reaction-diffusion | |
| CN111738942A (en) | Generation countermeasure network image defogging method fusing feature pyramid | |
| CN108416745B (en) | Image self-adaptive defogging enhancement method with color constancy | |
| CN106846261A (en) | Underwater picture processing method based on convolutional neural networks | |
| CN110288550A (en) | The single image defogging method of confrontation network is generated based on priori knowledge guiding conditions | |
| CN109255758A (en) | Image enchancing method based on full 1*1 convolutional neural networks | |
| CN110517203A (en) | A kind of defogging method rebuild based on reference picture | |
| Ma et al. | Image-based air pollution estimation using hybrid convolutional neural network | |
| CN109509156A (en) | A kind of image defogging processing method based on generation confrontation model | |
| Zhang et al. | Fast learning radiance fields by shooting much fewer rays | |
| CN109829868A (en) | A kind of lightweight deep learning model image defogging method, electronic equipment and medium | |
| Zheng et al. | Neural relightable participating media rendering | |
| Zamuda et al. | Environmental framework to visualize emergent artificial forest ecosystems | |
| CN116777805A (en) | Rain scene image synthesis method and device learned from rendering and storage medium | |
| CN108876728A (en) | Single image to the fog method based on residual error study | |
| CN112258420A (en) | DQN-based image enhancement processing method and device | |
| CN108269241A (en) | With reference to greasy weather road image scene depth and the defogging algorithm of road image feature | |
| CN107833198A (en) | A kind of Outdoor Scene weight illumination method decomposed based on large scale | |
| Darles et al. | Accelerating and enhancing rendering of realistic ocean scenes | |
| Barile et al. | Animated drawings rendered by genetic programming | |
| CN102867325B (en) | Plant leaf scab rendering method |
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 | ||
| WW01 | Invention patent application withdrawn after publication |
Application publication date: 20181016 |