CN110443759A - A kind of image defogging method based on deep learning - Google Patents
A kind of image defogging method based on deep learning Download PDFInfo
- Publication number
- CN110443759A CN110443759A CN201910614615.6A CN201910614615A CN110443759A CN 110443759 A CN110443759 A CN 110443759A CN 201910614615 A CN201910614615 A CN 201910614615A CN 110443759 A CN110443759 A CN 110443759A
- Authority
- CN
- China
- Prior art keywords
- mist
- layer
- defogging
- concentration map
- neural networks
- 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
- 238000000034 method Methods 0.000 title claims abstract description 57
- 238000013135 deep learning Methods 0.000 title claims abstract description 15
- 239000003595 mist Substances 0.000 claims abstract description 63
- 238000012549 training Methods 0.000 claims abstract description 30
- 238000013527 convolutional neural network Methods 0.000 claims abstract description 17
- 238000009792 diffusion process Methods 0.000 claims abstract description 10
- 230000015572 biosynthetic process Effects 0.000 claims abstract description 6
- 238000003786 synthesis reaction Methods 0.000 claims abstract description 6
- 238000000605 extraction Methods 0.000 claims description 9
- 230000004927 fusion Effects 0.000 claims description 9
- 238000013507 mapping Methods 0.000 claims description 7
- 238000013528 artificial neural network Methods 0.000 claims description 4
- 238000003475 lamination Methods 0.000 claims description 3
- 238000005070 sampling Methods 0.000 claims description 3
- 239000000203 mixture Substances 0.000 claims 1
- 238000012805 post-processing Methods 0.000 abstract description 6
- 230000003044 adaptive effect Effects 0.000 abstract description 4
- 230000000694 effects Effects 0.000 description 8
- 238000009826 distribution Methods 0.000 description 3
- 238000005516 engineering process Methods 0.000 description 3
- 230000007547 defect Effects 0.000 description 2
- 230000013016 learning Effects 0.000 description 2
- 238000010606 normalization Methods 0.000 description 2
- 238000003860 storage Methods 0.000 description 2
- 238000002834 transmittance Methods 0.000 description 2
- ORILYTVJVMAKLC-UHFFFAOYSA-N Adamantane Natural products C1C(C2)CC3CC1CC2C3 ORILYTVJVMAKLC-UHFFFAOYSA-N 0.000 description 1
- 230000004913 activation Effects 0.000 description 1
- 238000000149 argon plasma sintering Methods 0.000 description 1
- 230000005540 biological transmission Effects 0.000 description 1
- 150000001875 compounds Chemical class 0.000 description 1
- 238000010276 construction Methods 0.000 description 1
- 238000011161 development Methods 0.000 description 1
- 239000000284 extract Substances 0.000 description 1
- 230000006870 function Effects 0.000 description 1
- 238000012544 monitoring process Methods 0.000 description 1
- 230000003287 optical effect Effects 0.000 description 1
- 230000010181 polygamy Effects 0.000 description 1
- 238000002360 preparation method Methods 0.000 description 1
- 238000011160 research Methods 0.000 description 1
- 238000007619 statistical method Methods 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
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T5/00—Image enhancement or restoration
- G06T5/90—Dynamic range modification of images or parts thereof
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T2207/00—Indexing scheme for image analysis or image enhancement
- G06T2207/10—Image acquisition modality
- G06T2207/10004—Still image; Photographic image
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T2207/00—Indexing scheme for image analysis or image enhancement
- G06T2207/10—Image acquisition modality
- G06T2207/10016—Video; Image sequence
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T2207/00—Indexing scheme for image analysis or image enhancement
- G06T2207/20—Special algorithmic details
- G06T2207/20081—Training; Learning
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T2207/00—Indexing scheme for image analysis or image enhancement
- G06T2207/20—Special algorithmic details
- G06T2207/20084—Artificial neural networks [ANN]
Landscapes
- Physics & Mathematics (AREA)
- General Physics & Mathematics (AREA)
- Engineering & Computer Science (AREA)
- Theoretical Computer Science (AREA)
- Image Processing (AREA)
- Image Analysis (AREA)
Abstract
The invention discloses a kind of image defogging method based on deep learning.Then the present invention generates the transmissivity and atmosphere light of different values, and manually have mist figure as training set by the synthesis of atmosphere photon diffusion models by collecting nature picture as fogless figure at random in the different zones of picture；Construct the convolutional neural networks for predicting mist concentration map；The training set training convolutional neural networks are utilized later；Recently enter it is to be processed have mist figure, mist concentration map is calculated using the convolutional neural networks, obtains final defogging figure with there is mist figure to subtract mist concentration map.The present invention is a kind of adaptive defogging method, the defogging result of generation compare naturally, and strong robustness, it is applied widely, can be applied to simultaneously in indoor and natural scene defogging；This method is a kind of full-automatic, defogging method end to end, does not need post-processing step.
Description
Technical field
The present invention relates to deep learnings and computer vision field, and in particular to a kind of image defogging based on deep learning
Method.
Background technique
Image defogging method is the important research content of computer vision field, is widely used in video monitoring, automatically
The fields such as driving.In recent years, serious with haze, the visibility and ornamental value of picture are seriously affected, to image defogging skill
More stringent requirements are proposed for the effect and versatility of art.Existing image defogging method can be roughly divided into two kinds: be based on image
The defogging method of priori and defogging method based on deep learning.
Based on the defogging method of image prior by the contrast of image, atmosphere light scattering is acquired the features such as distribution of color
Transmittance figure in model, and then air light value is gone out by the smallest point estimation of a part of transmissivity, mould is scattered using atmosphere light
Type recovers defogging figure.Method based on priori is limited to the applicability of priori, cannot be effective in the scene for not meeting priori
Defogging.Such method is poor there is also operation efficiency and transmissivity estimates the problems such as inaccurate, and defog effect is difficult to reach each
The requirement of kind computer vision application.
With the development of depth learning technology, there are many methods for carrying out defogging using neural network.It is most of this
Class method is based on atmosphere photon diffusion models, by estimating that transmissivity and air light value restore to obtain fogless figure in mist figure from having.By
It is accurate to estimate that transmittance figure and air light value are extremely difficult in the diversification and complexity of defogging scene.Recent method passes through
The effect of defogging is improved using fairly large network structure or by post-processing step, increase network structure can be brought general
Change ability is poor, the low problem of operational efficiency.Meanwhile by the insufficient influence of training set, the model that complicated network training goes out is past
It is poor toward the extensive effect on other pictures.And increase post-processing step then and be since the capability of fitting of network is insufficient, it cannot
Realize adaptive end-to-end defogging, the overhead of post-processing equally reduces the efficiency of defogging.
The way for estimating parameter according to atmosphere photon diffusion models has been abandoned there are also certain methods, then has directly been indicated with network
From there is mapping relations of the mist figure to fogless figure.These methods can improve the quality of image to a certain extent, but due to discontented
Sufficient physical model, there are larger differences for the result and true picture of these methods, be easy to cause color to deflect, it cannot be guaranteed that defogging
The quality of figure.
To sum up, although the defogging method based on deep learning is achieved better than traditional effect based on transcendental method, but is gone
There are also very big rooms for promotion for fog effect and efficiency.
An existing method is the defogging method of Combined estimator atmosphere light and transmissivity, comes from paper AOD-Net:All-
in-One Dehazing Network.The technology deforms atmosphere photon diffusion models, converts parameter for defogging problem
Then forecasting problem constructs 7 with multiple short circuit connections first according to atmosphere photon diffusion models compound training data
Layer network structure, training network obtain defogging model.
The shortcomings that this method, is:
First, training data only has more than 1000 in above-mentioned defogging method, and all it is indoor scene, trains
Model is poor in the defog effect of various outdoor scenes.
Second, the number of plies of network only has 7 layers, and capability of fitting is poor in above-mentioned defogging method, the defogging when mistiness degree is higher
As a result middle fog residual is obvious.
Summary of the invention
The purpose of the present invention is overcoming the shortcomings of existing methods, a kind of image defogging side based on deep learning is proposed
Method.There are two problems solved by the invention is main: lacking training data under true outdoor scene for existing method, lead to model
The defect poor to real scene effect, the present invention utilize high definition outdoor images, according to atmosphere photon diffusion models in image difference
Each region of scale adds mist respectively, constructs the data set including more than 54000 outdoor foggy images.For existing
The defect of light-duty defogging network capability of fitting difference, the present invention utilize short-cut structure, and the residual error for devising similar U-Net is gone
The output of mist network, network is mist concentration map, has mist figure to subtract mist concentration map and obtains defogging result.
To solve the above-mentioned problems, the invention proposes a kind of the image defogging method based on deep learning, the method
Include:
Collect nature picture as fogless figure, then the different zones of picture generate at random different values transmissivity and
Atmosphere light, and manually have mist figure as training set by the synthesis of atmosphere photon diffusion models；
Construct the convolutional neural networks for predicting mist concentration map；
It is described for predicting the convolutional neural networks of mist concentration map using training set training；
Input it is to be processed have mist figure, mistiness degree is calculated using the convolutional neural networks for predicting mist concentration map
Figure, obtains final defogging figure with there is mist figure to subtract mist concentration map.
Preferably, the synthesis manually has mist figure to specifically include as training set:
Collect all Outdoor Scene pictures in SUN data set；
The Outdoor Scene picture is divided into the rectangular area of 128*128 size；
It is random to generate atmosphere light and transmissivity for each rectangular area, to form the Outdoor Scene picture
It is corresponding to have mist figure；
By Outdoor Scene picture and corresponding there is mist figure to zoom to 128*128 size.What is generated has the fogless figure of mist figure to structure
At one group of training data, all data constitute whole training sets.
It is preferably, described for predicting the convolutional neural networks of mist concentration map, specifically:
The network is divided into feature extraction layer, Fusion Features layer, Feature Mapping layer, shares 10 layers, is a full convolutional network；
In characteristic extraction part, sufficiently being extracted using the convolutional layer that 4 convolution kernel sizes are 3 has the feature in mist figure to believe
Breath, and the convolutional layer for being 1 by two convolution kernels carries out down-sampling twice；
In Fusion Features part, characteristic pattern is amplified to original image size using the warp lamination that two step-lengths are 2, is made simultaneously
The shallow-layer information in fusion the deep information and characteristic extraction part is connected with short circuit；
In Feature Mapping part, the dimension of characteristic pattern is reduced using the convolutional layer that two convolution kernel sizes are 1, is finally obtained
Mist concentration map of the output of 128*128*3 as prediction；
The network architecture reference structure of residual error neural network, in network third layer and layer 7, the second layer and the 8th layer
Between be provided with two short-circuit structures, third, two layers of characteristic pattern are incorporated to the seven, the eight layers.
A kind of image defogging method based on deep learning proposed by the present invention, this method are a kind of adaptive defogging sides
The defogging result of method, generation compare naturally, and strong robustness, it is applied widely, can be applied to simultaneously indoor and natural scene
In defogging；This method is a kind of full-automatic, defogging method end to end, does not need post-processing step.
Detailed description of the invention
In order to more clearly explain the embodiment of the invention or the technical proposal in the existing technology, to embodiment or will show below
There is attached drawing needed in technical description to be briefly described, it should be apparent that, the accompanying drawings in the following description is only this
Some embodiments of invention for those of ordinary skill in the art without creative efforts, can be with
Other attached drawings are obtained according to these attached drawings.
Fig. 1 is the image defogging method overview flow chart of the embodiment of the present invention；
Fig. 2 is the network structure of the embodiment of the present invention.
Specific embodiment
Following will be combined with the drawings in the embodiments of the present invention, and technical solution in the embodiment of the present invention carries out clear, complete
Site preparation description, it is clear that described embodiments are only a part of the embodiments of the present invention, instead of all the embodiments.It is based on
Embodiment in the present invention, it is obtained by those of ordinary skill in the art without making creative efforts every other
Embodiment shall fall within the protection scope of the present invention.
Fig. 1 is the image defogging method overview flow chart of the embodiment of the present invention, as shown in Figure 1, this method comprises:
S1 collects nature picture as fogless figure, then generates the transmission of different values at random in the different zones of picture
Rate and atmosphere light, and manually have mist figure as training set by the synthesis of atmosphere photon diffusion models；
S2 constructs the convolutional neural networks for predicting mist concentration map；
S3, it is described for predicting the convolutional neural networks of mist concentration map using training set training；
S4, input is to be processed mist figure, and mist is calculated using the convolutional neural networks for predicting mist concentration map
Concentration map obtains final defogging figure with there is mist figure to subtract mist concentration map.
Step S1, specific as follows:
S1-1: all Outdoor Scene pictures in collection SUN data set, more than totally 54000.
S1-2: the picture in S1-1 is divided into the rectangular area of 128*128 size.
S1-3: random to generate atmosphere light and transmissivity for each rectangular area in S1-2.By statistical analysis, greatly
The main value range of gas light is between [0.7,1], and the main value range of transmissivity is between [0.3,0.7], therefore this hair
It is bright that the value range of the two is fixed in above-mentioned section when randomly selecting.According to the characteristic of transmissivity local continuous, this hair
It is bright to keep the atmosphere light in each rectangular area identical with transmissivity.It is produced by the formula of following atmosphere photon diffusion models by original image J
There is mist figure I.X in formula represents the position of each pixel in picture, and t (x) represents transmissivity, and A represents air light value.
I (x)=J (x) t (x)+A (1-t (x))
S1-4: the picture in S1-1 and S1-3 is zoomed into 128*128 size.The set for having the fogless figure pair of mist figure generated
One group of training data is constituted, all 54000 groups of data constitute whole training sets.Not due to the image size in SUN data set
It unanimously, the rectangular area for the 128*128 size chosen in S1-2 can be big with different proportion in representative image after zooming in and out
Small region, so that the data diversity in training set is more abundant.Therefore it is trained using training set proposed by the present invention
Defogging network more meets the requirement of the image defogging in complicated natural scene.
Data set construction method proposed by the present invention, it is contemplated that the diversity of mist in natural scene, in obtained data set,
Area size, regional location, atmosphere light and the transmissivity of every picture fog distribution be all it is random, it is multiple to meet defogging scene
The requirement of polygamy also increases the robustness of network.The atmosphere optical range being arranged in S1-3 is by dark channel prior method
Estimation obtains from a large amount of pictures, can satisfy the versatility of model.In addition, since the data set is based in SUN data set
Outdoor Scene picture building, the quality of original image is preferable, avoids model and is interfered by noise etc..
Step S2, network structure is as shown in Fig. 2, specific as follows:
What is proposed in the present invention is used to predict that the convolutional neural networks of mist concentration map are divided into feature extraction layer, Fusion Features
Three layer, Feature Mapping layer part, share 10 layers, are a full convolutional networks.In characteristic extraction part, 4 convolution kernels are used
The convolutional layer that size is 3 sufficiently extracts the characteristic information having in mist figure, and the convolutional layer for being 1 by two convolution kernels carries out twice
Down-sampling.In Fusion Features part, characteristic pattern is amplified to original image size using the warp lamination that two step-lengths are 2, is made simultaneously
The shallow-layer information in fusion the deep information and characteristic extraction part is connected with short circuit.In Feature Mapping part, two convolution are used
The convolutional layer that core size is 1 reduces the dimension of characteristic pattern, finally obtains mist concentration map of the output of 128*128*3 as prediction.
The details of above-mentioned network structure is as shown in the table, in addition, being added to batch normalization layer behind each convolutional layer
(batch normalization) and Relu activation primitive layer, is omitted in table:
Network structure scale of the invention is smaller, and the number of convolution kernel is fewer, and having only used size is 3,2 and
1 convolution kernel saves the cost of training and prediction and the quantity of parameter, while keeping the generalization ability of model more preferable.Network knot
Structure is provided with two articles with reference to the structure of residual error neural network between network third layer and layer 7, the second layer and the 8th layer
Third, two layers of characteristic pattern are incorporated to the seven, the eight layers by short-circuit structure.This short-circuit structure helps to merge the figure of low middle higher-dimension
As feature, network-evaluated mist concentration distribution is made to be more nearly legitimate reading.
Step S3, specific as follows:
Defogging method in the present invention is needed using caffe frame training network by following steps:
S3-1: convolutional neural networks structure described in S2 is realized using caffe frame.
S3-2: trained loss function is set are as follows:
Loss=∑ | | J- (I-Iconv)||2
Wherein IconvIt is the mist concentration map predicted for the output of convolutional neural networks.
S3-3: setting learning rate is 0.01, and with Adam gradient optimizing method undated parameter, momentum is set as 0.9, is passed through
Final model is obtained after 95000 iteration.
Step S4, specific as follows:
S4-1: input has mist picture I.
S4-2: there is mist picture by the model treatment generated in S3, the image of output is the mist concentration map predicted
Iconv, final defogging figure is arrived with there is mist figure to subtract mist concentration mapThis process may be expressed as:
A kind of image defogging method based on deep learning that the embodiment of the present invention proposes, this method are a kind of adaptive
The defogging result of defogging method, generation compare naturally, and strong robustness, it is applied widely, can be applied to simultaneously indoor and natural
In the defogging of scene；This method is a kind of full-automatic, defogging method end to end, does not need post-processing step.
Those of ordinary skill in the art will appreciate that all or part of the steps in the various methods of above-described embodiment is can
It is completed with instructing relevant hardware by program, which can be stored in a computer readable storage medium, storage
Medium may include: read-only memory (ROM, Read Only Memory), random access memory (RAM, Random
Access Memory), disk or CD etc..
It has been carried out in detail in addition, being provided for the embodiments of the invention a kind of image defogging method based on deep learning above
Thin to introduce, used herein a specific example illustrates the principle and implementation of the invention, and above embodiments are said
It is bright to be merely used to help understand method and its core concept of the invention；At the same time, for those skilled in the art, foundation
Thought of the invention, there will be changes in the specific implementation manner and application range, in conclusion the content of the present specification is not
It is interpreted as limitation of the present invention.
Claims (3)
1. a kind of image defogging method based on deep learning, which is characterized in that the described method includes:
Nature picture is collected as fogless figure, then generates the transmissivity and atmosphere of different values at random in the different zones of picture
Light, and manually have mist figure as training set by the synthesis of atmosphere photon diffusion models；
Construct the convolutional neural networks for predicting mist concentration map；
It is described for predicting the convolutional neural networks of mist concentration map using training set training；
Input it is to be processed have mist figure, mist concentration map is calculated using the convolutional neural networks for predicting mist concentration map,
Final defogging figure is obtained with there is mist figure to subtract mist concentration map.
2. a kind of image defogging method based on deep learning as described in claim 1, which is characterized in that the synthesis is artificial
There is the step of mist figure is as training set, specifically include:
Collect all Outdoor Scene pictures in SUN data set；
The Outdoor Scene picture is divided into the rectangular area of 128*128 size；
It is random to generate atmosphere light and transmissivity for each rectangular area, so that it is corresponding to form the Outdoor Scene picture
Have mist figure；
By Outdoor Scene picture and corresponding there is mist figure to zoom to 128*128 size.What is generated has the fogless figure of mist figure to composition one
Group training data, all data constitute whole training sets.
3. a kind of image defogging method based on deep learning as described in claim 1, which is characterized in that described for predicting
The convolutional neural networks of mist concentration map, specifically:
The network is divided into feature extraction layer, Fusion Features layer, Feature Mapping layer, shares 10 layers, is a full convolutional network；
In characteristic extraction part, the characteristic information having in mist figure is sufficiently extracted using the convolutional layer that 4 convolution kernel sizes are 3, and
Down-sampling twice is carried out by the convolutional layer that two convolution kernels are 1；
In Fusion Features part, characteristic pattern is amplified to original image size by the warp lamination for the use of two step-lengths being 2, while using short
Shallow-layer information in road connection fusion the deep information and characteristic extraction part；
In Feature Mapping part, the dimension of characteristic pattern is reduced using the convolutional layer that two convolution kernel sizes are 1, finally obtains 128*
Mist concentration map of the output of 128*3 as prediction；
The network architecture reference structure of residual error neural network, between network third layer and layer 7, the second layer and the 8th layer
Provided with two short-circuit structures, third, two layers of characteristic pattern are incorporated to the seven, the eight layers.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201910614615.6A CN110443759A (en) | 2019-07-09 | 2019-07-09 | A kind of image defogging method based on deep learning |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201910614615.6A CN110443759A (en) | 2019-07-09 | 2019-07-09 | A kind of image defogging method based on deep learning |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN110443759A true CN110443759A (en) | 2019-11-12 |
Family
ID=68429957
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201910614615.6A Pending CN110443759A (en) | 2019-07-09 | 2019-07-09 | A kind of image defogging method based on deep learning |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN110443759A (en) |
Cited By (6)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN111091516A (en) * | 2019-12-24 | 2020-05-01 | 广州柏视医疗科技有限公司 | Anti-scattering grating method and device based on artificial intelligence |
| CN111539250A (en) * | 2020-03-12 | 2020-08-14 | 上海交通大学 | Image fog concentration estimation method, system and terminal based on neural network |
| CN111681180A (en) * | 2020-05-25 | 2020-09-18 | 厦门大学 | Priori-driven deep learning image defogging method |
| CN111861923A (en) * | 2020-07-21 | 2020-10-30 | 济南大学 | Target identification method and system based on lightweight residual error network image defogging |
| CN113052124A (en) * | 2021-04-09 | 2021-06-29 | 济南博观智能科技有限公司 | Identification method and device for fogging scene and computer-readable storage medium |
| CN114648467A (en) * | 2022-05-18 | 2022-06-21 | 中山大学深圳研究院 | Image defogging method and device, terminal equipment and computer readable storage medium |
-
2019
- 2019-07-09 CN CN201910614615.6A patent/CN110443759A/en active Pending
Non-Patent Citations (1)
| Title |
|---|
| YINGHONG LIAO 等: ""HDP-Net: Haze Density Prediction Network for Nighttime Dehazing"", 《PCM 2018: ADVANCES IN MULTIMEDIA INFORMATION PROCESSING – PCM 2018》 * |
Cited By (9)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN111091516A (en) * | 2019-12-24 | 2020-05-01 | 广州柏视医疗科技有限公司 | Anti-scattering grating method and device based on artificial intelligence |
| CN111539250A (en) * | 2020-03-12 | 2020-08-14 | 上海交通大学 | Image fog concentration estimation method, system and terminal based on neural network |
| CN111539250B (en) * | 2020-03-12 | 2024-02-27 | 上海交通大学 | Image fog concentration estimation method, system and terminal based on neural network |
| CN111681180A (en) * | 2020-05-25 | 2020-09-18 | 厦门大学 | Priori-driven deep learning image defogging method |
| CN111681180B (en) * | 2020-05-25 | 2022-04-26 | 厦门大学 | Priori-driven deep learning image defogging method |
| CN111861923A (en) * | 2020-07-21 | 2020-10-30 | 济南大学 | Target identification method and system based on lightweight residual error network image defogging |
| CN113052124A (en) * | 2021-04-09 | 2021-06-29 | 济南博观智能科技有限公司 | Identification method and device for fogging scene and computer-readable storage medium |
| CN113052124B (en) * | 2021-04-09 | 2023-02-10 | 济南博观智能科技有限公司 | Identification method and device for fogging scene and computer readable storage medium |
| CN114648467A (en) * | 2022-05-18 | 2022-06-21 | 中山大学深圳研究院 | Image defogging method and device, terminal equipment and computer readable storage medium |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN110443759A (en) | A kind of image defogging method based on deep learning | |
| CN106910175B (en) | Single image defogging algorithm based on deep learning | |
| CN109447907B (en) | Single image enhancement method based on full convolution neural network | |
| CN111539887B (en) | Channel attention mechanism and layered learning neural network image defogging method based on mixed convolution | |
| CN111489372B (en) | Video foreground and background separation method based on cascade convolution neural network | |
| Chen et al. | Haze removal using radial basis function networks for visibility restoration applications | |
| CN111161277A (en) | Natural image matting method based on deep learning | |
| CN111739082B (en) | Stereo vision unsupervised depth estimation method based on convolutional neural network | |
| CN112070690B (en) | Single image rain removing method based on convolution neural network double-branch attention generation | |
| CN110956094A (en) | RGB-D multi-mode fusion personnel detection method based on asymmetric double-current network | |
| CN109146810A (en) | A kind of image defogging method based on end-to-end deep learning | |
| CN102741879A (en) | Method for generating depth maps from monocular images and systems using the same | |
| CN109584188B (en) | Image defogging method based on convolutional neural network | |
| CN111079507B (en) | Behavior recognition method and device, computer device and readable storage medium | |
| CN109509156A (en) | A kind of image defogging processing method based on generation confrontation model | |
| CN107749048B (en) | Image correction system and method, and color blindness image correction system and method | |
| CN113077505A (en) | Optimization method of monocular depth estimation network based on contrast learning | |
| CN110705366A (en) | Real-time human head detection method based on stair scene | |
| CN111598793A (en) | Method and system for defogging image of power transmission line and storage medium | |
| Qian et al. | CIASM-Net: a novel convolutional neural network for dehazing image | |
| CN109685735B (en) | Single picture defogging method based on fog layer smoothing prior | |
| Zhao et al. | Image dehazing based on haze degree classification | |
| CN118334365A (en) | Novel RGB-D image saliency target detection method | |
| Kerim et al. | NOVA: Rendering virtual worlds with humans for computer vision tasks | |
| CN117456449A (en) | Efficient cross-modal crowd counting method based on specific information |
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: 20191112 |