CN109801232A - A kind of single image to the fog method based on deep learning - Google Patents
A kind of single image to the fog method based on deep learning Download PDFInfo
- Publication number
- CN109801232A CN109801232A CN201811611677.3A CN201811611677A CN109801232A CN 109801232 A CN109801232 A CN 109801232A CN 201811611677 A CN201811611677 A CN 201811611677A CN 109801232 A CN109801232 A CN 109801232A
- Authority
- CN
- China
- Prior art keywords
- image
- neural networks
- convolutional neural
- mist
- depth convolutional
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Pending
Links
Landscapes
- Image Analysis (AREA)
- Image Processing (AREA)
Abstract
The present invention provides a kind of single image to the fog method based on deep learning.This method comprises: obtaining the data set of indoor and outdoor synthesis mist and true mist, which includes: training set, test set and verifying collection；The depth convolutional neural networks of end-to-end (end-to-end) of the construction based on residual error network (Residual Network), the training set and verifying collection are inputted into the depth convolutional neural networks and training pattern, training terminates after reaching maximum number of iterations, optimal models of the depth convolutional neural networks in current iteration are obtained, the defog effect of the depth convolutional neural networks and optimal models is tested with test set；Defogging processing is carried out using the foggy image that depth convolutional neural networks and optimal models treat defogging, obtains fog free images.Method of the invention realizes to the synthesis mist figure of various concentration and outdoor true mist figure and restores image sharpening and visualization end to end, has good defog effect and practical application value.
Description
Technical field
The present invention relates to technical field of image processing more particularly to a kind of single image defogging sides based on deep learning
Method.
Background technique
As smart city and security protection are fast-developing, video monitoring spreads over the every nook and cranny of our lives, however the greasy weather
Gas proposes huge challenge to the intellectual analysis of video image.Because picture quality brings very big shadow to object detection and recognition
It rings, and deteriroation of image quality caused by bad weather factor is particularly acute.Mist reduces the brightness and contrast of image, leads to object
Body thickens.Therefore, under severe weather conditions, the clear image of high quality how is obtained to realize the accurate of target object
Detection, this problem have important researching value in video monitoring intelligent recognition field.
Currently, image enhancement in the prior art and restored method have classical way and deep learning method, wherein classical
Method is divided into image enchancing method and the image recovery method based on physical model again.Image restoration research based on physical model
In, the method for greasy weather scene generated to analyze demisting is mainly simulated by establishing atmospheric scattering physical model.Wherein most write
The algorithm of name is the defogging algorithm preferential based on dark of the propositions such as He Kaiming, and the emphasis of such algorithm is by estimating environment
Light and transmission graph parameter, backstepping export have the corresponding fog free images of mist figure.Although the method for dark channel prior increases compared to image
Strong method can effectively remove haze, but such algorithm is very white in image chunk region, and dark pixel is not present when very bright, so this
It is also the limitation of dark channel prior.
Enhance image comparison by using various methods such as Gamma transformation, Retinex theory etc. based on image enchancing method
It spends to improve the visualization of image, but essentially there is no the physical angle defoggings of the formation from mist for such method, instead
And the partial pixel information that will lead to image is lost.
Summary of the invention
The embodiment provides a kind of single image to the fog method based on deep learning, to overcome the prior art
The problem of.
To achieve the goals above, this invention takes following technical solutions.
A kind of single image to the fog method based on deep learning, comprising:
The data set of indoor and outdoor synthesis mist and true mist is obtained, which includes: training set, test set and verifying collection；
The depth convolutional Neural of end-to-end (end-to-end) of the construction based on residual error network (Residual Network)
The training set and verifying collection are inputted the depth convolutional neural networks and training pattern, reach maximum number of iterations by network
Training terminates afterwards, obtains optimal models of the depth convolutional neural networks in current iteration, tests institute with the test set
State the defog effect of depth convolutional neural networks and optimal models；
Defogging processing is carried out using the foggy image that the depth convolutional neural networks and optimal models treat defogging, is obtained
Fog free images.
Further, the data set of the described synthesis of acquisition indoor and outdoor mist and true mist, the data set include: training set,
Test set and verifying collection, comprising:
Using ChinaMM2018-dehazing racing data collection as data set, which includes: training set, test
Collection and verifying collection；
The training set includes that fog free images 1300 are opened, and corresponding artificial synthesized mist image is 13000 pictures, every nothing
The image 10 that mist image corresponds to the mist of various concentration is opened；The verifying collection is opened comprising fog free images 99, corresponding artificial synthesized mist
Image is 990, and the image 10 that every fog free images correspond to the mist of various concentration is opened；The test set includes 200 indoor conjunctions
Mist formation figure, 200 outdoor synthesis mist figures and the true mist figure in 4469 outdoors.
Further, the construction is based on end-to-end (end-to-end) of residual error network (Residual Network)
Depth convolutional neural networks, the training set and verifying collection are inputted into the depth convolutional neural networks and training pattern, reached
Training terminates after to maximum number of iterations, obtains optimal models of the depth convolutional neural networks in current iteration, comprising:
The depth convolutional neural networks end to end based on residual error network are constructed, which includes 53
Residual block, each residual block include 2 Conv, 2 BN and 1 Prelu, and the depth convolutional neural networks include 106 volumes
The convolution kernel size of lamination, the convolutional layer is 3 × 3, inputs the RGB image in 3 channels, and the convolutional layer schemes the RGB of input
As carrying out feature extraction, batch normalization, down-sampled operation and up-sampling operation；
Network construction, hyper parameter and the loss function of the depth convolutional neural networks are set, and the loss function uses
The training set and verifying collection are inputted the depth convolutional neural networks, utilize PyTorch frame by the mode that L1 and L2 is combined
The training depth convolutional neural networks, in the training process each time grey iterative generation training pattern and defogging as a result, until reaching
Training to the maximum number of iterations of setting, the depth convolutional neural networks terminates；
Mapping relations between the depth convolutional neural networks output foggy image and fog free images, according to the mapping
Relationship exports the verifying and collects corresponding defogging as a result, seeking structural similarity to the corresponding defogging result of verifying collection and original image
(SSIM) value generates SSIM curve, selects the corresponding training pattern of maximum SSIM value in the SSIM curve as the depth
Spend optimal models of the convolutional neural networks in current iteration.
Further, the convolutional layer carries out feature extraction to the RGB image of input, comprising:
The convolutional layer carries out convolution algorithm using the RGB image of 64 3 × 3 filters and 3 channels of input, obtains
To the characteristic pattern in 64 channels, then by after the characteristic pattern in 64 channels and 33 × 3 filters progress convolution algorithms, export 3 channels
RGB image, N number of filter can extract the different characteristics of image of N kind in each convolution operation.
Further, the defogging for testing the depth convolutional neural networks and optimal models with the test set is imitated
Fruit, comprising:
Foggy image is tested using the depth convolutional neural networks and optimal models, the foggy image includes indoor conjunction
Mist formation figure, outdoor synthesis mist figure and outdoor true mist figure, the optimal models of the depth convolutional neural networks export fog free images,
Calculate Y-PSNR (PSNR) and the structure self similarity between the foggy image of input and the fog free images of corresponding output
Property (SSIM), the value of comprehensive SSIM and PSNR measures the defog effects of the depth convolutional neural networks and optimal models.
Further, it is described using the depth convolutional neural networks and optimal models treat the foggy image of defogging into
The processing of row defogging, obtains fog free images, comprising:
Defogging processing, output are carried out using the foggy image that the depth convolutional neural networks and optimal models treat defogging
Fog free images, using Faster R-CNN pre-training model to the foggy image to defogging and corresponding fog free images into
Row target detection, compares the testing result of same target in foggy image and corresponding fog free images to defogging, described in acquisition
The defog effect of foggy image to defogging embodies the importance of defogging processing.
As can be seen from the technical scheme provided by the above-mentioned embodiment of the present invention, the embodiment of the present invention based on deep learning
Image defogging method compared with the Image Restoration Algorithm based on physical model, the present invention uses deep learning algorithm end to end
It realizes and the defogging of haze image is handled, extract the synthesis to various concentration with artificial parameter Estimation without artificial design feature
Mist figure and outdoor true mist figure realize and restore image sharpening and visualization end to end.
The additional aspect of the present invention and advantage will be set forth in part in the description, these will become from the following description
Obviously, or practice through the invention is recognized.
Detailed description of the invention
In order to illustrate the technical solution of the embodiments of the present invention more clearly, required use in being described below to embodiment
Attached drawing be briefly described, it should be apparent that, drawings in the following description are only some embodiments of the invention, for this
For the those of ordinary skill of field, without creative efforts, it can also be obtained according to these attached drawings others
Attached drawing.
Fig. 1 is a kind of overall flow of single image to the fog method based on deep learning provided in an embodiment of the present invention
Figure.
Fig. 2 is a kind of end-to-end for being based on Residual Network (residual error network) provided in an embodiment of the present invention
The structural schematic diagram of the depth convolutional neural networks of (end-to-end).
Fig. 3 is that one kind provided in an embodiment of the present invention selects optimal models procedure chart from SSIM curve.
Fig. 4 (a), 4 (c) are a kind of indoor and outdoors synthesis mist figures provided in an embodiment of the present invention, and 4 (b), 4 (d) be to utilize
Image defogging algorithm of the present invention carries out the correspondence effect picture after defogging.
Fig. 5 (a), 5 (c), 5 (e) are target detection knots before a kind of outdoor true mist figure defogging provided in an embodiment of the present invention
Fruit schematic diagram, Fig. 5 (b), 5 (d), 5 (f) are the testing result schematic diagrames after a kind of defogging provided in an embodiment of the present invention.
Specific embodiment
Embodiments of the present invention are described below in detail, the example of the embodiment is shown in the accompanying drawings, wherein from beginning
Same or similar element or element with the same or similar functions are indicated to same or similar label eventually.Below by ginseng
The embodiment for examining attached drawing description is exemplary, and for explaining only the invention, and is not construed as limiting the claims.
Those skilled in the art of the present technique are appreciated that unless expressly stated, singular " one " used herein, " one
It is a ", " described " and "the" may also comprise plural form.It is to be further understood that being arranged used in specification of the invention
Diction " comprising " refer to that there are the feature, integer, step, operation, element and/or component, but it is not excluded that in the presence of or addition
Other one or more features, integer, step, operation, element, component and/or their group.It should be understood that when we claim member
Part is " connected " or when " coupled " to another element, it can be directly connected or coupled to other elements, or there may also be
Intermediary element.In addition, " connection " used herein or " coupling " may include being wirelessly connected or coupling.Wording used herein
"and/or" includes one or more associated any cells for listing item and all combinations.
Those skilled in the art of the present technique are appreciated that unless otherwise defined, all terms used herein (including technology art
Language and scientific term) there is meaning identical with the general understanding of those of ordinary skill in fields of the present invention.Should also
Understand, those terms such as defined in the general dictionary, which should be understood that, to be had and the meaning in the context of the prior art
The consistent meaning of justice, and unless defined as here, it will not be explained in an idealized or overly formal meaning.
In order to facilitate understanding of embodiments of the present invention, it is done by taking several specific embodiments as an example below in conjunction with attached drawing further
Explanation, and each embodiment does not constitute the restriction to the embodiment of the present invention.
Recently as the fast development of software and hardware, depth convolutional neural networks obtain fast in image field of image processing
Speed development and extensive use.Defogging algorithm based on deep learning realizes that image defogging end to end, i.e. input foggy image are defeated
The effect of fog free images out, the defogging of such algorithm is better than above-mentioned all algorithms.
Under the conditions of the severe greasy weather, image viewing be deteriorated, target often due to image atomization etc. interference cause detection and
Recognition accuracy is low, and false detection rate is high.So in order to improve the accuracy of Target detection and identification in image under mist weather condition, it is right
Haze image pretreatment is very necessary, so the embodiment of the present invention proposes a kind of single image defogging side based on deep learning
Method.
A kind of processing flow schematic diagram of single image to the fog method based on deep learning provided in an embodiment of the present invention
As shown in Figure 1, step specific as follows:
Step 1, data set used in the embodiment of the present invention is ChinaMM2018-dehazing racing data collection；
Step 1.1, training set and verifying collection are off-the-air pictures, and wherein training set is 13000 images, and verifying collection is
990 images.
Step 1.2, it is 1300 that training set, which includes fog free images, and corresponding artificial synthesized mist image is 13000 pictures,
The image 10 that every fog free images correspond to the mist of various concentration is opened；
Step 1.3, verifying collection is 99 comprising fog free images, and corresponding artificial synthesized mist image is 990, and every fogless
The image 10 that image corresponds to the mist of various concentration is opened；
Step 1.4, test set is 200 indoor synthesis mist figures, and 200 outdoor synthesis mist figures and 4469 outdoors are true
Mist figure.
Step 2, design end-to-end (end-to-end) of the one kind based on Residual Network (residual error network)
DCNN (Deep Convolutional Neural Networks, depth convolutional neural networks).
Step 2.1, a kind of framework of above-mentioned depth convolutional neural networks provided in an embodiment of the present invention is as shown in Fig. 2, packet
Containing 53 residual blocks, each residual block includes 2 Conv, 2 BN and 1 Prelu.So the depth convolutional neural networks have
106 convolutional layers.
Step 2.2, the convolution kernel size of the convolutional layer (Conv, convolution) of the depth convolutional neural networks be 3 ×
3, inputting as RGB image is 3 channels,
Extraction characteristic procedure is first converted to 64 channels by 3 channels and is finally reconverted into the output of 3 channels, which is first by 64
After a 3 × 3 filter and 3 channel convolution 64 channels, finally will export 3 after 64 channels and 33 × 3 filter convolution again
The picture in channel, N number of filter can extract the different characteristics of image of N kind in each convolution operation.
Step 2.3, normalization (BN, Batch Normalization) is criticized, gradient explosion occurs in order to prevent, accelerates mind
The training speed of convergence rate and model through network, using batch normalization operation after convolution operation.
Step 2.4, activation primitive activates the Nonlinear Mapping of depth convolutional neural networks model, full in order to reduce gradient
Linear positive unit PReLU (Parametric Rectified Linear Uint) is used with, activation primitive.
Step 2.5, in order to extract the feature of wider scope and reduce the calculating memory of GPU, down-sampled operation is used here
(average pond, Average-pooling), core size is 2 × 2, step-length 2.
Step 2.6, up-sampling operation (Upsample) reduces, institute since front uses down-sampled operating characteristics figure
Purpose with up-sampling is to restore the size of characteristic pattern.
Step 3, the hyper parameter for setting network is being instructed using the above-mentioned depth convolutional neural networks of PyTorch frame training
As a result, until reaching maximum number of iterations, network training terminates for grey iterative generation training pattern and defogging each time during practicing.
Step 3.1, maximum number of iterations is set as 370, Bach size and is set as 13000；
Step 3.2, learning rate is set as 1 × 10-4, to avoid not restraining.
Step 3.3, objective function both loss function in such a way that L1 and L2 are combined.
Step 3.4, the mapping relations between foggy image and fog free images are obtained, final output verifying collects corresponding defogging
As a result.
Step 4, optimal models are selected.
Step 4.1, corresponding defogging result is collected to the verifying and original image seeks SSIM, generate SSIM curve, selected described
Optimal models of the corresponding training pattern of maximum SSIM value as the depth convolutional neural networks in SSIM curve.Fig. 3 is
Optimal models procedure chart is selected in a kind of curve SSIM provided in an embodiment of the present invention.
Step 5, the depth convolutional neural networks and optimal models test indoor synthesis mist figure, outdoor synthesis mist figure are utilized
With outdoor true mist figure, fog free images are exported, herein only include forward-propagating.
Step 5.1, the present invention using indoor and outdoor synthesis mist image and true mist image measurement depth convolutional neural networks and
Optimal models performance.
Step 5.2, using the depth convolutional neural networks and optimal models test foggy image (indoor synthesis mist figure,
Outdoor synthesis mist figure and outdoor true mist figure), fog free images are exported, herein only include forward-propagating.
Step 5.3, output test result as shown in figure 4, Fig. 4 (a), 4 (c) being foggy image, Fig. 4 (b), 4 (d) are to utilize
Image defogging algorithm of the present invention carries out the fog free images obtained after defogging processing.
Step 5.4, the Y-PSNR between the foggy image of input and the fog free images of corresponding output is calculated
(PSNR) and structure self-similarity (SSIM), the value of comprehensive SSIM and PSNR measures depth convolutional neural networks and optimal
The defog effect of model.The PSNR that outdoor can synthesize the more indoor difference of mist figure test result by the following table 1 is low, the reason is that the instruction of this algorithm
Practicing collection is entirely indoor synthesis mist figure.
Table 1
| Defogging algorithm of the present invention | PSNR | SSIM |
| The test of indoor synthesis mist figure | 26.5306 | 0.9275 |
| Outdoor synthesis mist figure test | 23.2270 | 0.9223 |
Step 6, the target before and after detection defogging in image
Step 6.1, defogging is carried out using the foggy image that the depth convolutional neural networks and optimal models treat defogging
Processing exports fog free images.
Step 6.2, using the pre-training model of Faster R-CNN to the foggy image to defogging and corresponding nothing
Mist image carries out target detection, compares the testing result of same target in foggy image and corresponding fog free images to defogging,
The defog effect for obtaining the foggy image to defogging, embodies the importance of defogging processing.
Testing result as shown in figure 5, before outdoor true mist figure defogging object detection results such as Fig. 5 (a), 5 (c), 5 (e) and
Testing result such as Fig. 5 (b), 5 (d), 5 (f) after defogging.From testing result it can be seen that image detection accuracy rate after defogging
It is high, it is seen that it is very significant for first going haze to detect again image under severe greasy weather gas.
In conclusion the image defogging method based on deep learning of the embodiment of the present invention and the image based on physical model
Recovery algorithms are compared, and the present invention is realized using deep learning algorithm end to end and handled the defogging of haze image, without artificial
Design feature extracts and artificial parameter Estimation, and the synthesis mist figure and outdoor true mist figure to various concentration realize extensive end to end
Complex pattern sharpening and visualization have good defog effect and practical application value.
Compared with deep learning algorithm, the image defogging method based on deep learning of the embodiment of the present invention combines more losses
Function, network model have good generalization ability.Feature that network structure of the invention extracts first block and last
It extracts feature to combine, reduces the loss of characteristic information in transmittance process.
Those of ordinary skill in the art will appreciate that: attached drawing is the schematic diagram of one embodiment, module in attached drawing or
Process is not necessarily implemented necessary to the present invention.
As seen through the above description of the embodiments, those skilled in the art can be understood that the present invention can
It realizes by means of software and necessary general hardware platform.Based on this understanding, technical solution of the present invention essence
On in other words the part that contributes to existing technology can be embodied in the form of software products, the computer software product
It can store in storage medium, such as ROM/RAM, magnetic disk, CD, including some instructions are used so that a computer equipment
(can be personal computer, server or the network equipment etc.) executes the certain of each embodiment or embodiment of the invention
Method described in part.
All the embodiments in this specification are described in a progressive manner, same and similar portion between each embodiment
Dividing may refer to each other, and each embodiment focuses on the differences from other embodiments.Especially for device or
For system embodiment, since it is substantially similar to the method embodiment, so describing fairly simple, related place is referring to method
The part of embodiment illustrates.Apparatus and system embodiment described above is only schematical, wherein the conduct
The unit of separate part description may or may not be physically separated, component shown as a unit can be or
Person may not be physical unit, it can and it is in one place, or may be distributed over multiple network units.It can root
According to actual need that some or all of the modules therein is selected to achieve the purpose of the solution of this embodiment.Ordinary skill
Personnel can understand and implement without creative efforts.
The foregoing is only a preferred embodiment of the present invention, but scope of protection of the present invention is not limited thereto,
In the technical scope disclosed by the present invention, any changes or substitutions that can be easily thought of by anyone skilled in the art,
It should be covered by the protection scope of the present invention.Therefore, protection scope of the present invention should be with scope of protection of the claims
Subject to.
Claims (6)
1. a kind of single image to the fog method based on deep learning characterized by comprising
The data set of indoor and outdoor synthesis mist and true mist is obtained, which includes: training set, test set and verifying collection；
The depth convolutional neural networks for constructing the end-to-end end-to-end based on residual error network Residual Network, by institute
It states training set and verifying collection inputs the depth convolutional neural networks and training pattern, reach training knot after maximum number of iterations
Beam obtains optimal models of the depth convolutional neural networks in current iteration, tests the depth with the test set and rolls up
The defog effect of product neural network and optimal models；
Defogging processing is carried out using the foggy image that the depth convolutional neural networks and optimal models treat defogging, is obtained fogless
Image.
2. the method according to claim 1, wherein the data of acquisition indoor and outdoor the synthesis mist and true mist
Collection, the data set include: training set, test set and verifying collection, comprising:
Using ChinaMM2018-dehazing racing data collection as data set, the data set include: training set, test set and
Verifying collection；
The training set includes that fog free images 1300 are opened, and corresponding artificial synthesized mist image is 13000 pictures, every fogless figure
As the image 10 of the mist of corresponding various concentration is opened；The verifying collection is opened comprising fog free images 99, corresponding artificial synthesized mist image
It is 990, the image 10 that every fog free images correspond to the mist of various concentration is opened；The test set includes 200 indoor synthesis mists
Figure, 200 outdoor synthesis mist figures and the true mist figure in 4469 outdoors.
3. according to claim 1 or claim 2, which is characterized in that the construction is end-to-end based on residual error network
Depth convolutional neural networks, the training set and verifying collection are inputted into the depth convolutional neural networks and training pattern, reached
Training terminates after to maximum number of iterations, obtains optimal models of the depth convolutional neural networks in current iteration, comprising:
The depth convolutional neural networks end to end based on residual error network are constructed, which includes 53 residual errors
Block, each residual block include 2 Conv, 2 BN and 1 Prelu, and the depth convolutional neural networks include 106 convolutional layers,
The convolution kernel size of the convolutional layer is 3 × 3, inputs the RGB image in 3 channels, and the convolutional layer carries out the RGB image of input
Feature extraction, batch normalization, down-sampled operation and up-sampling operation；
Network construction, hyper parameter and the loss function of the depth convolutional neural networks be set, the loss function using L1 and
The training set and verifying collection are inputted the depth convolutional neural networks, utilize the training of PyTorch frame by the mode that L2 is combined
The depth convolutional neural networks, in the training process each time grey iterative generation training pattern and defogging as a result, until reaching and setting
The training of fixed maximum number of iterations, the depth convolutional neural networks terminates；
Mapping relations between the depth convolutional neural networks output foggy image and fog free images, according to the mapping relations
It exports the verifying and collects corresponding defogging as a result, seeking structural similarity SSIM to the corresponding defogging result of verifying collection and original image
Value generates SSIM curve, selects the corresponding training pattern of maximum SSIM value in the SSIM curve as the depth convolution
Optimal models of the neural network in current iteration.
4. according to the method described in claim 3, it is characterized in that, the convolutional layer carries out feature to the RGB image of input
It extracts, comprising:
The convolutional layer carries out convolution algorithm using the RGB image of 64 3 × 3 filters and 3 channels of input, obtains 64
The characteristic pattern in channel, then by after the characteristic pattern in 64 channels and 33 × 3 filters progress convolution algorithms, the RGB in 3 channels of output schemes
Picture, N number of filter can extract the different characteristics of image of N kind in each convolution operation.
5. according to the method described in claim 3, it is characterized in that, described test the depth convolution mind with the test set
Defog effect through network and optimal models, comprising:
Foggy image is tested using the depth convolutional neural networks and optimal models, the foggy image includes indoor synthesis mist
Figure, outdoor synthesis mist figure and outdoor true mist figure, the optimal models of the depth convolutional neural networks export fog free images, calculate
Y-PSNR (PSNR) and structure self-similarity between the fog free images of the foggy image and corresponding output that input out
(SSIM), the value of SSIM and PSNR is integrated to measure the defog effect of the depth convolutional neural networks and optimal models.
6. according to the method described in claim 5, it is characterized in that, described using the depth convolutional neural networks and optimal
The foggy image that model treats defogging carries out defogging processing, obtains fog free images, comprising:
Defogging processing is carried out using the foggy image that the depth convolutional neural networks and optimal models treat defogging, is exported fogless
Image carries out mesh to the foggy image to defogging and corresponding fog free images using the pre-training model of Faster R-CNN
Mark detection, compares the testing result of same target in foggy image and corresponding fog free images to defogging, obtains described wait go
The defog effect of the foggy image of mist embodies the importance of defogging processing.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201811611677.3A CN109801232A (en) | 2018-12-27 | 2018-12-27 | A kind of single image to the fog method based on deep learning |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201811611677.3A CN109801232A (en) | 2018-12-27 | 2018-12-27 | A kind of single image to the fog method based on deep learning |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN109801232A true CN109801232A (en) | 2019-05-24 |
Family
ID=66557741
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201811611677.3A Pending CN109801232A (en) | 2018-12-27 | 2018-12-27 | A kind of single image to the fog method based on deep learning |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN109801232A (en) |
Cited By (9)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN110310238A (en) * | 2019-06-18 | 2019-10-08 | 华南农业大学 | A kind of removing rain based on single image method based on the compression rewards and punishments neural network for reusing raw information |
| CN110517203A (en) * | 2019-08-30 | 2019-11-29 | 山东工商学院 | A kind of defogging method rebuild based on reference picture |
| CN110738622A (en) * | 2019-10-17 | 2020-01-31 | 温州大学 | Lightweight neural network single image defogging method based on multi-scale convolution |
| CN111046828A (en) * | 2019-12-20 | 2020-04-21 | 西安交通大学 | Dust removal and noise reduction neural network method for mine underground monitoring image |
| CN111539896A (en) * | 2020-04-30 | 2020-08-14 | 华中科技大学 | Domain-adaptive-based image defogging method and system |
| CN111598793A (en) * | 2020-04-24 | 2020-08-28 | 云南电网有限责任公司电力科学研究院 | Method and system for defogging image of power transmission line and storage medium |
| CN111681178A (en) * | 2020-05-22 | 2020-09-18 | 厦门大学 | Knowledge distillation-based image defogging method |
| CN111899169A (en) * | 2020-07-02 | 2020-11-06 | 佛山市南海区广工大数控装备协同创新研究院 | Network segmentation method of face image based on semantic segmentation |
| CN112634171A (en) * | 2020-12-31 | 2021-04-09 | 上海海事大学 | Image defogging method based on Bayes convolutional neural network and storage medium |
Citations (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN107256541A (en) * | 2017-06-15 | 2017-10-17 | 北京航空航天大学 | A kind of multi-spectral remote sensing image defogging method based on convolutional neural networks |
| CN108230264A (en) * | 2017-12-11 | 2018-06-29 | 华南农业大学 | A kind of single image to the fog method based on ResNet neural networks |
| CN108876728A (en) * | 2018-04-20 | 2018-11-23 | 南京理工大学 | Single image to the fog method based on residual error study |
| CN108876737A (en) * | 2018-06-06 | 2018-11-23 | 武汉大学 | A kind of image de-noising method of joint residual error study and structural similarity |
-
2018
- 2018-12-27 CN CN201811611677.3A patent/CN109801232A/en active Pending
Patent Citations (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN107256541A (en) * | 2017-06-15 | 2017-10-17 | 北京航空航天大学 | A kind of multi-spectral remote sensing image defogging method based on convolutional neural networks |
| CN108230264A (en) * | 2017-12-11 | 2018-06-29 | 华南农业大学 | A kind of single image to the fog method based on ResNet neural networks |
| CN108876728A (en) * | 2018-04-20 | 2018-11-23 | 南京理工大学 | Single image to the fog method based on residual error study |
| CN108876737A (en) * | 2018-06-06 | 2018-11-23 | 武汉大学 | A kind of image de-noising method of joint residual error study and structural similarity |
Non-Patent Citations (1)
| Title |
|---|
| JINJIANG LI，ET AL: "Image Dehazing Using Residual-Based Deep CNN", 《DIGITAL OBJECT IDENTIFIER》 * |
Cited By (14)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN110310238A (en) * | 2019-06-18 | 2019-10-08 | 华南农业大学 | A kind of removing rain based on single image method based on the compression rewards and punishments neural network for reusing raw information |
| CN110517203A (en) * | 2019-08-30 | 2019-11-29 | 山东工商学院 | A kind of defogging method rebuild based on reference picture |
| CN110517203B (en) * | 2019-08-30 | 2023-06-23 | 山东工商学院 | Defogging method based on reference image reconstruction |
| CN110738622A (en) * | 2019-10-17 | 2020-01-31 | 温州大学 | Lightweight neural network single image defogging method based on multi-scale convolution |
| CN111046828A (en) * | 2019-12-20 | 2020-04-21 | 西安交通大学 | Dust removal and noise reduction neural network method for mine underground monitoring image |
| CN111598793A (en) * | 2020-04-24 | 2020-08-28 | 云南电网有限责任公司电力科学研究院 | Method and system for defogging image of power transmission line and storage medium |
| CN111539896B (en) * | 2020-04-30 | 2022-05-27 | 华中科技大学 | Domain-adaptive-based image defogging method and system |
| CN111539896A (en) * | 2020-04-30 | 2020-08-14 | 华中科技大学 | Domain-adaptive-based image defogging method and system |
| CN111681178A (en) * | 2020-05-22 | 2020-09-18 | 厦门大学 | Knowledge distillation-based image defogging method |
| CN111681178B (en) * | 2020-05-22 | 2022-04-26 | 厦门大学 | Knowledge distillation-based image defogging method |
| CN111899169A (en) * | 2020-07-02 | 2020-11-06 | 佛山市南海区广工大数控装备协同创新研究院 | Network segmentation method of face image based on semantic segmentation |
| CN111899169B (en) * | 2020-07-02 | 2024-01-26 | 佛山市南海区广工大数控装备协同创新研究院 | Method for segmenting network of face image based on semantic segmentation |
| CN112634171A (en) * | 2020-12-31 | 2021-04-09 | 上海海事大学 | Image defogging method based on Bayes convolutional neural network and storage medium |
| CN112634171B (en) * | 2020-12-31 | 2023-09-29 | 上海海事大学 | Image defogging method and storage medium based on Bayesian convolutional neural network |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN109801232A (en) | A kind of single image to the fog method based on deep learning | |
| CN110414670A (en) | A kind of image mosaic tampering location method based on full convolutional neural networks | |
| CN112699786B (en) | Video behavior identification method and system based on space enhancement module | |
| CN109615582A (en) | A kind of face image super-resolution reconstruction method generating confrontation network based on attribute description | |
| CN111080567A (en) | Remote sensing image fusion method and system based on multi-scale dynamic convolution neural network | |
| CN111046821B (en) | Video behavior recognition method and system and electronic equipment | |
| CN110570363A (en) | Image defogging method based on Cycle-GAN with pyramid pooling and multi-scale discriminator | |
| CN112767279B (en) | Underwater image enhancement method for generating countermeasure network based on discrete wavelet integration | |
| CN112507920B (en) | Examination abnormal behavior identification method based on time displacement and attention mechanism | |
| CN113810611B (en) | Data simulation method and device for event camera | |
| CN112257741B (en) | Method for detecting generative anti-false picture based on complex neural network | |
| CN109949200A (en) | Steganalysis framework establishment method based on filter subset selection and CNN | |
| CN113628297A (en) | COVID-19 deep learning diagnosis system based on attention mechanism and transfer learning | |
| CN116994044A (en) | Construction method of image anomaly detection model based on mask multi-mode generation countermeasure network | |
| CN113627504B (en) | Multi-mode multi-scale feature fusion target detection method based on generation of countermeasure network | |
| CN117391938B (en) | Infrared image super-resolution reconstruction method, system, equipment and terminal | |
| CN109409305A (en) | A kind of facial image clarity evaluation method and device | |
| Krishnan et al. | A novel underwater image enhancement technique using ResNet | |
| CN114820395B (en) | Underwater image enhancement method based on multi-field information fusion | |
| CN116703750A (en) | Image defogging method and system based on edge attention and multi-order differential loss | |
| CN116309171A (en) | Method and device for enhancing monitoring image of power transmission line | |
| CN115984949A (en) | Low-quality face image recognition method and device with attention mechanism | |
| Jiang et al. | Haze relevant feature attention network for single image dehazing | |
| CN116415019A (en) | Virtual reality VR image recognition method and device, electronic equipment and storage medium | |
| CN114332103A (en) | Image segmentation method based on improved FastFCN |
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 |
Application publication date: 20190524 |
|
| RJ01 | Rejection of invention patent application after publication |