CN110211052A - A kind of single image to the fog method based on feature learning - Google Patents
A kind of single image to the fog method based on feature learning Download PDFInfo
- Publication number
- CN110211052A CN110211052A CN201910246074.6A CN201910246074A CN110211052A CN 110211052 A CN110211052 A CN 110211052A CN 201910246074 A CN201910246074 A CN 201910246074A CN 110211052 A CN110211052 A CN 110211052A
- Authority
- CN
- China
- Prior art keywords
- image
- model
- fog
- defogging
- picture
- 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 46
- 230000005540 biological transmission Effects 0.000 claims abstract description 26
- 238000013135 deep learning Methods 0.000 claims abstract description 10
- 238000000605 extraction Methods 0.000 claims abstract description 8
- 238000013507 mapping Methods 0.000 claims abstract description 5
- 238000013136 deep learning model Methods 0.000 claims abstract description 3
- 238000013527 convolutional neural network Methods 0.000 claims description 18
- 238000012545 processing Methods 0.000 claims description 18
- 239000003595 mist Substances 0.000 claims description 12
- 238000012549 training Methods 0.000 claims description 12
- 230000008030 elimination Effects 0.000 claims description 9
- 238000003379 elimination reaction Methods 0.000 claims description 9
- 230000011218 segmentation Effects 0.000 claims description 9
- 230000004913 activation Effects 0.000 claims description 6
- 238000011084 recovery Methods 0.000 claims description 4
- 230000004048 modification Effects 0.000 claims description 3
- 238000012986 modification Methods 0.000 claims description 3
- 238000005457 optimization Methods 0.000 claims description 3
- 238000013461 design Methods 0.000 claims description 2
- 230000008676 import Effects 0.000 claims description 2
- 238000012360 testing method Methods 0.000 claims description 2
- 230000003044 adaptive effect Effects 0.000 abstract description 4
- 238000005516 engineering process Methods 0.000 description 4
- 230000008569 process Effects 0.000 description 4
- 230000000593 degrading effect Effects 0.000 description 3
- 230000000694 effects Effects 0.000 description 3
- 239000002245 particle Substances 0.000 description 3
- 230000000007 visual effect Effects 0.000 description 3
- 230000008859 change Effects 0.000 description 2
- 238000010586 diagram Methods 0.000 description 2
- 230000006870 function Effects 0.000 description 2
- 238000003707 image sharpening Methods 0.000 description 2
- 238000012544 monitoring process Methods 0.000 description 2
- 230000009467 reduction Effects 0.000 description 2
- 230000001133 acceleration Effects 0.000 description 1
- 238000004458 analytical method Methods 0.000 description 1
- 230000008901 benefit Effects 0.000 description 1
- 230000000052 comparative effect Effects 0.000 description 1
- 238000011161 development Methods 0.000 description 1
- 230000018109 developmental process Effects 0.000 description 1
- 235000013399 edible fruits Nutrition 0.000 description 1
- 230000035800 maturation Effects 0.000 description 1
- 230000007246 mechanism Effects 0.000 description 1
- 230000000149 penetrating effect Effects 0.000 description 1
- 238000003672 processing method Methods 0.000 description 1
- 238000011160 research Methods 0.000 description 1
- 239000000725 suspension Substances 0.000 description 1
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T3/00—Geometric image transformations in the plane of the image
- G06T3/40—Scaling of whole images or parts thereof, e.g. expanding or contracting
-
- 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
- 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]
-
- 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
- G06T2207/20092—Interactive image processing based on input by user
Landscapes
- Physics & Mathematics (AREA)
- General Physics & Mathematics (AREA)
- Engineering & Computer Science (AREA)
- Theoretical Computer Science (AREA)
- Image Analysis (AREA)
Abstract
The invention discloses a kind of single image to the fog method based on feature learning, this method goes out transmission plot according to pre-estimations such as certain features such as dark feature, maximum color contrast of fog image, is then obtained atmosphere light intensity value based on atmospherical scattering model again and is recovered fog free images.In order to improve the adaptivity of algorithm, the present invention improves traditional physical model, predicts transmission plot using deep learning method, improve it is traditional based on the assumption that method, obtain more true transmission plot.The core of image defogging is to estimate transmission plot, and deep learning has powerful feature extraction and learning ability, can train the mapping model between fog image and transmission plot.After going out the transmission image of fog image using deep learning model prediction, fog free images can be recovered further according to atmospherical scattering model, while improving the adaptive adjustment capability of defogging algorithm, obtain higher defogging quality, and have lower complexity.
Description
Technical field
The present invention relates to a kind of method for image progress defogging more particularly to a kind of single width figures based on feature learning
As defogging method.
Background technique
In the case where meteorological condition is bad, the digital picture of outdoor shooting will receive in air suspended particles (such as mist,
Haze etc.) influence be degraded, this is because suspended particles can to light generate scattering, so that the light of scene reflections is decayed,
Scatter ambient light is mixed into the received light of observer simultaneously, changes the features such as contrast, color of image after being imaged
Become.Pass through the color-values distribution map and three Color Histograms to foggy image and clear image.As can be seen that the color of foggy image
Distribution value Relatively centralized is in centre, and the gray value DYNAMIC DISTRIBUTION range of tri- color of pixel RGB is relatively narrow, and histogram is also relatively concentrated,
The color-values of clear image are distributed more widely, and color is remarkably reinforced, and histogram distribution is also relatively uniform.For being dropped by greasy weather weather
The image of matter needs pair in order to which the visual effect of beautifying picture and prominent characteristics of image are so as to computer vision system processing
Image carries out defogging processing.Image defogging just refers to specific ways and means, make air-borne particulates to image not
Good influence reduction even is eliminated.
With increasingly extensive and computer technology the development and maturation of computer vision application, image defogging technology is gradually
Receive the concern of whole world researcher.
Current misty image sharpening processing method has very much, can be mainly divided into two major classes: the first kind is based on figure
The method of image intensifying, such methods are enhanced the image being degraded, and the quality of image is improved.This method advantage is
Common algorithm for image enhancement targetedly can be used and be changed using existing mature image processing algorithm
It is good, enhance the contrast of image, still, this method may be made by the feature of scenery and valuable information in prominent image
At the loss of image portion information, making the second class of image fault is the method based on physical model, and this method is big by research
Atmospherical scattering model is established in scattering process of the gas suspension particle to light, understands the Physical Mechanism that image is degenerated, and inversion restoration goes out
Image before not degrading.This is the method for a kind of image restoration specifically for Misty Image, and the image effect for restoring out is true
Real, the scenery original scape close to before degrading, preferable to the image processing effect of complex scene, image information is more completely saved.
So far, researcher mainly studies this problem of image defogging from both direction: based on image enhancement
Direction and direction based on physical model.
The defogging method in image enhancement direction it is not intended that there is mist air that image is made to degenerate the actual physics process to degrade, and
It is that the contrast of image is improved with image procossing method for the image itself after degrading, the feature of prominent image improves image
Visual effect and analysis and processing convenient for computer vision system to image, the universal comparative maturity of such methods is efficient, place
The result of reason is also able to satisfy system to the clarity demand of image after processing.But such methods do not adapt to different scenes
And image, the especially more image of scenery change in depth, and importantly, such methods based on the increasing to image
It by force, can not be fog " removal " to restore scape it is not intended that the process that fog degrades, can only limitedly improve image definition
The style of object makes image fault, and image ornamental value is lower after processing, after being also unfavorable for computer vision system to image
Continuous processing.
The ideal processing currently, the method based on physical model can make comparisons for existing single image,
And such methods can generally make effective processing to the biggish image of change in depth, and the image of recovery relatively connects with former scenery
Closely, feature is obvious, and visual effect is good.But such methods generally need to be tested according to different pictures, are tied according to test
To more progress, manually parameter adjusts fruit, can not automatically be handled, and need larger amount of operation mostly, when
Between and space complexity it is higher, can not achieve real-time processing.
Summary of the invention
It is analyzed based on above-mentioned technology, the object of the invention is first of all for the adaptive adjustment capability for improving algorithm, existing algorithm
It does not ensure that and is suitable for all scene or image, or need to manually adjust parameter;However, many computer vision systems
System, such as safety monitoring system and military surveillance system etc. require that algorithm is automatically handled different images, without
Need or need seldom manually adjust.Ideal defogging algorithm should be able to automatically analyze the data of single width foggy image,
Adaptive adjustment is made for different scenes and different weather conditions, meets the defogging and image sharpening of different scenes
Demand.
Secondly, improving the quality of defogging algorithm process.Still more or less there is distortion in current image defogging technology, especially
It is that this problem is especially prominent in the processing to thick fog image.Ideal defogging algorithm should be able to be according in degraded image
The information for being included restores the scenery before being degraded by fog as much as possible, can protrude scene features to meet computer
The demand of vision system processing, can also improve the ornamental value of image.
Again, the complexity of defogging algorithm is reduced.The computer vision systems such as safety monitoring system, military surveillance system are all
Often require that image processing algorithm has relatively high real-time, still, existing defogging algorithm, especially single image defogging
The preferable algorithm of quality, the excessively high problem of all generally existing Space-time Complexity, the defogging algorithm thought should be can be applied to greatly
Width image is handled in real time, this requires defogging algorithm while guaranteeing defogging quality, Time & Space Complexity has larger
The reduction of amplitude, or utilize large amount of complex data processing problem that is hardware-accelerated, often occurring in acceleration processing defogging algorithm.
To achieve the goals above, the technical solution adopted by the present invention is a kind of single image defogging based on feature learning
Method, steps are as follows for the realization of this method:
S1 builds the pre-training model of mist elimination image I, treats mist elimination image I by pre-training model and carries out feature extraction.
S2 pre-training model is realized using convolutional neural networks, wherein to improve convolutional neural networks on image procossing
Feature learning ability constructs the deep learning mould of transmission image prediction using the convolutional neural networks group of three groups of different scales
Type.
S2.1 is based on AlexNet model refinement and obtains first group of CNN, and changes its output layer structure, makes Alex Net
Model becomes network model end to end (i.e. input terminal and output end are all images).
S2.2 increases the CNNs of two groups of different scales.
S3 obtains dark channel image by the convolutional neural networks successive optimization of three groups of different scales in S2;It uses
Maxout nonlinear activation function simulates extreme value filter, and it is special that dark is extracted from the image of input atmospherical scattering model
Sign.
According to dark channel prior knowledge, using former mist elimination image I's and dark channel image, solve atmosphere light A.
S4 predicts transmission plot t: the coding stage of depth model using deep learning method, use SENet154 as
Foundation structure.The transmission image t of fog image is split by decoding stage by modification routine FPN network.Melted using FPN
Multiresolution features are closed, the segmentation precision of zonule fog image is improved;On the basis of FPN, hypercolumn mould is introduced
Block further merges the multiresolution features of atmosphere light A;Global average pond layer and classification head are eventually adding in encoder;
In addition, introducing classification auxiliary loss in segmentation network;In the level of each resolution ratio of decoder, segmentation auxiliary loss is introduced,
The training of each level parameter is further adjusted, final realize predicts transmission plot t.
S5 according to the atmospheric light value A in the S4 mapping model exported and S3, build atmospherical scattering model J (x)=
(I(x)-A)/t(x)+A。
S6 is to building atmospherical scattering model into after the dark feature extraction of image, then in parallel through three groups of different scales
Convolution kernel, then carry out pond and nonlinear activation, recovery obtains fog free images.
The main flow of traditional physical model image defogging be according to certain features such as dark feature of fog image,
The pre-estimations such as maximum color contrast go out transmission plot, are then obtained atmosphere light intensity value based on atmospherical scattering model again and are recovered
Fog free images.In order to improve the adaptivity of algorithm, the present invention improves traditional physical model, uses deep learning side
Method predicts transmission plot, improve it is traditional based on the assumption that method, obtain more true transmission plot.Image defogging
Core is to estimate transmission plot, and deep learning has powerful feature extraction and learning ability, can train fog image and
Mapping model between transmission plot.After going out the transmission image of fog image using deep learning model prediction, dissipated further according to atmosphere
Fog free images can be recovered by penetrating model.
In addition, predict transmission plot t using deep learning method, improve it is traditional based on the assumption that method, obtain
More true transmission plot is based on the defogging network model of the preferential principle of dark (Dark Channel Prior, DCP) feature
Foundation, be the innovation of this method, the adaptive adjustment capability of defogging algorithm, higher defogging quality can be improved, and have
Lower complexity.
Detailed description of the invention
Fig. 1 is implementation flow chart of the present invention.
Fig. 2 is exemplary diagram one before and after defogging of the present invention.
Fig. 3 is exemplary diagram two before and after defogging of the present invention.
Specific embodiment
This method is described in detail below in conjunction with drawings and examples.
As shown in Figure 1, a kind of single image to the fog method based on feature learning, steps are as follows for the realization of this method:
S1 builds the pre-training model of mist elimination image I, treats mist elimination image I by pre-training model and carries out feature extraction.
S2 pre-training model is realized using convolutional neural networks, wherein to improve convolutional neural networks on image procossing
Feature learning ability constructs the deep learning mould of transmission image prediction using the convolutional neural networks group of three groups of different scales
Type.
S2.1 is based on AlexNet model refinement and obtains first group of CNN, and changes its output layer structure, makes Alex Net
Model becomes network model end to end (i.e. input terminal and output end are all images).
S2.2 increases the CNNs of two groups of different scales.
S3 obtains dark channel image by the convolutional neural networks successive optimization of three groups of different scales in S2;It uses
Maxout nonlinear activation function simulates extreme value filter, and it is special that dark is extracted from the image of input atmospherical scattering model
Sign.
According to dark channel prior knowledge, using former mist elimination image I's and dark channel image, solve atmosphere light A.
S4 predicts transmission plot t: the coding stage of depth model using deep learning method, use SENet154 as
Foundation structure.The transmission image t of fog image is split by decoding stage by modification routine FPN network.Melted using FPN
Multiresolution features are closed, the segmentation precision of zonule fog image is improved;On the basis of FPN, hypercolumn mould is introduced
Block further merges the multiresolution features of atmosphere light A;Global average pond layer and classification head are eventually adding in encoder;
In addition, introducing classification auxiliary loss in segmentation network;In the level of each resolution ratio of decoder, segmentation auxiliary loss is introduced,
The training of each level parameter is further adjusted, final realize predicts transmission plot t.
S5 according to the atmospheric light value A in the S4 mapping model exported and S3, build atmospherical scattering model J (x)=
(I(x)-A)/t(x)+A。
S6 is to building atmospherical scattering model into after the dark feature extraction of image, then in parallel through three groups of different scales
Convolution kernel, then carry out pond and nonlinear activation, recovery obtains fog free images.
Embodiment: the input parameter of defogging module is picture file.
Picture file comes from video, and the video that VAM Video Access Module is obtained is converted to the picture of a frame frame, calls defogging
Module dynamic base carries out defogging processing.Pictures to be treated are loaded, data library button is connected to, it can be directly from data
Library obtains data.The picture that defogging has been handled is by saving picture, and by treated, picture is saved.
Left and right is able to carry out to picture to pull up and down, can amplify the operation such as diminution.After plurality of pictures is selected, lead to
Verifying is crossed, design requirement is met.
Picture is imported from database, such as clickthrough database, database is configured and carries out parameter selection.
After successful connection, it will import after a series of pictures chooses picture, menu image button can be clicked, click image defogging
Button.Effect after image defogging is as shown in Figure 2 and Figure 3.
Claims (4)
1. a kind of single image to the fog method based on feature learning, steps are as follows for the realization of this method:
S1 builds the pre-training model of mist elimination image I, treats mist elimination image I by pre-training model and carries out feature extraction;
S2 pre-training model is realized using convolutional neural networks, wherein to improve feature of the convolutional neural networks on image procossing
Learning ability constructs the deep learning model of transmission image prediction using the convolutional neural networks group of three groups of different scales;
S3 obtains dark channel image by the convolutional neural networks successive optimization of three groups of different scales in S2;It is non-using Maxout
Linear activation primitive simulates extreme value filter, and dark feature is extracted from the image of input atmospherical scattering model;
According to dark channel prior knowledge, using former mist elimination image I's and dark channel image, solve atmosphere light A;
S4 predicts transmission plot t: the coding stage of depth model using deep learning method, based on SENet154
Structure;The transmission image t of fog image is split by decoding stage by modification routine FPN network;It is merged using FPN more
Resolution characteristics improve the segmentation precision of zonule fog image;On the basis of FPN, hypercolumn module is introduced,
Further merge the multiresolution features of atmosphere light A;Global average pond layer and classification head are eventually adding in encoder;This
Outside, in segmentation network, classification auxiliary loss is introduced;In the level of each resolution ratio of decoder, segmentation auxiliary loss is introduced, into
The training of each level parameter of one successive step, final realize predict transmission plot t;
S5 builds atmospherical scattering model J (x)=(I according to the atmospheric light value A in the S4 mapping model exported and S3
(x)-A)/t(x)+A;
S6 is to building atmospherical scattering model into after the dark feature extraction of image, then in parallel through the volume of three groups of different scales
Product core, then carries out pond and nonlinear activation, and recovery obtains fog free images.
2. a kind of single image to the fog method based on feature learning according to claim 1, it is characterised in that: end-to-end
Network model in, i.e., input terminal and output end are all images.
3. a kind of single image to the fog method based on feature learning according to claim 1, it is characterised in that: S2.1 base
First group of CNN is obtained in Alex Net model refinement, and changes its output layer structure, Alex Net model is made to become end-to-end
Network model;
S2.2 increases the CNN of two groups of different scales.
4. a kind of single image to the fog method based on feature learning according to claim 1, it is characterised in that: picture text
Part comes from video, and the video that VAM Video Access Module is obtained is converted to the picture of a frame frame, calls defogging module dynamic base, into
The processing of row defogging;Pictures to be treated are loaded, data library button is connected to, directly can obtain data from database;It goes
The picture that mist has been handled is by saving picture, and by treated, picture is saved;
Left and right is able to carry out to picture to pull up and down, can amplify the operation such as diminution;After plurality of pictures is selected, by testing
Card, meets design requirement;
Picture is imported from database, such as clickthrough database, database is configured and carries out parameter selection;
After successful connection, it will import after a series of pictures chooses picture, menu image button can be clicked, image defogging is clicked and press
Button.
Priority Applications (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201910246074.6A CN110211052A (en) | 2019-03-29 | 2019-03-29 | A kind of single image to the fog method based on feature learning |
Applications Claiming Priority (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201910246074.6A CN110211052A (en) | 2019-03-29 | 2019-03-29 | A kind of single image to the fog method based on feature learning |
Publications (1)
Publication Number | Publication Date |
---|---|
CN110211052A true CN110211052A (en) | 2019-09-06 |
Family
ID=67785244
Family Applications (1)
Application Number | Title | Priority Date | Filing Date |
---|---|---|---|
CN201910246074.6A Pending CN110211052A (en) | 2019-03-29 | 2019-03-29 | A kind of single image to the fog method based on feature learning |
Country Status (1)
Country | Link |
---|---|
CN (1) | CN110211052A (en) |
Cited By (7)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN110738623A (en) * | 2019-10-18 | 2020-01-31 | 电子科技大学 | multistage contrast stretching defogging method based on transmission spectrum guidance |
CN110766640A (en) * | 2019-11-05 | 2020-02-07 | 中山大学 | Image defogging method based on depth semantic segmentation |
CN111369472A (en) * | 2020-03-12 | 2020-07-03 | 北京字节跳动网络技术有限公司 | Image defogging method and device, electronic equipment and medium |
CN111861939A (en) * | 2020-07-30 | 2020-10-30 | 四川大学 | Single image defogging method based on unsupervised learning |
CN112508814A (en) * | 2020-12-07 | 2021-03-16 | 重庆邮电大学 | Image tone restoration type defogging enhancement method based on unmanned aerial vehicle at low altitude view angle |
CN113436127A (en) * | 2021-03-25 | 2021-09-24 | 上海志御软件信息有限公司 | Method and device for constructing automatic liver segmentation model based on deep learning, computer equipment and storage medium |
CN117392009A (en) * | 2023-12-06 | 2024-01-12 | 国网山东省电力公司淄博供电公司 | Automatic fog penetrating processing method, system, terminal and storage medium for image |
Citations (4)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN107749052A (en) * | 2017-10-24 | 2018-03-02 | 中国科学院长春光学精密机械与物理研究所 | Image defogging method and system based on deep learning neutral net |
CN107958460A (en) * | 2016-10-18 | 2018-04-24 | 奥多比公司 | Instance-level semantic segmentation system |
CN107958465A (en) * | 2017-10-23 | 2018-04-24 | 华南农业大学 | A kind of single image to the fog method based on depth convolutional neural networks |
CN108460735A (en) * | 2018-02-06 | 2018-08-28 | 中国科学院光电技术研究所 | Improvement dark defogging method based on single image |
-
2019
- 2019-03-29 CN CN201910246074.6A patent/CN110211052A/en active Pending
Patent Citations (4)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN107958460A (en) * | 2016-10-18 | 2018-04-24 | 奥多比公司 | Instance-level semantic segmentation system |
CN107958465A (en) * | 2017-10-23 | 2018-04-24 | 华南农业大学 | A kind of single image to the fog method based on depth convolutional neural networks |
CN107749052A (en) * | 2017-10-24 | 2018-03-02 | 中国科学院长春光学精密机械与物理研究所 | Image defogging method and system based on deep learning neutral net |
CN108460735A (en) * | 2018-02-06 | 2018-08-28 | 中国科学院光电技术研究所 | Improvement dark defogging method based on single image |
Non-Patent Citations (5)
Title |
---|
BAOPING YUAN ET AL: ""Single Image Defogging Method based on Deep Learning"", 《INTERNATIONAL CONFERENCE ON MECHANICAL, ELECTRONIC, CONTROL AND AUTOMATION ENGINEERING》 * |
HU J ET AL: ""Squeeze-and-Excitation Network"", 《COMPUTER VISION AND PATTERN RECOGNITION》 * |
VIREDERY: ""计算机视觉-语义分割(二)"", 《博客园》 * |
YONG XU ET AL: ""Review of video and image defogging algorithms and related studies on image restoration and enhancement"", 《IEEE》 * |
袁保平: ""基于卷积神经网络的图像去雾方法研究"", 《中国优秀硕士学位论文全文数据库 信息科技辑》 * |
Cited By (10)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN110738623A (en) * | 2019-10-18 | 2020-01-31 | 电子科技大学 | multistage contrast stretching defogging method based on transmission spectrum guidance |
CN110766640A (en) * | 2019-11-05 | 2020-02-07 | 中山大学 | Image defogging method based on depth semantic segmentation |
CN111369472A (en) * | 2020-03-12 | 2020-07-03 | 北京字节跳动网络技术有限公司 | Image defogging method and device, electronic equipment and medium |
CN111369472B (en) * | 2020-03-12 | 2021-04-23 | 北京字节跳动网络技术有限公司 | Image defogging method and device, electronic equipment and medium |
CN111861939A (en) * | 2020-07-30 | 2020-10-30 | 四川大学 | Single image defogging method based on unsupervised learning |
CN112508814A (en) * | 2020-12-07 | 2021-03-16 | 重庆邮电大学 | Image tone restoration type defogging enhancement method based on unmanned aerial vehicle at low altitude view angle |
CN112508814B (en) * | 2020-12-07 | 2022-05-20 | 重庆邮电大学 | Image tone restoration type defogging enhancement method based on unmanned aerial vehicle at low altitude visual angle |
CN113436127A (en) * | 2021-03-25 | 2021-09-24 | 上海志御软件信息有限公司 | Method and device for constructing automatic liver segmentation model based on deep learning, computer equipment and storage medium |
CN117392009A (en) * | 2023-12-06 | 2024-01-12 | 国网山东省电力公司淄博供电公司 | Automatic fog penetrating processing method, system, terminal and storage medium for image |
CN117392009B (en) * | 2023-12-06 | 2024-03-19 | 国网山东省电力公司淄博供电公司 | Automatic fog penetrating processing method, system, terminal and storage medium for image |
Similar Documents
Publication | Publication Date | Title |
---|---|---|
CN110211052A (en) | A kind of single image to the fog method based on feature learning | |
CN108986050B (en) | Image and video enhancement method based on multi-branch convolutional neural network | |
CN109685072B (en) | Composite degraded image high-quality reconstruction method based on generation countermeasure network | |
CN108615226B (en) | Image defogging method based on generation type countermeasure network | |
CN112070690B (en) | Single image rain removing method based on convolution neural network double-branch attention generation | |
CN110544213A (en) | Image defogging method based on global and local feature fusion | |
CN109993804A (en) | A kind of road scene defogging method generating confrontation network based on condition | |
CN111915530A (en) | End-to-end-based haze concentration self-adaptive neural network image defogging method | |
CN112465727A (en) | Low-illumination image enhancement method without normal illumination reference based on HSV color space and Retinex theory | |
CN114037938B (en) | NFL-Net-based low-illumination target detection method | |
CN111079864A (en) | Short video classification method and system based on optimized video key frame extraction | |
CN114120093A (en) | Coal gangue target detection method based on improved YOLOv5 algorithm | |
CN110930327A (en) | Video denoising method based on cascade depth residual error network | |
CN112070688A (en) | Single image defogging method for generating countermeasure network based on context guidance | |
CN114627269A (en) | Virtual reality security protection monitoring platform based on degree of depth learning target detection | |
CN111709888A (en) | Aerial image defogging method based on improved generation countermeasure network | |
CN111369477A (en) | Method for pre-analysis and tool self-adaptation of video recovery task | |
CN115880177A (en) | Full-resolution low-illumination image enhancement method for aggregating context and enhancing details | |
CN112446376A (en) | Intelligent segmentation and compression method for industrial image | |
CN112102214B (en) | Image defogging method based on histogram and neural network | |
CN113255464A (en) | Airplane action recognition method and system | |
CN113628143A (en) | Weighted fusion image defogging method and device based on multi-scale convolution | |
CN110738624B (en) | Area-adaptive image defogging system and method | |
CN112132207A (en) | Target detection neural network construction method based on multi-branch feature mapping | |
CN116468625A (en) | Single image defogging method and system based on pyramid efficient channel attention mechanism |
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: 20190906 |