CN112785523A - Semi-supervised image rain removing method and device for sub-band network bridging - Google Patents

Semi-supervised image rain removing method and device for sub-band network bridging Download PDF

Info

Publication number
CN112785523A
CN112785523A CN202110088761.7A CN202110088761A CN112785523A CN 112785523 A CN112785523 A CN 112785523A CN 202110088761 A CN202110088761 A CN 202110088761A CN 112785523 A CN112785523 A CN 112785523A
Authority
CN
China
Prior art keywords
image
network
rain
sub
learning
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.)
Granted
Application number
CN202110088761.7A
Other languages
Chinese (zh)
Other versions
CN112785523B (en
Inventor
刘家瑛
杨文瀚
胡煜章
郭宗明
Current Assignee (The listed assignees may be inaccurate. Google has not performed a legal analysis and makes no representation or warranty as to the accuracy of the list.)
Peking University
Original Assignee
Peking University
Priority date (The priority date 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 date listed.)
Filing date
Publication date
Application filed by Peking University filed Critical Peking University
Priority to CN202110088761.7A priority Critical patent/CN112785523B/en
Publication of CN112785523A publication Critical patent/CN112785523A/en
Application granted granted Critical
Publication of CN112785523B publication Critical patent/CN112785523B/en
Active legal-status Critical Current
Anticipated expiration legal-status Critical

Links

Images

Classifications

    • G06T5/73—
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F18/00—Pattern recognition
    • G06F18/20—Analysing
    • G06F18/21—Design or setup of recognition systems or techniques; Extraction of features in feature space; Blind source separation
    • G06F18/214—Generating training patterns; Bootstrap methods, e.g. bagging or boosting
    • G06F18/2155—Generating training patterns; Bootstrap methods, e.g. bagging or boosting characterised by the incorporation of unlabelled data, e.g. multiple instance learning [MIL], semi-supervised techniques using expectation-maximisation [EM] or naïve labelling
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
    • G06N3/00—Computing arrangements based on biological models
    • G06N3/02—Neural networks
    • G06N3/04—Architecture, e.g. interconnection topology
    • G06N3/045—Combinations of networks
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • 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
    • Y—GENERAL TAGGING OF NEW TECHNOLOGICAL DEVELOPMENTS; GENERAL TAGGING OF CROSS-SECTIONAL TECHNOLOGIES SPANNING OVER SEVERAL SECTIONS OF THE IPC; TECHNICAL SUBJECTS COVERED BY FORMER USPC CROSS-REFERENCE ART COLLECTIONS [XRACs] AND DIGESTS
    • Y02—TECHNOLOGIES OR APPLICATIONS FOR MITIGATION OR ADAPTATION AGAINST CLIMATE CHANGE
    • Y02A—TECHNOLOGIES FOR ADAPTATION TO CLIMATE CHANGE
    • Y02A90/00—Technologies having an indirect contribution to adaptation to climate change
    • Y02A90/10—Information and communication technologies [ICT] supporting adaptation to climate change, e.g. for weather forecasting or climate simulation

Abstract

The invention discloses a semi-supervised image rain removing method and device for subband network bridging, which are used for carrying out semi-supervised learning on images in rainy days through deep learning and provide a recursive frequency band representation for connecting unsupervised and fully-supervised frames. A series of frequency bands from coarse to fine are extracted, and enhancement is carried out through recursive end-to-end learning, and rain drop removal and detail correction are carried out. Under the competitive learning guided by the perception quality, the depth frequency band representation is used for reconstruction, and a final restoration result is generated. The invention extracts a series of frequency band representations from coarse to fine, enhances the frequency band representations through end-to-end learning of recursion, removes raindrop and corrects details, and provides a recursion frequency band representation for connecting unsupervised and fully supervised frameworks.

Description

Semi-supervised image rain removing method and device for sub-band network bridging
Technical Field
The invention belongs to the field of image processing and enhancement, and particularly relates to a method and a device for removing rain from a semi-supervised image bridged by a sub-band network.
Background
The deep learning rain-removing era started in 2017. Yang et al construct a network that combines rain mark detection and removal to handle heavy rain, overlapping rain marks and rain fog. The network can detect the position of rain by predicting a binary mask, and remove rain marks by adopting a recursive framework to gradually remove rain fog. The method achieves good effect under the condition of heavy rain. However, the method may erroneously remove the vertical texture and cause underexposure.
In the same year, Fu et al tried to remove rain marks by constructing a deep detail network. The network takes only high frequency details as input and predicts rain marks and clean, rain-free images. This work shows that removing background information from the network input facilitates network training.
Following the work of Yang and Fu et al, in subsequent work, a number of convolutional neural network-based approaches have been proposed. These methods employ more advanced network structures and embed new priors associated with rain, yielding better results in both quantitative and qualitative analysis. However, since these methods are limited by the supervised learning paradigm (i.e. use of a composite rain map), they may fail in dealing with real rain scenarios never seen during training.
Disclosure of Invention
In view of the above problems and the disadvantages of the related methods, the present invention provides a method and an apparatus for removing rain from a semi-supervised image bridged by a subband network. The overall framework is shown in figure 1, and the method constructs an effective feature representation, namely a learning-based sub-band representation, and connects supervised learning and unsupervised learning to realize efficient deep learning and semi-supervised rain removal. The supervised learning part of the model fully utilizes paired data and loss measurement based on signal fidelity to learn the raindrop removal and detail correction processes. The semi-supervised learning part utilizes unpaired data and counterstudy to learn the image quality enhancement process, thereby improving the visibility and comfort of the image.
The technical scheme adopted by the invention comprises the following steps:
a semi-supervised image rain removal method for subband network bridging, comprising the steps of:
1) generating a plurality of rainy-day images y based on the plurality of samples without rain marks and rain fog, constructing a paired image data set, collecting sample images with different qualities, obtaining an image quality label of each sample image, and constructing a non-paired image quality data set;
2) constructing an image rain removal model, and training the image rain removal model by utilizing a paired image data set and a non-paired image quality data set to obtain a trained image rain removal model;
the image rain removal model comprises an iterative sub-band learning network and an iterative sub-band reconstruction network, wherein the iterative sub-band learning network is used for learning the rain image y or reconstructing the sub-band signals of the image in rainy days; training an iterative subband learning network by using the paired image data sets, and training an iterative subband reconstruction network by using the paired image data sets and the trained quality evaluation network;
an iterative sub-band learning network is constructed by the following strategies:
A) constructing a plurality of deep networks similar to U-Net as sub-networks;
B) each subnetwork recovers the result of the previous cycle by using the rain image y
Figure BDA0002911948240000021
Cascading as input, and taking the rain day image y and the restoration result
Figure BDA0002911948240000022
Mapping the cascade to a feature space, and then performing feature transformation through a plurality of convolution layers;
C) in the middle layer, firstly, the spatial resolution of the features is downsampled through convolution and deconvolution with step length, and then upsampled;
D) connecting the shallow and deep features of the same spatial resolution of each sub-network using a hopping connection;
an iterative subband reconstruction network is constructed by the following strategy:
a) constructing a plurality of deep networks similar to U-Net as sub-networks;
b) connecting the shallow and deep features of the same spatial resolution of each sub-network using a hopping connection;
the trained quality evaluation network is obtained by utilizing an unpaired rainy day image quality data set for training; the structure of the quality evaluation network includes: replacing the last layer with a network of VGG16 with n cell full connectivity layers and a softmax layer, where n is the number of categories of image quality labels;
3) and inputting the image to be processed into the trained image rain removal model to obtain the image with rain removed.
Further, the method for generating rain marks and rain fog comprises the following steps: a rain mark appearance model was used.
Further, the parameters of the rain fog include: light transmittance and background light.
Further, the rainy day image y is x (1-t) + t α + s, where x is a sample no-rain image, s is rain mark, t is light transmittance, and α is background light.
Further, the iterative subband learning network learns the subband signals in the rainy-day image y or the restored image by the following steps:
1) mapping the rainy day image as a feature, or accumulating the cross-cycle feature residual generated by utilizing the restored image to obtain a feature;
2) generating a cross-scale feature residual error by using the long-time and short-time memory network and the features, and accumulating the cross-scale feature residual error to obtain a cross-scale feature residual error accumulation result;
3) and mapping the cross-scale feature residual accumulated result into enhancement results under different scales to obtain the rainy-day image y or the sub-band signal in the restored image.
Further, when the iterative subband learning network is trained by using the paired image data sets, the learning of the iterative subband learning network is constrained by using a multi-scale loss function, wherein the multi-scale loss function
Figure BDA0002911948240000023
Figure BDA0002911948240000031
Phi (-) is the structural similarity index of the calculated image, siIs a given scaling factor, FDIs downsampledDistance, λ1Is a first weight parameter, λ2Is a second weight parameter.
Further, the iterative subband reconstruction network reconstructs the subband signals to generate a restored image by:
1) mapping the sub-band signals into signal recombination weights;
2) using the signal recombination weight to weight the sub-band signal recombination to generate a new enhancement result;
3) and recombining the new enhanced result to generate a restored image.
Further, when the iterative subband reconstruction network is trained by using the paired image data set and the trained quality evaluation network, a loss function L is usedSBRLearning of constrained sub-band reconstruction networks, where LSBR＝LPercept+λ3LDetail+λ4LQualityFunction of perceptual loss
Figure BDA0002911948240000032
Signal fidelity metric
Figure BDA0002911948240000033
Function of mass loss
Figure BDA0002911948240000034
λ3Is a third weight parameter, λ4Is a fourth weight parameter, Fp(. is a depth feature extracted from a pre-trained VGG network, phi () is a structural similarity index for computed images, lrFor random numbers, D (-) is the trained quality assessment network.
A storage medium having a computer program stored therein, wherein the computer program is arranged to perform the above-mentioned method when executed.
An electronic device comprising a memory having a computer program stored therein and a processor arranged to run the computer to perform the method as described above.
Compared with the prior art, the invention has the following advantages:
1) a recursive frequency band representation is provided to connect an unsupervised framework and a fully supervised framework, and the method has the advantages of both the supervised learning method and the unsupervised learning image enhancement method, namely: better detail resilience and overall visibility and visual comfort;
2) a series of frequency bands from coarse to fine are extracted, and enhancement is carried out through recursive end-to-end learning, and rain drop removal and detail correction are carried out.
Drawings
FIG. 1 is a diagram of the deep recursive subband network framework of the present invention.
Fig. 2 is a block diagram of a sub-band learning network of the present invention.
FIG. 3 is a block diagram of a subband reorganization network of the present invention.
Detailed Description
In order to further explain the technical method of the invention, the invention is further described in detail by combining the drawings and the specific examples in the specification.
The semi-supervised image rain removing method of the invention uses a deep recursion sub-band network as shown in figure 1, and comprises the following steps:
step 1: a total of 1800 rain/no rain image pairs were constructed for the rain/no rain training dataset. According to a rainless image x, generating corresponding rain marks s and rain fog parameters (light transmittance t and background light alpha) based on a rain mark appearance model (random sampling generation illumination direction parameters, viewing angle parameters and rain drop vibration parameters) [ Garg and Nayar,2016], superposing related variables to generate a rainy day image y:
y＝x(1-t)+tα+s. (1)
step 2: an unpaired image quality data set is constructed, and 1000 images with different qualities and corresponding image quality labels are collected through an open channel (1-10 grades, 10 represents the highest quality, and 1 represents the lowest quality).
And step 3: an iterative sub-band learning network is constructed as shown in fig. 2. The primary objective is to fully utilize the paired training data generated in step 1 to learn each time of the restored image (output of iterative subband learning network, target fitting rain-free image)A subband signal. As shown in FIG. 1, a series of deep networks like U-Net are constructed. Recovery of each subnetwork with y and last cycle
Figure BDA0002911948240000041
Cascading as input, mapping it to a feature space, and then performing a feature transformation over several convolutional layers, where si(i ═ 1,2,3) is the scaling factor, and t-1 is the number of previous cycles. In the middle layer, the spatial resolution of the features is first downsampled by convolution and deconvolution with a step size, and then upsampled. The use of skip-connects to connect shallow and deep features of the same spatial resolution helps to bring the local information contained in the shallow features to the output. Each subnet is respectively at s1＝1/4、ss＝1/2 and s3Three features are produced on the scale of 1.
1) The first cycle of recursive learning is first described below:
Figure BDA0002911948240000042
wherein ,
Figure BDA0002911948240000043
is a feature extracted at a corresponding scale,
Figure BDA0002911948240000044
is the characteristic of the long and short memory network after being enhanced.
Figure BDA0002911948240000045
And
Figure BDA0002911948240000046
respectively corresponding sub-band learning network and long-short term memory network processes,
Figure BDA0002911948240000047
in order to record the memory information of all the loop states,
Figure BDA0002911948240000048
is the cross-scale feature residual.
Figure BDA0002911948240000049
And
Figure BDA00029119482400000410
is a mapping process that projects the enhanced features back into the image domain. FU(. cndot.) is an upsampling process. The image is first at a coarser granularity scale s1And (4) performing upper reconstruction. The residuals of the image signals are then predicted on a finer-grained scale and then combined into the entire image. In the step (2), the first line formula maps the rainy-day image as a feature, the second line formula generates a cross-scale feature residual by using a long-time and short-time memory network, the third line formula accumulates the cross-scale feature residual, and finally the third line formula maps the feature as an enhancement result under different scales.
2) Thereafter, at the t-th cycle, residual features and images are learned under the direction of the previous estimation results. Recovery of each subnetwork with y and last cycle
Figure BDA0002911948240000051
The cascade serves as input:
Figure BDA0002911948240000052
wherein
Figure BDA0002911948240000053
Is the cross-loop feature residual. In the step (3), a first line formula generates a cross-cycle characteristic residual error, a second line formula accumulates the cross-cycle characteristic residual error, a third line formula generates a cross-scale characteristic residual error by using a long-time and short-time memory network, a fourth line formula accumulates the cross-scale characteristic residual error, and finally the three line formula maps the characteristics into enhancement results under different scales. This formula ties all sub-band features together tightly, forming a joint optimization of all sub-bands.
And 4, step 4: an iterative sub-band reconstruction network is constructed, as in fig. 3, which again employs a U-Net like network structure, except that in the sub-band reconstruction network, shallow and deep features of the same spatial resolution are connected using hopping connections. With the paired data, the band restoration process from the rainy image to the normal light image can be well learned, while the details can be well restored and the rain marks can be suppressed. Since signal fidelity is not always well consistent with human visual perception, especially for certain global properties of the image (e.g., visibility, contrast, color illumination distribution, etc.). Therefore, the model is further used as a constraint through a perception quality evaluation method based on a neural network, and the recovery model learns better recovery enhancement mapping. The sub-band signals are recombined using another U-Net like network, using FRC(-) represents the process, generating the following coefficients to recombine the subband signals:
Figure BDA0002911948240000061
wherein T is the total cycle number,
Figure BDA0002911948240000062
are subband signals. In the first line formula in (4), FRC(. the) Signal recombination Module maps the sub-band signals into Signal recombination weights { omega1,ω2,ω3}. In the second line of equations in (4), new enhancement results are generated using signal rebinning weight weighted subband signal rebinning
Figure BDA0002911948240000063
And 5: the quality assessment network D is trained using unpaired image quality datasets. D uses the network structure of VGG16 and makes the last layer into an FC layer with 10 units. Then softmax is followed. The network was pre-trained on ImageNet and then refined using AVA Dataset. The AVA Dataset contains 255,000 pictures, each scored by approximately 200 skilled photographers. Each picture is associated with a certain game theme (a total of approximately 900 themes). The score range [1,10], 10 is the highest score.
Step 6: and training an iterative sub-band learning network by using a pair of image data sets, and constraining the learning of the network by using a multi-scale loss function. The loss function can be expressed as:
Figure BDA0002911948240000064
wherein ,FD(. is a downsampling process, s)iIs a given scaling factor. Phi (-) calculates the structural similarity index of the image. Lambda [ alpha ]1 and λ2Is a weight parameter.
And 7: training a sub-band reconstruction network using paired image data sets and quality assessment network constraints, using a perceptual loss function LPerceptSignal fidelity metric LDetailAnd a mass loss function LQualityAnd (5) restricting the learning of the network. The loss function can be expressed as:
Figure BDA0002911948240000065
wherein ,λ3 and λ4Is a weight parameter. lrIs a random number between 7 and 12, where 10 represents the highest quality in the database. FP() is a deep feature extracted from a pre-trained VGG network. D (-) is a trained NIMA quality assessment network (Talebi and Milanfar, 2018).
Figure 1 summarizes the overall process of the invention. It will be apparent to those skilled in the art that various changes and modifications may be made in the present invention without departing from the spirit and scope of the invention. Thus, if such modifications and variations of the present invention fall within the scope of the claims of the present invention and their equivalents, the present invention is also intended to include such modifications and variations.

Claims (10)

1. A semi-supervised image rain removal method for subband network bridging, comprising the steps of:
1) generating a plurality of rainy-day images y based on the plurality of samples without rain marks and rain fog, constructing a paired image data set, collecting sample images with different qualities, obtaining an image quality label of each sample image, and constructing a non-paired image quality data set;
2) constructing an image rain removal model, and training the image rain removal model by utilizing a paired image data set and a non-paired image quality data set to obtain a trained image rain removal model;
the image rain removal model comprises an iterative sub-band learning network and an iterative sub-band reconstruction network, wherein the iterative sub-band learning network is used for learning the rain image y or reconstructing the sub-band signals of the image in rainy days; training an iterative subband learning network by using the paired image data sets, and training an iterative subband reconstruction network by using the paired image data sets and the trained quality evaluation network;
an iterative sub-band learning network is constructed by the following strategies:
A) constructing a plurality of deep networks similar to U-Net as sub-networks;
B) each subnetwork recovers the result of the previous cycle by using the rain image y
Figure FDA0002911948230000011
Cascading as input, and taking the rain day image y and the restoration result
Figure FDA0002911948230000012
Mapping the cascade to a feature space, and then performing feature transformation through a plurality of convolution layers;
C) in the middle layer, firstly, the spatial resolution of the features is downsampled through convolution and deconvolution with step length, and then upsampled;
D) connecting the shallow and deep features of the same spatial resolution of each sub-network using a hopping connection;
an iterative subband reconstruction network is constructed by the following strategy:
a) constructing a plurality of deep networks similar to U-Net as sub-networks;
b) connecting the shallow and deep features of the same spatial resolution of each sub-network using a hopping connection;
the trained quality evaluation network is obtained by utilizing an unpaired rainy day image quality data set for training; the structure of the quality evaluation network includes: replacing the last layer with a network of VGG16 with n cell full connectivity layers and a softmax layer, where n is the number of categories of image quality labels;
3) and inputting the image to be processed into the trained image rain removal model to obtain the image with rain removed.
2. The method of claim 1, wherein the method of generating rain marks and fog comprises: a rain mark appearance model was used.
3. The method of claim 2, wherein the parameters of the rain fog comprise: light transmittance and background light.
4. The method of claim 1, wherein the rain image y ═ x (1-t) + t α + s, where x is a sample no rain image, s is rain mark, t is light transmittance, and α is background light.
5. The method of claim 1, wherein the iterative subband learning network learns the rain image y or restores the subband signals in the image by:
1) mapping the rainy day image as a feature, or accumulating the cross-cycle feature residual generated by utilizing the restored image to obtain a feature;
2) generating a cross-scale feature residual error by using the long-time and short-time memory network and the features, and accumulating the cross-scale feature residual error to obtain a cross-scale feature residual error accumulation result;
3) and mapping the cross-scale feature residual accumulated result into enhancement results under different scales to obtain the rainy-day image y or the sub-band signal in the restored image.
6. The method of claim 1, wherein training the iterative subband learning network with the paired-image dataset constrains learning of the iterative subband learning network using a multi-scale loss function, wherein the multi-scale loss function
Figure FDA0002911948230000021
Figure FDA0002911948230000022
Phi (-) is the structural similarity index of the calculated image, siIs a given scaling factor, FD(. is a downsampling process, λ1Is a first weight parameter, λ2Is a second weight parameter.
7. The method of claim 1, wherein the iterative subband reconstruction network reconstructs the subband signals to generate the restored image by:
1) mapping the sub-band signals into signal recombination weights;
2) using the signal recombination weight to weight the sub-band signal recombination to generate a new enhancement result;
3) and recombining the new enhanced result to generate a restored image.
8. The method of claim 1, wherein a loss function L is used in training an iterative subband reconstruction network using a paired image dataset and a trained quality assessment networkSBRLearning of constrained sub-band reconstruction networks, where LSBR＝LPercept+λ3LDetail+λ4LQualityFunction of perceptual loss
Figure FDA0002911948230000023
Signal fidelity metric
Figure FDA0002911948230000024
Function of mass loss
Figure FDA0002911948230000025
λ3Is a third weight parameter, λ4Is a fourth weight parameter, Fp(. is a depth feature extracted from a pre-trained VGG network, phi () is a structural similarity index for computed images, lrFor random numbers, D (-) is the trained quality assessment network.
9. A storage medium having a computer program stored thereon, wherein the computer program is arranged to, when run, perform the method of any of claims 1-8.
10. An electronic device comprising a memory having a computer program stored therein and a processor arranged to run the computer program to perform the method according to any of claims 1-8.
CN202110088761.7A 2021-01-22 2021-01-22 Semi-supervised image rain removing method and device for sub-band network bridging Active CN112785523B (en)

Priority Applications (1)

Application Number Priority Date Filing Date Title
CN202110088761.7A CN112785523B (en) 2021-01-22 2021-01-22 Semi-supervised image rain removing method and device for sub-band network bridging

Applications Claiming Priority (1)

Application Number Priority Date Filing Date Title
CN202110088761.7A CN112785523B (en) 2021-01-22 2021-01-22 Semi-supervised image rain removing method and device for sub-band network bridging

Publications (2)

Publication Number Publication Date
CN112785523A true CN112785523A (en) 2021-05-11
CN112785523B CN112785523B (en) 2023-10-17

Family

ID=75758601

Family Applications (1)

Application Number Title Priority Date Filing Date
CN202110088761.7A Active CN112785523B (en) 2021-01-22 2021-01-22 Semi-supervised image rain removing method and device for sub-band network bridging

Country Status (1)

Country Link
CN (1) CN112785523B (en)

Citations (2)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN109086807A (en) * 2018-07-16 2018-12-25 哈尔滨工程大学 A kind of semi-supervised light stream learning method stacking network based on empty convolution
CN111062892A (en) * 2019-12-26 2020-04-24 华南理工大学 Single image rain removing method based on composite residual error network and deep supervision

Patent Citations (2)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN109086807A (en) * 2018-07-16 2018-12-25 哈尔滨工程大学 A kind of semi-supervised light stream learning method stacking network based on empty convolution
CN111062892A (en) * 2019-12-26 2020-04-24 华南理工大学 Single image rain removing method based on composite residual error network and deep supervision

Also Published As

Publication number Publication date
CN112785523B (en) 2023-10-17

Similar Documents

Publication Publication Date Title
CN108062754B (en) Segmentation and identification method and device based on dense network image
CN111259905B (en) Feature fusion remote sensing image semantic segmentation method based on downsampling
Li et al. A comprehensive benchmark analysis of single image deraining: Current challenges and future perspectives
CN110969577A (en) Video super-resolution reconstruction method based on deep double attention network
CN113450288B (en) Single image rain removing method and system based on deep convolutional neural network and storage medium
CN109872278B (en) Image cloud layer removing method based on U-shaped network and generation countermeasure network
CN111275638B (en) Face repairing method for generating confrontation network based on multichannel attention selection
CN107749048B (en) Image correction system and method, and color blindness image correction system and method
CN112488935B (en) Method for generating anti-finger vein image restoration based on texture constraint and poisson fusion
Chen et al. U-net like deep autoencoders for deblurring atmospheric turbulence
CN110796623A (en) Infrared image rain removing method and device based on progressive residual error network
CN115357944A (en) Image tampering detection method based on shallow feature enhancement and attention mechanism
Ran et al. RGAN: Rethinking generative adversarial networks for cloud removal
CN112767277B (en) Depth feature sequencing deblurring method based on reference image
CN112785523B (en) Semi-supervised image rain removing method and device for sub-band network bridging
Mandal et al. Neural architecture search for image dehazing
CN115861749A (en) Remote sensing image fusion method based on window cross attention
Song et al. Deep semantic-aware remote sensing image deblurring
CN112614073A (en) Image rain removing method based on visual quality evaluation feedback and electronic device
CN114331894A (en) Face image restoration method based on potential feature reconstruction and mask perception
CN114155165A (en) Image defogging method based on semi-supervision
Poomani et al. RETRACTED ARTICLE: Wiener filter based deep convolutional network approach for classification of satellite images
Sun et al. Kinect depth recovery via the cooperative profit random forest algorithm
Wang et al. A CBAM‐GAN‐based method for super‐resolution reconstruction of remote sensing image
Fakhari et al. An image restoration architecture using abstract features and generative models

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
GR01 Patent grant
GR01 Patent grant