CN111127354A - Single-image rain removing method based on multi-scale dictionary learning - Google Patents

Single-image rain removing method based on multi-scale dictionary learning Download PDF

Info

Publication number
CN111127354A
CN111127354A CN201911300714.3A CN201911300714A CN111127354A CN 111127354 A CN111127354 A CN 111127354A CN 201911300714 A CN201911300714 A CN 201911300714A CN 111127354 A CN111127354 A CN 111127354A
Authority
CN
China
Prior art keywords
rain
image
network model
rough
images
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
CN201911300714.3A
Other languages
Chinese (zh)
Other versions
CN111127354B (en
Inventor
余磊
何敬伟
袁琼雯
罗美露
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.)
Wuhan University WHU
Original Assignee
Wuhan University WHU
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 Wuhan University WHU filed Critical Wuhan University WHU
Priority to CN201911300714.3A priority Critical patent/CN111127354B/en
Publication of CN111127354A publication Critical patent/CN111127354A/en
Application granted granted Critical
Publication of CN111127354B publication Critical patent/CN111127354B/en
Active legal-status Critical Current
Anticipated expiration legal-status Critical

Links

Images

Classifications

    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
    • G06T5/00—Image enhancement or restoration
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F18/00—Pattern recognition
    • G06F18/20—Analysing
    • G06F18/28—Determining representative reference patterns, e.g. by averaging or distorting; Generating dictionaries
    • G06T5/73—
    • 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]

Abstract

The invention provides a multi-scale dictionary single-image rain removing method, which comprises the following steps: step 1, taking each pair of clean images and synthesized rain-carrying images as a training sample pair, and establishing a training set; step 2, constructing a network model, wherein the network model comprises a rough rain line extraction module and a fine rain line purification module, the rough rain line extraction module comprises two convolution layers and two ReLU activation function layers, and the rough rain line extraction module is used for realizing rough rain line extraction of combined rain images; the fine rain line purification module comprises seven convolution layers and four ReLU activation functions and is used for recovering a fine rain line graph from a noisy rain line image; and 3, training the network model by using the training set constructed in the step 1, and 4, inputting the image with rain to be tested into the trained network model to obtain a corresponding rain removing image. The invention comprehensively utilizes the sparse theory and the CNN learning ability, and greatly improves the solving efficiency and the reconstruction precision of the SC problem.

Description

Single-image rain removing method based on multi-scale dictionary learning
Technical Field
The invention belongs to the field of image processing, and particularly relates to a method for removing rain from a single-frame image by using a dictionary learning method.
Background
In reality, most computer vision algorithms assume that input is clear, however, for most outdoor vision systems such as video monitoring and automatic driving, the rainy environment can seriously affect imaging quality, which causes problems of image blurring, deformation, poor visibility and the like, and the performance of the system can be greatly reduced, so that the effective elimination of the influence of rainwater on the image has important application value, and especially, the raining removal of a single image is important in a task of removing rain. At present, rain removing algorithms for single images are mainly divided into two types: a priori based and learning based methods.
The prior-based method usually needs to observe the characteristics of the rain line in advance, and design specific prior information, for example, the rain line is more inclined straight line in a certain range or has low-rank characteristics, but the rain removing performance of the algorithm depends on the selection of the prior information to a great extent, and the complex rainfall condition in reality is difficult to process.
Learning-based methods are the research focus in recent years, and mainly include Convolutional Neural Network (CNN) -based methods, which treat an image rain-removing task as a pixel-by-pixel regression task, take the whole image as input, take a rain-removing image as output, and perform end-to-end training and testing. The method does not need manual design prior, and utilizes the strong learning ability of the CNN to enable the network to learn the rain line characteristics by self, but the method is lack of a guidance scheme and interpretability during network design and is not beneficial to network improvement and promotion.
Disclosure of Invention
Based on the analysis, the invention aims to provide a single-image rain removing method for multi-scale dictionary learning, which introduces sparse prior into CNN network design and greatly improves rain removing effect
The invention provides a multi-scale dictionary single-image rain removing method, which comprises the following specific steps:
step 1, according to the existing clean images, obtaining corresponding synthetic rain images by adding rain lines, taking each pair of clean images and synthetic rain images as a training sample pair, and establishing a training set;
step 2, constructing a network model, wherein the network model comprises a rough rain line extraction module and a fine rain line purification module, the rough rain line extraction module comprises two convolution layers and two ReLU activation function layers, and the rough rain line extraction module is used for realizing rough rain line extraction of combined rain images; the fine rain line purification module comprises seven convolution layers and four ReLU activation functions and is used for recovering a fine rain line graph from a noisy rain line graph;
step 3, training the network model by using the training set constructed in the step 1,
and 4, inputting the rain-carrying image to be tested into the trained network model to obtain a corresponding rain-removing image.
Further, the rough rain line extraction module in step 2 is implemented as follows,
Figure BDA0002321719900000021
wherein E is1，E2The number of the rolling layers is two,
Figure BDA0002321719900000022
for convolution operations, r∈For the extracted noisy rain line image, y is the composite rain image.
Further, the detailed implementation manner of the fine rain line purification module in the step 2 is as follows,
the fine rain line purification module comprises a sparse coding solving part and an HR characteristic reconstruction part, wherein the sparse coding solves the following minimization problem in a convolution mode:
Figure BDA0002321719900000023
wherein f isj，iI-th filter kernel, f, representing j-th dictionaryj，iThree filter groups are provided, namely j is 3, and the practical meaning is three rain line dictionaries with different scales, which are marked as S1、S2、S3Its transpose dictionary is denoted G1、G2、G3In which S is1、S2、 S3、G1、G2、G3All implemented by convolutional layers, c is the number of channels decomposed, zj，iFor convolutional sparse coding to be solved, | |1， ||.||2Respectively represent l1Norm and l2Norm, lambda is sparse penalty coefficient;
after sparse coding is obtained, reconstructing a denoised rainchart:
Figure BDA0002321719900000024
wherein the content of the first and second substances,
Figure BDA0002321719900000025
E3is a convolution layer, r is the final recovered fine rain image;
the specific process is as follows: firstly, the first step is to
Figure BDA0002321719900000026
Are all initialized to r∈，r∈Extracting a noisy rainline image for the feature extraction module; computing
Figure BDA0002321719900000027
Respectively pass through S1、S2、S3From the sum of (a) and (b), adding this sum from r∈Subtracting, and passing the difference value through G1、 G2、G3Then respectively and
Figure BDA0002321719900000028
add to obtain
Figure BDA0002321719900000029
Repeating the above process to obtain
Figure BDA00023217199000000210
The final convolution sparse code is obtained, wherein t is iteration times; at the time of reconstruction, calculating
Figure BDA00023217199000000211
Respectively pass through S1、S2、S3After ReLU and then E3I.e. to recover a fine rain image.
Furthermore, the formula (2) is converted into the traditional sparse coding problem by utilizing the multiplication relation of convolution and matrix, and the solution is carried out by adopting the ISTA algorithm under the assumption of non-negative sparse coding,
Figure BDA0002321719900000031
wherein the content of the first and second substances,
Figure BDA0002321719900000032
in order to be the middle symbol,
Figure BDA0002321719900000033
refers to a threshold value, and t is the number of iterations.
Further, in step 3, global residual learning is introduced into the network model, and an MSE loss function is selected, and with the minimum loss function as a training target, the expression of the MSE loss function is as follows:
Figure BDA0002321719900000034
wherein, theta refers to the network model parameter, l is the index of the training sample in the training set, and yl-rlRain-removed image output for network model, and real clean image xlAnd performing difference and accumulation to obtain a final error, so that the final error is minimized to realize optimization of the network model.
Further, in the step 1, the number of images in a training set is increased by means of turning, rotating, scaling and cutting, then rain lines are added to each clean image through Photoshop to obtain a composite rain image, and the corresponding clean image and the composite rain image are used as a training sample pair.
The invention provides a single image rain removing algorithm, which comprehensively utilizes a sparse theory and CNN learning capacity, obtains an iterative formula by solving an optimization problem in SC, and is realized by using CNN, thereby greatly improving the solving efficiency and the reconstruction precision of the SC problem.
Drawings
Fig. 1 is a general flowchart of network model construction according to an embodiment of the present invention.
Fig. 2 is a schematic diagram of sparse coding solution.
Fig. 3 is a Rain removal comparison graph of various algorithms for cat images in the Rain12 data set.
FIG. 4 is a diagram of Rain removal contrast for each algorithm for forest images in the Rain1200 data set.
Detailed Description
In order that the present invention may be more clearly understood, the following detailed description is provided.
As shown in fig. 1, the single-image rain removing method based on multi-scale dictionary learning provided by the present invention specifically includes four steps:
step 1, according to the existing clean images, obtaining corresponding synthetic rain images by adding rain lines, taking each pair of clean images and synthetic rain images as a training sample pair, and establishing a training set;
due to the limited training data set, data enhancement methods are required to effectively utilize the limited HR images. Data enhancement is an efficient way to expand the size of data samples. Deep learning is a data-driven method, and the larger the training data set is, the stronger the generalization ability of the trained model is. However, in practice, it is difficult to cover all scenes when data is collected, and the collection of data requires a large cost, which results in a limited training set in practice. If various training data can be generated according to the existing data, better open source throttling can be achieved, and the purpose of data enhancement is achieved.
Common data enhancement techniques are:
(1) turning: the flipping includes a horizontal flipping and a vertical flipping.
(2) Rotating: rotation is clockwise or counter-clockwise, and it is noted that rotation is preferably 90-180 ° during rotation, otherwise dimensional problems may occur.
(3) Zooming: the image may be enlarged or reduced. When enlarged, the size of the enlarged image will be larger than the original size. Most image processing architectures crop the enlarged image to its original size.
(4) Cutting: the region of interest of the picture is cut, and different regions are cut out randomly and are expanded to the original size again usually during training.
(5) Translation: translation is the movement of the image in either the x or y direction (or both). We need to make assumptions about the background during panning, such as black, etc., because some images are empty during panning, and because the objects in the images may appear at arbitrary positions, the panning enhancement method is very useful.
(6) Noise addition: overfitting usually occurs when the neural network learns high frequency features (because low frequency features are easily learned by the neural network and high frequency features are learned only at the last time) which may not help the task the neural network does and may have an effect on low frequency features which we randomly add noisy data to eliminate.
First, in order to train the CNN model, the present embodiment needs to construct training sample pairs. And adding rain lines to the clean image through Photoshop software to obtain a synthetic rain-carrying image. After obtaining a clean and synthetic image pair, the number of training sample pairs is increased by adopting the means of turning, rotating, zooming and cutting, and then the model can be input for training.
And 2, constructing a network model, specifically comprising a rough rain line extraction module and a fine rain line purification module.
Step 2a, the rough rain line extraction module is realized by a simple two-layer convolution layer and two ReLU layers, and the rough rain line extraction with rain images is realized;
Figure BDA0002321719900000041
wherein E is1，E2The number of the rolling layers is two,
Figure BDA0002321719900000051
for convolution operations, r∈For the extracted noisy rain line image, y is the composite rain image.
Step 2b, constructing a fine rain line purification module which comprises seven convolution layers and four ReLU activation functions and is used for denoising the rough (noisy) rain line image to obtain a clean rain line;
the core content of the invention is a fine rain line purification module, and the key point is to solve three dictionaries S1、S2、S3And its transposed dictionary G1、G2、G3：
Figure BDA0002321719900000052
Wherein f isj，iI-th filter kernel, f, representing j-th dictionaryj，iThree filter groups are provided, namely j is 3, and the practical meaning is three rain line dictionaries with different scales, which are marked as S1、S2、S3Its transpose dictionary is denoted G1、G2、G3(S1、S2、S3、 G1、G2、G3All implemented by convolutional layers), c is the number of channels decomposed, zj，iFor convolutional sparse coding to be solved, | |1，||.||2Respectively represent l1Norm and l2The norm, λ, is a sparse penalty coefficient, which is set to 1 in this embodiment.
After sparse coding is obtained, a de-noised rain image can be reconstructed;
Figure BDA0002321719900000053
wherein the content of the first and second substances,
Figure BDA0002321719900000054
E3is a convolution layer, r is the final recovered fine rain image;
for the first minimization problem, the conventional sparse coding problem can be converted by using the convolution and matrix multiplication relation, and under the assumption of non-negative sparse coding, the solution is carried out by adopting an ISTA algorithm:
Figure BDA0002321719900000055
wherein the content of the first and second substances,
Figure BDA0002321719900000056
in order to be the middle symbol,
Figure BDA0002321719900000057
finger threshold, t is the number of iterations, S1、S2、S3For dictionaries of different dimensions, corresponding to transpose dictionary as G1、G2、G3。
Accordingly, a sparse code iterative solution process in the fine rain purification module can be obtained, a CNN implementation schematic diagram is shown in figure 2, and S1、S2、S3And G1、G2、G3All can be realized by a convolution layer. After passing through the rough rain line extraction module, the rough rain line image r∈Is input into a fine rain purification module part which is mainly divided into two steps: sparse coding solving and HR characteristic reconstruction, solving sparse coding corresponding to the graph 2, and outputting the optimal sparse coding after a certain number of times through an ISTA algorithm realized by iterative convolution.
Finally, the whole network flow is as follows: inputting the composite rain-bearing image y through two convolutions E1，E2And two ReLUs to obtain a rough rainline image r∈Will be
Figure BDA0002321719900000061
Are all initialized to r∈. Computing
Figure BDA0002321719900000062
Respectively pass through S1、S2、S3From the sum of (a) and (b), adding this sum from r∈Subtracting, and passing the difference value through G1、G2、G3Then respectively and
Figure BDA0002321719900000063
add to obtain
Figure BDA0002321719900000064
Repeating the above process to obtain
Figure BDA0002321719900000065
The final convolution sparse code is obtained, wherein t is iteration times, and the value of t is preferably 25. At the time of reconstruction, calculating
Figure BDA0002321719900000066
Respectively pass through S1、S2、S3After ReLU and then E3The fine rain image can be recovered.
And 3, training the network model by using the training set constructed in the step 1. Meanwhile, in the embodiment, an MSE loss function is selected:
Figure BDA0002321719900000067
wherein, theta refers to the network model parameter, l is the index of training data in the training set, and yl-rlRain-removed image output for network model, and real clean image xlAnd performing difference and accumulation to obtain a final error, and performing network optimization according to the final error.
And 4, inputting the rain-carrying image to be tested into the trained network model to obtain a corresponding rain-removing image.
In the test process, a peak signal-to-noise ratio (PSNR) and a Structural Similarity (SSIM) are used as measurement standards, and the measurement standards are specifically defined as follows:
PSNR＝10*log10(2552/mean(mean((X-Y)2)))
SSIM＝[L(X,Y)a]×[C(X,Y)b]×[S(X,Y)c]
wherein the content of the first and second substances,
Figure BDA0002321719900000068
μXand muYRespectively represent the mean values, σ, of X and YX、σYAnd σXYRepresenting the variance of X and Y and the covariance of the two, respectively.
Wherein, the higher the PSNR and SSIM values are, the better the reconstruction effect is.
In the test process, CNN, JORDER and DIDMDN are selected as comparison algorithms, visual comparison is shown in an attached figure 4, the method can remove rain lines in the image more easily, good detail information is stored at the same time, the rain removing effect of the comparison algorithms is not ideal, the rain is removed incompletely or a fuzzy result is generated, and even artifacts are generated. For the quantitative index, two commonly used data sets (Rain12 and Rain1200) were selected as the test set, and the test results are shown in table 1, and it can be seen that: the method of the patent greatly improves the PSNR and SSIM of the rain removing result, and the effectiveness of the method of the patent is demonstrated.
TABLE 1 test results
Figure BDA0002321719900000071
It should be understood that the above-mentioned embodiments are described in some detail, and not intended to limit the scope of the invention, and those skilled in the art may make substitutions and modifications within the scope of the invention without departing from the scope of the invention as defined by the appended claims.

Claims (6)

1. A single image rain removing method based on multi-scale dictionary learning is characterized by comprising the following steps:
step 1, according to the existing clean images, obtaining corresponding synthetic rain-carrying images by adding rain lines, taking each pair of clean images and synthetic rain-carrying images as a training sample pair, and establishing a training set;
step 2, constructing a network model, wherein the network model comprises a rough rain line extraction module and a fine rain line purification module, the rough rain line extraction module comprises two convolution layers and two ReLU activation function layers, and the rough rain line extraction module is used for realizing rough rain line extraction of combined rain images; the fine rain line purification module comprises seven convolution layers and four ReLU activation functions and is used for recovering a fine rain line graph from a noisy rain line image;
step 3, training the network model by using the training set constructed in the step 1,
and 4, inputting the rain-carrying image to be tested into the trained network model to obtain a corresponding rain-removing image.
2. The single-image rain removal method based on multi-scale dictionary learning as claimed in claim 1, wherein: the rough rain line extraction module in step 2 is implemented as follows,
Figure FDA0002321719890000011
wherein E is1，E2The number of the rolling layers is two,
Figure FDA0002321719890000012
for convolution operations, r∈For the extracted noisy rain line image, y is the composite rain image.
3. The single-image rain removal method based on multi-scale dictionary learning as claimed in claim 1, wherein: the detailed implementation manner of the fine rain line purification module in the step 2 is as follows,
the fine rain line purification module comprises a sparse coding solving part and an HR characteristic reconstruction part, wherein the sparse coding solves the following minimization problem in a convolution mode:
Figure FDA0002321719890000013
wherein f isj，iI < th > representing j < th > dictionaryFilter kernel, fj，i is three filter groups, i.e. j is 3, the actual meaning is three rain line dictionaries with different scales, which is marked as S1、S2、S3Its transpose dictionary is denoted G1、G2、G3In which S is1、S2、S3、G1、G2、G3All implemented by convolutional layers, c is the number of channels decomposed, zj，iFor convolutional sparse coding to be solved, | |1，||.||2Respectively represent l1Norm and l2Norm, lambda is sparse penalty coefficient;
after sparse coding is obtained, reconstructing a denoised rainchart:
Figure FDA0002321719890000014
wherein the content of the first and second substances,
Figure FDA0002321719890000021
E3is a convolution layer, r is the final recovered fine rain image;
the specific process is as follows: firstly, the first step is to
Figure FDA0002321719890000022
Are all initialized to r∈，r∈Extracting a noisy rainline image for the feature extraction module; computing
Figure FDA0002321719890000023
Respectively pass through S1、S2、S3From the sum of (a) and (b), adding this sum from r∈Subtracting, and passing the difference value through G1、G2、G3Then respectively and
Figure FDA0002321719890000024
add to obtain
Figure FDA0002321719890000025
Is repeated onThe process described above to obtain
Figure FDA0002321719890000026
The final convolution sparse code is obtained, wherein t is iteration times; at the time of reconstruction, calculating
Figure FDA0002321719890000027
Respectively pass through S1、S2、S3After ReLU and then E3I.e. to recover a fine rain image.
4. The single-image rain removal method based on multi-scale dictionary learning as claimed in claim 3, wherein: the formula (2) is converted into the traditional sparse coding problem by utilizing the multiplication relation of convolution and matrix, and is solved by adopting an ISTA algorithm under the assumption of non-negative sparse coding,
Figure FDA0002321719890000028
wherein the content of the first and second substances,
Figure FDA0002321719890000029
in order to be the middle symbol,
Figure FDA00023217198900000210
refers to a threshold value, and t is the number of iterations.
5. The single-image rain removal method based on multi-scale dictionary learning as claimed in claim 1, wherein: in step 3, global residual learning is introduced into the network model, an MSE loss function is selected, the minimum loss function is taken as a training target, and the expression of the MSE loss function is as follows:
Figure FDA00023217198900000211
wherein, theta refers to network model parameters, and l is a training setIndex of middle training sample, yl-rlRain-removed image output for network model, and real clean image xlAnd performing difference and accumulation to obtain a final error, so that the final error is minimized to realize optimization of the network model.
6. The single-image rain removal method based on multi-scale dictionary learning as claimed in claim 1, wherein: in the step 1, the number of images in a training set is increased by means of turning, rotating, zooming and cutting, then rain lines are added to each clean image through Photoshop to obtain a composite rain image, and the corresponding clean image and the composite rain image are used as a training sample pair.
CN201911300714.3A 2019-12-17 2019-12-17 Single-image rain removing method based on multi-scale dictionary learning Active CN111127354B (en)

Priority Applications (1)

Application Number Priority Date Filing Date Title
CN201911300714.3A CN111127354B (en) 2019-12-17 2019-12-17 Single-image rain removing method based on multi-scale dictionary learning

Applications Claiming Priority (1)

Application Number Priority Date Filing Date Title
CN201911300714.3A CN111127354B (en) 2019-12-17 2019-12-17 Single-image rain removing method based on multi-scale dictionary learning

Publications (2)

Publication Number Publication Date
CN111127354A true CN111127354A (en) 2020-05-08
CN111127354B CN111127354B (en) 2022-07-26

Family

ID=70498195

Family Applications (1)

Application Number Title Priority Date Filing Date
CN201911300714.3A Active CN111127354B (en) 2019-12-17 2019-12-17 Single-image rain removing method based on multi-scale dictionary learning

Country Status (1)

Country Link
CN (1) CN111127354B (en)

Cited By (7)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN111738932A (en) * 2020-05-13 2020-10-02 合肥师范学院 Automatic rain removing method for photographed image of vehicle-mounted camera
CN111861935A (en) * 2020-07-29 2020-10-30 天津大学 Rain removing method based on image restoration technology
CN112712486A (en) * 2020-11-26 2021-04-27 贵州大学 Radio astronomical image reconstruction method based on activation function
CN113222888A (en) * 2021-03-19 2021-08-06 复旦大学 Textile yarn weaving size detection method based on depth texture characteristics
CN113902931A (en) * 2021-09-17 2022-01-07 淮阴工学院 Image rain removing method based on learning type convolution sparse coding
WO2023202265A1 (en) * 2022-04-19 2023-10-26 腾讯科技（深圳）有限公司 Image processing method and apparatus for artifact removal, and device, product and medium
CN112712486B (en) * 2020-11-26 2024-04-26 贵州大学 Radio astronomical image reconstruction method based on activation function

Citations (6)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
US20110188775A1 (en) * 2010-02-01 2011-08-04 Microsoft Corporation Single Image Haze Removal Using Dark Channel Priors
CN107133935A (en) * 2017-05-25 2017-09-05 华南农业大学 A kind of fine rain removing method of single image based on depth convolutional neural networks
CN107657593A (en) * 2017-04-20 2018-02-02 湘潭大学 A kind of removing rain based on single image method
CN108765344A (en) * 2018-05-30 2018-11-06 南京信息工程大学 A method of the single image rain line removal based on depth convolutional neural networks
CN109360155A (en) * 2018-08-17 2019-02-19 上海交通大学 Single-frame images rain removing method based on multi-scale feature fusion
CN110570351A (en) * 2019-08-01 2019-12-13 武汉大学 Image super-resolution reconstruction method based on convolution sparse coding

Patent Citations (6)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
US20110188775A1 (en) * 2010-02-01 2011-08-04 Microsoft Corporation Single Image Haze Removal Using Dark Channel Priors
CN107657593A (en) * 2017-04-20 2018-02-02 湘潭大学 A kind of removing rain based on single image method
CN107133935A (en) * 2017-05-25 2017-09-05 华南农业大学 A kind of fine rain removing method of single image based on depth convolutional neural networks
CN108765344A (en) * 2018-05-30 2018-11-06 南京信息工程大学 A method of the single image rain line removal based on depth convolutional neural networks
CN109360155A (en) * 2018-08-17 2019-02-19 上海交通大学 Single-frame images rain removing method based on multi-scale feature fusion
CN110570351A (en) * 2019-08-01 2019-12-13 武汉大学 Image super-resolution reconstruction method based on convolution sparse coding

Non-Patent Citations (2)

* Cited by examiner, † Cited by third party
Title
YANYAN WEI ET.AL: "A Coarse-to-fine Multi-stream Hybrid Deraining Network for Single Image Deraining", 《ARXIV》 *
郭继昌 等: "多尺度卷积神经网络的单幅图像去雨方法", 《哈尔滨工业大学学报》 *

Cited By (8)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN111738932A (en) * 2020-05-13 2020-10-02 合肥师范学院 Automatic rain removing method for photographed image of vehicle-mounted camera
CN111861935A (en) * 2020-07-29 2020-10-30 天津大学 Rain removing method based on image restoration technology
CN111861935B (en) * 2020-07-29 2022-06-03 天津大学 Rain removing method based on image restoration technology
CN112712486A (en) * 2020-11-26 2021-04-27 贵州大学 Radio astronomical image reconstruction method based on activation function
CN112712486B (en) * 2020-11-26 2024-04-26 贵州大学 Radio astronomical image reconstruction method based on activation function
CN113222888A (en) * 2021-03-19 2021-08-06 复旦大学 Textile yarn weaving size detection method based on depth texture characteristics
CN113902931A (en) * 2021-09-17 2022-01-07 淮阴工学院 Image rain removing method based on learning type convolution sparse coding
WO2023202265A1 (en) * 2022-04-19 2023-10-26 腾讯科技（深圳）有限公司 Image processing method and apparatus for artifact removal, and device, product and medium

Also Published As

Publication number Publication date
CN111127354B (en) 2022-07-26

Similar Documents

Publication Publication Date Title
CN111127354B (en) Single-image rain removing method based on multi-scale dictionary learning
CN112200750B (en) Ultrasonic image denoising model establishing method and ultrasonic image denoising method
Wei et al. 3-D quasi-recurrent neural network for hyperspectral image denoising
Xu et al. Image inpainting by patch propagation using patch sparsity
CN111915530B (en) End-to-end-based haze concentration self-adaptive neural network image defogging method
CN112308860A (en) Earth observation image semantic segmentation method based on self-supervision learning
CN110570351B (en) Image super-resolution reconstruction method based on convolution sparse coding
CN112150354B (en) Single image super-resolution method combining contour enhancement and denoising statistical prior
CN112419174B (en) Image character removing method, system and device based on gate cycle unit
CN114723631A (en) Image denoising method, system and device based on depth context prior and multi-scale reconstruction sub-network
CN113191969A (en) Unsupervised image rain removing method based on attention confrontation generation network
Xu et al. Deep sparse representation based image restoration with denoising prior
CN114723630A (en) Image deblurring method and system based on cavity double-residual multi-scale depth network
CN114820352A (en) Hyperspectral image denoising method and device and storage medium
CN115063318A (en) Adaptive frequency-resolved low-illumination image enhancement method and related equipment
CN113962905A (en) Single image rain removing method based on multi-stage feature complementary network
Hussain et al. Image denoising to enhance character recognition using deep learning
CN113066025A (en) Image defogging method based on incremental learning and feature and attention transfer
CN116205821A (en) Single-image rain removing method based on vertical stripe characteristic extraction cross convolution
Li et al. Distribution-transformed network for impulse noise removal
CN115272113A (en) Image deblurring method based on multi-scale frequency separation network
CN115170812A (en) Image denoising model training and denoising method, device and storage medium thereof
CN110288525B (en) Multi-dictionary super-resolution image reconstruction method
CN114219738A (en) Single-image multi-scale super-resolution reconstruction network structure and method
CN114723614B (en) Image restoration method based on wavelet domain ADMM depth network

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