CN111145125A - Image denoising method based on residual learning and convolutional neural network - Google Patents
Image denoising method based on residual learning and convolutional neural network Download PDFInfo
- Publication number
- CN111145125A CN111145125A CN201911391449.4A CN201911391449A CN111145125A CN 111145125 A CN111145125 A CN 111145125A CN 201911391449 A CN201911391449 A CN 201911391449A CN 111145125 A CN111145125 A CN 111145125A
- Authority
- CN
- China
- Prior art keywords
- neural network
- convolutional neural
- image
- residual
- 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.)
- Pending
Links
- 238000013527 convolutional neural network Methods 0.000 title claims abstract description 67
- 238000000034 method Methods 0.000 title claims abstract description 61
- 230000008569 process Effects 0.000 claims abstract description 15
- ORILYTVJVMAKLC-UHFFFAOYSA-N Adamantane Natural products C1C(C2)CC3CC1CC2C3 ORILYTVJVMAKLC-UHFFFAOYSA-N 0.000 claims abstract description 13
- 238000010606 normalization Methods 0.000 claims abstract description 9
- 238000013507 mapping Methods 0.000 claims description 20
- 238000012549 training Methods 0.000 claims description 16
- 230000006870 function Effects 0.000 claims description 14
- 210000002569 neuron Anatomy 0.000 claims description 14
- 238000012545 processing Methods 0.000 claims description 14
- 238000003062 neural network model Methods 0.000 claims description 10
- 230000004913 activation Effects 0.000 claims description 9
- 238000013528 artificial neural network Methods 0.000 claims description 7
- 230000001105 regulatory effect Effects 0.000 claims description 6
- 238000004364 calculation method Methods 0.000 claims description 3
- 238000012937 correction Methods 0.000 claims description 3
- 230000000694 effects Effects 0.000 abstract description 11
- 230000005540 biological transmission Effects 0.000 abstract description 6
- 230000015556 catabolic process Effects 0.000 abstract description 3
- 238000006731 degradation reaction Methods 0.000 abstract description 3
- 230000014759 maintenance of location Effects 0.000 abstract description 2
- 238000010586 diagram Methods 0.000 description 3
- 238000005516 engineering process Methods 0.000 description 3
- 230000008859 change Effects 0.000 description 2
- 238000012423 maintenance Methods 0.000 description 2
- 238000011176 pooling Methods 0.000 description 2
- 238000011160 research Methods 0.000 description 2
- 230000035945 sensitivity Effects 0.000 description 2
- 230000002411 adverse Effects 0.000 description 1
- 230000009286 beneficial effect Effects 0.000 description 1
- 230000006835 compression Effects 0.000 description 1
- 238000007906 compression Methods 0.000 description 1
- 238000013135 deep learning Methods 0.000 description 1
- 230000007850 degeneration Effects 0.000 description 1
- 238000001514 detection method Methods 0.000 description 1
- 238000001035 drying Methods 0.000 description 1
- 230000008030 elimination Effects 0.000 description 1
- 238000003379 elimination reaction Methods 0.000 description 1
- 230000002708 enhancing effect Effects 0.000 description 1
- 238000000605 extraction Methods 0.000 description 1
- 230000006872 improvement Effects 0.000 description 1
- 239000011159 matrix material Substances 0.000 description 1
- 238000007781 pre-processing Methods 0.000 description 1
- 230000009467 reduction Effects 0.000 description 1
- 238000011946 reduction process Methods 0.000 description 1
- 230000000717 retained effect Effects 0.000 description 1
- 238000013519 translation Methods 0.000 description 1
Images
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T5/00—Image enhancement or restoration
- G06T5/70—Denoising; Smoothing
-
- 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
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/08—Learning methods
- G06N3/084—Backpropagation, e.g. using gradient descent
Landscapes
- Engineering & Computer Science (AREA)
- Physics & Mathematics (AREA)
- Theoretical Computer Science (AREA)
- General Physics & Mathematics (AREA)
- Data Mining & Analysis (AREA)
- Molecular Biology (AREA)
- Biophysics (AREA)
- Computational Linguistics (AREA)
- Artificial Intelligence (AREA)
- Evolutionary Computation (AREA)
- General Health & Medical Sciences (AREA)
- Biomedical Technology (AREA)
- Computing Systems (AREA)
- General Engineering & Computer Science (AREA)
- Life Sciences & Earth Sciences (AREA)
- Mathematical Physics (AREA)
- Software Systems (AREA)
- Health & Medical Sciences (AREA)
- Image Processing (AREA)
Abstract
本发明公开了一种基于残差学习与卷积神经网络的图像去噪方法，包括以下步骤：先分别采用卷积神经网络方法、残差学习的方法对输入的含噪图像去噪；再采用卷积神经网络方法与残差学习方法相结合的方法对含噪图像去噪，在卷积神经网络中添加padding，进行批规范化操作，在网络中加入浅层到深层的跨越连接结构，再采用Adam算法的卷积神经网络进行训练；最后，将去噪后的图像输出。本发明的方法不仅能拓展网络深度，有效避免网络退化以及信息在传递的工程中的丢失与损耗，提高卷积神经网络去噪模型深度和结构信息的保持效果。
The invention discloses an image denoising method based on residual learning and convolutional neural network. The method of combining the convolutional neural network method and the residual learning method denoises the noisy image, adding padding to the convolutional neural network, performing batch normalization operations, adding a shallow-to-deep cross-connection structure to the network, and then using The convolutional neural network of the Adam algorithm is trained; finally, the denoised image is output. The method of the invention can not only expand the network depth, effectively avoid network degradation and information loss and loss in the process of transmission, and improve the depth of the convolutional neural network denoising model and the retention effect of structural information.
Description
Technical Field
The invention relates to an image processing technology, in particular to an image denoising method based on residual learning and a convolutional neural network.
Background
In the process of image acquisition, coding and transmission, some noises are introduced for various reasons, which leads to serious reduction of image quality, so that the image is denoised, and the improvement of image quality is an important research part in the image processing technology. Through some noise processing technologies, image noise is removed as much as possible, so that information is obtained more efficiently, and further processing such as feature extraction, signal detection and image compression is facilitated for the image. However, some conventional denoising methods always cause loss of information in the transmission process, and when a convolutional neural network is used for image processing operation, preprocessing operation required for inputting information is greatly relaxed, so that the processing process is simplified. The most key in the Convolutional Neural Network (CNN) is convolution operation, which makes full use of the information of adjacent areas in the picture and greatly reduces the scale of the parameter matrix by means of sparse connection and weight sharing.
However, the convolutional neural network cannot well maintain the image structure information, and in the process of deepening the network depth, the training difficulty is increased and the training effect is reduced, so that research on related contents is urgently needed to improve the maintenance effect of the convolutional neural network structure information.
Disclosure of Invention
The purpose of the invention is as follows: the invention aims to provide an image denoising method based on residual error learning and a convolutional neural network, which can increase the stability of network training, avoid information loss and loss in the transmission process and improve the depth of a convolutional neural network denoising model and the structural information retention effect.
The technical scheme is as follows: the invention discloses an image denoising method based on residual error learning and a convolutional neural network, which comprises the following steps of:
(1) denoising the input noisy image by adopting a convolutional neural network method, and constructing a convolutional neural network model comprising a plurality of convolutional layers;
(2) denoising an input noisy image by adopting a residual error learning method, superposing a network mapping layer on a neural network, and equating an original deep neural network model to a shallow neural network model;
(3) denoising the noisy image by adopting an image denoising method combining a convolutional neural network method and a residual error learning method, adding padding in the convolutional neural network, performing batch normalization operation, adding a shallow-to-deep crossing connection structure in the obtained convolutional neural network, and then training by adopting the convolutional neural network of an Adam algorithm;
(4) and outputting the denoised image.
In the step (2), the residual function formula of the network mapping layer is as follows:
F(x)＝H(x)-x
wherein, F(x)Representing a residual function, H(x)Representing a network mapping function, and x representing an identity mapping layer; the depth of the image can be deepened, and the phenomenon of degradation can not occur.
In the step (3), padding is added to the convolutional neural network to fill zero around the image block; not only can prevent the size of the output characteristic graph from changing, but also can repeatedly utilize the edge information.
In the step (3), the normalized specific calculation method is as follows:
wherein, aiIs a certain neuron's original activation value, ai normIs a normalized value, gamma, after a normalization operationi、βiTwo regulating factors of the neuron in the training process are adopted, the regulating factors slightly regulate the numerical values which are normalized to have a mean value of 0 and a variance of 1, and then a condition that the mean value is βiVariance is gammai 2The distribution value, thereby preventing the interference of the internal covariates.
When the batch of normalized training networks is carried out, a neuron set S is selected firstly, and then the mean value and the variance of the neuron activation values are calculated, wherein the specific formula is as follows:
Wherein m is the size of the set S, and epsilon is a positive number with a numerical value close to zero; avoiding the situation that the gradient is not defined at a certain position, and after the normalization operation, the output result can be normalized to be in the range no matter what the change of the neuron occurs.
In the step (2), the convolutional neural network adopting the Adam algorithm is used for training, specifically, the Adam algorithm is used for updating the weight, the estimation with the deviation is calculated, the estimation after the deviation correction is calculated later is improved, and the parameters in the Adam algorithm are set as default values; the algorithm parameter adjustment process is simple, and most problems in the denoising process can be solved.
And each layer of convolution kernels of the convolution layers in the convolution neural network model is identical in number and size.
And the convolution layer in the convolutional neural network model adopts a ReLU activation function.
In the step (1), the original image is subjected to gray scale processing before the noisy image is input, and is converted into a gray scale image, and the gray scale image is subjected to noise processing to obtain a noisy image and then is input.
The noise adding process is to add Gaussian noise with a fixed variance value in the gray level image, so that the noise removing effect is better.
Has the advantages that: compared with the prior art, the invention has the beneficial effects that: (1) loss and loss of information in the transmission process are avoided, and the depth of a convolutional neural network denoising model and the keeping effect of structural information are improved; (2) the stability of network training is increased; (3) enhancing the sensitivity of the convolutional neural network to the shallow and deep features of the image and improving the learning effect of the drying mapping; (4) the sizes of the feature images after convolution are not changed, and images with the same size are finally output.
Drawings
FIG. 1 is a flowchart of an image denoising method based on residual learning and convolutional neural network according to the present invention;
FIG. 2 is a frame diagram of the image denoising method based on residual learning and convolutional neural network according to the present invention;
FIG. 3 is a gray scale image obtained by performing gray scale processing on an original image according to the present invention;
FIG. 4 is a noisy image obtained after a noise process is applied to a gray scale image in accordance with the present invention;
FIG. 5 is a diagram showing a structure of a residual error unit according to the present invention;
FIG. 6 is a diagram illustrating the effect of denoising a noisy image according to the present invention;
Detailed Description
The invention is described in further detail below with reference to specific embodiments and the attached drawing figures.
As shown in fig. 1, the image denoising method based on residual learning and convolutional neural network of the present invention includes the following steps:
(1) denoising the input noisy image by adopting a convolutional neural network method, and constructing a convolutional neural network model comprising a plurality of convolutional layers; the convolution kernels of each layer of convolution layers in the convolution neural network model are identical in number and size, convolution is carried out by adopting convolution kernels of 3 x 3, and the number of the convolution kernels of each layer is 64; a convolution layer in the convolutional neural network model adopts a ReLU activation function;
(2) denoising an input noisy image by adopting a residual error learning method, superposing a network mapping layer on a neural network, and equating an original deep neural network model to a shallow neural network model, so that the depth is deepened, and the image is not degraded;
(3) denoising the noisy image by adopting an image denoising method combining a convolutional neural network method and a residual error learning method, adding padding in the convolutional neural network, performing batch normalization operation, adding a shallow-to-deep crossing connection structure in the obtained convolutional neural network, and then training by adopting the convolutional neural network of an Adam algorithm;
the padding added in the convolutional neural network is zero padding around the image block, so that the size change of the output feature map can be prevented, and the edge information can be used for multiple times. Because the size of the adopted convolution kernel is 3 multiplied by 3, and padding with the size of 1 is added to each layer, the size of the feature image after convolution is unchanged, and finally the image with the same size can be output;
the specific calculation method of the normalization is as follows:
wherein, aiFor a certain value of the original activation of a neuron,
is a normalized value, gamma, after a normalization operationi、βiTwo regulating factors of the neuron in the training process are adopted, the regulating factors slightly regulate the numerical values which are normalized to have a mean value of 0 and a variance of 1, and then a condition that the mean value is βiVariance is gammai 2The value of the distribution, thereby preventing interference of the internal covariates;
when the training network is normalized in batches, a neuron set S is selected firstly, and then the mean value and the variance of the neuron activation value are calculated, wherein the specific formula is as follows:
Wherein m is the size of the set S, and epsilon is a positive number with a numerical value close to zero; avoiding the situation that the gradient is not defined at a certain position;
after standardized operation, no matter what the neuron changes, the output result is normalized to the range, in addition, the translation parameter and the scaling parameter can be self-adaptively learned, the introduced parameters are few, and the parameters can be updated by using back propagation, so that the convergence speed of neural network training can be accelerated to a certain extent, and the degree of dependence on the network parameter initialization process is reduced;
as shown in fig. 2, the head-to-tail crossing connection from the shallow layer to the deep layer is added to the convolutional neural network, so that the problem that the shallow image features are lost due to the higher-dimensional feature image obtained after the input image is subjected to multiple convolution operations in the forward propagation process in the network can be avoided, the shallow information of the image can be retained, the sensitivity of the convolutional neural network to the shallow and deep image features can be enhanced, and the learning effect of denoising mapping can be improved; in addition, the convolutional neural network adopting the Adam algorithm is trained, specifically, the weight is updated by adopting the Adam algorithm, the estimation with deviation is calculated, the estimation after the deviation correction is calculated later is improved, and the parameters in the Adam algorithm are set as default values;
(4) and outputting the denoised image.
As shown in fig. 2, the convolutional neural network model of the present invention only includes a plurality of convolutional layers, because when denoising is performed using the convolutional neural network, the dimensionality reduction process of the pooling operation filters out part of the features, which adversely affects the denoising, and therefore, the network model of the present invention removes the pooling layer and only retains the convolutional layers.
As shown in fig. 3 and 4, in step (1), before the noisy image is input, the original image is subjected to gray scale processing to be converted into a gray scale image, and the gray scale image is subjected to noise processing to obtain a noisy image, and then the noisy image is input into each model to be subjected to noise removal processing respectively. In this embodiment, the noise addition process is to add gaussian noise having a variance value of 25 to the grayscale image.
In order to increase the depth of the network and ensure that the network does not degenerate, the residual is used in deep learning, an identity mapping layer (y-x layer) is superimposed on a neural network, and the original deep neural network model is equivalent to a shallow neural network, so that the depth is deepened, and the phenomenon of degeneration does not occur. Since it is not easy to directly implement an identity mapping function h (x) ═ x in the neural network layer, the mapping of the network is changed to h (x) ═ f (x) + x, as shown in fig. 5, the model includes two mapping, one is identification mapping performed by using a short connection structure on the right side, and the other is residual part f (x) of residual mapping, and the model can convert the problem into learning a residual function, so the residual function formula of the network mapping layer in step (2) is expressed as:
F(x)＝H(x)-x
wherein, F(x)Representing a residual function, H(x)Representing the network mapping function and x representing the identity mapping layer.
The present invention is compared with a three-dimensional block matching algorithm (BM3D), a convolutional neural network denoising method including 5 convolutional layers (novel method 1), and a convolutional neural network denoising method including 19 convolutional layers (novel method 2):
in the above four comparison methods, the sizes of the convolution kernels are all 3 × 3. After the same group of gray level images are denoised, the peak signal-to-noise ratio (PSNR) and the used time are compared, and the results clearly show that the PSNR is obviously improved and the consumed time is obviously reduced by adopting the denoising method disclosed by the invention, so that the image denoising method based on residual learning and the convolutional neural network has the best denoising effect, as shown in fig. 6, the residual learning is combined into the convolutional neural network, the network depth can be expanded, the high-dimensional characteristic is extracted, the network degradation and the loss and loss of information in the transmission project are effectively avoided, the noise elimination is considered, and the depth of the denoising model of the convolutional neural network and the maintenance effect of structural information are improved.
Claims (10)
1. An image denoising method based on residual learning and a convolutional neural network is characterized by comprising the following steps:
(1) denoising the input noisy image by adopting a convolutional neural network method, and constructing a convolutional neural network model comprising a plurality of convolutional layers;
(2) denoising an input noisy image by adopting a residual error learning method, superposing a network mapping layer on a neural network, and equating an original deep neural network model to a shallow neural network model;
(3) denoising the noisy image by adopting an image denoising method combining a convolutional neural network method and a residual error learning method, adding padding in the convolutional neural network, performing batch normalization operation, adding a shallow-to-deep crossing connection structure in the obtained convolutional neural network, and then training by adopting the convolutional neural network of an Adam algorithm;
(4) and outputting the denoised image.
2. The method for denoising an image based on residual error learning and convolutional neural network of claim 1, wherein in step (2), the residual error function formula of the network mapping layer is:
F(x)＝H(x)-x
wherein, F(x)Representing a residual function, H(x)Representing the network mapping function and x representing the identity mapping layer.
3. The image denoising method based on residual learning and convolutional neural network of claim 1, wherein: in the step (3), padding is added to the convolutional neural network to fill zeros around the image block.
4. The image denoising method based on residual learning and convolutional neural network of claim 1, wherein: in the step (3), the normalized specific calculation method is as follows:
wherein, aiFor a certain value of the original activation of a neuron,
is a normalized value, gamma, after a normalization operationi、βiTwo regulating factors of the neuron in the training process are adopted, the regulating factors slightly regulate the numerical values which are normalized to have a mean value of 0 and a variance of 1, and then a condition that the mean value is βiVariance is gammai 2The value of the distribution.
5. The image denoising method based on residual learning and convolutional neural network of claim 1, wherein: when the batch of normalized training networks is carried out, a neuron set S is selected firstly, and then the mean value and the variance of the neuron activation values are calculated, wherein the specific formula is as follows:
where m is the size of the set S and ε is a positive number with a value close to zero.
6. The image denoising method based on residual error learning and convolutional neural network of claim 1, wherein in the step (2), the convolutional neural network training using Adam algorithm is specifically to use Adam algorithm to update weights, and by calculating the estimation with deviation, the estimation after the deviation correction is calculated later is improved, and the parameters in Adam algorithm are set as default values.
7. The image denoising method based on residual learning and convolutional neural network of claim 1, wherein: and each layer of convolution kernels of the convolution layers in the convolution neural network model is identical in number and size.
8. The image denoising method based on residual learning and convolutional neural network of claim 1, wherein: and the convolution layer in the convolutional neural network model adopts a ReLU activation function.
9. The image denoising method based on residual learning and convolutional neural network of claim 1, wherein: in the step (1), before the noisy image is input, the original image is subjected to gray scale processing to be converted into a gray scale image, and the gray scale image is subjected to noise processing to obtain a noisy image and then input.
10. The image denoising method based on residual learning and convolutional neural network of claim 9, wherein: the noise adding process is to add Gaussian noise with a fixed variance value in the gray level image.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201911391449.4A CN111145125A (en) | 2019-12-30 | 2019-12-30 | Image denoising method based on residual learning and convolutional neural network |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201911391449.4A CN111145125A (en) | 2019-12-30 | 2019-12-30 | Image denoising method based on residual learning and convolutional neural network |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN111145125A true CN111145125A (en) | 2020-05-12 |
Family
ID=70521721
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201911391449.4A Pending CN111145125A (en) | 2019-12-30 | 2019-12-30 | Image denoising method based on residual learning and convolutional neural network |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN111145125A (en) |
Cited By (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN111580162A (en) * | 2020-05-21 | 2020-08-25 | 长江大学 | Seismic data random noise suppression method based on residual convolutional neural network |
| CN114241188A (en) * | 2021-11-26 | 2022-03-25 | 电子科技大学长三角研究院(湖州) | A multi-scale target detection method and system for smooth transfer of semantic information |
| CN114845042A (en) * | 2022-03-14 | 2022-08-02 | 南京大学 | Camera automatic focusing method based on image information entropy |
| CN116703772A (en) * | 2023-06-15 | 2023-09-05 | 山东财经大学 | Image denoising method, system and terminal based on adaptive interpolation algorithm |
Citations (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US20080240607A1 (en) * | 2007-02-28 | 2008-10-02 | Microsoft Corporation | Image Deblurring with Blurred/Noisy Image Pairs |
| CN108564553A (en) * | 2018-05-07 | 2018-09-21 | 南方医科大学 | Low-dose CT image noise suppression method based on convolutional neural networks |
| CN109978778A (en) * | 2019-03-06 | 2019-07-05 | 浙江工业大学 | Convolutional neural networks medicine CT image denoising method based on residual error study |
-
2019
- 2019-12-30 CN CN201911391449.4A patent/CN111145125A/en active Pending
Patent Citations (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US20080240607A1 (en) * | 2007-02-28 | 2008-10-02 | Microsoft Corporation | Image Deblurring with Blurred/Noisy Image Pairs |
| CN108564553A (en) * | 2018-05-07 | 2018-09-21 | 南方医科大学 | Low-dose CT image noise suppression method based on convolutional neural networks |
| CN109978778A (en) * | 2019-03-06 | 2019-07-05 | 浙江工业大学 | Convolutional neural networks medicine CT image denoising method based on residual error study |
Non-Patent Citations (1)
| Title |
|---|
| XIAO-JIAO MAO等: "Image Restoration Using Very Deep Convolutional Encoder-Decoder Networks with Symmetric Skip Connections" * |
Cited By (6)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN111580162A (en) * | 2020-05-21 | 2020-08-25 | 长江大学 | Seismic data random noise suppression method based on residual convolutional neural network |
| CN114241188A (en) * | 2021-11-26 | 2022-03-25 | 电子科技大学长三角研究院(湖州) | A multi-scale target detection method and system for smooth transfer of semantic information |
| CN114845042A (en) * | 2022-03-14 | 2022-08-02 | 南京大学 | Camera automatic focusing method based on image information entropy |
| CN114845042B (en) * | 2022-03-14 | 2023-03-10 | 南京大学 | Camera automatic focusing method based on image information entropy |
| CN116703772A (en) * | 2023-06-15 | 2023-09-05 | 山东财经大学 | Image denoising method, system and terminal based on adaptive interpolation algorithm |
| CN116703772B (en) * | 2023-06-15 | 2024-03-15 | 山东财经大学 | An image denoising method, system and terminal based on adaptive interpolation algorithm |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN110599409B (en) | Convolutional neural network image denoising method based on multi-scale convolutional groups and parallel | |
| CN109658344B (en) | Image denoising method, device, device and storage medium based on deep learning | |
| CN111145125A (en) | Image denoising method based on residual learning and convolutional neural network | |
| CN110045419B (en) | A Denoising Method for Seismic Data Based on Perceptron Residual Self-encoding Network | |
| CN108510485B (en) | A reference-free image quality assessment method based on convolutional neural network | |
| CN109087273B (en) | Image restoration method, storage medium and system based on enhanced neural network | |
| CN109961396B (en) | An image super-resolution reconstruction method based on convolutional neural network | |
| CN109410127B (en) | Image denoising method based on deep learning and multi-scale image enhancement | |
| Wang et al. | Channel and space attention neural network for image denoising | |
| CN110781895A (en) | Image semantic segmentation method based on convolutional neural network | |
| CN113763268B (en) | Blind restoration method and system for face image | |
| CN115063318A (en) | Low-light image enhancement method and related equipment based on adaptive frequency decomposition | |
| CN110111266B (en) | An Improved Approximate Information Transfer Algorithm Based on Deep Learning Denoising | |
| CN110232668B (en) | A multi-scale image enhancement method | |
| CN112581397B (en) | Degraded image restoration method, system, medium and equipment based on image priori information | |
| CN113362241B (en) | Depth map denoising method combining high-low frequency decomposition and two-stage fusion strategy | |
| CN113191983A (en) | Image denoising method and device based on deep learning attention mechanism | |
| CN112991198A (en) | Blind denoising method based on noise modeling | |
| CN113888405B (en) | Denoising and Demosaicing Method Based on Clustering Adaptive Dilated Convolutional Neural Network | |
| Li et al. | Stimulating diffusion model for image denoising via adaptive embedding and ensembling | |
| CN114359076A (en) | Self-adaptive weighted Gaussian curvature filtering method based on image edge indication function | |
| CN111598788B (en) | Single image defogging method based on quadtree decomposition and non-local prior | |
| CN115358910A (en) | Digital watermark attack method and system based on convolutional neural network denoising algorithm | |
| Xing et al. | Vision graph convolutional network for underwater image enhancement | |
| CN114998666B (en) | A degradation kernel extraction method for image blind super-resolution enhancement 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 | ||
| RJ01 | Rejection of invention patent application after publication |
Application publication date: 20200512 |
|
| RJ01 | Rejection of invention patent application after publication |