CN110503608A - The image de-noising method of convolutional neural networks based on multi-angle of view - Google Patents
The image de-noising method of convolutional neural networks based on multi-angle of view Download PDFInfo
- Publication number
- CN110503608A CN110503608A CN201910632475.5A CN201910632475A CN110503608A CN 110503608 A CN110503608 A CN 110503608A CN 201910632475 A CN201910632475 A CN 201910632475A CN 110503608 A CN110503608 A CN 110503608A
- Authority
- CN
- China
- Prior art keywords
- image
- network
- angle
- view
- denoising
- 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
Links
- 238000000034 method Methods 0.000 title claims abstract description 43
- 238000013527 convolutional neural network Methods 0.000 title claims abstract description 22
- 230000000694 effects Effects 0.000 claims description 3
- 238000010606 normalization Methods 0.000 claims description 3
- 238000005516 engineering process Methods 0.000 abstract description 11
- 201000010099 disease Diseases 0.000 abstract description 4
- 208000037265 diseases, disorders, signs and symptoms Diseases 0.000 abstract description 4
- 230000007812 deficiency Effects 0.000 abstract description 2
- 230000004927 fusion Effects 0.000 abstract description 2
- 230000006870 function Effects 0.000 description 12
- 238000012545 processing Methods 0.000 description 6
- 238000000605 extraction Methods 0.000 description 5
- 238000012549 training Methods 0.000 description 4
- 238000011084 recovery Methods 0.000 description 3
- 230000007423 decrease Effects 0.000 description 2
- 230000001815 facial effect Effects 0.000 description 2
- 238000005457 optimization Methods 0.000 description 2
- 241000208340 Araliaceae Species 0.000 description 1
- 241000282326 Felis catus Species 0.000 description 1
- 235000005035 Panax pseudoginseng ssp. pseudoginseng Nutrition 0.000 description 1
- 235000003140 Panax quinquefolius Nutrition 0.000 description 1
- 230000003044 adaptive effect Effects 0.000 description 1
- 230000006835 compression Effects 0.000 description 1
- 238000007906 compression Methods 0.000 description 1
- 238000007796 conventional method Methods 0.000 description 1
- 238000013135 deep learning Methods 0.000 description 1
- 238000003745 diagnosis Methods 0.000 description 1
- 235000008434 ginseng Nutrition 0.000 description 1
- 238000013507 mapping Methods 0.000 description 1
- 230000002441 reversible effect Effects 0.000 description 1
- 238000012360 testing method Methods 0.000 description 1
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
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T2207/00—Indexing scheme for image analysis or image enhancement
- G06T2207/20—Special algorithmic details
- G06T2207/20021—Dividing image into blocks, subimages or windows
-
- 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
- G06T2207/20048—Transform domain processing
- G06T2207/20056—Discrete and fast Fourier transform, [DFT, FFT]
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T2207/00—Indexing scheme for image analysis or image enhancement
- G06T2207/20—Special algorithmic details
- G06T2207/20081—Training; Learning
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T2207/00—Indexing scheme for image analysis or image enhancement
- G06T2207/20—Special algorithmic details
- G06T2207/20084—Artificial neural networks [ANN]
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T2207/00—Indexing scheme for image analysis or image enhancement
- G06T2207/30—Subject of image; Context of image processing
- G06T2207/30196—Human being; Person
- G06T2207/30201—Face
-
- 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
- Y02T—CLIMATE CHANGE MITIGATION TECHNOLOGIES RELATED TO TRANSPORTATION
- Y02T10/00—Road transport of goods or passengers
- Y02T10/10—Internal combustion engine [ICE] based vehicles
- Y02T10/40—Engine management systems
Landscapes
- Physics & Mathematics (AREA)
- General Physics & Mathematics (AREA)
- Engineering & Computer Science (AREA)
- Theoretical Computer Science (AREA)
- Image Processing (AREA)
- Image Analysis (AREA)
Abstract
The invention discloses a kind of image de-noising methods of convolutional neural networks based on multi-angle of view.The present invention is accelerated to denoise the convergence of network instead of BN with INBN technology；This method can effectively mend the deficiency with regard to BN and accelerate network convergence, can handle real noise image, blind noise and Gaussian noise.The present invention is only denoised with 20 layers of network, reduces the calculating cost of network.In addition, linear data is more preferably transformed to nonlinear data with a kind of new technology GF technology by the present invention；Denoising model is preferably trained with Smooth function.In addition, it enhances network performance using the Fusion Features of multi-angle of view.The present invention diagnoses a disease and is of great significance for disaster relief, aviation exploration and medical treatment in reality.
Description
Technical field
The present invention relates to technical field of image processing, are based especially on the image denoising side of the convolutional neural networks of multi-angle of view
Method.
Background technique
Digital image device has been widely applied to multiple fields, such as medical diagnosis on disease, person identification and calamity
Hardly possible rescue.And vision facilities is usually influenced by camera shake, low light, greasy weather rainy day etc. in shooting photo, this causes to shoot
The photo come is unintelligible, and Image Denoising Technology is exactly these not to be known image restoring to highly clearly image, this process
Referred to as reversible process, it is y=x+ μ that it, which denoises the method relied primarily on, and wherein y is noise image, and x is the clean figure recovered to
Picture, μ are noises.From the perspective of Bayes, priori knowledge is crucial for image denoising, while some scholars are at this
Aspect has done many work, as Sparse methods have good robustness for image denoising task.It is adaptive with NOT logic
Property optimizes Sparse methods and applies in denoising task.Dictionary learning is used to removal noise, can also efficiently reduce calculating
Cost.Full variational regularization method can make image become more smooth, this is conducive to restore clean image.In addition, Ma Erke
The Block- matching filter method of husband, Weighted Kernel norm minimum and 3 dimensions are the denoising methods of mainstream.Although these methods are in image
Fine performance is had been achieved in denoising, but these methods still suffer from following challenge:
(1) these methods need artificial manual setting parameter to obtain optimal performance；
(2) these methods need to need complicated optimization algorithm in test phase, this greatly increases the calculating of these methods
Cost；
(3) this method can only solve this problem for a kind of situation one model of training, if level of noise is 25
Gaussian noise, they can only have a model to solve this problem, and noise image is complicated in life, this is limited significantly
Their application range.
In recent years, deep learning is due to image processing unit (Graphic Processing Unit, GPU) and big data
The reason of become more and more popular.Wherein convolutional neural networks (Convolutional Neural Network) are typical
Depth learning technology, its also known as the reason of becoming more and more popular following points:
(1) structure of CNN is end-to-end connection, it is very flexibly, it can set for the characteristic of task
Determine structure；
(2) CNN can realize regularization by basic plug-in unit, and basic plug-in unit includes linear amending unit
(Rectifier Linear Unit, ReLU) and convolutional layer (Convolution, Conv)；
(3) CNN carries out parallel computation by GPU, greatly improves operational efficiency.
CNN is not since with powerful self-learning capability, it needs to adjust ginseng manually, while it can quickly be handled by GPU
Image, therefore it is also a kind of effectively processing image recovery method, such as: SRCNN network carries out processing super-resolution using three layers
Task.Although it is better than conventional method in super-resolution task, when network depth is more than that three layers of performance will decline, so side
Method lacks flexibility.Then, CNN also progress of making a breakthrough property on image denoising, as CNN is used image for the first time by DnCNN
Denoising, and multiple tasks are handled with a model: Gauss denoising, super-resolution and recovery compression image.FFDNet noise
The input as denoising network, this method effectively handle blind noise together for mapping and noise image.IRCNN for the first time will optimization
Algorithm and method of discrimination are combined together, and this method has definite meaning in processing real noise.MLWC by space domain characteristic with
CNN in conjunction with come solve super-resolution and denoising etc. tasks.He Kaiming proposition is recycled+is operated to promote the performance of image denoising.
The above method all obtains certain progress in terms of image denoising, but still has following challenge not yet to solve:
(1) most of method uses BN technology more than, and BN technology relies on the size of batch very much, when batch is big, BN
Contrast in technical performance is good；When batch is smaller, the decline of BN technical performance.Therefore BN technology does not have good robust
Property；
(2) it is referred to above to method cannot be more with a model treatment real noise, Gaussian noise and blind noise etc.
A task；
(3) it is by repeatedly being mentioned using+operation that method part referred to above, which is by deepening the network number of plies, Part Methods,
High denoising performance, this greatly increases the calculating cost of network.
Summary of the invention
Technical problem to be solved by the invention is to provide a kind of image denoisings of convolutional neural networks based on multi-angle of view
Method, this method can effectively mend the deficiency with regard to BN and accelerate network convergence, can handle real noise image, blind noise and height
This noise.Disaster relief, aviation exploration and medical treatment in reality are diagnosed a disease and be of great significance.
The present invention is implemented as follows: the image de-noising method of the convolutional neural networks based on multi-angle of view, includes following step
It is rapid to carry out:
1) after original image being extracted feature with fft algorithm, then image reconstruction is carried out, by reconstructed image and original image piecemeal
Afterwards collectively as the input of denoising network；
2) reconstructed image and original image piecemeal are by denoising network, by the corresponding output of FFT and original image corresponding network
After output is added, the clean image of denoising is exported.
The denoising network is formed by 19 layers, and the 1st layer is made of convolutional layer and ReLU；2-18 layers by convolutional layer,
INBN and GF function composition, the 19th layer is made of convolutional layer；It is by denoising network processes, the corresponding network of obtained FFT is defeated
Input after being added out with the output of original image corresponding network, as the 19th layer of convolutional layer.
The input size of the network is 256 × 1 × 40 × 40, and output size is 256 × 1 × 40 × 40, and convolution is big
Small is 3 × 3, wherein 256 × 1 × 40 × 40 to represent batchsize be 256, output channel 1 is highly 40 with width.
The INBN layer is the channel half an of convolutional layer by IN, the other half by passing through+operation again after BN
It is merged together；The wherein expression of IN such as formula (1):
In formula (1), μ is average value, and σ is standard deviation, and ε is constant, and H is that height W is depth；
Shown in the expression of BN such as formula (2):
(2) represent average value (3) represent variance (4) represent normalization (5) represent data reconstruction.
The GF function is GF (x)=ReLU (x) × Tanh (x), and wherein ReLU is φ (x)=max (0, x), ReLU
Effect be exactly the data of linear transfor be nonlinear data.
Wherein, tanh (x) is
Smooth function is used as objective function to train denoising model, shown in Smooth function such as formula (6):
Image reconstruction: be multiplied and can be obtained with feature with the parameter of model training training after the feature for extracting image
To characteristic pattern.Characteristic pattern and original image be it is distinguishing, characteristic pattern is the extraction to original image key, such as: a secondary facial image, I
Extraction feature, can with human eye, nose, mouth feature etc. reconstruct a sub-picture band indicate whole picture facial image.
Image block: being that entire image is divided into multiple fritters, helps the direct local feature of rapidly extracting image in this way.
Input is to input FFT characteristic image with original image, and characteristic image and original image are carried out piecemeal.
The present invention is accelerated to denoise the convergence of network instead of BN with INBN technology；This method can effectively be mended with regard to BN not
Foot and quickening network convergence, can handle real noise image, blind noise and Gaussian noise.The present invention only uses 20 layers of network
It is denoised, reduces the calculating cost of network.In addition, the present invention is with a kind of new technology GF technology come more preferably linear data
It is transformed to nonlinear data；Denoising model is preferably trained with Smooth function.In addition, it is melted using the feature of multi-angle of view
It closes to enhance network performance.The present invention diagnoses a disease and is of great significance for disaster relief, aviation exploration and medical treatment in reality.
Detailed description of the invention
Fig. 1 is the overall flow figure of network of the invention；
Fig. 2 is the structure chart of denoising network of the invention；
Fig. 3 is the former noise image of the embodiment of the present invention and the image of FFT extraction space domain characteristic；
Fig. 4 is 2 pieces of figures in the former noise block image of the embodiment of the present invention；
Fig. 5 is 2 blocks of images in the FFT image block image of the embodiment of the present invention；
Fig. 6 is 2 blocks of images of clean image of the embodiment of the present invention.
Fig. 7 is the ratio of the clean image of the original image of the embodiment of the present invention, the image of FFT extraction space characteristics and recovery
Compared with figure.
Specific embodiment
The embodiment of the present invention: the image de-noising method of the convolutional neural networks based on multi-angle of view, this implementation is by Smooth
Objective function of the function as training denoising network.By taking Gaussian noise rank 75 as an example.
Include the following steps:
1) original image is subjected to image reconstruction after FFT carries out feature extraction, reconstructed image piecemeal, also original image
Input of the piecemeal as network, as shown in Figure 3.
2) reconstructed image and original image piecemeal are by denoising network, by the corresponding output of FFT and original image corresponding network
After output is added, the clean image of denoising is exported.
The network is formed by 19 layers, and the 1st layer is made of convolutional layer and ReLU；2-18 layers by convolutional layer, INBN
It is formed with GF function, the 19th layer is made of convolutional layer；It is operated with "+" by step 1, obtains Fusion Features as shown in Figure 2 to together
It is connected afterwards with the 19th layer of convolutional layer.
The input size of the network is 256 × 1 × 40 × 40, and output size is 256 × 1 × 40 × 40, and convolution is big
Small is 3 × 3, wherein 256 × 1 × 40 × 40 to represent batchsize be 256, output channel 1 is highly 40 with width.
The INBN layer is the channel half an of convolutional layer by IN, the other half by passing through cat+ again after BN
Operation is merged together；The wherein expression of IN such as formula (1):
In formula (1), μ is average value, and σ is standard deviation, and ε is constant, and H is that height W is depth；
Shown in the expression of BN such as formula (2):
(2) represent average value (3) represent variance (4) represent normalization (5) represent data reconstruction.
The GF function is GF (x)=ReLU (x) × Tanh (x), and wherein ReLU is φ (x)=max (0, x), ReLU
Effect be exactly the data of linear transfor be nonlinear data.
Wherein, tanh (x) is
The present invention uses Smooth function as objective function to train denoising model, and Smooth can be such that image becomes more
Smoothly.
Claims (6)
1. a kind of image de-noising method of the convolutional neural networks based on multi-angle of view, which is characterized in that comprise the following steps progress:
1) after original image being extracted feature with fft algorithm, then image reconstruction is carried out, it will be total after reconstructed image and original image piecemeal
With the input as denoising network；
2) reconstructed image and original image piecemeal are by denoising network, by the corresponding network output of FFT and original image corresponding network
After output is added, the clean image of denoising is exported.
2. the image de-noising method of the convolutional neural networks according to claim 1 based on multi-angle of view, it is characterised in that: institute
The denoising network stated is formed by 19 layers, and the 1st layer is made of convolutional layer and ReLU；2-18 layers by convolutional layer, INBN and GF letter
Array is at the 19th layer is made of convolutional layer；By denoising network processes, by the corresponding network output of obtained FFT and original image
After corresponding network output is added, the input as the 19th layer of convolutional layer.
3. the image de-noising method of the convolutional neural networks according to claim 1 based on multi-angle of view, it is characterised in that: institute
The input size for the network stated is 256 × 1 × 40 × 40, and output size is 256 × 1 × 40 × 40, and convolution size is 3 × 3,
In 256 × 1 × 40 × 40 to represent batchsize be 256, output channel 1 is highly 40 with width.
4. the image de-noising method of the convolutional neural networks according to claim 2 based on multi-angle of view, it is characterised in that: institute
The INBN layer stated is the channel half an of convolutional layer by IN, the other half is merged together by passing through again after BN+operating；
The wherein expression of IN such as formula (1):
In formula (1), μ is average value, and σ is standard deviation, and ε is constant, and H is that height W is depth；
Shown in the expression of BN such as formula (2):
(2) represent average value (3) represent variance (4) represent normalization (5) represent data reconstruction.
5. the image de-noising method of the convolutional neural networks according to claim 2 based on multi-angle of view, it is characterised in that: institute
The GF function stated is GF (x)=ReLU (x) × Tanh (x), and wherein ReLU is φ (x)=max (0, x), and the effect of ReLU is exactly
It is nonlinear data the data of linear transfor.Wherein, tanh (x) is
6. the image de-noising method of the convolutional neural networks according to claim 2 based on multi-angle of view, it is characterised in that: make
Smooth function is used as objective function to train denoising model, shown in Smooth function such as formula (6):
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201910632475.5A CN110503608B (en) | 2019-07-13 | 2019-07-13 | Image denoising method based on multi-view convolutional neural network |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201910632475.5A CN110503608B (en) | 2019-07-13 | 2019-07-13 | Image denoising method based on multi-view convolutional neural network |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN110503608A true CN110503608A (en) | 2019-11-26 |
| CN110503608B CN110503608B (en) | 2023-08-08 |
Family
ID=68585482
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201910632475.5A Active CN110503608B (en) | 2019-07-13 | 2019-07-13 | Image denoising method based on multi-view convolutional neural network |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN110503608B (en) |
Cited By (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN111045084A (en) * | 2020-01-06 | 2020-04-21 | 中国石油化工股份有限公司 | Multi-wave self-adaptive subtraction method based on prediction feature extraction |
| CN112634159A (en) * | 2020-12-23 | 2021-04-09 | 中国海洋大学 | Hyperspectral image denoising method based on blind noise estimation |
| CN112883759A (en) * | 2019-11-29 | 2021-06-01 | 杭州海康威视数字技术股份有限公司 | Method for detecting image noise of biological characteristic part |
Citations (5)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US6207957B1 (en) * | 1998-09-18 | 2001-03-27 | The Regents Of The University Of California | System to quantify gamma-ray radial energy deposition in semiconductor detectors |
| CN108460726A (en) * | 2018-03-26 | 2018-08-28 | 厦门大学 | A kind of magnetic resonance image super-resolution reconstruction method based on enhancing recurrence residual error network |
| CN108986047A (en) * | 2018-07-13 | 2018-12-11 | 中国科学技术大学 | Image denoising method |
| CN109410127A (en) * | 2018-09-17 | 2019-03-01 | 西安电子科技大学 | A kind of image de-noising method based on deep learning and multi-scale image enhancing |
| CN109658344A (en) * | 2018-11-12 | 2019-04-19 | 哈尔滨工业大学（深圳） | Image de-noising method, device, equipment and storage medium based on deep learning |
-
2019
- 2019-07-13 CN CN201910632475.5A patent/CN110503608B/en active Active
Patent Citations (5)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US6207957B1 (en) * | 1998-09-18 | 2001-03-27 | The Regents Of The University Of California | System to quantify gamma-ray radial energy deposition in semiconductor detectors |
| CN108460726A (en) * | 2018-03-26 | 2018-08-28 | 厦门大学 | A kind of magnetic resonance image super-resolution reconstruction method based on enhancing recurrence residual error network |
| CN108986047A (en) * | 2018-07-13 | 2018-12-11 | 中国科学技术大学 | Image denoising method |
| CN109410127A (en) * | 2018-09-17 | 2019-03-01 | 西安电子科技大学 | A kind of image de-noising method based on deep learning and multi-scale image enhancing |
| CN109658344A (en) * | 2018-11-12 | 2019-04-19 | 哈尔滨工业大学（深圳） | Image de-noising method, device, equipment and storage medium based on deep learning |
Non-Patent Citations (7)
| Title |
|---|
| HAROLD C. BURGER 等: ""Image denoising: Can plain neural networks compete with BM3D"", 《HTTPS://IEEEXPLORE.IEEE.ORG/STAMP/STAMP.JSP?TP=&ARNUMBER=6247952》 * |
| KUANG GONG: ""PET Image Denoising Using a Deep Neural Network Through Fine Tuning"", 《HTTPS://IEEEXPLORE.IEEE.ORG/STAMP/STAMP.JSP?TP=&ARNUMBER=8502864》 * |
| 孙利雷等: "基于深度学习的乳腺X射线影像分类方法研究", 《计算机工程与应用》 * |
| 新智元: ""【实战】利用卷积自编码器实现图片降噪（代码开源）"", 《HTTPS://WWW.SOHU.COM/A/157810470_473283》 * |
| 胡拓: ""结合图像降噪处理的低剂量CT重建"", 《中国优秀硕士学位论文全文数据库 医药卫生科技辑》 * |
| 马国喻: "一种求解差分方程的好方法――生成函数法", 《北京化工大学学报(自然科学版)》 * |
| 马红强等: "基于改进栈式稀疏去噪自编码器的自适应图像去噪", 《光学学报》 * |
Cited By (5)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN112883759A (en) * | 2019-11-29 | 2021-06-01 | 杭州海康威视数字技术股份有限公司 | Method for detecting image noise of biological characteristic part |
| CN112883759B (en) * | 2019-11-29 | 2023-09-26 | 杭州海康威视数字技术股份有限公司 | Method for detecting image noise of biological feature part |
| CN111045084A (en) * | 2020-01-06 | 2020-04-21 | 中国石油化工股份有限公司 | Multi-wave self-adaptive subtraction method based on prediction feature extraction |
| CN112634159A (en) * | 2020-12-23 | 2021-04-09 | 中国海洋大学 | Hyperspectral image denoising method based on blind noise estimation |
| CN112634159B (en) * | 2020-12-23 | 2022-07-26 | 中国海洋大学 | Hyperspectral image denoising method based on blind noise estimation |
Also Published As
| Publication number | Publication date |
|---|---|
| CN110503608B (en) | 2023-08-08 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| WO2023092813A1 (en) | Swin-transformer image denoising method and system based on channel attention | |
| Tian et al. | Deep learning on image denoising: An overview | |
| Remez et al. | Class-aware fully convolutional Gaussian and Poisson denoising | |
| CN108520503B (en) | Face defect image restoration method based on self-encoder and generation countermeasure network | |
| CN110782399B (en) | Image deblurring method based on multitasking CNN | |
| WO2020087607A1 (en) | Bi-skip-net-based image deblurring method | |
| CN111127308B (en) | Mirror image feature rearrangement restoration method for single sample face recognition under partial shielding | |
| Liu et al. | Knowledge-driven deep unrolling for robust image layer separation | |
| CN108765338A (en) | Spatial target images restored method based on convolution own coding convolutional neural networks | |
| CN111915530A (en) | End-to-end-based haze concentration self-adaptive neural network image defogging method | |
| Huang et al. | Deep hyperspectral image fusion network with iterative spatio-spectral regularization | |
| CN111161360B (en) | Image defogging method of end-to-end network based on Retinex theory | |
| CN110232653A (en) | The quick light-duty intensive residual error network of super-resolution rebuilding | |
| CN111161178A (en) | Single low-light image enhancement method based on generation type countermeasure network | |
| CN110503608A (en) | The image de-noising method of convolutional neural networks based on multi-angle of view | |
| CN111861894A (en) | Image motion blur removing method based on generating type countermeasure network | |
| CN112270654A (en) | Image denoising method based on multi-channel GAN | |
| CN112991199B (en) | Image high-low frequency decomposition noise removal method based on residual dense network | |
| CN113450290A (en) | Low-illumination image enhancement method and system based on image inpainting technology | |
| CN107194380A (en) | The depth convolutional network and learning method of a kind of complex scene human face identification | |
| CN114202460A (en) | Super-resolution high-definition reconstruction method, system and equipment facing different damage images | |
| CN104200194B (en) | A kind of face identification method | |
| CN116703750A (en) | Image defogging method and system based on edge attention and multi-order differential loss | |
| Liang et al. | Deep convolution neural networks for automatic eyeglasses removal | |
| CN113378980B (en) | Mask face shielding recovery method based on self-adaptive context attention mechanism |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| PB01 | Publication | ||
| PB01 | Publication | ||
| SE01 | Entry into force of request for substantive examination | ||
| SE01 | Entry into force of request for substantive examination | ||
| GR01 | Patent grant | ||
| GR01 | Patent grant |