CN112669214A - Fuzzy image super-resolution reconstruction method based on alternative direction multiplier algorithm - Google Patents
Fuzzy image super-resolution reconstruction method based on alternative direction multiplier algorithm Download PDFInfo
- Publication number
- CN112669214A CN112669214A CN202110002643.XA CN202110002643A CN112669214A CN 112669214 A CN112669214 A CN 112669214A CN 202110002643 A CN202110002643 A CN 202110002643A CN 112669214 A CN112669214 A CN 112669214A
- Authority
- CN
- China
- Prior art keywords
- image
- resolution
- features
- super
- resolution image
- 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 51
- 238000012545 processing Methods 0.000 claims abstract description 7
- 238000005070 sampling Methods 0.000 claims description 16
- 238000006731 degradation reaction Methods 0.000 claims description 12
- 238000012549 training Methods 0.000 claims description 9
- 230000015556 catabolic process Effects 0.000 claims description 8
- 230000006870 function Effects 0.000 claims description 8
- 238000005457 optimization Methods 0.000 claims description 8
- 239000000654 additive Substances 0.000 claims description 7
- 230000000996 additive effect Effects 0.000 claims description 7
- 238000013527 convolutional neural network Methods 0.000 claims description 4
- 238000004088 simulation Methods 0.000 claims description 4
- 238000000605 extraction Methods 0.000 claims description 3
- 230000002194 synthesizing effect Effects 0.000 claims description 3
- ORILYTVJVMAKLC-UHFFFAOYSA-N Adamantane Natural products C1C(C2)CC3CC1CC2C3 ORILYTVJVMAKLC-UHFFFAOYSA-N 0.000 claims description 2
- 230000003190 augmentative effect Effects 0.000 claims description 2
- 239000011159 matrix material Substances 0.000 claims description 2
- 230000000593 degrading effect Effects 0.000 claims 1
- 230000000694 effects Effects 0.000 abstract description 4
- 238000010586 diagram Methods 0.000 description 7
- 101100365548 Caenorhabditis elegans set-14 gene Proteins 0.000 description 2
- 230000005540 biological transmission Effects 0.000 description 2
- 238000013136 deep learning model Methods 0.000 description 2
- 238000005516 engineering process Methods 0.000 description 2
- 238000003384 imaging method Methods 0.000 description 2
- 230000000750 progressive effect Effects 0.000 description 2
- 238000012360 testing method Methods 0.000 description 2
- 238000013459 approach Methods 0.000 description 1
- 230000009286 beneficial effect Effects 0.000 description 1
- 238000002790 cross-validation Methods 0.000 description 1
- 238000013135 deep learning Methods 0.000 description 1
- 230000007547 defect Effects 0.000 description 1
- 238000004519 manufacturing process Methods 0.000 description 1
- 238000012544 monitoring process Methods 0.000 description 1
- 230000003287 optical effect Effects 0.000 description 1
- 230000008707 rearrangement Effects 0.000 description 1
- 230000021317 sensory perception Effects 0.000 description 1
Images
Landscapes
- Image Processing (AREA)
Abstract
The invention belongs to the technical field of image processing, and discloses a fuzzy image super-resolution reconstruction method based on an alternating direction multiplier algorithm. According to the invention, a forward observation model is constructed, so that the degraded low-resolution image can be closer to the degraded low-resolution image in a real scene, a blurred image super-resolution reconstruction model based on an alternating direction multiplier algorithm is built, an image deblurring process is separated from a high-quality characteristic image obtaining process, and a better blurred image super-resolution reconstruction effect is obtained. Compared with a super-resolution reconstruction method by directly performing bicubic interpolation and downsampling, the method can better process low-resolution image reconstruction in a real scene.
Description
Technical Field
The invention belongs to the technical field of image processing, and relates to a fuzzy image super-resolution reconstruction method of an alternative direction multiplier algorithm.
Background
Super resolution techniques are techniques that convert a given low resolution image or sequence of images into a corresponding high resolution image or sequence of images. The low-resolution image has lower definition, less details, and the high-resolution image processed by the super-resolution algorithm has clear details and abundant textures, and can be consistent with the low-resolution image in the whole structure. Therefore, the sensory perception of people on the image is greatly improved under the condition of not influencing the whole content of the image. The image super-resolution reconstruction technology has important value and is widely applied to the fields of security monitoring, medical images, movie and television production, satellite imaging and the like.
However, most super-resolution methods obtain a corresponding low-resolution image by performing bicubic interpolation down-sampling on a high-resolution image. The image acquisition process typically experiences a natural loss of spatial resolution. The natural loss is mainly caused by the problems of optical distortion, object movement, incorrect focusing, limitation of shutter speed, noise occurring in the transmission process, noise in the sensor, poor density of the sensor and the like, so that the obtained image is degraded due to noise, blurring and undersampling effects, and therefore the deep learning model trained based on the bicubic interpolation method cannot well process the problem of true image super-resolution reconstruction.
Disclosure of Invention
The invention provides a fuzzy image super-resolution reconstruction method based on an alternative direction multiplier method, aiming at the defects of the technology, and solves the problem that a deep learning model processes true image super-resolution reconstruction.
The specific technical scheme of the invention is as follows:
a blurred image super-resolution reconstruction method based on an alternating direction multiplier algorithm comprises the following steps:
step 1: constructing a simulation data set of a pair of the low-resolution image and the original high-resolution image;
step 2: step 2: and (3) constructing an image super-resolution reconstruction model based on an alternating direction multiplier algorithm, wherein the network model is divided into two parts, namely a step-by-step refinement network and a deblurring network, and finally, a sub-pixel convolution module is used for realizing the process of reconstructing the characteristic image into a high-quality image.
Step 2.1: building a step-by-step refinement network to obtain rough features and detailed features of the image; the detail features are regarded as high-frequency features obtained by continuously performing convolution operation extraction, the features are used for synthesizing a feature map of a high-quality image, and the rough features are connected along the channel direction and enter a deblurring network; according to input features FinputThis process is represented as:
wherein ,FcoarseShowing a coarse feature, FrefinedDetail features are shown, C represents convolution operation, Split represents channel splitting operation, and Concat represents channel connection operation;
step 2.2: and constructing a deblurring network aiming at the rough features, and selecting an alternating direction multiplier method as an optimization method aiming at the rough feature deblurring, wherein the alternating direction multiplier method is one of the most common convex optimization methods in image processing. It transforms the optimization of the original nonlinear problem into a saddle point problem that finds the classical lagrangian function.
Blurred images containing rough features are knownFuzzy kernelDown-sampling the factor s and the noise level σ to solve a deblurred image containing coarse featuresBased on the maximum posterior probability theory, the method is converted into the following equation:
step 2.2.1: firstly, introducing an auxiliary variable z to obtain an equivalent constraint optimization formula:
step 2.2.2: and then converting the optimization problem into an unconstrained problem, and obtaining the following equation by using an augmented Lagrange function:
step 2.2.3: and finally, converting the saddle point problem for solving the Lagrangian function into the following three subproblems:
step 2.2.4: assuming convolution under circular boundary conditions, equation (6) is solved using Fast Fourier Transform (FFT):
wherein F (-) represents a fast Fourier transform, F-1(. cndot.) represents the inverse transform,is the complex conjugate of F (-);
for a blurred image with coarse features, equation (10) shows that under the prior assumption, the noise removal level isContains a blurred image with rough features and minimizesAndthe residual error between; a feature map without roughness features is obtained.
Step 2.3: building a sub-pixel convolution module, and reconstructing the feature map without the rough features obtained in the step 2.2 into a high-quality image;
and step 3: and training the whole image super-resolution reconstruction network until the network converges.
Further, the step 1: constructing a simulation data set of the pair of the low-resolution image and the original high-resolution image: in combination with the degradation process of image acquisition, the degradation process of the high resolution image is expressed by the following formula:
wherein ,which represents a 2D convolution of the image,a blur kernel is represented by the number of pixels,representing additive noise and ↓sWhich represents the down-sampling of the sample,for the purpose of a high-resolution image,is a low resolution image corresponding to the high resolution image.
Step 1.1: only the isotropic gaussian blur kernel is considered in the blur processing, and the blur kernel width ranges are set to [0.2,2], [0.2,3] and [0.2,4] at 2,3, 4-fold super-resolution. A low resolution image with blur degradation is constructed.
Step 1.2: the degradation model uses a bicubic interpolation approach to downsample low resolution images with blur degradation. By constructing a sampling weight matrix, 16 pixel points adjacent to each pixel are taken as sampling points, and the sampling step size is 2,3 and 4 times of the down-sampling multiplying power.
Step 1.3: noise may occur during imaging due to noise occurring during transmission, noise within the sensor. This process is modeled in the degradation model by additive white gaussian noise. The image was degraded using additive white gaussian noise with a mean of 0 and variance range of 0, 15.
Step 1.4: the low resolution image and the corresponding high resolution image are randomly cropped. The low-resolution image block size is 48 × 48 × 3, and the corresponding high-resolution image block sizes are 96 × 96 × 3, 144 × 144 × 3, and 192 × 192 × 3 at 2,3, and 4 times super-resolution magnification, respectively.
Step 1.5: and carrying out data enhancement processing on the image blocks, and expanding a training data set through rotation and turning operations.
The method has the beneficial effect that the fuzzy image super-resolution reconstruction method based on the alternative direction multiplier algorithm has better performance on PSNR indexes. Compared with a super-resolution reconstruction method by directly performing bicubic interpolation and downsampling, the method can better process low-resolution image reconstruction in a real scene.
Drawings
Fig. 1 is a network architecture of the present invention.
Fig. 2 is a diagram of a step-by-step refinement network architecture of the present invention.
FIG. 3 is a deblurring network structure for coarse features of the present invention.
Fig. 4 is a super-resolution reconstruction effect diagram of the present invention.
Detailed Description
The following detailed description of embodiments of the invention refers to the accompanying drawings.
In the method of the embodiment, the software environment is the Ubuntu16.04 system, and the deep learning framework is PyTorch.
Step 1: urban100 was used as training data, which included 90 training sets and 10 cross-validation sets. And constructing a low-resolution image and a high-resolution image pair through the established degradation model for end-to-end training. At 2,3,4 times super-resolution, the blur kernel width ranges are set to [0.2,2], [0.2,3] and [0.2,4 ]; the blur kernel size is fixed to 15 × 15 pixels; in the down-sampling mode, a bicubic interpolation mode is used for down-sampling; for the degenerate noise, additive white gaussian noise with mean 0 and variance range 0,15 is used.
Step 2: and constructing an image super-resolution model based on the convolutional neural network. The network architecture is shown in fig. 1.
Step 2.1: in the progressive refinement network, a 3 × 3 convolutional layer is adopted to extract input features, the number of picture channels is increased, and the convolutional layer result is activated by using a LeakyReLU function. Then, splitting the channel on the basis of convolution, wherein the split characteristics are divided into two parts: and (4) carrying out rough feature and detail feature, and carrying out convolution operation on the detail feature to extract the feature, and reserving the rough feature. And finally, the detail features are regarded as high-frequency features obtained by continuously performing convolution operation extraction, the features are used for synthesizing a feature map of a high-quality image, and the rough features are connected along the channel direction and enter a deblurring network. The network structure is progressively refined as shown in fig. 2. The structure and detailed parameters of each layer of the stepwise refinement network are shown in the following table 1:
TABLE 1 progressive refinement of network layer structure and detailed parameters
Step 2.2: in the deblurring network aiming at the rough features, the total iteration number is set to be 6. Firstly, the initial value is pre-calculated by utilizing fast Fourier transformIn code implementation, the fast fourier transform and its inverse transform are used with torch. And secondly, combining the deblurring process and the residual error updating process by adopting a residual error network. In each iteration, the initial value of λ is set to 0.05, the initial value of the parameter μ is set to 0.01, and μ is increased as the number of iterations increases until convergence. The deblurring network structure for coarse features is shown in fig. 3.
Step 2.3: and (4) building a sub-pixel convolution module, and reconstructing the feature map without the rough features obtained in the step 2.2 into a high-quality image. The sub-pixel convolution module consists of two 3 × 3 convolutional layers, the first of which converts the dimensions of the feature map from cxhxw to(s)2C) xHxW, wherein C represents the number of characteristic diagram channels, H represents the height of the characteristic diagram, and W represents the width of the characteristic diagram; the dimension of the characteristic diagram is changed from(s) to(s) by the up-sampling layer in a pixel rearrangement mode2C) The xHxW is converted into CxsHxsW, wherein s represents super-resolution magnification, the second convolution layer converts the characteristic diagram into an RGB color image, and a final super-resolution result is output.
And step 3: in the training process, setting the learning rate to be initialized to 0.0001, the learning rate attenuation step length to be 200 and the attenuation rate to be 0.5; setting a loss function as MSE loss; and selecting an Adam algorithm to optimize network parameters of each layer. The reconstruction effect is shown in fig. 4.
And 4, step 4: in the testing process, because the image super-resolution model based on the convolutional neural network adopts a residual error module, SR-ResNet is selected as a comparison model; test data used Set5 and Set 14; the blur kernel width is set to 2.6; PSNR index was tested on the super-resolution x 4 task. The PSNR values were 24.615, 23.046 when the SR-ResNet model processed Set5 and Set14 data sets with a blur kernel width of 2.6, respectively, and 29.425, 26.340 when the convolutional neural network-based image super-resolution model processed Set5 and Set14 data sets with a blur kernel width of 2.6, respectively.
In conclusion, the blurred image super-resolution reconstruction method based on the alternating direction multiplier algorithm has better performance on the PSNR index. Compared with a super-resolution reconstruction method by directly performing bicubic interpolation and downsampling, the method can better process low-resolution image reconstruction in a real scene.
Claims (3)
1. A blurred image super-resolution reconstruction method based on an alternating direction multiplier algorithm is characterized by comprising the following steps,
step 1: constructing a simulation data set of a pair of the low-resolution image and the original high-resolution image;
step 2: constructing an image super-resolution model based on a convolutional neural network;
step 2.1: building a step-by-step refinement network to obtain rough features and detailed features of the image; the detail features are regarded as high-frequency features obtained by continuously performing convolution operation extraction, the features are used for synthesizing a feature map of a high-quality image, and the rough features are connected along the channel direction and enter a deblurring network; according to input features FinputThis process is represented as:
wherein ,FcoarseShowing a coarse feature, FrefinedDetail features are shown, C represents convolution operation, Split represents channel splitting operation, and Concat represents channel connection operation;
step 2.2: building a deblurring network aiming at the rough characteristics, and selecting an alternative direction multiplier method as an optimization method aiming at the rough characteristic deblurring: blurred images containing rough features are knownFuzzy kernelDown-sampling the factor s and the noise level σ to solve a deblurred image containing coarse featuresBased on the maximum posterior probability theory, the method is converted into the following equation:
step 2.2.1: firstly, introducing an auxiliary variable z to obtain an equivalent constraint optimization formula:
step 2.2.2: and then converting the optimization problem into an unconstrained problem, and obtaining the following equation by using an augmented Lagrange function:
step 2.2.3: and finally, converting the saddle point problem for solving the Lagrangian function into the following three subproblems:
step 2.2.4: assuming convolution under circular boundary conditions, equation (6) is solved using fast fourier transform:
wherein F (-) represents a fast Fourier transform, F-1(. cndot.) represents the inverse transform,is the complex conjugate of F (-);
for a blurred image with coarse features, equation (10) shows that under the prior assumption, the noise removal level isContains a blurred image with rough features and minimizesAndthe residual error between; obtaining a feature map without roughness features;
step 2.3: building a sub-pixel convolution module, and reconstructing the feature map without the rough features obtained in the step 2.2 into a high-quality image;
and step 3: and training the whole image super-resolution reconstruction network until the network converges.
2. The method for super-resolution reconstruction of blurred images based on the alternative direction multiplier method as claimed in claim 1, wherein the simulation data set of the pair of low-resolution image and original high-resolution image is constructed by the following steps:
in combination with the degradation process of image acquisition, the degradation process of the LR image is expressed by the following formula:
wherein ,which represents a 2D convolution of the image,a blur kernel is represented by the number of pixels,representing additive noise and ↓sWhich represents the down-sampling of the sample,for high resolutionThe image is a picture of a person to be imaged,a low resolution image corresponding to the high resolution image;
step 1.1: only an isotropic Gaussian blur kernel is considered in the blur processing, under the super-resolution magnification of 2 times, 3 times and 4 times, the width range of the blur kernel is set to be [0.2,2], [0.2,3] and [0.2,4], and a low-resolution image with blur degradation is constructed;
step 1.2: a bicubic interpolation mode used by the degradation model is used for carrying out downsampling on the high-resolution image; by constructing a sampling weight matrix, 16 pixel points adjacent to each pixel are taken as sampling points, and the sampling step size is 2,3 and 4 times of the down-sampling multiplying power;
step 1.3: modeling the process through additive white Gaussian noise in a degradation model, and degrading the image by using the additive white Gaussian noise with the mean value of 0 and the variance range of [0,15 ];
step 1.4: randomly cropping the low resolution image and the corresponding high resolution image; the size of the low-resolution image block is 48 multiplied by 3, and under 2,3 and 4 times super-resolution magnification, the sizes of the corresponding high-resolution image blocks are 96 multiplied by 3, 144 multiplied by 3 and 192 multiplied by 3 respectively;
step 1.5: and carrying out data enhancement processing on the image blocks, and expanding a training data set through rotation and turning operations.
3. The method for super-resolution reconstruction of blurred images according to claim 1, wherein the step 3 is performed as follows:
step 3.1: in the training process, setting the learning rate to be initialized to 0.0001, the learning rate attenuation step length to be 200 and the attenuation rate to be 0.5;
step 3.2: setting a loss function as MSE loss;
step 3.3: and selecting an Adam algorithm to optimize network parameters of each layer.
Priority Applications (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN202110002643.XA CN112669214B (en) | 2021-01-04 | 2021-01-04 | Fuzzy image super-resolution reconstruction method based on alternating direction multiplier algorithm |
Applications Claiming Priority (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN202110002643.XA CN112669214B (en) | 2021-01-04 | 2021-01-04 | Fuzzy image super-resolution reconstruction method based on alternating direction multiplier algorithm |
Publications (2)
Publication Number | Publication Date |
---|---|
CN112669214A true CN112669214A (en) | 2021-04-16 |
CN112669214B CN112669214B (en) | 2023-09-19 |
Family
ID=75412637
Family Applications (1)
Application Number | Title | Priority Date | Filing Date |
---|---|---|---|
CN202110002643.XA Active CN112669214B (en) | 2021-01-04 | 2021-01-04 | Fuzzy image super-resolution reconstruction method based on alternating direction multiplier algorithm |
Country Status (1)
Country | Link |
---|---|
CN (1) | CN112669214B (en) |
Cited By (5)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN113469884A (en) * | 2021-07-15 | 2021-10-01 | 长视科技股份有限公司 | Video super-resolution method, system, equipment and storage medium based on data simulation |
CN113538245A (en) * | 2021-08-03 | 2021-10-22 | 四川启睿克科技有限公司 | Degradation model-based super-resolution image reconstruction method and system |
CN116544146A (en) * | 2023-05-22 | 2023-08-04 | 浙江固驰电子有限公司 | Vacuum sintering equipment and method for power semiconductor device |
WO2023155305A1 (en) * | 2022-02-16 | 2023-08-24 | 平安科技(深圳)有限公司 | Image reconstruction method and apparatus, and electronic device and storage medium |
WO2024120224A1 (en) * | 2022-12-05 | 2024-06-13 | 杭州堃博生物科技有限公司 | Image processing method, apparatus, electronic device and storage medium |
Citations (6)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
US20110115934A1 (en) * | 2009-11-19 | 2011-05-19 | Sen Wang | Increasing image resolution using combined differential image |
US20140354886A1 (en) * | 2013-05-29 | 2014-12-04 | Yeda Research & Development Co. Ltd. | Device, system, and method of blind deblurring and blind super-resolution utilizing internal patch recurrence |
CN109767386A (en) * | 2018-12-22 | 2019-05-17 | 昆明理工大学 | A kind of rapid image super resolution ratio reconstruction method based on deep learning |
CN110705699A (en) * | 2019-10-18 | 2020-01-17 | 厦门美图之家科技有限公司 | Super-resolution reconstruction method and device, electronic equipment and readable storage medium |
US20200034948A1 (en) * | 2018-07-27 | 2020-01-30 | Washington University | Ml-based methods for pseudo-ct and hr mr image estimation |
AU2020100462A4 (en) * | 2020-03-26 | 2020-04-30 | Hu, Xiaoyan MISS | Edge-preserving image super-resolution via low rank and total variation model |
-
2021
- 2021-01-04 CN CN202110002643.XA patent/CN112669214B/en active Active
Patent Citations (6)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
US20110115934A1 (en) * | 2009-11-19 | 2011-05-19 | Sen Wang | Increasing image resolution using combined differential image |
US20140354886A1 (en) * | 2013-05-29 | 2014-12-04 | Yeda Research & Development Co. Ltd. | Device, system, and method of blind deblurring and blind super-resolution utilizing internal patch recurrence |
US20200034948A1 (en) * | 2018-07-27 | 2020-01-30 | Washington University | Ml-based methods for pseudo-ct and hr mr image estimation |
CN109767386A (en) * | 2018-12-22 | 2019-05-17 | 昆明理工大学 | A kind of rapid image super resolution ratio reconstruction method based on deep learning |
CN110705699A (en) * | 2019-10-18 | 2020-01-17 | 厦门美图之家科技有限公司 | Super-resolution reconstruction method and device, electronic equipment and readable storage medium |
AU2020100462A4 (en) * | 2020-03-26 | 2020-04-30 | Hu, Xiaoyan MISS | Edge-preserving image super-resolution via low rank and total variation model |
Non-Patent Citations (3)
Title |
---|
HAESOL PARK 等: "Joint Estimation of Camera Pose, Depth, Deblurring, and Super-Resolution From a Blurred Image Sequence", 《PROCEEDINGS OF THE IEEE INTERNATIONAL CONFERENCE ON COMPUTER VISION (ICCV)》, pages 4613 - 4621 * |
张伟: "基于稀疏表示的图像超分辨率重构算法研究", 《中国优秀硕士学位论文全文数据库 信息科技辑》, no. 10, pages 138 - 550 * |
李岚 等: "基于改进残差亚像素卷积神经网络的超分辨率图像重建方法研究", 《长春师范大学学报》, vol. 39, no. 8, pages 23 - 29 * |
Cited By (6)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN113469884A (en) * | 2021-07-15 | 2021-10-01 | 长视科技股份有限公司 | Video super-resolution method, system, equipment and storage medium based on data simulation |
CN113538245A (en) * | 2021-08-03 | 2021-10-22 | 四川启睿克科技有限公司 | Degradation model-based super-resolution image reconstruction method and system |
WO2023155305A1 (en) * | 2022-02-16 | 2023-08-24 | 平安科技(深圳)有限公司 | Image reconstruction method and apparatus, and electronic device and storage medium |
WO2024120224A1 (en) * | 2022-12-05 | 2024-06-13 | 杭州堃博生物科技有限公司 | Image processing method, apparatus, electronic device and storage medium |
CN116544146A (en) * | 2023-05-22 | 2023-08-04 | 浙江固驰电子有限公司 | Vacuum sintering equipment and method for power semiconductor device |
CN116544146B (en) * | 2023-05-22 | 2024-04-09 | 浙江固驰电子有限公司 | Vacuum sintering equipment and method for power semiconductor device |
Also Published As
Publication number | Publication date |
---|---|
CN112669214B (en) | 2023-09-19 |
Similar Documents
Publication | Publication Date | Title |
---|---|---|
CN112669214B (en) | Fuzzy image super-resolution reconstruction method based on alternating direction multiplier algorithm | |
CN106683067B (en) | Deep learning super-resolution reconstruction method based on residual sub-images | |
CN106709875B (en) | Compressed low-resolution image restoration method based on joint depth network | |
CN111028177B (en) | Edge-based deep learning image motion blur removing method | |
CN111709895A (en) | Image blind deblurring method and system based on attention mechanism | |
CN111598778B (en) | Super-resolution reconstruction method for insulator image | |
CN111784582B (en) | DEC-SE-based low-illumination image super-resolution reconstruction method | |
CN110796622B (en) | Image bit enhancement method based on multi-layer characteristics of series neural network | |
CN108492270B (en) | Super-resolution method based on fuzzy kernel estimation and variational reconstruction | |
CN103136728B (en) | Based on the image super-resolution method of dictionary learning and non local total variance | |
CN116051428B (en) | Deep learning-based combined denoising and superdivision low-illumination image enhancement method | |
CN113724134B (en) | Aerial image blind super-resolution reconstruction method based on residual distillation network | |
CN112001843A (en) | Infrared image super-resolution reconstruction method based on deep learning | |
CN116029902A (en) | Knowledge distillation-based unsupervised real world image super-resolution method | |
Chen et al. | Image denoising via deep network based on edge enhancement | |
CN117274059A (en) | Low-resolution image reconstruction method and system based on image coding-decoding | |
CN115345791A (en) | Infrared image deblurring algorithm based on attention mechanism residual error network model | |
CN111754399A (en) | Image super-resolution method for keeping geometric structure based on gradient | |
CN112435165B (en) | Two-stage video super-resolution reconstruction method based on generation countermeasure network | |
CN111986102B (en) | Digital pathological image deblurring method | |
CN113962882A (en) | JPEG image compression artifact eliminating method based on controllable pyramid wavelet network | |
CN113240581A (en) | Real world image super-resolution method for unknown fuzzy kernel | |
CN111899166A (en) | Medical hyperspectral microscopic image super-resolution reconstruction method based on deep learning | |
CN116188265A (en) | Space variable kernel perception blind super-division reconstruction method based on real degradation | |
CN112348745B (en) | Video super-resolution reconstruction method based on residual convolutional 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 |