CN114998666B - Degradation kernel extraction method for image blind super-resolution enhancement network - Google Patents
Degradation kernel extraction method for image blind super-resolution enhancement network Download PDFInfo
- Publication number
- CN114998666B CN114998666B CN202210467732.6A CN202210467732A CN114998666B CN 114998666 B CN114998666 B CN 114998666B CN 202210467732 A CN202210467732 A CN 202210467732A CN 114998666 B CN114998666 B CN 114998666B
- Authority
- CN
- China
- Prior art keywords
- degradation
- image
- kernel
- resolution
- rlr
- 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.)
- Active
Links
- 238000006731 degradation reaction Methods 0.000 title claims abstract description 130
- 230000015556 catabolic process Effects 0.000 title claims abstract description 128
- 238000000605 extraction Methods 0.000 title claims abstract description 20
- 238000012512 characterization method Methods 0.000 claims abstract description 21
- 238000012549 training Methods 0.000 claims abstract description 21
- 238000012545 processing Methods 0.000 claims abstract description 15
- 238000000034 method Methods 0.000 claims description 22
- 230000006870 function Effects 0.000 claims description 20
- ORILYTVJVMAKLC-UHFFFAOYSA-N Adamantane Natural products C1C(C2)CC3CC1CC2C3 ORILYTVJVMAKLC-UHFFFAOYSA-N 0.000 claims description 3
- 238000013528 artificial neural network Methods 0.000 claims description 3
- 230000009977 dual effect Effects 0.000 claims description 3
- 238000012952 Resampling Methods 0.000 claims description 2
- 238000005070 sampling Methods 0.000 claims description 2
- 230000000694 effects Effects 0.000 abstract description 12
- 238000010586 diagram Methods 0.000 description 5
- 238000013507 mapping Methods 0.000 description 4
- 238000013527 convolutional neural network Methods 0.000 description 2
- 238000012986 modification Methods 0.000 description 2
- 230000004048 modification Effects 0.000 description 2
- 238000011176 pooling Methods 0.000 description 2
- 230000003044 adaptive effect Effects 0.000 description 1
- 230000009286 beneficial effect Effects 0.000 description 1
- 238000011161 development Methods 0.000 description 1
- 238000003384 imaging method Methods 0.000 description 1
- 238000011160 research Methods 0.000 description 1
- 230000000007 visual effect Effects 0.000 description 1
Classifications
-
- 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
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T3/00—Geometric image transformations in the plane of the image
- G06T3/40—Scaling of whole images or parts thereof, e.g. expanding or contracting
- G06T3/4046—Scaling of whole images or parts thereof, e.g. expanding or contracting using neural networks
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T3/00—Geometric image transformations in the plane of the image
- G06T3/40—Scaling of whole images or parts thereof, e.g. expanding or contracting
- G06T3/4053—Scaling of whole images or parts thereof, e.g. expanding or contracting based on super-resolution, i.e. the output image resolution being higher than the sensor resolution
Landscapes
- Engineering & Computer Science (AREA)
- Physics & Mathematics (AREA)
- Theoretical Computer Science (AREA)
- General Physics & Mathematics (AREA)
- Artificial Intelligence (AREA)
- Evolutionary Computation (AREA)
- Computational Linguistics (AREA)
- Molecular Biology (AREA)
- Biomedical Technology (AREA)
- Biophysics (AREA)
- Health & Medical Sciences (AREA)
- Data Mining & Analysis (AREA)
- General Health & Medical Sciences (AREA)
- Life Sciences & Earth Sciences (AREA)
- Computing Systems (AREA)
- General Engineering & Computer Science (AREA)
- Mathematical Physics (AREA)
- Software Systems (AREA)
- Image Processing (AREA)
- Image Analysis (AREA)
Abstract
The invention discloses a degradation kernel extraction method for an image blind super-resolution enhancement network, which comprises the following steps of: s1, performing fuzzy processing on a high-resolution image by using a random parameter anisotropic Gaussian kernel as a real degradation kernel, and performing downsampling to obtain a low-resolution image; s2, inputting the low-resolution image into a degradation characterization learning network to extract degradation kernels thereof; s3, respectively carrying out fuzzy processing on the low-resolution image by using the extracted degradation kernel and the corresponding real degradation kernel, and downsampling to obtain a resampled image pair; and S4, calculating an error value between the resampled image pairs and an error value between the degradation kernels, and performing constraint training on the degradation characterization learning network by using the weighted sum of the error value and the error value as a total error value. The degradation kernel extraction method for the image blind super-resolution enhancement network can effectively extract the degradation kernel in the low-resolution image, and can obtain better reconstruction effect when being applied to the image blind super-resolution enhancement technology.
Description
Technical Field
The invention belongs to the technical field of image processing, and particularly relates to a degradation kernel extraction method for an image blind super-resolution enhancement network.
Background
Image super-resolution enhancement is a research problem which is paid attention to in the field of image processing, and has become a key ring in the image signal processing flow. For the low-resolution imaging equipment, the image super-resolution enhancement can effectively improve the image resolution through an algorithm means under the condition of not changing sensor hardware, so that the visual effect of the image is optimized, the characteristic information of the image is enriched, and the expression capability of the image is improved at lower cost. In recent years, the algorithm effect of image super-resolution enhancement is rapidly improved along with the development of convolutional neural networks.
The convolutional neural network-based super-resolution enhancement algorithm initially uses high-low resolution image pairs as training sets to learn the mapping of low resolution images to high resolution images with the strong characterizations capabilities of the network model. However, high-low resolution image pairs of the real world are generally difficult to obtain, and most of low resolution images in common training sets are obtained by downsampling high resolution images by using a bicubic interpolation algorithm, so that the super-resolution enhancement effect of the network model in the real world scene is difficult to meet the actual application requirements. Subsequently, the non-blind super-resolution enhancement algorithm tries to guide the super-resolution enhancement process of the image by using a preset fixed degradation kernel, but such algorithm has difficulty in ensuring that the preset degradation kernel is matched with the actual degradation process of the low-resolution image to be processed, and the super-resolution enhancement effect is unstable. In order to solve the problem, a special degradation kernel extraction module is designed for the blind super-resolution network algorithm, and the purpose is to obtain a real degradation kernel of a low-resolution image to be processed, and the real degradation kernel is used for replacing a fixed degradation kernel preset in a non-blind super-resolution neural network. Clearly, designing a network algorithm to accurately extract the degradation kernel of a low resolution image is a key to achieving blind super resolution enhancement of a high quality image.
Disclosure of Invention
The invention aims to provide a degradation kernel extraction method for an image blind super-resolution enhancement network, which can effectively extract degradation kernels in a low-resolution image and can obtain a better reconstruction effect when being applied to an image blind super-resolution enhancement technology. Firstly, using an anisotropic Gaussian kernel with random parameters as a real degradation kernel K GT to carry out fuzzy degradation on a high-resolution image, and downsampling to obtain a low-resolution image LR; then, inputting the image LR into a degradation characterization learning network to obtain an extracted degradation kernel K fake; then, blurring processing is carried out on the low-resolution image LR by using an extracted degradation kernel K fake and a corresponding real degradation kernel K GT respectively, and downsampling is carried out to obtain a resampled image pair RLR fake and RLR GT; finally, the error value between the resampled images RLR fake and RLR GT and the error value between the degradation kernels K fake and K GT are calculated, and the network parameters are iterated continuously to enable the weighted sum of the two to converge, so that training is completed. During reasoning, the image to be amplified is input into a trained degradation characterization learning network, and a degradation kernel of the image can be obtained.
In order to achieve the above purpose, the invention adopts the following technical scheme:
A degradation kernel extraction method for an image blind super-resolution enhancement network, comprising the steps of:
Firstly, using an anisotropic Gaussian kernel with random parameters as a real degradation kernel K GT to carry out fuzzy degradation on a high-resolution image, and downsampling to obtain a low-resolution image LR; then, inputting the image LR into a degradation characterization learning network to obtain an extracted degradation kernel K fake; then, blurring processing is carried out on the low-resolution image LR by using an extracted degradation kernel K fake and a corresponding real degradation kernel K GT respectively, and downsampling is carried out to obtain a resampled image pair RLR fake and RLR GT; finally, calculating an error value between the resampled images RLR fake and RLR GT and an error value between the degradation kernels K fake and K GT, continuously iterating the network parameters to enable the weighted sum of the two to be converged, and completing training; during reasoning, the image to be amplified is input into a trained degradation characterization learning network, and a degradation kernel of the image can be obtained.
The invention is further improved in that the method specifically comprises the following steps:
1) Generating a data set for network training
Firstly, using an anisotropic Gaussian kernel K GT with random parameters as a real degradation check to carry out HR convolution on a high-resolution image to obtain a degradation image, wherein the resolution of the degradation image is consistent with the high-resolution image; then, the degraded image is downsampled by using a bicubic interpolation algorithm to obtain a low-resolution image LR; thereby obtaining the data set used;
2) Extraction of degenerated kernels and resampling using neural networks
Firstly, inputting a low-resolution image LR into a degradation characterization learning network to obtain a degradation kernel K fake; then, the input low resolution blurring process is performed by using the real degradation kernel K GT and the extracted degradation kernel K fake, respectively, and resampled images RLR GT and RLR fake are obtained by sampling;
3) Training of co-constrained network models using dual loss functions
Calculating an error value between the degradation kernel K GT and the degradation kernel K fake by using an L1 loss function, calculating an error value between the resampled image RLR GT and the resampled image RLR fake by using an MSE loss function, wherein the weighted sum of the two is a total error value, updating network parameters to enable the total error value to be converged to the minimum, and completing training;
4) And inputting the image to be amplified into a trained degradation characterization learning network, and obtaining a degradation kernel of the image.
The invention is further improved in that in step 1), the data set used comprises in particular: a true degenerate kernel K GT and a low-resolution image LR.
A further improvement of the invention is that each low resolution image corresponds to a true degradation kernel.
The invention is further improved in that in step 2), resampled images RLR GT and RLR fake are obtained by downsampling using a bicubic interpolation algorithm.
The invention is further improved in that in step 2), after the degradation kernel of the low resolution image is extracted by using the degradation characterization learning network, the low resolution image is blurred again and downsampled to obtain a resampled image pair RLR GT and RLR fake.
The invention further improves in that in step 3), a multi-domain constraint loss function is designed, namely, the error value of the degradation kernel domain is calculated by using the L1 loss function, and the loss function of the image domain is calculated by using the MSE loss function.
A further improvement of the invention is that in step 3), the Adam optimizer is used to update the network parameters so that the total error value converges to a minimum value, thus completing the training.
Compared with the prior art, the invention has at least the following beneficial technical effects:
the invention provides a degradation kernel extraction method for an image blind super-resolution enhancement network, which uses a method more conforming to a real degradation process to generate data for training, so that the generalization capability of a network model is improved; the mapping range of the network is reduced by increasing constraint conditions, so that the extracted degradation core is closer to the degradation effect of the real degradation core. The degradation kernel extracted by the method can be widely applied to super-resolution enhancement tasks of real world scenes and can obtain better super-resolution enhancement effects.
In step 1), a random parameter anisotropic gaussian kernel is used as a real degradation kernel to carry out fuzzy processing on the high-resolution image, and then a bicubic interpolation algorithm is used to downsample the degraded image to obtain a low-resolution degraded image. Compared with a data set generation method which only downsamples and does not degrade, the method can better fit the real-world image degradation rule, and the generated low-resolution image can better simulate an actual scene. Meanwhile, the anisotropic Gaussian kernel corresponding to each image is randomly generated by an algorithm, so that the obtained degradation sample covered by the low-resolution image set is wide, and the trained network generalization capability is strong.
Further, in step 2), the constraint condition of the network is increased, that is, the low-resolution image to be processed is degraded again by using a degradation kernel algorithm to obtain a resampled image, and the resampled image is used for network training. Compared with other methods which only use the degradation kernel as the supervision sample, the method enriches the supervision sample of the network, reduces the mapping space of the network, accelerates the convergence speed of the network, and enables the network to extract the degradation kernel from the low-resolution image more quickly and accurately.
Further, in step 3), the error value between the two degenerate kernels is calculated using Loss Kernel, the error value between the two resampled images is calculated using Loss RLR, and the network parameters are optimized using the weighted sum of the two. The dual loss function is adopted to jointly restrict the network training, so that the extracted degradation core is closer to the degradation effect of the real degradation core, and the extracted degradation core can obtain a better super-resolution enhancement effect when applied to the blind super-resolution technology.
Drawings
FIG. 1 is a flow chart of an implementation method of the present invention.
FIG. 2 is a schematic diagram of a training data generation method used in the present invention.
FIG. 3 is a schematic diagram of a degradation characterization learning network structure according to the present invention.
Fig. 4 is a schematic diagram of the overall structure of the present invention.
FIG. 5 is a diagram showing an example of a training data generation method used in the present invention.
FIG. 6 is a graph comparing the effect of the present invention with the effect of the prior mainstream algorithm.
Detailed Description
Exemplary embodiments of the present disclosure will be described in more detail below with reference to the accompanying drawings. While exemplary embodiments of the present disclosure are shown in the drawings, it should be understood that the present disclosure may be embodied in various forms and should not be limited to the embodiments set forth herein. Rather, these embodiments are provided so that this disclosure will be thorough and complete, and will fully convey the scope of the disclosure to those skilled in the art. It should be noted that, without conflict, the embodiments of the present invention and features of the embodiments may be combined with each other. The invention will be described in detail below with reference to the drawings in connection with embodiments.
As shown in fig. 1, the degradation kernel extraction method for the image blind super-resolution enhancement network provided by the invention comprises the following steps:
S1, performing fuzzy processing on a high-resolution image by using a random parameter anisotropic Gaussian kernel as a real degradation kernel, and performing downsampling to obtain a low-resolution image;
s2, inputting the low-resolution image into a degradation characterization learning network to extract degradation kernels thereof;
S3, respectively carrying out fuzzy processing on the low-resolution image by using the extracted degradation kernel and the corresponding real degradation kernel, and downsampling to obtain a resampled image pair;
s4, calculating an error value between the resampled image pairs and an error value between the degradation kernels, and performing constraint training on the degradation characterization learning network by using the weighted sum of the error value and the degradation kernels as a total error value;
S5, inputting the image to be amplified into a trained degradation characterization learning network, and obtaining a degradation kernel of the image.
The steps are discussed in detail below:
S1: as shown in fig. 2, first, an anisotropic gaussian kernel with random parameters is generated as a true degenerate kernel K GT in the training set; then, using anisotropic Gaussian to check the convolution of the high-resolution image to obtain a degraded image, wherein the resolution of the degraded image is consistent with the high-resolution image; and finally, performing downsampling operation on the degraded image by adopting a bicubic interpolation algorithm to obtain a corresponding low-resolution image LR. The data set used in the present invention is thus obtained, comprising: a low resolution image LR and a corresponding true degenerate kernel K GT.
S2: as shown in fig. 3, the degradation characterization learning network is composed of a degradation feature extraction module and a degradation kernel reconstruction module, and the specific structure and the parameter diagram are marked. Taking as an example the extraction of a degradation kernel of size 19×19 from a low resolution image LR of resolution size 512×512: firstly, inputting an image LR into a degradation feature extraction module, and obtaining a feature tensor in the shape of 1 multiplied by 64 multiplied by 512 through the processing of a shallow feature extraction module; then, processing by two deep feature extraction modules to obtain a feature tensor in the form of 1×256×128×128, wherein the number of channels of the feature tensor is increased by 2 times and the length and width of the channel of the feature tensor are reduced by 2 times after each processing; inputting the feature tensor into the feature mapping module, wherein the size of the degradation kernel is 19×19, so that the number of channels after processing becomes 361, and the length and width are unchanged, and the shape is 1×361×128×128; then using an adaptive average pooling layer to carry out average pooling operation on the length and width dimensions of the feature tensor to obtain the feature tensor in the shape of 1 multiplied by 361 multiplied by 1; and finally, inputting the characteristic tensor into a degenerated kernel reconstruction module, remolding the characteristic tensor to 19 multiplied by 19, and optimizing again through a three-layer perceptron to obtain the final degenerated kernel.
S3: as shown in fig. 4, the low resolution image is blurred degraded using the extracted degradation kernel K fake and the true degradation kernel K GT, respectively, and the degraded image is downsampled using a bicubic interpolation algorithm to obtain resampled images RLR fake and RLR GT.
S4: the error value between the two degenerate kernels is calculated using Loss Kernel, the error value between the two resampled images is calculated using Loss RLR, and the network parameters are optimized using the weighted sum of the two. Specifically, the Loss function value Loss Kernel of the degradation kernel and the Loss function Loss RLR of the resampled image are multiplied by corresponding coefficients and added to obtain a total similarity function Loss total:
LossKernel＝L1(KGT,Kfake)
LossRLR＝MSE(RLRGT,RLRfake)
Losstotal＝λKernel×LossKernel+λRLR×LossRLR
Wherein λ Kernel is the coefficient of the degradation kernel loss function value, here taken as 10; lambada RLR is the coefficient of the resampled image loss function, here taken as 1.
In order to improve learning convergence speed and prevent sinking into local optimum points, an Adam optimizer is used for updating model parameters.
Examples
First, as shown in fig. 5, a true degradation kernel and a low resolution image are obtained through the method of step S1; the low resolution image is then fed into a degradation characterization learning network, which results in an extracted degradation kernel and a final enhanced image via steps S2, S3, S4. As shown in FIG. 6, compared with other methods, the degradation kernel extracted by the method is closest to the real degradation kernel, and the edges and textures of the object in the final enhanced image are clear and easy to distinguish.
While the invention has been described in detail in the foregoing general description and with reference to specific embodiments thereof, it will be apparent to one skilled in the art that modifications and improvements can be made thereto. Accordingly, such modifications or improvements may be made without departing from the spirit of the invention and are intended to be within the scope of the invention as claimed.
Claims (8)
1. A degradation kernel extraction method for an image blind super-resolution enhancement network, comprising the steps of:
Firstly, using an anisotropic Gaussian kernel with random parameters as a real degradation kernel K GT to carry out fuzzy degradation on a high-resolution image, and downsampling to obtain a low-resolution image LR; then, inputting the image LR into a degradation characterization learning network to obtain an extracted degradation kernel K fake; then, blurring processing is carried out on the low-resolution image LR by using an extracted degradation kernel K fake and a corresponding real degradation kernel K GT respectively, and downsampling is carried out to obtain a resampled image pair RLR fake and RLR GT; finally, calculating an error value between the resampled images RLR fake and RLR GT and an error value between the degradation kernels K fake and K GT, continuously iterating the network parameters to enable the weighted sum of the two to be converged, and completing training; during reasoning, the image to be amplified is input into a trained degradation characterization learning network, and a degradation kernel of the image can be obtained.
2. The degradation kernel extraction method for an image blind super-resolution enhancement network according to claim 1, wherein the method specifically comprises the following steps:
1) Generating a data set for network training
Firstly, using an anisotropic Gaussian kernel K GT with random parameters as a real degradation check to carry out HR convolution on a high-resolution image to obtain a degradation image, wherein the resolution of the degradation image is consistent with the high-resolution image; then, the degraded image is downsampled by using a bicubic interpolation algorithm to obtain a low-resolution image LR; thereby obtaining the data set used;
2) Extraction of degenerated kernels and resampling using neural networks
Firstly, inputting a low-resolution image LR into a degradation characterization learning network to obtain a degradation kernel K fake; then, the input low resolution blurring process is performed by using the real degradation kernel K GT and the extracted degradation kernel K fake, respectively, and resampled images RLR GT and RLR fake are obtained by sampling;
3) Training of co-constrained network models using dual loss functions
Calculating an error value between the degradation kernel K GT and the degradation kernel K fake by using an L1 loss function, calculating an error value between the resampled image RLR GT and the resampled image RLR fake by using an MSE loss function, wherein the weighted sum of the two is a total error value, updating network parameters to enable the total error value to be converged to the minimum, and completing training;
4) And inputting the image to be amplified into a trained degradation characterization learning network, and obtaining a degradation kernel of the image.
3. The method for extracting degradation kernel for image blind super-resolution enhancement network according to claim 2, wherein in step 1), the data set used specifically comprises: a true degenerate kernel K GT and a low-resolution image LR.
4. A degradation kernel extraction method for an image blind super-resolution enhancement network according to claim 3, wherein each low resolution image corresponds to a real degradation kernel.
5. The method for extracting the degradation kernel for the image blind super-resolution enhancement network according to claim 2, wherein in the step 2), resampled images RLR GT and RLR fake are obtained by downsampling using a bicubic interpolation algorithm.
6. The method for extracting degradation kernel of image blind super-resolution enhancement network according to claim 2, wherein in step 2), after extracting degradation kernel of low-resolution image using degradation characterization learning network, the low-resolution image is blurred again and downsampled to obtain a resampled image pair RLR GT and RLR fake.
7. The method according to claim 2, wherein in step 3), a multi-domain constraint loss function is designed, i.e. the error value of the degraded kernel domain is calculated using the L1 loss function, and the loss function of the image domain is calculated using the MSE loss function.
8. The method for extracting degradation kernel of image blind super-resolution enhancement network according to claim 2, wherein in step 3), the Adam optimizer is used to update network parameters to minimize the total error value, and training is completed.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202210467732.6A CN114998666B (en) | 2022-04-29 | 2022-04-29 | Degradation kernel extraction method for image blind super-resolution enhancement network |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202210467732.6A CN114998666B (en) | 2022-04-29 | 2022-04-29 | Degradation kernel extraction method for image blind super-resolution enhancement network |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN114998666A CN114998666A (en) | 2022-09-02 |
| CN114998666B true CN114998666B (en) | 2024-08-06 |
Family
ID=83025364
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202210467732.6A Active CN114998666B (en) | 2022-04-29 | 2022-04-29 | Degradation kernel extraction method for image blind super-resolution enhancement network |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN114998666B (en) |
Citations (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN113191959A (en) * | 2021-03-08 | 2021-07-30 | 浙江大学 | Digital imaging system limit image quality improving method based on degradation calibration |
| CN113888413A (en) * | 2021-12-03 | 2022-01-04 | 湖南大学 | Different-time different-source multispectral image blind spectrum super-resolution method and system |
Family Cites Families (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| TWI768364B (en) * | 2020-06-01 | 2022-06-21 | 宏碁股份有限公司 | Method and electronic device for processing images that can be played on a virtual device |
| CN114170088A (en) * | 2021-12-15 | 2022-03-11 | 中山大学 | Relational reinforcement learning system and method based on graph structure data |
-
2022
- 2022-04-29 CN CN202210467732.6A patent/CN114998666B/en active Active
Patent Citations (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN113191959A (en) * | 2021-03-08 | 2021-07-30 | 浙江大学 | Digital imaging system limit image quality improving method based on degradation calibration |
| CN113888413A (en) * | 2021-12-03 | 2022-01-04 | 湖南大学 | Different-time different-source multispectral image blind spectrum super-resolution method and system |
Also Published As
| Publication number | Publication date |
|---|---|
| CN114998666A (en) | 2022-09-02 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN109859147B (en) | Real image denoising method based on generation of antagonistic network noise modeling | |
| CN111028177B (en) | Edge-based deep learning image motion blur removing method | |
| Kim et al. | Unsupervised real-world super resolution with cycle generative adversarial network and domain discriminator | |
| CN111028163A (en) | Convolution neural network-based combined image denoising and weak light enhancement method | |
| CN110781895A (en) | Image semantic segmentation method based on convolutional neural network | |
| CN112150354B (en) | Single image super-resolution method combining contour enhancement and denoising statistical prior | |
| CN110796622B (en) | Image bit enhancement method based on multi-layer characteristics of series neural network | |
| CN110111251B (en) | Image super-resolution reconstruction method combining depth supervision self-coding and perception iterative back projection | |
| CN112581397B (en) | Degraded image restoration method, system, medium and equipment based on image priori information | |
| CN114331886B (en) | Image deblurring method based on depth features | |
| CN112164011A (en) | Motion image deblurring method based on self-adaptive residual error and recursive cross attention | |
| CN112669214A (en) | Fuzzy image super-resolution reconstruction method based on alternative direction multiplier algorithm | |
| CN113096013B (en) | Blind image super-resolution reconstruction method and system based on imaging modeling and knowledge distillation | |
| CN116645287B (en) | Diffusion model-based image deblurring method | |
| CN115345791A (en) | Infrared image deblurring algorithm based on attention mechanism residual error network model | |
| CN117036880A (en) | Sea surface temperature image complement method and network based on denoising diffusion generation model | |
| CN114998107A (en) | Image blind super-resolution network model, method, equipment and storage medium | |
| CN111145125A (en) | Image denoising method based on residual learning and convolutional neural network | |
| Kas et al. | DLL-GAN: Degradation-level-based learnable adversarial loss for image enhancement | |
| CN116862795A (en) | Multistage motion blur removing method based on pixel-by-pixel degradation prediction network | |
| CN113096032A (en) | Non-uniform blur removing method based on image area division | |
| CN114764750A (en) | Image denoising method based on self-adaptive consistency prior depth network | |
| CN114998666B (en) | Degradation kernel extraction method for image blind super-resolution enhancement network | |
| CN116188265A (en) | Space variable kernel perception blind super-division reconstruction method based on real degradation | |
| CN112132756A (en) | Attention mechanism-based single raindrop image enhancement method |
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 |