CN118052711A - Single-patch image super-resolution recapture algorithm based on coordinate attention and residual error network mechanism technology - Google Patents
Single-patch image super-resolution recapture algorithm based on coordinate attention and residual error network mechanism technology Download PDFInfo
- Publication number
- CN118052711A CN118052711A CN202410195789.4A CN202410195789A CN118052711A CN 118052711 A CN118052711 A CN 118052711A CN 202410195789 A CN202410195789 A CN 202410195789A CN 118052711 A CN118052711 A CN 118052711A
- Authority
- CN
- China
- Prior art keywords
- image
- resolution
- attention
- residual
- network
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Pending
Links
- 238000005516 engineering process Methods 0.000 title claims abstract description 26
- 230000007246 mechanism Effects 0.000 title claims abstract description 24
- 238000012545 processing Methods 0.000 claims abstract description 44
- 238000004458 analytical method Methods 0.000 claims abstract description 16
- 238000005070 sampling Methods 0.000 claims abstract description 15
- 238000005457 optimization Methods 0.000 claims abstract description 10
- 238000000605 extraction Methods 0.000 claims abstract description 8
- 238000007781 pre-processing Methods 0.000 claims abstract description 7
- 238000013527 convolutional neural network Methods 0.000 claims description 14
- 230000006870 function Effects 0.000 claims description 13
- 238000004364 calculation method Methods 0.000 claims description 11
- 238000012937 correction Methods 0.000 claims description 10
- 238000005282 brightening Methods 0.000 claims description 9
- 238000013507 mapping Methods 0.000 claims description 9
- 238000000034 method Methods 0.000 claims description 9
- 208000037170 Delayed Emergence from Anesthesia Diseases 0.000 claims description 8
- 238000010606 normalization Methods 0.000 claims description 6
- 238000011176 pooling Methods 0.000 claims description 6
- 230000000007 visual effect Effects 0.000 claims description 6
- 230000004913 activation Effects 0.000 claims description 3
- 230000002776 aggregation Effects 0.000 claims description 3
- 238000004220 aggregation Methods 0.000 claims description 3
- 238000012549 training Methods 0.000 claims description 3
- 230000009466 transformation Effects 0.000 claims description 3
- 230000002146 bilateral effect Effects 0.000 claims description 2
- 230000000087 stabilizing effect Effects 0.000 claims description 2
- 230000001629 suppression Effects 0.000 claims description 2
- 230000009286 beneficial effect Effects 0.000 description 3
- 230000009467 reduction Effects 0.000 description 3
- 238000013528 artificial neural network Methods 0.000 description 2
- 238000010586 diagram Methods 0.000 description 2
- 230000008569 process Effects 0.000 description 2
- 238000003672 processing method Methods 0.000 description 2
- 230000003213 activating effect Effects 0.000 description 1
- 230000004075 alteration Effects 0.000 description 1
- 230000003321 amplification Effects 0.000 description 1
- 230000008901 benefit Effects 0.000 description 1
- 230000008033 biological extinction Effects 0.000 description 1
- 238000000354 decomposition reaction Methods 0.000 description 1
- 238000013135 deep learning Methods 0.000 description 1
- 230000007547 defect Effects 0.000 description 1
- 238000002059 diagnostic imaging Methods 0.000 description 1
- 230000000694 effects Effects 0.000 description 1
- 238000012986 modification Methods 0.000 description 1
- 230000004048 modification Effects 0.000 description 1
- 238000003199 nucleic acid amplification method Methods 0.000 description 1
- 230000000452 restraining effect Effects 0.000 description 1
- 238000006467 substitution reaction Methods 0.000 description 1
- 238000012546 transfer Methods 0.000 description 1
Landscapes
- Image Processing (AREA)
Abstract
The invention relates to the technical field of single-patch image super-resolution processing, and discloses a single-patch image super-resolution re-picking algorithm based on a coordinate attention and residual error network mechanism technology, which comprises the following steps: firstly, an image acquisition module acquires a low-resolution image through a preprocessing unit, and forms an image data set after standardized processing; step two, the analysis and re-picking module numbers the image dataset according to the characteristics of the image dataset, inputs the image dataset into a residual error network unit for characteristic extraction, and calculates and generates a residual error characteristic group Catz; thirdly, the coordinate attention unit calculates and generates an attention feature group Zytz according to the residual feature group Catz; step four, the up-sampling unit samples and generates a high-resolution image according to the attention feature group Zytz; and fifthly, the vision processing module performs optimization processing for two times according to the high-resolution image, and outputs the image through network connection.
Description
Technical Field
The invention relates to the technical field of single-patch image super-resolution processing, in particular to a single-patch image super-resolution re-picking algorithm based on a coordinate attention and residual error network mechanism technology.
Background
A single patch image refers to a single type of map, a type of graphical data for a 3D model surface. The map is used in various combinations to define the properties of the object surface, such as color, brightness, self-luminosity, opacity, etc. In game making, different types and numbers of maps may be used to achieve the desired visual effect, as required by the project asset. Therefore, a super-resolution processing technology is required, the definition and quality of the image are improved under the condition of not increasing the original image data, the distortion problem during image amplification is better solved, and meanwhile, the natural sense and detail richness of the image are maintained.
In the field of image processing, super-resolution technology refers to a process of reconstructing a high-resolution image from a low-resolution image. Such algorithms typically incorporate a variety of deep learning techniques, including attention mechanisms, residual networks (ResNet), and Convolutional Neural Networks (CNNs), to improve performance and efficiency. The residual network solves the problem of gradient extinction in deep networks by introducing "jump connections" (also called residual connections or shortcut connections). These connections enable the network to directly transfer signals from one layer to the subsequent layer, so that the network can learn the residual mapping between inputs and outputs. The attention mechanism mimics the visual attention of humans, allowing models to focus on more important information. In image super resolution this means that the network can focus on the high frequency details of the image, thereby recovering lost details more effectively. The super-resolution technology is widely applied in the fields of medical imaging, satellite image processing, video streaming, personal photo enhancement and the like. It has important practical significance for improving the usability of the image, especially when the original high-resolution image cannot be acquired.
At present, traditional single-patch image super-resolution recapturing algorithms rely on signal processing technologies, such as wavelet transform interpolation methods and the like, attempt to recover high-resolution details of images through decomposition and resynthesis of image characteristic information, although the image size can be improved, the high-frequency details cannot be recovered well, the result images are blurred, textures and edge sharpness are lacked, the traditional algorithms cannot learn complex nonlinear mapping relations autonomously, the calculation amount is large, the real-time processing requirements are difficult to meet, and the performance and applicability of the traditional single-patch image super-resolution recapturing algorithms are limited to a certain extent.
Disclosure of Invention
(One) solving the technical problems
Aiming at the defects of the prior art, the invention provides a single-patch image super-resolution restoration algorithm based on a coordinate attention and residual error network mechanism technology, which has the advantages of high fidelity restoration of image detail textures, wide autonomous learning applicability and the like, and solves the problems that the traditional single-patch image super-resolution restoration algorithm has low detail texture restoration degree and cannot autonomously learn processing performance to be limited.
(II) technical scheme
In order to achieve the above purpose, the present invention provides the following technical solutions: the single-patch image super-resolution recapture algorithm based on the coordinate attention and residual error network mechanism technology carries out super-resolution processing through an image recapture system, the image recapture system comprises an image acquisition module, an analysis recapture module and a vision processing module, and the single-patch image super-resolution recapture algorithm based on the coordinate attention and residual error network mechanism technology comprises the following steps:
Firstly, an image acquisition module acquires a low-resolution image through a preprocessing unit, and forms an image data set after standardized processing, and the image acquisition module is connected with an analysis and re-pickup module through a network;
Step two, the analysis and re-picking module numbers the image dataset according to the characteristics of the image dataset, inputs the image dataset into a residual error network unit for characteristic extraction, calculates and generates a residual error characteristic group Catz, and transmits the residual error characteristic group Catz to a coordinate attention unit through a network;
Thirdly, the coordinate attention unit calculates and generates an attention feature group Zytz according to the residual feature group Catz, and transmits the attention feature group Zytz to the up-sampling unit through a network;
Step four, the up-sampling unit samples and generates a high-resolution image according to the attention feature group Zytz, and transmits the high-resolution image to the vision processing module through a network;
And fifthly, the vision processing module performs optimization processing for two times according to the high-resolution image, and outputs the image through network connection.
Preferably, the preprocessing unit performs standardization processing through a bilateral filter, performs pixel denoising on a single or multiple low-resolution images, and protects image edges and texture areas to form an image data set.
Preferably, the analysis and re-picking module comprises a residual error network unit, a coordinate attention unit and an up-sampling unit, and the analysis and re-picking module numbers the image data set according to the characteristics of the image data set, wherein the image data set is numbered as TX 1、TX2、TX3、…TXn and is transmitted to the residual error network unit through a network.
Preferably, the residual network unit is provided with a deep convolutional neural network, and the deep convolutional neural network comprises a convolutional layer JU, an activation function JH, a batch normalization layer GY, a pooling layer CH and a full connection layer QL.
Preferably, the residual network unit inputs the image dataset into a deep convolutional neural network, performs feature extraction between each residual block in a jump connection manner, and calculates and generates a residual feature group Catz, and the calculation formula is as follows:
In the formula, catz represents a residual characteristic group, pi JUTXn represents characteristic data obtained by multiplying a convolution kernel of a convolution layer with image data point by point and summing the result, maxJH (0, TX n) represents that the image data is substituted into an activated function algorithm to obtain nonlinear complex characteristics, sigma GYTXn mu represents that when the image data amount is large, a batch normalization layer stabilizing network training gradient is substituted, TX n-CH|e-n represents that the image data is substituted into a pooling layer to obtain key characteristic data, and U QLTXn+∞ represents that the full connection layer integrates the image data characteristics extracted by a plurality of residual blocks and finally outputs the key characteristic data.
Preferably, the coordinate attention unit is provided with a reference plane coordinate model ZB and a reference mapping function model YH.
Preferably, the coordinate attention unit calculates and generates an attention feature group Zytz according to the residual feature group Catz, and the calculation formula is as follows:
wherein Zytz denotes the set of attention features, Representing attention weight images calculated by substituting residual feature set data into a reference mapping function and a reference plane coordinate model transformation, including attention coordinate coding and attention feature aggregation,/>Feature weighting data representing suppression of secondary features and highlighting important features for the attention weighted image.
Preferably, the up-sampling unit increases the number of pixels of the image by a nearest neighbor interpolation method according to the attention feature set Zytz, and samples the image to generate a high-resolution image.
Preferably, the vision processing module comprises a color correction unit and a sharpening and brightening unit, wherein the color correction unit is connected with computer image software according to the high-resolution image to perform first optimization processing and is transmitted to the sharpening and brightening unit through a network.
Preferably, the sharpening and brightening unit is connected with computer image software for performing a second optimization process according to the high-resolution image after color correction, and outputs an optimal image through a network.
Compared with the prior art, the invention provides a single-patch image super-resolution re-picking algorithm based on a coordinate attention and residual error network mechanism technology, which has the following beneficial effects:
1. According to the invention, the image acquisition module acquires the low-resolution image through the preprocessing unit, performs standardization processing, performs pixel denoising on a single image or a plurality of low-resolution images, protects image edges and texture areas to form an image data set, and is beneficial to improving the calculation efficiency of super-resolution re-picking of the subsequent image while keeping details undistorted when the original image data of the standardization processing is subjected to standardization processing, wherein the residual error network unit, the coordinate attention unit and the up-sampling unit are arranged in the analysis re-picking module, and the deep convolution neural network is arranged in the residual error network unit, so that the problems of distortion and edge jaggy in the traditional image super-resolution re-picking processing method are solved, the key characteristics of the captured image are independently learned and captured, and the fidelity reduction degree of the image detail textures is high.
2. According to the invention, the image dataset is transmitted to the deep convolutional neural network of the residual network unit through the analysis and re-picking module, feature extraction is carried out between each residual block in a jump connection mode, a residual feature group Catz is calculated and generated, a complex visual task can be stably carried out, the coordinate attention unit is provided with a reference plane coordinate model ZB and a reference mapping function model YH, an attention feature group Zytz is calculated and generated, the image features are better understood by utilizing the spatial position relation, so that the super-resolution re-picking performance and the generalization capability of the image are improved, the convolutional feature expression capability of the deep convolutional neural network is effectively improved, the autonomous learning applicability is wide, the up-sampling unit increases the number of image pixels through a nearest neighbor interpolation method, the sampling generates a high-resolution image, and the vision processing module is matched for carrying out color correction and sharpening to improve the brightness and output optimal image, and the visual effect is further improved while the fidelity reduction degree of the detail texture of the image is high.
Drawings
FIG. 1 is a step diagram of an algorithm of the present invention;
FIG. 2 is a schematic diagram of the system of the present invention.
Detailed Description
The following description of the embodiments of the present invention will be made clearly and completely with reference to the accompanying drawings, in which it is apparent that the embodiments described are only some embodiments of the present invention, but not all embodiments. All other embodiments, which can be made by those skilled in the art based on the embodiments of the invention without making any inventive effort, are intended to be within the scope of the invention.
Referring to fig. 1, a single-patch image super-resolution recapture algorithm based on a coordinate attention and residual error network mechanism technology performs super-resolution processing through an image recapture system, wherein the image recapture system comprises an image acquisition module, an analysis recapture module and a vision processing module, and the single-patch image super-resolution recapture algorithm based on the coordinate attention and residual error network mechanism technology comprises the following steps:
The method comprises the steps that firstly, an image acquisition module acquires low-resolution images through a preprocessing unit, performs standardization processing, performs pixel denoising on single or multiple low-resolution images, protects image edges and texture areas to form an image data set, and is beneficial to improving the super-resolution recapture calculation efficiency of subsequent images when the standardization processing initial image data keeps details undistorted, and the image acquisition module is connected with an analysis recapture module through a network;
The analysis and re-picking module comprises a residual network unit, a coordinate attention unit and an up-sampling unit, the analysis and re-picking module numbers the image data set according to the characteristics of the image data set, the image data set is numbered as TX 1、TX2、TX3、…TXn, the numbers correspond to the number of the images to be processed one by one, the residual network unit is input to the residual network unit for characteristic extraction, the residual network unit is provided with a deep convolutional neural network, the deep convolutional neural network comprises a convolutional layer JU, an activating function JH, a batch normalization layer GY, a pooling layer CH and a full-connection layer QL, the deep convolutional neural network solves the problems of distortion and edge saw-tooth in the traditional image super-resolution re-picking processing method, the key characteristics of the images are captured by autonomous learning, the selection characteristics are not required to be set manually, the calculation difficulty is reduced, the residual network unit inputs the image data set into the deep convolutional neural network, the characteristic extraction is carried out between each residual block in a jump connection mode, the data processing mode is flexible and efficient, a residual characteristic group Catz is generated by calculation and is transmitted to the coordinate attention unit through the network, and the calculation formula is as follows:
In the formula, catz represents a residual feature group, pi JUTXn represents feature data obtained by carrying out point-by-point multiplication and summation on convolution kernels of a convolution layer and image data, maxJH (0, TX n) represents that the image data is substituted into an activation function algorithm to obtain nonlinear complex features, sigma GYTXn mu represents that when the image data amount is large, a batch normalization layer stable network training gradient is substituted, I TX n-CH|e-n represents that the image data is substituted into a pooling layer to obtain key feature data, U QLTXn+∞ represents that the full-connection layer integrates image data features extracted by a plurality of residual blocks to carry out final output, and according to the residual feature group Catz, a plurality of residual blocks work cooperatively, so that rich features can be effectively extracted from an image, a complex visual task is carried out, and the self-learning applicability is wide;
the third step, the coordinate attention unit is provided with a reference plane coordinate model ZB and a reference mapping function model YH, and calculates and generates an attention feature group Zytz according to the residual feature group Catz, and transmits the attention feature group Zytz to the up-sampling unit through a network, wherein the calculation formula is as follows:
wherein Zytz denotes the set of attention features, Representing attention weight images calculated by substituting residual feature set data into a reference mapping function and a reference plane coordinate model transformation, including attention coordinate coding and attention feature aggregation,/>The feature weighting data obtained by restraining secondary features and highlighting important features on the attention weighted image is represented, and according to the attention feature group Zytz, the system can better understand the image features by utilizing the spatial position relation, so that the super-resolution recapturing performance and generalization capability of the image are improved, and the convolution feature expression capability of a deep convolution neural network can be effectively improved;
The up-sampling unit increases the number of image pixels by a nearest neighbor interpolation method according to the attention feature group Zytz, samples and generates a high-resolution image, assigns the high-resolution image to a target pixel according to the nearest pixel point of the target pixel in the source image, has simple operation and easy realization, retains detail features, is suitable for images with different multiples of each type and is transmitted to a vision processing module through a network;
and fifthly, the vision processing module comprises a color correction unit and a sharpening and brightening unit, the color correction unit is connected with computer image software according to the high-resolution image to perform first optimization processing and is transmitted to the sharpening and brightening unit through a network, the sharpening and brightening unit is connected with the computer image software according to the high-resolution image after color correction to perform second optimization processing, the high-resolution image is connected with the network to output an optimal image after two optimization processing, and the fidelity reduction degree of the image detail texture is high, and meanwhile, the vision effect is further improved.
Although embodiments of the present invention have been shown and described, it will be understood by those skilled in the art that various changes, modifications, substitutions and alterations can be made therein without departing from the principles and spirit of the invention, the scope of which is defined in the appended claims and their equivalents.
Claims (10)
1. A single-patch image super-resolution recapture algorithm based on a coordinate attention and residual error network mechanism technology is characterized in that: the single-patch image super-resolution recapture algorithm based on the coordinate attention and residual error network mechanism technology carries out super-resolution processing through an image recapture system, the image recapture system comprises an image acquisition module, an analysis recapture module and a vision processing module, and the single-patch image super-resolution recapture algorithm based on the coordinate attention and residual error network mechanism technology comprises the following steps:
Firstly, an image acquisition module acquires a low-resolution image through a preprocessing unit, and forms an image data set after standardized processing, and the image acquisition module is connected with an analysis and re-pickup module through a network;
Step two, the analysis and re-picking module numbers the image dataset according to the characteristics of the image dataset, inputs the image dataset into a residual error network unit for characteristic extraction, calculates and generates a residual error characteristic group Catz, and transmits the residual error characteristic group Catz to a coordinate attention unit through a network;
Thirdly, the coordinate attention unit calculates and generates an attention feature group Zytz according to the residual feature group Catz, and transmits the attention feature group Zytz to the up-sampling unit through a network;
Step four, the up-sampling unit samples and generates a high-resolution image according to the attention feature group Zytz, and transmits the high-resolution image to the vision processing module through a network;
And fifthly, the vision processing module performs optimization processing for two times according to the high-resolution image, and outputs the image through network connection.
2. The single patch image super-resolution re-picking algorithm based on the coordinate attention and residual network mechanism technology as claimed in claim 1, wherein the single patch image super-resolution re-picking algorithm is characterized in that: the preprocessing unit performs standardization processing through a bilateral filter, performs pixel denoising on a single or a plurality of low-resolution images, and protects image edges and texture areas to form an image data set.
3. The single patch image super-resolution re-picking algorithm based on the coordinate attention and residual network mechanism technology as claimed in claim 2, wherein the single patch image super-resolution re-picking algorithm is characterized in that: the analysis and re-picking module comprises a residual error network unit, a coordinate attention unit and an up-sampling unit, and is used for numbering the image data set according to the characteristics of the image data set, wherein the image data set is numbered as TX 1、TX2、TX3、…TXn and is transmitted to the residual error network unit through a network.
4. The single patch image super-resolution re-picking algorithm based on the coordinate attention and residual network mechanism technology as claimed in claim 1, wherein the single patch image super-resolution re-picking algorithm is characterized in that: the residual error network unit is provided with a deep convolutional neural network, and the deep convolutional neural network comprises a convolutional layer JU, an activation function JH, a batch normalization layer GY, a pooling layer CH and a full connection layer QL.
5. The single patch image super-resolution re-picking algorithm based on the coordinate attention and residual network mechanism technology according to claim 4, wherein the single patch image super-resolution re-picking algorithm is characterized in that: the residual network unit inputs the image data set into the deep convolutional neural network, performs feature extraction between each residual block in a jump connection manner, calculates and generates a residual feature group Catz, and the calculation formula is as follows:
In the formula, catz represents a residual characteristic group, pi JUTXn represents characteristic data obtained by multiplying a convolution kernel of a convolution layer with image data point by point and summing the result, maxJH (0, TX n) represents that the image data is substituted into an activated function algorithm to obtain nonlinear complex characteristics, sigma GYTXn mu represents that when the image data amount is large, a batch normalization layer stabilizing network training gradient is substituted, TX n-CH|e-n represents that the image data is substituted into a pooling layer to obtain key characteristic data, and U QLTXn+∞ represents that the full connection layer integrates the image data characteristics extracted by a plurality of residual blocks and finally outputs the key characteristic data.
6. The single patch image super-resolution re-picking algorithm based on the coordinate attention and residual network mechanism technology as claimed in claim 1, wherein the single patch image super-resolution re-picking algorithm is characterized in that: the coordinate attention unit is provided with a reference plane coordinate model ZB and a reference mapping function model YH.
7. The single patch image super-resolution re-picking algorithm based on the coordinate attention and residual network mechanism technology as claimed in claim 6, wherein the single patch image super-resolution re-picking algorithm is characterized in that: the coordinate attention unit calculates and generates an attention feature group Zytz according to the residual feature group Catz, and the calculation formula is as follows:
wherein Zytz denotes the set of attention features, Representing attention weight images calculated by substituting residual feature set data into a reference mapping function and a reference plane coordinate model transformation, including attention coordinate coding and attention feature aggregation,/>Feature weighting data representing suppression of secondary features and highlighting important features for the attention weighted image.
8. The single patch image super-resolution re-picking algorithm based on the coordinate attention and residual network mechanism technology as claimed in claim 7, wherein: the up-sampling unit increases the number of pixels of the image by a nearest neighbor interpolation method according to the attention feature group Zytz, and samples the image to generate a high-resolution image.
9. The single patch image super-resolution re-picking algorithm based on the coordinate attention and residual network mechanism technology as claimed in claim 8, wherein: the visual processing module comprises a color correction unit and a sharpening and brightening unit, wherein the color correction unit is connected with computer image software according to a high-resolution image to perform first optimization processing and is transmitted to the sharpening and brightening unit through a network.
10. The single patch image super-resolution re-picking algorithm based on the coordinate attention and residual network mechanism technology as claimed in claim 9, wherein: and the sharpening and brightening unit is connected with computer image software for performing secondary optimization processing according to the high-resolution image after color correction, and outputs an optimal image through a network.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202410195789.4A CN118052711A (en) | 2024-02-22 | 2024-02-22 | Single-patch image super-resolution recapture algorithm based on coordinate attention and residual error network mechanism technology |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202410195789.4A CN118052711A (en) | 2024-02-22 | 2024-02-22 | Single-patch image super-resolution recapture algorithm based on coordinate attention and residual error network mechanism technology |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN118052711A true CN118052711A (en) | 2024-05-17 |
Family
ID=91047841
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202410195789.4A Pending CN118052711A (en) | 2024-02-22 | 2024-02-22 | Single-patch image super-resolution recapture algorithm based on coordinate attention and residual error network mechanism technology |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN118052711A (en) |
-
2024
- 2024-02-22 CN CN202410195789.4A patent/CN118052711A/en active Pending
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN109859147B (en) | Real image denoising method based on generation of antagonistic network noise modeling | |
| CN112233038B (en) | True image denoising method based on multi-scale fusion and edge enhancement | |
| CN107123089B (en) | Remote sensing image super-resolution reconstruction method and system based on depth convolution network | |
| CN113222822B (en) | Hyperspectral image super-resolution reconstruction method based on multi-scale transformation | |
| CN114757832B (en) | Face super-resolution method and device based on cross convolution attention pair learning | |
| CN113284051B (en) | Face super-resolution method based on frequency decomposition multi-attention machine system | |
| CN111127336A (en) | Image signal processing method based on self-adaptive selection module | |
| CN113096017A (en) | Image super-resolution reconstruction method based on depth coordinate attention network model | |
| CN111932461A (en) | Convolutional neural network-based self-learning image super-resolution reconstruction method and system | |
| CN110136075B (en) | Remote sensing image defogging method for generating countermeasure network based on edge sharpening cycle | |
| CN110533614B (en) | Underwater image enhancement method combining frequency domain and airspace | |
| Han et al. | Underwater image enhancement based on a spiral generative adversarial framework | |
| CN111861886B (en) | Image super-resolution reconstruction method based on multi-scale feedback network | |
| CN112598602A (en) | Mask-based method for removing Moire of deep learning video | |
| CN112163998A (en) | Single-image super-resolution analysis method matched with natural degradation conditions | |
| CN115393191A (en) | Method, device and equipment for reconstructing super-resolution of lightweight remote sensing image | |
| CN115797176A (en) | Image super-resolution reconstruction method | |
| CN115222614A (en) | Priori-guided multi-degradation-characteristic night light remote sensing image quality improving method | |
| CN109993701B (en) | Depth map super-resolution reconstruction method based on pyramid structure | |
| CN115511708A (en) | Depth map super-resolution method and system based on uncertainty perception feature transmission | |
| CN113379606B (en) | Face super-resolution method based on pre-training generation model | |
| Zhang et al. | Enhanced visual perception for underwater images based on multistage generative adversarial network | |
| CN111899166A (en) | Medical hyperspectral microscopic image super-resolution reconstruction method based on deep learning | |
| CN116823610A (en) | Deep learning-based underwater image super-resolution generation method and system | |
| CN118052711A (en) | Single-patch image super-resolution recapture algorithm based on coordinate attention and residual error network mechanism technology |
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 |