CN115936985A - Image super-resolution reconstruction method based on high-order degradation cycle generation countermeasure network - Google Patents
Image super-resolution reconstruction method based on high-order degradation cycle generation countermeasure network Download PDFInfo
- Publication number
- CN115936985A CN115936985A CN202211539202.4A CN202211539202A CN115936985A CN 115936985 A CN115936985 A CN 115936985A CN 202211539202 A CN202211539202 A CN 202211539202A CN 115936985 A CN115936985 A CN 115936985A
- Authority
- CN
- China
- Prior art keywords
- image
- resolution
- network
- 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.)
- Pending
Links
- 238000006731 degradation reaction Methods 0.000 title claims abstract description 41
- 238000000034 method Methods 0.000 title claims abstract description 39
- 230000015556 catabolic process Effects 0.000 title claims abstract description 38
- 238000013507 mapping Methods 0.000 claims abstract description 18
- 238000012549 training Methods 0.000 claims abstract description 13
- 238000013527 convolutional neural network Methods 0.000 claims abstract description 5
- 230000006870 function Effects 0.000 claims description 19
- 230000006835 compression Effects 0.000 claims description 17
- 238000007906 compression Methods 0.000 claims description 17
- 239000013598 vector Substances 0.000 claims description 10
- 125000004122 cyclic group Chemical group 0.000 claims description 9
- 238000012545 processing Methods 0.000 claims description 7
- 238000005070 sampling Methods 0.000 claims description 6
- 238000006243 chemical reaction Methods 0.000 claims description 5
- 238000013528 artificial neural network Methods 0.000 claims description 3
- 230000009467 reduction Effects 0.000 claims description 2
- 230000003042 antagnostic effect Effects 0.000 claims 1
- 230000007850 degeneration Effects 0.000 claims 1
- 230000000694 effects Effects 0.000 abstract description 5
- 238000010586 diagram Methods 0.000 description 5
- 230000015572 biosynthetic process Effects 0.000 description 2
- 238000002474 experimental method Methods 0.000 description 2
- 238000003786 synthesis reaction Methods 0.000 description 2
- 238000012360 testing method Methods 0.000 description 2
- 238000009825 accumulation Methods 0.000 description 1
- 230000009286 beneficial effect Effects 0.000 description 1
- 239000003054 catalyst Substances 0.000 description 1
- 239000008358 core component Substances 0.000 description 1
- 238000013135 deep learning Methods 0.000 description 1
- 230000000593 degrading effect Effects 0.000 description 1
- 238000013461 design Methods 0.000 description 1
- 238000001514 detection method Methods 0.000 description 1
- 238000002059 diagnostic imaging Methods 0.000 description 1
- 238000003384 imaging method Methods 0.000 description 1
- 230000006872 improvement Effects 0.000 description 1
- 238000012986 modification Methods 0.000 description 1
- 230000004048 modification Effects 0.000 description 1
- 238000012544 monitoring process Methods 0.000 description 1
- 230000008569 process Effects 0.000 description 1
Images
Classifications
-
- Y—GENERAL TAGGING OF NEW TECHNOLOGICAL DEVELOPMENTS; GENERAL TAGGING OF CROSS-SECTIONAL TECHNOLOGIES SPANNING OVER SEVERAL SECTIONS OF THE IPC; TECHNICAL SUBJECTS COVERED BY FORMER USPC CROSS-REFERENCE ART COLLECTIONS [XRACs] AND DIGESTS
- Y02—TECHNOLOGIES OR APPLICATIONS FOR MITIGATION OR ADAPTATION AGAINST CLIMATE CHANGE
- Y02T—CLIMATE CHANGE MITIGATION TECHNOLOGIES RELATED TO TRANSPORTATION
- Y02T10/00—Road transport of goods or passengers
- Y02T10/10—Internal combustion engine [ICE] based vehicles
- Y02T10/40—Engine management systems
Landscapes
- Image Processing (AREA)
Abstract
The invention discloses an image super-resolution reconstruction method for generating a countermeasure network based on a high-order degradation Cycle, which comprises the steps of firstly generating a low-resolution image from a high-resolution image by using a high-order image degradation model, forming an image pair, then constructing a Cycle generation network model, utilizing a generator to realize the generation from the low-resolution image to the high-resolution image and then judging, constructing a network loss function, ensuring the stability of the established mapping relation through a classical GAN network loss function and a Cycle loss function which is not changed in Cycle, finally training the network model, configuring various hyper-parameters, and enabling the network to be converged to obtain a better generation effect. Compared with the traditional hyper-resolution reconstruction method based on image interpolation and the method based on the general convolutional neural network, the method has better performance of hyper-resolution reconstruction of the image.
Description
Technical Field
The invention belongs to the technical field of image processing, relates to a super-resolution reconstruction method for weak and small image targets, and particularly relates to an image super-resolution reconstruction method for generating a countermeasure network based on high-order degradation cycle.
Background
In the field of application of a large number of electronic images, high resolution means that the pixel density of the images is high, more detail information can be provided, the cost for improving the low-resolution images from a hardware level is huge, the imaging model is analyzed by using a software algorithm, and an over-resolution model reconstructed from the low-resolution images into the high-resolution images is constructed, so that the method has considerable flexibility and adaptability, and has wide application prospects in the fields of remote sensing military application, auxiliary meteorological detection, geographical environment analysis, medical imaging, video monitoring and the like.
When a hyper-resolution reconstruction algorithm model is constructed, a network structure of a high-order degradation model and a classical Cycle-GAN is usually adopted, the Cycle-GAN does not require that a training data set is a strictly aligned image pair, the mapping relation between a low-resolution image and a high-resolution image is more stable by using paired image pairs, and the data set in the prior art has fewer samples.
Disclosure of Invention
Aiming at the problems that image target pixels in some fields are too small to be identified, the invention provides an image super-resolution reconstruction method for generating a countermeasure network based on high-order degradation cycle.
The technical scheme adopted by the invention for solving the technical problems is as follows: an image super-resolution reconstruction method based on a high-order degradation cycle generation countermeasure network comprises the following steps:
step S1, multiplexing the first-order degradation model to obtain a second-order degradation model, and degrading the high-resolution image by adopting the second-order degradation model: the first-order degradation model firstly convolves an original high-resolution image y with a fuzzy function k, then performs down-sampling operation of a scale factor r, finally adds noise n to obtain a low-resolution image x, and obtains a low-resolution image x through a degradation formula
Compressing the JPEG image to obtain a low-resolution image x, wherein r represents a scale factor, ↓, and r a down-sampling factor is represented by a factor, n represents a noise operation value] JPEG The compression processing is performed on the obtained result by using a JPEG method, and the low-resolution image x and the high-resolution image y are used as an image pair LR and HR;
s2, establishing a hyper-resolution mapping model from a low-resolution image x to a high-resolution image y by using a Cycle-GAN network based on a generator of a cyclic loss function cyclloss; the generator is composed of an encoder, a converter and a decoder, a high-resolution image y is generated from a low-resolution image x, the generated image is judged by using a discriminator, the Cycle-GAN network trains two pairs of generator-discriminator models to convert the image from one field to the other field, and the hyper-resolution mapping model comprises two mapping functions G: x- - > Y and F: y- - > X, and associated countermeasure discriminators D _ Y and D _ X; d _ Y encourages G to convert the low-resolution image X into a high-resolution image Y, and the discriminator D _ X encourages G to convert the high-resolution image Y into the low-resolution image X;
step S3, two Cycle coordination loss functions are introduced to further standardize the mapping to construct a Cycle generation network model, the converted image can be ensured to return to the state before processing after reverse conversion, and the stability number of the established mapping relation is ensured through the classical GAN network loss function and the Cycle invariant Cycle loss function: l (G, F, D) X ，D Y )＝L GAN (G，D Y ，X，Y)+L GAN (D X ，F，Y，X)+λL cyc (G, F), wherein:
s4, training a cyclic generation network model, configuring various hyper-parameters, and enabling the network to be converged to obtain a good generation effect: the input image obtained from the domain DA is passed to a first generator GeneratorA → B, whose task is to convert a given image from the domain DA into an image in the target domain DB, and then the newly generated image is passed to another generator GeneratorB → a, whose task is to convert back in the original domain DA an image CyclicA similar to the original input image.
According to the image super-resolution reconstruction method based on the high-order degradation cycle generation countermeasure network, a fuzzy kernel k in the step S1 uses an isotropic Gaussian fuzzy kernel and an anisotropic Gaussian fuzzy kernel; one of 3 interpolation modes of cubic interpolation, bilinear interpolation or regional interpolation is randomly selected to reduce the scale factor r; and simultaneously adding Gaussian noise and noise distributed from Poisson in the noise operation n, and selecting the number of channels of the image to be super-resolution to add noise to the color image and add noise to the gray image.
The image super-resolution reconstruction method based on the high-order degradation cycle generation countermeasure network comprises the step S1] JPEG Is obtained by the reaction of a catalyst selected from [0,100 ]]The compression quality is selected from the range to carry out JPEG compression on the image, wherein 0 represents the worst quality after compression, and 100 represents the best quality after compression.
According to the image super-resolution reconstruction method based on the high-order degradation cycle generation countermeasure network, a first-order degradation model adopts a sinc filter in the last step of fuzzy processing and synthesis, and the sequence of the last sinc filter and JPEG compression is randomly exchanged, so that a larger degradation space is obtained.
Furthermore, the encoder coding is to extract features from the input image by using the convolutional neural network, and then compress the image into 256 feature vectors of 64 × 64.
Further, the converter converts the image feature vector in the DA domain into the feature vector in the DB domain by combining the dissimilar features of the image using 6-layer Reset modules, each Reset module being a neural network layer composed of two convolutional layers.
Further, the decoder decoding is to use the deconvolution layer deconvolution to complete the work of restoring low-level features from the feature vectors, and finally, a generated image is obtained.
Further, the discriminator takes an image as input and tries to predict whether it is the original image or the output image of the generator.
The invention has the beneficial effects that:
the method carries out the super-resolution reconstruction on the image, the adopted HD-Cycle-GAN network model fuses the structural characteristics of a high-order degradation model and the Cycle-GAN, the high-order degradation model degrades the high-resolution image into an unknown and complex low-resolution image through a blind super-resolution technology, a sufficient low-resolution image pair is provided for the training of the network model, and then a stable mapping relation between the low-resolution image and the high-resolution image is established through a cyclic confrontation generation network, so that the super-resolution reconstruction enhancement of the low-resolution image is realized.
Compared with the traditional hyper-resolution reconstruction method based on image interpolation and the method based on the general convolutional neural network, the method has better performance of hyper-resolution reconstruction of the image.
Drawings
FIG. 1 is a block diagram of an HD-Cycle-GAN network according to the present invention;
FIG. 2 is a flow chart of the high resolution image block training and online reasoning algorithm of the present invention;
FIG. 3 is a block diagram of a cycle generating network model according to the present invention;
fig. 4 is a configuration diagram of an encoder, a converter, and a decoder;
FIG. 5 is a block diagram of the discriminator of the present invention;
FIG. 6 is a comparison graph of the effect of the super-resolution reconstruction performed by the method of the present invention.
Detailed Description
To further illustrate the objects and technical solutions of the present invention, the following detailed description will be given with reference to specific embodiments of the accompanying drawings.
The invention discloses an image super-resolution reconstruction method for generating a countermeasure network based on high-order degradation cycle.
Referring to fig. 1, the present invention degrades a high resolution image using a second order degradation model, which is a multiplexing of a classical first order degradation model.
(1) For a first-order degradation model, generally, a high-resolution image y is firstly convolved with a fuzzy kernel k, then down-sampling operation of a scale factor r is carried out, and noise n is added to obtain a low-resolution image x. The degradation formula is
Where y denotes the original image, k denotes the blur function, ↓ r Represents a downsampling factor, n represents noise] JPEG The obtained result is compressed by the JPEG method. Therefore, the meaning of the whole equation is to perform down-sampling after a high-resolution image is processed by blurring, then add noise, and finally obtain a low-resolution image x by JPEG compression.
It can be seen from fig. 1 that the degradation model for each layer comprises four degradation processes, and each degradation process has a specific degradation operation.
For blur kernel k, the present invention uses gaussian blur kernels of isotropy (isotropic) and anisotropy (anisotropic).
2, for the reduction operation r, when the operation is executed, the invention randomly selects one of 3 interpolation modes, namely cubic interpolation, bilinear interpolation and regional interpolation.
3, for the add noise operation n, the present invention adds both gaussian noise and noise that obeys poisson distribution. Meanwhile, according to the number of channels of the image to be super-resolved, the operation of adding noise is divided into adding noise to the color image and adding noise to the gray image.
4, JPEG compression, wherein 0 represents the worst quality after compression and 100 represents the best quality after compression.
And 5, adopting a sinc filter in the last step of the fuzzy processing and the synthesis, and randomly exchanging the sequence of the last sinc filter and JPEG compression to obtain larger degradation space.
(2) After obtaining the image pair of LR (low resolution) image and HR (high resolution) image, a hyper-resolution mapping model from LR to HR is established below using a Cycle-GAN network, the main design of which is to train two pairs of generator-discriminator models to convert the image from one domain to the other, in which process cyclic consistency is required, i.e. after applying the generators sequentially, an image similar to the original L1 loss should be obtained. Therefore, a cyclic loss function (cyclloss) is needed, which ensures that the generator establishes a stable mapping between the image pairs.
The hyper-resolution mapping model comprises two mapping functions G: x- - > Y and F: y- - > X, and associated countermeasure discriminators D _ Y and D _ X; discriminator D _ Y encourages G to convert low resolution image X to high resolution image Y and vice versa.
(3) In order to further standardize mapping, the invention also introduces two cycle coordination loss functions to ensure that the converted image can return to the state before processing after inverse conversion, and the flow of the high-resolution image block training and online reasoning algorithm is shown in fig. 2: the loss function of Cycle-GAN is divided into two parts: one part is GAN loss and the other part is Cycle loss, and the total loss function is formulated as
L(G，F，D X ，D Y )＝L GAN (G，D Y ，X，Y)+L GAN (D X ，F，Y，X)+λL cyc (G, F), wherein,
and then optimizing the total loss:
the specific framework of the cycle generation network model is shown in fig. 3: the model takes an input image from the domain DA, which is passed to a first generator GeneratorA → B, the task of which is to convert a given image from the domain DA into an image in the target domain DB. This newly generated image is then passed to another generator GeneratorB → a, whose task is to convert back to image CyclicA in the original domain DA, this output image must be similar to the original input image to ensure stability of the image mapping relationship. The core component of the method comprises a generator and a discriminator, wherein the generator is composed of an encoder, a converter and a decoder, and the structure is shown in figure 4.
And (3) encoding by an encoder: the first step is to extract features from the input image using a convolutional neural network. The image is compressed into 256 64 by 64 feature vectors.
The converter converts: the feature vectors of the image in the DA domain are converted to feature vectors in the DB domain by combining the dissimilar features of the images. The goal of preserving the original image features at the same time during conversion can be achieved using 6-layer Reset modules, each of which is a neural network layer consisting of two convolutional layers.
Decoding by a decoder: and (4) finishing the work of restoring low-level features from the feature vector by using a deconvolution (deconvolution), and finally obtaining a generated image.
The discriminator discriminates: an image is taken as input and an attempt is made to predict whether it is the original image or the output image of the generator. As shown in fig. 5, the discriminator itself belongs to a convolutional network, and it is necessary to extract features from the image and then determine whether the extracted features belong to a particular class by adding a convolutional layer that produces a one-dimensional output.
The method comprises the steps that firstly, a high-resolution image is subjected to high-order image degradation model to generate a low-resolution image, and an LR and SR image pair is formed; designing and constructing a cyclic generation network model, constructing a generator by utilizing an encoder, a converter and a decoder to realize the generation from a low-resolution image to a high-resolution image, and judging the generated image by utilizing a discriminator; thirdly, constructing a network loss function, and ensuring the stability of the established mapping relation through a classical GAN network loss function and a Cycle loss function with constant circulation; and fourthly, training a network model, and configuring various hyper-parameters so that the network can be converged to obtain a better generation effect.
The method selects various data sets including RealSR-Camon, realSR-Nikon, DRearSR, DPED-iphone, OST300, imageNeval and ADE20Kval as training test sets, randomly crawls 100 pictures from the Internet, divides the training sets and the test sets according to the ratio of 9: 1, verifies the super-resolution reconstruction function of the network, and carries out comparison experiments under the index of a Natural Image Quality Evaluator (NIQE) by methods of Bicubic, ESRGAN, DAN, realSR, CDC, BSRGAN and the like, thereby verifying the method advancement of the invention. The following table shows the NIQE scores for the invention and methods on each data set:
| Bicubic | ESRGAN | DAN | RealSR | CDC | BSRGAN | HD-CycleGAN | |
| RealSR-Camon | 6.1269 | 6.7715 | 6.5282 | 6.8692 | 6.1488 | 5.7489 | 4.4312 |
| RealSR-Nikon | 6.3607 | 6.7480 | 6.6063 | 6.7390 | 6.3265 | 5.9920 | 5.0138 |
| DRealSR | 6.5766 | 8.6335 | 7.0720 | 7.7213 | 6.6359 | 6.1362 | 4.9156 |
| DPED-iphone | 6.0121 | 5.7363 | 6.1414 | 5.5855 | 6.2738 | 5.9906 | 5.1326 |
| OST300 | 4.4440 | 3.5245 | 5.0232 | 4.5715 | 4.7441 | 4.1662 | 2.8192 |
| ImageNet val | 7.4985 | 3.6474 | 6.0932 | 3.8303 | 7.0441 | 4.3528 | 4.7438 |
| ADE20K val | 7.5239 | 3.6905 | 6.3839 | 3.4102 | 6.9219 | 3.9434 | 3.8728 |
as can be seen from the table above, the HD-Cycle-GAN algorithm adopted by the invention obtains the best NIQE score by comparing with the data sets such as Bicubic, ESRGAN, DAN, realSR, CDC, BSRGAN and the like on RealSR-Camon, realSR-Nikon, DRealSR, DPED-iphone, OST300 and the like, and the network has better over-resolution reconstruction effect on low-resolution images.
FIG. 6 is a comparison of the results of the hyper-resolution reconstruction of images randomly crawled from the web using the method of the present invention. It can be easily found from the contrast diagram that the present invention can effectively perform the enhancement of the super-resolution reconstruction for the LR images with low quality, and the result is the result of using four times of super-resolution, that is, for the 100 × 100 image, the resolution is expanded to 200 × 200, and the pixel quality is also enhanced.
The data set and platform for this experiment were as follows: the data set adopts high-definition video images downloaded from the internet, about 10000 frames are extracted from the high-definition video images, then the corresponding low-resolution images are obtained through high-order image degradation, therefore, the image pair required by training can be obtained, if the data accumulation of the low-resolution image pair exists in the earlier stage, the high-order degradation process can be skipped, the network training is carried out, and the software and hardware platform comprises:
intel (R) Core (TM) i7-8700CPU@3.20GHz*12; GPU: forceGTX1080Ti; operating the system: ubuntu16.04LTS; a deep learning framework: a pytorech. Training the network, setting the initial learning rate of the weight value to be 0.001, setting the attenuation coefficient to be 0.0005, adopting a cosinlr learning rate strategy, and setting the epoch to be 145.
It will be readily understood by those skilled in the art that the foregoing is only a preferred embodiment of this invention and is not intended to limit the invention, and that any modification, equivalent replacement or improvement made within the spirit and scope of the invention shall be included therein.
Claims (8)
1. An image super-resolution reconstruction method based on a high-order degradation cycle generation countermeasure network is characterized in that: comprises the following steps
S1, convolving an original high-resolution image y and a fuzzy kernel k through a first-order degradation model, then performing down-sampling operation of a scale factor r, and finally performing degradation formula
Compressing the JPEG image to obtain a low-resolution image x, wherein r represents a scale factor, ↓, and r a down-sampling factor is represented by a factor, n represents a noise operation value] JPEG The image compression method is characterized in that the obtained result is compressed by a JPEG mode, and a low-resolution image x and a high-resolution image y are used as an image pair;
s2, establishing a hyper-resolution mapping model from a low-resolution image x to a high-resolution image y based on a generator of a cyclic loss function cyclicloss; the generator consists of an encoder, a converter and a decoder, a high-resolution image Y is generated from a low-resolution image X, the generated image is judged by a discriminator, and the hyper-resolution mapping model comprises mapping functions G, X, Y and F, Y, X and related confrontation discriminators D _ Y and D _ X; d _ Y encourages G to convert the low-resolution image X into a high-resolution image Y, and D _ X encourages G to convert the high-resolution image Y into the low-resolution image X;
step S3, constructing a cyclic generation network model based on a loss function, and ensuring that the converted image can return to the state before processing after reverse conversion:
L(G,F,D X ,D Y )＝L GAN (G,D Y ,X,Y)+L GAN (D X ,F,Y,X)+λL cyc (G, F), wherein:
by the formula G * ,
Optimizing total losses；
S4, training a cyclic generation network model: the acquired input image is passed to a first generator GeneratorA → B, the given image is converted into an image in the target domain, then the newly generated image is passed to another generator GeneratorB → a, converted back to an image CyclicA similar to the original input image.
2. The image super-resolution reconstruction method based on the countermeasure network generated by the high-order degradation loop is characterized in that the blur kernel k in the step S1 uses a Gaussian blur kernel with isotropy and anisotropy; carrying out reduction operation on the scale factor r by adopting a cubic interpolation mode, a bilinear interpolation mode or a regional interpolation mode; and simultaneously adding Gaussian noise and noise distributed from Poisson in the noise operation n, and selecting the number of channels of the image to be super-resolution to add noise to the color image and add noise to the gray image.
3. The method for reconstructing super-resolution image based on generation of countermeasure network by higher-order degeneration loop according to claim 2, wherein the value of [ 2 ], [ 2 ]] JPEG Is obtained by reacting from [0,100 ]]The compression quality is selected from the range to carry out JPEG compression on the image, wherein 0 represents the worst quality after compression, and 100 represents the best quality after compression.
4. The method for reconstructing the super-resolution image based on the high-order degradation loop generation countermeasure network of claim 3, wherein a sinc filter is adopted in the last step of the first-order degradation model, and the order of the last sinc filter and JPEG compression is randomly exchanged.
5. The method for image super-resolution reconstruction based on the antagonistic network generated by the higher-order degenerate cycle of claim 4, wherein the encoder coding is to extract features from the input image by using the convolutional neural network, and compress the image into 256 feature vectors of 64 × 64.
6. The method for reconstructing super-resolution images based on generation of countermeasure networks by high-order degradation loop as claimed in claim 4, wherein said converter converts the eigenvectors of the images in DA domain into eigenvectors in DB domain by combining the dissimilar characteristics of the images using 6 layers of Reset modules, each Reset module being a neural network layer composed of two convolutional layers.
7. The method as claimed in claim 4, wherein the decoder decoding is to use deconvolution to complete the work of recovering low-level features from the feature vectors, and finally obtain the generated image.
8. The method for reconstructing the super-resolution image based on the generation of the countermeasure network by the higher order degradation loop as claimed in claim 4, wherein the discriminator is to take an image as an input and try to predict the image as an original image or an output image of the generator.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202211539202.4A CN115936985A (en) | 2022-12-01 | 2022-12-01 | Image super-resolution reconstruction method based on high-order degradation cycle generation countermeasure network |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202211539202.4A CN115936985A (en) | 2022-12-01 | 2022-12-01 | Image super-resolution reconstruction method based on high-order degradation cycle generation countermeasure network |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN115936985A true CN115936985A (en) | 2023-04-07 |
Family
ID=86551613
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202211539202.4A Pending CN115936985A (en) | 2022-12-01 | 2022-12-01 | Image super-resolution reconstruction method based on high-order degradation cycle generation countermeasure network |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN115936985A (en) |
Cited By (5)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN116503258A (en) * | 2023-06-20 | 2023-07-28 | 中国科学院生物物理研究所 | Super-resolution computing imaging method, device, electronic equipment and storage medium |
| CN117036161A (en) * | 2023-06-13 | 2023-11-10 | 河海大学 | Dam defect recovery method based on generation type countermeasure network |
| CN117197627A (en) * | 2023-09-13 | 2023-12-08 | 长春理工大学 | Multi-mode image fusion method based on high-order degradation model |
| CN117196957A (en) * | 2023-11-03 | 2023-12-08 | 广东省电信规划设计院有限公司 | Image resolution conversion method and device based on artificial intelligence |
| CN117745725A (en) * | 2024-02-20 | 2024-03-22 | 阿里巴巴达摩院(杭州)科技有限公司 | Image processing method, image processing model training method, three-dimensional medical image processing method, computing device, and storage medium |
Citations (5)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN109543548A (en) * | 2018-10-26 | 2019-03-29 | 桂林电子科技大学 | A kind of face identification method, device and storage medium |
| US20200111194A1 (en) * | 2018-10-08 | 2020-04-09 | Rensselaer Polytechnic Institute | Ct super-resolution gan constrained by the identical, residual and cycle learning ensemble (gan-circle) |
| CN111382601A (en) * | 2018-12-28 | 2020-07-07 | 河南中原大数据研究院有限公司 | Illumination face image recognition preprocessing system and method for generating confrontation network model |
| CN111915513A (en) * | 2020-07-10 | 2020-11-10 | 河海大学 | Image denoising method based on improved adaptive neural network |
| CN113643183A (en) * | 2021-10-14 | 2021-11-12 | 湖南大学 | Non-matching remote sensing image weak supervised learning super-resolution reconstruction method and system |
-
2022
- 2022-12-01 CN CN202211539202.4A patent/CN115936985A/en active Pending
Patent Citations (5)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US20200111194A1 (en) * | 2018-10-08 | 2020-04-09 | Rensselaer Polytechnic Institute | Ct super-resolution gan constrained by the identical, residual and cycle learning ensemble (gan-circle) |
| CN109543548A (en) * | 2018-10-26 | 2019-03-29 | 桂林电子科技大学 | A kind of face identification method, device and storage medium |
| CN111382601A (en) * | 2018-12-28 | 2020-07-07 | 河南中原大数据研究院有限公司 | Illumination face image recognition preprocessing system and method for generating confrontation network model |
| CN111915513A (en) * | 2020-07-10 | 2020-11-10 | 河海大学 | Image denoising method based on improved adaptive neural network |
| CN113643183A (en) * | 2021-10-14 | 2021-11-12 | 湖南大学 | Non-matching remote sensing image weak supervised learning super-resolution reconstruction method and system |
Cited By (9)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN117036161A (en) * | 2023-06-13 | 2023-11-10 | 河海大学 | Dam defect recovery method based on generation type countermeasure network |
| CN116503258A (en) * | 2023-06-20 | 2023-07-28 | 中国科学院生物物理研究所 | Super-resolution computing imaging method, device, electronic equipment and storage medium |
| CN116503258B (en) * | 2023-06-20 | 2023-11-03 | 中国科学院生物物理研究所 | Super-resolution computing imaging method, device, electronic equipment and storage medium |
| CN117197627A (en) * | 2023-09-13 | 2023-12-08 | 长春理工大学 | Multi-mode image fusion method based on high-order degradation model |
| CN117197627B (en) * | 2023-09-13 | 2024-03-01 | 长春理工大学 | Multi-mode image fusion method based on high-order degradation model |
| CN117196957A (en) * | 2023-11-03 | 2023-12-08 | 广东省电信规划设计院有限公司 | Image resolution conversion method and device based on artificial intelligence |
| CN117196957B (en) * | 2023-11-03 | 2024-03-22 | 广东省电信规划设计院有限公司 | Image resolution conversion method and device based on artificial intelligence |
| CN117745725A (en) * | 2024-02-20 | 2024-03-22 | 阿里巴巴达摩院(杭州)科技有限公司 | Image processing method, image processing model training method, three-dimensional medical image processing method, computing device, and storage medium |
| CN117745725B (en) * | 2024-02-20 | 2024-05-14 | 阿里巴巴达摩院(杭州)科技有限公司 | Image processing method, image processing model training method, three-dimensional medical image processing method, computing device, and storage medium |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN111768342B (en) | Human face super-resolution method based on attention mechanism and multi-stage feedback supervision | |
| CN115936985A (en) | Image super-resolution reconstruction method based on high-order degradation cycle generation countermeasure network | |
| CN109903228B (en) | Image super-resolution reconstruction method based on convolutional neural network | |
| Yu et al. | Deep iterative down-up cnn for image denoising | |
| Cai et al. | Pix2nerf: Unsupervised conditional p-gan for single image to neural radiance fields translation | |
| CN109035142B (en) | Satellite image super-resolution method combining countermeasure network with aerial image prior | |
| CN109949222B (en) | Image super-resolution reconstruction method based on semantic graph | |
| CN113284051B (en) | Face super-resolution method based on frequency decomposition multi-attention machine system | |
| CN111080567A (en) | Remote sensing image fusion method and system based on multi-scale dynamic convolution neural network | |
| CN109949217B (en) | Video super-resolution reconstruction method based on residual learning and implicit motion compensation | |
| CN111861886B (en) | Image super-resolution reconstruction method based on multi-scale feedback network | |
| CN116681584A (en) | Multistage diffusion image super-resolution algorithm | |
| Min et al. | Blind deblurring via a novel recursive deep CNN improved by wavelet transform | |
| Rivadeneira et al. | Thermal image super-resolution challenge-pbvs 2021 | |
| CN115880158B (en) | Blind image super-resolution reconstruction method and system based on variation self-coding | |
| Dai et al. | Image super-resolution via residual block attention networks | |
| CN114972024A (en) | Image super-resolution reconstruction device and method based on graph representation learning | |
| CN112927137A (en) | Method, device and storage medium for acquiring blind super-resolution image | |
| CN112884650A (en) | Image mixing super-resolution method based on self-adaptive texture distillation | |
| Zha et al. | The power of triply complementary priors for image compressive sensing | |
| CN112200752B (en) | Multi-frame image deblurring system and method based on ER network | |
| Wen et al. | The power of complementary regularizers: Image recovery via transform learning and low-rank modeling | |
| Yu et al. | Scale-aware frequency attention network for super-resolution | |
| CN116188272B (en) | Two-stage depth network image super-resolution reconstruction method suitable for multiple fuzzy cores | |
| CN115439849B (en) | Instrument digital identification method and system based on dynamic multi-strategy GAN 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 |