CN109903226B - Image super-resolution reconstruction method based on symmetric residual convolution neural network - Google Patents
Image super-resolution reconstruction method based on symmetric residual convolution neural network Download PDFInfo
- Publication number
- CN109903226B CN109903226B CN201910093918.8A CN201910093918A CN109903226B CN 109903226 B CN109903226 B CN 109903226B CN 201910093918 A CN201910093918 A CN 201910093918A CN 109903226 B CN109903226 B CN 109903226B
- Authority
- CN
- China
- Prior art keywords
- image
- residual
- resolution
- residual block
- training
- 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
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
Abstract
The invention discloses an image super-resolution reconstruction method based on a symmetrical residual convolution neural network, which specifically comprises the following steps: selecting a training sample and a test sample; preprocessing a data set; extracting features; feature fusion; and (5) reconstructing an image. The technical scheme of the invention solves the problems of multiple parameters of the deep reconstruction network, great training difficulty and easy gradient disappearance, and realizes the full utilization of information in the block and extraction of richer local features by establishing symmetrical short-jump connection in the residual block; long jump connection is established outside the residual block, so that global feature fusion is realized; the image reconstructed by the method has clearer texture, richer details, better subjective visual effect and wide development prospect in the aspect of image processing application.
Description
Technical Field
The invention relates to the technical field of image processing, in particular to an image super-resolution reconstruction method based on a symmetrical residual convolution neural network.
Background
Image Super-resolution (SR) reconstruction is an important branch in image restoration, and is a method for restoring a corresponding High Resolution (HR) image from a Low Resolution (LR) image. With the remarkable achievement of image super-resolution in the fields of satellite imaging, medical imaging, security and monitoring, image generation and other computer vision tasks and image processing, reconstructing a higher-resolution and clearer image by using a super-resolution technology becomes a big research hot spot in the current image restoration field.
The image super-resolution reconstruction method mainly comprises 3 types: interpolation-based methods, reconstruction-based methods, and learning-based methods. The super-resolution reconstruction method based on the difference value is simple in calculation and easy to realize, but has larger dependence on priori knowledge of natural images, has poorer recovery effect on image details and is easy to generate edge effect. The reconstruction-based method is to obtain an LR image according to a known degradation model, restrict the generation of an HR image by extracting key pixel point information in the LR image, and obtain a corresponding reconstruction result by combining prior knowledge in the HR image. However, since the obtained prior knowledge is limited, more detailed information cannot be recovered for a complex image, and the reconstruction performance is limited. The learning-based method completes the reconstruction of the high-resolution image by establishing a mapping relation between the high-resolution image and the low-resolution image and then utilizing prior knowledge obtained by learning. Compared with other reconstruction algorithms, the super-resolution algorithm based on the learning image can obtain relatively good reconstruction effect, so that the super-resolution algorithm becomes a hot spot of the current super-resolution reconstruction research. The current learning-based methods mainly comprise sparse representation-based methods, neighbor embedding-based methods and deep learning-based methods. Yang et al propose a method based on sparse representation and dictionary learning by learning LR image blocks and an overcomplete dictionary corresponding to the HR image blocks for image reconstruction. However, because the learning requirement on the overcomplete dictionary is higher, the practicability is poor, timofte and the like combine the sparse dictionary with the field embedding, an anchoring field regression (ANR) algorithm and an improved anchoring field regression (A+) algorithm are provided, the calculation efficiency is improved, and the image detail recovery effect is poor.
The efficient learning ability of convolutional neural networks (Convolutional Neural Network, CNN) has been widely used in the super-resolution problem in recent years. Dong et al adopts a 3-layer convolutional neural network to reconstruct the image super-resolution for the first time, has a simple network structure and is easy to realize, but the network cannot extract the deep features of the image due to the defects of few convolutional layers, small receptive field, poor generalization capability and the like, so that the reconstruction performance is limited. The convolutional network proposed by Kim et al improves the network depth to 20 layers, and introduces residual error learning to improve the feature extraction capability. Meanwhile, kim et al also propose the deep recursive convolutional network, adopt the recursive learning to realize the parameter sharing of the deep network, reduce the training difficulty of the network. Although both methods proposed by Kim et al achieve better reconstruction performance, the more pronounced the gradient is and the network is degraded as the network deepens. The depth residual error coding and decoding network proposed by Mao and the like adopts a symmetrical mode to jump and connect a convolution layer and a deconvolution layer, so that the training convergence speed is faster, the local optimum with higher quality is achieved, but the gradient vanishes or the network degradation problem still exists along with the increase of the network depth.
Therefore, in combination with the above problems, it is a problem that needs to be solved by those skilled in the art to provide an image super-resolution reconstruction method based on a symmetric residual convolution neural network.
Disclosure of Invention
In view of the above, the invention provides an image super-resolution reconstruction method based on a symmetric residual convolution neural network, which realizes full utilization of intra-block information, extracts richer local features, establishes long-jump connection outside a residual block, realizes global feature fusion, and makes up for serious loss of degradation of depth network image detail information.
In order to achieve the above purpose, the technical scheme adopted by the invention is as follows:
an image super-resolution reconstruction method based on a symmetric residual convolution neural network, the method comprising the following steps:
s1, selecting a training sample and a test sample:
training samples: acquiring 91 image data sets and 200 Berkeley Segmentation data sets, rotating 291 images by 90 degrees, 180 degrees and 270 degrees, horizontally overturning, shrinking the images in the order of factors of 0.9, 0.8, 0.7 and 0.6, and obtaining 11640 training images after data enhancement processing is carried out on the training images;
test sample: testing with standard data sets Set5, set14, and BSD100 respectively;
s2, preprocessing a data set: sampling the original image by a factor k (k=2, 3, 4) by bicubic interpolation to generate a corresponding LR image, and cropping the LR training image into a set of l sub ×l sub Corresponding HR training image is cropped to kl sub ×kl sub Obtaining a low-resolution subgraph and a high-resolution subgraph training image pair;
s3, extracting features: extracting features on the low-resolution subgraph obtained in the step S2, wherein the features comprise two layers of convolution layers, and each layer comprises 64 convolution kernels with the size of 3 multiplied by 3;
s4, feature fusion: the method comprises the steps of including 4 residual blocks, wherein each residual block consists of an enhancement unit and a compression unit, the enhancement unit comprises 6 layers of convolution layers with the size of 3 multiplied by 3, the enhancement unit adopts symmetrical residual errors to realize identical mapping of shallow layer and deep layer information, fuses network context information, extracts characteristics, and the compression unit consists of 1 layer of convolution layers with the size of 1 multiplied by 1, so that key information is extracted for a subsequent network;
s5, image reconstruction: the method comprises a deconvolution layer without an activation function, and the reconstruction calculation relation is as follows:
y＝R(F n (B n-1 ))+U(x) (3)
wherein R, U are reconstruction and bicubic interpolation functions, respectively.
Preferably, the step S3 of feature extraction and calculation includes:
B 0 ＝f(x) (1)
wherein f is a feature extraction function, B 0 Is an extracted feature.
Preferably, the step S4 calculates the relationship as:
B m ＝F m (B m-1 )，m＝1,…,n (2)
wherein F is m As a function corresponding to the mth residual block, B m-1 And B m Respectively the input and output of the mth residual block.
Preferably, the calculation relationship is:
wherein B is m-1 For the output of the current residual block, and for the input of the next residual block, D is a dimension reduction operation,is the convolution operation of the z-th layer convolution layer in the m-1 th residual block.
Compared with the prior art, the invention has the following beneficial effects:
in order to reduce network parameters of a deep network, prevent gradient elimination and overfitting, the technical scheme of the invention realizes full utilization of information in a block and extraction of richer local features by establishing symmetrical short-jump connection in a residual block, and simultaneously, realizes global feature fusion by establishing long-jump connection outside the residual block so as to make up for serious loss of degradation of detail information of a deep network image, and realizes up-sampling by using a deconvolution layer, thereby restoring a high-resolution image. The invention further researches the application of the convolutional neural network in the super-resolution reconstruction of the depth image, and provides an image super-resolution reconstruction method based on the symmetric residual convolutional neural network. The full utilization of local information in residual blocks is realized by adopting a symmetrical residual method, and gradient disappearance caused by network depth is effectively improved; and meanwhile, fusion of global features outside the residual blocks is realized to make up for high-frequency details lost by network deepening.
Drawings
In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the drawings that are required to be used in the embodiments or the description of the prior art will be briefly described below, and it is obvious that the drawings in the following description are only embodiments of the present invention, and that other drawings can be obtained according to the provided drawings without inventive effort for a person skilled in the art.
FIG. 1 is a diagram of a symmetric residual convolutional neural network;
FIG. 2 is a diagram of a residual intra-block enhancement and compression unit framework;
FIG. 3 is a graph of PSNR convergence using symmetry and non-symmetry over a Set5 test Set;
FIG. 4 is a view showing a 2-fold reconstruction of the button_GT blur in the Set5 dataset;
FIG. 5 is a graph of a reconstruction of the Set14 dataset with zebra blur by a factor of 4.
Detailed Description
The following description of the embodiments of the present invention will be made clearly and completely, and it is apparent that the described embodiments 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.
The invention provides an image super-resolution reconstruction method based on a symmetrical residual convolution neural network, which comprises the following steps:
step 1, selecting a training sample and a test sample, wherein:
training samples: using 91 image datasets, this embodiment uses 200 datasets from YANG JC, WRIGHT J, HUANG T S, et al image super-resolution via sparse representation [ J ]. IEEE transactions on image processing,2010,19 (11): 2861-2873 and Berkeley Segmentation (BSD) as training samples, and obtains 11640 training images by performing 90 DEG, 180 DEG and 270 DEG rotations, horizontal flipping and sequential reduction by factors of 0.9, 0.8, 0.7 and 0.6 on 291 images, and performing data enhancement processing on the training images.
Test sample: to further verify the validity of the network herein, a wide range of standard data sets are selected: set5, set14 and BSD100 are tested, and the data Set contains rich natural scenes, so that the performance of the network can be effectively tested.
Step 2, preprocessing a data set: downsampling the original image by a factor k (k=2, 3, 4) using bicubic interpolation to generate a corresponding LR image, and then cropping the LR training image into a set of l sub ×l sub The corresponding HR training image is cropped to kl sub ×kl sub Since the present invention trains using the Caffe deep learning framework, the output size produced by the deconvolution kernel is set to (kl) sub -k+1) 2 Rather than (kl) sub ) 2 . Therefore, the pixel boundaries of the HR sub-graph should be clipped (k-1). When training is performed by using the Caffe framework, the training sample with a larger learning rate is larger in size, and the training process is larger in sizeUnstable. Since there is no overlap when clipping image blocks, in order to fully utilize 291 images, when the blurring factor k=3, more edge image blocks are to be obtained, the sub-image sampling step size is set to 15, and the training phase is trained using a training pair of size 152/432. In this way, training image pairs of different blur factor low resolution subgraphs and high resolution subgraphs are obtained.
Step 3, extracting features: features are extracted on the original low resolution image, comprising two convolution layers, each layer comprising 64 convolution kernels of size 3 x 3, the feature extraction process can be expressed as:
B 0 ＝f(x) (1)
wherein f represents a feature extraction function, B 0 Representing the extracted features;
step 4, feature fusion: the method comprises the steps of including 4 residual blocks, reading the features extracted in the step 3 into the residual blocks, wherein each residual block is composed of an Enhancement Unit (EU) and a Compression Unit (CU), the Enhancement Unit comprises 6 layers of 3×3 convolution layers, the symmetrical residual is adopted to realize identical mapping of shallow and deep information, network context information is fused, and richer features are extracted. The compression unit is composed of 1 layer 1×1 convolution layer, and extracts key information for subsequent network.
This process can be expressed as:
B m ＝F m (B m-1 )，m＝1,…,n (2)
wherein F is m Representing a function corresponding to the mth residual block, B m-1 And B m Representing the input and output of the mth residual block, respectively.
Wherein B is m-1 Represents the output of the current residual block, while D represents the dimension reduction operation,representing the convolution operation of the z-th layer convolution layer in the m-1 th residual block.
Step 5, image reconstruction: the method comprises a deconvolution layer without an activation function, wherein the local features and the global features extracted in the step 4 are subjected to up-sampling operation by adopting a deconvolution core of 17 multiplied by 17, a final output high-resolution image is obtained, and a reconstruction result is expressed as follows:
y＝R(F n (B n-1 ))+U(x) (4)
wherein R, U represent the reconstruction and bicubic interpolation functions, respectively.
The network training example of the present invention is illustrated as follows:
the initial learning rate adopted by the invention is set to be 10-4, the learning rate is reduced by 10-1 every 250000 times in the training process, and the total training times are 600000 times. The hardware used for training is configured as an Intel (R) Xeon (R) CPU E5-1650 v4@3.60GHz X12 processor, a Tesla K20c display card, a 64GB memory, a Matlab R2016a software and a caffe deep learning framework and a development packet CUDA8.0 for calling the GPU. Table 1 shows training sub-graph sizes for different blurring factors, and Table 2 shows average PSNR/SSIM for different super-resolution reconstruction methods on Set5, set14 and BSD100 datasets.
TABLE 1 training sub-graph sizes for different blur factors
TABLE 2 average PSNR/SSIM for different super-resolution reconstruction methods on Set5, set14 and BSD100 datasets
The invention provides an image super-resolution reconstruction method based on a symmetric residual convolution neural network, aiming at the problems that the image super-resolution reconstruction method based on the convolution neural network has high reconstruction performance, but the network parameters are more, the training difficulty is high, gradient disappearance, gradient explosion or overfitting is easy to occur, and the like. The local feature fusion is realized by symmetrically integrating the local features into the residual block and adopting symmetrical connection, so that as many valuable features as possible are extracted; and the global feature fusion is realized by adopting jump connection outside the residual block, so that the image reconstruction quality is improved. According to the invention, the global and local feature fusion is realized by adopting the symmetrical long-short jump connection, the stability of network training and the high efficiency of feature extraction are realized, 4 residual blocks are arranged, the richer robust features are extracted in the residual blocks by adopting a symmetrical residual mode, the global feature fusion is realized by the long-jump connection outside the residual blocks, the problem of insufficient feature extraction caused by simply stacking a plurality of residual blocks is solved, and the gradient elimination and network degradation phenomena are effectively relieved. The network structure is shown in fig. 2. Meanwhile, the method and the related method are compared with each other in terms of image quality of image super-resolution reconstruction.
In the present specification, each embodiment is described in a progressive manner, and each embodiment is mainly described in a different point from other embodiments, and identical and similar parts between the embodiments are all enough to refer to each other. For the device disclosed in the embodiment, since it corresponds to the method disclosed in the embodiment, the description is relatively simple, and the relevant points refer to the description of the method section.
The previous description of the disclosed embodiments is provided to enable any person skilled in the art to make or use the present invention. Various modifications to these embodiments will be readily apparent to those skilled in the art, and the generic principles defined herein may be applied to other embodiments without departing from the spirit or scope of the invention. Thus, the present invention is not intended to be limited to the embodiments shown herein but is to be accorded the widest scope consistent with the principles and novel features disclosed herein.
Claims (4)
1. An image super-resolution reconstruction method based on a symmetric residual convolution neural network is characterized by comprising the following steps of:
s1, selecting a training sample and a test sample:
training samples: acquiring 91 image data sets and 200 Berkeley Segmentation data sets, rotating 291 images by 90 degrees, 180 degrees and 270 degrees, horizontally overturning, shrinking the images in the order of factors of 0.9, 0.8, 0.7 and 0.6, and obtaining 11640 training images after data enhancement processing is carried out on the training images;
test sample: testing with standard data sets Set5, set14, and BSD100 respectively;
s2, preprocessing a data set: sampling an original image by a factor k through a bicubic interpolation method, wherein the value of k is 2,3 and 4 respectively, generating a corresponding LR image, and cutting the LR training image into a group of l sub ×l sub Corresponding HR training image is cropped to kl sub ×kl sub Obtaining a low-resolution subgraph and a high-resolution subgraph training image pair;
s3, extracting features: extracting features on the low-resolution subgraph obtained in the step S2, wherein the features comprise two layers of convolution layers, and each layer comprises 64 convolution kernels with the size of 3 multiplied by 3;
s4, feature fusion: the method comprises the steps of including 4 residual blocks, wherein each residual block consists of an enhancement unit and a compression unit, the enhancement unit comprises 6 layers of convolution layers with the size of 3 multiplied by 3, the enhancement unit adopts symmetrical residual errors to realize identical mapping of shallow layer and deep layer information, fuses network context information, extracts characteristics, and the compression unit consists of 1 layer of convolution layers with the size of 1 multiplied by 1, so that key information is extracted for a subsequent network;
s5, image reconstruction: the method comprises a deconvolution layer without an activation function, and the reconstruction calculation relation is as follows:
y＝R(F n (B n-1 ))+U(x) (3)
wherein R, U are reconstruction and bicubic interpolation functions respectively; f (F) n Representing the function corresponding to the nth residual block, B n-1 Representing the input of the nth residual block, F n (B n-1 )＝B n ，B n Is the output of the nth residual block.
2. The method for reconstructing the super-resolution image based on the symmetric residual convolutional neural network according to claim 1, wherein the step S3 feature extraction calculation relationship is as follows:
B 0 ＝f(x) (1)
wherein f is a feature extraction function, B 0 Is an extracted feature.
3. The method for reconstructing the super-resolution image based on the symmetric residual convolutional neural network according to claim 1, wherein the step S4 calculates the relationship as follows:
B m ＝F m (B m-1 )，m＝1,...,n (2)
wherein F is m As a function corresponding to the mth residual block, B m-1 And B m Respectively the input and output of the mth residual block.
4. The method for reconstructing an image super-resolution based on a symmetric residual convolutional neural network according to claim 3, wherein the calculated relationship is:
wherein B is m-1 For the output of the current residual block, and for the input of the next residual block, D is a dimension reduction operation,and (3) performing convolution operation on a z-th layer convolution layer in the m-1 th residual block, wherein the value of z is respectively 1-6.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201910093918.8A CN109903226B (en) | 2019-01-30 | 2019-01-30 | Image super-resolution reconstruction method based on symmetric residual convolution neural network |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201910093918.8A CN109903226B (en) | 2019-01-30 | 2019-01-30 | Image super-resolution reconstruction method based on symmetric residual convolution neural network |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN109903226A CN109903226A (en) | 2019-06-18 |
| CN109903226B true CN109903226B (en) | 2023-08-15 |
Family
ID=66944440
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201910093918.8A Active CN109903226B (en) | 2019-01-30 | 2019-01-30 | Image super-resolution reconstruction method based on symmetric residual convolution neural network |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN109903226B (en) |
Families Citing this family (20)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN109191378A (en) * | 2018-07-25 | 2019-01-11 | 成都信息工程大学 | Based on ERGAN network single image super-resolution reconstruction method |
| CN110298791B (en) * | 2019-07-08 | 2022-10-28 | 西安邮电大学 | Super-resolution reconstruction method and device for license plate image |
| CN110689483B (en) * | 2019-09-24 | 2022-07-01 | 重庆邮电大学 | Image super-resolution reconstruction method based on depth residual error network and storage medium |
| CN111080533B (en) * | 2019-10-21 | 2023-05-16 | 南京航空航天大学 | Digital zooming method based on self-supervision residual sensing network |
| CN113538225A (en) * | 2020-04-14 | 2021-10-22 | 阿里巴巴集团控股有限公司 | Model training method, image conversion method, device, equipment and storage medium |
| CN111681296B (en) * | 2020-05-09 | 2024-03-22 | 上海联影智能医疗科技有限公司 | Image reconstruction method, image reconstruction device, computer equipment and storage medium |
| CN111667407B (en) * | 2020-05-18 | 2023-04-07 | 武汉大学 | Image super-resolution method guided by depth information |
| CN111861961B (en) * | 2020-07-25 | 2023-09-22 | 安徽理工大学 | Single image super-resolution multi-scale residual error fusion model and restoration method thereof |
| CN111932461B (en) * | 2020-08-11 | 2023-07-25 | 西安邮电大学 | Self-learning image super-resolution reconstruction method and system based on convolutional neural network |
| CN112102166B (en) * | 2020-08-26 | 2023-12-01 | 上海交通大学 | Combined super-resolution, color gamut expansion and inverse tone mapping method and equipment |
| CN112070676B (en) * | 2020-09-10 | 2023-10-27 | 东北大学秦皇岛分校 | Picture super-resolution reconstruction method of double-channel multi-perception convolutional neural network |
| CN112053287B (en) * | 2020-09-11 | 2023-04-18 | 北京邮电大学 | Image super-resolution method, device and equipment |
| CN114387161B (en) * | 2020-10-16 | 2023-07-07 | 四川大学 | Video super-resolution reconstruction method |
| CN112348839B (en) * | 2020-10-27 | 2024-03-15 | 重庆大学 | Image segmentation method and system based on deep learning |
| CN112734642B (en) * | 2021-01-12 | 2023-03-10 | 武汉工程大学 | Remote sensing satellite super-resolution method and device of multi-scale texture transfer residual error network |
| CN112767255B (en) * | 2021-03-04 | 2022-11-29 | 山东大学 | Image super-resolution reconstruction method and system based on feature separation fusion network |
| CN113116361A (en) * | 2021-03-09 | 2021-07-16 | 山东大学 | Sleep staging method based on single-lead electroencephalogram |
| CN113409195A (en) * | 2021-07-06 | 2021-09-17 | 中国标准化研究院 | Image super-resolution reconstruction method based on improved deep convolutional neural network |
| CN113850269B (en) * | 2021-12-01 | 2022-03-15 | 西南石油大学 | Denoising method based on multi-branch selective kernel nested connection residual error network |
| CN114693547A (en) * | 2022-03-03 | 2022-07-01 | 大连海事大学 | Radio frequency image enhancement method and radio frequency image identification method based on image super-resolution |
Citations (16)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN105929843A (en) * | 2016-04-22 | 2016-09-07 | 天津城建大学 | Robot path planning method based on improved ant colony algorithm |
| CN106874898A (en) * | 2017-04-08 | 2017-06-20 | 复旦大学 | Extensive face identification method based on depth convolutional neural networks model |
| CN107240066A (en) * | 2017-04-28 | 2017-10-10 | 天津大学 | Image super-resolution rebuilding algorithm based on shallow-layer and deep layer convolutional neural networks |
| CN107274347A (en) * | 2017-07-11 | 2017-10-20 | 福建帝视信息科技有限公司 | A kind of video super-resolution method for reconstructing based on depth residual error network |
| CN107358575A (en) * | 2017-06-08 | 2017-11-17 | 清华大学 | A kind of single image super resolution ratio reconstruction method based on depth residual error network |
| CN107578377A (en) * | 2017-08-31 | 2018-01-12 | 北京飞搜科技有限公司 | A kind of super-resolution image reconstruction method and system based on deep learning |
| CN108460726A (en) * | 2018-03-26 | 2018-08-28 | 厦门大学 | A kind of magnetic resonance image super-resolution reconstruction method based on enhancing recurrence residual error network |
| CN108492270A (en) * | 2018-03-23 | 2018-09-04 | 沈阳理工大学 | A kind of super-resolution method reconstructed based on fuzzy kernel estimates and variation |
| CN108537733A (en) * | 2018-04-11 | 2018-09-14 | 南京邮电大学 | Super resolution ratio reconstruction method based on multipath depth convolutional neural networks |
| CN108647775A (en) * | 2018-04-25 | 2018-10-12 | 陕西师范大学 | Super-resolution image reconstruction method based on full convolutional neural networks single image |
| CN108734660A (en) * | 2018-05-25 | 2018-11-02 | 上海通途半导体科技有限公司 | A kind of image super-resolution rebuilding method and device based on deep learning |
| CN108921789A (en) * | 2018-06-20 | 2018-11-30 | 华北电力大学 | Super-resolution image reconstruction method based on recurrence residual error network |
| CN108921786A (en) * | 2018-06-14 | 2018-11-30 | 天津大学 | Image super-resolution reconstructing method based on residual error convolutional neural networks |
| CN109064396A (en) * | 2018-06-22 | 2018-12-21 | 东南大学 | A kind of single image super resolution ratio reconstruction method based on depth ingredient learning network |
| CN109064398A (en) * | 2018-07-14 | 2018-12-21 | 深圳市唯特视科技有限公司 | A kind of image super-resolution implementation method based on residual error dense network |
| CN109118432A (en) * | 2018-09-26 | 2019-01-01 | 福建帝视信息科技有限公司 | A kind of image super-resolution rebuilding method based on Rapid Circulation convolutional network |
Family Cites Families (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| TWI624804B (en) * | 2016-11-07 | 2018-05-21 | 盾心科技股份有限公司 | A method and system for providing high resolution image through super-resolution reconstrucion |
-
2019
- 2019-01-30 CN CN201910093918.8A patent/CN109903226B/en active Active
Patent Citations (16)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN105929843A (en) * | 2016-04-22 | 2016-09-07 | 天津城建大学 | Robot path planning method based on improved ant colony algorithm |
| CN106874898A (en) * | 2017-04-08 | 2017-06-20 | 复旦大学 | Extensive face identification method based on depth convolutional neural networks model |
| CN107240066A (en) * | 2017-04-28 | 2017-10-10 | 天津大学 | Image super-resolution rebuilding algorithm based on shallow-layer and deep layer convolutional neural networks |
| CN107358575A (en) * | 2017-06-08 | 2017-11-17 | 清华大学 | A kind of single image super resolution ratio reconstruction method based on depth residual error network |
| CN107274347A (en) * | 2017-07-11 | 2017-10-20 | 福建帝视信息科技有限公司 | A kind of video super-resolution method for reconstructing based on depth residual error network |
| CN107578377A (en) * | 2017-08-31 | 2018-01-12 | 北京飞搜科技有限公司 | A kind of super-resolution image reconstruction method and system based on deep learning |
| CN108492270A (en) * | 2018-03-23 | 2018-09-04 | 沈阳理工大学 | A kind of super-resolution method reconstructed based on fuzzy kernel estimates and variation |
| CN108460726A (en) * | 2018-03-26 | 2018-08-28 | 厦门大学 | A kind of magnetic resonance image super-resolution reconstruction method based on enhancing recurrence residual error network |
| CN108537733A (en) * | 2018-04-11 | 2018-09-14 | 南京邮电大学 | Super resolution ratio reconstruction method based on multipath depth convolutional neural networks |
| CN108647775A (en) * | 2018-04-25 | 2018-10-12 | 陕西师范大学 | Super-resolution image reconstruction method based on full convolutional neural networks single image |
| CN108734660A (en) * | 2018-05-25 | 2018-11-02 | 上海通途半导体科技有限公司 | A kind of image super-resolution rebuilding method and device based on deep learning |
| CN108921786A (en) * | 2018-06-14 | 2018-11-30 | 天津大学 | Image super-resolution reconstructing method based on residual error convolutional neural networks |
| CN108921789A (en) * | 2018-06-20 | 2018-11-30 | 华北电力大学 | Super-resolution image reconstruction method based on recurrence residual error network |
| CN109064396A (en) * | 2018-06-22 | 2018-12-21 | 东南大学 | A kind of single image super resolution ratio reconstruction method based on depth ingredient learning network |
| CN109064398A (en) * | 2018-07-14 | 2018-12-21 | 深圳市唯特视科技有限公司 | A kind of image super-resolution implementation method based on residual error dense network |
| CN109118432A (en) * | 2018-09-26 | 2019-01-01 | 福建帝视信息科技有限公司 | A kind of image super-resolution rebuilding method based on Rapid Circulation convolutional network |
Non-Patent Citations (1)
| Title |
|---|
| Stable and symmetric convolutional nearal network;Raymond Alexander Yeh;《Http://www.ideals.illinois.edu/items/95106》;20160831;全文 * |
Also Published As
| Publication number | Publication date |
|---|---|
| CN109903226A (en) | 2019-06-18 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN109903226B (en) | Image super-resolution reconstruction method based on symmetric residual convolution neural network | |
| Zhang et al. | Designing a practical degradation model for deep blind image super-resolution | |
| Zhang et al. | Residual dense network for image super-resolution | |
| CN110570353B (en) | Super-resolution reconstruction method for generating single image of countermeasure network by dense connection | |
| CN110033410B (en) | Image reconstruction model training method, image super-resolution reconstruction method and device | |
| CN110120011B (en) | Video super-resolution method based on convolutional neural network and mixed resolution | |
| Qin et al. | Multi-scale feature fusion residual network for single image super-resolution | |
| CN108647775B (en) | Super-resolution image reconstruction method based on full convolution neural network single image | |
| CN107240066A (en) | Image super-resolution rebuilding algorithm based on shallow-layer and deep layer convolutional neural networks | |
| CN110276721A (en) | Image super-resolution rebuilding method based on cascade residual error convolutional neural networks | |
| CN111105352A (en) | Super-resolution image reconstruction method, system, computer device and storage medium | |
| CN110705699A (en) | Super-resolution reconstruction method and device, electronic equipment and readable storage medium | |
| WO2016019484A1 (en) | An apparatus and a method for providing super-resolution of a low-resolution image | |
| CN111768340B (en) | Super-resolution image reconstruction method and system based on dense multipath network | |
| CN111340744B (en) | Attention double-flow depth network-based low-quality image down-sampling method and system | |
| CN112200724B (en) | Single-image super-resolution reconstruction system and method based on feedback mechanism | |
| CN111402128A (en) | Image super-resolution reconstruction method based on multi-scale pyramid network | |
| CN113554058A (en) | Method, system, device and storage medium for enhancing resolution of visual target image | |
| CN111800630A (en) | Method and system for reconstructing video super-resolution and electronic equipment | |
| CN115953294A (en) | Single-image super-resolution reconstruction method based on shallow channel separation and aggregation | |
| CN113538246A (en) | Remote sensing image super-resolution reconstruction method based on unsupervised multi-stage fusion network | |
| Tang et al. | Deep residual networks with a fully connected reconstruction layer for single image super-resolution | |
| CN111754399A (en) | Image super-resolution method for keeping geometric structure based on gradient | |
| CN115526779A (en) | Infrared image super-resolution reconstruction method based on dynamic attention mechanism | |
| Zhou et al. | Deep fractal residual network for fast and accurate single image super resolution |
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 | ||
| CB03 | Change of inventor or designer information | ||
| CB03 | Change of inventor or designer information |
Inventor after: Liu Shudong Inventor after: Wang Xiaomin Inventor after: Zhang Yan Inventor before: Liu Shudong |
|
| GR01 | Patent grant | ||
| GR01 | Patent grant |