CN111402128A - Image super-resolution reconstruction method based on multi-scale pyramid network - Google Patents
Image super-resolution reconstruction method based on multi-scale pyramid network Download PDFInfo
- Publication number
- CN111402128A CN111402128A CN202010108622.1A CN202010108622A CN111402128A CN 111402128 A CN111402128 A CN 111402128A CN 202010108622 A CN202010108622 A CN 202010108622A CN 111402128 A CN111402128 A CN 111402128A
- Authority
- CN
- China
- Prior art keywords
- layer
- image
- resolution
- features
- scale
- 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
- 238000000034 method Methods 0.000 title claims abstract description 25
- 230000004927 fusion Effects 0.000 claims abstract description 14
- 238000005070 sampling Methods 0.000 claims abstract description 13
- 238000000605 extraction Methods 0.000 claims abstract description 4
- 230000004913 activation Effects 0.000 claims description 15
- 230000006870 function Effects 0.000 claims description 12
- 238000003475 lamination Methods 0.000 claims description 12
- 230000006835 compression Effects 0.000 claims description 9
- 238000007906 compression Methods 0.000 claims description 9
- 208000037170 Delayed Emergence from Anesthesia Diseases 0.000 claims description 6
- 238000012549 training Methods 0.000 claims description 6
- 230000017105 transposition Effects 0.000 claims description 3
- 238000004364 calculation method Methods 0.000 claims description 2
- 230000008569 process Effects 0.000 claims description 2
- 230000000694 effects Effects 0.000 description 3
- 230000015556 catabolic process Effects 0.000 description 2
- 238000013135 deep learning Methods 0.000 description 2
- 238000006731 degradation reaction Methods 0.000 description 2
- 238000011161 development Methods 0.000 description 2
- 238000010586 diagram Methods 0.000 description 2
- 238000012986 modification Methods 0.000 description 2
- 230000004048 modification Effects 0.000 description 2
- 238000006467 substitution reaction Methods 0.000 description 2
- 230000007547 defect Effects 0.000 description 1
- 230000008034 disappearance Effects 0.000 description 1
- 230000006872 improvement Effects 0.000 description 1
- 238000012545 processing Methods 0.000 description 1
- 230000002787 reinforcement Effects 0.000 description 1
- 238000011160 research Methods 0.000 description 1
Images
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T3/00—Geometric image transformations in the plane of the image
- G06T3/40—Scaling of whole images or parts thereof, e.g. expanding or contracting
- G06T3/4053—Scaling of whole images or parts thereof, e.g. expanding or contracting based on super-resolution, i.e. the output image resolution being higher than the sensor resolution
- G06T3/4076—Scaling of whole images or parts thereof, e.g. expanding or contracting based on super-resolution, i.e. the output image resolution being higher than the sensor resolution using the original low-resolution images to iteratively correct the high-resolution images
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/04—Architecture, e.g. interconnection topology
- G06N3/045—Combinations of networks
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/08—Learning methods
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T3/00—Geometric image transformations in the plane of the image
- G06T3/40—Scaling of whole images or parts thereof, e.g. expanding or contracting
- G06T3/4007—Scaling of whole images or parts thereof, e.g. expanding or contracting based on interpolation, e.g. bilinear interpolation
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T5/00—Image enhancement or restoration
- G06T5/50—Image enhancement or restoration using two or more images, e.g. averaging or subtraction
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T7/00—Image analysis
- G06T7/50—Depth or shape recovery
Landscapes
- Engineering & Computer Science (AREA)
- Physics & Mathematics (AREA)
- Theoretical Computer Science (AREA)
- General Physics & Mathematics (AREA)
- General Health & Medical Sciences (AREA)
- Molecular Biology (AREA)
- Biophysics (AREA)
- Computational Linguistics (AREA)
- Data Mining & Analysis (AREA)
- Evolutionary Computation (AREA)
- Artificial Intelligence (AREA)
- Biomedical Technology (AREA)
- Computing Systems (AREA)
- General Engineering & Computer Science (AREA)
- Life Sciences & Earth Sciences (AREA)
- Mathematical Physics (AREA)
- Software Systems (AREA)
- Health & Medical Sciences (AREA)
- Computer Vision & Pattern Recognition (AREA)
- Image Processing (AREA)
Abstract
The invention discloses an image super-resolution reconstruction method based on a multi-scale pyramid network, which comprises the following steps: s1, shallow feature extraction is carried out on the input image; s2, performing feature fusion and feature enhancement on the shallow features through K multi-scale residual error modules to obtain richer deep features; s3, performing up-sampling on the deep level features by using the transposed convolution; s4, reconstructing the image by residual error learning; and S5, taking the reconstructed image as the output of the current pyramid network and the input of the next pyramid network, and continuing to train by adopting the steps S1-S4 to obtain an image with higher resolution. The invention adopts a multi-scale residual error module to fuse the characteristics to obtain richer characteristics; meanwhile, a Laplacian pyramid network is adopted to gradually up-sample and reconstruct a high-resolution image; by the method, the image with richer details and higher quality can be reconstructed.
Description
Technical Field
The invention relates to the technical field of computer vision and image processing, in particular to an image super-resolution reconstruction method based on a multi-scale pyramid network.
Background
With the development of information technology, the number of pictures on a network is continuously increased, and the images are used as a main medium for people to learn the world and are applied to various scenes. In a plurality of fields, the image quality is as large as that of medical images, satellite remote sensing, and people's cameras, mobile phones and the like. People have increasingly high requirements on the image quality of images. Therefore, the improvement of the resolution of the image is of great significance in real life.
Image super-resolution reconstruction aims to recover a high-resolution image by using one or more low-resolution images, and has been developed in recent years as one of the research hotspots in the field of computer vision. At present, super-resolution reconstruction algorithms are divided into two categories, namely interpolation-based and learning-based, in the aspect of reconstruction algorithms. The interpolation-based algorithm is simple and fast, but cannot meet the increasing image quality requirements of people. The super-resolution reconstruction method based on learning learns the prior by means of extra training samples to reduce the ill-posed property of the super-resolution problem and obtain better effect, such as a method based on sparse coding and a method based on neighborhood embedding. However, these methods only solve sparse coding coefficients and learn embedding space on the primary feature space of the image, so that sparsity and manifold assumptions are difficult to strictly satisfy, directly resulting in degradation of image reconstruction quality. With the rapid development of deep learning, researchers widely apply the deep learning algorithm to image super-resolution reconstruction and obtain a reconstruction result superior to an interpolation algorithm. However, the mainstream methods at present are based on the theory that the deeper the network is, the better the reconstruction effect is, and with the increase of the network depth, the problems of gradient disappearance or grid degradation still exist, and most methods are to reach the specified size through one-time up-sampling, so that the quality of the reconstructed high-resolution image needs to be improved.
Disclosure of Invention
The invention aims to solve the defects in the prior art and provides an image super-resolution reconstruction method based on a multi-scale pyramid network.
The purpose of the invention can be achieved by adopting the following technical scheme:
an image super-resolution reconstruction method based on a multi-scale pyramid network comprises the following steps:
s1, shallow feature extraction is carried out on the input image;
s2, performing feature fusion and feature enhancement on the shallow features through K multi-scale residual error modules to obtain deep features;
s3, performing up-sampling on the deep level features by using the transposed convolution;
s4, reconstructing the image by residual error learning;
and S5, taking the reconstructed image as the output of the current pyramid network and simultaneously as the input of the next pyramid network, and continuously repeating the training from the step S1 to the step S4 to obtain an image with higher resolution.
Further, the step S1 is as follows:
extracting shallow features from the input low resolution image using a layer of 3 × 3 convolutional layer followed by a non-linear activation unit, the expression is as follows:
F0=σ(W1*ILR) (1)
wherein, ILRRepresenting the input low resolution image, sigma represents the nonlinear activation function Re L U, W1Convolution kernel representing the convolution layer of 3 × 3, F0Representing features extracted by convolutional layers.
Further, each multi-scale residual module in S2 comprises a feature enhancement unit, a compression unit and a residual learning unit, wherein the feature enhancement unit comprises 2 convolution layers of 3 × 3 followed by nonlinear activation units and 2 convolution layers of 5 × 5 followed by nonlinear activation units, the compression unit comprises a layer of convolution layers with the size of 1 × 1, compared with a single-scale convolution kernel, features of different scales can be extracted by convolution kernels with different sizes, so that a filter can extract and learn richer image information.
Further, the step S2 is as follows:
firstly, the shallow layer features extracted in the step S1 are processed by a feature enhancement unit to obtain two different features, then the two different features are processed by feature fusion by a compression unit, the fused features are further processed by learning of a convolution layer, and finally the fused features and the shallow layer features are added to form a residual block, wherein the expression of the calculation process is as follows:
M=W1×1*[T2,P2](6)
B=σ(W*M) (7)
Fm=B+Fm-1(8)
wherein, T1Characteristic after lamination by the first layer 3 × 3, T2Characteristic of the post-lamination by the second layer 3 × 3, P1Characteristic of the post-lamination layer after lamination by the first layer 5 × 5, P2For the characteristics after the layer was wrapped by the second layer 5 × 5, σ represents the nonlinear activation function Re L U function,the convolution kernel of the first layer 3 × 3 convolutional layer,the convolution kernel of the second layer 3 × 3 convolutional layer,the convolution kernel of the first layer 5 × 5 convolutional layer,convolution kernel, W, of the second layer 5 × 5 convolution layer1×1Denotes a convolution kernel of 1 [1 × 1 ] convolution layer, W denotes a convolution kernel of the last learning layer, [ alpha ], [ alpha]Represents a feature fusion function, M represents a feature after fusion by 1 × 1 convolutional layers,Brepresenting features obtained by the last learning layer, Fm-1And FmRespectively representing the input and output of the mth multi-scale residual block.
Further, the step S3 is as follows:
using a layer of transposition convolution layer to up-sample the deep-level features passing through the K multi-scale residual error modules to obtain a high-resolution image, wherein the expression is as follows:
IHR_conv=fdeconv(FK) (9)
in the above formula, IHR_convFor high resolution images after upsampling, fdeconvFor up-sampling operations, FKIs the output of the kth multi-scale residual module.
Further, the step S4 is as follows:
firstly, carrying out double cubic interpolation on a low-resolution image to obtain a high-resolution image IHR_bicuThen the high resolution image IHR_bicuWith up-sampled high-resolution image IHR_convAdding to obtain a depth map I with spatial resolution amplified by two timesHRThe expression is as follows:
IHR=IHR_bicu+IHR_deconv(10)
further, compared with a network in which an image with a specified size is obtained by only one-time upsampling, the pyramid network can be used for gradual upsampling, so that the training difficulty of the network (especially the training of large-scale factors) can be reduced, and a picture with higher quality can be obtained. The pyramid network comprises N levels in total, if the input low-resolution image is a low-resolution image with the down-sampling rate of 1/S times, and S is an up-sampling scale factor, N is log2S; each stage reconstructs an image output by a previous stage into a high-resolution image of the stage.
Compared with the prior art, the invention has the following advantages and effects:
according to the invention, the multi-scale residual error module is adopted to extract various characteristics from the image, the characteristics are reinforced by fusing the characteristics, so that the extracted characteristics are richer, and the image is gradually up-sampled and reconstructed in a pyramid network mode, so that the high-resolution image quality is higher.
Drawings
FIG. 1 is a schematic diagram of an image super-resolution reconstruction method based on a multi-scale pyramid network disclosed in the present invention;
fig. 2 is a multi-scale residual module framework diagram in the present invention.
Detailed Description
In order to make the objects, technical solutions and advantages of the embodiments of the present invention clearer, the technical solutions in the embodiments of the present invention will be clearly and completely described below with reference to the drawings in the embodiments of the present invention, and it is obvious that the described embodiments are some, but not all, embodiments of the present invention. All other embodiments, which can be derived by a person skilled in the art from the embodiments given herein without making any creative effort, shall fall within the protection scope of the present invention.
Examples
As shown in fig. 1, the present embodiment provides an image super-resolution reconstruction method based on a multi-scale pyramid network, which performs fusion and reinforcement on extracted features through a multi-scale residual module, and gradually performs up-sampling on the pyramid network to gradually reconstruct an image, and specifically includes the following steps:
s1, shallow feature extraction is carried out on the input image, and the method specifically comprises the following steps:
shallow features are extracted from the input low resolution image using a layer of 3 × 3 convolutional layer followed by a non-linear activation unit, as follows:
F0=σ(W1*ILR) (1)
wherein, ILRRepresenting the input low resolution image, sigma represents the nonlinear activation function Re L U, W1Convolution kernel representing the convolution layer of 3 × 3, F0Representing features extracted by convolutional layers.
S2, performing feature fusion and feature enhancement on the shallow features through a plurality of K multi-scale residual error modules to obtain deep features, specifically:
the multi-scale residual error module can extract features of different scales by using convolution kernels of different sizes, so that the filter can extract and learn richer image information. As shown in fig. 1, by inputting the shallow feature into K multi-scale residual error modules, the shallow feature can be enhanced to obtain richer and deeper features, where K is 2 in this embodiment, but the value of K does not limit the technical solution of the present invention.
Multi-scale residual modules as shown in fig. 2, each multi-scale residual module comprises a feature enhancement unit, a compression unit and a residual learning, wherein the feature enhancement unit comprises 2 convolutional layers of 3 × 3 followed by nonlinear activation units and 2 convolutional layers of 5 × 5 followed by nonlinear activation units, the compression unit is composed of one layer of convolutional layer with the size of 1 × 1, and the utilization of the residual learning makes the network easier to optimize.
Firstly, the shallow layer features extracted in the step S1 are processed by a feature enhancement unit to obtain two different features, then the two different features are processed by feature fusion by a compression unit, the fused features are further processed by learning of a convolution layer, and finally the feature is added with the shallow layer features to form a residual block; the expression is as follows:
M=W1×1*[T2,P2](6)
B=σ(W*M) (7)
Fm=B+Fm-1(8)
wherein, T1Characteristic after lamination by the first layer 3 × 3, T2Characteristic of the post-lamination by the second layer 3 × 3, P1Characteristic of the post-lamination layer after lamination by the first layer 5 × 5, P2For the characteristics after the layer was wrapped by the second layer 5 × 5, σ represents the nonlinear activation function Re L U function,the convolution kernel of the first layer 3 × 3 convolutional layer,the convolution kernel of the second layer 3 × 3 convolutional layer,the convolution kernel of the first layer 5 × 5 convolutional layer,convolution kernel, W, of the second layer 5 × 5 convolution layer1×1Denotes a convolution kernel of 1 [1 × 1 ] convolution layer, W denotes a convolution kernel of the last learning layer, [ alpha ], [ alpha]Represents a feature fusion function, M represents a feature after fusion by 1 × 1 convolutional layers,Brepresenting features obtained by the last learning layer, Fm-1And FmRespectively representing the input and output of the mth multi-scale residual block. In this embodiment, m is [1,2 ]]。
S3, using a layer of transposition convolution layer to up-sample the deep-level features passing through the 2 multi-scale residual modules to obtain a high-resolution image, wherein the expression is as follows:
IHR_conv=fdeconv(F2) (9)
in the above formula, IHR_convFor high resolution images after upsampling, fdeconvFor up-sampling operations, F2Is the output of the 2 nd multi-scale residual module.
S4, reconstructing the image by residual learning, specifically:
firstly, carrying out double cubic interpolation on a low-resolution image to obtain a high-resolution image IHR_bicuThen the high resolution image IHR_bicuWith up-sampled high-resolution image IHR_convAdding to obtain a depth map I with spatial resolution amplified by two timesHRThe expression is as follows:
IHR=IHR_bicu+IHR_deconv(10)
and S5, taking the reconstructed image as the output of the current pyramid network layer and the input of the next pyramid network layer, and continuing to adopt the training from the step S1 to the step S4 to obtain an image with higher resolution.
The pyramid network comprises N levels in total, if the input low-resolution image is a down-sampling 1/S times low-resolution image, and S is an up-sampling scale factor, N is log2S; each stage reconstructs an image output by a previous stage into a high-resolution image of the stage.
The above embodiments are preferred embodiments of the present invention, but the present invention is not limited to the above embodiments, and any other changes, modifications, substitutions, combinations, and simplifications which do not depart from the spirit and principle of the present invention should be construed as equivalents thereof, and all such changes, modifications, substitutions, combinations, and simplifications are intended to be included in the scope of the present invention.
Claims (7)
1. The image super-resolution reconstruction method based on the multi-scale pyramid network is characterized by comprising the following steps of:
s1, shallow feature extraction is carried out on the input image;
s2, performing feature fusion and feature enhancement on the shallow features through K multi-scale residual error modules to obtain deep features;
s3, performing up-sampling on the deep level features by using the transposed convolution;
s4, reconstructing the image by residual error learning;
and S5, taking the reconstructed image as the output of the current pyramid network and simultaneously as the input of the next pyramid network, and continuously repeating the training from the step S1 to the step S4 to obtain an image with higher resolution.
2. The method for reconstructing image resolution based on multi-scale pyramid network as claimed in claim 1, wherein said step S1 is performed as follows:
extracting shallow features from the input low resolution image using a layer of 3 × 3 convolutional layer followed by a non-linear activation unit, the expression is as follows:
F0=σ(W1*ILR) (1)
wherein, ILRRepresenting the input low resolution image, sigma represents the nonlinear activation function Re L U, W1Convolution kernel representing the convolution layer of 3 × 3, F0Representing features extracted by convolutional layers.
3. The method as claimed in claim 1, wherein each multi-scale residual error module comprises a feature enhancement unit, a compression unit and a residual error learning, wherein the feature enhancement unit comprises 2 convolutional layers of 3 × 3 followed by nonlinear activation units and 2 convolutional layers of 5 × 5 followed by nonlinear activation units, and the compression unit comprises a convolutional layer with a size of 1 × 1.
4. The method for reconstructing image resolution based on multi-scale pyramid network as claimed in claim 3, wherein said step S2 is as follows:
firstly, the shallow layer features extracted in the step S1 are processed by a feature enhancement unit to obtain two different features, then the two different features are processed by feature fusion by a compression unit, the fused features are further processed by learning of a convolution layer, and finally the fused features and the shallow layer features are added to form a residual block, wherein the expression of the calculation process is as follows:
M=W1×1*[T2,P2](6)
B=σ(W*M) (7)
Fm=B+Fm-1(8)
wherein, T1Characteristic after lamination by the first layer 3 × 3, T2Characteristic of the post-lamination by the second layer 3 × 3, P1Characteristic of the post-lamination layer after lamination by the first layer 5 × 5, P2For the characteristics after the layer was wrapped by the second layer 5 × 5, σ represents the nonlinear activation function Re L U function,the convolution kernel of the first layer 3 × 3 convolutional layer,the convolution kernel of the second layer 3 × 3 convolutional layer,the convolution kernel of the first layer 5 × 5 convolutional layer,convolution kernel, W, of the second layer 5 × 5 convolution layer1×1Denotes a convolution kernel of 1 [1 × 1 ] convolution layer, W denotes a convolution kernel of the last learning layer, [ alpha ], [ alpha]Represents a feature fusion function, M represents a feature after fusion by 1 × 1 convolutional layers, B represents a feature obtained by the last learning layer, and Fm-1And FmRespectively representing the input and output of the mth multi-scale residual block.
5. The method for reconstructing image resolution based on multi-scale pyramid network as claimed in claim 1, wherein said step S3 is performed as follows:
using a layer of transposition convolution layer to up-sample the deep-level features passing through the K multi-scale residual error modules to obtain a high-resolution image, wherein the expression is as follows:
IHR_conv=fdeconv(FK) (9)
in the above formula, IHR_convFor high resolution images after upsampling, fdeconvFor up-sampling operations, FKIs the output of the kth multi-scale residual module.
6. The method for reconstructing image resolution based on multi-scale pyramid network as claimed in claim 1, wherein said step S4 is performed as follows:
firstly, carrying out double cubic interpolation on a low-resolution image to obtain a high-resolution image IHR_bicuThen the high resolution image IHR_bicuWith up-sampled high-resolution image IHR_convAdding to obtain a depth map I with spatial resolution amplified by two timesHRThe expression is as follows:
IHR=IHR_bicu+IHR_deconv(10)
7. the method as claimed in any one of claims 1 to 6, wherein the pyramid network comprises a total of N levels, and if the input low-resolution image is a down-sampled 1/S times low-resolution image and S is an up-sampling scale factor, then N is log2S; each stage reconstructs an image output by a previous stage into a high-resolution image of the stage.
Priority Applications (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN202010108622.1A CN111402128A (en) | 2020-02-21 | 2020-02-21 | Image super-resolution reconstruction method based on multi-scale pyramid network |
Applications Claiming Priority (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN202010108622.1A CN111402128A (en) | 2020-02-21 | 2020-02-21 | Image super-resolution reconstruction method based on multi-scale pyramid network |
Publications (1)
Publication Number | Publication Date |
---|---|
CN111402128A true CN111402128A (en) | 2020-07-10 |
Family
ID=71430440
Family Applications (1)
Application Number | Title | Priority Date | Filing Date |
---|---|---|---|
CN202010108622.1A Pending CN111402128A (en) | 2020-02-21 | 2020-02-21 | Image super-resolution reconstruction method based on multi-scale pyramid network |
Country Status (1)
Country | Link |
---|---|
CN (1) | CN111402128A (en) |
Cited By (9)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN111861961A (en) * | 2020-07-25 | 2020-10-30 | 安徽理工大学 | Multi-scale residual error fusion model for single image super-resolution and restoration method thereof |
CN112070702A (en) * | 2020-09-14 | 2020-12-11 | 中南民族大学 | Image super-resolution reconstruction system and method for multi-scale residual error feature discrimination enhancement |
CN112634136A (en) * | 2020-12-24 | 2021-04-09 | 华南理工大学 | Image super-resolution method and system based on image characteristic quick splicing |
CN113034381A (en) * | 2021-02-08 | 2021-06-25 | 浙江大学 | Single image denoising method and device based on cavitated kernel prediction network |
CN113222821A (en) * | 2021-05-24 | 2021-08-06 | 南京航空航天大学 | Image super-resolution processing method for annular target detection |
CN113378972A (en) * | 2021-06-28 | 2021-09-10 | 成都恒创新星科技有限公司 | License plate recognition method and system in complex scene |
CN113421187A (en) * | 2021-06-10 | 2021-09-21 | 山东师范大学 | Super-resolution reconstruction method, system, storage medium and equipment |
CN113989800A (en) * | 2021-12-29 | 2022-01-28 | 南京南数数据运筹科学研究院有限公司 | Intestinal plexus auxiliary identification method based on improved progressive residual error network |
CN116385280A (en) * | 2023-01-09 | 2023-07-04 | 爱芯元智半导体(上海)有限公司 | Image noise reduction system and method and noise reduction neural network training method |
Citations (7)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
US20170293825A1 (en) * | 2016-04-08 | 2017-10-12 | Wuhan University | Method and system for reconstructing super-resolution 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 |
CN109767386A (en) * | 2018-12-22 | 2019-05-17 | 昆明理工大学 | A kind of rapid image super resolution ratio reconstruction method based on deep learning |
CN109829855A (en) * | 2019-01-23 | 2019-05-31 | 南京航空航天大学 | A kind of super resolution ratio reconstruction method based on fusion multi-level features figure |
CN109993701A (en) * | 2019-04-09 | 2019-07-09 | 福州大学 | A method of the depth map super-resolution rebuilding based on pyramid structure |
CN110197468A (en) * | 2019-06-06 | 2019-09-03 | 天津工业大学 | A kind of single image Super-resolution Reconstruction algorithm based on multiple dimensioned residual error learning network |
CN110473144A (en) * | 2019-08-07 | 2019-11-19 | 南京信息工程大学 | A kind of image super-resolution rebuilding method based on laplacian pyramid network |
-
2020
- 2020-02-21 CN CN202010108622.1A patent/CN111402128A/en active Pending
Patent Citations (7)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
US20170293825A1 (en) * | 2016-04-08 | 2017-10-12 | Wuhan University | Method and system for reconstructing super-resolution 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 |
CN109767386A (en) * | 2018-12-22 | 2019-05-17 | 昆明理工大学 | A kind of rapid image super resolution ratio reconstruction method based on deep learning |
CN109829855A (en) * | 2019-01-23 | 2019-05-31 | 南京航空航天大学 | A kind of super resolution ratio reconstruction method based on fusion multi-level features figure |
CN109993701A (en) * | 2019-04-09 | 2019-07-09 | 福州大学 | A method of the depth map super-resolution rebuilding based on pyramid structure |
CN110197468A (en) * | 2019-06-06 | 2019-09-03 | 天津工业大学 | A kind of single image Super-resolution Reconstruction algorithm based on multiple dimensioned residual error learning network |
CN110473144A (en) * | 2019-08-07 | 2019-11-19 | 南京信息工程大学 | A kind of image super-resolution rebuilding method based on laplacian pyramid network |
Non-Patent Citations (2)
Title |
---|
WEI-SHENG LAI 等: "Deep Laplacian Pyramid Networks for Fast and Accurate Super-Resolution", 《2017 IEEE CONFERENCE ON COMPUTER VISION AND PATTERN RECOGNITION》 * |
应自炉 等: "多尺度密集残差网络的单幅图像超分辨率重建", 《中国图象图形学报》 * |
Cited By (16)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN111861961A (en) * | 2020-07-25 | 2020-10-30 | 安徽理工大学 | Multi-scale residual error fusion model for single image super-resolution and restoration method thereof |
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 |
CN112070702A (en) * | 2020-09-14 | 2020-12-11 | 中南民族大学 | Image super-resolution reconstruction system and method for multi-scale residual error feature discrimination enhancement |
CN112070702B (en) * | 2020-09-14 | 2023-10-03 | 中南民族大学 | Image super-resolution reconstruction system and method for multi-scale residual error characteristic discrimination enhancement |
CN112634136A (en) * | 2020-12-24 | 2021-04-09 | 华南理工大学 | Image super-resolution method and system based on image characteristic quick splicing |
CN112634136B (en) * | 2020-12-24 | 2023-05-23 | 华南理工大学 | Image super-resolution method and system based on image feature rapid stitching |
CN113034381B (en) * | 2021-02-08 | 2022-06-21 | 浙江大学 | Single image denoising method and device based on cavitated kernel prediction network |
CN113034381A (en) * | 2021-02-08 | 2021-06-25 | 浙江大学 | Single image denoising method and device based on cavitated kernel prediction network |
CN113222821A (en) * | 2021-05-24 | 2021-08-06 | 南京航空航天大学 | Image super-resolution processing method for annular target detection |
CN113421187A (en) * | 2021-06-10 | 2021-09-21 | 山东师范大学 | Super-resolution reconstruction method, system, storage medium and equipment |
CN113421187B (en) * | 2021-06-10 | 2023-01-03 | 山东师范大学 | Super-resolution reconstruction method, system, storage medium and equipment |
CN113378972A (en) * | 2021-06-28 | 2021-09-10 | 成都恒创新星科技有限公司 | License plate recognition method and system in complex scene |
CN113378972B (en) * | 2021-06-28 | 2024-03-22 | 成都恒创新星科技有限公司 | License plate recognition method and system under complex scene |
CN113989800A (en) * | 2021-12-29 | 2022-01-28 | 南京南数数据运筹科学研究院有限公司 | Intestinal plexus auxiliary identification method based on improved progressive residual error network |
CN116385280A (en) * | 2023-01-09 | 2023-07-04 | 爱芯元智半导体(上海)有限公司 | Image noise reduction system and method and noise reduction neural network training method |
CN116385280B (en) * | 2023-01-09 | 2024-01-23 | 爱芯元智半导体(上海)有限公司 | Image noise reduction system and method and noise reduction neural network training method |
Similar Documents
Publication | Publication Date | Title |
---|---|---|
CN111402128A (en) | Image super-resolution reconstruction method based on multi-scale pyramid network | |
CN109903226B (en) | Image super-resolution reconstruction method based on symmetric residual convolution neural network | |
CN113096017B (en) | Image super-resolution reconstruction method based on depth coordinate attention network model | |
CN110992270A (en) | Multi-scale residual attention network image super-resolution reconstruction method based on attention | |
CN108475415B (en) | Method and system for image processing | |
CN112750082B (en) | Human face super-resolution method and system based on fusion attention mechanism | |
CN111369440B (en) | Model training and image super-resolution processing method, device, terminal and storage medium | |
CN110348487B (en) | Hyperspectral image compression method and device based on deep learning | |
CN113139907A (en) | Generation method, system, device and storage medium for visual resolution enhancement | |
CN110689483B (en) | Image super-resolution reconstruction method based on depth residual error network and storage medium | |
CN109146788A (en) | Super-resolution image reconstruction method and device based on deep learning | |
CN111768340B (en) | Super-resolution image reconstruction method and system based on dense multipath network | |
CN111652804B (en) | Super-resolution reconstruction method based on expansion convolution pyramid and bottleneck network | |
CN110349087B (en) | RGB-D image high-quality grid generation method based on adaptive convolution | |
CN115953303B (en) | Multi-scale image compressed sensing reconstruction method and system combining channel attention | |
CN112700460B (en) | Image segmentation method and system | |
CN114331831A (en) | Light-weight single-image super-resolution reconstruction method | |
CN116433914A (en) | Two-dimensional medical image segmentation method and system | |
CN113554058A (en) | Method, system, device and storage medium for enhancing resolution of visual target image | |
CN104899835A (en) | Super-resolution processing method for image based on blind fuzzy estimation and anchoring space mapping | |
CN116664397B (en) | TransSR-Net structured image super-resolution reconstruction method | |
CN111654621B (en) | Dual-focus camera continuous digital zooming method based on convolutional neural network model | |
CN112949636A (en) | License plate super-resolution identification method and system and computer readable medium | |
Hui et al. | Two-stage convolutional network for image super-resolution | |
CN115936992A (en) | Garbage image super-resolution method and system of lightweight transform |
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 | ||
RJ01 | Rejection of invention patent application after publication | ||
RJ01 | Rejection of invention patent application after publication |
Application publication date: 20200710 |