CN108734660A - A kind of image super-resolution rebuilding method and device based on deep learning - Google Patents

A kind of image super-resolution rebuilding method and device based on deep learning Download PDF

Info

Publication number
CN108734660A
CN108734660A CN201810511777.2A CN201810511777A CN108734660A CN 108734660 A CN108734660 A CN 108734660A CN 201810511777 A CN201810511777 A CN 201810511777A CN 108734660 A CN108734660 A CN 108734660A
Authority
CN
China
Prior art keywords
resolution
multiple dimensioned
image
low
residual error
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
Application number
CN201810511777.2A
Other languages
Chinese (zh)
Inventor
陈涛
王洪剑
林江
Current Assignee (The listed assignees may be inaccurate. Google has not performed a legal analysis and makes no representation or warranty as to the accuracy of the list.)
SHANGHAI TONGTU SEMICONDUCTOR TECHNOLOGY Co Ltd
Original Assignee
SHANGHAI TONGTU SEMICONDUCTOR TECHNOLOGY Co Ltd
Priority date (The priority date 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 date listed.)
Filing date
Publication date
Application filed by SHANGHAI TONGTU SEMICONDUCTOR TECHNOLOGY Co Ltd filed Critical SHANGHAI TONGTU SEMICONDUCTOR TECHNOLOGY Co Ltd
Priority to CN201810511777.2A priority Critical patent/CN108734660A/en
Publication of CN108734660A publication Critical patent/CN108734660A/en
Pending legal-status Critical Current

Links

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
    • 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
    • G06N3/084—Backpropagation, e.g. using gradient descent
    • 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
    • G06T7/10—Segmentation; Edge detection
    • G06T7/11—Region-based segmentation
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
    • G06T2207/00—Indexing scheme for image analysis or image enhancement
    • G06T2207/20—Special algorithmic details
    • G06T2207/20212—Image combination
    • G06T2207/20221—Image fusion; Image merging

Landscapes

  • Engineering & Computer Science (AREA)
  • Physics & Mathematics (AREA)
  • Theoretical Computer Science (AREA)
  • General Physics & Mathematics (AREA)
  • Computational Linguistics (AREA)
  • General Health & Medical Sciences (AREA)
  • Life Sciences & Earth Sciences (AREA)
  • Artificial Intelligence (AREA)
  • Biomedical Technology (AREA)
  • Biophysics (AREA)
  • Computer Vision & Pattern Recognition (AREA)
  • Data Mining & Analysis (AREA)
  • Evolutionary Computation (AREA)
  • Health & Medical Sciences (AREA)
  • Molecular Biology (AREA)
  • Computing Systems (AREA)
  • General Engineering & Computer Science (AREA)
  • Mathematical Physics (AREA)
  • Software Systems (AREA)
  • Image Analysis (AREA)
  • Image Processing (AREA)

Abstract

The invention discloses a kind of image super-resolution rebuilding method and device based on deep learning, described method includes following steps：Step S1 builds multiple dimensioned residual error convolutional neural networks, and carrying out multiple dimensioned residual error convolutional neural networks using training sample trains, to obtain the nonlinear correspondence relation of low high-definition picture block；Step S2 carries out Nonlinear Mapping using trained multiple dimensioned residual error convolutional neural networks to the low-resolution image of input, with the high-definition picture reconstructed, through the invention, the super-resolution image of reconstruction can be made closer to true image.

Description

A kind of image super-resolution rebuilding method and device based on deep learning
Technical field
The present invention relates to Digital Image Processing, deep learning and artificial intelligence fields, more particularly to one kind based on deep Spend the image super-resolution rebuilding method and device of study.
Background technology
The super-resolution rebuilding of image refers to being eliminated due to imaging system using the method for signal processing and computer software Deteriroation of image quality caused by the factors such as inaccurate, motion blur and non-ideal sampling is focused, it is high-resolution clear to obtain Clear image.It has important application valence in video monitoring, satellite image, medical image and some high-resolution display fields etc. Value.
Since deep neural network has superpower nonlinear characteristic to indicate ability, the super resolution technology energy based on deep learning Enough effects for improving image super-resolution well, at this stage, the super-resolution image reconstruction effect based on deep learning is Pervious non-deep learning super-resolution technique is surmounted.But the image of resolution reconstruction at this stage and true image There are still larger gaps, therefore, it is really necessary to propose a kind of technological means, so that the super-resolution image rebuild is closer True image.
Invention content
In order to overcome the deficiencies of the above existing technologies, purpose of the present invention is to provide a kind of figures based on deep learning As super resolution ratio reconstruction method and device, by combining residual error network and multiple dimensioned convolution advantage, using symmetrical more convolution kernels, make The super-resolution image of reconstruction is closer to true image.
In view of the above and other objects, the present invention proposes a kind of image super-resolution rebuilding method based on deep learning, Include the following steps：
Step S1 builds multiple dimensioned residual error convolutional neural networks, and multiple dimensioned residual error convolutional Neural is carried out using training sample Network training, to obtain the nonlinear correspondence relation of low high-definition picture block；
Step S2 carries out the low-resolution image of input using trained multiple dimensioned residual error convolutional neural networks non-thread Property mapping, with the high-definition picture reconstructed.
Preferably, step S1 further comprises：
Step S100 obtains training sample by the high-resolution and low-resolution image that several include identical content；
Step S101 builds multiple dimensioned residual error convolutional neural networks, and multiple dimensioned residual error is carried out using the training sample of acquisition Convolutional neural networks learn, to establish the nonlinear correspondence relation between low-resolution image block and high-definition picture block.
Preferably, in step S100, the high-resolution and low-resolution image that several include identical content is obtained, to every Width low-resolution image extracts low-resolution image block, extraction and low-resolution image block position in corresponding high-definition picture Corresponding high-definition picture block is set, to obtain training sample.
Preferably, in step S100, the high-definition picture of high quality is first chosen, then to high-definition picture elder generation Down-sampling up-samples again, obtains the low-resolution image with the high-definition picture same size.
Preferably, in step S101, the input of the multiple dimensioned residual error convolutional neural networks is and high-definition picture An equal amount of low-resolution image block exports the high-definition picture block for reconstruct, the multiple dimensioned residual error convolutional Neural net If network includes dried layer, the 1st layer and last layer are convolutional layer, if intermediate dried layer is multiple dimensioned residual error layer, the 1st layer of output with most The output of later layer is added as the high-resolution result of reconstruct.
Preferably, if the centre dried layer of the multiple dimensioned residual error convolutional neural networks is with mutually isostructural multiple dimensioned residual Poor unit, each multiple dimensioned residual unit include two multiple dimensioned convolution units, and centre is activation primitive, each multiple dimensioned convolution Unit includes the convolutional network of 3 scales, and first scale is the convolutional network that core size is 1 × 1, and second scale is that core is big The small convolutional network for being 3 × 3, third scale are 3 × 3 pondizations and 3 × 3 convolutional networks, and the channel figure of three scales passes through grade Connection mode composition characteristic figure.
Preferably, the multiple dimensioned residual error convolutional neural networks are trained study using following loss function：
Wherein, N is training sample number, YiFor high-definition picture block,For the image block of network reconnection, | | | |1Table Show L1 norms.
Preferably, in step S101, the damage is minimized using the backpropagation of stochastic gradient descent method and network Function is lost, adjustment is optimized to the convolution nuclear parameter of multiple dimensioned residual error convolutional neural networks.
Preferably, step S2 further comprises：
Step S200 pre-processes the low-resolution image of input, to obtain the low resolution of target image size Image；
The low-resolution image is divided into low-resolution image block to be reconstructed by step S201；
Step S202, using trained multiple dimensioned residual error convolutional neural networks to low-resolution image block respectively to be reconstructed Carry out resolution reconstruction；
Step S203 carries out carry out fusion treatment to the high-definition picture block of all reconstruct of acquisition, is inputted The reconstruct high-definition picture of image.
In order to achieve the above objectives, the present invention also provides a kind of image super-resolution rebuilding device based on deep learning, packet It includes：
Training unit carries out multiple dimensioned residual error volume for building multiple dimensioned residual error convolutional neural networks using training sample Product neural metwork training, to obtain the nonlinear correspondence relation of low high-definition picture block；
Reconstruction unit, for using trained multiple dimensioned residual error convolutional neural networks to the low-resolution image of input into Row Nonlinear Mapping, with the high-definition picture reconstructed.
The prior art is compared, and the present invention a kind of image super-resolution rebuilding method and device based on deep learning pass through structure It builds multiple dimensioned residual error convolutional neural networks, and combines residual error network and multiple dimensioned convolution advantage, by using symmetrical more convolution kernels, The super-resolution reconstruction effect closer to true high-definition picture is achieved, it is closer with the super-resolution image for reaching reconstruction The purpose of true image.
Description of the drawings
Fig. 1 is a kind of step flow chart of the image super-resolution rebuilding method based on deep learning of the present invention；
Fig. 2 is the detailed flowchart of step S1 in the specific embodiment of the invention；
Fig. 3 is the configuration diagram of multiple dimensioned residual error convolutional neural networks constructed in the specific embodiment of the invention；
Fig. 4 is the structural schematic diagram of multiple dimensioned residual unit in the specific embodiment of the invention；
Fig. 5 is the structural schematic diagram of multiple dimensioned convolution unit in the specific embodiment of the invention；
Fig. 6 is the detailed flowchart of step S2 in the specific embodiment of the invention；
Fig. 7 is a kind of system architecture diagram of the image super-resolution rebuilding device based on deep learning of the present invention；
Fig. 8 is the detail structure chart of training unit in the specific embodiment of the invention；
Fig. 9 is the detail structure chart of reconstruction unit in the specific embodiment of the invention.
Specific implementation mode
Below by way of specific specific example and embodiments of the present invention are described with reference to the drawings, those skilled in the art can Understand the further advantage and effect of the present invention easily by content disclosed in the present specification.The present invention can also pass through other differences Specific example implemented or applied, details in this specification can also be based on different perspectives and applications, without departing substantially from Various modifications and change are carried out under the spirit of the present invention.
Depth convolutional neural networks structure is the core of image super-resolution, it has most critical to final quality reconstruction Influence.The present invention is directed to key problem, it is proposed that utilizes multiple dimensioned residual error convolutional neural networks so that super-resolution reconstruction is imitated Level of the fruit close to true high-definition picture.
Fig. 1 is a kind of step flow chart of the image super-resolution rebuilding method based on deep learning of the present invention.Such as Fig. 1 institutes Show, a kind of image super-resolution rebuilding method based on deep learning of the present invention includes the following steps：
Step S1 builds multiple dimensioned residual error convolutional neural networks, and multiple dimensioned residual error convolutional Neural is carried out using training sample Network training, to obtain the nonlinear correspondence relation of low high-definition picture block.In the specific embodiment of the invention, by more Width includes that the high-resolution and low-resolution image of identical content obtains the multiple dimensioned residual error convolutional neural networks of training sample progress It practises, to obtain the nonlinear correspondence relation of low high-definition picture block.
Specifically, as shown in Fig. 2, step S1 further comprises：
Step S100 obtains training sample by the high-resolution and low-resolution image that several include identical content.At this In invention specific embodiment, the high-resolution and low-resolution image that several include identical content is obtained, to every width low resolution figure As extracting low-resolution image block by the method for sliding window, extraction and low resolution in correspondence image high-definition picture The corresponding high-definition picture block of tile location, to obtain training sample.
In fact, due to the high-definition picture that can not often obtain Same Scene simultaneously and corresponding low resolution figure Picture, the present invention take analog form to obtain.Specifically, the high-definition picture of high quality is first chosen, for example 1000 width can be chosen The high-definition picture that size is 1920 × 1080；Then full resolution pricture elder generation down-sampling is up-sampled again, is obtained and high-resolution The low-resolution image of rate image same size.The method of upper down-sampling can be bilinearity, cubic spline, iterative backward projection (IBP, iterative back-projection) etc..Finally use a certain size step-length (such as 60) by corresponding height Image in different resolution is divided into the image block of 64 × 64 sizes respectively, as training image blocks sample
Step S101 builds multiple dimensioned residual error convolutional neural networks, and multiple dimensioned residual error is carried out using the training sample of acquisition Convolutional neural networks learn, to establish the nonlinear correspondence relation between low-resolution image block and high-definition picture block.
Since residual error network is avoided that gradient disappears, so as to design the network of deeper number, preferably indicate non-thread Property correspondence.In addition objects in images size is uncertain, and Analysis On Multi-scale Features can completely portray object.Symmetrical multireel product Core increases the robustness of character representation, and therefore, the present invention establishes multiple dimensioned residual error convolutional neural networks, in conjunction with residual error network With multiple dimensioned convolution advantage, symmetrical more convolution kernels are used.
Multiple dimensioned residual error convolutional neural networks constructed by the present invention are as shown in figure 3, its input is and high-definition picture An equal amount of low-resolution image block exports the high-definition picture block for reconstruct, the multiple dimensioned residual error convolutional neural networks One shares 34 layers, the 1st layer and the 34th layer as convolutional layer, and intermediate 32 layers are multiple dimensioned residual error layer.
1st layer and the 34th layer is all 3 × 3 convolutional layers.1st layer of output is added as the height of reconstruct with the 34th layer of output Resolution ratio is as a result, their port number is all 3, corresponding tri- colors of RGB.
2nd layer to the 33rd layer is with mutually isostructural multiple dimensioned residual unit, internal structure is as shown in Figure 4.They Channel number all be 256.Each multiple dimensioned residual unit contains two multiple dimensioned convolution units, and centre is activation primitive. Activation primitive is common ReLU (Rectified Linear Units) function, is defined as follows：
F (x)=max (0, x),
Wherein, x is image convolution result.
Multiple dimensioned convolution unit is as shown in Figure 5.Multiple dimensioned convolution unit includes the convolutional network of 3 scales, first ruler Degree is the convolutional network that core size is 1 × 1, and output channel figure number is 96, and second scale is the convolution net that core size is 3 × 3 Network, channel figure number are 96, and third scale is 3 × 3 ponds (taking 3x3 maximum values) and 3 × 3 convolutional networks, and channel figure number is 64, The channel figure of three scales constitutes the characteristic pattern in 256 channels by cascade system.
Usually, four basic kinds can be symmetrically divided into：Symmetrical, horizontal symmetrical, vertical symmetry, diagonal line pair Claim, then this four symmetrically by combination, 2 can be obtained4=16 kinds symmetrical, in this way, second ruler in multiple dimensioned convolution unit It is original 1 that degree and third scale, which need 3 × 3 convolution kernel numbers preserved only,
16.When carrying out convolution operation, 3 × 3 convolution kernels of storage obtain 16 times of required storages by symmetry transformation 3 × 3 convolution kernels of number.
Specifically, the present invention is as follows by high-definition picture block and the image block counting loss function of reconstruction：
Wherein, N is training sample number, YiFor high-definition picture block,For the image block of network reconnection, | | | |1Table Show L1 norms.Counting loss function is then based on the backpropagation of stochastic gradient descent method and network to adjust multiple dimensioned residual error The convolution nuclear parameter of convolutional neural networks.This process is constantly repeated, counting loss functional value adjusts convolution nuclear parameter, until Until loss function value is very small.
Then, it is minimized using the backpropagation (BP, Back Propagation) of stochastic gradient descent method and network Loss function optimizes adjustment to the convolution nuclear parameter of multiple dimensioned residual error convolutional neural networks, constantly repeats this process, meter Loss function value is calculated, convolution nuclear parameter is adjusted, until loss function value is very small.
Step S2 carries out the low-resolution image of input using trained multiple dimensioned residual error convolutional neural networks non-thread Property mapping, to obtain the reconstruct high-definition picture of high quality.
Specifically, as shown in fig. 6, step S2 further comprises：
Step S200 pre-processes the low-resolution image of input, to obtain the low resolution of target image size Image.In the specific embodiment of the invention, 3 channel RGB color images of the low resolution of input are up-sampled to obtain target figure As the low-resolution image of size.Here top sampling method is identical as top sampling method when neural metwork training, for 2 times Image super-resolution is it is necessary to which to 2 times of up-samplings of input low-resolution image, amplification method can be cubic spline interpolation, herein It will not go into details.
Low-resolution image is divided into low-resolution image block to be reconstructed by step S201.Specifically, by low resolution Image is divided into low-resolution image block to be reconstructed according to certain step-length, such as the size of image block can be 64 × 64, Horizontal step-length and vertical step-length can all be 60, but invention is not limited thereto.
Step S202, using trained multiple dimensioned residual error convolutional neural networks to low-resolution image block respectively to be reconstructed Carry out resolution reconstruction.Specifically, the low-resolution image block to be reconstructed divided is inputted into trained multiple dimensioned residual error Convolutional neural networks, by network mapping, output is the high-definition picture block of reconstruct, that is, inputs the low resolution for 3 channels Rate RGB color image block obtains high-resolution RGB image block by 34 layers of Processing with Neural Network.
Step S203 carries out carry out fusion treatment to the high-definition picture block of all reconstruct of acquisition, is inputted The reconstruct high-definition picture of image.Specifically, the high-definition picture block of all reconstruct of acquisition is pressed into low resolution figure It is averaged after being superimposed as block corresponding position, obtains the reconstruct high-definition picture of input picture.
Fig. 7 is a kind of system architecture diagram of the image super-resolution rebuilding device based on deep learning of the present invention.Step stream Cheng Tu.As shown in fig. 7, a kind of image super-resolution rebuilding device based on deep learning of the present invention, including：
Training unit 70, for obtain several include identical content high-resolution and low-resolution image carry out it is multiple dimensioned Residual error convolutional neural networks are trained, i.e., obtain training sample by the high-resolution and low-resolution image that several include identical content The multiple dimensioned residual error convolutional neural networks study of this progress, to obtain the nonlinear correspondence relation of low high-definition picture block.
Specifically, as shown in figure 8, training unit 70 further comprises：
Training sample acquiring unit 701, for obtaining training sample.In the specific embodiment of the invention, training sample obtains Take unit 701 by obtaining the high-resolution and low-resolution image that several include identical content, it is logical to every width low-resolution image Cross the method extraction low-resolution image block of sliding window, extraction and low-resolution image in correspondence image high-definition picture The corresponding high-definition picture block in block position, to obtain training sample.
In fact, due to the high-definition picture that can not often obtain Same Scene simultaneously and corresponding low resolution figure Picture, the present invention take analog form to obtain.Specifically, training sample acquiring unit 701 first chooses the high resolution graphics of high quality Picture, for example the high-definition picture that 1000 width sizes are 1920 × 1080 can be chosen；Then again to full resolution pricture elder generation down-sampling Up-sampling, obtains the low-resolution image with high-definition picture same size.The method of upper down-sampling can be bilinearity, three Secondary batten, iterative backward projection (IBP, iterative back-projection) etc..Finally use a certain size step-length Corresponding high-low resolution image is divided into the image block of 64 × 64 sizes by (such as 60) respectively, as training image blocks sample
Network struction and training unit 702 utilize the instruction of acquisition for building multiple dimensioned residual error convolutional neural networks Practice sample and carry out multiple dimensioned residual error convolutional neural networks study, to establish between low-resolution image block and high-definition picture block Nonlinear correspondence relation.
It is convolutional layer that multiple dimensioned residual error convolutional neural networks one constructed by the present invention, which share 34 layers, the 1st layer and the 34th layer, Intermediate 32 layers are multiple dimensioned residual error layer, input be with an equal amount of low-resolution image block of high-definition picture, export and be The high-definition picture block of reconstruct.1st layer and the 34th layer is 3 × 3 convolutional layers.1st layer of output is added with the 34th layer of output For reconstruct high-resolution as a result, their port number is all 3, tri- colors of corresponding RGB.
2nd layer to the 33rd layer is, with mutually isostructural multiple dimensioned residual unit, their channel number is all 256.Often A multiple dimensioned residual unit contains two multiple dimensioned convolution units, and centre is activation primitive.Activation primitive is common ReLU (Rectified Linear Units) function, is defined as follows：
F (x)=max (0, x).
Each multiple dimensioned convolution unit includes the convolutional network of 3 scales, and first scale is the convolution that core size is 1x1 Network, output channel figure number are 96, and second scale is the convolutional network that core size is 3x3, and channel figure number is 96, third ruler Degree is 3 × 3 ponds (taking 3x3 maximum values) and 3x3 convolutional networks, and channel figure number is 64, and the channel figure of three scales passes through cascade Mode constitutes the characteristic pattern in 256 channels.
Usually, four basic kinds can be symmetrically divided into：Symmetrical, horizontal symmetrical, vertical symmetry, diagonal line pair Claim, then this four symmetrically by combination, 2 can be obtained4=16 kinds symmetrical, in this way, second ruler in multiple dimensioned convolution unit It is original that degree and third scale, which need 3 × 3 convolution kernel numbers preserved only,When carrying out convolution operation, the 3 of storage × 3 convolution kernels obtain 3 × 3 convolution kernels of 16 times of required storage numbers by symmetry transformation.
Specifically, for better resolution reconstruction effect, the present invention passes through high-definition picture block and the figure of reconstruction As block counting loss function is as follows：
Wherein, N is training sample number, YiFor high-definition picture block,For the image block of network reconnection, | | | |1Table Show L1 norms.
Then, network struction and training unit 702 using stochastic gradient descent method and network backpropagation (BP, Back Propagation) loss function is minimized, the convolution nuclear parameter of multiple dimensioned residual error convolutional neural networks is carried out excellent Change adjustment, constantly repeat this process, counting loss functional value adjusts convolution nuclear parameter, is until loss function value is very small Only.
Reconstruction unit 71, for the low-resolution image using trained multiple dimensioned residual error convolutional neural networks to input Nonlinear Mapping is carried out, to obtain the reconstruct high-definition picture of high quality.
Specifically, as shown in figure 9, reconstruction unit 71 further comprises：
Image pre-processing unit 710 is pre-processed for the low-resolution image to input, big to obtain target image Small low-resolution image.In the specific embodiment of the invention, image pre-processing unit 710 leads to the 3 of the low resolution of input Road RGB color image up-samples to obtain the low-resolution image of target image size.
Image segmentation unit 711, for pretreated low-resolution image to be divided into low resolution figure to be reconstructed As block.Specifically, low-resolution image after pretreatment is divided by image segmentation unit 711 according to certain step-length waits for weight The low-resolution image block of structure, such as the size of image block can be 64 × 64, horizontal step-length and vertical step-length can all be 60.
Resolution reconstruction unit 712, for using trained multiple dimensioned residual error convolutional neural networks to respectively to be reconstructed Low-resolution image block carries out resolution reconstruction.Specifically, the low resolution to be reconstructed that resolution reconstruction unit 712 will have been divided Rate image block inputs trained multiple dimensioned residual error convolutional neural networks, and by network mapping, output is the high-resolution of reconstruct Rate image block, that is, input and obtain height by 34 layers of Processing with Neural Network for the low resolution RGB color image block in 3 channels Differentiate RGB image block.
Fusion treatment unit 713, the high-definition picture block for all reconstruct to acquisition carry out fusion treatment, obtain To the reconstruct high-definition picture of input picture.Specifically, fusion treatment unit 713 is by the high-resolution of all reconstruct of acquisition Rate image block is averaged after being superimposed by low-resolution image block corresponding position, obtains the reconstruct high resolution graphics of input picture Picture.
In conclusion the present invention a kind of image super-resolution rebuilding method and device based on deep learning are more by building Scale residual error convolutional neural networks, and residual error network and multiple dimensioned convolution advantage are combined, by using symmetrical more convolution kernels, obtain Closer to the super-resolution reconstruction effect of true high-definition picture, to reach the super-resolution image of reconstruction closer to true Image purpose.
The above-described embodiments merely illustrate the principles and effects of the present invention, and is not intended to limit the present invention.Any Field technology personnel can without violating the spirit and scope of the present invention, and modifications and changes are made to the above embodiments.Therefore, The scope of the present invention, should be as listed in the claims.

Claims (10)

1. a kind of image super-resolution rebuilding method based on deep learning, includes the following steps：
Step S1 builds multiple dimensioned residual error convolutional neural networks, and multiple dimensioned residual error convolutional neural networks are carried out using training sample Training, to obtain the nonlinear correspondence relation of low high-definition picture block；
Step S2 carries out non-linear reflect using trained multiple dimensioned residual error convolutional neural networks to the low-resolution image of input It penetrates, with the high-definition picture reconstructed.
2. a kind of image super-resolution rebuilding method based on deep learning as described in claim 1, which is characterized in that step S1 further comprises：
Step S100 obtains training sample by the high-resolution and low-resolution image that several include identical content；
Step S101 builds multiple dimensioned residual error convolutional neural networks, and multiple dimensioned residual error convolution is carried out using the training sample of acquisition Neural network learning, to establish the nonlinear correspondence relation between low-resolution image block and high-definition picture block.
3. a kind of image super-resolution rebuilding method based on deep learning as claimed in claim 2, it is characterised in that：Yu Bu In rapid S100, the high-resolution and low-resolution image that several include identical content is obtained, every width low-resolution image is extracted low Image in different resolution block extracts high-definition picture corresponding with low-resolution image block position in corresponding high-definition picture Block, to obtain training sample.
4. a kind of image super-resolution rebuilding method based on deep learning as claimed in claim 3, it is characterised in that：Yu Bu In rapid S100, the high-definition picture of high quality is first chosen, then down-sampling up-samples again to the high-definition picture elder generation, obtains With the low-resolution image of the high-definition picture same size.
5. a kind of image super-resolution rebuilding method based on deep learning as claimed in claim 2, it is characterised in that：Yu Bu In rapid S101, the input of the multiple dimensioned residual error convolutional neural networks is and an equal amount of low resolution figure of high-definition picture As block, export the high-definition picture block for reconstruct, if the multiple dimensioned residual error convolutional neural networks include dried layer, the 1st layer and Last layer is convolutional layer, if intermediate dried layer is multiple dimensioned residual error layer, the 1st layer of output and the output of last layer are added as The high-resolution result of reconstruct.
6. a kind of image super-resolution rebuilding method based on deep learning as claimed in claim 5, it is characterised in that：It is described It is each multiple dimensioned residual if the centre dried layer of multiple dimensioned residual error convolutional neural networks is with mutually isostructural multiple dimensioned residual unit Poor unit includes two multiple dimensioned convolution units, and centre is activation primitive, and each multiple dimensioned convolution unit includes the volume of 3 scales Product network, first scale is the convolutional network that core size is 1 × 1, and second scale is the convolutional network that core size is 3 × 3, Third scale is 3 × 3 pondizations and 3 × 3 convolutional networks, and the channel figure of three scales passes through cascade system composition characteristic figure.
7. a kind of image super-resolution rebuilding method based on deep learning as claimed in claim 6, which is characterized in that described Multiple dimensioned residual error convolutional neural networks are trained study using following loss function：
Wherein, N is training sample number, YiFor high-definition picture block,For the image block of network reconnection, | | | |1Indicate L1 Norm.
8. a kind of image super-resolution rebuilding method based on deep learning as claimed in claim 7, it is characterised in that:Yu Bu In rapid S101, the loss function is minimized using the backpropagation of stochastic gradient descent method and network, to multiple dimensioned residual error The convolution nuclear parameter of convolutional neural networks optimizes adjustment.
9. a kind of image super-resolution rebuilding method based on deep learning as described in claim 1, it is characterised in that:Step S2 further comprises：
Step S200 pre-processes the low-resolution image of input, to obtain the low-resolution image of target image size；
The low-resolution image is divided into low-resolution image block to be reconstructed by step S201；
Step S202 carries out low-resolution image block respectively to be reconstructed using trained multiple dimensioned residual error convolutional neural networks Resolution reconstruction；
Step S203 carries out carry out fusion treatment to the high-definition picture block of all reconstruct of acquisition, obtains input picture Reconstruct high-definition picture.
10. a kind of image super-resolution rebuilding device based on deep learning, including：
Training unit carries out multiple dimensioned residual error convolution god for building multiple dimensioned residual error convolutional neural networks using training sample Through network training, to obtain the nonlinear correspondence relation of low high-definition picture block；
Reconstruction unit, it is non-for being carried out to the low-resolution image of input using trained multiple dimensioned residual error convolutional neural networks Linear Mapping, with the high-definition picture reconstructed.
CN201810511777.2A 2018-05-25 2018-05-25 A kind of image super-resolution rebuilding method and device based on deep learning Pending CN108734660A (en)

Priority Applications (1)

Application Number Priority Date Filing Date Title
CN201810511777.2A CN108734660A (en) 2018-05-25 2018-05-25 A kind of image super-resolution rebuilding method and device based on deep learning

Applications Claiming Priority (1)

Application Number Priority Date Filing Date Title
CN201810511777.2A CN108734660A (en) 2018-05-25 2018-05-25 A kind of image super-resolution rebuilding method and device based on deep learning

Publications (1)

Publication Number Publication Date
CN108734660A true CN108734660A (en) 2018-11-02

Family

ID=63936126

Family Applications (1)

Application Number Title Priority Date Filing Date
CN201810511777.2A Pending CN108734660A (en) 2018-05-25 2018-05-25 A kind of image super-resolution rebuilding method and device based on deep learning

Country Status (1)

Country Link
CN (1) CN108734660A (en)

Cited By (71)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN109375186A (en) * 2018-11-22 2019-02-22 中国人民解放军海军航空大学 Radar target identification method based on the multiple dimensioned one-dimensional convolutional neural networks of depth residual error
CN109493308A (en) * 2018-11-14 2019-03-19 吉林大学 The medical image synthesis and classification method for generating confrontation network are differentiated based on condition more
CN109584162A (en) * 2018-11-30 2019-04-05 江苏网进科技股份有限公司 A method of based on the image super-resolution reconstruct for generating network
CN109685717A (en) * 2018-12-14 2019-04-26 厦门理工学院 Image super-resolution rebuilding method, device and electronic equipment
CN109887018A (en) * 2019-02-28 2019-06-14 中国计量大学 A kind of photon 3D imaging system based on deep learning
CN109903219A (en) * 2019-02-28 2019-06-18 深圳市商汤科技有限公司 Image processing method and device, electronic equipment, computer readable storage medium
CN109903226A (en) * 2019-01-30 2019-06-18 天津城建大学 Image super-resolution rebuilding method based on symmetrical residual error convolutional neural networks
CN109903228A (en) * 2019-02-28 2019-06-18 合肥工业大学 A kind of image super-resolution rebuilding method based on convolutional neural networks
CN109903225A (en) * 2019-01-29 2019-06-18 中南大学 A kind of medical image Enhancement Method based on deep learning
CN109920013A (en) * 2019-01-30 2019-06-21 北京交通大学 Image reconstructing method and device based on gradual convolution measurement network
CN109934771A (en) * 2019-03-08 2019-06-25 北京航空航天大学 Unsupervised Remote sensed image super-resolution reconstruction method based on Recognition with Recurrent Neural Network
CN110044262A (en) * 2019-05-09 2019-07-23 哈尔滨理工大学 Contactless precision measuring instrument and measurement method based on image super-resolution rebuilding
CN110111256A (en) * 2019-04-28 2019-08-09 西安电子科技大学 Image Super-resolution Reconstruction method based on residual error distillation network
CN110111257A (en) * 2019-05-08 2019-08-09 哈尔滨工程大学 A kind of super resolution image reconstruction method based on the weighting of feature channel adaptive
CN110175953A (en) * 2019-05-24 2019-08-27 鹏城实验室 A kind of image super-resolution method and system
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
CN110211038A (en) * 2019-04-29 2019-09-06 南京航空航天大学 Super resolution ratio reconstruction method based on dirac residual error deep neural network
CN110232653A (en) * 2018-12-12 2019-09-13 天津大学青岛海洋技术研究院 The quick light-duty intensive residual error network of super-resolution rebuilding
CN110264410A (en) * 2019-05-07 2019-09-20 西安理工大学 A kind of image super-resolution rebuilding method based on minutia
CN110288524A (en) * 2019-05-09 2019-09-27 广东启迪图卫科技股份有限公司 Deep learning super-resolution method based on enhanced up-sampling and discrimination syncretizing mechanism
CN110310227A (en) * 2019-06-27 2019-10-08 电子科技大学 A kind of image super-resolution rebuilding method decomposed based on high and low frequency information
CN110415201A (en) * 2019-07-30 2019-11-05 北京理工大学 Single exposure super-resolution imaging method and device based on structure light and deep learning
CN110473247A (en) * 2019-07-30 2019-11-19 中国科学院空间应用工程与技术中心 Solid matching method, device and storage medium
CN110580681A (en) * 2019-09-12 2019-12-17 杭州海睿博研科技有限公司 high-resolution cardiac motion pattern analysis device and method
CN110930306A (en) * 2019-10-28 2020-03-27 杭州电子科技大学 Depth map super-resolution reconstruction network construction method based on non-local perception
CN110929811A (en) * 2019-10-21 2020-03-27 北京理工大学 Deep learning method for acquiring high-resolution data of full-waveform laser radar
CN111027487A (en) * 2019-12-11 2020-04-17 山东大学 Behavior recognition system, method, medium, and apparatus based on multi-convolution kernel residual network
CN111062434A (en) * 2019-12-13 2020-04-24 国网重庆市电力公司永川供电分公司 Multi-scale fusion detection method for unmanned aerial vehicle inspection
CN111062867A (en) * 2019-11-21 2020-04-24 浙江大华技术股份有限公司 Video super-resolution reconstruction method
CN111080516A (en) * 2019-11-26 2020-04-28 广东石油化工学院 Super-resolution image reconstruction method based on self-sampling enhancement
CN111160413A (en) * 2019-12-12 2020-05-15 天津大学 Thyroid nodule classification method based on multi-scale feature fusion
CN111223046A (en) * 2019-11-20 2020-06-02 中国科学院遥感与数字地球研究所 Image super-resolution reconstruction method and device
CN111239731A (en) * 2020-01-06 2020-06-05 南京航空航天大学 Synthetic aperture radar rapid imaging method and device based on neural network
CN111260558A (en) * 2020-01-22 2020-06-09 武汉大学 Image super-resolution network model with variable magnification
CN111353940A (en) * 2020-03-31 2020-06-30 成都信息工程大学 Image super-resolution reconstruction method based on deep learning iterative up-down sampling
CN111369466A (en) * 2020-03-05 2020-07-03 福建帝视信息科技有限公司 Image distortion correction enhancement method of convolutional neural network based on deformable convolution
CN111402131A (en) * 2020-03-10 2020-07-10 北京师范大学 Method for acquiring super-resolution land cover classification map based on deep learning
CN111402128A (en) * 2020-02-21 2020-07-10 华南理工大学 Image super-resolution reconstruction method based on multi-scale pyramid network
CN111415299A (en) * 2020-03-26 2020-07-14 浙江科技学院 High-resolution image style migration method
CN111476719A (en) * 2020-05-06 2020-07-31 Oppo广东移动通信有限公司 Image processing method, image processing device, computer equipment and storage medium
CN111476714A (en) * 2020-03-30 2020-07-31 清华大学 Cross-scale image splicing method and device based on PSV neural network
CN111521989A (en) * 2019-02-01 2020-08-11 通用汽车环球科技运作有限责任公司 Deep learning for super-resolution in radar systems
CN111562305A (en) * 2019-02-14 2020-08-21 爱科维申科技（天津）有限公司 Automobile shock absorber piston defect detection method based on electromagnetic tomography technology
CN111598778A (en) * 2020-05-13 2020-08-28 云南电网有限责任公司电力科学研究院 Insulator image super-resolution reconstruction method
CN111598808A (en) * 2020-05-18 2020-08-28 腾讯科技（深圳）有限公司 Image processing method, device and equipment and training method thereof
CN111640067A (en) * 2020-06-10 2020-09-08 华侨大学 Single image super-resolution reconstruction method based on three-channel convolutional neural network
WO2020177108A1 (en) * 2019-03-01 2020-09-10 北京大学深圳研究生院 Video frame interpolation method, apparatus and device
CN111696032A (en) * 2020-04-27 2020-09-22 平安科技（深圳）有限公司 Image processing method and device based on unsupervised learning and computer equipment
CN111696042A (en) * 2020-06-04 2020-09-22 四川轻化工大学 Image super-resolution reconstruction method based on sample learning
WO2020199593A1 (en) * 2019-04-04 2020-10-08 平安科技（深圳）有限公司 Image segmentation model training method and apparatus, image segmentation method and apparatus, and device and medium
CN111754406A (en) * 2020-06-22 2020-10-09 北京大学深圳研究生院 Image resolution processing method, device and equipment and readable storage medium
CN111798377A (en) * 2020-07-08 2020-10-20 广东工业大学 Magnetic resonance image super-resolution reconstruction method based on multi-resolution learning strategy
CN111861886A (en) * 2020-07-15 2020-10-30 南京信息工程大学 Image super-resolution reconstruction method based on multi-scale feedback network
CN111915481A (en) * 2020-06-08 2020-11-10 北京大米未来科技有限公司 Image processing method, image processing apparatus, electronic device, and medium
CN111915484A (en) * 2020-07-06 2020-11-10 天津大学 Reference image guiding super-resolution method based on dense matching and self-adaptive fusion
CN111932461A (en) * 2020-08-11 2020-11-13 西安邮电大学 Convolutional neural network-based self-learning image super-resolution reconstruction method and system
CN111951164A (en) * 2020-08-11 2020-11-17 哈尔滨理工大学 Image super-resolution reconstruction network structure and image reconstruction effect analysis method
CN112070667A (en) * 2020-08-14 2020-12-11 西安理工大学 Multi-scale feature fusion video super-resolution reconstruction method
CN112184552A (en) * 2020-09-23 2021-01-05 国电南瑞科技股份有限公司 Sub-pixel convolution image super-resolution method based on high-frequency feature learning
CN112686807A (en) * 2021-01-12 2021-04-20 中国科学院长春光学精密机械与物理研究所 Image super-resolution reconstruction method and system
CN112750076A (en) * 2020-04-13 2021-05-04 奕目(上海)科技有限公司 Light field multi-view image super-resolution reconstruction method based on deep learning
CN112967184A (en) * 2021-02-04 2021-06-15 西安理工大学 Super-resolution amplification method based on double-scale convolutional neural network
CN113223001A (en) * 2021-05-07 2021-08-06 西安智诊智能科技有限公司 Image segmentation method based on multi-resolution residual error network
CN113808017A (en) * 2021-08-24 2021-12-17 北京理工大学 Structured light imaging method and device based on illumination optimization
CN113837944A (en) * 2021-09-29 2021-12-24 清华大学 DEM super-resolution method and device based on residual error network
CN114022348A (en) * 2021-09-30 2022-02-08 浪潮(北京)电子信息产业有限公司 Super-resolution method, system and related device for image
CN114266697A (en) * 2020-09-16 2022-04-01 广州虎牙科技有限公司 Image processing and model training method and device, electronic equipment and storage medium
WO2022111368A1 (en) * 2020-11-26 2022-06-02 上海健康医学院 Deep-learning-based super-resolution reconstruction method for microscopic image, and medium and electronic device
CN116468812A (en) * 2023-05-16 2023-07-21 山东省计算中心（国家超级计算济南中心） Image compressed sensing reconstruction method and system based on multiple branches and multiple scales
CN117853502A (en) * 2023-12-15 2024-04-09 北京透彻未来科技有限公司 Pathological image segmentation model and method based on multi-scale mixed analysis
CN118195903A (en) * 2024-05-20 2024-06-14 四川新视创伟超高清科技有限公司 Super-resolution image restoration method and restoration system based on deep learning

Citations (5)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
US20170256033A1 (en) * 2016-03-03 2017-09-07 Mitsubishi Electric Research Laboratories, Inc. Image Upsampling using Global and Local Constraints
CN107182216A (en) * 2015-12-30 2017-09-19 中国科学院深圳先进技术研究院 A kind of rapid magnetic resonance imaging method and device based on depth convolutional neural networks
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
CN107507134A (en) * 2017-09-21 2017-12-22 大连理工大学 Super-resolution method based on convolutional neural networks

Patent Citations (5)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN107182216A (en) * 2015-12-30 2017-09-19 中国科学院深圳先进技术研究院 A kind of rapid magnetic resonance imaging method and device based on depth convolutional neural networks
US20170256033A1 (en) * 2016-03-03 2017-09-07 Mitsubishi Electric Research Laboratories, Inc. Image Upsampling using Global and Local Constraints
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
CN107507134A (en) * 2017-09-21 2017-12-22 大连理工大学 Super-resolution method based on convolutional neural networks

Non-Patent Citations (2)

* Cited by examiner, † Cited by third party
Title
孙跃文等: "基于深度学习的辐射图像超分辨率重建方法", 《原子能科学技术》 *
李伟等: "基于卷积神经网络的深度图像超分辨率重建方法", 《电子测量与仪器学报》 *

Cited By (109)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN109493308A (en) * 2018-11-14 2019-03-19 吉林大学 The medical image synthesis and classification method for generating confrontation network are differentiated based on condition more
CN109375186A (en) * 2018-11-22 2019-02-22 中国人民解放军海军航空大学 Radar target identification method based on the multiple dimensioned one-dimensional convolutional neural networks of depth residual error
CN109584162A (en) * 2018-11-30 2019-04-05 江苏网进科技股份有限公司 A method of based on the image super-resolution reconstruct for generating network
CN110232653A (en) * 2018-12-12 2019-09-13 天津大学青岛海洋技术研究院 The quick light-duty intensive residual error network of super-resolution rebuilding
CN109685717A (en) * 2018-12-14 2019-04-26 厦门理工学院 Image super-resolution rebuilding method, device and electronic equipment
CN109903225A (en) * 2019-01-29 2019-06-18 中南大学 A kind of medical image Enhancement Method based on deep learning
CN109903226B (en) * 2019-01-30 2023-08-15 天津城建大学 Image super-resolution reconstruction method based on symmetric residual convolution neural network
CN109920013B (en) * 2019-01-30 2020-11-10 北京交通大学 Image reconstruction method and device based on progressive convolution measurement network
CN109903226A (en) * 2019-01-30 2019-06-18 天津城建大学 Image super-resolution rebuilding method based on symmetrical residual error convolutional neural networks
CN109920013A (en) * 2019-01-30 2019-06-21 北京交通大学 Image reconstructing method and device based on gradual convolution measurement network
CN111521989B (en) * 2019-02-01 2023-12-29 通用汽车环球科技运作有限责任公司 Deep learning for super resolution in radar systems
CN111521989A (en) * 2019-02-01 2020-08-11 通用汽车环球科技运作有限责任公司 Deep learning for super-resolution in radar systems
CN111562305A (en) * 2019-02-14 2020-08-21 爱科维申科技（天津）有限公司 Automobile shock absorber piston defect detection method based on electromagnetic tomography technology
CN111562305B (en) * 2019-02-14 2024-01-26 爱科维申科技（天津）有限公司 Automobile shock absorber piston defect detection method based on electromagnetic tomography technology
CN109903228B (en) * 2019-02-28 2023-03-24 合肥工业大学 Image super-resolution reconstruction method based on convolutional neural network
CN109903228A (en) * 2019-02-28 2019-06-18 合肥工业大学 A kind of image super-resolution rebuilding method based on convolutional neural networks
CN109903219A (en) * 2019-02-28 2019-06-18 深圳市商汤科技有限公司 Image processing method and device, electronic equipment, computer readable storage medium
CN109887018A (en) * 2019-02-28 2019-06-14 中国计量大学 A kind of photon 3D imaging system based on deep learning
US11354541B2 (en) 2019-03-01 2022-06-07 Peking University Shenzhen Graduate School Method, apparatus, and device for video frame interpolation
WO2020177108A1 (en) * 2019-03-01 2020-09-10 北京大学深圳研究生院 Video frame interpolation method, apparatus and device
CN109934771A (en) * 2019-03-08 2019-06-25 北京航空航天大学 Unsupervised Remote sensed image super-resolution reconstruction method based on Recognition with Recurrent Neural Network
CN109934771B (en) * 2019-03-08 2021-03-12 北京航空航天大学 Unsupervised remote sensing image super-resolution reconstruction method based on recurrent neural network
WO2020199593A1 (en) * 2019-04-04 2020-10-08 平安科技（深圳）有限公司 Image segmentation model training method and apparatus, image segmentation method and apparatus, and device and medium
CN110111256A (en) * 2019-04-28 2019-08-09 西安电子科技大学 Image Super-resolution Reconstruction method based on residual error distillation network
CN110111256B (en) * 2019-04-28 2023-03-31 西安电子科技大学 Image super-resolution reconstruction method based on residual distillation network
CN110211038A (en) * 2019-04-29 2019-09-06 南京航空航天大学 Super resolution ratio reconstruction method based on dirac residual error deep neural network
CN110264410A (en) * 2019-05-07 2019-09-20 西安理工大学 A kind of image super-resolution rebuilding method based on minutia
CN110111257B (en) * 2019-05-08 2023-01-03 哈尔滨工程大学 Super-resolution image reconstruction method based on characteristic channel adaptive weighting
CN110111257A (en) * 2019-05-08 2019-08-09 哈尔滨工程大学 A kind of super resolution image reconstruction method based on the weighting of feature channel adaptive
CN110288524A (en) * 2019-05-09 2019-09-27 广东启迪图卫科技股份有限公司 Deep learning super-resolution method based on enhanced up-sampling and discrimination syncretizing mechanism
CN110044262A (en) * 2019-05-09 2019-07-23 哈尔滨理工大学 Contactless precision measuring instrument and measurement method based on image super-resolution rebuilding
CN110175953B (en) * 2019-05-24 2023-04-18 鹏城实验室 Image super-resolution method and system
CN110175953A (en) * 2019-05-24 2019-08-27 鹏城实验室 A kind of image super-resolution method and system
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
CN110310227A (en) * 2019-06-27 2019-10-08 电子科技大学 A kind of image super-resolution rebuilding method decomposed based on high and low frequency information
CN110310227B (en) * 2019-06-27 2020-09-08 电子科技大学 Image super-resolution reconstruction method based on high-low frequency information decomposition
CN110415201B (en) * 2019-07-30 2022-03-22 北京理工大学 Single-exposure super-resolution imaging method and device based on structured light and deep learning
CN110415201A (en) * 2019-07-30 2019-11-05 北京理工大学 Single exposure super-resolution imaging method and device based on structure light and deep learning
CN110473247A (en) * 2019-07-30 2019-11-19 中国科学院空间应用工程与技术中心 Solid matching method, device and storage medium
CN110580681A (en) * 2019-09-12 2019-12-17 杭州海睿博研科技有限公司 high-resolution cardiac motion pattern analysis device and method
CN110929811B (en) * 2019-10-21 2023-07-28 北京理工大学 Deep learning method for acquiring full waveform laser radar high-resolution data
CN110929811A (en) * 2019-10-21 2020-03-27 北京理工大学 Deep learning method for acquiring high-resolution data of full-waveform laser radar
CN110930306A (en) * 2019-10-28 2020-03-27 杭州电子科技大学 Depth map super-resolution reconstruction network construction method based on non-local perception
CN110930306B (en) * 2019-10-28 2023-09-12 杭州电子科技大学 Depth map super-resolution reconstruction network construction method based on non-local perception
CN111223046B (en) * 2019-11-20 2023-04-25 中国科学院遥感与数字地球研究所 Image super-resolution reconstruction method and device
CN111223046A (en) * 2019-11-20 2020-06-02 中国科学院遥感与数字地球研究所 Image super-resolution reconstruction method and device
CN111062867A (en) * 2019-11-21 2020-04-24 浙江大华技术股份有限公司 Video super-resolution reconstruction method
CN111080516B (en) * 2019-11-26 2023-04-28 广东石油化工学院 Super-resolution image reconstruction method based on self-sample enhancement
CN111080516A (en) * 2019-11-26 2020-04-28 广东石油化工学院 Super-resolution image reconstruction method based on self-sampling enhancement
CN111027487A (en) * 2019-12-11 2020-04-17 山东大学 Behavior recognition system, method, medium, and apparatus based on multi-convolution kernel residual network
CN111027487B (en) * 2019-12-11 2023-04-28 山东大学 Behavior recognition system, method, medium and equipment based on multi-convolution kernel residual error network
CN111160413B (en) * 2019-12-12 2023-11-17 天津大学 Thyroid nodule classification method based on multi-scale feature fusion
CN111160413A (en) * 2019-12-12 2020-05-15 天津大学 Thyroid nodule classification method based on multi-scale feature fusion
CN111062434A (en) * 2019-12-13 2020-04-24 国网重庆市电力公司永川供电分公司 Multi-scale fusion detection method for unmanned aerial vehicle inspection
CN111239731A (en) * 2020-01-06 2020-06-05 南京航空航天大学 Synthetic aperture radar rapid imaging method and device based on neural network
CN111260558A (en) * 2020-01-22 2020-06-09 武汉大学 Image super-resolution network model with variable magnification
CN111260558B (en) * 2020-01-22 2021-10-22 武汉大学 Image super-resolution network model with variable magnification
CN111402128A (en) * 2020-02-21 2020-07-10 华南理工大学 Image super-resolution reconstruction method based on multi-scale pyramid network
CN111369466A (en) * 2020-03-05 2020-07-03 福建帝视信息科技有限公司 Image distortion correction enhancement method of convolutional neural network based on deformable convolution
CN111369466B (en) * 2020-03-05 2023-06-16 福建帝视信息科技有限公司 Image distortion correction enhancement method of convolutional neural network based on deformable convolution
CN111402131A (en) * 2020-03-10 2020-07-10 北京师范大学 Method for acquiring super-resolution land cover classification map based on deep learning
CN111402131B (en) * 2020-03-10 2022-04-01 北京师范大学 Method for acquiring super-resolution land cover classification map based on deep learning
CN111415299B (en) * 2020-03-26 2023-03-28 浙江科技学院 High-resolution image style migration method
CN111415299A (en) * 2020-03-26 2020-07-14 浙江科技学院 High-resolution image style migration method
CN111476714A (en) * 2020-03-30 2020-07-31 清华大学 Cross-scale image splicing method and device based on PSV neural network
CN111476714B (en) * 2020-03-30 2022-10-28 清华大学 Cross-scale image splicing method and device based on PSV neural network
CN111353940A (en) * 2020-03-31 2020-06-30 成都信息工程大学 Image super-resolution reconstruction method based on deep learning iterative up-down sampling
CN112750076B (en) * 2020-04-13 2022-11-15 奕目(上海)科技有限公司 Light field multi-view image super-resolution reconstruction method based on deep learning
CN112750076A (en) * 2020-04-13 2021-05-04 奕目(上海)科技有限公司 Light field multi-view image super-resolution reconstruction method based on deep learning
CN111696032A (en) * 2020-04-27 2020-09-22 平安科技（深圳）有限公司 Image processing method and device based on unsupervised learning and computer equipment
CN111476719B (en) * 2020-05-06 2023-05-12 Oppo广东移动通信有限公司 Image processing method, device, computer equipment and storage medium
CN111476719A (en) * 2020-05-06 2020-07-31 Oppo广东移动通信有限公司 Image processing method, image processing device, computer equipment and storage medium
CN111598778A (en) * 2020-05-13 2020-08-28 云南电网有限责任公司电力科学研究院 Insulator image super-resolution reconstruction method
CN111598778B (en) * 2020-05-13 2023-11-21 云南电网有限责任公司电力科学研究院 Super-resolution reconstruction method for insulator image
CN111598808A (en) * 2020-05-18 2020-08-28 腾讯科技（深圳）有限公司 Image processing method, device and equipment and training method thereof
CN111696042B (en) * 2020-06-04 2023-06-27 四川轻化工大学 Image super-resolution reconstruction method based on sample learning
CN111696042A (en) * 2020-06-04 2020-09-22 四川轻化工大学 Image super-resolution reconstruction method based on sample learning
CN111915481B (en) * 2020-06-08 2024-03-29 北京大米未来科技有限公司 Image processing method, device, electronic equipment and medium
CN111915481A (en) * 2020-06-08 2020-11-10 北京大米未来科技有限公司 Image processing method, image processing apparatus, electronic device, and medium
CN111640067A (en) * 2020-06-10 2020-09-08 华侨大学 Single image super-resolution reconstruction method based on three-channel convolutional neural network
CN111640067B (en) * 2020-06-10 2022-08-09 华侨大学 Single image super-resolution reconstruction method based on three-channel convolutional neural network
CN111754406B (en) * 2020-06-22 2024-02-23 北京大学深圳研究生院 Image resolution processing method, device, equipment and readable storage medium
CN111754406A (en) * 2020-06-22 2020-10-09 北京大学深圳研究生院 Image resolution processing method, device and equipment and readable storage medium
CN111915484A (en) * 2020-07-06 2020-11-10 天津大学 Reference image guiding super-resolution method based on dense matching and self-adaptive fusion
CN111915484B (en) * 2020-07-06 2023-07-07 天津大学 Reference image guiding super-resolution method based on dense matching and self-adaptive fusion
CN111798377A (en) * 2020-07-08 2020-10-20 广东工业大学 Magnetic resonance image super-resolution reconstruction method based on multi-resolution learning strategy
CN111798377B (en) * 2020-07-08 2023-07-28 广东工业大学 Magnetic resonance image super-resolution reconstruction method based on multi-resolution learning strategy
CN111861886A (en) * 2020-07-15 2020-10-30 南京信息工程大学 Image super-resolution reconstruction method based on multi-scale feedback network
CN111861886B (en) * 2020-07-15 2023-08-08 南京信息工程大学 Image super-resolution reconstruction method based on multi-scale feedback network
CN111932461A (en) * 2020-08-11 2020-11-13 西安邮电大学 Convolutional neural network-based self-learning image super-resolution reconstruction method and system
CN111951164A (en) * 2020-08-11 2020-11-17 哈尔滨理工大学 Image super-resolution reconstruction network structure and image reconstruction effect analysis method
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
CN112070667A (en) * 2020-08-14 2020-12-11 西安理工大学 Multi-scale feature fusion video super-resolution reconstruction method
CN114266697A (en) * 2020-09-16 2022-04-01 广州虎牙科技有限公司 Image processing and model training method and device, electronic equipment and storage medium
CN112184552B (en) * 2020-09-23 2022-11-11 国电南瑞科技股份有限公司 Sub-pixel convolution image super-resolution method based on high-frequency feature learning
CN112184552A (en) * 2020-09-23 2021-01-05 国电南瑞科技股份有限公司 Sub-pixel convolution image super-resolution method based on high-frequency feature learning
WO2022111368A1 (en) * 2020-11-26 2022-06-02 上海健康医学院 Deep-learning-based super-resolution reconstruction method for microscopic image, and medium and electronic device
CN112686807A (en) * 2021-01-12 2021-04-20 中国科学院长春光学精密机械与物理研究所 Image super-resolution reconstruction method and system
CN112967184A (en) * 2021-02-04 2021-06-15 西安理工大学 Super-resolution amplification method based on double-scale convolutional neural network
CN112967184B (en) * 2021-02-04 2022-12-13 西安理工大学 Super-resolution amplification method based on double-scale convolutional neural network
CN113223001A (en) * 2021-05-07 2021-08-06 西安智诊智能科技有限公司 Image segmentation method based on multi-resolution residual error network
CN113808017A (en) * 2021-08-24 2021-12-17 北京理工大学 Structured light imaging method and device based on illumination optimization
CN113837944A (en) * 2021-09-29 2021-12-24 清华大学 DEM super-resolution method and device based on residual error network
CN114022348A (en) * 2021-09-30 2022-02-08 浪潮(北京)电子信息产业有限公司 Super-resolution method, system and related device for image
CN116468812A (en) * 2023-05-16 2023-07-21 山东省计算中心（国家超级计算济南中心） Image compressed sensing reconstruction method and system based on multiple branches and multiple scales
CN117853502A (en) * 2023-12-15 2024-04-09 北京透彻未来科技有限公司 Pathological image segmentation model and method based on multi-scale mixed analysis
CN117853502B (en) * 2023-12-15 2024-06-14 北京透彻未来科技有限公司 Pathological image segmentation model and method based on multi-scale mixed analysis
CN118195903A (en) * 2024-05-20 2024-06-14 四川新视创伟超高清科技有限公司 Super-resolution image restoration method and restoration system based on deep learning
CN118195903B (en) * 2024-05-20 2024-07-30 四川新视创伟超高清科技有限公司 Super-resolution image restoration method and restoration system based on deep learning

Similar Documents

Publication Publication Date Title
CN108734660A (en) A kind of image super-resolution rebuilding method and device based on deep learning
WO2021056969A1 (en) Super-resolution image reconstruction method and device
CN111861961B (en) Single image super-resolution multi-scale residual error fusion model and restoration method thereof
CN108765296B (en) Image super-resolution reconstruction method based on recursive residual attention network
US10019642B1 (en) Image upsampling system, training method thereof and image upsampling method
CN108734659A (en) A kind of sub-pix convolved image super resolution ratio reconstruction method based on multiple dimensioned label
CN108765291A (en) Super resolution ratio reconstruction method based on dense neural network and two-parameter loss function
CN109146788A (en) Super-resolution image reconstruction method and device based on deep learning
CN107240066A (en) Image super-resolution rebuilding algorithm based on shallow-layer and deep layer convolutional neural networks
CN111784582B (en) DEC-SE-based low-illumination image super-resolution reconstruction method
CN109886871A (en) The image super-resolution method merged based on channel attention mechanism and multilayer feature
CN107369189A (en) The medical image super resolution ratio reconstruction method of feature based loss
CN108475415A (en) Method and system for image procossing
CN107358575A (en) A kind of single image super resolution ratio reconstruction method based on depth residual error network
CN102722865A (en) Super-resolution sparse representation method
Chen et al. Single image super-resolution using deep CNN with dense skip connections and inception-resnet
Li et al. Underwater image high definition display using the multilayer perceptron and color feature-based SRCNN
CN112801904A (en) Hybrid degraded image enhancement method based on convolutional neural network
RU2697928C1 (en) Superresolution of an image imitating high detail based on an optical system, performed on a mobile device having limited resources, and a mobile device which implements
CN109949224A (en) A kind of method and device of the connection grade super-resolution rebuilding based on deep learning
CN114841856A (en) Image super-pixel reconstruction method of dense connection network based on depth residual channel space attention
CN108765290A (en) A kind of super resolution ratio reconstruction method based on improved dense convolutional neural networks
CN110246085A (en) A kind of single-image super-resolution method
CN110533591A (en) Super resolution image reconstruction method based on codec structure
Guan et al. Srdgan: learning the noise prior for super resolution with dual generative adversarial networks

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
WD01 Invention patent application deemed withdrawn after publication
WD01 Invention patent application deemed withdrawn after publication

Application publication date: 20181102