CN107369189A - The medical image super resolution ratio reconstruction method of feature based loss - Google Patents
The medical image super resolution ratio reconstruction method of feature based loss Download PDFInfo
- Publication number
- CN107369189A CN107369189A CN201710600369.XA CN201710600369A CN107369189A CN 107369189 A CN107369189 A CN 107369189A CN 201710600369 A CN201710600369 A CN 201710600369A CN 107369189 A CN107369189 A CN 107369189A
- Authority
- CN
- China
- Prior art keywords
- network
- medical image
- image
- reconstruction method
- feature based
- 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 52
- 210000002569 neuron Anatomy 0.000 claims abstract description 17
- 238000013528 artificial neural network Methods 0.000 claims abstract description 14
- 238000005070 sampling Methods 0.000 claims description 24
- 238000012549 training Methods 0.000 claims description 21
- 208000037170 Delayed Emergence from Anesthesia Diseases 0.000 claims description 14
- 230000008569 process Effects 0.000 claims description 9
- 238000004364 calculation method Methods 0.000 claims description 7
- 239000011248 coating agent Substances 0.000 claims description 5
- 238000000576 coating method Methods 0.000 claims description 5
- 230000000694 effects Effects 0.000 claims description 5
- 230000003321 amplification Effects 0.000 claims description 4
- 238000003199 nucleic acid amplification method Methods 0.000 claims description 4
- 235000013399 edible fruits Nutrition 0.000 claims 1
- 230000006870 function Effects 0.000 description 12
- 238000013527 convolutional neural network Methods 0.000 description 9
- 210000004556 brain Anatomy 0.000 description 5
- 230000014509 gene expression Effects 0.000 description 4
- 230000001537 neural effect Effects 0.000 description 3
- 238000005457 optimization Methods 0.000 description 3
- 238000012545 processing Methods 0.000 description 3
- 238000012360 testing method Methods 0.000 description 3
- 230000017105 transposition Effects 0.000 description 3
- 241000208340 Araliaceae Species 0.000 description 1
- 235000005035 Panax pseudoginseng ssp. pseudoginseng Nutrition 0.000 description 1
- 235000003140 Panax quinquefolius Nutrition 0.000 description 1
- 238000012098 association analyses Methods 0.000 description 1
- 230000009286 beneficial effect Effects 0.000 description 1
- 230000000747 cardiac effect Effects 0.000 description 1
- 238000006243 chemical reaction Methods 0.000 description 1
- 238000000205 computational method Methods 0.000 description 1
- 238000002591 computed tomography Methods 0.000 description 1
- 238000007796 conventional method Methods 0.000 description 1
- 230000000593 degrading effect Effects 0.000 description 1
- 238000010586 diagram Methods 0.000 description 1
- 238000005516 engineering process Methods 0.000 description 1
- 238000002474 experimental method Methods 0.000 description 1
- 238000000605 extraction Methods 0.000 description 1
- 235000008434 ginseng Nutrition 0.000 description 1
- 238000003384 imaging method Methods 0.000 description 1
- 238000012886 linear function Methods 0.000 description 1
- 238000010801 machine learning Methods 0.000 description 1
- 238000013507 mapping Methods 0.000 description 1
- 238000012986 modification Methods 0.000 description 1
- 230000004048 modification Effects 0.000 description 1
- 210000004218 nerve net Anatomy 0.000 description 1
- 230000007935 neutral effect Effects 0.000 description 1
- 238000010606 normalization Methods 0.000 description 1
- 230000008447 perception Effects 0.000 description 1
- 230000000644 propagated effect Effects 0.000 description 1
- 238000006467 substitution reaction Methods 0.000 description 1
- 238000012876 topography Methods 0.000 description 1
- 230000007704 transition Effects 0.000 description 1
- 230000000007 visual effect Effects 0.000 description 1
- 230000016776 visual perception Effects 0.000 description 1
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T11/00—2D [Two Dimensional] image generation
- G06T11/003—Reconstruction from projections, e.g. tomography
- G06T11/008—Specific post-processing after tomographic reconstruction, e.g. voxelisation, metal artifact correction
-
- 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
- G06T3/4046—Scaling of whole images or parts thereof, e.g. expanding or contracting using neural networks
-
- 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
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T2207/00—Indexing scheme for image analysis or image enhancement
- G06T2207/10—Image acquisition modality
- G06T2207/10072—Tomographic images
- G06T2207/10081—Computed x-ray tomography [CT]
-
- 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
- G06T2207/10—Image acquisition modality
- G06T2207/10072—Tomographic images
- G06T2207/10088—Magnetic resonance imaging [MRI]
-
- 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
- G06T2207/30—Subject of image; Context of image processing
- G06T2207/30004—Biomedical image processing
- G06T2207/30016—Brain
Landscapes
- Engineering & Computer Science (AREA)
- Physics & Mathematics (AREA)
- General Physics & Mathematics (AREA)
- Theoretical Computer Science (AREA)
- Artificial Intelligence (AREA)
- Evolutionary Computation (AREA)
- Image Analysis (AREA)
- Image Processing (AREA)
Abstract
The invention discloses the medical image super resolution ratio reconstruction method of feature based loss, including image switching network fw, described image switching network fwIt is the full Connection Neural Network of feedforward, the full Connection Neural Network of the feedforward is by each neuron in network, it is divided into different groups by the priority of receive information, each group is regarded as an Internet, neuron in each layer receives the number output of preceding layer neuron as the input of oneself, the output of oneself is input to next layer again, the information in whole network is to propagate in one direction;Image switching network fwThe low resolution medical image for H/4 × W/4 sizes that feedforward neural network is sent is received, the low resolution medical image of H/4 × W/4 sizes is converted into the high-definition picture of H × W sizes.
Description
Technical field
The present invention relates to a kind of super-resolution rebuilding, and in particular to the medical image super-resolution rebuilding of feature based loss
Method.
Background technology
In recent years, super-resolution rebuilding mainly has two class methods:One kind is the method based on reconstruction;Another kind of is based on
The method of habit.Method based on reconstruction, it is by being modeled to the acquisition process of low-resolution image, utilizing regularization method
The prior-constrained of high-definition picture is constructed, estimates that high-definition picture, reconstruction are lost during degrading by low-resolution image
High-frequency signal, the problem of most problem is converted into cost function optimization under constraints at last;Another kind of is based on study
Super-resolution rebuilding, the basic thought of this method is by learning to obtain reflecting between high-definition picture and ground resolution image
Relation is penetrated, in this kind of method, the process of study is crucial, is learnt using rarefaction representation scheduling theory, and study terminates can
Low resolution is instructed to be rebuild with the priori independent of people.
Lifted along with the raising of the level of informatization and graphics processor computing capability so that the acquisition of view data and place
Reason also becomes easy, and the super resolution ratio reconstruction method based on study receives extensive concern.Yang etc. is to low resolution and high score
The image library that resolution image block is formed carries out rarefaction representation, and finds low resolution and high-resolution by the method for joint training
Corresponding excessively complete dictionary, the contact between foundation between image block;Rueda, Wang et al. are using based on rarefaction representation
Method produces high-resolution brain MR image from low resolution brain MR image, and Huang Haofeng, white Fu also uses same method
Rebuild for different types of medical image;Done et al. establishes the convolutional Neural net of an only hidden layer
Network SRCNN, network is considered as one and mapped end to end, one end is that the low-resolution image other end is high-definition picture, is obtained
Obtained preferable natural image super-resolution rebuilding effect;Bahrami et al. devises one five layers of Three dimensional convolution nerve net
Network, class 7T brain images are reconstructed from 3T brain images;Oktay et al. is based on residual error network from two-dimensional cardiac MR image sequence weights
Build out high-resolution 3-D view;Other medical image super-resolution methods, as Burgos goes out a kind of topography's similitude
Method go out CT images from MR image reconstructions;Bahrami proposes a kind of method for combining typical association analysis to be schemed using 3TMR
Method as reconstructing 7TMR images.
Super-resolution rebuilding problem is that solve the problems, such as to rebuild high-definition picture by low-resolution image, wherein low resolution
The acquisition methods of rate image are usually to carry out down-sampling to high-definition picture, by the high-definition picture of down-sampling as weight
The data source built, original high-resolution image is as the target rebuild.Sample rate is more than the sampling for obtaining the data signal originally
Rate is referred to as up-sampling, and the main purpose of up-sampling is enlarged drawing, so as to be shown on the display device of higher resolution.
SRCNN is also required to image using the up-sampling of image as the image procossing before training convolutional neural networks in test phase
Carry out same up-sampling processing.The low-resolution image for so in advance obtaining down-sampling carries out up-sampling and is used further to convolutional Neural
The training of network is cumbersome, and the process of up-sampling is to be dissolved into convolutional neural networks, so as to independent of fixed upper
Sample interpolation function, the process of up-sampling is also served as into the part that network can learn so that the network trained has more
Universal applicability.
The up-sampling of image is realized in convolutional neural networks, conventional method is exactly to set transposition convolutional layer or sub-pixel
Convolutional layer.Transposition convolution is also named deconvolution, but in using transposition convolution process, it may occur that convolution is uneven to cause image some
The problem of color of position is more deeper than other positions, there is the artifact similar to checker-wise, such artifact is in convolution kernel
Size is more obvious when can not be divided exactly by step-length;Sub-pixel convolution convolution nuclear energy is divided exactly by step-length, but although this method has
Help, but still easily produce the artifact similar to checker-wise.
The content of the invention
When the technical problems to be solved by the invention are traditional super-resolution rebuildings, it can produce similar to checker-wise
Artifact solves traditional Super-resolution reconstruction, and it is an object of the present invention to provide the medical image super resolution ratio reconstruction method of feature based loss
When building, the problem of artifact similar to checker-wise can be produced.
The present invention is achieved through the following technical solutions:
The medical image super resolution ratio reconstruction method of feature based loss, including
The medical image super resolution ratio reconstruction method of feature based loss, it is characterised in that:Including image switching network fw,
Described image switching network fwIt is the full Connection Neural Network of feedforward, the feedforward neural network presses each neuron in network
The priority of receive information is divided into different groups, and each group is regarded as an Internet, and the neuron in each layer receives preceding layer
The number output of neuron is input to next layer as the input of oneself, then by the output of oneself, and the information in whole network is court
Propagate in one direction;Image switching network fwThe low resolution medical image of H/4 × W/4 sizes is handled, by H/4 × W/4 sizes
Low resolution medical image be converted into the high-definition pictures of H × W sizes, the H and W are natural number.Feedforward neural network
Each neuron in network is divided into different groups by the priority of receive information, each group can be regarded as an Internet, often
Neuron in one layer receives the number output of preceding layer neuron as the input of oneself, then the output of oneself is input to next
Layer.Information in whole network is propagated in one direction.Feedforward network is considered as one and passes through simple non-linear functions
Multiple combination, realize the input space to output space complex mappings.Image switching network is a full connection feed forward neural
Network, full connection can make network use the image of any size in test phase, and the input for not requiring network is fixed size.
The function of realization is that the low resolution medical image of size is converted into the high-definition picture of size.
Described image switching network fwIncluding two amplification convolutional layers, by amplifying convolutional layer twice, image is just enlarged into
Input 4 times of size.Convolutional layer uses the size for first adjusting network middle level, such as:Inserted using closest interpolation method or bilinearity
Value method, carry out convolution operation again afterwards.
In network fwMiddle setting residual block, the input using the output of preceding layer as residual block, input are passed through convolutional layer, swashed
Result after layer living is along with output of the input as residual block, and in the structure of network, network exceedes certain depth, it may appear that
Gradient disperse, even if using batch normalization operation, also result in very high training error, the method for solving this problem be
The residual block of addition input and the quick connection of the output of active coating in network.
The residual error number of blocks is 5.Residual block use standard feedforward convolutional neural networks, and add once skip it is several layers of
Connection, skip generation residual block every time, skip part result of calculation be added to input in be used as output result, include five
Such residual block, further, the preferred scheme as the present invention.
Also include up-sampling layer.Set up-sampling layer position when, according to typically first amplify after training network thinking,
Up-sampling layer is placed on to the leading portion of network.
The up-sampling layer is arranged on image switching network fwBack segment.Up-sampling layer is placed on to the leading portion of network.Meeting
The amount of calculation of Internet below is caused to increase, so reasonable manner is that up-sampling layer is placed on to the back segment of network so that on
Simply small-sized image is handled before sample level, layer is to the last up-sampled and is amplified again, reduce amount of calculation.
Also include loss network φ, described image switching network fwHigh-definition picture is sent to loss network φ, loss
Network φ inputs include original image and resolution chart, are damaged by the good image classification network calculations feature of training in advance.
Described image sorter network is VGG16 networks, performs 3 × 3 or 2 × 2 convolution, in image classification task, one
After individual network training is good, the different layers in network differ to the level of abstraction of image different characteristic.In order to increase image
The output of switching network and the similarity degree of true picture spatially, introduce the good volume for image classification of a training in advance
Product neutral net VGG16, it has highly uniform framework, from start to finish only performs 3 × 3 and 2 × 2 convolution.In order to more
Network internal graphical representation process is visually known, using the expression information of image in itself, using random noise as initial solution, instead
Neural network characteristics are involved in expression, the different layers of network are visualized.
The weight lost in network φ is fixed value in the training process.Image switching network fwIn weight in training
It is activity value.Weight represents different layer parameters in network, and these parameters are being randomly generated at the beginning, then by training,
Constantly amendment so that these parameters can complete task of instructing low-resolution image to reconstruct full resolution pricture.
The present invention compared with prior art, has the following advantages and advantages:
1st, the medical image super resolution ratio reconstruction method of feature based loss of the present invention, the full connection convolutional Neural of feedforward is used
Network carries out 4 times of magnetic resonance brain medical image super-resolution rebuildings, efficiently convenient;
2nd, the medical image super resolution ratio reconstruction method of feature based loss of the present invention, can be applied to other medical images,
Such as CT scan CT;The super-resolution rebuilding of human body different parts imaging is can be used for, applicable surface is extensive;
3rd, the medical image super resolution ratio reconstruction method of feature based loss of the present invention,.
Brief description of the drawings
Accompanying drawing described herein is used for providing further understanding the embodiment of the present invention, forms one of the application
Point, do not form the restriction to the embodiment of the present invention.In the accompanying drawings:
Fig. 1 is the artifact of checker-wise pattern of the present invention;
Fig. 2 is schematic structural view of the invention;
Fig. 3 is feature of present invention costing bio disturbance flow chart;
Fig. 4 is residual error block structural diagram of the present invention.
Embodiment
For the object, technical solutions and advantages of the present invention are more clearly understood, with reference to embodiment and accompanying drawing, to this
Invention is described in further detail, and exemplary embodiment of the invention and its explanation are only used for explaining the present invention, do not make
For limitation of the invention.
As Figure 1-4:
Embodiment 1
The medical image super resolution ratio reconstruction method of feature based loss, including image switching network fw, described image turn
Switching network fwIt is the full Connection Neural Network of feedforward, the feedforward neural network is by each neuron in network, by receive information
Successively it is divided into different groups, each group is regarded as an Internet, and the neuron in each layer receives the number of preceding layer neuron
Output is input to next layer as the input of oneself, then by the output of oneself, and the information in whole network is to pass in one direction
Broadcast;Image switching network fwThe low resolution medical image of H/4 × W/4 sizes is handled, by the low resolution of H/4 × W/4 sizes
Medical image is converted into the high-definition picture of H × W sizes, and the H and W are natural number.Described image switching network fwIncluding
Two amplification convolutional layers, by amplifying convolutional layer twice, image is just enlarged into 4 times of input size.In network fwMiddle setting is residual
Poor block, the input using the output of preceding layer as residual block, inputs and passes through convolutional layer, the result after active coating is made along with input
For the output of residual block.The residual error number of blocks is 5.Also include up-sampling layer.The up-sampling layer is arranged on image transition net
Network fwBack segment.Solve the problems, such as that convolutional neural networks processing image easily produces chessboard artifact.
Embodiment 2
The medical image super resolution ratio reconstruction method of feature based loss, including image switching network fw, described image turn
Switching network fwIt is the full Connection Neural Network of feedforward, the feedforward neural network is by each neuron in network, by receive information
Successively it is divided into different groups, each group is regarded as an Internet, and the neuron in each layer receives the number of preceding layer neuron
Output is input to next layer as the input of oneself, then by the output of oneself, and the information in whole network is to pass in one direction
Broadcast;Image switching network fwThe low resolution medical image of H/4 × W/4 sizes is handled, by the low resolution of H/4 × W/4 sizes
Medical image is converted into the high-definition picture of H × W sizes, and the H and W are natural number.Described image switching network fwIncluding
Two amplification convolutional layers, by amplifying convolutional layer twice, image is just enlarged into 4 times of input size.In network fwMiddle setting is residual
Poor block, the input using the output of preceding layer as residual block, inputs and passes through convolutional layer, the result after active coating is along with input
Output as residual block.The residual error number of blocks is 5.Also include up-sampling layer.The up-sampling layer is arranged on image conversion
Network fwBack segment.Also include loss network φ, described image switching network fwSend high-definition picture to lose network φ,
Loss network φ inputs include original image and high resolution graphics, pass through the good image classification network calculations feature of training in advance
Damage.Described image sorter network is VGG16 networks, performs 3 × 3 or 2 × 2 convolution.Weight in loss network φ is being trained
During be fixed value, image switching network fwIn weight be activity value in training.
By analyzing VGG16 network structures, when image passes through VGG16 lower levels, operation saves image main contents letter
Breath, and image absolute position feature is weakened, using this characteristic, by the output of super-resolution rebuilding convolutional neural networks
Input in VGG16 networks, use again with target image yWith MSEs of the y in VGG16 lower levels as loss function, phase
Hope them have similar character representation in VGG16, increase output and the similitude of target image spatially of network, do not make list
One compares pixel-by-pixel, and such loss function is defined as characteristic loss, computational methods such as formula (1):
In formula, the good VGG16 networks of φ expressions training in advance, l ∈ { Relu1_1, Relu1_2 ..., Relu5_3 }, this
Test l=Relu2_2,, Hl, Wl, ClIt is illustrated respectively in a certain active coating output in VGG16
Length and width and port number.
In having the machine learning task of supervision, the image of high quality, target letter can be generated by optimization object function
Number is for estimating the difference degree between the predicted value of model and actual value, and it is a nonnegative number function, optimization aim letter
Number is exactly to find parameter w to lose J minimums.In the application of super-resolution rebuilding, the loss function that is defined using formula (3)
Training convolutional neural networks export by networkVery close target image y, but be not to allow them to accomplish completely to match,
Simply visually it can less be distinguished with y so that reconstruct the visual signature that the image come more conforms to human eye.In the training of network
During, φ weight is all fixed in network, Super-resolution reconstruction establishing network fwIn weight w be not in the training process
Disconnected study amendment.
In order to prevent over-fitting, total variation regularization terms are added by loss function, constrain the ginseng to be optimized
W is counted, thus the object function of super-resolution rebuilding can be expressed as formula (2):
λ=10 in formula-3Regularization factors are represented, because image is discrete distribution, are substituted using finite-difference approximation, it is right
In image x ∈ RH×W, limited R difference is defined as formula (3):
In formula, xi,jThe pixel value of (i, j) is put in expression in image x.
Devising the full connection convolutional neural networks of a feedforward realizes single width medical image super-resolution rebuilding, from experiment
As a result it can be seen that solving the problems, such as that convolutional neural networks processing image easily produces chessboard artifact;In order to obtain preferably
Image perception effect, without using the single mean square error loss function based on pixel, but by the good image of training in advance
Sorter network VGG16, from its lower level extraction feature calculation mean square error as loss function, from experimental result it can be seen that.
Most of image is resumed, and some are smoothed similar to the details of noise, is more conformed to the visual perception of human eye, is visually tested
The validity of method is demonstrate,proved
Above-described embodiment, the purpose of the present invention, technical scheme and beneficial effect are carried out further
Describe in detail, should be understood that the embodiment that the foregoing is only the present invention, be not intended to limit the present invention
Protection domain, within the spirit and principles of the invention, any modification, equivalent substitution and improvements done etc., all should include
Within protection scope of the present invention.
Claims (10)
1. the medical image super resolution ratio reconstruction method of feature based loss, it is characterised in that:Including image switching network fw, institute
State image switching network fwIt is the full Connection Neural Network of feedforward, the feedforward neural network is by each neuron in network, by connecing
The priority of breath of collecting mail is divided into different groups, and each group is regarded as an Internet, and the neuron in each layer receives preceding layer god
Number output through member is input to next layer as the input of oneself, then by the output of oneself, and the information in whole network is towards one
Propagate in individual direction;
Image switching network fwThe low resolution medical image of H/4 × W/4 sizes is handled, by the low resolution of H/4 × W/4 sizes
Medical image is converted into the high-definition picture of H × W sizes, and the H and W are natural number.
2. the medical image super resolution ratio reconstruction method of feature based loss according to claim 1, it is characterised in that:Institute
State image switching network fwIncluding two amplification convolutional layers, by amplifying convolutional layer twice, image is just enlarged into the 4 of input size
Times.
3. the medical image super resolution ratio reconstruction method of feature based loss according to claim 1, it is characterised in that:
Network fwMiddle setting residual block, the input using the output of preceding layer as residual block, inputs and passes through convolutional layer, the knot after active coating
Fruit is along with output of the input as residual block.
4. the medical image super resolution ratio reconstruction method of feature based loss according to claim 3, it is characterised in that:Institute
Residual error number of blocks is stated as 5.
5. the medical image super resolution ratio reconstruction method of feature based loss according to claim 1, it is characterised in that:Also
Including up-sampling layer.
6. the medical image super resolution ratio reconstruction method of feature based loss according to claim 5, it is characterised in that:Institute
State up-sampling layer and be arranged on image switching network fwBack segment.
7. the medical image super resolution ratio reconstruction method of feature based loss according to claim 1, it is characterised in that:Also
Including losing network φ, described image switching network fwHigh-definition picture is sent to loss network φ, the φ inputs of loss network
End includes original image and high resolution graphics, is damaged by the good image classification network calculations feature of training in advance.
8. the medical image super resolution ratio reconstruction method of feature based loss according to claim 7, it is characterised in that:Institute
It is VGG16 networks to state image classification network, and network performs 3 × 3 or 2 × 2 convolution.
9. the medical image super resolution ratio reconstruction method of feature based loss according to claim 7, it is characterised in that:Damage
The weight lost in network φ is fixed value in the training process.
10. the medical image super resolution ratio reconstruction method of feature based loss according to claim 7, it is characterised in that:
Image switching network fwIn weight be activity value in training.
Priority Applications (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201710600369.XA CN107369189A (en) | 2017-07-21 | 2017-07-21 | The medical image super resolution ratio reconstruction method of feature based loss |
Applications Claiming Priority (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201710600369.XA CN107369189A (en) | 2017-07-21 | 2017-07-21 | The medical image super resolution ratio reconstruction method of feature based loss |
Publications (1)
Publication Number | Publication Date |
---|---|
CN107369189A true CN107369189A (en) | 2017-11-21 |
Family
ID=60308079
Family Applications (1)
Application Number | Title | Priority Date | Filing Date |
---|---|---|---|
CN201710600369.XA Pending CN107369189A (en) | 2017-07-21 | 2017-07-21 | The medical image super resolution ratio reconstruction method of feature based loss |
Country Status (1)
Country | Link |
---|---|
CN (1) | CN107369189A (en) |
Cited By (28)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN107948529A (en) * | 2017-12-28 | 2018-04-20 | 北京麒麟合盛网络技术有限公司 | Image processing method and device |
CN108090871A (en) * | 2017-12-15 | 2018-05-29 | 厦门大学 | A kind of more contrast MR image reconstruction methods based on convolutional neural networks |
CN108596994A (en) * | 2018-04-24 | 2018-09-28 | 朱高杰 | A kind of Diffusion-weighted imaging method being in harmony certainly based on deep learning and data |
CN108629784A (en) * | 2018-05-08 | 2018-10-09 | 上海嘉奥信息科技发展有限公司 | A kind of CT image intracranial vessel dividing methods and system based on deep learning |
CN108921788A (en) * | 2018-06-20 | 2018-11-30 | 华北电力大学 | Image super-resolution method, device and storage medium based on deep layer residual error CNN |
CN109003229A (en) * | 2018-08-09 | 2018-12-14 | 成都大学 | Magnetic resonance super resolution ratio reconstruction method based on three-dimensional enhancing depth residual error network |
CN109523584A (en) * | 2018-10-26 | 2019-03-26 | 上海联影医疗科技有限公司 | Image processing method, device, multi-mode imaging system, storage medium and equipment |
CN109712077A (en) * | 2018-12-29 | 2019-05-03 | 成都信息工程大学 | A kind of HARDI compressed sensing super resolution ratio reconstruction method based on depth dictionary learning |
CN109754357A (en) * | 2018-01-26 | 2019-05-14 | 京东方科技集团股份有限公司 | Image processing method, processing unit and processing equipment |
CN109816742A (en) * | 2018-12-14 | 2019-05-28 | 中国人民解放军战略支援部队信息工程大学 | Cone-Beam CT geometry artifact minimizing technology based on full connection convolutional neural networks |
CN109903353A (en) * | 2019-01-28 | 2019-06-18 | 华南理工大学 | A kind of CT image sparse method for reconstructing of iteration evolutionary model |
CN109993694A (en) * | 2017-12-29 | 2019-07-09 | Tcl集团股份有限公司 | A kind of method and device generating super-resolution image |
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 |
CN110135375A (en) * | 2019-05-20 | 2019-08-16 | 中国科学院宁波材料技术与工程研究所 | More people's Attitude estimation methods based on global information integration |
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 |
CN110322400A (en) * | 2018-03-30 | 2019-10-11 | 京东方科技集团股份有限公司 | Image processing method and device, image processing system and its training method |
CN110443755A (en) * | 2019-08-07 | 2019-11-12 | 杭州智团信息技术有限公司 | A method of the image super-resolution based on low-and high-frequency semaphore |
CN110599399A (en) * | 2019-07-26 | 2019-12-20 | 清华大学 | Fast two-photon imaging method and device based on convolutional neural network |
CN110826467A (en) * | 2019-11-22 | 2020-02-21 | 中南大学湘雅三医院 | Electron microscope image reconstruction system and method |
EP3637099A1 (en) * | 2018-10-08 | 2020-04-15 | Ecole Polytechnique Federale de Lausanne (EPFL) | Image reconstruction method based on a trained non-linear mapping |
WO2020079605A1 (en) * | 2018-10-16 | 2020-04-23 | Indian Institute Of Science | Device and method for enhancing readability of a low-resolution binary image |
CN111340682A (en) * | 2018-12-19 | 2020-06-26 | 通用电气公司 | Method and system for converting medical image into different-style image by using deep neural network |
CN111839574A (en) * | 2020-09-08 | 2020-10-30 | 南京安科医疗科技有限公司 | CT ultralow-dose automatic three-dimensional positioning scanning method and system |
CN112419192A (en) * | 2020-11-24 | 2021-02-26 | 北京航空航天大学 | Convolutional neural network-based ISMS image restoration and super-resolution reconstruction method and device |
CN113888410A (en) * | 2021-09-30 | 2022-01-04 | 北京百度网讯科技有限公司 | Image super-resolution method, apparatus, device, storage medium, and program product |
US11449751B2 (en) | 2018-09-30 | 2022-09-20 | Boe Technology Group Co., Ltd. | Training method for generative adversarial network, image processing method, device and storage medium |
US11551333B2 (en) * | 2017-12-20 | 2023-01-10 | Huawei Technologies Co., Ltd. | Image reconstruction method and device |
WO2024032075A1 (en) * | 2022-08-08 | 2024-02-15 | 华为技术有限公司 | Training method for image processing network, and coding method, decoding method, and electronic device |
Citations (3)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN104112263A (en) * | 2014-06-28 | 2014-10-22 | 南京理工大学 | Method for fusing full-color image and multispectral image based on deep neural network |
CN105072373A (en) * | 2015-08-28 | 2015-11-18 | 中国科学院自动化研究所 | Bilateral-circulation convolution network-based video super-resolution method and system |
CN105976318A (en) * | 2016-04-28 | 2016-09-28 | 北京工业大学 | Image super-resolution reconstruction method |
-
2017
- 2017-07-21 CN CN201710600369.XA patent/CN107369189A/en active Pending
Patent Citations (3)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN104112263A (en) * | 2014-06-28 | 2014-10-22 | 南京理工大学 | Method for fusing full-color image and multispectral image based on deep neural network |
CN105072373A (en) * | 2015-08-28 | 2015-11-18 | 中国科学院自动化研究所 | Bilateral-circulation convolution network-based video super-resolution method and system |
CN105976318A (en) * | 2016-04-28 | 2016-09-28 | 北京工业大学 | Image super-resolution reconstruction method |
Non-Patent Citations (8)
Title |
---|
CHAO DONG等: ""Image super-resolution using deep convolutional networks"", 《IEEE TRANSACTIONS ON PATTERN ANALYSIS AND MACHINE INTELLIGENCE》 * |
EVAN SHELHAMER等: ""Fully convolutional networks for semantic segmentation"", 《IEEE TRANSACTIONS ON PATTERN ANALYSIS AND MACHINE INTELLIGENCE》 * |
JUSTIN JOHNSON等: ""Perceptual Losses for Real-Time Style Transfer and Super-Resolution"", 《EUROPEAN CONFERENCE ON COMPUTER VISION 2016》 * |
KAIMING HE等: ""Deep residual learning for image recognition"", 《PROCEEDINGS OF THE IEEE CONFERENCE ON COMPUTER VISION AND PATTERN RECOGNITION》 * |
KATHIRAVAN SRINIVASAN等: ""Super-resolution of magnetic resonance images using deep convolutional neural networks"", 《2017 IEEE INTERNATIONAL CONFERENCE ON CONSUMER ELECTRONICS-TAIWAN (ICCE-TW)》 * |
YIZHEN HUANG等: ""Super-resolution using neural networks based on the optimal recovery theory"", 《PROCEEDINGS OF THE 2006 16TH IEEE SIGNAL PROCESSING SOCIETY WORKSHOP ON MACHINE LEARNING FOR SIGNAL PROCESSING》 * |
唐云岚等: ""基于跨越连接的多层前馈神经网络结构分析"", 《计算机工程与应用》 * |
赵小乐: ""单幅图像超分辨技术研究"", 《万方数据企业知识服务平台》 * |
Cited By (48)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN108090871A (en) * | 2017-12-15 | 2018-05-29 | 厦门大学 | A kind of more contrast MR image reconstruction methods based on convolutional neural networks |
CN108090871B (en) * | 2017-12-15 | 2020-05-08 | 厦门大学 | Multi-contrast magnetic resonance image reconstruction method based on convolutional neural network |
US11551333B2 (en) * | 2017-12-20 | 2023-01-10 | Huawei Technologies Co., Ltd. | Image reconstruction method and device |
CN107948529A (en) * | 2017-12-28 | 2018-04-20 | 北京麒麟合盛网络技术有限公司 | Image processing method and device |
CN107948529B (en) * | 2017-12-28 | 2020-11-06 | 麒麟合盛网络技术股份有限公司 | Image processing method and device |
CN109993694A (en) * | 2017-12-29 | 2019-07-09 | Tcl集团股份有限公司 | A kind of method and device generating super-resolution image |
US11281938B2 (en) | 2018-01-26 | 2022-03-22 | Boe Technology Group Co., Ltd. | Image processing method, processing apparatus and processing device |
CN109754357A (en) * | 2018-01-26 | 2019-05-14 | 京东方科技集团股份有限公司 | Image processing method, processing unit and processing equipment |
CN110322400A (en) * | 2018-03-30 | 2019-10-11 | 京东方科技集团股份有限公司 | Image processing method and device, image processing system and its training method |
US11189013B2 (en) | 2018-03-30 | 2021-11-30 | Boe Technology Group Co., Ltd. | Image processing apparatus, image processing method thereof, image processing system, and training method thereof |
CN110322400B (en) * | 2018-03-30 | 2021-04-27 | 京东方科技集团股份有限公司 | Image processing method and device, image processing system and training method thereof |
CN108596994A (en) * | 2018-04-24 | 2018-09-28 | 朱高杰 | A kind of Diffusion-weighted imaging method being in harmony certainly based on deep learning and data |
CN108596994B (en) * | 2018-04-24 | 2022-05-03 | 朱高杰 | Magnetic resonance diffusion weighted imaging method based on deep learning and data self-consistency |
CN108629784A (en) * | 2018-05-08 | 2018-10-09 | 上海嘉奥信息科技发展有限公司 | A kind of CT image intracranial vessel dividing methods and system based on deep learning |
CN108921788A (en) * | 2018-06-20 | 2018-11-30 | 华北电力大学 | Image super-resolution method, device and storage medium based on deep layer residual error CNN |
CN109003229B (en) * | 2018-08-09 | 2022-12-13 | 成都大学 | Magnetic resonance super-resolution reconstruction method based on three-dimensional enhanced depth residual error network |
CN109003229A (en) * | 2018-08-09 | 2018-12-14 | 成都大学 | Magnetic resonance super resolution ratio reconstruction method based on three-dimensional enhancing depth residual error network |
US11449751B2 (en) | 2018-09-30 | 2022-09-20 | Boe Technology Group Co., Ltd. | Training method for generative adversarial network, image processing method, device and storage medium |
EP3637099A1 (en) * | 2018-10-08 | 2020-04-15 | Ecole Polytechnique Federale de Lausanne (EPFL) | Image reconstruction method based on a trained non-linear mapping |
CN112771374A (en) * | 2018-10-08 | 2021-05-07 | 洛桑联邦理工学院 | Image reconstruction method based on training nonlinear mapping |
WO2020074181A1 (en) * | 2018-10-08 | 2020-04-16 | Ecole Polytechnique Federale De Lausanne (Epfl) | Image reconstruction method based on a trained non-linear mapping |
US11846608B2 (en) | 2018-10-08 | 2023-12-19 | Ecole Polytechnique Federale De Lausanne (Epfl) | Image reconstruction method based on a trained non-linear mapping |
WO2020079605A1 (en) * | 2018-10-16 | 2020-04-23 | Indian Institute Of Science | Device and method for enhancing readability of a low-resolution binary image |
CN109523584A (en) * | 2018-10-26 | 2019-03-26 | 上海联影医疗科技有限公司 | Image processing method, device, multi-mode imaging system, storage medium and equipment |
CN109816742A (en) * | 2018-12-14 | 2019-05-28 | 中国人民解放军战略支援部队信息工程大学 | Cone-Beam CT geometry artifact minimizing technology based on full connection convolutional neural networks |
CN109816742B (en) * | 2018-12-14 | 2022-10-28 | 中国人民解放军战略支援部队信息工程大学 | Cone beam CT geometric artifact removing method based on fully-connected convolutional neural network |
CN111340682B (en) * | 2018-12-19 | 2023-12-05 | 通用电气公司 | Method and system for converting medical image into different-style image by deep neural network |
CN111340682A (en) * | 2018-12-19 | 2020-06-26 | 通用电气公司 | Method and system for converting medical image into different-style image by using deep neural network |
CN109712077B (en) * | 2018-12-29 | 2020-07-17 | 成都信息工程大学 | Depth dictionary learning-based HARDI (hybrid automatic repeat-based) compressed sensing super-resolution reconstruction method |
CN109712077A (en) * | 2018-12-29 | 2019-05-03 | 成都信息工程大学 | A kind of HARDI compressed sensing super resolution ratio reconstruction method based on depth dictionary learning |
CN109903353B (en) * | 2019-01-28 | 2023-02-14 | 华南理工大学 | CT image sparse reconstruction method of iterative evolution model |
CN109903353A (en) * | 2019-01-28 | 2019-06-18 | 华南理工大学 | A kind of CT image sparse method for reconstructing of iteration evolutionary model |
CN110264410B (en) * | 2019-05-07 | 2021-06-15 | 西安理工大学 | Image super-resolution reconstruction method based on detail features |
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 |
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 |
CN110135375A (en) * | 2019-05-20 | 2019-08-16 | 中国科学院宁波材料技术与工程研究所 | More people's Attitude estimation methods based on global information integration |
CN110599399B (en) * | 2019-07-26 | 2022-02-18 | 清华大学 | Fast two-photon imaging method and device based on convolutional neural network |
CN110599399A (en) * | 2019-07-26 | 2019-12-20 | 清华大学 | Fast two-photon imaging method and device based on convolutional neural network |
CN110443755B (en) * | 2019-08-07 | 2023-05-30 | 杭州智团信息技术有限公司 | Image super-resolution method based on high-low frequency signal quantity |
CN110443755A (en) * | 2019-08-07 | 2019-11-12 | 杭州智团信息技术有限公司 | A method of the image super-resolution based on low-and high-frequency semaphore |
CN110826467B (en) * | 2019-11-22 | 2023-09-29 | 中南大学湘雅三医院 | Electron microscope image reconstruction system and method thereof |
CN110826467A (en) * | 2019-11-22 | 2020-02-21 | 中南大学湘雅三医院 | Electron microscope image reconstruction system and method |
CN111839574B (en) * | 2020-09-08 | 2023-10-31 | 南京安科医疗科技有限公司 | CT ultralow-dose automatic three-dimensional positioning scanning method and system |
CN111839574A (en) * | 2020-09-08 | 2020-10-30 | 南京安科医疗科技有限公司 | CT ultralow-dose automatic three-dimensional positioning scanning method and system |
CN112419192A (en) * | 2020-11-24 | 2021-02-26 | 北京航空航天大学 | Convolutional neural network-based ISMS image restoration and super-resolution reconstruction method and device |
CN112419192B (en) * | 2020-11-24 | 2022-09-09 | 北京航空航天大学 | Convolutional neural network-based ISMS image restoration and super-resolution reconstruction method and device |
CN113888410A (en) * | 2021-09-30 | 2022-01-04 | 北京百度网讯科技有限公司 | Image super-resolution method, apparatus, device, storage medium, and program product |
WO2024032075A1 (en) * | 2022-08-08 | 2024-02-15 | 华为技术有限公司 | Training method for image processing network, and coding method, decoding method, and electronic device |
Similar Documents
Publication | Publication Date | Title |
---|---|---|
CN107369189A (en) | The medical image super resolution ratio reconstruction method of feature based loss | |
CN106683067B (en) | Deep learning super-resolution reconstruction method based on residual sub-images | |
McCann et al. | Convolutional neural networks for inverse problems in imaging: A review | |
WO2021077997A1 (en) | Multi-generator generative adversarial network learning method for image denoising | |
CN104008538B (en) | Based on single image super-resolution method | |
CN102142137B (en) | High-resolution dictionary based sparse representation image super-resolution reconstruction method | |
CN107464216A (en) | A kind of medical image ultra-resolution ratio reconstructing method based on multilayer convolutional neural networks | |
CN105046672B (en) | A kind of image super-resolution rebuilding method | |
CN110310227A (en) | A kind of image super-resolution rebuilding method decomposed based on high and low frequency information | |
CN109671023A (en) | A kind of secondary method for reconstructing of face image super-resolution | |
CN110189253A (en) | A kind of image super-resolution rebuilding method generating confrontation network based on improvement | |
CN108416821B (en) | A kind of CT Image Super-resolution Reconstruction method of deep neural network | |
CN106934766A (en) | A kind of infrared image super resolution ratio reconstruction method based on rarefaction representation | |
WO2021022929A1 (en) | Single-frame image super-resolution reconstruction method | |
CN109544457A (en) | Image super-resolution method, storage medium and terminal based on fine and close link neural network | |
Chen et al. | Single image super-resolution using deep CNN with dense skip connections and inception-resnet | |
Xiao et al. | A dual-UNet with multistage details injection for hyperspectral image fusion | |
Yang et al. | Image super-resolution based on deep neural network of multiple attention mechanism | |
CN110322402A (en) | Medical image super resolution ratio reconstruction method based on dense mixing attention network | |
CN111899165A (en) | Multi-task image reconstruction convolution network model based on functional module | |
CN108921783A (en) | A kind of satellite image super resolution ratio reconstruction method based on losses by mixture function constraint | |
CN107845065A (en) | Super-resolution image reconstruction method and device | |
CN110533591A (en) | Super resolution image reconstruction method based on codec structure | |
CN104825161B (en) | High-quality lung MR imaging method based on excessively complete dictionary Yu priori | |
He et al. | Remote sensing image super-resolution using deep–shallow cascaded convolutional neural 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 | ||
RJ01 | Rejection of invention patent application after publication |
Application publication date: 20171121 |
|
RJ01 | Rejection of invention patent application after publication |