CN110390647A - The OCT image denoising method and device for generating network are fought based on annular - Google Patents
The OCT image denoising method and device for generating network are fought based on annular Download PDFInfo
- Publication number
- CN110390647A CN110390647A CN201910515611.2A CN201910515611A CN110390647A CN 110390647 A CN110390647 A CN 110390647A CN 201910515611 A CN201910515611 A CN 201910515611A CN 110390647 A CN110390647 A CN 110390647A
- Authority
- CN
- China
- Prior art keywords
- image
- oct image
- oct
- sample
- network model
- 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 47
- 238000012549 training Methods 0.000 claims abstract description 45
- 238000010276 construction Methods 0.000 claims description 45
- 230000004069 differentiation Effects 0.000 claims description 43
- 230000006870 function Effects 0.000 claims description 21
- 238000003475 lamination Methods 0.000 claims description 4
- 230000000694 effects Effects 0.000 abstract description 6
- 238000013135 deep learning Methods 0.000 abstract description 5
- 238000013473 artificial intelligence Methods 0.000 abstract description 2
- 238000012014 optical coherence tomography Methods 0.000 description 188
- 230000008569 process Effects 0.000 description 5
- 238000012545 processing Methods 0.000 description 5
- 238000010586 diagram Methods 0.000 description 4
- 238000004891 communication Methods 0.000 description 3
- 238000005516 engineering process Methods 0.000 description 3
- 238000003062 neural network model Methods 0.000 description 3
- ORILYTVJVMAKLC-UHFFFAOYSA-N Adamantane Natural products C1C(C2)CC3CC1CC2C3 ORILYTVJVMAKLC-UHFFFAOYSA-N 0.000 description 2
- 238000013528 artificial neural network Methods 0.000 description 2
- 238000013527 convolutional neural network Methods 0.000 description 2
- 239000004973 liquid crystal related substance Substances 0.000 description 2
- 238000013507 mapping Methods 0.000 description 2
- 230000005055 memory storage Effects 0.000 description 2
- 238000009394 selective breeding Methods 0.000 description 2
- 108010001267 Protein Subunits Proteins 0.000 description 1
- 206010064930 age-related macular degeneration Diseases 0.000 description 1
- 230000009286 beneficial effect Effects 0.000 description 1
- 230000003115 biocidal effect Effects 0.000 description 1
- 238000006243 chemical reaction Methods 0.000 description 1
- 208000021921 corneal disease Diseases 0.000 description 1
- 238000012631 diagnostic technique Methods 0.000 description 1
- 230000005611 electricity Effects 0.000 description 1
- 208000002780 macular degeneration Diseases 0.000 description 1
- 238000005457 optimization Methods 0.000 description 1
- 230000009467 reduction Effects 0.000 description 1
- 238000012360 testing method Methods 0.000 description 1
- 238000012546 transfer Methods 0.000 description 1
- 230000009466 transformation Effects 0.000 description 1
- 230000000007 visual effect Effects 0.000 description 1
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/04—Architecture, e.g. interconnection topology
- G06N3/045—Combinations of networks
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/08—Learning methods
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T5/00—Image enhancement or restoration
- G06T5/70—Denoising; Smoothing
-
- 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
- G06T2207/10101—Optical tomography; Optical coherence tomography [OCT]
-
- 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
- G06T2207/20081—Training; Learning
-
- 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
- G06T2207/20084—Artificial neural networks [ANN]
Landscapes
- Engineering & Computer Science (AREA)
- Physics & Mathematics (AREA)
- Theoretical Computer Science (AREA)
- General Physics & Mathematics (AREA)
- Data Mining & Analysis (AREA)
- Molecular Biology (AREA)
- Biophysics (AREA)
- Computational Linguistics (AREA)
- Artificial Intelligence (AREA)
- Evolutionary Computation (AREA)
- General Health & Medical Sciences (AREA)
- Biomedical Technology (AREA)
- Computing Systems (AREA)
- General Engineering & Computer Science (AREA)
- Life Sciences & Earth Sciences (AREA)
- Mathematical Physics (AREA)
- Software Systems (AREA)
- Health & Medical Sciences (AREA)
- Image Analysis (AREA)
Abstract
The invention belongs to field of artificial intelligence, disclose a kind of OCT image denoising method and device fought based on annular and generate network, the OCT image denoising method includes: to obtain OCT image to be denoised;The annular confrontation that the OCT image input to be denoised is obtained by training is generated into network model;The OCT image of network model output denoising is generated by the annular confrontation.The present invention generates network model by annular confrontation and carries out denoising to OCT image, effectively converts clearly OCT image for the OCT image of strong noise, distinguishes image in order to doctor or OCT image is used for software analysis.Also, the deep learning limitation that must match of training data in denoising application before avoiding, is conducive to obtain a large amount of data and is trained, to improve the denoising effect of model.
Description
Technical field
The present invention relates to field of artificial intelligence more particularly to a kind of OCT images that generation network is fought based on annular
Denoising method and device.
Background technique
Optical coherence tomography (Optical Coherence tomography, OCT) is as a kind of emerging optics
Diagnostic techniques can be used to check the finer position such as eyes, help patient earlier, more accurately diagnosing and treating green light
Eye, disease of cornea, age related macular degeneration etc..But OCT image is easy to produce noise, either for doctor distinguish image or
It is to be analyzed for software, all brings huge challenge.It generally requires and strong noise image is first denoised, then give doctor and read
Piece carries out software analysis.Currently, carrying out the technology of Denoising disposal to figure includes computer vision and two kinds of deep learning
There is the problems such as processing overlong time by taking BM3D as an example in direction, computer vision methods, and as deep learning is applied to image
Denoising field, for example, by using multilayer perceptron, autocoder and convolutional neural networks method, so that denoising effect is more traditional
Computer vision methods are improved.
Generating confrontation neural network (Generative Adversarial Networks, GAN) is in deep learning algorithm
A kind of new network, by the generation network that is constructed by convolutional neural networks and judge the training of network progress confrontation type, quilt
It is widely applied to the fields such as image conversion, image procossing.But at present mostly with condition confrontation neural network to OCT image
Denoising is carried out, needs noise image, clear image during being trained to condition confrontation neural network model
It matches one by one, is unfavorable for obtaining a large amount of OCT image data for training, so that training effect is bad, so as to cause condition confrontation
The effect that neural network model carries out OCT image denoising is undesirable.
Summary of the invention
The present invention provides a kind of OCT image denoising method, device and medium for fighting based on annular and generating network, to solve
Use condition confrontation neural network model carries out the undesirable problem of denoising effect to OCT image in the prior art.
To achieve the goals above, it is an aspect of the invention to provide a kind of OCT that generation network is fought based on annular
Image de-noising method, comprising: obtain OCT image to be denoised;The OCT image input to be denoised is obtained by training
Annular confrontation generates network model;The OCT image of network model output denoising is generated by the annular confrontation.
Preferably, the annular confrontation generates network model including generating network model and differentiating network model, the life
It include two generators, respectively the first generator and the second generator at network model, the differentiation network model includes two
A arbiter, respectively the first arbiter and the second arbiter, the corresponding arbiter of each generator, wherein described first
Generator and second generator are made of several convolutional layers with step-length and warp lamination;First arbiter and institute
The second arbiter is stated to be made of several convolutional layers with step-length and full articulamentum.
Preferably, the method also includes building training sample set, including noise image sample and clear image samples;
Network model is generated to the annular confrontation based on the training sample set to be trained.
Preferably, the step of constructing training sample set includes: to obtain OCT image sample database;Selection criteria OCT image;It will
The standard OCT image traverses the sample in the OCT image sample database, seeks each OCT figure in the OCT image sample database
Decent Y-PSNR index;The Y-PSNR index is sorted from high to low;Select corresponding Y-PSNR
The predetermined number OCT image sample that index sorts forward selects corresponding Y-PSNR index as clear image sample
The predetermined number OCT image sample of sequence rearward is as noise image sample.
Preferably, the step of constructing training sample set includes: to obtain OCT image sample database;Multiple standards OCT is selected to scheme
Picture;Multiple described standard OCT images are successively traversed into the sample in the OCT image sample database, seek the OCT image respectively
Each OCT image sample corresponds to the Y-PSNR index of every standard OCT image in sample database;By an OCT image sample
Originally multiple Y-PSNR indexes that multiple standard OCT images obtain are corresponded respectively to and seek mean value, obtain one OCT figure
Decent Y-PSNR Mean value of index;The Y-PSNR Mean value of index is sorted from high to low;Select corresponding peak
The predetermined number OCT image sample that value signal-to-noise ratio Mean value of index sorts forward selects corresponding peak as clear image sample
It is worth the predetermined number OCT image sample of signal-to-noise ratio Mean value of index sequence rearward as noise image sample.
Preferably, the Y-PSNR index of OCT image sample is sought by following formula:
In formula, PSNR indicates the Y-PSNR of OCT image sample;MSE indicates standard OCT image and current OCT image
Mean square error between sample;X indicates the bit number of each sampled value.
The training sample set is preferably based on to be trained the annular confrontation generation network model, comprising:
The noise image sample is inputted into first generator, exports corresponding first construction image;
The first construction image is inputted into second generator, exports corresponding second construction image;
The noise image sample and the second construction image are inputted into second arbiter, sentenced by described second
Other device exports the probability value that the second construction image is determined as true noise image;
The first construction image and clear image sample are inputted into first arbiter, pass through first arbiter
Export the probability value that the first construction image is determined as true clear image;
Judge whether the differentiation network model restrains, if the differentiation network model is restrained, by the differentiation network
Denoising image of the first construction image of model output as the noise image sample;If the differentiation network model is not received
It holds back, then updates the parameter for generating network model until the differentiation network model is restrained.
Preferably, judge whether the differentiation network model restrains, comprising:
Loss function model is constructed, is shown below:
L(GA2B,GB2A,DA,DB)
=LGAN(GA2B,DB,A,B)
+LGAN(GB2A,DA,B,A)+λLcyc(GA2B,GB2A,A,B)
Wherein, L indicates penalty values, and GAN indicates that confrontation generates network, and G indicates that generator, D indicate arbiter, and A expression is made an uproar
Sound spectrogram image field, B indicate clear image domain, and A2B indicates that image area A is mapped to image area B, and B2A indicates that image area B is mapped to image
Domain A, λ take empirical value, LcycIndicate circulation consistency loss;
Loss function model based on construction judges whether the differentiation network model restrains.
Preferably, judge whether the differentiation network model restrains, comprising: judgement is differentiated by the first arbiter or second
Whether the probability value of device output is all larger than respective preset probability threshold value, if being all larger than preset probability threshold value, described in judgement
Differentiate network model convergence, if the probability value of at least one arbiter output is less than corresponding predetermined probabilities threshold value, determines institute
It states and differentiates that network model is not restrained.
To achieve the goals above, another aspect of the present invention is to provide a kind of electronic device, which includes:
Processor;Memory includes fighting to generate the OCT image of network and denoise program, the OCT figure based on annular in the memory
The step of realizing OCT image denoising method as described above when being executed as denoising program by the processor.
To achieve the goals above, another aspect of the invention is to provide a kind of computer readable storage medium, described
It include that the OCT image denoising program for generating network is fought based on annular in computer readable storage medium, the OCT image denoising
When program is executed by processor, the step of realizing OCT image denoising method as described above.
Compared with the existing technology, the present invention has the following advantages and beneficial effects:
The present invention generates network model by annular confrontation and carries out denoising to OCT image, effectively by strong noise
OCT image is converted into clearly OCT image, distinguishes image in order to doctor or the OCT image after denoising is used for software analysis.
Also, present invention reduces the times spent by image noise reduction, improve efficiency.Also, it is generated at network model using annular confrontation
Image is managed, the deep learning limitation that training data must match in denoising application, no longer needs to make an uproar when training before avoiding
The pairing one by one of acoustic image and clear image, is conducive to obtain a large amount of data and is trained, and obtains good training result, from
And improve the denoising effect of model.
Detailed description of the invention
Fig. 1 is the flow diagram of the present invention for being fought based on annular and generating the OCT image denoising method of network;
Fig. 2 is the module diagram that the OCT image for fighting generation network based on annular in the present invention denoises program.
The embodiments will be further described with reference to the accompanying drawings for the realization, the function and the advantages of the object of the present invention.
Specific embodiment
Embodiment of the present invention described below with reference to the accompanying drawings.Those skilled in the art may recognize that
It arrives, it without departing from the spirit and scope of the present invention, can be with a variety of different modes or combinations thereof to described
Embodiment is modified.Therefore, attached drawing and description are regarded as illustrative in nature, and are only used to explain the present invention, rather than are used
In limitation scope of protection of the claims.In addition, in the present specification, attached drawing is drawn not in scale, and identical attached drawing mark
Note indicates identical part.
Fig. 1 is the flow diagram of the present invention for being fought based on annular and generating the OCT image denoising method of network, is such as schemed
It is of the present invention that the OCT image denoising method for generating network is fought based on annular shown in 1, comprising the following steps:
Step S1, OCT image to be denoised is obtained;
Step S2, the annular confrontation that the OCT image input to be denoised is obtained by training is generated into network model, it is right
The OCT image carries out denoising;
Step S3, the OCT image of network model output denoising is generated by the annular confrontation.
The present invention generates network model by annular confrontation and carries out denoising to OCT image, effectively by strong noise
OCT image is converted into clearly OCT image, distinguishes image in order to doctor or OCT image is used for software analysis.
Annular confrontation generates network model including generating network model and differentiating network model, wherein generates network model
For noise picture to be transformed into clear picture, the input for generating network model is noise-containing OCT image, output be with not
Image after the denoising as similar as possible of the OCT image of Noise differentiates that network model is used to judge the clear picture that transformation obtains
Whether it is true clear image data, differentiates that the input of network model is an image pair, image is to including noise-containing
Noise OCT image and by generate network denoising after clear OCT image, not the OCT image of Noise and by generate network
Clear OCT image after denoising.Generating network model, deception differentiates network model as much as possible in the training process, and differentiates net
Network model makes correct differentiation to image as far as possible.The generation network model includes two generators, and respectively first is raw
It grows up to be a useful person and the second generator, the differentiation network model includes two arbiters, respectively the first arbiter and the second arbiter,
The corresponding arbiter of each generator, wherein first generator and second generator are by several with step-length
Convolutional layer and warp lamination composition;First arbiter and second arbiter are by several convolutional layers and Quan Lian with step-length
Connect layer composition.
In order to improve the performance that annular fights generation network model, need to generate network model progress sufficiently to annular confrontation
Training so that the weight parameter in model is optimal value, Optimized model performance.Preferably, the method also includes: to institute
It states annular confrontation and generates the step of network model is trained, specifically, comprising: building training sample set, including noise image
Sample and clear image sample;It generates network model to the annular confrontation based on the training sample set to be trained, training
When annular confrontation generates network, it is no longer necessary to the one-to-one correspondence of noise image sample and clear image sample.
In an alternative embodiment of the invention, building training sample set includes: to obtain OCT image sample database, for example, figure
As including about 10000 OCT image samples in sample database, image pattern building training sample is selected from image pattern library
Collection;Selection criteria OCT image carries out artificial selection to OCT image, high quality, clearly is selected except OCT image sample database
OCT image is as standard picture;The standard OCT image is traversed into the sample in the OCT image sample database, is sought described
Y-PSNR index (PSNR, Peak Signal the to Noise of each OCT image sample in OCT image sample database
Ratio), wherein PSNR index is higher, and corresponding OCT image sample is more clear, noise is fewer, and PSNR index is lower, corresponding
The noise of OCT image sample is more serious;The PSNR index is sorted from high to low;Select corresponding PSNR index sequence forward
The predetermined number OCT image sample of (PSNR index is higher) selects corresponding PSNR index to sort as clear image sample
The predetermined number OCT image sample of (PSNR index is lower) is as noise image sample rearward.Wherein, predetermined number is according to instruction
Practice in sample set depending on required sample size, the numerical value that predetermined number can be some determination is also possible to the ratio of setting
Example value, when value is determined predetermined number according to set proportion, noise pattern required for being concentrated according to training sample
As ratio is set separately in sample number and clear image sample size, for example, can set selection percentage is 3%, PSNR is referred to
After number sequence, according to 3% forward image pattern of selected and sorted from high to low as clear image sample, according to from low to high
3% image pattern of selected and sorted rearward is as noise image sample.
Preferably, the PSNR index of OCT image sample is sought by following formula:
In formula, PSNR indicates the Y-PSNR of OCT image sample;MSE indicates standard OCT image and current OCT image
Mean square error between sample;X indicates the bit number of each sampled value.
Wherein, mean square deviation MSE is obtained by following formula:
In formula, I is standard OCT image, and K is current OCT image sample, and m*n is the size of image, standard OCT image and
Current OCT image size is identical.
Preferably, building training sample set includes: to obtain OCT image sample database;Select multiple standard OCT images;It will be more
Zhang Suoshu standard OCT image successively traverses the sample in the OCT image sample database, seeks the OCT image sample database respectively
In each OCT image sample correspond to every standard OCT image PSNR index;One OCT image sample is corresponded respectively to
Multiple PSNR indexes that multiple standard OCT images obtain seek mean value, and the PSNR index for obtaining one OCT image sample is equal
Value;The PSNR Mean value of index is sorted from high to low;Corresponding PSNR Mean value of index is selected to sort forward predetermined number
OCT image sample selects the predetermined number OCT image of corresponding PSNR Mean value of index sequence rearward as clear image sample
Sample is as noise image sample.For example, selecting 16 high quality, when clearly OCT image is as standard picture, 16 are marked
Quasi- image successively traverses the data in the OCT image sample database, and 16 standard pictures are asked with the OCT image sample of processing respectively
A PSNR index is obtained, 16 PSNR indexes are sought into mean value, the final PSNR for representing currently evaluated OCT image sample refers to
Number, constantly repeats this step, until all OCT image sample standard deviation is evaluated, obtained final PSNR index is higher, illustrates pair
The OCT image sample answered is more clear, noise is fewer, and final PSNR index is lower, and the noise of corresponding OCT image sample is tighter
Weight.For PSNR index seek and the selection mode of final sample is similar with foregoing embodiments, details are not described herein.
In the present invention, the clear image sample size of selection is identical as noise image sample size, in OCT image sample database
In occupy same ratio, for example, in OCT image sample database include about 10000 OCT images when, select 3% data as clearly
Sample (about 300) equally selects 3% data as noise sample (about 300).
In an alternative embodiment of the invention, network model is generated to the annular confrontation based on the training sample set
It is trained, comprising: noise image sample is inputted into first generator, exports corresponding first construction image;
The first construction image is inputted into second generator, exports corresponding second construction image;
The noise image sample and the second construction image are inputted into second arbiter, sentenced by described second
Other device exports the probability value that the second construction image is determined as true noise image;
The first construction image and clear image sample are inputted into first arbiter, pass through first arbiter
Export the probability value that the first construction image is determined as true clear image;
Judge whether the differentiation network model restrains, if the differentiation network model is restrained, by the differentiation network
Denoising image of the first construction image of model output as the noise image sample;If the differentiation network model is not received
It holds back, then updates the parameter for generating network model until the differentiation network model is restrained.Using Adam Optimization Learning method,
Use learning rate for 0.0002, crowd size batch size is 1, and the training of 200 wheels is carried out to it.
In an alternative embodiment of the invention, judge whether the differentiation network model restrains, comprising: construction loss letter
Exponential model;Loss function model based on construction judges whether the differentiation network model restrains.Further, the loss of construction
Function model includes: generational loss function model, obtains the loss of generator;Differentiate loss function model, obtains arbiter
Loss;Circulation loss function model obtains circulation consistency loss, is realized by the circulation consistency loss of introducing and schemes one
Data in image field are uniformly mapped in another image area, avoid the chance phenomenon of mapping relations (for example, by an image
Data in domain are all mapped to a certain picture in another image area) generation.
The loss function that the annular confrontation generates network model is shown below:
L(GA2B,GB2A,DA,DB)
=LGAN(GA2B,DB,A,B)
+LGAN(GB2A,DA,B,A)+λLcyc(GA2B,GB2A,A,B)
Wherein, L indicates penalty values, and GAN indicates that confrontation generates network, and G indicates that generator, D indicate arbiter, and A expression is made an uproar
Sound spectrogram image field, B indicate clear image domain, and A2B indicates that image area A is mapped to image area B, and B2A indicates that image area B is mapped to image
Domain A, λ can use empirical value, λ=10, LcycIndicate circulation consistency loss.
Wherein, each penalty values are obtained by following formula respectively:
Wherein,It indicates that a is derived from and is distributed as paNoise pattern image field,It indicates that b is derived from and is distributed as pbIt is clear
Image area.
In one embodiment of the present of invention, judge whether the differentiation network model restrains, comprising: judgement is sentenced by first
Whether other device or the probability value of the second arbiter output are all larger than respective preset probability threshold value, if being all larger than preset probability threshold
Value then determines the differentiation network model convergence, if the probability value of at least one arbiter output is less than corresponding predetermined probabilities
Threshold value then determines that the differentiation network model is not restrained.
It is to be appreciated that " first " and " second " is only used for distinguishing two same or similar objects, without
Represent the meanings such as specific sequence or precedence.The data used in this way are interchangeable under appropriate circumstances, so as to herein
The embodiment of the present invention of description can be implemented with handling the sequence other than those of illustrating or describe herein.
In order to verify the validity of OCT image denoising method of the present invention, in noisy Heidelberg OCT image data
On done related denoising test, will verify, the clear and legible knowledge of the noise-reduced image after Style Transfer, for the analysis of further software and
Artificial diagosis lays the foundation.
OCT image denoising method of the present invention that generate network of being fought based on annular is applied to electronic device, the electricity
Sub-device can be the terminal devices such as television set, smart phone, tablet computer, computer.
The electronic device includes: processor;Memory fights the OCT image for generating network based on annular for storing
Program is denoised, processor executes the OCT image for generating network based on annular confrontation and denoises program, realizes below based on ring
The step of shape confrontation generates the OCT image denoising method of network:
Obtain OCT image to be denoised;The annular that the OCT image input to be denoised is obtained by training is to antibiosis
At network model;The OCT image of network model output denoising is generated by the annular confrontation.
The electronic device further includes network interface and communication bus etc..Wherein, network interface may include having for standard
Line interface, wireless interface, communication bus is for realizing the connection communication between various components.
Memory includes the readable storage medium storing program for executing of at least one type, can be that flash memory, hard disk, CD etc. are non-volatile to be deposited
Storage media is also possible to plug-in type hard disk etc., and is not limited to this, can be in a manner of non-transitory store instruction or software with
And any associated data file and to processor provide instruction or software program so that the processor be able to carry out instruction or
Any device of software program.In the present invention, the software program of memory storage includes that the OCT for generating network is fought based on annular
Image denoising program, and OCT image denoising program can be provided to processor, so that processor can execute OCT figure
As denoising program, the step of OCT image denoising method for generating network is fought based on annular is realized.
Processor can be central processing unit, microprocessor or other data processing chips etc., can be in run memory
Storage program, for example, in the present invention based on annular fight generate network OCT image denoise program.
The electronic device can also include display, and display is referred to as display screen or display unit.Some
Display can be light-emitting diode display, liquid crystal display, touch-control liquid crystal display and Organic Light Emitting Diode in embodiment
(Organic Light-Emitting Diode, OLED) touches device etc..Display is used to show and handle in an electronic
Information and for showing visual working interface.
The electronic device can also include user interface, and user interface may include input unit (such as keyboard), language
Sound output device (such as sound equipment, earphone) etc..
In other embodiments, the OCT image denoising program for fighting generation network based on annular can also be divided into one
A or multiple modules, one or more module is stored in memory, and is executed by processor, to complete the present invention.
The so-called module of the present invention is the series of computation machine program instruction section for referring to complete specific function.Fig. 2 is base in the present invention
The module diagram for generating the OCT image denoising program of network is fought in annular, as shown in Fig. 2, the OCT image denoises program
It can be divided into: obtain module 1, denoising module 2 and output module 3.The functions or operations step that above-mentioned module is realized is equal
It is similar as above, such as wherein:
Module 1 is obtained, OCT image to be denoised is obtained;
Module 2 is denoised, the annular confrontation that the OCT image input to be denoised is obtained by training is generated into network mould
Type carries out denoising to the OCT image;
Output module 3 generates the OCT image of network model output denoising by the annular confrontation.
It includes generating network model and differentiating network mould that the confrontation of annular used in the denoising module, which generates network model,
Type, wherein generate network model and be used to noise picture being transformed into clear picture, the input for generating network model is containing noise
OCT image, output is image after denoising as similar as possible with the OCT image of not Noise, differentiates network model for sentencing
It is disconnected to convert whether obtained clear picture is true clear image data, differentiate that the input of network model is an image pair,
Image is to including noise-containing noise OCT image and by generating the clear OCT image after network denoising, not Noise
OCT image and pass through generate network denoising after clear OCT image.Network model is generated to cheat as much as possible in the training process
Differentiate network model, and differentiates network model and correct differentiation is made to image as far as possible.The generation network model includes two
A generator, respectively the first generator and the second generator, the differentiation network model include two arbiters, respectively the
One arbiter and the second arbiter, the corresponding arbiter of each generator, wherein first generator and second life
It grows up to be a useful person and is made of several convolutional layers with step-length and warp lamination;First arbiter and second arbiter are by several
Convolutional layer with step-length and full articulamentum composition.
It is to be appreciated that " first " and " second " is only used for distinguishing two generators or arbiter, without generation
Table priority or other meanings.
In order to improve the performance that annular fights generation network model, need to generate network model progress sufficiently to annular confrontation
Training so that the weight parameter in model is optimal value, Optimized model performance.Preferably, the electronic device further includes
Sample set constructs module, constructs training sample set, including noise image sample and clear image sample;Training module is based on institute
It states training sample set to be trained the annular confrontation generation network model, when training annular confrontation generates network, no longer need
Want the one-to-one correspondence of noise image sample Yu clear image sample.
In an alternative embodiment of the invention, the sample set building module includes: sample database acquiring unit, is obtained
OCT image sample database is selected from image pattern library for example, including about 10000 OCT image samples in image pattern library
Image pattern constructs training sample set;First selecting unit, selection criteria OCT image carry out artificial selection, in OCT image sample
High quality is selected except this library, clearly OCT image is as standard picture;First Y-PSNR acquiring unit, by the mark
Quasi- OCT image traverses the sample in the OCT image sample database, seeks each OCT image sample in the OCT image sample database
Y-PSNR index (Peak Signal to Noise Ratio, PSNR), wherein PSNR index is higher, corresponding OCT
Image pattern is more clear, noise is fewer, and PSNR index is lower, and the noise of corresponding OCT image sample is more serious;First sequence is single
Member sorts the PSNR index from high to low;First construction unit selects corresponding PSNR index to sort forward default
Several OCT image samples select the predetermined number OCT image of corresponding PSNR index sequence rearward as clear image sample
Sample is as noise image sample.When selection, it can be selected according to setting ratio, required for being concentrated according to training sample
Noise image sample number and clear image sample size set selection percentage, for example, can set selection percentage is 3%, it will
After the sequence of PSNR index, according to selecting 3% image pattern from high to low as clear image sample, according to selecting from low to high
3% image pattern is as noise image sample.
Preferably, the Y-PSNR acquiring unit seeks the PSNR index of OCT image sample by following formula:
In formula, PSNR indicates the Y-PSNR of OCT image sample;MSE indicates standard OCT image and current OCT image
Mean square error between sample;X indicates the bit number of each sampled value.
Wherein, mean square deviation MSE is obtained by following formula:
In formula, I is standard OCT image, and K is current OCT image sample, and m*n is the size of image, standard OCT image and
Current OCT image size is identical.
Preferably, sample set building module includes: sample database acquiring unit, obtains OCT image sample database;Second selection is single
Member selects multiple standard OCT images;Multiple described standard OCT images are successively traversed institute by the second Y-PSNR acquiring unit
The sample in OCT image sample database is stated, each OCT image sample in the OCT image sample database is sought respectively and corresponds to every
The PSNR index of standard OCT image;One OCT image sample is corresponded respectively to multiple standard OCT images by mean value acquiring unit
Obtained multiple PSNR indexes seek mean value, obtain the PSNR Mean value of index of one OCT image sample;Second sequence is single
Member sorts the PSNR Mean value of index from high to low;Second construction unit selects corresponding PSNR Mean value of index sequence forward
Predetermined number OCT image sample as clear image sample, select corresponding PSNR Mean value of index sequence rearward default
A several OCT image samples are as noise image sample.For example, selecting 16 high quality, clearly OCT image is as standard drawing
When picture, 16 standard pictures are successively traversed into the data in the OCT image sample database, 16 standard pictures respectively with processing
OCT image sample acquires a PSNR index, and 16 PSNR indexes are sought mean value, represent currently evaluated OCT image sample
This final PSNR index, constantly repeats this step, until all OCT image sample standard deviations are evaluated, obtained final PSNR
Index is higher, illustrates that corresponding OCT image sample is more clear, noise is fewer, final PSNR index is lower, corresponding OCT image
The noise of sample is more serious.For PSNR index seek and the first construction unit and the second construction unit final sample
Selection mode is similar with foregoing embodiments, and details are not described herein.
In the present invention, the clear image sample size of selection is identical as noise image sample size, in OCT image sample database
In occupy same ratio, for example, in OCT image sample database include about 10000 OCT images when, select 3% data as clearly
Sample (about 300) equally selects 3% data as noise sample (about 300).
In an alternative embodiment of the invention, the training module includes:
Noise image sample is inputted first generator by the first generation unit, exports corresponding first construction image;
The first construction image is inputted second generator, exports corresponding second construction by the second generation unit
Image;
The noise image sample and the second construction image are inputted second arbiter by the first judgement unit,
The probability value that the second construction image is determined as true noise image is exported by second arbiter;
The first construction image and clear image sample are inputted first arbiter, passed through by the second judgement unit
First arbiter exports the probability value that the first construction image is determined as true clear image;
Judging unit, judges whether the differentiation network model restrains, will be described if the differentiation network model is restrained
Differentiate denoising image of the first construction image of network model output as the noise image sample;If the differentiation network mould
Type does not restrain, then updates the parameter for generating network model until the differentiation network model is restrained.Optimized using Adam and is learned
Learning method uses learning rate for 0.0002, and crowd size batch size is 1, and the training of 200 wheels is carried out to it.
In an alternative embodiment of the invention, the judging unit includes: construction subelement, constructs loss function mould
Type;Judgment sub-unit, the loss function model based on construction judge whether the differentiation network model restrains.Further, institute
The loss function model for stating construction subelement construction includes: generational loss function model, obtains the loss of generator;Differentiate loss
Function model obtains the loss of arbiter;Circulation loss function model obtains circulation consistency loss, passes through the circulation of introducing
Consistency loss realize the data in an image area are uniformly mapped in another image area, avoid mapping relations with
The generation of machine phenomenon (for example, the data in an image area are all mapped to a certain picture in another image area).
The loss function that the annular confrontation generates network model is shown below:
L(GA2B,GB2A,DA,DB)
=LGAN(GA2B,DB,A,B)
+LGAN(GB2A,DA,B,A)+λLcyc(GA2B,GB2A,A,B)
Wherein, L indicates penalty values, and GAN indicates that confrontation generates network, and G indicates that generator, D indicate arbiter, and A expression is made an uproar
Sound spectrogram image field, B indicate clear image domain, and A2B indicates that image area A is mapped to image area B, and B2A indicates that image area B is mapped to image
Domain A, λ can use empirical value, λ=10, LcycIndicate circulation consistency loss.
Wherein, each penalty values are obtained by following formula respectively:
Wherein,It indicates that a is derived from and is distributed as paNoise pattern image field,It indicates that b is derived from and is distributed as pbIt is clear
Image area.
In one embodiment of the present of invention, the judging unit can also judge the differentiation network mould by following manner
Whether type restrains, comprising: it is respectively default whether judgement is all larger than by the probability value that the first arbiter or the second arbiter export
Probability threshold value differentiation network model convergence is determined, if at least one arbiter if being all larger than preset probability threshold value
The probability value of output is less than corresponding predetermined probabilities threshold value, then determines that the differentiation network model is not restrained.
In one embodiment of the present of invention, computer readable storage medium, which can be, any includes or storage program or instruction
Tangible medium, program therein can be performed, and pass through the corresponding function of the relevant hardware realization of the program instruction of storage.Example
Such as, computer readable storage medium can be computer disk, hard disk, random access memory, read-only memory etc..The present invention
It is not limited to this, can be in a manner of non-transitory store instruction or software and any associated data files or data structure simultaneously
And processor is provided to so that processor executes any device of program therein or instruction.The computer-readable storage medium
It include that the OCT image denoising program for generating network is fought based on annular in matter, it is described that the OCT figure for generating network is fought based on annular
When being executed by processor as denoising program, the following OCT image denoising method fought based on annular and generate network is realized:
Obtain OCT image to be denoised;
The annular confrontation that the OCT image input to be denoised is obtained by training is generated into network model;
The OCT image of network model output denoising is generated by the annular confrontation.
The specific embodiment of the computer readable storage medium of the present invention and above-mentioned fought based on annular generate network
OCT image denoising method, the specific embodiment of electronic device are roughly the same, and details are not described herein.
It should be noted that, in this document, the terms "include", "comprise" or its any other variant are intended to non-row
His property includes, so that the process, device, article or the method that include a series of elements not only include those elements, and
And further include other elements that are not explicitly listed, or further include for this process, device, article or method institute it is intrinsic
Element.In the absence of more restrictions, the element limited by sentence "including a ...", it is not excluded that including being somebody's turn to do
There is also other identical elements in the process, device of element, article or method.
The serial number of the above embodiments of the invention is only for description, does not represent the advantages or disadvantages of the embodiments.Pass through above embodiment party
The description of formula, it is required general that those skilled in the art can be understood that above-described embodiment method can add by software
The mode of hardware platform is realized, naturally it is also possible to which by hardware, but in many cases, the former is more preferably embodiment.It is based on
Such understanding, substantially the part that contributes to existing technology can be with software product in other words for technical solution of the present invention
Form embody, which is stored in a storage medium (such as ROM/RAM, magnetic disk, light as described above
Disk) in, including some instructions use is so that a terminal device (can be mobile phone, computer, server or the network equipment
Deng) execute method described in each embodiment of the present invention.
The above is only a preferred embodiment of the present invention, is not intended to limit the scope of the invention, all to utilize this hair
Equivalent structure or equivalent flow shift made by bright specification and accompanying drawing content is applied directly or indirectly in other relevant skills
Art field, is included within the scope of the present invention.
Claims (10)
1. a kind of fight the OCT image denoising method for generating network based on annular, it is applied to electronic device, which is characterized in that packet
It includes:
Obtain OCT image to be denoised;
The annular confrontation that the OCT image input to be denoised is obtained by training is generated into network model;
The OCT image of network model output denoising is generated by the annular confrontation;
Wherein, the annular confrontation generates network model including generating network model and differentiating network model, the generation network
Model includes two generators, respectively the first generator and the second generator, and the differentiation network model includes two differentiations
Device, respectively the first arbiter and the second arbiter, the corresponding arbiter of each generator, wherein first generator
It is made of with second generator several convolutional layers with step-length and warp lamination, first arbiter and described second
Arbiter is made of several convolutional layers with step-length and full articulamentum.
2. according to claim 1 fight the OCT image denoising method for generating network based on annular, which is characterized in that institute
The method of stating further includes following steps:
Construct training sample set, including noise image sample and clear image sample;
Network model is generated to the annular confrontation based on the training sample set to be trained.
3. according to claim 2 fight the OCT image denoising method for generating network based on annular, which is characterized in that institute
Stating the step of constructing training sample set includes:
Obtain OCT image sample database;
Selection criteria OCT image;
The standard OCT image is traversed into the sample in the OCT image sample database, is sought each in the OCT image sample database
The Y-PSNR index of a OCT image sample;
The Y-PSNR index is sorted from high to low;
The predetermined number OCT image sample for selecting corresponding Y-PSNR index to sort forward as clear image sample,
Select the predetermined number OCT image sample of corresponding Y-PSNR index as noise image sample.
4. according to claim 2 fight the OCT image denoising method for generating network based on annular, which is characterized in that institute
Stating the step of constructing training sample set includes:
Obtain OCT image sample database;
Select multiple standard OCT images;
Multiple described standard OCT images are successively traversed into the sample in the OCT image sample database, seek the OCT figure respectively
As OCT image sample each in sample database corresponds to the Y-PSNR index of every standard OCT image;
One OCT image sample is corresponded respectively to multiple Y-PSNR indexes that multiple standard OCT images obtain to seek
Value, obtains the Y-PSNR Mean value of index of one OCT image sample;
The Y-PSNR Mean value of index is sorted from high to low;
The predetermined number OCT image sample for selecting corresponding Y-PSNR Mean value of index to sort forward is as clear image sample
This, selects the predetermined number OCT image sample of corresponding Y-PSNR Mean value of index sequence rearward decent as noise pattern
This.
5. according to claim 3 or 4 fight the OCT image denoising method for generating network based on annular, which is characterized in that
The Y-PSNR index of OCT image sample is sought by following formula:
In formula, PSNR indicates the Y-PSNR of OCT image sample;MSE indicates standard OCT image and current OCT image sample
Between mean square error;X indicates the bit number of each sampled value.
6. according to claim 2 fight the OCT image denoising method for generating network based on annular, which is characterized in that base
Network model is generated to the annular confrontation in the training sample set to be trained, comprising:
The noise image sample is inputted into first generator, exports corresponding first construction image;
The first construction image is inputted into second generator, exports corresponding second construction image;
The noise image sample and the second construction image are inputted into second arbiter, pass through second arbiter
Export the probability value that the second construction image is determined as true noise image;
The first construction image and clear image sample are inputted into first arbiter, exported by first arbiter
The first construction image is determined as the probability value of true clear image;
Judge whether the differentiation network model restrains, if the differentiation network model is restrained, by the differentiation network model
Denoising image of the first construction image of output as the noise image sample;If the differentiation network model is not restrained,
The parameter for generating network model is updated until the differentiation network model is restrained.
7. according to claim 6 fight the OCT image denoising method for generating network based on annular, which is characterized in that sentence
Whether the differentiation network model that breaks restrains, comprising:
Loss function model is constructed, is shown below:
L(GA2B,GB2A,DA,DB)=LGAN(GA2B,DB,A,B)+LGAN(GB2A,DA,B,A)+λLcyc(GA2B,GB2A,A,B)
Wherein, L indicates penalty values, and GAN indicates that confrontation generates network, and G indicates that generator, D indicate that arbiter, A indicate noise pattern
Image field, B indicate clear image domain, and A2B indicates that image area A is mapped to image area B, and B2A indicates that image area B is mapped to image area A,
λ takes empirical value, LcycIndicate circulation consistency loss;
Loss function model based on construction judges whether the differentiation network model restrains.
8. according to claim 6 fight the OCT image denoising method for generating network based on annular, which is characterized in that sentence
Whether the differentiation network model that breaks restrains, comprising:
Whether judgement is all larger than respective preset probability threshold value by the probability value that the first arbiter or the second arbiter export, if
It is all larger than preset probability threshold value, then determines the differentiation network model convergence, if the probability value of at least one arbiter output
Less than corresponding predetermined probabilities threshold value, then determine that the differentiation network model is not restrained.
9. a kind of electronic device, which is characterized in that the electronic device includes:
Processor;
Memory includes fighting to generate the OCT image of network and denoise program, the OCT image based on annular in the memory
Denoising program realizes the step such as OCT image denoising method described in any item of the claim 1 to 8 when being executed by the processor
Suddenly.
10. a kind of computer readable storage medium, which is characterized in that include based on annular in the computer readable storage medium
The OCT image that confrontation generates network denoises program and realizes when the OCT image denoising program is executed by processor as right is wanted
Described in asking any one of 1 to 8 the step of OCT image denoising method.
Priority Applications (2)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201910515611.2A CN110390647A (en) | 2019-06-14 | 2019-06-14 | The OCT image denoising method and device for generating network are fought based on annular |
PCT/CN2019/117815 WO2020248492A1 (en) | 2019-06-14 | 2019-11-13 | Method and device for denoising oct image based on cyclic generative adversarial network |
Applications Claiming Priority (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201910515611.2A CN110390647A (en) | 2019-06-14 | 2019-06-14 | The OCT image denoising method and device for generating network are fought based on annular |
Publications (1)
Publication Number | Publication Date |
---|---|
CN110390647A true CN110390647A (en) | 2019-10-29 |
Family
ID=68285447
Family Applications (1)
Application Number | Title | Priority Date | Filing Date |
---|---|---|---|
CN201910515611.2A Pending CN110390647A (en) | 2019-06-14 | 2019-06-14 | The OCT image denoising method and device for generating network are fought based on annular |
Country Status (2)
Country | Link |
---|---|
CN (1) | CN110390647A (en) |
WO (1) | WO2020248492A1 (en) |
Cited By (11)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN110428385A (en) * | 2019-08-09 | 2019-11-08 | 南京工业大学 | A kind of SD-OCT denoising method based on unsupervised confrontation neural network |
CN110840445A (en) * | 2019-12-12 | 2020-02-28 | 山东省计算中心(国家超级计算济南中心) | Automatic noise reduction method for dynamic electrocardiosignals |
CN110852424A (en) * | 2019-11-15 | 2020-02-28 | 广东工业大学 | Processing method and device for confrontation generation network |
CN111899185A (en) * | 2020-06-18 | 2020-11-06 | 深圳先进技术研究院 | Training method and device of image noise reduction model, electronic equipment and storage medium |
WO2020248492A1 (en) * | 2019-06-14 | 2020-12-17 | 平安科技(深圳)有限公司 | Method and device for denoising oct image based on cyclic generative adversarial network |
CN113160160A (en) * | 2021-04-12 | 2021-07-23 | 哈尔滨医科大学 | Calcification identification method based on OCT light attenuation image |
CN113191960A (en) * | 2021-03-18 | 2021-07-30 | 北京理工大学 | Optical coherence tomography image denoising model training method, denoising method and denoising device |
CN113516238A (en) * | 2020-11-25 | 2021-10-19 | 阿里巴巴集团控股有限公司 | Model training method, denoising method, model, device and storage medium |
WO2021253316A1 (en) * | 2020-06-18 | 2021-12-23 | 深圳先进技术研究院 | Method and apparatus for training image noise reduction model, electronic device, and storage medium |
CN114841878A (en) * | 2022-04-27 | 2022-08-02 | 广东博迈医疗科技股份有限公司 | Speckle denoising method and device for optical coherence tomography image and electronic equipment |
CN117710373A (en) * | 2024-02-05 | 2024-03-15 | 中国科学院宁波材料技术与工程研究所 | Anti-interference OCTA retinal vessel extraction method and device |
Families Citing this family (9)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN112699885A (en) * | 2020-12-21 | 2021-04-23 | 杭州反重力智能科技有限公司 | Semantic segmentation training data augmentation method and system based on antagonism generation network GAN |
TR202022517A1 (en) * | 2020-12-30 | 2022-07-21 | Istanbul Medipol Ueniversitesi | A METHOD FOR MODEL-INDEPENDENT SOLUTION OF REVERSE PROBLEMS IN IMAGE/VIDEO PROCESSING WITH DEEP LEARNING |
CN113688912B (en) * | 2021-08-26 | 2024-01-05 | 平安国际智慧城市科技股份有限公司 | Method, device, equipment and medium for generating countermeasure sample based on artificial intelligence |
CN114529480B (en) * | 2022-03-02 | 2024-06-18 | 北京大学 | Unsupervised one-pot multi-frame image denoising method based on random pairing |
CN114947792B (en) * | 2022-05-19 | 2024-05-03 | 北京航空航天大学 | Video-based physiological signal measurement and enhancement method |
CN114998124B (en) * | 2022-05-23 | 2024-06-18 | 北京航空航天大学 | Image sharpening processing method for target detection |
CN115546848B (en) * | 2022-10-26 | 2024-02-02 | 南京航空航天大学 | Challenge generation network training method, cross-equipment palmprint recognition method and system |
CN116250844B (en) * | 2023-03-03 | 2024-04-26 | 山东大学 | Electrocardiosignal noise reduction optimization method and system based on condition generation countermeasure network |
CN117934323B (en) * | 2024-03-22 | 2024-06-07 | 摩尔线程智能科技(北京)有限责任公司 | Image generation method, device, electronic equipment and storage medium |
Citations (3)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN108198154A (en) * | 2018-03-19 | 2018-06-22 | 中山大学 | Image de-noising method, device, equipment and storage medium |
CN109345469A (en) * | 2018-09-07 | 2019-02-15 | 苏州大学 | It is a kind of that speckle denoising method in the OCT image of confrontation network is generated based on condition |
CN109493303A (en) * | 2018-05-30 | 2019-03-19 | 湘潭大学 | A kind of image defogging method based on generation confrontation network |
Family Cites Families (1)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN110390647A (en) * | 2019-06-14 | 2019-10-29 | 平安科技(深圳)有限公司 | The OCT image denoising method and device for generating network are fought based on annular |
-
2019
- 2019-06-14 CN CN201910515611.2A patent/CN110390647A/en active Pending
- 2019-11-13 WO PCT/CN2019/117815 patent/WO2020248492A1/en active Application Filing
Patent Citations (3)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN108198154A (en) * | 2018-03-19 | 2018-06-22 | 中山大学 | Image de-noising method, device, equipment and storage medium |
CN109493303A (en) * | 2018-05-30 | 2019-03-19 | 湘潭大学 | A kind of image defogging method based on generation confrontation network |
CN109345469A (en) * | 2018-09-07 | 2019-02-15 | 苏州大学 | It is a kind of that speckle denoising method in the OCT image of confrontation network is generated based on condition |
Cited By (13)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
WO2020248492A1 (en) * | 2019-06-14 | 2020-12-17 | 平安科技(深圳)有限公司 | Method and device for denoising oct image based on cyclic generative adversarial network |
CN110428385A (en) * | 2019-08-09 | 2019-11-08 | 南京工业大学 | A kind of SD-OCT denoising method based on unsupervised confrontation neural network |
CN110852424A (en) * | 2019-11-15 | 2020-02-28 | 广东工业大学 | Processing method and device for confrontation generation network |
CN110840445B (en) * | 2019-12-12 | 2021-07-20 | 山东省计算中心(国家超级计算济南中心) | Automatic noise reduction method for dynamic electrocardiosignals |
CN110840445A (en) * | 2019-12-12 | 2020-02-28 | 山东省计算中心(国家超级计算济南中心) | Automatic noise reduction method for dynamic electrocardiosignals |
WO2021253316A1 (en) * | 2020-06-18 | 2021-12-23 | 深圳先进技术研究院 | Method and apparatus for training image noise reduction model, electronic device, and storage medium |
CN111899185A (en) * | 2020-06-18 | 2020-11-06 | 深圳先进技术研究院 | Training method and device of image noise reduction model, electronic equipment and storage medium |
CN113516238A (en) * | 2020-11-25 | 2021-10-19 | 阿里巴巴集团控股有限公司 | Model training method, denoising method, model, device and storage medium |
CN113191960A (en) * | 2021-03-18 | 2021-07-30 | 北京理工大学 | Optical coherence tomography image denoising model training method, denoising method and denoising device |
CN113160160A (en) * | 2021-04-12 | 2021-07-23 | 哈尔滨医科大学 | Calcification identification method based on OCT light attenuation image |
CN114841878A (en) * | 2022-04-27 | 2022-08-02 | 广东博迈医疗科技股份有限公司 | Speckle denoising method and device for optical coherence tomography image and electronic equipment |
CN117710373A (en) * | 2024-02-05 | 2024-03-15 | 中国科学院宁波材料技术与工程研究所 | Anti-interference OCTA retinal vessel extraction method and device |
CN117710373B (en) * | 2024-02-05 | 2024-06-04 | 中国科学院宁波材料技术与工程研究所 | Anti-interference OCTA retinal vessel extraction method and device |
Also Published As
Publication number | Publication date |
---|---|
WO2020248492A1 (en) | 2020-12-17 |
Similar Documents
Publication | Publication Date | Title |
---|---|---|
CN110390647A (en) | The OCT image denoising method and device for generating network are fought based on annular | |
CN108460356A (en) | A kind of facial image automated processing system based on monitoring system | |
CN108805860A (en) | The generation method of method for analyzing image, device, program and deep learning algorithm | |
CN110443364A (en) | A kind of deep neural network multitask hyperparameter optimization method and device | |
CN109147010A (en) | Band attribute Face image synthesis method, apparatus, system and readable storage medium storing program for executing | |
CN111340819A (en) | Image segmentation method, device and storage medium | |
CN108765319A (en) | A kind of image de-noising method based on generation confrontation network | |
CN107977629A (en) | A kind of facial image aging synthetic method of feature based separation confrontation network | |
CN110363226A (en) | Ophthalmology disease classifying identification method, device and medium based on random forest | |
CN108198147A (en) | A kind of method based on the multi-source image fusion denoising for differentiating dictionary learning | |
CN110070935A (en) | Medical image synthetic method, classification method and device based on confrontation neural network | |
CN108053030A (en) | A kind of transfer learning method and system of Opening field | |
CN110223292A (en) | Image evaluation method, device and computer readable storage medium | |
CN110110600A (en) | The recognition methods of eye OCT image lesion, device and storage medium | |
CN113505855B (en) | Training method for challenge model | |
CN111026267B (en) | VR brain electrical idea control interface system | |
CN110363229A (en) | A kind of characteristics of human body's parameter selection method combined based on improvement RReliefF and mRMR | |
Shan et al. | Residual learning of deep convolutional neural networks for image denoising | |
CN112052874A (en) | Physiological data classification method and system based on generation countermeasure network | |
CN113421243B (en) | Method and device for detecting fundus image type based on knowledge distillation network | |
Kunanbayev et al. | Data augmentation for p300-based brain-computer interfaces using generative adversarial networks | |
CN110176050B (en) | Aesthetic optimization method for text generated image | |
Liu et al. | Cells image generation method based on VAE-SGAN | |
CN107944366A (en) | A kind of finger vein identification method and device based on attribute study | |
CN112101438B (en) | Left-right eye classification method, device, server and storage medium |
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 |