CN109754376A - Image de-noising method and device - Google Patents
Image de-noising method and device Download PDFInfo
- Publication number
- CN109754376A CN109754376A CN201811629202.7A CN201811629202A CN109754376A CN 109754376 A CN109754376 A CN 109754376A CN 201811629202 A CN201811629202 A CN 201811629202A CN 109754376 A CN109754376 A CN 109754376A
- Authority
- CN
- China
- Prior art keywords
- noise
- image
- value
- denoising
- sample
- 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.)
- Granted
Links
Landscapes
- Image Processing (AREA)
Abstract
The embodiment of the present application provides a kind of image de-noising method and device, implement principle are as follows: noise image sample set is input in convolutional neural networks and is trained, the denoising image pattern collection that output training obtains, corresponding mean square error loss function value and maximum a posteriori loss function value are calculated separately according to noise-free picture sample set and denoising image pattern collection, and total losses functional value is obtained according to mean square error loss function value and maximum a posteriori loss function value, finally based on the total losses functional value repetitive exercise convolutional neural networks being calculated, when meeting training termination condition, the image denoising model that output training obtains, and denoising image is treated according to image denoising model and carries out denoising.As a result, by combining mean square error loss function value and maximum a posteriori loss function value repetitive exercise convolutional neural networks, the removal effect of real noise can be greatly promoted.
Description
Technical field
This application involves field of computer technology, in particular to a kind of image de-noising method and device.
Background technique
Currently, image denoising (Image Denoising) is all that image restores the very important project in field all the time,
No matter learning value and landing application in all have very high researching value.But current image de-noising method is all universal
It establishes on the Gaussian noise model of additivity, however inventor studies the real noise model in discovery real image not such as
This, so current Image denoising algorithm often cannot all remove the noise on true picture well.
Summary of the invention
In order to overcome above-mentioned deficiency in the prior art, the application's is designed to provide a kind of image de-noising method and dress
It sets, to solve or improve the above problem.
To achieve the goals above, the embodiment of the present application the technical solution adopted is as follows:
In a first aspect, the embodiment of the present application provides a kind of image de-noising method, it is applied to electronic equipment, the method packet
It includes:
Obtain training sample set, the training sample set include noise image sample set and with the noise image sample
Collect corresponding noise-free picture sample set；
The noise image sample set is input in convolutional neural networks and is trained, the denoising figure that output training obtains
As sample set；
Corresponding mean square error loss is calculated separately according to the noise-free picture sample set and the denoising image pattern collection
Functional value and maximum a posteriori loss function value, and be worth according to the mean square error loss function value and maximum a posteriori loss function
To total losses functional value；
Based on convolutional neural networks described in the total losses functional value repetitive exercise being calculated, item is terminated until meeting training
When part, the image denoising model that output training obtains, and denoising image is treated according to described image denoising model and is carried out at denoising
Reason.
In a kind of possible embodiment, the step of the acquisition training sample set, comprising:
Multiframe noise image sample that image capture device is continuously shot based on the image acquisition parameter of configuration is obtained with structure
Noise image sample set is built, described image acquisition parameter includes image exposure parameter and filming frequency parameter；
For every frame noise image sample in the noise image sample set, to each of the frame noise image sample
The pixel value of pixel is ranked up, and is greater than pixel value less than the first presetted pixel value and pixel value according to ranking results
The extraordinary image vegetarian refreshments of second presetted pixel value is rejected, the noise image sample after obtaining rejecting abnormalities pixel；
Calculate the average value of each pixel of the noise image sample after rejecting each frame extraordinary image vegetarian refreshments；
According to the relationship between preset noise image sample and corresponding noise-free picture sample and reject each frame exception
The average value of each pixel of noise image sample after pixel obtains nothing corresponding with the noise image sample set and makes an uproar
Image pattern collection.
In a kind of possible embodiment, there is such as ShiShimonoseki with corresponding noise-free picture sample in the noise image sample
System:
xn=clip (yn+εn(yn))
εn(yn)~N (0, σn 2(yn))
Wherein, xnFor noise image sample, ynFor corresponding noise-free picture sample, εn(yn) it is ynNoise profile function,
σn 2(yn) it is the variance that noise profile function is obeyed, wherein σn 2(yn)=β1 nyn+β2 n, β1And β2Equipment is acquired with described image
Imaging sensor parameter and sensitivity numerical value it is related.
In a kind of possible embodiment, it is described by the noise image sample set be input in convolutional neural networks into
The step of row training, the denoising image pattern collection that output training obtains, comprising:
The noise image sample set is input in convolutional neural networks, each layer in the convolutional neural networks is passed through
Convolutional layer successively extracts the image feature information of each noise image sample in the noise image sample set；
The image feature information for each noise image sample that every layer of convolutional layer extracts is sent to by articulamentum
In the symmetrically arranged warp lamination of this layer of convolutional layer, by the warp lamination according to the image of each noise image sample
Characteristic information generates the denoising image pattern collection after deconvolution.
It is described according to the noise-free picture sample set and the denoising image pattern collection in a kind of possible embodiment
The step of calculating separately corresponding mean square error loss function value and maximum a posteriori loss function value, comprising:
Mean square error loss function value is calculated according to the noise-free picture sample set and the denoising image pattern collection, specifically
Are as follows:
Calculate separately in noise-free picture sample set every frame noise-free picture sample and the denoising image pattern concentrate with it is every
The square value of each pixel difference between the corresponding denoising image pattern of frame noise-free picture sample, and it is each by what is be calculated
The mean value of the square value of pixel difference is as the mean square error loss function value；And
Maximum a posteriori loss function value is calculated according to the noise-free picture sample set and the denoising image pattern collection, specifically
Are as follows:
Add according to each noise-free picture sample in the noise-free picture sample set and in each noise-free picture sample
Enter the poisson noise image after poisson noise, each noise-free picture sample calculated in the noise-free picture sample set has pool
The posterior probability of loose noise；
According in each noise-free picture sample be added poisson noise after poisson noise image, in each noise-free picture
The gaussian noise image after Gaussian noise is added in sample, calculating each poisson noise image, there are the posteriority of Gaussian noise is general
Rate；
According to each noise-free picture sample, there are the posterior probability of poisson noise and each poisson noise image there is height
The posterior probability of this noise, each noise-free picture sample is calculated, and there are the maximum a posteriori probabilities of Poisson Gaussian noise；
Using each noise-free picture sample, there are the maximum a posteriori probabilities of Poisson Gaussian noise as the negative logarithm of independent variable for calculating
Functional value, and using the negative logarithmic function value as the maximum a posteriori loss function value.
It is described that letter is lost according to the mean square error loss function value and maximum a posteriori in a kind of possible embodiment
Numerical value obtains the step of total losses functional value, comprising:
According to default weight proportion calculate separately the first weight loss function value of the mean square error loss function value with
And the second weight loss function value of the maximum a posteriori loss function value；
The total losses function is obtained according to the first weight loss function value and the second weight loss function value
Value.
Second aspect, the embodiment of the present application also provide a kind of image denoising device, are applied to electronic equipment, described device packet
It includes:
Obtain module, for obtaining training sample set, the training sample set include noise image sample set and with institute
State the corresponding noise-free picture sample set of noise image sample set；
Training module is trained for the noise image sample set to be input in convolutional neural networks, output instruction
The denoising image pattern collection got；
Computing module, it is corresponding for being calculated separately according to the noise-free picture sample set and the denoising image pattern collection
Mean square error loss function value and maximum a posteriori loss function value, and according to the mean square error loss function value and maximum a posteriori
Loss function value obtains total losses functional value；
Denoising module, for based on convolutional neural networks described in the total losses functional value repetitive exercise being calculated,
When meeting training termination condition, the image denoising model that output training obtains, and treated according to described image denoising model
It denoises image and carries out denoising.
The third aspect, the embodiment of the present application also provide a kind of readable storage medium storing program for executing, are stored thereon with computer program, described
Computer program, which is performed, realizes above-mentioned image de-noising method.
In terms of existing technologies, the application has the advantages that
The embodiment of the present application provides a kind of image de-noising method and device, implements principle are as follows: by noise image sample
Collection, which is input in convolutional neural networks, to be trained, the denoising image pattern collection that output training obtains, according to noise-free picture sample
Collection and denoising image pattern collection calculate separately corresponding mean square error loss function value and maximum a posteriori loss function value, and according to
Mean square error loss function value and maximum a posteriori loss function value obtain total losses functional value, finally based on the total damage being calculated
Functional value repetitive exercise convolutional neural networks are lost, when meeting training termination condition, obtained image denoising mould is trained in output
Type, and denoising image is treated according to image denoising model and carries out denoising.As a result, by combining mean square error loss function value
With maximum a posteriori loss function value repetitive exercise convolutional neural networks, the removal effect of real noise can be greatly promoted.
Detailed description of the invention
Technical solution in ord to more clearly illustrate embodiments of the present application, below will be to needed in the embodiment attached
Figure is briefly described, it should be understood that the following drawings illustrates only some embodiments of the application, therefore is not construed as pair
The restriction of range for those of ordinary skill in the art without creative efforts, can also be according to this
A little attached drawings obtain other relevant attached drawings.
Fig. 1 is the flow diagram of image de-noising method provided by the embodiments of the present application；
Fig. 2 is the schematic network structure of convolutional neural networks provided by the embodiments of the present application；
Fig. 3 is the functional block diagram of image denoising device provided by the embodiments of the present application；
Fig. 4 is the structural representation frame of the electronic equipment provided by the embodiments of the present application for realizing above-mentioned image de-noising method
Figure.
Specific embodiment
Below in conjunction with the attached drawing in the embodiment of the present application, technical solutions in the embodiments of the present application carries out clear, complete
Site preparation description, it is clear that described embodiment is some embodiments of the present application, instead of all the embodiments.Usually herein
The component of the embodiment of the present application described and illustrated in place's attached drawing can be arranged and be designed with a variety of different configurations.
Therefore, the detailed description of the embodiments herein provided in the accompanying drawings is not intended to limit below claimed
Scope of the present application, but be merely representative of the selected embodiment of the application.Based on the embodiment in the application, this field is common
Technical staff's all other embodiment obtained without creative labor belongs to the application protection
Range.
It should also be noted that similar label and letter indicate similar terms in following attached drawing, therefore, once a certain Xiang Yi
It is defined in a attached drawing, does not then need that it is further defined and explained in subsequent attached drawing.
The technical issues of knowing such as aforementioned background art, present inventor have found in the course of the research, with depth
The sudden emergence of (Deep Learning, DL) in the fields such as Classification and Identification is practised, image restores field also quickly using DL as weight
The research tool wanted.Due to being supervised learning, while using true picture noise and corresponding noise-free picture (Ground
Truth, GT) it is used as training sample set, such DL Denoising Algorithm also can be really used in real life.
However inventor has found current DL Denoising Algorithm in the course of the research that there are the following problems:
First, when acquiring training sample, noise image sample is easy to collect, but its corresponding noise-free picture is extremely difficult obtains
, if traditional Denoising Algorithm is used to come after denoising to noise image sample as noise-free picture, it is clear that the noise-free picture is not
Accurately, the image denoising ability upper limit obtained finally by deep learning is also only traditional Denoising Algorithm.
Second, image denoising and image classification identification are different, and deep learning algorithm is in the explanatory of image denoising at present
It is insufficient, it is difficult to determine that just centainly most suitable image denoising often occurs deep learning algorithm during hands-on at present
Phenomena such as training is not easy to restrain, and image detail loss is serious after training.
Third, since current all image de-noising methods all assume that true picture noise is Gaussian Profile, so
Loss function loss function of the mean square error loss function value MSE Loss as image denoising generally can be all used, but
It is actual true picture noise is not Gaussian Profile, the deep learning algorithm that training obtains in this way can go out
Now as described in aforementioned background art, the problem of noise on true picture cannot be removed well.
Discovery based on above-mentioned technical problem, present inventor propose that following technical proposals are above-mentioned to solve or improve
Problem.It is inventor by practice and careful it should be noted that defect present in the above scheme in the prior art
It is being obtained after research as a result, therefore, the discovery procedure of the above problem and hereinafter the embodiment of the present application institute regarding to the issue above
The solution of proposition all should be the contribution that inventor makes the application during innovation and creation.
Referring to Fig. 1, being a kind of flow diagram of image de-noising method provided by the embodiments of the present application, should illustrate
It is that image de-noising method provided by the embodiments of the present application is not limitation with Fig. 1 and specific order below.The specific stream of this method
Journey is as follows:
Step S210 obtains training sample set.
In the present embodiment, the training sample set include noise image sample set and with the noise image sample set pair
The noise-free picture sample set answered.Wherein, in order to obtain accurate noise-free picture sample set, present inventor is by long-term
Research proposes following proposal:
Firstly, obtaining the multiframe noise image sample that image capture device is continuously shot based on the image acquisition parameter of configuration
To construct noise image sample set, described image acquisition parameter includes image exposure parameter and filming frequency parameter.
Then, since the variance of the distribution of the picture noise of noise image sample each in noise image sample set is not true
Fixed, therefore the pixel value size of a part of picture noise of noise image sample is actually except 0~255, still
The problem of being indicated due to real figure, final pixel value still can be fixed on 0~255.If decent using these noise patterns
The average value of this each pixel is actually to send out for being subsequently generated corresponding noise-free picture sample, noise-free picture sample
Offset is given birth to, to cause the inaccuracy of noise-free picture sample.In consideration of it, inventor proposes that the noise image can be directed to
Every frame noise image sample in sample set is ranked up the pixel value of each pixel in the frame noise image sample,
And pixel value is greater than to the extraordinary image of the second presetted pixel value according to ranking results less than the first presetted pixel value and pixel value
Vegetarian refreshments is rejected, the noise image sample after obtaining rejecting abnormalities pixel.Then, making an uproar after rejecting each frame extraordinary image vegetarian refreshments is calculated
The average value of each pixel of acoustic image sample.
Finally, according to the relationship between preset noise image sample and corresponding noise-free picture sample and rejecting each frame
The average value of each pixel of noise image sample after extraordinary image vegetarian refreshments obtains corresponding with the noise image sample set
Noise-free picture sample set.
Wherein, there are following relationships with corresponding noise-free picture sample for the noise image sample:
xn=clip (yn+εn(yn))
εn(yn)~N (0, σn 2(yn))
Wherein, xnFor noise image sample, ynFor corresponding noise-free picture sample, εn(yn) it is ynNoise profile function,
σn 2(yn) it is the variance that noise profile function is obeyed, wherein σn 2(yn)=β1 nyn+β2 n, β1And β2Equipment is acquired with described image
Imaging sensor parameter and sensitivity numerical value it is related.
Collected noise-free picture sample accuracy is high through the above scheme, can effectively improve subsequently through deep learning
Obtained image denoising ability.
The noise image sample set is input in convolutional neural networks and is trained by step S220, exports trained
The denoising image pattern collection arrived.
The convergence in view of algorithm is needed in the design of convolutional neural networks, and how to guarantee the same of denoising effect
When, reduce the loss of image detail texture.Present inventor is directed to the demand, takes coding-decoding mode, eliminates biography
The normalization layer of system convolutional neural networks, and a large amount of Long-Term Link articulamentums are added.In detail, in a kind of possible reality
It applies in mode, please refers to Fig. 2, include multiple convolutional layer Convolution in the convolutional neural networks of the present embodiment design
As the part coding Encoder of convolutional neural networks and multiple warp lamination Deconvolution as convolutional Neural net
The part decoding Decoder of network.The main function of convolutional layer Convolution is that the noise image sample that will be inputted passes through largely
Convolutional layer extracts the characteristic information of noise image sample, removes picture noise；The effect of warp lamination Deconvolution is extensive
Image detail in the characteristic information of multiple noise image sample.
In order to accelerate the convergence of convolutional neural networks in the training process, the present embodiment is added greatly in convolutional neural networks
The Long-Term Link articulamentum of amount, after the convolutional layer Convolution characteristic information extracted in front is added directly into
In the warp lamination Deconvolution in face.Thus, it is possible to which reducing convolutional neural networks is using back propagation Back
Gradient disappearance problem under Propagation.Meanwhile Long-Term Link articulamentum may be considered front convolutional layer
The medium-high frequency image feature information that Convolution is extracted is added back to subsequent warp lamination Deconvolution, and these
The grain details of medium-high frequency image feature information i.e. noise image sample, so Long-Term Link articulamentum is added can
The problem of greatly to alleviate grain details loss.
Based on above-mentioned analysis, the present embodiment is led to by the way that the noise image sample set to be input in convolutional neural networks
Each layer convolutional layer crossed in the convolutional neural networks successively extracts each noise image sample in the noise image sample set
Image feature information, and the image feature information for each noise image sample that every layer of convolutional layer extracts is passed through into connection
Layer be sent to in the symmetrically arranged warp lamination of this layer of convolutional layer, it is decent according to each noise pattern by the warp lamination
This image feature information generates the denoising image pattern collection after deconvolution.
Step S230 is calculated separately corresponding square according to the noise-free picture sample set and the denoising image pattern collection
Error loss function value and maximum a posteriori loss function value, and lost according to the mean square error loss function value and maximum a posteriori
Functional value obtains total losses functional value.
It is calculated in a kind of possible embodiment according to the noise-free picture sample set and the denoising image pattern collection
Mean square error loss function value, is specifically as follows:
Calculate separately in noise-free picture sample set every frame noise-free picture sample and the denoising image pattern concentrate with it is every
The square value of each pixel difference between the corresponding denoising image pattern of frame noise-free picture sample, and it is each by what is be calculated
The mean value of the square value of pixel difference is as the mean square error loss function value MSE LOSS.
In a kind of possible embodiment, calculated according to the noise-free picture sample set and the denoising image pattern collection
Maximum a posteriori loss function value, is specifically as follows:
Firstly, according to each noise-free picture sample GT in the noise-free picture sample set and in each noise-free picture sample
The poisson noise image poi after poisson noise is added in this, calculates each noise-free picture in the noise-free picture sample set
There are the maximum a posteriori probability P of poisson noise (poi | GT) for sample.The specific formula for calculation of P (poi | GT) may include:
Then, according in each noise-free picture sample be added poisson noise after poisson noise image poi, each
The gaussian noise image I after Gaussian noise is added in noise-free picture samplenoise(noise), each poisson noise image is calculated
There are the maximum a posteriori probability P (I of Gaussian noise by poinoise|poi).Specific formula for calculation may include:
Inoise=poi+ ∈, ∈~N (0, σ)
According to each noise-free picture sample, there are the posterior probability P of poisson noise (poi | GT) and each poisson noise figure
As there are the posterior probability P (I of Gaussian noisenoise| poi), calculate each noise-free picture sample in the noise-free picture sample set
There are the maximum a posteriori probability P (I of Poisson Gaussian noise by this GTnoise|GT).Specific formula for calculation may include:
Finally, calculating, using each noise-free picture sample, there are the maximum a posteriori probabilities of Poisson Gaussian noise as the negative of independent variable
Logarithmic function value-logP (Inoise| GT), and by the negative logarithmic function value-logP (Inoise| GT) it is used as the maximum a posteriori
Loss function value MAP LOSS, and constantly minimize maximum a posteriori loss function value MAP LOSS.
Then, the first of the mean square error loss function value MSE LOSS can be calculated separately according to default weight proportion
Second weight loss function value of weight loss function value and the maximum a posteriori loss function value MAP LOSS, then basis
The first weight loss function value and the second weight loss function value obtain the total losses functional value.For example, can be with
The sum of the first weight loss function value and the second weight loss function value are calculated to obtain the total losses functional value.
Step S240, based on convolutional neural networks described in the total losses functional value repetitive exercise being calculated, until meeting
When training termination condition, the image denoising model that output training obtains, and denoising image is treated according to described image denoising model
Carry out denoising.
In detail, based on the total losses functional value being calculated, the network parameter of the convolutional neural networks is constantly updated,
And repeat above-mentioned training step and be iterated training, until meet training termination condition (such as total losses functional value no longer decline,
Repetitive exercise number reaches preset times etc.) when, the image denoising model that output training obtains, the image denoising model can at this time
To possess image denoising ability, and denoising image is treated according to described image denoising model and carries out denoising.
The present embodiment is by combining mean square error loss function value and maximum a posteriori loss function value repetitive exercise to roll up as a result,
Product neural network, can greatly promote the removal effect of real noise.
Further, referring to Fig. 3, the embodiment of the present application also provides a kind of image denoising device 200, image denoising dress
Setting 200 may include:
Obtain module 210, for obtaining training sample set, the training sample set include noise image sample set and with
The corresponding noise-free picture sample set of the noise image sample set.
Training module 220 is trained for the noise image sample set to be input in convolutional neural networks, output
The denoising image pattern collection that training obtains.
Computing module 230, for being calculated separately pair according to the noise-free picture sample set and the denoising image pattern collection
The mean square error loss function value and maximum a posteriori loss function value answered, and according to the mean square error loss function value and maximum
Posterior loss functional value obtains total losses functional value.
Denoising module 240, for based on convolutional Neural net described in the total losses functional value repetitive exercise being calculated
Network, when meeting training termination condition, the image denoising model that output training obtains, and according to described image denoising model pair
Image to be denoised carries out denoising.
In a kind of possible embodiment, the acquisition module 210 can specifically obtain trained sample in the following manner
This collection:
Multiframe noise image sample that image capture device is continuously shot based on the image acquisition parameter of configuration is obtained with structure
Noise image sample set is built, described image acquisition parameter includes image exposure parameter and filming frequency parameter；
For every frame noise image sample in the noise image sample set, to each of the frame noise image sample
The pixel value of pixel is ranked up, and is greater than pixel value less than the first presetted pixel value and pixel value according to ranking results
The extraordinary image vegetarian refreshments of second presetted pixel value is rejected, the noise image sample after obtaining rejecting abnormalities pixel；
Calculate the average value of each pixel of the noise image sample after rejecting each frame extraordinary image vegetarian refreshments；
According to the relationship between preset noise image sample and corresponding noise-free picture sample and reject each frame exception
The average value of each pixel of noise image sample after pixel obtains nothing corresponding with the noise image sample set and makes an uproar
Image pattern collection；
There are following relationships with corresponding noise-free picture sample for the noise image sample:
xn=clip (yn+εn(yn))
εn(yn)~N (0, σn 2(yn))
Wherein, xnFor noise image sample, ynFor corresponding noise-free picture sample, εn(yn) it is ynNoise profile function,
σn 2(yn) it is the variance that noise profile function is obeyed, wherein σn 2(yn)=β1 nyn+β2 n, β1And β2Equipment is acquired with described image
Imaging sensor parameter and sensitivity numerical value it is related.
In a kind of possible embodiment, the training module 230 can specifically export trained in the following manner
The denoising image pattern collection arrived:
The noise image sample set is input in convolutional neural networks, each layer in the convolutional neural networks is passed through
Convolutional layer successively extracts the image feature information of each noise image sample in the noise image sample set；
The image feature information for each noise image sample that every layer of convolutional layer extracts is sent to by articulamentum
In the symmetrically arranged warp lamination of this layer of convolutional layer, by the warp lamination according to the image of each noise image sample
Characteristic information generates the denoising image pattern collection after deconvolution.
In a kind of possible embodiment, the computing module 240 can specifically calculate separately pair in the following manner
The mean square error loss function value and maximum a posteriori loss function value answered:
Mean square error loss function value is calculated according to the noise-free picture sample set and the denoising image pattern collection, specifically
Are as follows:
Calculate separately in noise-free picture sample set every frame noise-free picture sample and the denoising image pattern concentrate with it is every
The square value of each pixel difference between the corresponding denoising image pattern of frame noise-free picture sample, and it is each by what is be calculated
The mean value of the square value of pixel difference is as the mean square error loss function value；And
Maximum a posteriori loss function value is calculated according to the noise-free picture sample set and the denoising image pattern collection, specifically
Are as follows:
Add according to each noise-free picture sample in the noise-free picture sample set and in each noise-free picture sample
Enter the poisson noise image after poisson noise, each noise-free picture sample calculated in the noise-free picture sample set has pool
The posterior probability of loose noise；
According in each noise-free picture sample be added poisson noise after poisson noise image, in each noise-free picture
The gaussian noise image after Gaussian noise is added in sample, calculating each poisson noise image, there are the posteriority of Gaussian noise is general
Rate；
According to each noise-free picture sample, there are the posterior probability of poisson noise and each poisson noise image there is height
The posterior probability of this noise, each noise-free picture sample is calculated, and there are the maximum a posteriori probabilities of Poisson Gaussian noise；
Using each noise-free picture sample, there are the maximum a posteriori probabilities of Poisson Gaussian noise as the negative logarithm of independent variable for calculating
Functional value, and using the negative logarithmic function value as the maximum a posteriori loss function value.
It is understood that the concrete operation method of each functional module in the present embodiment can refer to above method embodiment
The detailed description of middle corresponding steps, it is no longer repeated herein.
Further, referring to Fig. 4, being the electronic equipment provided by the embodiments of the present application for above-mentioned image de-noising method
100 structural schematic block diagram.In the present embodiment, the electronic equipment 100 can be made general bus architecture knot by bus 110
Structure is realized.According to the concrete application of electronic equipment 100 and overall design constraints condition, bus 110 may include any amount
Interconnection bus and bridge joint.Together by various circuit connections, these circuits include processor 120, storage medium to bus 110
130 and bus interface 140.Optionally, bus interface 140 can be used, and network adapter 150 is equal via total in electronic equipment 100
Line 110 connects.Network adapter 150 can be used for realizing the signal processing function of physical layer in electronic equipment 100, and pass through antenna
Realize sending and receiving for radiofrequency signal.User interface 160 can connect external equipment, such as: keyboard, display, mouse or
Person's control stick etc..Bus 110 can also connect various other circuits, such as timing source, peripheral equipment, voltage regulator or power
Circuit etc. is managed, these circuits are known in the art, therefore are no longer described in detail.
It can replace, electronic equipment 100 may also be configured to generic processing system, such as be commonly referred to as chip, the general place
Reason system includes: to provide the one or more microprocessors of processing function, and provide at least part of of storage medium 130
External memory, it is all these all to be linked together by external bus architecture and other support circuits.
Alternatively, following realize can be used in electronic equipment 100: having processor 120, bus interface 140, user
The ASIC (specific integrated circuit) of interface 160；And it is integrated at least part of the storage medium 130 in one single chip, or
Following realize can be used in person, electronic equipment 100: one or more FPGA (field programmable gate array), PLD are (programmable
Logical device), controller, state machine, gate logic, discrete hardware components, any other suitable circuit or be able to carry out this
Any combination of the circuit of various functions described in application in the whole text.
Wherein, processor 120 is responsible for management bus 110 and general processing (is stored on storage medium 130 including executing
Software).One or more general processors and/or application specific processor can be used to realize in processor 120.Processor 120
Example includes microprocessor, microcontroller, dsp processor and the other circuits for being able to carry out software.It should be by software broadly
It is construed to indicate instruction, data or any combination thereof, regardless of being called it as software, firmware, middleware, microcode, hard
Part description language or other.
Storage medium 130 is illustrated as separating with processor 120 in Fig. 4, however, those skilled in the art be easy to it is bright
White, storage medium 130 or its arbitrary portion can be located at except electronic equipment 100.For example, storage medium 130 may include
Transmission line, the carrier waveform modulated with data, and/or the computer product that separates with radio node, these media can be with
It is accessed by processor 120 by bus interface 140.Alternatively, storage medium 130 or its arbitrary portion can integrate everywhere
It manages in device 120, for example, it may be cache and/or general register.
Above-described embodiment can be performed in the processor 120, specifically, can store in the storage medium 130 described
Image denoising device 200, the processor 120 can be used for executing described image denoising device 200.
Further, the embodiment of the present application also provides a kind of nonvolatile computer storage media, the computer is deposited
Storage media is stored with computer executable instructions, which can be performed the figure in above-mentioned any means embodiment
As denoising method.
In conclusion the embodiment of the present application provides a kind of image de-noising method and device, principle is implemented are as follows: by noise
Image pattern collection, which is input in convolutional neural networks, to be trained, and the denoising image pattern collection that output training obtains is made an uproar according to nothing
Image pattern collection and denoising image pattern collection calculate separately corresponding mean square error loss function value and maximum a posteriori loss function
Value, and total losses functional value is obtained according to mean square error loss function value and maximum a posteriori loss function value, finally based on calculating
Obtained total losses functional value repetitive exercise convolutional neural networks, when meeting training termination condition, output training is obtained
Image denoising model, and denoising image is treated according to image denoising model and carries out denoising.As a result, by combining mean square error
Loss function value and maximum a posteriori loss function value repetitive exercise convolutional neural networks, can greatly promote the removal of real noise
Effect.
In embodiment provided herein, it should be understood that disclosed device and method, it can also be by other
Mode realize.Device and method embodiment described above is only schematical, for example, flow chart and frame in attached drawing
Figure shows the system frame in the cards of the system of multiple embodiments according to the application, method and computer program product
Structure, function and operation.In this regard, each box in flowchart or block diagram can represent a module, section or code
A part, a part of the module, section or code includes one or more for implementing the specified logical function
Executable instruction.It should also be noted that function marked in the box can also be with not in some implementations as replacement
It is same as the sequence marked in attached drawing generation.For example, two continuous boxes can actually be basically executed in parallel, they have
When can also execute in the opposite order, this depends on the function involved.It is also noted that in block diagram and or flow chart
Each box and the box in block diagram and or flow chart combination, can function or movement as defined in executing it is dedicated
Hardware based system realize, or can realize using a combination of dedicated hardware and computer instructions.
In addition, each functional module in each embodiment of the application can integrate one independent portion of formation together
Point, it is also possible to modules individualism, an independent part can also be integrated to form with two or more modules.
It can replace, can be realized wholly or partly by software, hardware, firmware or any combination thereof.When
When using software realization, can entirely or partly it realize in the form of a computer program product.The computer program product
Including one or more computer instructions.It is all or part of when loading on computers and executing the computer program instructions
Ground is generated according to process or function described in the embodiment of the present application.The computer can be general purpose computer, special purpose computer,
Computer network or other programmable devices.The computer instruction may be stored in a computer readable storage medium, or
Person is transmitted from a computer readable storage medium to another computer readable storage medium, for example, the computer instruction
Wired (such as coaxial cable, optical fiber, digital subscriber can be passed through from a web-site, computer, server or data center
Line (DSL)) or wireless (such as infrared, wireless, microwave etc.) mode to another web-site, computer, server or data
It is transmitted at center.The computer readable storage medium can be any usable medium that computer can access and either wrap
The data storage devices such as electronic equipment, server, the data center integrated containing one or more usable mediums.The usable medium
It can be magnetic medium, (for example, floppy disk, hard disk, tape), optical medium (for example, DVD) or semiconductor medium (such as solid-state
Hard disk Solid State Disk (SSD)) etc..
It should be noted that, in this document, term " including ", " including " or its any other variant are intended to non-row
Its property includes, so that the process, method, article or equipment for including a series of elements not only includes those elements, and
And further include the other elements being not explicitly listed, or further include for this process, method, article or equipment institute it is intrinsic
Element.In the absence of more restrictions, the element limited by sentence " including one ... ", it is not excluded that including institute
State in the process, method, article or equipment of element that there is also other identical elements.
It is obvious to a person skilled in the art that the application is not limited to the details of above-mentioned exemplary embodiment, Er Qie
In the case where without departing substantially from spirit herein or essential characteristic, the application can be realized in other specific forms.Therefore, no matter
From the point of view of which point, the present embodiments are to be considered as illustrative and not restrictive, and scope of the present application is by appended power
Benefit requires rather than above description limits, it is intended that all by what is fallen within the meaning and scope of the equivalent elements of the claims
Variation is included in the application.Any reference signs in the claims should not be construed as limiting the involved claims.
Claims (10)
1. a kind of image de-noising method, which is characterized in that be applied to electronic equipment, which comprises
Obtain training sample set, the training sample set include noise image sample set and with the noise image sample set pair
The noise-free picture sample set answered；
The noise image sample set is input in convolutional neural networks and is trained, the denoising image sample that output training obtains
This collection；
Corresponding mean square error loss function is calculated separately according to the noise-free picture sample set and the denoising image pattern collection
Value and maximum a posteriori loss function value, and obtained always according to the mean square error loss function value and maximum a posteriori loss function value
Loss function value；
Based on convolutional neural networks described in the total losses functional value repetitive exercise being calculated, until meeting training termination condition
When, the image denoising model that output training obtains, and denoising image is treated according to described image denoising model and carries out denoising.
2. image de-noising method according to claim 1, which is characterized in that the step of the acquisition training sample set, packet
It includes:
Image capture device is obtained to make an uproar based on the multiframe noise image sample that the image acquisition parameter of configuration is continuously shot to construct
Acoustic image sample set, described image acquisition parameter include image exposure parameter and filming frequency parameter；
For every frame noise image sample in the noise image sample set, to each pixel in the frame noise image sample
The pixel value of point is ranked up, and pixel value is greater than second less than the first presetted pixel value and pixel value according to ranking results
The extraordinary image vegetarian refreshments of presetted pixel value is rejected, the noise image sample after obtaining rejecting abnormalities pixel；
Calculate the average value of each pixel of the noise image sample after rejecting each frame extraordinary image vegetarian refreshments；
According to the relationship between preset noise image sample and corresponding noise-free picture sample and reject each frame abnormal pixel
The average value of each pixel of noise image sample after point obtains noise-free picture corresponding with the noise image sample set
Sample set.
3. image de-noising method according to claim 2, which is characterized in that the noise image sample is made an uproar with corresponding nothing
There are following relationships for image pattern:
xn=clip (yn+εn(yn))
εn(yn)~N (0, σn 2(yn))
Wherein, xnFor noise image sample, ynFor corresponding noise-free picture sample, εn(yn) it is ynNoise profile function, σn 2
(yn) it is the variance that noise profile function is obeyed, wherein σn 2(yn)=β1 nyn+β2 n, β1And β2With the figure of described image acquisition equipment
As sensor parameters and sensitivity numerical value are related.
4. image de-noising method according to claim 1, which is characterized in that described to input the noise image sample set
The step of being trained into convolutional neural networks, exporting the denoising image pattern collection that training obtains, comprising:
The noise image sample set is input in convolutional neural networks, each layer convolution in the convolutional neural networks is passed through
Layer successively extracts the image feature information of each noise image sample in the noise image sample set；
The image feature information for each noise image sample that every layer of convolutional layer extracts is sent to and is somebody's turn to do by articulamentum
In the layer symmetrically arranged warp lamination of convolutional layer, by the warp lamination according to the characteristics of image of each noise image sample
Information generates the denoising image pattern collection after deconvolution.
5. image de-noising method according to claim 1, which is characterized in that it is described according to the noise-free picture sample set and
The denoising image pattern collection calculates separately the step of corresponding mean square error loss function value and maximum a posteriori loss function value,
Include:
Mean square error loss function value is calculated according to the noise-free picture sample set and the denoising image pattern collection, specifically:
Calculate separately in noise-free picture sample set every frame noise-free picture sample and the denoising image pattern concentrate with every frame without
The square value of each pixel difference between the corresponding denoising image pattern of image pattern of making an uproar, and each pixel that will be calculated
The mean value of the square value of point difference is as the mean square error loss function value；And
Maximum a posteriori loss function value is calculated according to the noise-free picture sample set and the denoising image pattern collection, specifically:
Pool is added according to each noise-free picture sample in the noise-free picture sample set and in each noise-free picture sample
Poisson noise image after loose noise, there are Poissons to make an uproar for each noise-free picture sample for calculating in the noise-free picture sample set
The posterior probability of sound；
According in each noise-free picture sample be added poisson noise after poisson noise image, in each noise-free picture sample
Gaussian noise image after middle addition Gaussian noise, calculating each poisson noise image, there are the posterior probability of Gaussian noise；
According to each noise-free picture sample, there are the posterior probability of poisson noise and each poisson noise image, there are Gausses to make an uproar
The posterior probability of sound, each noise-free picture sample is calculated, and there are the maximum a posteriori probabilities of Poisson Gaussian noise；
Using each noise-free picture sample, there are the maximum a posteriori probabilities of Poisson Gaussian noise as the negative logarithmic function of independent variable for calculating
Value, and using the negative logarithmic function value as the maximum a posteriori loss function value.
6. image de-noising method according to claim 1, which is characterized in that described according to the mean square error loss function
The step of value and maximum a posteriori loss function value obtain total losses functional value, comprising:
The first weight loss function value and the institute of the mean square error loss function value are calculated separately according to default weight proportion
State the second weight loss function value of maximum a posteriori loss function value；
The total losses functional value is obtained according to the first weight loss function value and the second weight loss function value.
7. a kind of image denoising device, which is characterized in that be applied to electronic equipment, described device includes:
Module is obtained, for obtaining training sample set, the training sample set includes noise image sample set and makes an uproar with described
The corresponding noise-free picture sample set of acoustic image sample set；
Training module is trained for the noise image sample set to be input in convolutional neural networks, is exported trained
The denoising image pattern collection arrived；
Computing module, it is corresponding square for being calculated separately according to the noise-free picture sample set and the denoising image pattern collection
Error loss function value and maximum a posteriori loss function value, and lost according to the mean square error loss function value and maximum a posteriori
Functional value obtains total losses functional value；
Denoising module, for based on convolutional neural networks described in the total losses functional value repetitive exercise being calculated, until
When meeting training termination condition, the image denoising model that output training obtains, and denoising is treated according to described image denoising model
Image carries out denoising.
8. image denoising device according to claim 7, which is characterized in that the acquisition module is especially by following manner
Obtain training sample set:
Image capture device is obtained to make an uproar based on the multiframe noise image sample that the image acquisition parameter of configuration is continuously shot to construct
Acoustic image sample set, described image acquisition parameter include image exposure parameter and filming frequency parameter；
For every frame noise image sample in the noise image sample set, to each pixel in the frame noise image sample
The pixel value of point is ranked up, and pixel value is greater than second less than the first presetted pixel value and pixel value according to ranking results
The extraordinary image vegetarian refreshments of presetted pixel value is rejected, the noise image sample after obtaining rejecting abnormalities pixel；
Calculate the average value of each pixel of the noise image sample after rejecting each frame extraordinary image vegetarian refreshments；
According to the relationship between preset noise image sample and corresponding noise-free picture sample and reject each frame abnormal pixel
The average value of each pixel of noise image sample after point obtains noise-free picture corresponding with the noise image sample set
Sample set；
There are following relationships with corresponding noise-free picture sample for the noise image sample:
xn=clip (yn+εn(yn))
εn(yn)~N (0, σn 2(yn))
Wherein, xnFor noise image sample, ynFor corresponding noise-free picture sample, εn(yn) it is ynNoise profile function, σn 2
(yn) it is the variance that noise profile function is obeyed, wherein σn 2(yn)=β1 nyn+β2 n, β1And β2With the figure of described image acquisition equipment
As sensor parameters and sensitivity numerical value are related.
9. image denoising device according to claim 7, which is characterized in that the training module is especially by following manner
The denoising image pattern collection that output training obtains:
The noise image sample set is input in convolutional neural networks, each layer convolution in the convolutional neural networks is passed through
Layer successively extracts the image feature information of each noise image sample in the noise image sample set；
The image feature information for each noise image sample that every layer of convolutional layer extracts is sent to and is somebody's turn to do by articulamentum
In the layer symmetrically arranged warp lamination of convolutional layer, by the warp lamination according to the characteristics of image of each noise image sample
Information generates the denoising image pattern collection after deconvolution.
10. image denoising device according to claim 7, which is characterized in that the computing module is especially by with lower section
Formula calculates separately corresponding mean square error loss function value and maximum a posteriori loss function value:
Mean square error loss function value is calculated according to the noise-free picture sample set and the denoising image pattern collection, specifically:
Calculate separately in noise-free picture sample set every frame noise-free picture sample and the denoising image pattern concentrate with every frame without
The square value of each pixel difference between the corresponding denoising image pattern of image pattern of making an uproar, and each pixel that will be calculated
The mean value of the square value of point difference is as the mean square error loss function value；And
Maximum a posteriori loss function value is calculated according to the noise-free picture sample set and the denoising image pattern collection, specifically:
Pool is added according to each noise-free picture sample in the noise-free picture sample set and in each noise-free picture sample
Poisson noise image after loose noise, there are Poissons to make an uproar for each noise-free picture sample for calculating in the noise-free picture sample set
The posterior probability of sound；
According in each noise-free picture sample be added poisson noise after poisson noise image, in each noise-free picture sample
Gaussian noise image after middle addition Gaussian noise, calculating each poisson noise image, there are the posterior probability of Gaussian noise；
According to each noise-free picture sample, there are the posterior probability of poisson noise and each poisson noise image, there are Gausses to make an uproar
The posterior probability of sound, each noise-free picture sample is calculated, and there are the maximum a posteriori probabilities of Poisson Gaussian noise；
Using each noise-free picture sample, there are the maximum a posteriori probabilities of Poisson Gaussian noise as the negative logarithmic function of independent variable for calculating
Value, and using the negative logarithmic function value as the maximum a posteriori loss function value.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201811629202.7A CN109754376B (en) | 2018-12-28 | 2018-12-28 | Image denoising method and device |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201811629202.7A CN109754376B (en) | 2018-12-28 | 2018-12-28 | Image denoising method and device |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN109754376A true CN109754376A (en) | 2019-05-14 |
| CN109754376B CN109754376B (en) | 2020-11-10 |
Family
ID=66403223
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201811629202.7A Active CN109754376B (en) | 2018-12-28 | 2018-12-28 | Image denoising method and device |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN109754376B (en) |
Cited By (17)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN110163827A (en) * | 2019-05-28 | 2019-08-23 | 腾讯科技（深圳）有限公司 | Training method, image de-noising method, device and the medium of image denoising model |
| CN110276730A (en) * | 2019-06-13 | 2019-09-24 | Oppo广东移动通信有限公司 | Image processing method, device, electronic equipment |
| CN110290289A (en) * | 2019-06-13 | 2019-09-27 | Oppo广东移动通信有限公司 | Image denoising method, device, electronic equipment and storage medium |
| CN110490824A (en) * | 2019-08-23 | 2019-11-22 | 广东浪潮大数据研究有限公司 | The method and device of image denoising |
| CN110807741A (en) * | 2019-10-15 | 2020-02-18 | 腾讯科技（深圳）有限公司 | Training method of image processing network, image denoising method and device |
| CN111260579A (en) * | 2020-01-17 | 2020-06-09 | 北京理工大学 | Low-light-level image denoising and enhancing method based on physical noise generation model |
| CN111369456A (en) * | 2020-02-28 | 2020-07-03 | 深圳市商汤科技有限公司 | Image denoising method and device, electronic device and storage medium |
| CN111709891A (en) * | 2020-06-12 | 2020-09-25 | 北京小米松果电子有限公司 | Training method of image denoising model, image denoising method, device and medium |
| CN111861912A (en) * | 2020-06-30 | 2020-10-30 | 湖南傲英创视信息科技有限公司 | Line array peripheral scanning three-dimensional panoramic image denoising method, system and storage medium |
| CN111899199A (en) * | 2020-08-07 | 2020-11-06 | 深圳市捷顺科技实业股份有限公司 | Image processing method, device, equipment and storage medium |
| CN112950501A (en) * | 2021-02-26 | 2021-06-11 | 平安科技（深圳）有限公司 | Image noise reduction method, device and equipment based on noise field and storage medium |
| CN113160089A (en) * | 2021-05-10 | 2021-07-23 | 福州大学 | Image denoising method based on local density loss |
| CN115082357A (en) * | 2022-07-20 | 2022-09-20 | 深圳思谋信息科技有限公司 | Video denoising data set generation method and device, computer equipment and storage medium |
| US11540798B2 (en) | 2019-08-30 | 2023-01-03 | The Research Foundation For The State University Of New York | Dilated convolutional neural network system and method for positron emission tomography (PET) image denoising |
| CN116051427A (en) * | 2023-03-31 | 2023-05-02 | 季华实验室 | Point cloud denoising model acquisition method, point cloud fusion method and related equipment thereof |
| CN116681618A (en) * | 2023-06-13 | 2023-09-01 | 强联智创（北京）科技有限公司 | Image denoising method, electronic device and storage medium |
| CN117094909A (en) * | 2023-08-31 | 2023-11-21 | 青岛天仁微纳科技有限责任公司 | Nanometer stamping wafer image acquisition processing method |
Citations (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN106204468A (en) * | 2016-06-27 | 2016-12-07 | 深圳市未来媒体技术研究院 | A kind of image de-noising method based on ReLU convolutional neural networks |
| CN106683063A (en) * | 2017-01-22 | 2017-05-17 | 厦门美图之家科技有限公司 | Method and device of image denoising |
| CN106981055A (en) * | 2017-03-30 | 2017-07-25 | 西安交通大学 | A kind of ICCD image de-noising methods |
| CN109087255A (en) * | 2018-07-18 | 2018-12-25 | 中国人民解放军陆军工程大学 | Lightweight depth image denoising method based on mixed loss |
-
2018
- 2018-12-28 CN CN201811629202.7A patent/CN109754376B/en active Active
Patent Citations (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN106204468A (en) * | 2016-06-27 | 2016-12-07 | 深圳市未来媒体技术研究院 | A kind of image de-noising method based on ReLU convolutional neural networks |
| CN106683063A (en) * | 2017-01-22 | 2017-05-17 | 厦门美图之家科技有限公司 | Method and device of image denoising |
| CN106981055A (en) * | 2017-03-30 | 2017-07-25 | 西安交通大学 | A kind of ICCD image de-noising methods |
| CN109087255A (en) * | 2018-07-18 | 2018-12-25 | 中国人民解放军陆军工程大学 | Lightweight depth image denoising method based on mixed loss |
Cited By (26)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN110163827B (en) * | 2019-05-28 | 2023-01-10 | 腾讯科技（深圳）有限公司 | Training method of image denoising model, image denoising method, device and medium |
| CN110163827A (en) * | 2019-05-28 | 2019-08-23 | 腾讯科技（深圳）有限公司 | Training method, image de-noising method, device and the medium of image denoising model |
| CN110290289B (en) * | 2019-06-13 | 2021-07-09 | Oppo广东移动通信有限公司 | Image noise reduction method and device, electronic equipment and storage medium |
| CN110276730A (en) * | 2019-06-13 | 2019-09-24 | Oppo广东移动通信有限公司 | Image processing method, device, electronic equipment |
| CN110290289A (en) * | 2019-06-13 | 2019-09-27 | Oppo广东移动通信有限公司 | Image denoising method, device, electronic equipment and storage medium |
| CN110276730B (en) * | 2019-06-13 | 2023-03-28 | Oppo广东移动通信有限公司 | Image processing method and device and electronic equipment |
| CN110490824A (en) * | 2019-08-23 | 2019-11-22 | 广东浪潮大数据研究有限公司 | The method and device of image denoising |
| US11540798B2 (en) | 2019-08-30 | 2023-01-03 | The Research Foundation For The State University Of New York | Dilated convolutional neural network system and method for positron emission tomography (PET) image denoising |
| CN110807741A (en) * | 2019-10-15 | 2020-02-18 | 腾讯科技（深圳）有限公司 | Training method of image processing network, image denoising method and device |
| CN111260579B (en) * | 2020-01-17 | 2021-08-03 | 北京理工大学 | Low-light-level image denoising and enhancing method based on physical noise generation model |
| CN111260579A (en) * | 2020-01-17 | 2020-06-09 | 北京理工大学 | Low-light-level image denoising and enhancing method based on physical noise generation model |
| CN111369456A (en) * | 2020-02-28 | 2020-07-03 | 深圳市商汤科技有限公司 | Image denoising method and device, electronic device and storage medium |
| CN111709891A (en) * | 2020-06-12 | 2020-09-25 | 北京小米松果电子有限公司 | Training method of image denoising model, image denoising method, device and medium |
| CN111709891B (en) * | 2020-06-12 | 2023-11-24 | 北京小米松果电子有限公司 | Training method of image denoising model, image denoising method, device and medium |
| CN111861912A (en) * | 2020-06-30 | 2020-10-30 | 湖南傲英创视信息科技有限公司 | Line array peripheral scanning three-dimensional panoramic image denoising method, system and storage medium |
| CN111899199A (en) * | 2020-08-07 | 2020-11-06 | 深圳市捷顺科技实业股份有限公司 | Image processing method, device, equipment and storage medium |
| CN111899199B (en) * | 2020-08-07 | 2024-03-19 | 深圳市捷顺科技实业股份有限公司 | Image processing method, device, equipment and storage medium |
| CN112950501A (en) * | 2021-02-26 | 2021-06-11 | 平安科技（深圳）有限公司 | Image noise reduction method, device and equipment based on noise field and storage medium |
| CN112950501B (en) * | 2021-02-26 | 2023-10-13 | 平安科技（深圳）有限公司 | Noise field-based image noise reduction method, device, equipment and storage medium |
| CN113160089A (en) * | 2021-05-10 | 2021-07-23 | 福州大学 | Image denoising method based on local density loss |
| CN115082357A (en) * | 2022-07-20 | 2022-09-20 | 深圳思谋信息科技有限公司 | Video denoising data set generation method and device, computer equipment and storage medium |
| CN116051427B (en) * | 2023-03-31 | 2023-07-04 | 季华实验室 | Point cloud denoising model acquisition method, point cloud fusion method and related equipment thereof |
| CN116051427A (en) * | 2023-03-31 | 2023-05-02 | 季华实验室 | Point cloud denoising model acquisition method, point cloud fusion method and related equipment thereof |
| CN116681618A (en) * | 2023-06-13 | 2023-09-01 | 强联智创（北京）科技有限公司 | Image denoising method, electronic device and storage medium |
| CN117094909A (en) * | 2023-08-31 | 2023-11-21 | 青岛天仁微纳科技有限责任公司 | Nanometer stamping wafer image acquisition processing method |
| CN117094909B (en) * | 2023-08-31 | 2024-04-02 | 青岛天仁微纳科技有限责任公司 | Nanometer stamping wafer image acquisition processing method |
Also Published As
| Publication number | Publication date |
|---|---|
| CN109754376B (en) | 2020-11-10 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN109754376A (en) | Image de-noising method and device | |
| CN110399929B (en) | Fundus image classification method, fundus image classification apparatus, and computer-readable storage medium | |
| CN111369563B (en) | Semantic segmentation method based on pyramid void convolutional network | |
| CN110826519B (en) | Face shielding detection method and device, computer equipment and storage medium | |
| CN106529447B (en) | Method for identifying face of thumbnail | |
| CN109409503B (en) | Neural network training method, image conversion method, device, equipment and medium | |
| CN107633522A (en) | Brain image dividing method and system based on local similarity movable contour model | |
| Wu et al. | Residual attention based network for hand bone age assessment | |
| CN108304765A (en) | Multitask detection device for face key point location and semantic segmentation | |
| US9152926B2 (en) | Systems, methods, and media for updating a classifier | |
| CN110992351B (en) | sMRI image classification method and device based on multi-input convolution neural network | |
| CN112183501B (en) | Depth counterfeit image detection method and device | |
| CN109359539A (en) | Attention appraisal procedure, device, terminal device and computer readable storage medium | |
| CN106845621A (en) | Dense population number method of counting and system based on depth convolutional neural networks | |
| CN108062543A (en) | A kind of face recognition method and device | |
| Ding et al. | Brain Medical Image Fusion Based on Dual‐Branch CNNs in NSST Domain | |
| CN108986115A (en) | Medical image cutting method, device and intelligent terminal | |
| CN109146825A (en) | Photography style conversion method, device and readable storage medium storing program for executing | |
| CN111784699B (en) | Method and device for carrying out target segmentation on three-dimensional point cloud data and terminal equipment | |
| CN112396605B (en) | Network training method and device, image recognition method and electronic equipment | |
| CN110503078A (en) | A kind of remote face identification method and system based on deep learning | |
| CN112163609A (en) | Image block similarity calculation method based on deep learning | |
| CN110859642A (en) | Method, device, equipment and storage medium for realizing medical image auxiliary diagnosis based on AlexNet network model | |
| Lu et al. | Image-specific prior adaptation for denoising | |
| CN113837965B (en) | Image definition identification method and device, electronic equipment 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 | ||
| GR01 | Patent grant | ||
| GR01 | Patent grant |