CN109146824A - A kind of image noise reduction processing method and a kind of storage equipment - Google Patents
A kind of image noise reduction processing method and a kind of storage equipment Download PDFInfo
- Publication number
- CN109146824A CN109146824A CN201811129933.5A CN201811129933A CN109146824A CN 109146824 A CN109146824 A CN 109146824A CN 201811129933 A CN201811129933 A CN 201811129933A CN 109146824 A CN109146824 A CN 109146824A
- Authority
- CN
- China
- Prior art keywords
- picture
- noise
- noiseless
- value
- pixel
- 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
Classifications
-
- 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
Landscapes
- Physics & Mathematics (AREA)
- General Physics & Mathematics (AREA)
- Engineering & Computer Science (AREA)
- Theoretical Computer Science (AREA)
- Image Processing (AREA)
Abstract
The present invention relates to picture noise process field, in particular to a kind of image noise reduction processing method and a kind of storage equipment.A kind of image noise reduction processing method, comprising steps of obtaining the photo exposed under short exposure mode；To each pixel of the n after short exposure photos, the random noise on average value processing removal pixel is carried out, a noiseless picture is obtained；Intermediate dynamic process number of pictures under short exposure mode is obtained, average value processing is carried out to the photo of acquisition, obtains a noise picture；It inputs noiseless picture and noise picture to neural network unit and carries out deep learning training, obtain final picture；Norm is sought according to the difference of the noise figure of final picture and noise picture and noiseless picture；Neural network unit is adjusted according to norm.Neural network after being adjusted can remove spot block distortion, color in real image well and each noise like such as make an uproar, and promote picture degree of purity.
Description
Technical field
The present invention relates to picture noise process field, in particular to a kind of image noise reduction processing method and a kind of storage are set
It is standby.
Background technique
In the prior art, image noise reduction process neural network based is the depth by noisy image Jing Guo full convolution
Learning method obtains muting image after carrying out reversed gradient derivation.And the generation for noise image, it is all that use will be caught
The picture grasped, artificial addition white Gaussian noise generate so-called noise picture in turn.But true picture noise depends on
In the form of sensor, scene changes etc. are substantially unlikely to be white Gaussian noise, and the training of the image based on white Gaussian noise
Neural network unit out can only handle dotted noise, and the noise in real life is substantially plaquelike noise,
Therefore this kind of mode trains the neural network unit come and has no idea to handle noise in real life.
Summary of the invention
For this reason, it may be necessary to a kind of image noise reduction processing method be provided, to solve existing image noise reduction neural network based
Process can not handle the problem of noise in real life.Specific technical solution is as follows:
A kind of image noise reduction processing method, comprising steps of obtaining the n photos exposed under short exposure mode；To short exposure
Each pixel of n afterwards photos carries out the random noise on average value processing removal pixel, obtains a noiseless figure
Piece；Intermediate dynamic process number of pictures under short exposure mode is obtained, average value processing is carried out to the photo of acquisition, obtains a noise
Picture；It inputs the noiseless picture and noise picture to neural network unit and carries out deep learning training, obtain final picture；
Obtain the noise figure of final picture；Obtain the difference of the noise picture and noiseless picture；According to making an uproar for the final picture
Sound value and the difference of the noise picture and noiseless picture seek norm；Neural network unit is adjusted according to the norm.
Further, described " to each pixel of the n after short exposure photos, to carry out average value processing and remove pixel
On random noise, obtain a noiseless picture ", further comprise the steps of: the pixel for taking the pixel of same position of n picture
Value is added, then divided by n, obtains pixel value mean value, and the new picture for taking the corresponding pixel of pixel value mean value to constitute is as nothing
Noise picture.
Further, described " to input the noiseless picture and noise picture to neural network unit and carry out deep learning
Training, obtains final picture ", the picture after further comprising the steps of: input deep learning training is to image signal processing unit into one
Step processing, obtains a noiseless picture, and described be further processed includes: white balance, tone mapping, brightness adjustment and go to Marseille
Gram.
Further, described " according to the difference of the noise figure of the final picture and the noise picture and noiseless picture
Value seeks norm ", the noise figure of the final picture is further comprised the steps of:, the difference of the noise picture and noiseless picture is subtracted,
Obtain value L, value L's square is L2 norm.
Further, the n value size increases with the reduction of scene brightness, and presetting n value minimum value is 15, and maximum value is
60。
To solve the above problems, additionally providing a kind of storage equipment, specific technical solution is as follows:
A kind of storage equipment, wherein being stored with instruction set, described instruction collection is for executing: obtaining and exposes under short exposure mode
N photos；To each pixel of the n after short exposure photos, the random noise on average value processing removal pixel is carried out,
Obtain a noiseless picture；Intermediate dynamic process number of pictures under short exposure mode is obtained, mean value is carried out to the photo of acquisition
Processing, obtains a noise picture；It inputs the noiseless picture and noise picture to neural network unit and carries out deep learning
Training, obtains final picture；Obtain the noise figure of final picture；Obtain the difference of the noise picture and noiseless picture；Root
Norm is sought according to the difference of the noise figure of the final picture and the noise picture and noiseless picture；It is adjusted according to the norm
Neural network unit.
Further, described instruction collection is also used to execute: described " each pixel of photo is opened to the n after short exposure, into
Row average value processing removes the random noise on pixel, obtains a noiseless picture ", it further comprises the steps of: and takes the same of n picture
The pixel value of the pixel of one position is added, then divided by n, is obtained pixel value mean value, taken the corresponding pixel of pixel value mean value
The new picture that point is constituted is as noiseless picture.
Further, described instruction collection is also used to execute: described " to input the noiseless picture and noise picture to nerve
Network unit carries out deep learning training, obtains final picture ", the picture after further comprising the steps of: input deep learning training is extremely
Image signal processing unit is further processed, and obtains a noiseless picture, described be further processed includes: white balance, tone
Mapping, brightness adjustment and demosaicing.
Further, described instruction collection is also used to execute: described " according to the noise figure of the final picture and the noise
The difference of picture and noiseless picture seeks norm ", the noise figure of the final picture is further comprised the steps of:, the noise pattern is subtracted
The difference of piece and noiseless picture, obtains value L, and value L's square is L2 norm.
Further, described instruction collection is also used to execute: the n value size increases with the reduction of scene brightness, presets n
Being worth minimum value is 15, maximum value 60.
The beneficial effects of the present invention are: obtaining the photo exposed under short exposure mode, the photo after choosing exposure carries out equal
Value processing, obtains noiseless picture, chooses the number of pictures of dynamic process under exposure mode, carries out at mean value to the photo of acquisition
Reason, obtains a noise picture, the training material obtained by this kind of mode is closer to the noise in real life, by acquisition
Noiseless picture and noise picture carry out deep learning as training material input neural network unit and train, and obtain final figure
Piece, obtains the noise figure (theoretical noise value) of final picture, then obtains the difference (reality of the noise picture and noiseless picture
Noise figure), norm (i.e. basis is asked according to the difference of the noise figure of the final picture and the noise picture and noiseless picture
The difference of theoretical noise value and actual noise value seeks norm), norm is smaller to illustrate that neural network unit noise reduction accuracy is higher (i.e.
Theoretical noise value is closer to actual noise value), therefore neural network unit can be adjusted according to norm, or even can instruct by successive ignition
Practice, until adjusting neural network unit to goal-selling.Neural network after being adjusted can remove in real image well
Spot block distortion, color each noise like such as make an uproar, promote picture degree of purity.
Detailed description of the invention
Fig. 1 is a kind of flow chart of image noise reduction processing method described in specific embodiment；
Fig. 2 is a kind of module map for storing equipment described in specific embodiment.
Description of symbols:
200, equipment is stored.
Specific embodiment
Technology contents, construction feature, the objects and the effects for detailed description technical solution, below in conjunction with specific reality
It applies example and attached drawing is cooperated to be explained in detail.
Referring to Fig. 1, in the present embodiment, a kind of image noise reduction processing method be can be applicable in a kind of storage equipment,
The storage equipment includes but is not limited to: smart phone, tablet computer, Desktop PC, laptop, PDA etc..
The inventive concept of most critical of the present invention is: using the noise of actual acquisition as training material, is input to nerve net
Network unit carries out deep learning training, and then the noise that neural network prediction after training goes out is compared with actual noise,
The present invention seeks norm using the two difference, and norm is smaller, illustrates that trained precision is higher, therefore can be adjusted by norm
Whole neural network unit, it might even be possible to by successive ignition training, by neural network unit optimization to goal-selling.Specific skill
Art design is mainly: carrying out average value processing by short exposure exposure mode photo, and to photo, obtains noiseless picture and make an uproar
The two is input to neural network unit and carries out deep learning, obtained the noise figure of final picture, obtain the noise by sound spectrogram piece
The difference of picture and noiseless picture；According to the difference of the noise figure of the final picture and the noise picture and noiseless picture
Value seeks norm；Neural network unit is adjusted further according to norm.Norm is smaller, then reduces picture by the neural network unit and make an uproar
The precision of sound is then higher.
In the present embodiment, pass through short exposure exposure mode photo, wherein use of the short exposure mode closer to user
Scene.And the noise that obtains of short exposure is the superposition for the noise that all shooting structures are obtained, can not by way of software mould
It is quasi- to obtain.
In the present embodiment, the time for exposure is generally the 1/3 of constant exposure time, can solve blurred image ask in this way
Topic.The typical time for exposure is 1/30 second.
Explanation explained below is done to some nouns occurred in present embodiment first:
L2 norm: L2 norm refers to the quadratic sum and then extraction of square root of vector each element.Inside recurrence, someone's handle has it
Recurrence " ridge regression " (Ridge Regression), someone is also its " weight decay weight decay ".This is very
Mostly, because its powerful effect is to improve an extremely important problem inside machine learning: over-fitting.
In the present embodiment, a kind of specific embodiment of image noise reduction processing method is as follows:
Step S101: the n photos exposed under short exposure mode are obtained.It can be used such as under type: can be by local upload
The n exposed under short exposure mode photos, or the n photos exposed under the short exposure mode that remote terminal is sent are obtained, or from
The n photos exposed under short exposure mode are obtained on server.Wherein in the present embodiment, the short exposure mould of camera shooting can be used
Formula, in a short time with the identical time for exposure, exposes n picture, n under the stabilization of tripod or other devices
Depending on scene brightness, n is more than or equal to 30 under normal circumstances.
In the present embodiment, n value size increases with the reduction of scene brightness, and presetting n value minimum value is 15, maximum value
It is 60.The value of the namely darker implementation n of night scene is bigger.
Step S102: to each pixel of the photo of the n after short exposure, carry out on average value processing removal pixel with
Machine noise obtains a noiseless picture.It can be used such as under type: taking the pixel value of the pixel of the same position of n picture
It is added, then divided by n, obtains pixel value mean value, the new picture for taking the corresponding pixel of pixel value mean value to constitute is made an uproar as nothing
Sound spectrogram piece can obtain removing the random noise of the pixel by mean filter in this way, finally obtain a pure noiseless
Picture A.
Step S103: intermediate dynamic process number of pictures under short exposure mode is obtained, the photo of acquisition is carried out at mean value
Reason, obtains a noise picture.It can be used such as under type: the dynamic pilot process during all short exposures, such as the 15th
, or intermediate a few Zhang Jinhang average value processings, finally obtain a noise picture B.
In other embodiments, it can also be carried out using the median or fitting algorithm etc. for choosing same pixel
Average value processing noise, does not do any restrictions.
After obtaining noiseless picture A and noise picture B, executes step S104: inputting the noiseless picture and noise pattern
Piece to neural network unit carries out deep learning training, obtains final picture.
In the present embodiment, the picture after input deep learning is trained is further comprised the steps of: to image signal processing unit
(ISP) it is further processed, obtains a noiseless picture, described be further processed includes: white balance, tone mapping, brightness tune
Whole and demosaicing.
Step S105: the noise figure of final picture is obtained.The difference is the pixel that final picture subtracts noiseless picture A
Difference.
Step S106: the difference of the noise picture and noiseless picture is obtained.The difference is noise picture B and nothing is made an uproar
The pixel value difference that sound spectrogram piece A each pair of point point subtracts each other.
Step S107: model is asked according to the difference of the noise figure of the final picture and the noise picture and noiseless picture
Number.Can be used such as under type: the noise figure of the final picture subtracts the difference of the noise picture and noiseless picture, obtains
To value L, value L's square is L2 norm.
Step S108: neural network unit is adjusted according to the norm.In the present embodiment, it can be used such as under type:
When neural network module training, the parameter w that algorithm can adjust every layer to lose it is as small as possible, since data are deposited
In many interference or noise, it is easy to produce over-fitting, the effect for causing network to predict training data is preferable, and to survey
The prediction effect of examination and verify data is poor.
It turns out that decision surface is more complicated under identical network structure, the value of parameter w is often bigger, and w was more as a child,
Obtained decision surface is relatively gentle.L2 regularization is that L2 regularization term is increased in original loss function, defeated in optimization network
Also to allow w small as far as possible when error with true value out.
Therefore neural network unit can be adjusted according to the norm until reaching goal-selling.
The photo exposed under short exposure mode is obtained, the photo after choosing exposure carries out average value processing, obtains noiseless figure
Piece chooses the number of pictures of dynamic process under exposure mode, carries out average value processing to the photo of acquisition, obtains a noise pattern
Piece, the training material obtained by this kind of mode is closer to the noise in real life, by the noiseless picture and noise of acquisition
Picture carries out deep learning training as training material input neural network unit, obtains final picture, obtains final picture
Noise figure (theoretical noise value), then the difference (actual noise value) of the noise picture Yu noiseless picture is obtained, according to described
The noise figure and the noise picture of final picture and the difference of noiseless picture ask norm (i.e. according to theoretical noise value and reality
The difference of noise figure seeks norm), norm is smaller to be illustrated neural network unit noise reduction accuracy higher (i.e. theoretical noise value is closer
Actual noise value), therefore neural network unit can be adjusted according to norm, or even can be by successive ignition training, until adjustment nerve
Network unit is to goal-selling.Neural network after being adjusted can remove spot block distortion, color in real image well
It each noise like such as makes an uproar, promotes picture degree of purity.
Referring to Fig. 2, in the present embodiment, a kind of specific embodiment storing equipment 200 is as follows:
A kind of storage equipment 200, wherein being stored with instruction set, described instruction collection is for executing: obtaining under short exposure mode
The n of exposure photos；To each pixel of the n after short exposure photos, carry out random on average value processing removal pixel
Noise obtains a noiseless picture；Intermediate dynamic process number of pictures under short exposure mode is obtained, the photo of acquisition is carried out
Average value processing obtains a noise picture；It inputs the noiseless picture and noise picture to neural network unit and carries out depth
Learning training obtains final picture；Obtain the noise figure of final picture；Obtain the difference of the noise picture and noiseless picture
Value；Norm is sought according to the difference of the noise figure of the final picture and the noise picture and noiseless picture；According to the model
Number adjustment neural network unit.It can be used such as under type:
The n exposed under short exposure mode photo can be uploaded by local, or obtain the short exposure mould that remote terminal is sent
The n exposed under formula photos, or the n photos exposed under short exposure mode are obtained from server.Wherein in present embodiment
In, the short exposure mode of camera shooting can be used, under the stabilization of tripod or other devices, in a short time with identical
Time for exposure exposes n picture, and n depends on scene brightness, and n is more than or equal to 30 under normal circumstances.
Further, described instruction collection is also used to execute: the n value size increases with the reduction of scene brightness, presets n
Being worth minimum value is 15, maximum value 60.
Further, described instruction collection is also used to execute: described " each pixel of photo is opened to the n after short exposure, into
Row average value processing removes the random noise on pixel, obtains a noiseless picture ", it further comprises the steps of: and takes the same of n picture
The pixel value of the pixel of one position is added, then divided by n, is obtained pixel value mean value, taken the corresponding pixel of pixel value mean value
The new picture that point is constituted is as noiseless picture.It can obtain removing the random noise of the pixel by mean filter in this way,
Finally obtain a pure noiseless picture A.
At the dynamic pilot process during all short exposures, such as the 15th, or intermediate a few Zhang Jinhang mean values
Reason finally obtains a noise picture B.
In other embodiments, it can also be carried out using the median or fitting algorithm etc. for choosing same pixel
Average value processing noise, does not do any restrictions.
When neural network module training, the parameter w that algorithm can adjust every layer to lose it is as small as possible, since data are deposited
In many interference or noise, it is easy to produce over-fitting, the effect for causing network to predict training data is preferable, and to survey
The prediction effect of examination and verify data is poor.
It turns out that decision surface is more complicated under identical network structure, the value of parameter w is often bigger, and w was more as a child,
Obtained decision surface is relatively gentle.L2 regularization is that L2 regularization term is increased in original loss function, defeated in optimization network
Also to allow w small as far as possible when error with true value out.
Therefore neural network unit can be adjusted according to the norm until reaching goal-selling.
Further, described instruction collection is also used to execute: described " to input the noiseless picture and noise picture to nerve
Network unit carries out deep learning training, obtains final picture ", the picture after further comprising the steps of: input deep learning training is extremely
Image signal processing unit is further processed, and obtains a noiseless picture, described be further processed includes: white balance, tone
Mapping, brightness adjustment and demosaicing.
Further, described instruction collection is also used to execute: described " according to the noise figure of the final picture and the noise
The difference of picture and noiseless picture seeks norm ", the noise figure of the final picture is further comprised the steps of:, the noise pattern is subtracted
The difference of piece and noiseless picture, obtains value L, and value L's square is L2 norm.
It by the instruction set in storage equipment 200, executes to give an order: obtaining the photo exposed under short exposure mode, choosing
Photo after taking exposure carries out average value processing, obtains noiseless picture, chooses the number of pictures of dynamic process under exposure mode, right
The photo of acquisition carries out average value processing, obtains a noise picture, the training material obtained by this kind of mode is closer to reality
Noise in life carries out depth using the noiseless picture of acquisition and noise picture as training material input neural network unit
Learning training obtains final picture, obtains the noise figure (theoretical noise value) of final picture, then obtains the noise picture and nothing
The difference (actual noise value) of noise picture, according to the noise figure of the final picture and the noise picture and noiseless picture
Difference ask norm (norm is asked according to the difference of theoretical noise value and actual noise value), norm is smaller to illustrate neural network list
First noise reduction accuracy is higher (i.e. theoretical noise value is closer to actual noise value), therefore can adjust neural network unit according to norm,
It even can be by successive ignition training, until adjusting neural network unit to goal-selling.Neural network after being adjusted can
Spot block distortion, color in real image are removed well each noise like such as to make an uproar, and promote picture degree of purity.
It should be noted that being not intended to limit although the various embodiments described above have been described herein
Scope of patent protection of the invention.Therefore, it based on innovative idea of the invention, change that embodiment described herein is carried out and is repaired
Change, or using equivalent structure or equivalent flow shift made by description of the invention and accompanying drawing content, it directly or indirectly will be with
Upper technical solution is used in other related technical areas, is included within scope of patent protection of the invention.
Claims (10)
1. a kind of image noise reduction processing method, which is characterized in that comprising steps of
Obtain the n photos exposed under short exposure mode；
To each pixel of the n after short exposure photos, the random noise on average value processing removal pixel is carried out, obtains one
Open noiseless picture；
Intermediate dynamic process number of pictures under short exposure mode is obtained, average value processing is carried out to the photo of acquisition, one is obtained and makes an uproar
Sound spectrogram piece；
It inputs the noiseless picture and noise picture to neural network unit and carries out deep learning training, obtain final picture；
Obtain the noise figure of final picture；
Obtain the difference of the noise picture and noiseless picture；
Norm is sought according to the difference of the noise figure of the final picture and the noise picture and noiseless picture；
Neural network unit is adjusted according to the norm.
2. a kind of image noise reduction processing method according to claim 1, which is characterized in that
Described " to each pixel of the n after short exposure photos, the random noise on average value processing removal pixel is carried out,
Obtain a noiseless picture ", it further comprises the steps of:
It takes the pixel value of the pixel of the same position of n picture to be added, then divided by n, obtains pixel value mean value, capture element
It is worth the new picture of the corresponding pixel composition of mean value as noiseless picture.
3. a kind of image noise reduction processing method according to claim 1, which is characterized in that
It is described " to input the noiseless picture and noise picture to neural network unit and carry out deep learning training, obtain final
Picture " further comprises the steps of:
Picture to image signal processing unit after input deep learning training is further processed, and obtains a noiseless picture,
Described be further processed includes: white balance, tone mapping, brightness adjustment and demosaicing.
4. a kind of image noise reduction processing method according to claim 1, which is characterized in that
Described " seeking norm according to the difference of the noise figure of the final picture and the noise picture and noiseless picture ", also wraps
Include step:
The noise figure of the final picture subtracts the difference of the noise picture and noiseless picture, obtains value L, value L's
Square be L2 norm.
5. a kind of image noise reduction processing method according to claim 1, which is characterized in that
The n value size increases with the reduction of scene brightness, and presetting n value minimum value is 15, maximum value 60.
6. a kind of storage equipment, wherein being stored with instruction set, which is characterized in that described instruction collection is for executing:
Obtain the n photos exposed under short exposure mode；
To each pixel of the n after short exposure photos, the random noise on average value processing removal pixel is carried out, obtains one
Open noiseless picture；
Intermediate dynamic process number of pictures under short exposure mode is obtained, average value processing is carried out to the photo of acquisition, one is obtained and makes an uproar
Sound spectrogram piece；
It inputs the noiseless picture and noise picture to neural network unit and carries out deep learning training, obtain final picture；
Obtain the noise figure of final picture；
Obtain the difference of the noise picture and noiseless picture；
Norm is sought according to the difference of the noise figure of the final picture and the noise picture and noiseless picture；
Neural network unit is adjusted according to the norm.
7. a kind of storage equipment according to claim 6, which is characterized in that described instruction collection is also used to execute:
Described " to each pixel of the n after short exposure photos, the random noise on average value processing removal pixel is carried out,
Obtain a noiseless picture ", it further comprises the steps of:
It takes the pixel value of the pixel of the same position of n picture to be added, then divided by n, obtains pixel value mean value, capture element
It is worth the new picture of the corresponding pixel composition of mean value as noiseless picture.
8. a kind of storage equipment according to claim 6, which is characterized in that described instruction collection is also used to execute:
It is described " to input the noiseless picture and noise picture to neural network unit and carry out deep learning training, obtain final
Picture " further comprises the steps of:
Picture to image signal processing unit after input deep learning training is further processed, and obtains a noiseless picture,
Described be further processed includes: white balance, tone mapping, brightness adjustment and demosaicing.
9. a kind of storage equipment according to claim 6, which is characterized in that described instruction collection is also used to execute:
Described " seeking norm according to the difference of the noise figure of the final picture and the noise picture and noiseless picture ", also wraps
Include step:
The noise figure of the final picture subtracts the difference of the noise picture and noiseless picture, obtains value L, value L's
Square be L2 norm.
10. a kind of storage equipment according to claim 6, which is characterized in that described instruction collection is also used to execute:
The n value size increases with the reduction of scene brightness, and presetting n value minimum value is 15, maximum value 60.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201811129933.5A CN109146824A (en) | 2018-09-27 | 2018-09-27 | A kind of image noise reduction processing method and a kind of storage equipment |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201811129933.5A CN109146824A (en) | 2018-09-27 | 2018-09-27 | A kind of image noise reduction processing method and a kind of storage equipment |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN109146824A true CN109146824A (en) | 2019-01-04 |
Family
ID=64812748
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201811129933.5A Pending CN109146824A (en) | 2018-09-27 | 2018-09-27 | A kind of image noise reduction processing method and a kind of storage equipment |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN109146824A (en) |
Cited By (7)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN110119704A (en) * | 2019-05-08 | 2019-08-13 | 武汉大学 | A kind of text based on depth residual error network is revealed the exact details phenomenon minimizing technology |
| 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 |
| CN110517204A (en) * | 2019-08-30 | 2019-11-29 | 京东方科技集团股份有限公司 | A kind of noise cancellation method and device, detector of X-ray detector |
| CN111583144A (en) * | 2020-04-30 | 2020-08-25 | 深圳市商汤智能传感科技有限公司 | Image noise reduction method and device, electronic equipment and storage medium |
| WO2020192483A1 (en) * | 2019-03-25 | 2020-10-01 | 华为技术有限公司 | Image display method and device |
| WO2020207239A1 (en) * | 2019-04-09 | 2020-10-15 | Guangdong Oppo Mobile Telecommunications Corp., Ltd. | Method and apparatus for image processing |
| GB2585232A (en) * | 2019-07-04 | 2021-01-06 | Apical Ltd | Image data pre-processing for neural networks |
Citations (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN103839234A (en) * | 2014-02-21 | 2014-06-04 | 西安电子科技大学 | Double-geometry nonlocal average image denoising method based on controlled nuclear |
| US20160321523A1 (en) * | 2015-04-30 | 2016-11-03 | The Regents Of The University Of California | Using machine learning to filter monte carlo noise from images |
| CN106408522A (en) * | 2016-06-27 | 2017-02-15 | 深圳市未来媒体技术研究院 | Image de-noising method based on convolution pair neural network |
-
2018
- 2018-09-27 CN CN201811129933.5A patent/CN109146824A/en active Pending
Patent Citations (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN103839234A (en) * | 2014-02-21 | 2014-06-04 | 西安电子科技大学 | Double-geometry nonlocal average image denoising method based on controlled nuclear |
| US20160321523A1 (en) * | 2015-04-30 | 2016-11-03 | The Regents Of The University Of California | Using machine learning to filter monte carlo noise from images |
| CN106408522A (en) * | 2016-06-27 | 2017-02-15 | 深圳市未来媒体技术研究院 | Image de-noising method based on convolution pair neural network |
Non-Patent Citations (2)
| Title |
|---|
| JINGWEN CHEN 等: "Image Blind Denoising With Generative Adversarial Network Based Noise Modeling", 《HTTP://OPENACCESS.THECVF.COM/CONTENT_CVPR_2018/PAPERS/CHEN_IMAGE_BLIND_DENOISING_CVPR_2018_PAPER.PDF》 * |
| 李倩: "基于GANs训练去噪深度神经网络实现了良好的图像盲去噪效果", 《HTTP://WWW.ELECFANS.COM/D/700802.HTML》 * |
Cited By (14)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US11882357B2 (en) | 2019-03-25 | 2024-01-23 | Huawei Technologies Co., Ltd. | Image display method and device |
| WO2020192483A1 (en) * | 2019-03-25 | 2020-10-01 | 华为技术有限公司 | Image display method and device |
| US11403740B2 (en) | 2019-04-09 | 2022-08-02 | Guangdong Oppo Mobile Telecommunications Corp., Ltd. | Method and apparatus for image capturing and processing |
| WO2020207239A1 (en) * | 2019-04-09 | 2020-10-15 | Guangdong Oppo Mobile Telecommunications Corp., Ltd. | Method and apparatus for image processing |
| CN110119704A (en) * | 2019-05-08 | 2019-08-13 | 武汉大学 | A kind of text based on depth residual error network is revealed the exact details phenomenon minimizing technology |
| 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 |
| 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 |
| GB2585232A (en) * | 2019-07-04 | 2021-01-06 | Apical Ltd | Image data pre-processing for neural networks |
| GB2585232B (en) * | 2019-07-04 | 2021-12-08 | Apical Ltd | Image data pre-processing for neural networks |
| US11816813B2 (en) | 2019-07-04 | 2023-11-14 | Arm Limited | Image data pre-processing for neural networks |
| CN110517204B (en) * | 2019-08-30 | 2022-05-20 | 京东方科技集团股份有限公司 | Noise elimination method and device of X-ray detector and detector |
| CN110517204A (en) * | 2019-08-30 | 2019-11-29 | 京东方科技集团股份有限公司 | A kind of noise cancellation method and device, detector of X-ray detector |
| CN111583144A (en) * | 2020-04-30 | 2020-08-25 | 深圳市商汤智能传感科技有限公司 | Image noise reduction method and device, electronic equipment and storage medium |
| CN111583144B (en) * | 2020-04-30 | 2023-08-25 | 深圳市商汤智能传感科技有限公司 | Image noise reduction method and device, electronic equipment and storage medium |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN109146824A (en) | A kind of image noise reduction processing method and a kind of storage equipment | |
| CN111885295B (en) | A shooting method, device and equipment | |
| WO2020152521A1 (en) | Systems and methods for transforming raw sensor data captured in low-light conditions to well-exposed images using neural network architectures | |
| CN107464217B (en) | An image processing method and device | |
| JP2020144489A (en) | Image processing method, image processing device, program, method of producing trained model, and image processing system | |
| CN110619593A (en) | Double-exposure video imaging system based on dynamic scene | |
| CN109391775A (en) | A kind of intelligent shooting tripod head control method and system based on recognition of face | |
| CN105141841B (en) | Picture pick-up device and its method | |
| DE102020200310A1 (en) | Method and system for reducing haze for image processing | |
| CN114511487B (en) | Image fusion method and device, computer readable storage medium, and terminal | |
| Jaroensri et al. | Generating training data for denoising real rgb images via camera pipeline simulation | |
| CN111144491B (en) | Image processing method, device and electronic system | |
| CN110390646B (en) | A detail-preserving image denoising method | |
| WO2023081399A1 (en) | Integrated machine learning algorithms for image filters | |
| Chen et al. | Blind de-convolution of images degraded by atmospheric turbulence | |
| KR20200140713A (en) | Method and apparatus for training neural network model for enhancing image detail | |
| CN117351216B (en) | Image self-adaptive denoising method based on supervised deep learning | |
| CN120092246A (en) | Neural network training method and device, image processing method and device | |
| CN116416175A (en) | Image fusion method based on self-adaptive edge-preserving smooth pyramid | |
| CN111612707A (en) | A Neural Network Image Denoising Method Based on Wavelet Transform | |
| Chen et al. | CERL: A unified optimization framework for light enhancement with realistic noise | |
| Ma et al. | Low-Light Image Enhancement using Retinex-based Network with Attention Mechanism. | |
| CN112651911A (en) | High dynamic range imaging generation method based on polarization image | |
| Li et al. | Dualdn: Dual-domain denoising via differentiable isp | |
| CN111489300A (en) | Screen image de-moiré method based on unsupervised learning |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| PB01 | Publication | ||
| PB01 | Publication | ||
| SE01 | Entry into force of request for substantive examination | ||
| SE01 | Entry into force of request for substantive examination | ||
| RJ01 | Rejection of invention patent application after publication |
Application publication date: 20190104 |
|
| RJ01 | Rejection of invention patent application after publication |