CN108447036A - A kind of low light image Enhancement Method based on convolutional neural networks - Google Patents
A kind of low light image Enhancement Method based on convolutional neural networks Download PDFInfo
- Publication number
- CN108447036A CN108447036A CN201810246429.7A CN201810246429A CN108447036A CN 108447036 A CN108447036 A CN 108447036A CN 201810246429 A CN201810246429 A CN 201810246429A CN 108447036 A CN108447036 A CN 108447036A
- Authority
- CN
- China
- Prior art keywords
- image
- low light
- characteristic
- convolutional layer
- branch process
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Pending
Links
- 238000000034 method Methods 0.000 title claims abstract description 76
- 238000013527 convolutional neural network Methods 0.000 title claims abstract description 25
- 230000008569 process Effects 0.000 claims abstract description 33
- 238000013528 artificial neural network Methods 0.000 claims abstract description 16
- 238000012545 processing Methods 0.000 claims description 14
- 230000004913 activation Effects 0.000 claims description 7
- 238000003475 lamination Methods 0.000 claims description 3
- 230000000694 effects Effects 0.000 abstract description 10
- 230000002708 enhancing effect Effects 0.000 abstract description 10
- 238000005457 optimization Methods 0.000 abstract description 5
- 230000001965 increasing effect Effects 0.000 abstract description 3
- 230000010354 integration Effects 0.000 abstract description 2
- 230000006870 function Effects 0.000 description 20
- 238000005286 illumination Methods 0.000 description 8
- 238000012549 training Methods 0.000 description 6
- 238000005516 engineering process Methods 0.000 description 5
- 238000011156 evaluation Methods 0.000 description 4
- 238000012360 testing method Methods 0.000 description 3
- 230000008901 benefit Effects 0.000 description 2
- 230000008859 change Effects 0.000 description 2
- 230000008878 coupling Effects 0.000 description 2
- 238000010168 coupling process Methods 0.000 description 2
- 238000005859 coupling reaction Methods 0.000 description 2
- 238000010586 diagram Methods 0.000 description 2
- 239000003595 mist Substances 0.000 description 2
- 238000011160 research Methods 0.000 description 2
- 241000208340 Araliaceae Species 0.000 description 1
- 235000005035 Panax pseudoginseng ssp. pseudoginseng Nutrition 0.000 description 1
- 235000003140 Panax quinquefolius Nutrition 0.000 description 1
- 230000003044 adaptive effect Effects 0.000 description 1
- 238000004458 analytical method Methods 0.000 description 1
- 238000004364 calculation method Methods 0.000 description 1
- 238000000205 computational method Methods 0.000 description 1
- 230000007812 deficiency Effects 0.000 description 1
- 230000006866 deterioration Effects 0.000 description 1
- 238000011161 development Methods 0.000 description 1
- 238000000605 extraction Methods 0.000 description 1
- 239000004744 fabric Substances 0.000 description 1
- 235000008434 ginseng Nutrition 0.000 description 1
- 238000003384 imaging method Methods 0.000 description 1
- 235000013372 meat Nutrition 0.000 description 1
- 238000012544 monitoring process Methods 0.000 description 1
- 238000003062 neural network model Methods 0.000 description 1
- 238000012805 post-processing Methods 0.000 description 1
- 238000007781 pre-processing Methods 0.000 description 1
- 230000000750 progressive effect Effects 0.000 description 1
- 238000005070 sampling Methods 0.000 description 1
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T5/00—Image enhancement or restoration
- G06T5/73—Deblurring; Sharpening
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T2207/00—Indexing scheme for image analysis or image enhancement
- G06T2207/20—Special algorithmic details
- G06T2207/20081—Training; Learning
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T2207/00—Indexing scheme for image analysis or image enhancement
- G06T2207/20—Special algorithmic details
- G06T2207/20084—Artificial neural networks [ANN]
Landscapes
- Physics & Mathematics (AREA)
- General Physics & Mathematics (AREA)
- Engineering & Computer Science (AREA)
- Theoretical Computer Science (AREA)
- Image Processing (AREA)
- Image Analysis (AREA)
Abstract
Low light image Enhancement Method of the present invention based on convolutional neural networks, includes the following steps：Then the input feature vector image first in convolution module is handled by several convolution modules.In each convolution module, first time branch process is carried out to characteristic image all under several convolutional layer paths, and the first time branch process data after integration are subjected to second of branch process, obtain the output of convolution module.It after the last one convolution module, is calculated using convolutional layer, obtains the output of neural network, the difference of the image data and real image of network output is calculated using loss function, optimizes the parameter of convolutional neural networks according to difference.The present invention realizes the enhancing of low light image using convolutional neural networks, and constrained optimization is carried out to the parameter in neural network using loss function, reach the desired brightness for increasing image, contrast, the effect of image subjectivity aesthetic feeling is promoted, and largely keeps the original structure of image, detailed information.
Description
Technical field
The present invention relates to technical field of image processing more particularly to a kind of low light image increasings based on convolutional neural networks
Strong method.
Background technology
How the enhancing of low light image is mainly studied by computer progress operation, passes through the tune of certain parameter
Section, handles original image, promotes the characteristics such as brightness, the contrast of image, and keeps the texture of object in artwork, knot
Structure information, to meet people in each area research, the needs used.
With popularizing for digital product, the first-class equipment of camera, mobile phone, monitoring camera can very easily collect various
The image information of various kinds.The environment that acquisition obtains image information is varied.Different light are influenced under by varying environment, are shot
Brightness, the contrast of image out etc. are also different.In real life, a considerable amount of images are in insufficient light
Either night is shot, since image capture device fails to collect enough photons, imaging of the object in picture
A kind of feeling under-exposed, contrast is too low of people will be given.The image taken in dark conditions can generally also be introduced and be made an uproar
Sound, such as poisson noise can make image generation further degrade.
Therefore, brightness, the contrast for how improving image, enhance the image got under low illumination, make it in human eye and
Other field is attained by desired effect when applying, and is always the research hotspot of image processing field.Use image enhancement pair
Image is pre-processed, the common means of the effect and image processing field to promote some post analysis algorithms.With
The development of image processing techniques and the promotion of computer computation ability, the enhancing technology of low light image are these in insufficient light
Under conditions of the processing of image that captures provide good solution, can be in the line for keeping original image as far as possible
On the basis of reason, structural information, the features such as brightness, the contrast of image are improved, keep image more attractive, can be used as a kind of pre-
Processing method meets the needs that later stage other application uses.
The method of existing low light image enhancing is broadly divided into four major class.The first kind is the figure based on Retinex theories
Image intensifying technology.Second class is the image enhancement technique based on histogram equalization.Third class is the statistical property using image
The image enhancement technique of dark channel prior.4th class is the image enhancement technique based on neural network.
Based on the image enhancement technique of Retinex theories, the model in Retinex theories is utilized, i.e. piece image is
It is made of the reflection of illumination and object, is calculated according to existing image, estimate the light conditions in image, illumination is moved
Except the influence that can be removed due to image deterioration caused by illumination difference.When this method is applied in low light image, by
It is influenced caused by illumination deficiency in eliminating, so that it may to obtain the effect of image enhancement.
Method based on histogram equalization, it will usually which rough thinks that the histogram of normal illumination hypograph is uniformly to divide
Cloth, the numerical value of each pixel in image is redistributed using this statistical property of image.Each pixel
The numerical value of pixel replaces with meet the new value of statistical property after, you can obtain enhanced image.
Using the image enhancement technique of dark channel prior, uses for reference in the enhancing processing with mist image, utilized normal light
The characteristic for tending to 0 according to the dark numerical value of the image of lower shooting, in actually calculating it is estimated that in atmospherical scattering model
Parameters numerical value, substituting into parameters can calculate.With defogging algorithm the difference is that using dark channel prior to low light
When being enhanced according to image, the image of low illumination is first subjected to reverse phase, so that region gloomy in artwork becomes to whiten, meat
There is the effect with mist up soon, to carry out defogging.Reverse phase obtains enhanced figure again after the completion of the operation of defogging
Picture.
Image enhancement technique based on neural network is that Lore proposed that this method uses own coding in 2016
Device, image is encoded by self-encoding encoder, then is decoded, and after the calculating of parameter in layer, is obtained enhanced
Export image.The characteristics of this process employs study class methods, the output of neural network is constrained using common L2.
The method of existing non-study class, calculation amount is low, and arithmetic speed is fast, but the poor effect enhanced, usually can band
Carry out some bad results such as cross-color, the image supersaturation of image.Method based on study has certain complexity, this
Point can be accelerated by the parallel optimization of GPU, but this method can not utilize the characteristic information in image well at present,
Adequately retain the information such as texture, the structure in original image.
Invention content
The low light image Enhancement Method based on convolutional neural networks that the present invention provides a kind of, realizes low light image
Enhancing, while the structural informations such as edge, texture for keeping objects in images as far as possible.
To achieve the goals above, this invention takes following technical solutions：
The low light image Enhancement Method based on convolutional neural networks that the present invention provides a kind of, includes the following steps：
S1：The characteristic image for extracting real image, inputs the characteristic image, to described under several convolutional layer paths
Characteristic image carries out first time branch process, and after the first time branch process, by several convolutional layer roads
Handling result under diameter is integrated together to obtain characteristic image 1；
S2：Second of branch process is carried out to the characteristic image 1 under two convolutional layer paths, and at described second
After branch process, it is integrated together the handling result under two convolutional layer paths to obtain characteristic image 2；
S3：The difference that the characteristic image 2 and real image are calculated using setting loss function, is optimized according to the difference
The parameter of convolutional neural networks, and export and obtain enhanced image.
Further, the first time branch process is handled with second of branch process as one group of convolution module,
One or more groups of convolution module processing are carried out according to the characteristic information of input feature vector image.
Further, in the first time branch process, the quantity in convolutional layer path is according to the spy of input feature vector image
Reference breath is set, and the convolution layer number under every convolutional layer path is set according to the characteristic information of input feature vector image
It is fixed.
Further, in second of branch process, including two convolutional layer paths, wherein one is not any behaviour
Make, another sets there are two convolutional layer.
Further, the loss function is denoted as L, is：
L=λ leclidean+(1-λ)lSSIM
Wherein, λ indicates the weight between Euclidean distance loss function and structural similarity error.
Further, when the characteristic image of input is gray level image, in the output of last layer of convolutional layer of neural network
Afterwards, average pond layer processing is added.
Further, each convolutional layer matches an activation primitive layer, the characteristic image of the extraction real image
Including：
Real image is pre-processed to obtain characteristic image by lamination and activation primitive layer, then again by the characteristic pattern
As input convolution module is handled.
As seen from the above technical solution provided by the invention, the present invention is different from traditional image modeling either profit
With the method for image statistics, image prior, low light image is handled end to end using full convolutional neural networks,
The parameter of convolutional neural networks is constantly updated during training learns.With the side of the same self-encoding encoder using neural network
Method is compared, and the size of the characteristic pattern of image will not reduce during calculating, is avoided image information in down-sampling and is lost
The problem of, it can preferably keep the original structure of image, the information of texture.The present invention uses common PSNR indexs and SSIM
When index carries out evaluation test, show optimal.The evaluation indexes such as LOE, the SNM that can be also used in being enhanced using low light image
When being tested, method test data of the invention can also keep larger performance advantage compared to other algorithms.
The present invention, in practical application, only needing to input a low light image, can be obtained using simply in output end
To enhanced effect.During the use of model, intermediate calculating process need not adjust any parameter, so that it may with needle
Carrying out adaptive adjustment using the information of Image Multiscale to the different zones of image enhances.Still due to enhanced image
Can keep structure, the texture information of original image, and using simple, not just on the aesthetic measure of subjective assessment performance compared with
It is excellent, the guarantee of enough facilities and performance is also capable of providing when subsequently using other parsers.
The additional aspect of the present invention and advantage will be set forth in part in the description, these will become from the following description
Obviously, or practice through the invention is recognized.
Description of the drawings
In order to illustrate the technical solution of the embodiments of the present invention more clearly, required use in being described below to embodiment
Attached drawing be briefly described, it should be apparent that, drawings in the following description are only some embodiments of the invention, for this
For the those of ordinary skill of field, without having to pay creative labor, other are can also be obtained according to these attached drawings
Attached drawing.
Fig. 1 is the flow chart of the low light image Enhancement Method provided in an embodiment of the present invention；
Fig. 2 is the structural schematic diagram of convolution module described in one group provided in an embodiment of the present invention.
Specific implementation mode
Embodiments of the present invention are described below in detail, the example of the embodiment is shown in the accompanying drawings, wherein from beginning
Same or similar element or element with the same or similar functions are indicated to same or similar label eventually.Below by ginseng
The embodiment for examining attached drawing description is exemplary, and is only used for explaining the present invention, and is not construed as limiting the claims.
Those skilled in the art of the present technique are appreciated that unless expressly stated, singulative " one " used herein, " one
It is a ", " described " and "the" may also comprise plural form.It is to be further understood that is used in the specification of the present invention arranges
It refers to there are the feature, integer, step, operation, element and/or component, but it is not excluded that presence or addition to take leave " comprising "
Other one or more features, integer, step, operation, element, component and/or their group.It should be understood that when we claim member
Part is " connected " or when " coupled " to another element, it can be directly connected or coupled to other elements, or there may also be
Intermediary element.In addition, " connection " used herein or " coupling " may include being wirelessly connected or coupling.Wording used herein
"and/or" includes any cell of one or more associated list items and all combines.
Those skilled in the art of the present technique are appreciated that unless otherwise defined, all terms used herein (including technology art
Language and scientific terminology) there is meaning identical with the general understanding of the those of ordinary skill in fields of the present invention.Should also
Understand, those terms such as defined in the general dictionary, which should be understood that, to be had and the meaning in the context of the prior art
The consistent meaning of justice, and unless defined as here, will not be with idealizing or the meaning of too formal be explained.
The present invention provides a kind of low light image Enhancement Method based on convolutional neural networks, and this method, which is that one kind is special, to be set
The convolutional neural networks of structure are counted, and using a kind of loss function that can retain image texture and details in neural network
Parameter carries out constrained optimization, reaches the desired brightness for increasing image, contrast, promotes the effect of the subjective aesthetic feeling of image,
And largely keep the original structure of image, detailed information.
The solution details of the structure of the convolutional neural networks are as follows：
(1) special convolution module
Convolutional neural networks model can encounter the problem of gradient disappears, to influence neural network during training
Trained convergence process.In order to avoid the appearance of this problem, in conjunction with existing preferably network structure ResNet and
GoogLeNet, devises a kind of special convolution module, and as shown in the figure 2.
In order to realize modular operation, the port number of input, the output of the convolution module is consistent, to facilitate module in model
Stacking.
After characteristic image input, branch operation twice can be passed through, wherein：
In first time branch operation, image can pass through multichannel process, when the processing of different paths calculates, data by convolution
The quantity of layer differs.According to the calculating of convolution operation and receptive field it is recognised that this multichannel is the equal of right under different scale
The image of input is handled.After first time branch operation, the result that two-way calculates uses the side for connecting (concatenate)
Formula is stacked to together, followed by second of branch operation.
In second of branch operation, image can equally be handled by two lines, wherein do not do any operation all the way,
In addition it can be calculated all the way by two convolutional layers.After the completion of calculating, the results added of two-way obtains the output of module.
(2) present invention proposes a kind of new loss function, as follows：
In every application using depth convolutional neural networks, such as the super-resolution of image, in image denoising sound, use
Loss function is still the Euclidean distance between the output and actual value (ground truth) by calculating network and obtains
's.The neural network being trained using this loss function, desired network output is all consistent with ground truth, however
In image enhancement problem, in fact, many times there are one unified evaluation indexes.In the case of not overexposure, institute
Some image pixel values increase or reduce the numerical value of a very little, are all acceptables for image enhancement.Therefore, single
The pure application for being not suitable for low light image enhancing this respect using Euclidean distance.
In the algorithm of low light image enhancing, more concerned be image texture and minutia protected either with or without good
It handles.Therefore, the SSIM in common evaluation index is relatively suitble to this item application.For the pixel p in image, it
The computational methods of SSIM numerical value are as follows：
In formula, μxAnd μyRespectively represent the mean value of the image and true picture of image network output, σxAnd σyRepresentative image
The variance of the image and true picture of network output, C1 and C2 are constants.
The numerical value of SSIM be typically section (0,1] in, and after acquiring the SSIM numerical value of full figure, the numerical value of usual SSIM
It is bigger, it is meant that image is closer to desired value.Therefore, in the training process of neural network, using 1-SSIM as model
Trained loss function.Therefore, the calculating formula using SSIM as loss function is as follows：
The luminance information of texture, structural information and image in order to balance, it is total using original Euclidean distance and SSIM
Parameter with constraint network can obtain better effect, and therefore, the present invention proposes that a kind of novel effective loss function comes about
The training of the network model of the low light image enhancing of beam.The formula of the loss function is as follows：
L=λ leclidean+(1-λ)lSSIM
Wherein, λ indicates common Euclidean distance loss function (namely root-mean-square error MSE) and structural similarity error
(SSIM) weight between；
(3) overall structure of model
Using above-mentioned described convolution module, neural network model end to end is formed, it is as follows：
First, image is handled by a convolutional layer and an activation primitive layer, the characteristic pattern to be standardized
Picture, in this way, behind characteristic image, so that it may be handled with the convolution module for using several to define.The result handled according to
It is so the characteristic pattern in the channel containing standardization.Finally this result is integrated using a convolutional neural networks, is calculated
Obtain final network output.For coloured image, the port number of last layer of neural network output result is 3；For gray scale
Image adds an average pond layer again below in output.In the training process, it is utilized between the output and real image of network
Loss function calculating difference optimizes the parameter of convolutional neural networks according to difference.During the test, low illumination is inputted
Dim image, output can directly obtain enhanced image effect.
For ease of the understanding of the present invention, it is further explained by taking several specific embodiments as an example below in conjunction with attached drawing
Illustrate, and each embodiment does not constitute the restriction to the embodiment of the present invention.
The low light image Enhancement Method based on convolutional neural networks that the present embodiment provides a kind of, as shown in Figure 1, include with
Lower step：
S1：The characteristic image for extracting real image, inputs the characteristic image, to described under several convolutional layer paths
Characteristic image carries out first time branch process, and after the first time branch process, by several convolutional layer roads
Handling result under diameter is integrated together to obtain characteristic image 1；
S2：Second of branch process is carried out to the characteristic image 1 under two convolutional layer paths, and at described second
After branch process, it is integrated together the handling result under two convolutional layer paths to obtain characteristic image 2；
S3：The difference that the characteristic image 2 and real image are calculated using setting loss function, is optimized according to the difference
The parameter of convolutional neural networks；
S4：Output obtains enhanced image.
Specially：
S1：Input needs the image enhanced, is handled input picture using a convolutional layer, obtains characteristic image,
This feature image is by the input as the convolution module of special designing；
S2：Characteristic image is handled using the convolution module of several designs, in each convolution module, there is two
Secondary branch operation, first time branch operation, the convolutional layer that characteristic image can all pass through different number are handled, the result of processing
Input of the direct splicing as second of branch operation, carries out two-way processing in next branch operation, and characteristic image is logical
Cross several convolutional layers and initial value and splice the output for obtaining convolution module, the output as next convolution module input or
Post-processing convolutional layer is output to carry out that the final output image of network is calculated.
S3：After the last one convolution module, obtained characteristic image is calculated by a convolutional layer, is obtained final
Network exports image；
S4：In convolutional neural networks network training, calculates network output image using setting loss function and scheme with practical
The difference of picture optimizes the parameter of convolutional neural networks according to the difference, when using the convolutional neural networks, directly uses net
The output image of network is as the result images after image enhancement.
In the present embodiment, the first time branch process and second of branch process as one group of convolution module at
Reason carries out one or more groups of convolution modules according to the characteristic information of input feature vector image and handles, uses in the present embodiment
One group of convolution module handles image.
In the first time branch process, the quantity in convolutional layer path is carried out according to the characteristic information of input feature vector image
It sets, the convolution layer number under every convolutional layer path is set according to the characteristic information of input feature vector image.
In the present embodiment, the first time branch process is three convolutional layer paths, wherein one sets there are three convolution
Layer, one sets there are two convolutional layer, and another sets that there are one convolutional layers.
In second of branch process described in the present embodiment, including two convolutional layer paths, wherein one do not do it is any
Operation, another sets there are two convolutional layer.
The loss function is denoted as L, is：
L=λ leclidean+(1-λ)lSSIM
Wherein, λ indicates common Euclidean distance loss function (namely root-mean-square error MSE) and structural similarity error
(SSIM) weight between；
In the present embodiment, when the characteristic image of input is gray level image, average pond is added after loss function calculating
Change layer processing.
In this embodiment, each convolutional layer matches an activation primitive layer, and real image is by lamination and activation letter
Several layers are pre-processed to obtain characteristic image, then again handle characteristic image input convolution module.
In conclusion the convolution module of neural network proposed by the present invention, is divided into two-step pretreatment in the inside modules, passes through
The integration of different path handling results, can make full use of the multi-scale information in image, also, this connection type can be
The trained convolutional neural networks that can be built rapidly using the module for image enhancement.The invention also proposed be directed to simultaneously
A kind of optimization method of image enhancement can make neural network during handling input picture using the loss function of proposition
More focus on the reservation of image texture, structural information.Using the convolution module and optimization method in the present invention, may be constructed one it is complete
The whole solution for being directed to the enhancing of low light image end to end.
One of ordinary skill in the art will appreciate that：Attached drawing is the schematic diagram of one embodiment, module in attached drawing or
Flow is not necessarily implemented necessary to the present invention.
As seen through the above description of the embodiments, those skilled in the art can be understood that the present invention can
It is realized by the mode of software plus required general hardware platform.Based on this understanding, technical scheme of the present invention essence
On in other words the part that contributes to existing technology can be expressed in the form of software products, the computer software product
It can be stored in a storage medium, such as ROM/RAM, magnetic disc, CD, including some instructions are used so that a computer equipment
(can be personal computer, server either network equipment etc.) executes the certain of each embodiment of the present invention or embodiment
Method described in part.
Each embodiment in this specification is described in a progressive manner, identical similar portion between each embodiment
Point just to refer each other, and each embodiment focuses on the differences from other embodiments.Especially for device or
For system embodiment, since it is substantially similar to the method embodiment, so describing fairly simple, related place is referring to method
The part of embodiment illustrates.Apparatus and system embodiment described above is only schematical, wherein the conduct
The unit that separating component illustrates may or may not be physically separated, the component shown as unit can be or
Person may not be physical unit, you can be located at a place, or may be distributed over multiple network units.It can root
According to actual need that some or all of module therein is selected to achieve the purpose of the solution of this embodiment.Ordinary skill
Personnel are without creative efforts, you can to understand and implement.
The foregoing is only a preferred embodiment of the present invention, but scope of protection of the present invention is not limited thereto,
Any one skilled in the art in the technical scope disclosed by the present invention, the change or replacement that can be readily occurred in,
It should be covered by the protection scope of the present invention.Therefore, protection scope of the present invention should be with scope of the claims
Subject to.
Claims (7)
1. a kind of low light image Enhancement Method based on convolutional neural networks, which is characterized in that include the following steps：
S1：The characteristic image for extracting real image, inputs the characteristic image, to the feature under several convolutional layer paths
Image carries out first time branch process, and after the first time branch process, will be under several convolutional layer paths
Handling result be integrated together to obtain characteristic image 1；
S2：Second of branch process is carried out to the characteristic image 1 under two convolutional layer paths, and in second of branch
After treatment is integrated together the handling result under two convolutional layer paths to obtain characteristic image 2；
S3：The difference that the characteristic image 2 and real image are calculated using setting loss function optimizes convolution according to the difference
The parameter of neural network, and export and obtain enhanced image.
2. low light image Enhancement Method according to claim 1, which is characterized in that
The first time branch process is handled with second of branch process as one group of convolution module, according to input feature vector figure
The characteristic information of picture carries out one or more groups of convolution module processing.
3. low light image Enhancement Method according to claim 2, which is characterized in that
In the first time branch process, the quantity in convolutional layer path is set according to the characteristic information of input feature vector image
Fixed, the convolution layer number under every convolutional layer path is set according to the characteristic information of input feature vector image.
4. low light image Enhancement Method according to claim 3, which is characterized in that in second of branch process
In, including two convolutional layer paths, wherein one is not done any operation, another sets that there are two convolutional layers.
5. low light image Enhancement Method according to claim 4, which is characterized in that the loss function is denoted as L, is：
Wherein, λ indicates the weight between Euclidean distance loss function and structural similarity error.
6. low light image Enhancement Method according to claim 5, which is characterized in that when the characteristic image of input is gray scale
When image, after the output of last layer of convolutional layer of neural network, average pond layer processing is added.
7. low light image Enhancement Method according to claim 6, which is characterized in that each convolutional layer matches one
Activation primitive layer, the characteristic image for extracting real image include：
Real image is pre-processed to obtain characteristic image by lamination and activation primitive layer, then again that the characteristic image is defeated
Enter convolution module to be handled.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201810246429.7A CN108447036A (en) | 2018-03-23 | 2018-03-23 | A kind of low light image Enhancement Method based on convolutional neural networks |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201810246429.7A CN108447036A (en) | 2018-03-23 | 2018-03-23 | A kind of low light image Enhancement Method based on convolutional neural networks |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN108447036A true CN108447036A (en) | 2018-08-24 |
Family
ID=63196885
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201810246429.7A Pending CN108447036A (en) | 2018-03-23 | 2018-03-23 | A kind of low light image Enhancement Method based on convolutional neural networks |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN108447036A (en) |
Cited By (26)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN109410129A (en) * | 2018-09-28 | 2019-03-01 | 大连理工大学 | A kind of method of low light image scene understanding |
| CN109462732A (en) * | 2018-10-29 | 2019-03-12 | 努比亚技术有限公司 | A kind of image processing method, equipment and computer readable storage medium |
| CN109509248A (en) * | 2018-09-28 | 2019-03-22 | 北京大学 | A kind of Photon Mapping rendering method neural network based and system |
| CN109785252A (en) * | 2018-12-25 | 2019-05-21 | 山西大学 | Based on multiple dimensioned residual error dense network nighttime image enhancing method |
| CN109859120A (en) * | 2019-01-08 | 2019-06-07 | 北京交通大学 | Image defogging method based on multiple dimensioned residual error network |
| CN110163808A (en) * | 2019-03-28 | 2019-08-23 | 西安电子科技大学 | A kind of single frames high dynamic imaging method based on convolutional neural networks |
| CN110232661A (en) * | 2019-05-03 | 2019-09-13 | 天津大学 | Low illumination colour-image reinforcing method based on Retinex and convolutional neural networks |
| CN110298792A (en) * | 2018-03-23 | 2019-10-01 | 北京大学 | Low light image enhancing and denoising method, system and computer equipment |
| CN110334807A (en) * | 2019-05-31 | 2019-10-15 | 北京奇艺世纪科技有限公司 | Training method, device, equipment and the storage medium of deep learning network |
| CN110675336A (en) * | 2019-08-29 | 2020-01-10 | 苏州千视通视觉科技股份有限公司 | Low-illumination image enhancement method and device |
| WO2020053837A1 (en) * | 2018-09-13 | 2020-03-19 | Spectrum Optix Inc. | Photographic underexposure correction using a neural network |
| WO2020062074A1 (en) * | 2018-09-28 | 2020-04-02 | Hangzhou Hikvision Digital Technology Co., Ltd. | Reconstructing distorted images using convolutional neural network |
| CN111028163A (en) * | 2019-11-28 | 2020-04-17 | 湖北工业大学 | Convolution neural network-based combined image denoising and weak light enhancement method |
| CN111046887A (en) * | 2018-10-15 | 2020-04-21 | 华北电力大学（保定） | Method for extracting characteristics of image with noise |
| CN111681178A (en) * | 2020-05-22 | 2020-09-18 | 厦门大学 | Knowledge distillation-based image defogging method |
| CN111928799A (en) * | 2020-09-25 | 2020-11-13 | 南京理工大学 | Three-dimensional measurement method for realizing stripe image contrast enhancement based on deep learning |
| CN111932471A (en) * | 2020-07-24 | 2020-11-13 | 山西大学 | Double-path exposure degree fusion network model and method for low-illumination image enhancement |
| CN111986240A (en) * | 2020-09-01 | 2020-11-24 | 交通运输部水运科学研究所 | Drowning person detection method and system based on visible light and thermal imaging data fusion |
| CN112446247A (en) * | 2019-08-30 | 2021-03-05 | 北京大学 | Low-illumination face detection method based on multi-feature fusion and low-illumination face detection network |
| WO2021063118A1 (en) * | 2019-10-02 | 2021-04-08 | Guangdong Oppo Mobile Telecommunications Corp., Ltd. | Method and apparatus for image processing |
| CN112686823A (en) * | 2020-12-31 | 2021-04-20 | 广西慧云信息技术有限公司 | Automatic image enhancement method based on illumination transformation network |
| CN112804943A (en) * | 2018-10-03 | 2021-05-14 | 株式会社岛津制作所 | Method for creating learning completion model, luminance adjustment method, and image processing apparatus |
| CN113066019A (en) * | 2021-02-27 | 2021-07-02 | 华为技术有限公司 | Image enhancement method and related device |
| CN113095470A (en) * | 2020-01-08 | 2021-07-09 | 字节跳动有限公司 | Neural network training method, image processing method and device, and storage medium |
| CN113256528A (en) * | 2021-06-03 | 2021-08-13 | 中国人民解放军国防科技大学 | Low-illumination video enhancement method based on multi-scale cascade depth residual error network |
| WO2023050731A1 (en) * | 2021-09-28 | 2023-04-06 | 深圳市中兴微电子技术有限公司 | Method for training image enhancement model, image enhancement method, and readable medium |
-
2018
- 2018-03-23 CN CN201810246429.7A patent/CN108447036A/en active Pending
Cited By (41)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN110298792B (en) * | 2018-03-23 | 2021-08-17 | 北京大学 | Low-illumination image enhancement and denoising method, system and computer equipment |
| CN110298792A (en) * | 2018-03-23 | 2019-10-01 | 北京大学 | Low light image enhancing and denoising method, system and computer equipment |
| US11076103B2 (en) | 2018-09-13 | 2021-07-27 | Spectrum Optix Inc. | Photographic underexposure correction using a neural network |
| WO2020053837A1 (en) * | 2018-09-13 | 2020-03-19 | Spectrum Optix Inc. | Photographic underexposure correction using a neural network |
| WO2020062074A1 (en) * | 2018-09-28 | 2020-04-02 | Hangzhou Hikvision Digital Technology Co., Ltd. | Reconstructing distorted images using convolutional neural network |
| CN109509248A (en) * | 2018-09-28 | 2019-03-22 | 北京大学 | A kind of Photon Mapping rendering method neural network based and system |
| CN109410129A (en) * | 2018-09-28 | 2019-03-01 | 大连理工大学 | A kind of method of low light image scene understanding |
| CN109509248B (en) * | 2018-09-28 | 2023-07-18 | 北京大学 | Photon mapping rendering method and system based on neural network |
| CN112804943A (en) * | 2018-10-03 | 2021-05-14 | 株式会社岛津制作所 | Method for creating learning completion model, luminance adjustment method, and image processing apparatus |
| CN112804943B (en) * | 2018-10-03 | 2023-09-26 | 株式会社岛津制作所 | Learning completion model creation method, brightness adjustment method, and image processing apparatus |
| CN111046887A (en) * | 2018-10-15 | 2020-04-21 | 华北电力大学（保定） | Method for extracting characteristics of image with noise |
| CN109462732B (en) * | 2018-10-29 | 2021-01-15 | 努比亚技术有限公司 | Image processing method, device and computer readable storage medium |
| CN109462732A (en) * | 2018-10-29 | 2019-03-12 | 努比亚技术有限公司 | A kind of image processing method, equipment and computer readable storage medium |
| CN109785252A (en) * | 2018-12-25 | 2019-05-21 | 山西大学 | Based on multiple dimensioned residual error dense network nighttime image enhancing method |
| CN109785252B (en) * | 2018-12-25 | 2023-03-24 | 山西大学 | Night image enhancement method based on multi-scale residual error dense network |
| CN109859120A (en) * | 2019-01-08 | 2019-06-07 | 北京交通大学 | Image defogging method based on multiple dimensioned residual error network |
| CN110163808A (en) * | 2019-03-28 | 2019-08-23 | 西安电子科技大学 | A kind of single frames high dynamic imaging method based on convolutional neural networks |
| CN110163808B (en) * | 2019-03-28 | 2022-06-10 | 西安电子科技大学 | Single-frame high-dynamic imaging method based on convolutional neural network |
| CN110232661B (en) * | 2019-05-03 | 2023-01-06 | 天津大学 | Low-illumination color image enhancement method based on Retinex and convolutional neural network |
| CN110232661A (en) * | 2019-05-03 | 2019-09-13 | 天津大学 | Low illumination colour-image reinforcing method based on Retinex and convolutional neural networks |
| CN110334807B (en) * | 2019-05-31 | 2021-09-28 | 北京奇艺世纪科技有限公司 | Deep learning network training method, device, equipment and storage medium |
| CN110334807A (en) * | 2019-05-31 | 2019-10-15 | 北京奇艺世纪科技有限公司 | Training method, device, equipment and the storage medium of deep learning network |
| CN110675336A (en) * | 2019-08-29 | 2020-01-10 | 苏州千视通视觉科技股份有限公司 | Low-illumination image enhancement method and device |
| CN112446247B (en) * | 2019-08-30 | 2022-11-15 | 北京大学 | Low-illumination face detection method based on multi-feature fusion and low-illumination face detection network |
| CN112446247A (en) * | 2019-08-30 | 2021-03-05 | 北京大学 | Low-illumination face detection method based on multi-feature fusion and low-illumination face detection network |
| WO2021063118A1 (en) * | 2019-10-02 | 2021-04-08 | Guangdong Oppo Mobile Telecommunications Corp., Ltd. | Method and apparatus for image processing |
| CN111028163B (en) * | 2019-11-28 | 2024-02-27 | 湖北工业大学 | Combined image denoising and dim light enhancement method based on convolutional neural network |
| CN111028163A (en) * | 2019-11-28 | 2020-04-17 | 湖北工业大学 | Convolution neural network-based combined image denoising and weak light enhancement method |
| CN113095470A (en) * | 2020-01-08 | 2021-07-09 | 字节跳动有限公司 | Neural network training method, image processing method and device, and storage medium |
| CN113095470B (en) * | 2020-01-08 | 2024-04-23 | 字节跳动有限公司 | Training method, image processing method and device for neural network and storage medium |
| CN111681178A (en) * | 2020-05-22 | 2020-09-18 | 厦门大学 | Knowledge distillation-based image defogging method |
| CN111681178B (en) * | 2020-05-22 | 2022-04-26 | 厦门大学 | Knowledge distillation-based image defogging method |
| CN111932471B (en) * | 2020-07-24 | 2022-07-19 | 山西大学 | Double-path exposure degree fusion network model and method for low-illumination image enhancement |
| CN111932471A (en) * | 2020-07-24 | 2020-11-13 | 山西大学 | Double-path exposure degree fusion network model and method for low-illumination image enhancement |
| CN111986240A (en) * | 2020-09-01 | 2020-11-24 | 交通运输部水运科学研究所 | Drowning person detection method and system based on visible light and thermal imaging data fusion |
| CN111928799A (en) * | 2020-09-25 | 2020-11-13 | 南京理工大学 | Three-dimensional measurement method for realizing stripe image contrast enhancement based on deep learning |
| CN112686823A (en) * | 2020-12-31 | 2021-04-20 | 广西慧云信息技术有限公司 | Automatic image enhancement method based on illumination transformation network |
| CN113066019A (en) * | 2021-02-27 | 2021-07-02 | 华为技术有限公司 | Image enhancement method and related device |
| CN113256528A (en) * | 2021-06-03 | 2021-08-13 | 中国人民解放军国防科技大学 | Low-illumination video enhancement method based on multi-scale cascade depth residual error network |
| CN113256528B (en) * | 2021-06-03 | 2022-05-27 | 中国人民解放军国防科技大学 | Low-illumination video enhancement method based on multi-scale cascade depth residual error network |
| WO2023050731A1 (en) * | 2021-09-28 | 2023-04-06 | 深圳市中兴微电子技术有限公司 | Method for training image enhancement model, image enhancement method, and readable medium |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN108447036A (en) | A kind of low light image Enhancement Method based on convolutional neural networks | |
| CN107977932B (en) | Face image super-resolution reconstruction method based on discriminable attribute constraint generation countermeasure network | |
| CN114140353B (en) | Swin-Transformer image denoising method and system based on channel attention | |
| CN107492070B (en) | A kind of single image super-resolution calculation method of binary channels convolutional neural networks | |
| CN110992275B (en) | Refined single image rain removing method based on generation of countermeasure network | |
| CN108876735B (en) | Real image blind denoising method based on depth residual error network | |
| CN107123089B (en) | Remote sensing image super-resolution reconstruction method and system based on depth convolution network | |
| CN106204779B (en) | Check class attendance method based on plurality of human faces data collection strategy and deep learning | |
| CN111127336B (en) | Image signal processing method based on self-adaptive selection module | |
| Burger et al. | Learning how to combine internal and external denoising methods | |
| CN110188795A (en) | Image classification method, data processing method and device | |
| CN109064423B (en) | Intelligent image repairing method for generating antagonistic loss based on asymmetric circulation | |
| Hu et al. | Underwater image restoration based on convolutional neural network | |
| CN111080567A (en) | Remote sensing image fusion method and system based on multi-scale dynamic convolution neural network | |
| CN111161178A (en) | Single low-light image enhancement method based on generation type countermeasure network | |
| CN112270654A (en) | Image denoising method based on multi-channel GAN | |
| CN110148088B (en) | Image processing method, image rain removing method, device, terminal and medium | |
| Chen et al. | Densely connected convolutional neural network for multi-purpose image forensics under anti-forensic attacks | |
| CN109859120A (en) | Image defogging method based on multiple dimensioned residual error network | |
| CN112712481B (en) | Structure-texture sensing method aiming at low-light image enhancement | |
| CN111415304A (en) | Underwater vision enhancement method and device based on cascade deep network | |
| CN115393225A (en) | Low-illumination image enhancement method based on multilevel feature extraction and fusion | |
| Qiu et al. | Tone mapping for hdr image using optimization a new closed form solution | |
| CN109919954A (en) | target object identification method and device | |
| CN113284061A (en) | Underwater image enhancement method based on gradient network |
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 | ||
| RJ01 | Rejection of invention patent application after publication |
Application publication date: 20180824 |