CN108364270A - Colour cast color of image restoring method and device - Google Patents
Colour cast color of image restoring method and device Download PDFInfo
- Publication number
- CN108364270A CN108364270A CN201810496229.7A CN201810496229A CN108364270A CN 108364270 A CN108364270 A CN 108364270A CN 201810496229 A CN201810496229 A CN 201810496229A CN 108364270 A CN108364270 A CN 108364270A
- Authority
- CN
- China
- Prior art keywords
- image
- network
- triple channel
- true color
- generator
- 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
- 238000000034 method Methods 0.000 title claims abstract description 51
- 230000009467 reduction Effects 0.000 claims abstract description 13
- 238000005070 sampling Methods 0.000 claims description 19
- 230000006870 function Effects 0.000 claims description 15
- 238000013528 artificial neural network Methods 0.000 claims description 14
- 230000008569 process Effects 0.000 claims description 9
- 238000012549 training Methods 0.000 claims description 7
- 230000001052 transient effect Effects 0.000 claims description 6
- 241000208340 Araliaceae Species 0.000 claims description 3
- 235000005035 Panax pseudoginseng ssp. pseudoginseng Nutrition 0.000 claims description 3
- 235000003140 Panax quinquefolius Nutrition 0.000 claims description 3
- 238000004891 communication Methods 0.000 claims description 3
- 235000008434 ginseng Nutrition 0.000 claims description 3
- 230000000875 corresponding effect Effects 0.000 claims 13
- 238000004040 coloring Methods 0.000 description 6
- 230000004913 activation Effects 0.000 description 4
- 238000013527 convolutional neural network Methods 0.000 description 4
- 238000005516 engineering process Methods 0.000 description 4
- 230000003287 optical effect Effects 0.000 description 4
- 238000005457 optimization Methods 0.000 description 4
- 238000010586 diagram Methods 0.000 description 3
- 238000000605 extraction Methods 0.000 description 2
- 230000000007 visual effect Effects 0.000 description 2
- 239000011248 coating agent Substances 0.000 description 1
- 238000000576 coating method Methods 0.000 description 1
- 230000000694 effects Effects 0.000 description 1
- 238000012986 modification Methods 0.000 description 1
- 230000004048 modification Effects 0.000 description 1
- 210000005036 nerve Anatomy 0.000 description 1
- 238000010606 normalization Methods 0.000 description 1
- 239000013307 optical fiber Substances 0.000 description 1
- 238000012545 processing Methods 0.000 description 1
- 238000011084 recovery Methods 0.000 description 1
- 229920006395 saturated elastomer Polymers 0.000 description 1
- 239000007787 solid Substances 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
Landscapes
- Physics & Mathematics (AREA)
- General Physics & Mathematics (AREA)
- Engineering & Computer Science (AREA)
- Theoretical Computer Science (AREA)
- Image Analysis (AREA)
- Image Processing (AREA)
Abstract
A kind of colour cast color of image restoring method of offer of the embodiment of the present invention and device, this method include:Colour cast image is converted into gray level image, and triple channel image is obtained into row of channels overlap-add procedure to the gray level image;The triple channel image is input to generator network, obtains estimating coloured image.The present invention realizes effective reduction of colour cast image, so as to preferably restore operation real scene, is conducive to the precision and degree of safety of surgical.
Description
Technical field
The present embodiments relate to technical field of image processing more particularly to a kind of color of image restoring method and device.
Background technology
In laser surgey, the higher light source of output power can be such that the photosensitive element of endoscope is saturated, and then cause to show
Screen is a piece of in brilliant white.At this point, doctor, which is in blind, regards state, can not judge to treat position and focal reaction of the optical fiber in cavity,
Increase operation risk.In order to avoid risk that may be present under the above situation, optical filter is generally added on endoscope lens
To filter off strong light.This method can be a piece of in brilliant white to avoid display screen, but will produce by the endoscopic images that optical filter obtains
Serious colour cast can not restore operation real scene, can influence the precision and degree of safety of surgical.
Invention content
A kind of colour cast color of image restoring method of offer of the embodiment of the present invention and device, to solve to pass through in the prior art
The endoscopic images that optical filter obtains will produce serious colour cast, can not restore operation real scene, can influence surgical
The problem of precision and degree of safety.
The embodiment of the present invention provides a kind of colour cast color of image restoring method, including:Colour cast image is converted into gray-scale map
Picture, and triple channel image is obtained into row of channels overlap-add procedure to the gray level image;The triple channel image is input to generation
Device network obtains estimating coloured image.
The embodiment of the present invention provides a kind of colour cast color of image reduction apparatus, including:Triple channel image collection module and pre-
Estimate coloured image acquisition module;The triple channel image collection module, for colour cast image to be converted to gray level image, and to institute
Gray level image is stated into row of channels overlap-add procedure, obtains triple channel image;Described to estimate coloured image acquisition module, being used for will be described
Triple channel image is input to generator network, obtains estimating coloured image.
The embodiment of the present invention provides a kind of colour cast color of image reduction apparatus, including:At least one processor;And with institute
At least one processor of processor communication connection is stated, wherein:The memory is stored with the journey that can be executed by the processor
Sequence instructs, and the processor calls described program instruction to be able to carry out method as described above.
The embodiment of the present invention provides a kind of non-transient computer readable storage medium, the non-transient computer readable storage
Medium storing computer instructs, and the computer instruction makes the computer execute method as described above.
Colour cast color of image restoring method and device provided in an embodiment of the present invention, by the way that colour cast image is converted to gray scale
Image, and triple channel image is obtained into row of channels overlap-add procedure to the gray level image;The triple channel image is input to life
It grows up to be a useful person network, obtains estimating coloured image, realize effective reduction of colour cast image, so as to preferably restore the true field of operation
Scape is conducive to the precision and degree of safety of surgical.
Description of the drawings
In order to more clearly explain the embodiment of the invention or the technical proposal in the existing technology, to embodiment or will show below
There is attached drawing needed in technology description to be briefly described, it should be apparent that, the accompanying drawings in the following description is this hair
Some bright embodiments for those of ordinary skill in the art without creative efforts, can be with root
Other attached drawings are obtained according to these attached drawings.
Fig. 1 is colour cast color of image restoring method embodiment flow chart of the present invention;
Fig. 2 is the structural schematic diagram of arbiter network of the present invention;
Fig. 3 is the structural schematic diagram of generator network of the present invention;
Fig. 4 is colour cast color of image reduction apparatus example structure schematic diagram of the present invention.
Specific implementation mode
In order to make the object, technical scheme and advantages of the embodiment of the invention clearer, below in conjunction with the embodiment of the present invention
In attached drawing, technical scheme in the embodiment of the invention is clearly and completely described, it is clear that described embodiment is
A part of the embodiment of the present invention, instead of all the embodiments.Based on the embodiments of the present invention, those of ordinary skill in the art
The every other embodiment obtained without creative efforts, shall fall within the protection scope of the present invention.
As shown in Figure 1, the embodiment of the present invention provides a kind of colour cast color of image restoring method, including:101, by colour cast figure
As being converted to gray level image, and triple channel image is obtained into row of channels overlap-add procedure to the gray level image;102, by described three
Channel image is input to generator network, obtains estimating coloured image.
In the present embodiment, it handles for convenience, it can be by colour cast image cropping to pre-set dimension, such as 224*224.Colour cast
Image can be the endoscopic images of the generation colour cast obtained by optical filter.The input of generator network is to embody luminance information
Triple channel image, therefore colour cast image is converted into gray level image, row of channels of going forward side by side superposition generates triple channel image.Generator
Network is mainly used for extracting the advanced features information of image, and the advanced features information based on image establishes the luminance information of image
Correspondence between the colouring information of image, to more accurately carry out the color recovery of image.
Colour cast color of image restoring method provided in an embodiment of the present invention, by the way that colour cast image is converted to gray level image,
And triple channel image is obtained into row of channels overlap-add procedure to the gray level image;The triple channel image is input to generator
Network obtains estimating coloured image, realizes effective reduction of colour cast image, so as to preferably restore operation real scene,
Be conducive to the precision and degree of safety of surgical.
It is described that the triple channel image is input to generator network as a kind of alternative embodiment, it obtains estimating colour
Further include before image:Several true color images are converted to yuv space, up-sampling convolutional network is trained;According to
Up-sampling convolutional network after residual error neural network and training, obtains generator network.
In the present embodiment, true color image is cut to size identical as colour cast image:224*224.It is several true
Coloured image is bias colour normal picture.Since under yuv space, the luminance information and colouring information of image are separately shown
's.By converting several true color images to yuv space, can will scheme using the luminance information of image as sample data
The colouring information of picture is trained up-sampling convolutional network as label data.Residual error neural network is mainly used for extraction figure
The advanced features information of picture, up-sampling convolutional neural networks establish the brightness letter of image for the advanced features information based on image
Correspondence between breath and the colouring information of image.
It is described to convert several true color images to yuv space as a kind of alternative embodiment, to up-sampling convolution god
It is trained and specifically includes through network:Several true color images are converted to yuv space, each true color image pair is obtained
The YUV image and Y images answered;By the corresponding Y images of each true color image into row of channels overlap-add procedure, obtain each true
The corresponding triple channel image of coloured image;Using the corresponding triple channel image of all true color images as sample data, own
The corresponding U of true color image and V image are trained up-sampling convolutional neural networks as label data.
In the present embodiment, in order to enable generator network can preferably embody image colouring information and image it is bright
The correspondence between information is spent, needs to be trained the up-sampling convolutional neural networks in generator network.Due to unbiased
The correspondence of the colouring information of the luminance information and image of image is the most accurate in the true color image of color, therefore, uses
True color image is trained up-sampling convolutional neural networks.Under yuv space, Y images represent the brightness letter of image
Breath, i.e. gray level image.
As a kind of alternative embodiment, the up-sampling convolutional network after the residual error neural network and training is generated
Further include after device network:It is that all true color images and the generator network are generated with each true color image pair
The coloured image of estimating answered is input to arbiter network, carries out confrontation type study to optimize the parameter of the generator network.
In the present embodiment, the corresponding with each true color image of generator network generation estimates coloured image
The corresponding triple channel image input generator network of each true color image is obtained.In order to enable generator network is given birth to
At the color reduction effect for estimating coloured image it is more preferable, the parameter of generator network is optimized.Specifically, by by institute
There is the coloured image of estimating corresponding with each true color image that true color image and the generator network generate to input
Confrontation type study is carried out to arbiter network, obtains estimating between coloured image for bias colour true color image and generation
Color distortion for optimize generator network parameter.The structure of arbiter network can be as shown in Figure 2.Arbiter network packet
Containing eight convolutional layers, full articulamentum, activation primitive and full articulamentum are followed successively by after eight convolutional layers.Convolution kernel is preferably sized to 3*
3, it is preferred to use LeakyRelu activation primitives.With the increase of network depth, the characteristic pattern quantity of convolutional layer increases to from 64
512.Arbiter network can also be other structures, not limit the parameter optimization, it can be achieved that generator network herein.
As a kind of alternative embodiment, it is described it is that all true color images and the generator network are generated with it is each
The corresponding coloured image of estimating of true color image is input to arbiter network, carries out confrontation type study to optimize the generator
The parameter of network specifically includes:The parameter of the fixed generator network, by all true color images and the generator net
The coloured image of estimating corresponding with each true color image that network generates for the first time is input to arbiter network progress confrontation type
It practises, obtains the first Optimal Parameters of the arbiter network;It is the parameter of the arbiter network with first Optimal Parameters,
The parameter of the fixed arbiter network, minimizes the loss function of the generator network, obtains the generator network
Second Optimal Parameters;With the parameter that second Optimal Parameters are the generator network, the ginseng of the fixed generator network
Number is generating with each true color under second Optimal Parameters by all true color images and the generator network
The corresponding coloured image of estimating of image is input to the arbiter network progress confrontation type study, obtains the arbiter network
The first new Optimal Parameters;With the parameter that the first new Optimal Parameters are the arbiter network, the fixed arbiter
The parameter of network minimizes the loss function of the generator network, obtains the second new optimization ginseng of the generator network
Number;Repeat the acquisition process of the second new Optimal Parameters, until each true color image and with each true color figure
As corresponding coloured image of estimating is less than default error.
In the present embodiment, arbiter network, it is preferable that use srgan networks.Arbiter network carries out confrontation study,
The process of the first Optimal Parameters of arbiter network is obtained, the loss function by minimizing arbiter network is realized.Arbiter
The loss function of network can select according to actual needs, it is preferable that use cross entropy loss function.The loss letter of generator network
Number l includes content loss function lXWith confrontation loss function lGen, formula is:L=lX+10-3lGen.Wherein, content loss function lX
Represent the MSE losses l for estimating coloured image and true color imageMSEL is lost with visual similarityVGG;Fight loss function lXFor
The opposite number of arbiter network losses.MSE loses lMSEFormula be: Its
In, W, H respectively represent the width and height of image,Indicate true color image IRCIn at (x, y) pixel pixel value,It indicates to estimate coloured image I by what generator generatedGCIn at (x, y) pixel pixel value;Visual similarity loses lVGGPublic affairs
Formula is:Wherein
Indicate the I that 19 layers of VGG networks using pre-training extractRCJ-th of convolutional layer in corresponding network before i-th of maximum pond layer,It indicatesIn at (x, y) pixel pixel value,Indicate 19 using pre-training
The I of layer VGG networks extractionGCJ-th of convolutional layer in corresponding network before i-th of maximum pond layer,It indicatesIn at (x, y) pixel pixel value, Wi,jAnd Hi,jThe width and height of character pair figure respectively.Confrontation loss
Function lGenFormula be: What is indicated is that arbiter generates generator
ImageIt is determined as that the probability of true color image, N indicate the quantity of training image.Minimize the damage of the generator network
Function is lost, obtains the second new Optimal Parameters of the generator network, i.e.,It needs
Illustrate, generator network includes residual error neural network and up-sampling network.In the optimization of generator network, residual error nerve
The parameter of network is fixed value, only the parameter of optimization up-sampling network.
It is described that the triple channel image is input to generator network as a kind of alternative embodiment, it obtains estimating colour
Image specifically includes:The triple channel image is input to residual error neural network, obtains each layer of output characteristic pattern;It will most
Later layer is as current layer, and up-sampling to the preceding layer of current layer corresponds to after the output characteristic pattern of current layer is carried out convolution operation
The size of characteristic pattern is exported, and output characteristic pattern corresponding with the preceding layer of current layer is added, and obtains middle graph;Before current layer
One layer, as new current layer, the preceding layer of up-sampling to new current layer after middle graph progress convolution operation is corresponded to defeated
Go out the size of characteristic pattern, and output characteristic pattern corresponding with the preceding layer of new current layer is added, and obtains new middle graph;It repeats new
Middle graph acquisition process until new middle graph is identical as the triple channel picture size;According to the triple channel image
The identical new middle graph of size and the gray level image obtain described estimating coloured image.
In the present embodiment, triple channel image is inputted into generator network, the process for obtaining estimating coloured image can be as
Shown in Fig. 3.Wherein, triple channel image is:After colour cast image is converted to gray level image, it is 224* that gray level image, which is cut to size,
224, then be superimposed into row of channels, obtain the triple channel image that 3 sizes are 224*224.The triple channel image of 224*224*3 is defeated
Enter to residual error neural network, the output characteristic pattern size of obtained each layer is respectively 112*112,56*56,28*28,14*14
And 7*7, number are respectively 64,256,512,1024 and 2048.Then, using last layer as current layer, by the defeated of current layer
Go out characteristic pattern, i.e., the figure for being 7*7 by 2048 sizes that residual error neural network obtains uses size to be carried out for the convolution kernel of 3*3
Convolution operation obtains the figure that 1024 sizes are 7*7;It up-samples the figure that this 1024 sizes are 7*7 to obtain 1024 rulers again
The very little figure for 14*14, and output characteristic pattern corresponding with the preceding layer of current layer, i.e., 1024 rulers obtained by residual error neural network
The very little figure for 14*14 is added, and obtains the middle graph that 1024 sizes are 14*14.Using the preceding layer of current layer as newly current
Layer, i.e., the corresponding layer of figure for being 14*14 by 1024 sizes that residual error neural network obtains, uses size for the convolution kernel of 3*3
Convolution operation is carried out to the middle graph that 1024 sizes are 14*14, obtains the figure that 512 sizes are 14*14, then by this 512
The figure up-sampling that size is 14*14 obtains the figure that 512 sizes are 28*28, and output corresponding with the preceding layer of new current layer
Characteristic pattern, i.e., the figure that 512 sizes are 28*28 are added, and obtain the new middle graph that 512 sizes are 28*28.During repetition is new
Between figure acquisition process until new middle graph is identical with the triple channel picture size, be all 3 sizes for 224*224
Figure.The new middle graph for using size to be 224*224 for 3 sizes of convolution kernel pair of 3*3 carries out convolution operation and obtains 2 sizes
For the figure of 224*224, then the figure that this 2 sizes are 224*224 is superimposed with gray level image into row of channels, obtains estimating cromogram
Picture.
By each layer characteristic pattern size, the number of plies of residual error neural network that residual error neural network obtains it is solid in the above process
Definite value.When carrying out convolution operation, convolution kernel size is fixed value.In upsampling process, the activation primitive of network is preferably
Relu functions;The new middle graph progress convolution operation for being 224*224 to 3 sizes obtains the figure that 2 sizes are 224*224
When, the activation primitive of network is preferably sigmoid functions;A batch normalization layer is added after each active coating, i.e.,
Batchnorm layers.
As a kind of alternative embodiment, the basis new middle graph identical with the triple channel picture size and described
Gray level image, obtain it is described estimate coloured image, specifically include:Pair new middle graph identical with the triple channel picture size
It is superimposed into row of channels with the gray level image after carrying out convolution operation, obtains YUV image;The YUV image is converted to RGB skies
Between under, obtain estimating coloured image.
In the present embodiment, the gray-scale map that the new middle graph and size for being 224*224 according to 3 sizes is 224*224
Picture, obtain it is described estimate coloured image can be use size for 3 sizes of convolution kernel pair of 3*3 be 224*224 it is new in
Between figure carry out convolution operation obtain 2 sizes be 224*224 figure, then by this 2 sizes be 224*224 figure and gray level image
It is superimposed into row of channels, obtains YUV image;The YUV image is converted to rgb space, obtains estimating coloured image.
As shown in figure 4, the embodiment of the present invention provides a kind of colour cast color of image reduction apparatus, including:Triple channel image obtains
Modulus block 401 and estimate coloured image acquisition module 402;The triple channel image collection module 401, for turning colour cast image
It is changed to gray level image, and triple channel image is obtained into row of channels overlap-add procedure to the gray level image;It is described to estimate coloured image
Acquisition module 402 obtains estimating coloured image for the triple channel image to be input to generator network.
Colour cast color of image reduction apparatus provided in an embodiment of the present invention, by the way that colour cast image is converted to gray level image,
And triple channel image is obtained into row of channels overlap-add procedure to the gray level image;The triple channel image is input to generator
Network obtains estimating coloured image, realizes effective reduction of colour cast image, so as to preferably restore operation real scene,
Be conducive to the precision and degree of safety of surgical.
The embodiment of the present invention provides a kind of colour cast color of image reduction apparatus, including:At least one processor;And with institute
At least one processor of processor communication connection is stated, wherein:The memory is stored with the journey that can be executed by the processor
Sequence instructs, and the processor calls described program instruction to be able to carry out the method that above-mentioned each method embodiment is provided, such as wraps
It includes:101, colour cast image is converted into gray level image, and triple channel figure is obtained into row of channels overlap-add procedure to the gray level image
Picture;102, the triple channel image is input to generator network, obtains estimating coloured image.
The present embodiment provides a kind of non-transient computer readable storage medium, the non-transient computer readable storage medium
Computer instruction is stored, the computer instruction makes the computer execute the method that above-mentioned each method embodiment is provided, example
Such as include:101, colour cast image is converted into gray level image, and threeway is obtained into row of channels overlap-add procedure to the gray level image
Road image;102, the triple channel image is input to generator network, obtains estimating coloured image.
The apparatus embodiments described above are merely exemplary, wherein the unit illustrated as separating component can
It is physically separated with being or may not be, the component shown as unit may or may not be physics list
Member, you can be located at a place, or may be distributed over multiple network units.It can be selected according to the actual needs
In some or all of module achieve the purpose of the solution of this embodiment.Those of ordinary skill in the art are not paying creativeness
Labour in the case of, you can to understand and implement.
Through the above description of the embodiments, those skilled in the art can be understood that each embodiment can
It is realized by the mode of software plus required general hardware platform, naturally it is also possible to pass through hardware.Based on this understanding, on
Stating technical solution, substantially the part that contributes to existing technology can be expressed in the form of software products in other words, should
Computer software product can store in a computer-readable storage medium, such as ROM/RAM, magnetic disc, CD, including several fingers
It enables and using so that a computer equipment (can be personal computer, server or the network equipment etc.) executes each implementation
Method described in certain parts of example or embodiment.
Finally it should be noted that:The above embodiments are merely illustrative of the technical solutions of the present invention, rather than its limitations;Although
Present invention has been described in detail with reference to the aforementioned embodiments, it will be understood by those of ordinary skill in the art that:It still may be used
With technical scheme described in the above embodiments is modified or equivalent replacement of some of the technical features;
And these modifications or replacements, various embodiments of the present invention technical solution that it does not separate the essence of the corresponding technical solution spirit and
Range.
Claims (10)
1. a kind of colour cast color of image restoring method, which is characterized in that including:
Colour cast image is converted into gray level image, and triple channel image is obtained into row of channels overlap-add procedure to the gray level image;
The triple channel image is input to generator network, obtains estimating coloured image.
2. according to the method described in claim 1, it is characterized in that, described be input to generator net by the triple channel image
Network further includes before obtaining estimating coloured image:
Several true color images are converted to yuv space, up-sampling convolutional network is trained;
According to the up-sampling convolutional network after residual error neural network and training, generator network is obtained.
3. according to the method described in claim 2, it is characterized in that, described convert several true color images to yuv space,
Up-sampling convolutional network is trained and is specifically included:
Several true color images are converted to yuv space, each true color image corresponding Y, U and V image are obtained;
By the corresponding Y images of each true color image into row of channels overlap-add procedure, it is corresponding to obtain each true color image
Triple channel image;
Using the corresponding triple channel image of all true color images as sample data, the corresponding U and V of all true color images
Image is trained up-sampling convolutional network as label data.
4. according to the method described in claim 2, it is characterized in that, the up-sampling according to after residual error neural network and training
Convolutional network, obtaining generator network further includes later:
All true color images and the corresponding with each true color image of generator network generation are estimated into colour
Image is input to arbiter network, carries out confrontation type study to optimize the parameter of the generator network.
5. according to the method described in claim 4, it is characterized in that, described by all true color images and the generator net
The coloured image of estimating corresponding with each true color image that network generates is input to arbiter network, carry out confrontation type study with
The parameter for optimizing the generator network specifically includes:
The parameter of the fixed generator network, all true color images and the generator network are generated for the first time with it is every
The corresponding coloured image of estimating of one true color image is input to the progress confrontation type study of arbiter network, obtains the arbiter
First Optimal Parameters of network;With the parameter that first Optimal Parameters are the arbiter network, the fixed arbiter net
The parameter of network minimizes the loss function of the generator network, obtains the second Optimal Parameters of the generator network;
With the parameter that second Optimal Parameters are the generator network, the parameter of the fixed generator network will own
True color image and the generator network generate corresponding with each true color image under second Optimal Parameters
Coloured image of estimating be input to the arbiter network and carry out confrontation type study, obtain new first of the arbiter network
Optimal Parameters;With the parameter that the first new Optimal Parameters are the arbiter network, the ginseng of the fixed arbiter network
Number, minimizes the loss function of the generator network, obtains the second new Optimal Parameters of the generator network;
Repeat the acquisition process of the second new Optimal Parameters, until each true color image and with each true color figure
As corresponding coloured image of estimating is less than default error.
6. according to the method described in claim 1, it is characterized in that, described be input to generator net by the triple channel image
Network is obtained estimating coloured image, be specifically included:
The triple channel image is input to residual error neural network, obtains each layer of output characteristic pattern;
Using last layer as current layer, up-sampled to current layer after the output characteristic pattern of current layer is carried out convolution operation
The size of one layer of corresponding output characteristic pattern, and output characteristic pattern corresponding with the preceding layer of current layer is added, and obtains middle graph;
Using the preceding layer of current layer as new current layer, middle graph progress convolution operation is up-sampled to new current layer
Preceding layer correspond to output characteristic pattern size, and output characteristic pattern correspondings with the preceding layer of new current layer addition, obtain newly
Middle graph;
The acquisition process of new middle graph is repeated until new middle graph is identical as the triple channel picture size;
According to new middle graph identical with the triple channel picture size and the gray level image, obtain described estimating cromogram
Picture.
7. according to the method described in claim 1, it is characterized in that, the basis is identical with the triple channel picture size new
Middle graph and the gray level image, obtain it is described estimate coloured image, specifically include:
Pair identical with the triple channel picture size new middle graph is led to after carrying out convolution operation with the gray level image
Trace-stacking obtains YUV image;
The YUV image is converted to rgb space, obtains estimating coloured image.
8. a kind of colour cast color of image reduction apparatus, which is characterized in that including:Triple channel image collection module and estimate cromogram
As acquisition module;
The triple channel image collection module for colour cast image to be converted to gray level image, and carries out the gray level image
Channel overlap-add procedure obtains triple channel image;
It is described to estimate coloured image acquisition module, for the triple channel image to be input to generator network, obtain estimating coloured silk
Color image.
9. a kind of colour cast color of image reduction apparatus, which is characterized in that including:
At least one processor;And
At least one processor being connect with the processor communication, wherein:
The memory is stored with the program instruction that can be executed by the processor, and the processor calls described program to instruct energy
Enough methods executed as described in claim 1 to 7 is any.
10. a kind of non-transient computer readable storage medium, which is characterized in that the non-transient computer readable storage medium is deposited
Computer instruction is stored up, the computer instruction makes the computer execute the method as described in any one of claim 1-7.
Priority Applications (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201810496229.7A CN108364270B (en) | 2018-05-22 | 2018-05-22 | Color reduction method and device for color cast image |
Applications Claiming Priority (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201810496229.7A CN108364270B (en) | 2018-05-22 | 2018-05-22 | Color reduction method and device for color cast image |
Publications (2)
Publication Number | Publication Date |
---|---|
CN108364270A true CN108364270A (en) | 2018-08-03 |
CN108364270B CN108364270B (en) | 2020-11-06 |
Family
ID=63012232
Family Applications (1)
Application Number | Title | Priority Date | Filing Date |
---|---|---|---|
CN201810496229.7A Active CN108364270B (en) | 2018-05-22 | 2018-05-22 | Color reduction method and device for color cast image |
Country Status (1)
Country | Link |
---|---|
CN (1) | CN108364270B (en) |
Cited By (8)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN109859288A (en) * | 2018-12-25 | 2019-06-07 | 北京飞搜科技有限公司 | Based on the image painting methods and device for generating confrontation network |
CN110276731A (en) * | 2019-06-17 | 2019-09-24 | 艾瑞迈迪科技石家庄有限公司 | Endoscopic image color restoring method and device |
CN110930333A (en) * | 2019-11-22 | 2020-03-27 | 北京金山云网络技术有限公司 | Image restoration method and device, electronic equipment and computer-readable storage medium |
CN111353585A (en) * | 2020-02-25 | 2020-06-30 | 北京百度网讯科技有限公司 | Structure searching method and device of neural network model |
CN111898449A (en) * | 2020-06-30 | 2020-11-06 | 北京大学 | Pedestrian attribute identification method and system based on monitoring video |
CN111898448A (en) * | 2020-06-30 | 2020-11-06 | 北京大学 | Pedestrian attribute identification method and system based on deep learning |
US20210203900A1 (en) * | 2020-04-17 | 2021-07-01 | Beijing Baidu Netcom Science And Technology Co., Ltd. | Image processing method and apparatus, electronic device and computer-readable storage medium |
CN113870371A (en) * | 2021-12-03 | 2021-12-31 | 浙江霖研精密科技有限公司 | Picture color transformation device and method based on generation countermeasure network and storage medium |
Citations (4)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
TW200641723A (en) * | 2005-02-11 | 2006-12-01 | Hewlett Packard Development Co | Decreasing aliasing in electronic images |
CN101930596A (en) * | 2010-07-19 | 2010-12-29 | 赵全友 | Color constancy method in two steps under a kind of complex illumination |
CN107451963A (en) * | 2017-07-05 | 2017-12-08 | 广东欧谱曼迪科技有限公司 | Multispectral nasal cavity endoscope Real-time image enhancement method and endoscopic imaging system |
CN107527332A (en) * | 2017-10-12 | 2017-12-29 | 长春理工大学 | Enhancement Method is kept based on the low-light (level) image color for improving Retinex |
-
2018
- 2018-05-22 CN CN201810496229.7A patent/CN108364270B/en active Active
Patent Citations (5)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
TW200641723A (en) * | 2005-02-11 | 2006-12-01 | Hewlett Packard Development Co | Decreasing aliasing in electronic images |
US7525583B2 (en) * | 2005-02-11 | 2009-04-28 | Hewlett-Packard Development Company, L.P. | Decreasing aliasing in electronic images |
CN101930596A (en) * | 2010-07-19 | 2010-12-29 | 赵全友 | Color constancy method in two steps under a kind of complex illumination |
CN107451963A (en) * | 2017-07-05 | 2017-12-08 | 广东欧谱曼迪科技有限公司 | Multispectral nasal cavity endoscope Real-time image enhancement method and endoscopic imaging system |
CN107527332A (en) * | 2017-10-12 | 2017-12-29 | 长春理工大学 | Enhancement Method is kept based on the low-light (level) image color for improving Retinex |
Non-Patent Citations (1)
Title |
---|
CHRISTIAN LEDIG ET AL: ""Photo-Realistic Single Image Super-Resolution Using a Generative Adversarial Network"", 《IEEE》 * |
Cited By (11)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN109859288A (en) * | 2018-12-25 | 2019-06-07 | 北京飞搜科技有限公司 | Based on the image painting methods and device for generating confrontation network |
CN109859288B (en) * | 2018-12-25 | 2023-01-10 | 苏州飞搜科技有限公司 | Image coloring method and device based on generation countermeasure network |
CN110276731A (en) * | 2019-06-17 | 2019-09-24 | 艾瑞迈迪科技石家庄有限公司 | Endoscopic image color restoring method and device |
CN110930333A (en) * | 2019-11-22 | 2020-03-27 | 北京金山云网络技术有限公司 | Image restoration method and device, electronic equipment and computer-readable storage medium |
CN111353585A (en) * | 2020-02-25 | 2020-06-30 | 北京百度网讯科技有限公司 | Structure searching method and device of neural network model |
US20210203900A1 (en) * | 2020-04-17 | 2021-07-01 | Beijing Baidu Netcom Science And Technology Co., Ltd. | Image processing method and apparatus, electronic device and computer-readable storage medium |
US11930307B2 (en) * | 2020-04-17 | 2024-03-12 | Beijing Baidu Netcom Science Technology Co., Ltd. | Image processing method and apparatus, electronic device and computer-readable storage medium |
CN111898449A (en) * | 2020-06-30 | 2020-11-06 | 北京大学 | Pedestrian attribute identification method and system based on monitoring video |
CN111898448A (en) * | 2020-06-30 | 2020-11-06 | 北京大学 | Pedestrian attribute identification method and system based on deep learning |
CN111898448B (en) * | 2020-06-30 | 2023-10-24 | 北京大学 | Pedestrian attribute identification method and system based on deep learning |
CN113870371A (en) * | 2021-12-03 | 2021-12-31 | 浙江霖研精密科技有限公司 | Picture color transformation device and method based on generation countermeasure network and storage medium |
Also Published As
Publication number | Publication date |
---|---|
CN108364270B (en) | 2020-11-06 |
Similar Documents
Publication | Publication Date | Title |
---|---|---|
CN108364270A (en) | Colour cast color of image restoring method and device | |
CN110276731B (en) | Endoscopic image color reduction method and device | |
CN111080531B (en) | Super-resolution reconstruction method, system and device for underwater fish image | |
JP2020530920A (en) | Image lighting methods, devices, electronics and storage media | |
CN107798661B (en) | Self-adaptive image enhancement method | |
CN109829868B (en) | Lightweight deep learning model image defogging method, electronic equipment and medium | |
CN115223004A (en) | Method for generating confrontation network image enhancement based on improved multi-scale fusion | |
CN113284061B (en) | Underwater image enhancement method based on gradient network | |
CN109389667A (en) | A kind of efficient global illumination method for drafting based on deep learning | |
CN114067018B (en) | Infrared image colorization method for generating countermeasure network based on expansion residual error | |
CN114627002A (en) | Image defogging method based on self-adaptive feature fusion | |
CN115689962A (en) | Multi-exposure image fusion method based on multi-scale self-encoder | |
Wei et al. | Sidgan: Single image dehazing without paired supervision | |
CN116362998A (en) | Image enhancement device, image enhancement method, electronic device, and storage medium | |
JP5622971B2 (en) | Optical illusion analysis device, optical illusion-added image generation device that adds illusion to the perception of the original image, optical illusion analysis method, optical illusion-added image generation method that adds illusion to the perception of the original image, and program | |
Guo et al. | Reinforced depth-aware deep learning for single image dehazing | |
CN112200719B (en) | Image processing method, electronic device, and readable storage medium | |
CN112488926A (en) | System and method for neural network based color restoration | |
US20210374916A1 (en) | Storage medium storing program, image processing apparatus, and training method of machine learning model | |
Pavlidis | Mixed raster content: Segmentation, compression, transmission | |
CN113191971A (en) | Unmanned aerial vehicle image defogging method based on YUV color space | |
CN108429878A (en) | Deep sky object shoots intelligent optical imaging method, apparatus and system | |
CN111739008A (en) | Image processing method, device, equipment and readable storage medium | |
Akpinar et al. | Phase-coded computational imaging for accommodation-invariant near-eye displays | |
Marasakatla et al. | Under Water Image Enhancement Using CNN |
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 |