CN109447900A - A kind of image super-resolution rebuilding method and device - Google Patents
A kind of image super-resolution rebuilding method and device Download PDFInfo
- Publication number
- CN109447900A CN109447900A CN201811161605.3A CN201811161605A CN109447900A CN 109447900 A CN109447900 A CN 109447900A CN 201811161605 A CN201811161605 A CN 201811161605A CN 109447900 A CN109447900 A CN 109447900A
- Authority
- CN
- China
- Prior art keywords
- image
- resolution
- super
- residual error
- layer
- 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 51
- 230000004913 activation Effects 0.000 claims abstract description 33
- 230000006870 function Effects 0.000 claims description 23
- 238000004891 communication Methods 0.000 claims description 12
- 238000005457 optimization Methods 0.000 claims description 12
- 238000013528 artificial neural network Methods 0.000 claims description 9
- 238000003860 storage Methods 0.000 claims description 8
- 230000001052 transient effect Effects 0.000 claims description 6
- 238000005070 sampling Methods 0.000 claims description 4
- 238000003475 lamination Methods 0.000 claims description 2
- 238000013527 convolutional neural network Methods 0.000 description 11
- 238000010586 diagram Methods 0.000 description 7
- 238000004364 calculation method Methods 0.000 description 4
- 230000009286 beneficial effect Effects 0.000 description 3
- 238000009826 distribution Methods 0.000 description 3
- 230000008569 process Effects 0.000 description 3
- 230000000717 retained effect Effects 0.000 description 3
- 230000009467 reduction Effects 0.000 description 2
- 238000003745 diagnosis Methods 0.000 description 1
- 238000005516 engineering process Methods 0.000 description 1
- 230000002708 enhancing effect Effects 0.000 description 1
- 230000014759 maintenance of location Effects 0.000 description 1
- 238000003909 pattern recognition Methods 0.000 description 1
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T3/00—Geometric image transformations in the plane of the image
- G06T3/40—Scaling of whole images or parts thereof, e.g. expanding or contracting
- G06T3/4053—Scaling of whole images or parts thereof, e.g. expanding or contracting based on super-resolution, i.e. the output image resolution being higher than the sensor resolution
- G06T3/4076—Scaling of whole images or parts thereof, e.g. expanding or contracting based on super-resolution, i.e. the output image resolution being higher than the sensor resolution using the original low-resolution images to iteratively correct the high-resolution images
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/04—Architecture, e.g. interconnection topology
- G06N3/045—Combinations of networks
Landscapes
- Engineering & Computer Science (AREA)
- Physics & Mathematics (AREA)
- Theoretical Computer Science (AREA)
- General Physics & Mathematics (AREA)
- Data Mining & Analysis (AREA)
- Molecular Biology (AREA)
- Biophysics (AREA)
- Computational Linguistics (AREA)
- Artificial Intelligence (AREA)
- Evolutionary Computation (AREA)
- General Health & Medical Sciences (AREA)
- Biomedical Technology (AREA)
- Computing Systems (AREA)
- General Engineering & Computer Science (AREA)
- Life Sciences & Earth Sciences (AREA)
- Mathematical Physics (AREA)
- Software Systems (AREA)
- Health & Medical Sciences (AREA)
- Image Processing (AREA)
Abstract
The embodiment of the present invention provides a kind of image super-resolution rebuilding method and device.This method comprises: first resolution image input picture reconstruction model is exported second resolution image；Image reconstruction model is obtained after being trained based on first resolution sample image and its corresponding second resolution sample image；Wherein, first resolution is less than second resolution, and described image reconstruction model includes several layers residual error structure and one layer of super-pixel structure, and the residual error structure includes several convolutional layers and several nonlinear cascade activation primitive layers.The residual error structure of image super-resolution rebuilding method provided in an embodiment of the present invention, image reconstruction model replaces part convolutional layer using cascade nonlinear activation function CReLU, reduces the parameter of image reconstruction model, improves the speed of image super-resolution rebuilding.
Description
Technical field
The present embodiments relate to image reconstruction field more particularly to a kind of image super-resolution rebuilding methods and device.
Background technique
In electronic image application field, people often it is expected to obtain high-definition picture.High-resolution means in image
Pixel density it is high, be capable of providing more details, and these details are indispensable in many practical applications.For example, high score
It is very helpful that resolution medical image, which makes correctly diagnosis for doctor,；It is just easy to using high-resolution satellite image
Similar object is distinguished from homologue；If being capable of providing high-resolution image, the property of the pattern-recognition in computer vision
It can will greatly improve.Therefore, a kind of method for finding enhancing current resolution level is very necessary.
Image super-resolution, which refers to, recovers high-definition picture by a width low-resolution image or image sequence.Image is super
Resolution technique is divided into Super-Resolution and super-resolution rebuilding.Currently, image super-resolution research can be divided into 3 main models
Farmland: the method for reconstructing based on interpolation, the method based on study and frequency domain method.Method based on interpolation is on image
The value of each pixel is calculate with points several around it approaching, and obtained image is excessively smooth, is lost perhaps
More high frequency details.Parameters frequency domain method is a kind of important method in image super-resolution rebuilding, wherein it is mixed most importantly to disappear
Folded method for reconstructing.The aliasing method for reconstructing that disappears is to improve the spatial resolution realization Super-Resolution of image by solving aliasing,
Its shortcomings that is that frequency domain data lacks correlation.
Currently, there are also the super-resolution rebuilding algorithms based on convolutional neural networks, using a large amount of high resolution graphics
As learning of structure library generation learning model, introducing is obtained by learning model during restoring to low-resolution image
Priori knowledge, to obtain the high frequency detail of image.However image superficial feature is used only in the method based on study.Moreover, convolution
The convolution kernel size of neural network is larger, affects the speed of service of convolutional neural networks.
Summary of the invention
For the drawbacks described above of traditional images super resolution ratio reconstruction method.The embodiment of the present invention provides a kind of Image Super-resolution
Rate method for reconstructing.
In a first aspect, the embodiment of the present invention provides a kind of image super-resolution rebuilding method, comprising:
By first resolution image input picture reconstruction model, second resolution image is exported；Described image reconstruction model
It is to be obtained after being trained based on first resolution sample image and its corresponding second resolution sample image；Wherein,
One resolution ratio is less than second resolution, and image reconstruction model includes several layers residual error structure and one layer of super-pixel structure, residual error knot
Structure includes several convolutional layers and several nonlinear cascade activation primitive layers.
Second aspect, the embodiment of the present invention provide a kind of image super-resolution rebuilding device, comprising:
Image reconstruction module, for exporting second resolution image for first resolution image input picture reconstruction model；
Described image reconstruction model is trained based on first resolution sample image and its corresponding second resolution sample image
It obtains afterwards；Wherein, first resolution is less than second resolution, and image reconstruction model includes that several layers residual error structure and one layer are super
Dot structure, residual error structure include several convolutional layers and several nonlinear cascade activation primitive layers.
The third aspect, the embodiment of the invention provides a kind of electronic equipment, comprising:
At least one processor, at least one processor, communication interface and bus；Wherein,
The processor, memory, communication interface complete mutual communication by the bus；
The memory is stored with the program instruction that can be executed by the processor, and processor calls described program to instruct energy
Enough execute image super-resolution provided by any possible implementation in the various possible implementations of first aspect
Method for reconstructing.
Fourth aspect, the embodiment of the invention provides a kind of non-transient computer readable storage medium, non-transient computers
Readable storage medium storing program for executing stores computer instruction, and computer instruction makes the various possible implementations of computer execution first aspect
In image super-resolution rebuilding method provided by any possible implementation.
First resolution image input picture is rebuild mould by image super-resolution rebuilding method provided in an embodiment of the present invention
Type exports second resolution image.Image reconstruction model includes several layers residual error structure and one layer of super-pixel structure, residual error structure
It is made of several convolutional layers and several nonlinear cascade activation primitive layers.Image super-resolution rebuilding provided in an embodiment of the present invention
The residual error structure of method, image reconstruction model replaces part convolutional layer using cascade nonlinear activation function CReLU, reduces
The parameter of image reconstruction model improves the speed of image super-resolution rebuilding.
Detailed description of the invention
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 technical description to be briefly described, it should be apparent that, the accompanying drawings in the following description is this hair
Bright some 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 the flow diagram according to image super-resolution rebuilding method provided in an embodiment of the present invention；
Fig. 2 is the structural schematic diagram according to image reconstruction model provided in an embodiment of the present invention；
Fig. 3 is the structural block diagram according to image super-resolution rebuilding device provided in an embodiment of the present invention；
Fig. 4 is the structural schematic diagram according to electronic equipment provided in an embodiment of the present invention.
Specific embodiment
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
Every other embodiment obtained without creative efforts, shall fall within the protection scope of the present invention.
It should be noted that, in this document, relational terms such as first and second and the like are used merely to a reality
Body or operation are distinguished with another entity or operation, are deposited without necessarily requiring or implying between these entities or operation
In any actual relationship or order or sequence.Also, the terms "include", "comprise" or its any other variant are intended to
Non-exclusive inclusion, so that the process, method, article or equipment including a series of elements is not only wanted including those
Element, but also including other elements that are not explicitly listed, or further include for this process, method, article or equipment
Intrinsic element.In the absence of more restrictions, the element limited by sentence "including a ...", it is not excluded that
There is also other identical elements in process, method, article or equipment including the element.
Fig. 1 is the flow diagram according to image super-resolution rebuilding method provided in an embodiment of the present invention.According to Fig. 2
The structural schematic diagram of image reconstruction model provided in an embodiment of the present invention.Referring to Figures 1 and 2, the embodiment of the present invention provides one kind
Image super-resolution rebuilding method, comprising:
By first resolution image input picture reconstruction model, second resolution image is exported；Described image reconstruction model
It is to be obtained after being trained based on first resolution sample image and its corresponding second resolution sample image；Wherein,
One resolution ratio is less than second resolution, and described image reconstruction model includes several layers residual error structure and one layer of super-pixel structure, institute
Stating residual error structure includes several convolutional layers and several nonlinear cascade activation primitive layers.
Wherein, image resolution ratio (ImageResolution) refers to the information content stored in image.There are many this resolution ratio
Balancing method is typically measured with the pixel number of per inch.Certainly also have and to be measured with pixel number per cm.Image
Resolution ratio determines the quality of image output.Image super-resolution refers to be recovered by a width low-resolution image or image sequence
High-definition picture.
The present embodiment is instructed using based on first resolution sample image and its corresponding second resolution sample image
First resolution image input picture reconstruction model is exported second resolution image by the image reconstruction model obtained after white silk.From
And make first resolution image super-resolution rebuilding.The obtaining step of image reconstruction model specifically includes:
Image and second resolution sample graph after first resolution sample image is rebuild are calculated by optimization object function
As the average value of the pixel difference absolute value of each point；
Optimization object function are as follows:
In formula, ISRIndicate the pixel for the image that first resolution sample image is obtained by neural network reconstruction, IHRIt indicates
The pixel of second resolution sample image, the port number of C representative image, L1Indicate that optimization object function, n indicate the number of image
Mesh, k indicate port number from 1 to C, and j indicates that the pixel of width, H indicate the height of image, and W indicates the width of image, L1Value
It is smaller, illustrate that the image rebuild and original image similarity are higher；
According to the value of optimization object function, the parameter of neural network is adjusted, and utilizes the neural network pair after adjusting parameter
First resolution sample image and its corresponding second resolution sample image are trained, and obtain image reconstruction model.
Referring to Fig. 2, image reconstruction model includes several layers residual error structure and one layer of super-pixel structure, if residual error structure includes
Dry convolutional layer and several nonlinear cascade activation primitives (CReLU) layer.It should be noted that residual error structure sheaf and common stacking
Convolution is compared, and can preferably learn the marginal information and texture information of image, this is beneficial to image super-resolution rebuilding.It is next
The input object of layer residual error structure includes the output object of upper one layer of residual error structure and the input pair of upper one layer of residual error structure
As.Image is up-sampled using super-pixel layer.Super-pixel layer can reduction calculation amount, improve image reconstruction model fortune
Scanning frequency degree.Residual error structure sheaf before super-pixel layer is there is no the size of first resolution image is changed, so scheming
As by before super-pixel layer, the size of residual error structure output image and the size of first resolution image are consistent.
Super-pixel structure sheaf expands the size of residual error structure output image, for example, the picture to a H × W amplifies r times, wherein H is represented
The height of picture, W represent the width of picture, this picture can become (rH) × (rW) into after crossing super-pixel layer.
It should be noted that the parameter distribution of the convolutional layer in convolutional neural networks can approximation regard symmetrical as
's.For example, some convolutional neural networks for being made of four convolutional layers, is replaced with residual error knot provided in this embodiment
Structure can regard the parameter that the convolutional layer of the convolutional neural networks is retained to half as, the other half parameter uses nonlinear cascade
Activation primitive is substituted, and in the present embodiment, replaces part using nonlinear cascade activation primitive (CReLU) in residual error structure
Convolutional layer reduces the parameter of image reconstruction model, and the calculating of nonlinear cascade activation primitive is simpler, and it is super to improve image
The speed of resolution reconstruction.
First resolution image input picture is rebuild mould by image super-resolution rebuilding method provided in an embodiment of the present invention
Type exports second resolution image.Image reconstruction model includes several layers residual error structure and one layer of super-pixel structure, residual error structure
It is made of several convolutional layers and several nonlinear cascade activation primitive layers.Image super-resolution rebuilding provided in an embodiment of the present invention
The residual error structure of method, image reconstruction model replaces part convolutional layer using cascade nonlinear activation function CReLU, reduces
The parameter of image reconstruction model improves the speed of image super-resolution rebuilding.
On the basis of the above embodiments, the first resolution sample image and its corresponding second resolution sample graph
The obtaining step of picture includes:
Down-sampling is carried out using second image of the resize function in matlab to selection, obtains the first image；Wherein,
Down-sampling principle: for piece image I having a size of M*N, s times of down-sampling is carried out to it and is divided to get to (M/s) * (N/s) size
Resolution image, wherein s is the common divisor of M and N.
The central area of the first image is intercepted, first resolution sample image is obtained；The central area of the second image is intercepted,
Obtain second resolution sample image.
It should be noted that since the pixel value of the point of image border can lose information, in the present embodiment, selection interception the
The central area of one image intercepts the central area of second image as first resolution sample image, obtains second point
Resolution sample image.Wherein, first resolution is less than second resolution.
On the basis of the various embodiments described above, referring to Fig. 2, image reconstruction model successively includes four layers of residual error structure and one layer
Super-pixel structure, wherein super-pixel structure is used to expand the size of residual error structure output image.
Residual error structure sheaf compared with common stacking convolution, believe by the marginal information and texture that can preferably learn image
Breath, this is beneficial to image super-resolution rebuilding.The input object of next layer of residual error structure includes the output of upper one layer of residual error structure
The input object of object and upper one layer of residual error structure.Image is up-sampled using super-pixel layer.Super-pixel layer can subtract
Few calculation amount improves the speed of service of image reconstruction model.There is no to first point for residual error structure sheaf before super-pixel layer
The size of resolution image is changed, so before image is by super-pixel layer, the size of residual error structure output image
It is consistent with the size of first resolution image.The size of super-pixel structure sheaf expansion residual error structure output image.This implementation
In example, first resolution image is inputted to image reconstruction model as shown in Figure 2, by four layers of residual error structure and one layer of super-pixel
Structure, output expand 4 times of second resolution image.First resolution is less than second resolution.
On the basis of the various embodiments described above, residual error structure successively includes the first convolutional layer, the first cascade nonlinear activation
Function layer, the second convolutional layer and the second cascade nonlinear activation function layer.
It should be noted that the parameter distribution of the convolutional layer in convolutional neural networks can approximation regard symmetrical as
's.For example, some convolutional neural networks for being made of four convolutional layers, is replaced with residual error knot provided in this embodiment
Structure can regard the parameter that the convolutional layer of the convolutional neural networks is retained to half as, the other half parameter uses nonlinear cascade
Activation primitive is substituted, and in the present embodiment, residual error structure is successively by including first convolutional layer, the first cascade nonlinear activation
Function layer, the second convolutional layer and the second cascade nonlinear activation function layer are constituted, and are substituted using nonlinear cascade activation primitive
Convolutional layer reduces the parameter of image reconstruction model.The calculating of nonlinear cascade activation primitive is simpler, and it is super to improve image
The speed of resolution reconstruction.
On the basis of the various embodiments described above, referring to Fig. 2, super-pixel structure successively includes third convolutional layer, super-pixel layer
With Volume Four lamination.
Specifically, it is up-sampled using image of the super-pixel layer to residual error structure output.What super-pixel layer can be reduced
Calculation amount improves the speed of service of image reconstruction model.There is no to first resolution for residual error structure sheaf before super-pixel layer
The size of image is changed, so in image by before super-pixel layer, the size of residual error structure output image and the
The size of one image in different resolution is consistent.Super-pixel structure sheaf expands the size of residual error structure output image, for example, to one
The picture of H × W amplifies r times, and wherein H represents the height of picture, and W represents the width of picture, this picture into mistake super-pixel layer it
Afterwards, (rH) × (rW) can be become.
Fig. 3 is the structural block diagram according to image super-resolution rebuilding device provided in an embodiment of the present invention；Referring to Fig.1, Fig. 2
And Fig. 3, the embodiment of the present invention also provide a kind of image super-resolution rebuilding device, comprising:
Image reconstruction module 301, for exporting second resolution figure for first resolution image input picture reconstruction model
Picture；Described image reconstruction model is instructed based on first resolution sample image and its corresponding second resolution sample image
It is obtained after white silk；Wherein, described image reconstruction model includes several layers residual error structure and one layer of super-pixel structure, the residual error knot
Structure includes several convolutional layers and several nonlinear cascade activation primitive layers.
Image reconstruction module 301 is using based on first resolution sample image and its corresponding second resolution sample image
First resolution image input picture reconstruction model is exported second resolution by the image reconstruction model obtained after being trained
Image.To make first resolution image super-resolution rebuilding.The obtaining step of image reconstruction model specifically includes:
Image and second resolution sample graph after first resolution sample image is rebuild are calculated by optimization object function
As the average value of the pixel difference absolute value of each point；
Optimization object function are as follows:
In formula, ISRIndicate the pixel for the image that first resolution sample image is obtained by neural network reconstruction, IHRIt indicates
The pixel of second resolution sample image, the port number of C representative image, L1Indicate that optimization object function, n indicate the number of image
Mesh, k indicate port number from 1 to C, and j indicates that the pixel of width, H indicate the height of image, and W indicates the width of image, L1Value
It is smaller, illustrate that the image rebuild and original image similarity are higher；
According to the value of optimization object function, the parameter of neural network is adjusted, and utilizes the neural network pair after adjusting parameter
First resolution sample image and its corresponding second resolution sample image are trained, and obtain image reconstruction model.
Referring to Fig. 2, image reconstruction model includes several layers residual error structure and one layer of super-pixel structure, if residual error structure includes
Dry convolutional layer and several nonlinear cascade activation primitive layers.It should be noted that residual error structure sheaf and common stacking convolution phase
Than can preferably learn the marginal information and texture information of image, this is beneficial to image super-resolution rebuilding.Next layer of residual error
The input object of structure includes the output object of upper one layer of residual error structure and the input object of upper one layer of residual error structure.It uses
Super-pixel layer up-samples image.Super-pixel layer can reduction calculation amount, improve image reconstruction model the speed of service.
Residual error structure sheaf before super-pixel layer is there is no the size of first resolution image is changed, so in image by super
Before pixel layer, the size of residual error structure output image and the size of first resolution image are consistent.Super-pixel knot
Structure layer expands the size of residual error structure output image, for example, the picture to a H × W amplifies r times, wherein H represents the height of picture
Degree, W represent the width of picture, this picture can become (rH) × (rW) into after crossing super-pixel layer.
It should be noted that the parameter distribution of the convolutional layer in convolutional neural networks can approximation regard symmetrical as
's.For example, some convolutional neural networks for being made of four convolutional layers, is replaced with residual error knot provided in this embodiment
Structure can regard the parameter that the convolutional layer of the convolutional neural networks is retained to half as, the other half parameter uses nonlinear cascade
Activation primitive is substituted, and in the present embodiment, replaces part using nonlinear cascade activation primitive (CReLU) in residual error structure
Convolutional layer reduces the parameter of image reconstruction model, and the calculating of nonlinear cascade activation primitive is simpler, and it is super to improve image
The speed of resolution reconstruction.
First resolution image input picture is rebuild mould by image super-resolution rebuilding device provided in an embodiment of the present invention
Type exports second resolution image.Image reconstruction model includes several layers residual error structure and one layer of super-pixel structure, residual error structure
It is made of several convolutional layers and several nonlinear cascade activation primitive layers.Image super-resolution rebuilding provided in an embodiment of the present invention
The residual error structure of method, image reconstruction model replaces part convolutional layer using cascade nonlinear activation function CReLU, reduces
The parameter of image reconstruction model improves the speed of image super-resolution rebuilding.
The embodiment of the invention provides a kind of electronic equipment, as shown in figure 4, the electronic equipment includes:
At least one processor (processor) 401, communication interface (Communications Interface) 404, extremely
A few memory (memory) 402 and communication bus 403, wherein at least one processor 801, communication interface 404, at least
One memory 402 completes mutual communication by communication bus 403.At least one processor 401 can call at least one
Logical order in a memory 402, to execute following method: by first resolution image input picture reconstruction model, output
Second resolution image；Described image reconstruction model is based on first resolution sample image and its corresponding second resolution sample
What this image obtained after being trained；Wherein, first resolution is less than second resolution；Image reconstruction model includes that several layers are residual
Poor structure and one layer of super-pixel structure, residual error structure include several convolutional layers and several nonlinear cascade activation primitive layers.
The embodiment of the invention also provides a kind of non-transient computer readable storage medium, non-transient computer readable storages
Medium storing computer instruction, the computer instruction make computer execute image super-resolution rebuilding provided by corresponding embodiment
Method, for example, by first resolution image input picture reconstruction model, export second resolution image；Described image weight
Established model is obtained after being trained based on first resolution sample image and its corresponding second resolution sample image；Its
In, first resolution is less than second resolution；Image reconstruction model includes several layers residual error structure and one layer of super-pixel structure, residual
Poor structure includes several convolutional layers and several nonlinear cascade activation primitive layers.
Those of ordinary skill in the art will appreciate that: realize that all or part of the steps of above method embodiment can pass through
The relevant hardware of program instruction is completed, and program above-mentioned can be stored in a computer readable storage medium, the program
When being executed, step including the steps of the foregoing method embodiments is executed；And storage medium above-mentioned includes: ROM, RAM, magnetic disk or light
The various media that can store program code such as disk.
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, those skilled in the art should understand that: it still may be used
To modify the technical solutions described in the foregoing embodiments or equivalent replacement of some of the technical features；
And these are modified or replaceed, technical solution of various embodiments of the present invention that it does not separate the essence of the corresponding technical solution spirit and
Range.
Claims (10)
1. a kind of image super-resolution rebuilding method characterized by comprising
By first resolution image input picture reconstruction model, second resolution image is exported；Described image reconstruction model is base
It is obtained after first resolution sample image and its corresponding second resolution sample image are trained；
Wherein, first resolution is less than second resolution, and described image reconstruction model includes that several layers residual error structure and one layer are super
Dot structure, the residual error structure include several convolutional layers and several nonlinear cascade activation primitive layers.
2. image super-resolution rebuilding method according to claim 1, which is characterized in that the first resolution sample graph
As and its obtaining step of corresponding second resolution sample image include:
Down-sampling is carried out using second image of the resize function in matlab to selection, obtains the first image；
The central area of the first image is intercepted, first resolution sample image is obtained；Intercept the center of second image
Region obtains second resolution sample image.
3. image super-resolution rebuilding method according to claim 1, which is characterized in that described image reconstruction model is successively
Including four layers of residual error structure and one layer of super-pixel structure, the super-pixel structure is used to expand the ruler of residual error structure output image
It is very little.
4. image super-resolution rebuilding method according to claim 1 or 3, which is characterized in that the residual error structure is successively
Including the first convolutional layer, the first cascade nonlinear activation function layer, the second convolutional layer and the second cascade nonlinear activation function layer.
5. image super-resolution rebuilding method according to claim 1, which is characterized in that the input of next layer of residual error structure
Object includes the output object of upper one layer of residual error structure and the input object of upper one layer of residual error structure.
6. image super-resolution rebuilding method according to claim 5, which is characterized in that the super-pixel structure is successively wrapped
Include third convolutional layer, super-pixel layer and Volume Four lamination.
7. image super-resolution rebuilding method according to claim 6, which is characterized in that described image reconstruction model obtains
The step is taken to include:
The image after first resolution sample image is rebuild is calculated by optimization object function and second resolution sample image is each
The average value of the pixel difference absolute value of point；
The optimization object function are as follows:
In formula, ISRIndicate the pixel for the image that first resolution sample image is obtained by neural network reconstruction, IHRIndicate second
The pixel of resolution ratio sample image, the port number of C representative image, L1Indicate that optimization object function, n indicate the number of image, k table
Show port number from 1 to C, j indicates that the pixel of width, H indicate the height of image, and W indicates the width of image, L1Value it is smaller,
Illustrate that the image rebuild and original image similarity are higher；
According to the value of optimization object function, the parameter of neural network is adjusted, to first resolution sample image and its corresponding
Two resolution ratio sample images are trained, and obtain image reconstruction model.
8. a kind of image super-resolution rebuilding device characterized by comprising
Image reconstruction module, for exporting second resolution image for first resolution image input picture reconstruction model；It is described
Image reconstruction model is obtained after being trained based on first resolution sample image and its corresponding second resolution sample image
It arrives；Wherein, first resolution is less than second resolution, and described image reconstruction model includes that several layers residual error structure and one layer are super
Dot structure, the residual error structure include several convolutional layers and several nonlinear cascade activation primitive layers.
9. a kind of electronic equipment characterized by comprising
At least one processor, at least one processor, communication interface and bus；Wherein,
The processor, memory, communication interface complete mutual communication by the bus；
The memory is stored with the program instruction that can be executed by the processor, and the processor calls described program instruction,
To execute method as described in any one of claim 1 to 7.
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 claims 1 to 7.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201811161605.3A CN109447900A (en) | 2018-09-30 | 2018-09-30 | A kind of image super-resolution rebuilding method and device |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201811161605.3A CN109447900A (en) | 2018-09-30 | 2018-09-30 | A kind of image super-resolution rebuilding method and device |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN109447900A true CN109447900A (en) | 2019-03-08 |
Family
ID=65546140
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201811161605.3A Pending CN109447900A (en) | 2018-09-30 | 2018-09-30 | A kind of image super-resolution rebuilding method and device |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN109447900A (en) |
Cited By (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN110490807A (en) * | 2019-08-27 | 2019-11-22 | 中国人民公安大学 | Image rebuilding method, device and storage medium |
| CN114066722A (en) * | 2021-11-03 | 2022-02-18 | 北京字节跳动网络技术有限公司 | Method and device for acquiring image and electronic equipment |
Citations (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| DE102005052061A1 (en) * | 2005-11-01 | 2007-05-16 | Carl Zeiss Imaging Solutions G | Method and device for image processing |
| CN107239799A (en) * | 2017-05-26 | 2017-10-10 | 西安电子科技大学 | Polarization SAR image classification method with depth residual error net is decomposed based on Pauli |
| CN107578377A (en) * | 2017-08-31 | 2018-01-12 | 北京飞搜科技有限公司 | A kind of super-resolution image reconstruction method and system based on deep learning |
-
2018
- 2018-09-30 CN CN201811161605.3A patent/CN109447900A/en active Pending
Patent Citations (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| DE102005052061A1 (en) * | 2005-11-01 | 2007-05-16 | Carl Zeiss Imaging Solutions G | Method and device for image processing |
| CN107239799A (en) * | 2017-05-26 | 2017-10-10 | 西安电子科技大学 | Polarization SAR image classification method with depth residual error net is decomposed based on Pauli |
| CN107578377A (en) * | 2017-08-31 | 2018-01-12 | 北京飞搜科技有限公司 | A kind of super-resolution image reconstruction method and system based on deep learning |
Cited By (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN110490807A (en) * | 2019-08-27 | 2019-11-22 | 中国人民公安大学 | Image rebuilding method, device and storage medium |
| CN114066722A (en) * | 2021-11-03 | 2022-02-18 | 北京字节跳动网络技术有限公司 | Method and device for acquiring image and electronic equipment |
| CN114066722B (en) * | 2021-11-03 | 2024-03-19 | 抖音视界有限公司 | Method and device for acquiring image and electronic equipment |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| Li et al. | MDCN: Multi-scale dense cross network for image super-resolution | |
| US10776982B2 (en) | Systems and methods for providing non-parametric texture synthesis of arbitrary shape and/or material data in a unified framework | |
| Qin et al. | Multi-scale feature fusion residual network for single image super-resolution | |
| CN111369440B (en) | Model training and image super-resolution processing method, device, terminal and storage medium | |
| Choi et al. | A deep convolutional neural network with selection units for super-resolution | |
| CN109389556A (en) | The multiple dimensioned empty convolutional neural networks ultra-resolution ratio reconstructing method of one kind and device | |
| Kim et al. | Accurate image super-resolution using very deep convolutional networks | |
| CN108921789A (en) | Super-resolution image reconstruction method based on recurrence residual error network | |
| CN109146788A (en) | Super-resolution image reconstruction method and device based on deep learning | |
| CN107154023A (en) | Face super-resolution reconstruction method based on generation confrontation network and sub-pix convolution | |
| US11823322B2 (en) | Utilizing voxel feature transformations for view synthesis | |
| CN108475415A (en) | Method and system for image procossing | |
| Claus et al. | Metal-artifact reduction using deep-learning based sinogram completion: initial results | |
| CN116977522A (en) | Rendering method and device of three-dimensional model, computer equipment and storage medium | |
| CN113744136A (en) | Image super-resolution reconstruction method and system based on channel constraint multi-feature fusion | |
| CN116309070A (en) | Super-resolution reconstruction method and device for hyperspectral remote sensing image and computer equipment | |
| CN104899835A (en) | Super-resolution processing method for image based on blind fuzzy estimation and anchoring space mapping | |
| CN110428382A (en) | A kind of efficient video Enhancement Method, device and storage medium for mobile terminal | |
| CN109447900A (en) | A kind of image super-resolution rebuilding method and device | |
| Muqeet et al. | Hybrid residual attention network for single image super resolution | |
| CN109146792A (en) | Chip image super resolution ratio reconstruction method based on deep learning | |
| CN109064408A (en) | A kind of method and device of multi-scale image super-resolution rebuilding | |
| Li et al. | High-resolution network for photorealistic style transfer | |
| CN109559278B (en) | Super resolution image reconstruction method and system based on multiple features study | |
| CN109242919A (en) | A kind of image down sampling method |
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: 20190308 |