CN109978766A - Image magnification method and image amplifying device - Google Patents
Image magnification method and image amplifying device Download PDFInfo
- Publication number
- CN109978766A CN109978766A CN201910185936.9A CN201910185936A CN109978766A CN 109978766 A CN109978766 A CN 109978766A CN 201910185936 A CN201910185936 A CN 201910185936A CN 109978766 A CN109978766 A CN 109978766A
- Authority
- CN
- China
- Prior art keywords
- image
- pixel
- training
- resolution
- interpolation
- 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 30
- 230000003321 amplification Effects 0.000 claims abstract description 82
- 238000003199 nucleic acid amplification method Methods 0.000 claims abstract description 82
- 238000002156 mixing Methods 0.000 claims abstract description 62
- 230000007704 transition Effects 0.000 claims abstract description 57
- 230000004927 fusion Effects 0.000 claims abstract description 41
- 238000012546 transfer Methods 0.000 claims abstract description 34
- 238000003708 edge detection Methods 0.000 claims abstract description 29
- 238000012549 training Methods 0.000 claims description 132
- 238000009499 grossing Methods 0.000 claims description 17
- 239000011159 matrix material Substances 0.000 claims description 12
- 238000010801 machine learning Methods 0.000 claims description 11
- 238000012545 processing Methods 0.000 claims description 6
- 230000000694 effects Effects 0.000 abstract description 16
- 238000005516 engineering process Methods 0.000 description 5
- 238000013461 design Methods 0.000 description 3
- 238000011161 development Methods 0.000 description 2
- 238000012986 modification Methods 0.000 description 2
- 230000004048 modification Effects 0.000 description 2
- 241000238097 Callinectes sapidus Species 0.000 description 1
- 230000009286 beneficial effect Effects 0.000 description 1
- 230000000903 blocking effect Effects 0.000 description 1
- 238000004891 communication Methods 0.000 description 1
- 238000010586 diagram Methods 0.000 description 1
- 239000004744 fabric Substances 0.000 description 1
- 238000007689 inspection Methods 0.000 description 1
- 238000012360 testing method 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
- G06T3/4007—Scaling of whole images or parts thereof, e.g. expanding or contracting based on interpolation, e.g. bilinear interpolation
-
- 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
- G06T3/403—Edge-driven scaling; Edge-based scaling
Landscapes
- Physics & Mathematics (AREA)
- General Physics & Mathematics (AREA)
- Engineering & Computer Science (AREA)
- Theoretical Computer Science (AREA)
- Image Processing (AREA)
- Editing Of Facsimile Originals (AREA)
Abstract
The present invention provides a kind of image magnification method and image amplifying device.The image magnification method includes the following steps: to obtain the original image with first resolution；Interpolation amplification is carried out to original image by preset first interpolation algorithm, obtains the First Transition image with second resolution, second resolution is greater than first resolution；Interpolation amplification is carried out to original image by preset second interpolation algorithm, and the image after interpolation amplification is smoothed, obtains second transfer image acquisition with second resolution；Edge detection is carried out to original image, obtains the marginal information of original image；Weight output model is established, and the marginal information of original image is inputted into weight output model, generates the blending weight of target image；According to blending weight and preset fusion formula fusion First Transition image and the second transfer image acquisition, the target image with second resolution is obtained, image border smooth transition is can be realized, promotes image amplification effect, reduce image amplification cost.
Description
Technical field
The present invention relates to field of display technology more particularly to a kind of image magnification methods and image amplifying device.
Background technique
With computer technology, the high speed development of modern communications technology enters today of information age, people in human society
Demand to image information it is also more more and more urgent.Current image digitazation input equipment is all by micro- on sampled images
Zonule generates corresponding pixel, to form a dot matrixed image data, i.e., for fixed image input condition
For fixed image, retrievable data volume is relatively-stationary.
And with the development of technology with the variation of the market demand, requirement of the consumer to the display quality of display device be more next
Higher, correspondingly, the resolution ratio of display device is also higher and higher, the video and demand signals of high-res are also increasing.But
It is still to have many video file and signal source resolution ratio lower at present, the video file and signal source of these low resolution
When high resolution display part is shown, need by enhanced processing.Since the performance of image amplification directly determines view
The quality of frequency display, therefore video display system improves the vision body of user there is an urgent need to the image magnification method of high quality
It tests.
Currently, common image magnification method is generally interpolation amplification, typical interpolation amplification method includes closest inserts
Value, bilinear interpolation, bicubic interpolation and polynomial interopolation etc., wherein closest interpolation algorithm is most simple, but closest interpolation
It is discontinuous that algorithm is also most also easy to produce pixel value, and so as to cause blocking artifact, it is fuzzy in turn result in image, picture quality effect after amplification
It is general not ideal enough.Bilinear interpolation algorithm is complex, and bilinear interpolation algorithm is not in the discontinuous situation of pixel value,
Amplified picture quality is higher, but the edge contour of each theme and detail section in image can be made to become to a certain extent
It must obscure, and bicubic interpolation and the algorithm of polynomial interopolation are more complicated.
Further, in the prior art, usually the flat site of image and fringe region are carried out using identical algorithm
Interpolation amplification often makes in the image finally presented that there are apparent sawtooth when selecting better simply algorithm to carry out operation
Or distortion；When selecting more complex algorithm to carry out operation, although being avoided that the distortion of image, entire calculating process is consumed
The time taken is longer, and the requirement to hardware is higher, cannot be considered in terms of image amplification effect and amplification cost.
Summary of the invention
The purpose of the present invention is to provide a kind of image magnification methods, can be realized image border smooth transition, promote figure
As amplification effect, image amplification cost is reduced.
The object of the invention is also to provide a kind of image amplifying devices, can be realized image border smooth transition, are promoted
Image amplification effect reduces image amplification cost.
To achieve the above object, the present invention provides a kind of image magnification method, include the following steps:
Step S1, the original image with first resolution is obtained；
Step S2, interpolation amplification is carried out to the original image by preset first interpolation algorithm, obtains that there is second point
The First Transition image of resolution, the second resolution are greater than first resolution；
Step S3, by preset second interpolation algorithm to the original image carry out interpolation amplification, and to interpolation amplification after
Image be smoothed, obtain second transfer image acquisition with second resolution；
Step S4, edge detection is carried out to the original image, obtains the marginal information of the original image；
Step S5, weight output model is established, and the marginal information of original image is inputted into weight output model, generates target
The blending weight of image；
Step S6, the First Transition image and the second transition figure are merged according to blending weight and preset fusion formula
Picture obtains the target image with second resolution.
First interpolation algorithm is closest interpolation, bilinear interpolation, bicubic interpolation or polynomial interopolation algorithm, institute
Stating the second interpolation algorithm is closest interpolation algorithm；
In the step S3 mode of smoothing processing be using preset smoothing operator to interpolation amplification in step S3 after
Image carries out convolution；
Wherein, the smoothing operator is any of matrix 1 to matrix 5:
The original image includes multiple original pixel of array arrangement, and the First Transition image includes the multiple of array arrangement
First pixel, second transfer image acquisition include multiple second pixels of array arrangement, and the target image includes array arrangement
Multiple object pixels；
In the step S4, the marginal information of the original image includes the edge letter of each original pixel in the original image
Breath；
In the step S5, the corresponding marginal information of each original pixel is inputted into weight output model, is generated and the preimage
The blending weight of the corresponding object pixel in position of element；
In the step S6, the preset fusion formula are as follows:
Vp=(1- λ) × Vcb+ λ × Vs；
Wherein, the Vp is the gray value of object pixel, and Vcb is the first pixel corresponding with the position of the object pixel
Gray value, Vs be the second pixel corresponding with the position of the object pixel gray value, λ be the object pixel fusion weigh
Value, 0≤λ≤1.
The step of establishing weight output model in the step S5 specifically includes: obtaining a plurality of training data, and according to institute
It states a plurality of training data and the weight output model is generated by machine learning training；
Wherein, the method for obtaining a plurality of training data are as follows:
The training image with first resolution is provided, the training image includes multiple trained pixels of array arrangement；
Edge detection is carried out to the training image, obtains the marginal information of each trained pixel；
Interpolation amplification is carried out to the training image by preset first interpolation algorithm, obtains that there is second resolution
First Transition training image；
Interpolation amplification is carried out to the training image by preset second interpolation algorithm, and to the image after interpolation amplification
It is smoothed, obtains the second lead-in training image with second resolution；
Multiple and different blending weights is chosen, merges institute according to the fusion formula and the plurality of different blending weight
First Transition training image and the second lead-in training image are stated, generates multiple training objective images with second resolution, often
One training objective image includes multiple training objective pixels of array arrangement；
There is provided the training image the corresponding standard target image with second resolution, the standard target image includes
Multiple standard target pixels of array arrangement；
Determine that the sum of the grayscale values of each standard target pixel is in each training of same position with the standard target pixel
The smallest training objective pixel of the difference of the gray value of object pixel；
To generate the blending weight of the smallest training objective pixel of the difference as the standard target pixel pair with the position
The standard fusion weight answered；
The a plurality of training data for respectively corresponding each standard target pixel is formed, each training data includes a standard target
The marginal information of the corresponding standard fusion weight of pixel and trained pixel corresponding with the standard target pixel.
The step S5 further include: the target image is divided into multiple regions, calculates each mesh in each region
Mark pixel blending weight mean value, and using the mean value as the region in each object pixel blending weight.
The present invention also provides a kind of image amplifying devices, comprising: acquiring unit, first to be connected with the acquiring unit are put
Big unit, the second amplifying unit being connected with the acquiring unit, the edge detection unit being connected with the acquiring unit and institute
The connected weight of edge detection unit is stated to generate unit and generate with first amplifying unit, the second amplifying unit and weight
The connected integrated unit of unit；
The acquiring unit is used to obtain the original image with first resolution；
First amplifying unit is used to carry out interpolation amplification to the original image by preset first interpolation algorithm, obtains
To the First Transition image with second resolution, the second resolution is greater than first resolution；
Second amplifying unit is used to carry out interpolation amplification to the original image by preset second interpolation algorithm, and
Image after interpolation amplification is smoothed, second transfer image acquisition with second resolution is obtained；
The edge detection unit is used to carry out edge detection to the original image, generates the edge letter of the original image
Breath；
The weight generates unit and exports for establishing weight output model, and by the marginal information of original image input weight
Model generates the blending weight of target image；
The integrated unit is used to merge the First Transition image and the according to blending weight and preset fusion formula
Two transfer image acquisitions obtain the target image with second resolution.
First interpolation algorithm is closest interpolation, bilinear interpolation, bicubic interpolation or polynomial interopolation, described the
Two interpolation algorithms are closest interpolation；
The mode that second amplifying unit is smoothed is using preset smoothing operator to by the second amplification
Image after unit interpolation amplification carries out convolution；
Wherein, the smoothing operator is any of matrix 1 to matrix 5:
The original image includes multiple original pixel of array arrangement, and the First Transition image includes the multiple of array arrangement
First pixel, second transfer image acquisition include multiple second pixels of array arrangement, and the target image includes array arrangement
Multiple object pixels；
The marginal information for the original image that the edge detection unit generates specifically includes each original in the original image
The marginal information of pixel；
The weight generates unit and the corresponding marginal information of each original pixel is inputted weight output model, generates and the original
The blending weight of the corresponding object pixel in the position of pixel；
Preset fusion formula in the integrated unit are as follows:
Vp=(1- λ) × Vcb+ λ × Vs；
Wherein, the Vp is the gray value of object pixel, and Vcb is the first pixel corresponding with the position of the object pixel
Gray value, Vs be the second pixel corresponding with the position of the object pixel gray value, λ be the object pixel fusion weigh
Value, 0≤λ≤1.
The weight generates unit by obtaining a plurality of training data, and passes through engineering according to a plurality of training data
It practises training and generates the weight output model；
Wherein, a plurality of training data of acquisition specifically includes:
The training image with first resolution is provided, the training image includes multiple trained pixels of array arrangement；
Edge detection is carried out to the training image, obtains the marginal information of each trained pixel；
Interpolation amplification is carried out to the training image by preset first interpolation algorithm, obtains that there is second resolution
First Transition training image；
Interpolation amplification is carried out to the training image by preset second interpolation algorithm, and to the image after interpolation amplification
It is smoothed, obtains the second lead-in training image with second resolution；
Multiple and different blending weights is chosen, merges institute according to the fusion formula and the plurality of different blending weight
First Transition training image and the second lead-in training image are stated, generates multiple training objective images with second resolution, often
One training objective image includes multiple training objective pixels of array arrangement；
There is provided the training image the corresponding standard target image with second resolution, the standard target image includes
Multiple standard target pixels of array arrangement；
Determine that the sum of the grayscale values of each standard target pixel is in each training of same position with the standard target pixel
The smallest training objective pixel of the difference of the gray value of object pixel；
To generate the blending weight of the smallest training objective pixel of the difference as the standard target pixel pair with the position
The standard fusion weight answered；
The a plurality of training data for respectively corresponding each standard target pixel is formed, each training data includes a standard target
The marginal information of the corresponding standard fusion weight of pixel and trained pixel corresponding with the standard target pixel.
The weight generates unit and is also used to the target image being divided into multiple regions, and calculates in each region
The mean value of the blending weight of each object pixel, and using the mean value as the region in each object pixel blending weight.
Beneficial effects of the present invention: the present invention provides a kind of image magnification method.Described image amplification method includes as follows
Step: the original image with first resolution is obtained；Interpolation is carried out to the original image by preset first interpolation algorithm to put
Greatly, the First Transition image with second resolution is obtained, the second resolution is greater than first resolution；Pass through preset
Two interpolation algorithms carry out interpolation amplification to the original image, and are smoothed to the image after interpolation amplification, are had
Second transfer image acquisition of second resolution；Edge detection is carried out to the original image, obtains the marginal information of the original image；It builds
Write value output model, and the marginal information of original image is inputted into weight output model, generate the blending weight of target image；Root
The First Transition image and the second transfer image acquisition are merged according to blending weight and preset fusion formula, obtains having second to differentiate
The target image of rate can be realized image border smooth transition, promote image amplification effect, reduces image amplification cost.This hair
It is bright that a kind of image amplifying device is also provided, it can be realized image border smooth transition, promote image amplification effect, reduce image and put
Big cost.
Detailed description of the invention
For further understanding of the features and technical contents of the present invention, it please refers to below in connection with of the invention detailed
Illustrate and attached drawing, however, the drawings only provide reference and explanation, is not intended to limit the present invention.
In attached drawing,
Fig. 1 is the flow chart of image magnification method of the invention；
Fig. 2 is the schematic diagram of image amplifying device of the invention.
Specific embodiment
Further to illustrate technological means and its effect adopted by the present invention, below in conjunction with preferred implementation of the invention
Example and its attached drawing are described in detail.
Referring to Fig. 1, the present invention provides a kind of image magnification method, include the following steps:
Step S1, the original image with first resolution is obtained.
Specifically, the original image includes multiple original pixel of array arrangement.
Step S2, interpolation amplification is carried out to the original image by preset first interpolation algorithm, obtains that there is second point
The First Transition image of resolution, the second resolution are greater than first resolution.
Specifically, first interpolation algorithm is closest interpolation, bilinear interpolation, bicubic interpolation or polynomial interopolation
Algorithm.
Further, the First Transition image includes multiple first pixels of array arrangement.
Step S3, by preset second interpolation algorithm to the original image carry out interpolation amplification, and to interpolation amplification after
Image be smoothed, obtain second transfer image acquisition with second resolution.
Specifically, second transfer image acquisition includes multiple second pixels of array arrangement.
Further, second interpolation algorithm is closest interpolation algorithm.
Specifically, the mode of the smoothing processing is using preset smoothing operator to the figure after interpolation amplification in step S3
As carrying out convolution；
Wherein, the smoothing operator is any of matrix 1 to matrix 5:
Step S4, edge detection is carried out to the original image, obtains the marginal information of the original image.
Specifically, edge detection is carried out to the original image by Sobel (Sobel) operator in the step S4.
Specifically, in the step S4, the marginal information of the original image includes each original pixel in the original image
Marginal information.
Step S5, weight output model is established, and the marginal information of original image is inputted into weight output model, generates target
The blending weight of image.
Specifically, the target image includes multiple object pixels of array arrangement.
Specifically, weight output model is established by machine learning in the step S5.
Further, the step of establishing weight output model in the step S5 specifically includes: a plurality of training data is obtained,
And the weight output model is generated by machine learning training according to a plurality of training data, the training data can be anti-
Being associated with for the marginal information of original image and the blending weight of target image is reflected, described in generating by machine learning training
Weight output model.
Specifically, the method for obtaining a plurality of training data are as follows:
The training image with first resolution is provided, the training image includes multiple trained pixels of array arrangement；
Edge detection is carried out to the training image, obtains the marginal information of each trained pixel；
Interpolation amplification is carried out to the training image by preset first interpolation algorithm, obtains that there is second resolution
First Transition training image；
Interpolation amplification is carried out to the training image by preset second interpolation algorithm, and to the image after interpolation amplification
It is smoothed, obtains the second lead-in training image with second resolution；
Multiple and different blending weights is chosen, merges institute according to the fusion formula and the plurality of different blending weight
First Transition training image and the second lead-in training image are stated, generates multiple training objective images with second resolution, often
One training objective image includes multiple training objective pixels of array arrangement；
There is provided the training image the corresponding standard target image with second resolution, the standard target image includes
Multiple standard target pixels of array arrangement；
Determine that the sum of the grayscale values of each standard target pixel is in each training of same position with the standard target pixel
The smallest training objective pixel of the difference of the gray value of object pixel；
To generate the blending weight of the smallest training objective pixel of the difference as the standard target pixel pair with the position
The standard fusion weight answered；
The a plurality of training data for respectively corresponding each standard target pixel is formed, each training data includes a standard target
The marginal information of the corresponding standard fusion weight of pixel and trained pixel corresponding with the standard target pixel.
Specifically, in the step S5, the corresponding marginal information of each original pixel is inputted into weight output model, generate with
The blending weight of the corresponding object pixel in the position of the original pixel.
Step S6, the First Transition image and the second transition figure are merged according to blending weight and preset fusion formula
Picture obtains the target image with second resolution.
Specifically, in the step S6, the preset fusion formula are as follows:
Vp=(1- λ) × Vcb+ λ × Vs；
Wherein, the Vp is the gray value of object pixel, and Vcb is the first pixel corresponding with the position of the object pixel
Gray value, Vs be the second pixel corresponding with the position of the object pixel gray value, λ be the object pixel fusion weigh
Value, 0≤λ≤1.
Specifically, the original pixel, the first pixel, the second pixel and object pixel include red component, green component
And blue component, respectively to the ash of the red component of original pixel, blue component and green component in the step S2 and step S3
Angle value is handled to obtain the gray value of first pixel and the second pixel red component, blue component and green component, and
In the step S6, the gray scale of first pixel and the second pixel red component, blue component and green component is merged respectively
Value obtains the gray value of the red component of object pixel, green component and blue component.
Further, image transition is gentler, in the step S5 the blending weight for obtaining each object pixel it
Afterwards, include the steps that asking equal, the target image is specially divided into multiple regions, calculate each in each region
The mean value of the blending weight of object pixel, and using the mean value as the region in each object pixel blending weight, preferably
Ground, each region include 3 × 3 pixels.
It should be noted that of the invention amplifies original image generation First Transition image and second by two methods respectively
Transfer image acquisition, wherein First Transition image is more preferable for flat site amplification effect, and the second transfer image acquisition is directed to marginal zone
The amplification effect in domain is more preferable, then establishes a weight output model by machine learning algorithm, and output one and the edge of original image are believed
Relevant blending weight is ceased, the First Transition image and the second transfer image acquisition are merged, when object pixel is biased to fringe region,
The accounting of the second transfer image acquisition is larger when then First Transition image is merged with the second transfer image acquisition, when object pixel is biased to flat region
When domain, then the accounting of First Transition image is larger when First Transition image is merged with the second transfer image acquisition, can be realized image side
Edge smooth transition, promotes image amplification effect, reduces image amplification cost, conceptual design is simple, is easy to be fabricated to corresponding core
Piece, and cost is relatively low, the blending weight accuracy exported by the weight output model that machine learning is established is higher.
Referring to Fig. 2, the present invention also provides a kind of image amplifying devices, comprising: acquiring unit 10 and the acquiring unit
10 the first connected amplifying units 20, the second amplifying unit 30 being connected with the acquiring unit 10 and the acquiring unit 10
Connected edge detection unit 40, the weight being connected with the edge detection unit 40 generate unit 50 and put with described first
Big unit 20, the second amplifying unit 30 and weight generate the connected integrated unit 60 of unit 50；
The acquiring unit 10 is used to obtain the original image with first resolution；
First amplifying unit 20 is used to carry out interpolation amplification to the original image by preset first interpolation algorithm,
The First Transition image with second resolution is obtained, the second resolution is greater than first resolution；
Second amplifying unit 30 is used to carry out interpolation amplification to the original image by preset second interpolation algorithm,
And the image after interpolation amplification is smoothed, obtain second transfer image acquisition with second resolution；
The edge detection unit 40 is used to carry out edge detection to the original image, generates the edge letter of the original image
Breath；
The weight generates unit 50 for establishing weight output model, and the marginal information of original image input weight is defeated
Model out generates the blending weight of target image；
The integrated unit 60 be used to be merged according to blending weight and preset fusion formula the First Transition image and
Second transfer image acquisition obtains the target image with second resolution.
Specifically, first interpolation algorithm is that closest interpolation, bilinear interpolation, bicubic interpolation or multinomial are inserted
Value, second interpolation algorithm are closest interpolation；
The mode that second amplifying unit 30 is smoothed is to be put using preset smoothing operator to by second
Image after 30 interpolation amplification of big unit carries out convolution；
Wherein, the smoothing operator is any of matrix 1 to matrix 5:
Specifically, the edge detection unit 40 carries out edge inspection to the original image by Sobel (Sobel) operator
It surveys.
Specifically, the original image includes multiple original pixel of array arrangement, and the First Transition image includes array row
Multiple first pixels of cloth, second transfer image acquisition include multiple second pixels of array arrangement, and the target image includes
Multiple object pixels of array arrangement；
The marginal information for the original image that the edge detection unit 40 generates specifically includes each in the original image
The marginal information of original pixel；
The weight generates unit 50 and the corresponding marginal information of each original pixel is inputted weight output model, generates and is somebody's turn to do
The blending weight of the corresponding object pixel in the position of original pixel；
Preset fusion formula in the integrated unit 60 are as follows:
Vp=1- λ × Vcb+ λ × Vs；
Wherein, the Vp is the gray value of object pixel, and Vcb is the first pixel corresponding with the position of the object pixel
Gray value, Vs be the second pixel corresponding with the position of the object pixel gray value, λ be the object pixel fusion weigh
Value, 0≤λ≤1.
Specifically, the weight generates unit 50 by obtaining a plurality of training data, and according to a plurality of training data
The weight output model is generated by machine learning training；
Wherein, a plurality of training data of acquisition specifically includes:
The training image with first resolution is provided, the training image includes multiple trained pixels of array arrangement；
Edge detection is carried out to the training image, obtains the marginal information of each trained pixel；
Interpolation amplification is carried out to the training image by preset first interpolation algorithm, obtains that there is second resolution
First Transition training image；
Interpolation amplification is carried out to the training image by preset second interpolation algorithm, and to the image after interpolation amplification
It is smoothed, obtains the second lead-in training image with second resolution；
Multiple and different blending weights is chosen, merges institute according to the fusion formula and the plurality of different blending weight
First Transition training image and the second lead-in training image are stated, generates multiple training objective images with second resolution, often
One training objective image includes multiple training objective pixels of array arrangement；
There is provided the training image the corresponding standard target image with second resolution, the standard target image includes
Multiple standard target pixels of array arrangement；
Determine that the sum of the grayscale values of each standard target pixel is in each training of same position with the standard target pixel
The smallest training objective pixel of the difference of the gray value of object pixel；
To generate the blending weight of the smallest training objective pixel of the difference as the standard target pixel pair with the position
The standard fusion weight answered；
The a plurality of training data for respectively corresponding each standard target pixel is formed, each training data includes a standard target
The marginal information of the corresponding standard fusion weight of pixel and trained pixel corresponding with the standard target pixel.
Specifically, the original pixel, the first pixel, the second pixel and object pixel include red component, green component
And blue component, respectively to the red component of original pixel, blue point in first amplifying unit 20 and the second amplifying unit 30
The gray value of amount and green component is handled to obtain first pixel and the second pixel red component, blue component and green
The gray value of component, and the integrated unit 60 merges first pixel and the second pixel red component, blue component respectively
And the gray value of green component obtains the gray value of the red component of object pixel, green component and blue component.
Further, image transition is gentler, in the step S5 the blending weight for obtaining each object pixel it
Afterwards, include the steps that asking equal, the target image is specially divided into multiple regions, calculate each in each region
The mean value of the blending weight of object pixel, and using the mean value as the region in each object pixel blending weight, preferably
Ground, each region include 3 × 3 pixels.
It should be noted that of the invention amplifies original image generation First Transition image and second by two methods respectively
Transfer image acquisition, wherein First Transition image is more preferable for flat site amplification effect, and the second transfer image acquisition is directed to marginal zone
The amplification effect in domain is more preferable, then establishes a weight output model by machine learning algorithm, and output one and the edge of original image are believed
Relevant blending weight is ceased, the First Transition image and the second transfer image acquisition are merged, when object pixel is biased to fringe region,
The accounting of the second transfer image acquisition is larger when then First Transition image is merged with the second transfer image acquisition, when object pixel is biased to flat region
When domain, then the accounting of First Transition image is larger when First Transition image is merged with the second transfer image acquisition, can be realized image side
Edge smooth transition, promotes image amplification effect, reduces image amplification cost, conceptual design is simple, is easy to be fabricated to corresponding core
Piece, and cost is relatively low, the blending weight accuracy exported by the weight output model that machine learning is established is higher.
In conclusion the present invention provides a kind of image magnification method.Described image amplification method includes the following steps: to obtain
Original image with first resolution；Interpolation amplification is carried out to the original image by preset first interpolation algorithm, is had
There is the First Transition image of second resolution, the second resolution is greater than first resolution；It is calculated by preset second interpolation
Method carries out interpolation amplification to the original image, and is smoothed to the image after interpolation amplification, obtains having second to differentiate
Second transfer image acquisition of rate；Edge detection is carried out to the original image, obtains the marginal information of the original image；It is defeated to establish weight
Model out, and the marginal information of original image is inputted into weight output model, generate the blending weight of target image；It is weighed according to fusion
Value and preset fusion formula merge the First Transition image and the second transfer image acquisition, obtain the target with second resolution
Image can be realized image border smooth transition, promote image amplification effect, reduces image amplification cost.The present invention also provides
A kind of image amplifying device can be realized image border smooth transition, promote image amplification effect, reduces image amplification cost.
The above for those of ordinary skill in the art can according to the technique and scheme of the present invention and technology
Other various corresponding changes and modifications are made in design, and all these change and modification all should belong to the claims in the present invention
Protection scope.
Claims (10)
1. a kind of image magnification method, which comprises the steps of:
Step S1, the original image with first resolution is obtained；
Step S2, interpolation amplification is carried out to the original image by preset first interpolation algorithm, obtained with second resolution
First Transition image, the second resolution be greater than first resolution；
Step S3, interpolation amplification is carried out to the original image by preset second interpolation algorithm, and to the figure after interpolation amplification
As being smoothed, second transfer image acquisition with second resolution is obtained；
Step S4, edge detection is carried out to the original image, obtains the marginal information of the original image；
Step S5, weight output model is established, and the marginal information of original image is inputted into weight output model, generates target image
Blending weight；
Step S6, the First Transition image and the second transfer image acquisition are merged according to blending weight and preset fusion formula, obtained
To the target image with second resolution.
2. image magnification method as described in claim 1, which is characterized in that first interpolation algorithm be closest interpolation,
Bilinear interpolation, bicubic interpolation or polynomial interopolation algorithm, second interpolation algorithm are closest interpolation algorithm；
The mode of smoothing processing is using preset smoothing operator to the image after interpolation amplification in step S3 in the step S3
Carry out convolution；
Wherein, the smoothing operator is any of matrix 1 to matrix 5:
3. image magnification method as described in claim 1, which is characterized in that the original image includes multiple originals of array arrangement
Pixel, the First Transition image include multiple first pixels of array arrangement, and second transfer image acquisition includes array arrangement
Multiple second pixels, the target image includes multiple object pixels of array arrangement；
In the step S4, the marginal information of the original image includes the marginal information of each original pixel in the original image；
In the step S5, the corresponding marginal information of each original pixel is inputted into weight output model, is generated and the original pixel
The blending weight of the corresponding object pixel in position；
In the step S6, the preset fusion formula are as follows:
Vp=(1- λ) × Vcb+ λ × Vs；
Wherein, the Vp is the gray value of object pixel, and Vcb is the ash of the first pixel corresponding with the position of the object pixel
Angle value, Vs be the second pixel corresponding with the position of the object pixel gray value, λ be the object pixel blending weight, 0
≤λ≤1。
4. image magnification method as claimed in claim 3, which is characterized in that establish weight output model in the step S5
Step specifically includes: obtaining a plurality of training data, and is passed through described in machine learning training generation according to a plurality of training data
Weight output model；
Wherein, the method for obtaining a plurality of training data are as follows:
The training image with first resolution is provided, the training image includes multiple trained pixels of array arrangement；
Edge detection is carried out to the training image, obtains the marginal information of each trained pixel；
Interpolation amplification is carried out to the training image by preset first interpolation algorithm, obtains first with second resolution
Lead-in training image；
Interpolation amplification is carried out to the training image by preset second interpolation algorithm, and the image after interpolation amplification is carried out
Smoothing processing obtains the second lead-in training image with second resolution；
Multiple and different blending weights is chosen, according to the fusion formula and the plurality of different blending weight fusion described the
One lead-in training image and the second lead-in training image generate multiple training objective images with second resolution, Mei Yixun
Practice multiple training objective pixels that target image includes array arrangement；
There is provided the training image the corresponding standard target image with second resolution, the standard target image includes array
Multiple standard target pixels of arrangement；
Determine that the sum of the grayscale values of each standard target pixel is in each training objective of same position with the standard target pixel
The smallest training objective pixel of the difference of the gray value of pixel；
To generate the blending weight of the smallest training objective pixel of the difference as corresponding with the standard target pixel of the position
Standard fusion weight；
The a plurality of training data for respectively corresponding each standard target pixel is formed, each training data includes a standard target pixel
Corresponding standard fusion weight and trained pixel corresponding with the standard target pixel marginal information.
5. image magnification method as claimed in claim 3, which is characterized in that the step S5 further include: by the target figure
As being divided into multiple regions, calculate the mean value of the blending weight of each object pixel in each region, and using the mean value as
The blending weight of each object pixel in the region.
6. a kind of image amplifying device characterized by comprising acquiring unit (10) is connected with the acquiring unit (10)
First amplifying unit (20), the second amplifying unit (30) being connected with the acquiring unit (10), with the acquiring unit (10)
Connected edge detection unit (40), the weight being connected with the edge detection unit (40) generate unit (50) and with it is described
First amplifying unit (20), the second amplifying unit (30) and weight generate unit (50) connected integrated unit (60)；
The acquiring unit (10) is used to obtain the original image with first resolution；
First amplifying unit (20) is used to carry out interpolation amplification to the original image by preset first interpolation algorithm, obtains
To the First Transition image with second resolution, the second resolution is greater than first resolution；
Second amplifying unit (30) is used to carry out interpolation amplification to the original image by preset second interpolation algorithm, and
Image after interpolation amplification is smoothed, second transfer image acquisition with second resolution is obtained；
The edge detection unit (40) is used to carry out edge detection to the original image, generates the edge letter of the original image
Breath；
The weight generates unit (50) and exports for establishing weight output model, and by the marginal information of original image input weight
Model generates the blending weight of target image；
The integrated unit (60) is used to merge the First Transition image and the according to blending weight and preset fusion formula
Two transfer image acquisitions obtain the target image with second resolution.
7. image amplifying device as claimed in claim 6, which is characterized in that first interpolation algorithm be closest interpolation,
Bilinear interpolation, bicubic interpolation or polynomial interopolation, second interpolation algorithm are closest interpolation；
The mode that second amplifying unit (30) is smoothed is using preset smoothing operator to by the second amplification
Image after unit (30) interpolation amplification carries out convolution；
Wherein, the smoothing operator is any of matrix 1 to matrix 5:
8. image amplifying device as claimed in claim 6, which is characterized in that the original image includes multiple originals of array arrangement
Pixel, the First Transition image include multiple first pixels of array arrangement, and second transfer image acquisition includes array arrangement
Multiple second pixels, the target image includes multiple object pixels of array arrangement；
The marginal information for the original image that the edge detection unit (40) generates specifically includes each original in the original image
The marginal information of pixel；
The weight generates unit (50) and the corresponding marginal information of each original pixel is inputted weight output model, generates and the original
The blending weight of the corresponding object pixel in the position of pixel；
Preset fusion formula in the integrated unit (60) are as follows:
Vp=(1- λ) × Vcb+ λ × Vs；
Wherein, the Vp is the gray value of object pixel, and Vcb is the ash of the first pixel corresponding with the position of the object pixel
Angle value, Vs be the second pixel corresponding with the position of the object pixel gray value, λ be the object pixel blending weight, 0
≤λ≤1。
9. image amplifying device as claimed in claim 8, which is characterized in that it is more by obtaining that the weight generates unit (50)
Training data, and machine learning training is passed through according to a plurality of training data and generates the weight output model；
Wherein, a plurality of training data of acquisition specifically includes:
The training image with first resolution is provided, the training image includes multiple trained pixels of array arrangement；
Edge detection is carried out to the training image, obtains the marginal information of each trained pixel；
Interpolation amplification is carried out to the training image by preset first interpolation algorithm, obtains first with second resolution
Lead-in training image；
Interpolation amplification is carried out to the training image by preset second interpolation algorithm, and the image after interpolation amplification is carried out
Smoothing processing obtains the second lead-in training image with second resolution；
Multiple and different blending weights is chosen, according to the fusion formula and the plurality of different blending weight fusion described the
One lead-in training image and the second lead-in training image generate multiple training objective images with second resolution, Mei Yixun
Practice multiple training objective pixels that target image includes array arrangement；
There is provided the training image the corresponding standard target image with second resolution, the standard target image includes array
Multiple standard target pixels of arrangement；
Determine that the sum of the grayscale values of each standard target pixel is in each training objective of same position with the standard target pixel
The smallest training objective pixel of the difference of the gray value of pixel；
To generate the blending weight of the smallest training objective pixel of the difference as corresponding with the standard target pixel of the position
Standard fusion weight；
The a plurality of training data for respectively corresponding each standard target pixel is formed, each training data includes a standard target pixel
Corresponding standard fusion weight and trained pixel corresponding with the standard target pixel marginal information.
10. image amplifying device as claimed in claim 8, which is characterized in that the weight generate unit (50) be also used to by
The target image is divided into multiple regions, and calculates the mean value of the blending weight of each object pixel in each region, and
The blending weight of each object pixel in using the mean value as the region.
Priority Applications (2)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201910185936.9A CN109978766B (en) | 2019-03-12 | 2019-03-12 | Image amplifying method and image amplifying device |
| PCT/CN2019/085764 WO2020181641A1 (en) | 2019-03-12 | 2019-05-07 | Image magnification method and image magnification device |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201910185936.9A CN109978766B (en) | 2019-03-12 | 2019-03-12 | Image amplifying method and image amplifying device |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN109978766A true CN109978766A (en) | 2019-07-05 |
| CN109978766B CN109978766B (en) | 2020-10-16 |
Family
ID=67078601
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201910185936.9A Active CN109978766B (en) | 2019-03-12 | 2019-03-12 | Image amplifying method and image amplifying device |
Country Status (2)
| Country | Link |
|---|---|
| CN (1) | CN109978766B (en) |
| WO (1) | WO2020181641A1 (en) |
Cited By (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN112308808A (en) * | 2019-07-31 | 2021-02-02 | 北京金山云网络技术有限公司 | Image processing method and device and electronic equipment |
| CN112381714A (en) * | 2020-10-30 | 2021-02-19 | 南阳柯丽尔科技有限公司 | Image processing method, device, storage medium and equipment |
Citations (6)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US20050078883A1 (en) * | 2003-10-08 | 2005-04-14 | Yi Jong-Hyon | Digital image processing device and method |
| CN101911112A (en) * | 2007-12-25 | 2010-12-08 | 日本电气株式会社 | Image processing device, image processing method, image decompressing device, image compressing device, image transmission system, and storage medium |
| CN102800069A (en) * | 2012-05-22 | 2012-11-28 | 湖南大学 | Image super-resolution method for combining soft decision self-adaptation interpolation and bicubic interpolation |
| CN102842111A (en) * | 2012-07-09 | 2012-12-26 | 许丹 | Enlarged image compensation method and device |
| EP1947603A3 (en) * | 2007-01-22 | 2013-08-07 | Sharp Kabushiki Kaisha | Image upsampling technique |
| CN104299185A (en) * | 2014-09-26 | 2015-01-21 | 京东方科技集团股份有限公司 | Image magnification method, image magnification device and display device |
Family Cites Families (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US8180180B2 (en) * | 2008-12-29 | 2012-05-15 | Arcsoft Hangzhou Co., Ltd. | Method for magnifying images and videos |
| CN106204454B (en) * | 2016-01-26 | 2019-06-21 | 西北工业大学 | High-precision and fast image interpolation method based on texture edge adaptive data fusion |
| CN106709875B (en) * | 2016-12-30 | 2020-02-18 | 北京工业大学 | A Compressed Low-Resolution Image Restoration Method Based on Joint Deep Network |
-
2019
- 2019-03-12 CN CN201910185936.9A patent/CN109978766B/en active Active
- 2019-05-07 WO PCT/CN2019/085764 patent/WO2020181641A1/en active Application Filing
Patent Citations (6)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US20050078883A1 (en) * | 2003-10-08 | 2005-04-14 | Yi Jong-Hyon | Digital image processing device and method |
| EP1947603A3 (en) * | 2007-01-22 | 2013-08-07 | Sharp Kabushiki Kaisha | Image upsampling technique |
| CN101911112A (en) * | 2007-12-25 | 2010-12-08 | 日本电气株式会社 | Image processing device, image processing method, image decompressing device, image compressing device, image transmission system, and storage medium |
| CN102800069A (en) * | 2012-05-22 | 2012-11-28 | 湖南大学 | Image super-resolution method for combining soft decision self-adaptation interpolation and bicubic interpolation |
| CN102842111A (en) * | 2012-07-09 | 2012-12-26 | 许丹 | Enlarged image compensation method and device |
| CN104299185A (en) * | 2014-09-26 | 2015-01-21 | 京东方科技集团股份有限公司 | Image magnification method, image magnification device and display device |
Non-Patent Citations (2)
| Title |
|---|
| 王会鹏等: "《一种基于区域的双三次图像插值算法》", 《计算机工程》 * |
| 赵旦峰等: "《基于阈值判断的区域指导插值算法》", 《系统工程与电子技术》 * |
Cited By (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN112308808A (en) * | 2019-07-31 | 2021-02-02 | 北京金山云网络技术有限公司 | Image processing method and device and electronic equipment |
| CN112308808B (en) * | 2019-07-31 | 2025-01-10 | 北京金山云网络技术有限公司 | Image processing method, device and electronic equipment |
| CN112381714A (en) * | 2020-10-30 | 2021-02-19 | 南阳柯丽尔科技有限公司 | Image processing method, device, storage medium and equipment |
Also Published As
| Publication number | Publication date |
|---|---|
| CN109978766B (en) | 2020-10-16 |
| WO2020181641A1 (en) | 2020-09-17 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| US7791652B2 (en) | Image processing apparatus, image capture apparatus, image output apparatus, and method and program for these apparatus | |
| EP3076364B1 (en) | Image filtering based on image gradients | |
| US7792384B2 (en) | Image processing apparatus, image processing method, program, and recording medium therefor | |
| CN110264426B (en) | Image distortion correction method and device | |
| US10614551B2 (en) | Image interpolation methods and related image interpolation devices thereof | |
| US20130094781A1 (en) | Image processing apparatus | |
| US6813041B1 (en) | Method and apparatus for performing local color correction | |
| US10255665B2 (en) | Image processing device and method, image capturing device, program, and record medium | |
| JP4498361B2 (en) | How to speed up Retinex-type algorithms | |
| CN103942755A (en) | Image brightness adjusting method and device | |
| JP2006013558A (en) | Image processing apparatus and image processing program | |
| KR102337835B1 (en) | Directional Scaling Systems and Methods | |
| JP2004165840A (en) | Image processing program | |
| US7751642B1 (en) | Methods and devices for image processing, image capturing and image downscaling | |
| US20090022402A1 (en) | Image-resolution-improvement apparatus and method | |
| CN109978766A (en) | Image magnification method and image amplifying device | |
| TW201536029A (en) | Image downsampling apparatus and method | |
| US9928577B2 (en) | Image correction apparatus and image correction method | |
| JP4369030B2 (en) | Image correction method and apparatus, and computer-readable recording medium storing image correction program | |
| JP4992379B2 (en) | Image gradation conversion apparatus, program, electronic camera, and method thereof | |
| JP4147155B2 (en) | Image processing apparatus and method | |
| KR101585187B1 (en) | Image Processing Method and Apparatus for Integrated Multi-scale Retinex Based on CIELAB Color Space for Preserving Color | |
| TWI413019B (en) | Image adjusting circuit and image adjusting method | |
| JP4032200B2 (en) | Image data interpolation method, image data interpolation device, and computer readable recording medium recording image data interpolation program | |
| CN102447817B (en) | Image processing device and space image noise eliminating 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 | ||
| GR01 | Patent grant | ||
| GR01 | Patent grant | ||
| CP01 | Change in the name or title of a patent holder |
Address after: 9-2 Tangming Avenue, Guangming New District, Shenzhen City, Guangdong Province Patentee after: TCL China Star Optoelectronics Technology Co.,Ltd. Address before: 9-2 Tangming Avenue, Guangming New District, Shenzhen City, Guangdong Province Patentee before: Shenzhen China Star Optoelectronics Technology Co.,Ltd. |
|
| CP01 | Change in the name or title of a patent holder |