CN107169925B - The method for reconstructing of stepless zooming super-resolution image - Google Patents
The method for reconstructing of stepless zooming super-resolution image Download PDFInfo
- Publication number
- CN107169925B CN107169925B CN201710267753.2A CN201710267753A CN107169925B CN 107169925 B CN107169925 B CN 107169925B CN 201710267753 A CN201710267753 A CN 201710267753A CN 107169925 B CN107169925 B CN 107169925B
- Authority
- CN
- China
- Prior art keywords
- resolution
- image
- low
- super
- dictionary
- 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.)
- Active
Links
- 238000000034 method Methods 0.000 title claims abstract description 62
- 238000012549 training Methods 0.000 claims abstract description 23
- 238000012545 processing Methods 0.000 claims abstract description 14
- 238000001914 filtration Methods 0.000 claims description 18
- 230000009467 reduction Effects 0.000 claims description 12
- 239000011159 matrix material Substances 0.000 claims description 11
- 239000000284 extract Substances 0.000 claims description 5
- 230000015556 catabolic process Effects 0.000 claims description 4
- 238000006731 degradation reaction Methods 0.000 claims description 4
- 238000003708 edge detection Methods 0.000 claims description 4
- 238000005070 sampling Methods 0.000 claims description 4
- 230000017105 transposition Effects 0.000 claims description 4
- 230000006837 decompression Effects 0.000 claims description 3
- 238000004321 preservation Methods 0.000 claims description 3
- 238000013144 data compression Methods 0.000 claims description 2
- 230000008569 process Effects 0.000 abstract description 8
- 230000000694 effects Effects 0.000 abstract description 7
- 238000013507 mapping Methods 0.000 abstract description 5
- 230000008859 change Effects 0.000 abstract description 3
- 230000000052 comparative effect Effects 0.000 abstract description 3
- 238000012360 testing method Methods 0.000 abstract description 3
- 239000013598 vector Substances 0.000 description 5
- 238000010586 diagram Methods 0.000 description 2
- 238000000605 extraction Methods 0.000 description 2
- 230000006870 function Effects 0.000 description 2
- 238000003384 imaging method Methods 0.000 description 2
- 238000012634 optical imaging Methods 0.000 description 2
- 230000007613 environmental effect Effects 0.000 description 1
- 230000004048 modification Effects 0.000 description 1
- 238000012986 modification Methods 0.000 description 1
- 238000007781 pre-processing Methods 0.000 description 1
- 239000004576 sand Substances 0.000 description 1
- 230000011218 segmentation Effects 0.000 description 1
- 238000009394 selective breeding Methods 0.000 description 1
- 230000009466 transformation Effects 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
Landscapes
- Physics & Mathematics (AREA)
- General Physics & Mathematics (AREA)
- Engineering & Computer Science (AREA)
- Theoretical Computer Science (AREA)
- Editing Of Facsimile Originals (AREA)
- Image Processing (AREA)
Abstract
The present invention relates to a kind of method for reconstructing of stepless zooming super-resolution image, comprising: three dictionary training process, image reconstruction process, dimensional variation process processing steps.The present invention proposes the method for establishing image library training dictionary, by image set off-line training dictionary library, obtains the mapping relations of high-definition picture and low-resolution image, optimizes image reconstruction step, effectively reduce method runing time, has adaptivity and high efficiency.Threshold value in the present invention in method is not needed to be changed during place, in the case where not influencing image reconstruction effect, effectively raises image reconstruction efficiency based on fixation is optimized after a large amount of comparative tests.The super-resolution rebuilding image that the present invention obtains can carry out the change of scale of arbitrary size.It can satisfy demand in the case where difference in engineer application, there is preferable applicability.
Description
Technical field
The invention belongs to Image Reconstruction Technology fields, and in particular to a kind of reconstruction side of stepless zooming super-resolution image
Method.
Background technique
Currently, image resolution ratio will receive the influence of the factors such as imaging system, shooting environmental, and obtained image resolution ratio is past
Toward the demand for being unable to satisfy people.However, optical imaging system and its detector are limited to technical level at present and processing is multiple again
Miscellaneous degree can not improve system imaging quality from hardware point of view.Therefore, figure has been broken in the appearance of image super-resolution rebuilding technology
As the limitation of resolution ratio.In such a way that software is handled from a width or several low-resolution images degenerated, one is calculated
Panel height image in different resolution obtains detailed information more abundant.Super-resolution rebuilding technology is widely used in image procossing at present
And optical imaging field, obtain the attention of people.
Currently, the rarefaction representation image super-resolution rebuilding algorithm based on study can effectively improve the resolution ratio of image,
And it amplifies the scale that multiplying power is 3 to image to amplify.However, in certain special cases, the height obtained by this method
Required scale requirement is not achieved in image in different resolution.
Therefore, how to provide a kind of image rebuilding method allows the super-resolution rebuilding image obtained to carry out any ruler
The transformation of degree requires to become most important to reach related industries.
Summary of the invention
In order to solve the above-mentioned problems in the prior art, the present invention provides a kind of stepless zooming super-resolution images
Method for reconstructing, include the following steps:
Step 1, Fuzzy Processing and N times of down-sampling are carried out according to degradation model to high-resolution sample image, obtains low point
Resolution sample image；
Step 2, using K-SVD method, dictionary training is carried out using the low resolution sample image, obtains high-resolution word
Allusion quotation and low-resolution dictionary；
Step 3, low-resolution image to be reconstructed is pre-processed to obtain several low-resolution image blocks；
Step 4, fixed threshold is calculated using the low resolution sample image；
Step 5, judge whether the pixel value of each low-resolution image block is less than the fixed threshold；If so,
Then determine that the low-resolution image block is Poor information image block；If it is not, then determining that the low-resolution image block is
High information quantity image block；
Step 6, the high information quantity image block is used and rebuilds to obtain high letter based on dictionary learning and rarefaction representation algorithm
Breath amount rebuilds subregion；
Step 7, the Poor information image block is rebuild to obtain Poor information reconstruction subregion using interpolation algorithm；
Step 8, the high information quantity subregion subregion progress image mosaic is rebuild with the Poor information to be surpassed
Resolution reconstruction image；
Step 9, using interpolation algorithm by the super-resolution rebuilding image progress M times of interpolation obtain N*M (N=2,3,4；M
> 0) super-resolution rebuilding image again.
In one embodiment of the invention, the step 2 includes the following steps:
(21) characteristics of image that the low resolution sample image is extracted by feature extracting method, obtains extraterrestrial target
High-resolution features information and low resolution characteristic information.
(22) utilize the K-SVD method, to the high-resolution features information and the low resolution characteristic information into
Row joint training obtains the high-resolution dictionary and the low-resolution dictionary.
In one embodiment of the invention, the step (22) includes the following steps:
(221) sparse K-SVD method training low-resolution dictionary is utilized；
(222) high-resolution dictionary is calculated.
In one embodiment of the invention, the step 3 includes the following steps:
(31) the first image is obtained to the low-resolution image denoising to be reconstructed；
(32) the first image deblurring is obtained into the second image；
(33) second image is split, preservation processing according to fixed length and width, forms the low-resolution image
Block.
In one embodiment of the invention, the step 4 includes the following steps:
(41) divide low resolution sample image progress block to obtain low resolution sample image block；
(42) marginal information that the low resolution sample image block is extracted using Boundary extracting algorithm, is counted each described low
The pixel value distribution situation of the pixel value of resolution ratio sample image block and all low resolution sample image blocks；
(43) according to the pixel value of each low resolution sample image block and all low resolution sample image blocks
Pixel value distribution situation choose X (X > 1) a candidate thresholds；
(44) each candidate thresholds are carried out using the rarefaction representation image super-resolution rebuilding algorithm based on study
Calculating forms multiple high-resolution candidate images；
(45) according to the operation time of the rarefaction representation image super-resolution rebuilding algorithm based on study and corresponding shape
At the high-resolution candidate image resolution ratio, the information content threshold value is selected from X candidate thresholds.
In one embodiment of the invention, the Boundary extracting algorithm is Canny operator edge detection algorithm.
In one embodiment of the invention, the step 6 includes the following steps:
(61) image filtering processing is carried out for the high information quantity image block, and carries out high-frequency characteristic and extracts acquisition high frequency
Information；
(62) KPCA dimensionality reduction is carried out to the high information quantity image block using the high-frequency information, realizes high dimensional data pressure
Contracting；
(63) the compressed data of dimensionality reduction are rebuild using OMP algorithm to obtain the high information quantity reconstruction subregion.
In one embodiment of the invention, described image filtering operation uses two-dimensional filtering operator filtering device group, described
Filter group is f={ f1,f2,f3,f4, wherein
f1=[1, -1], f2=f1 T
f3=LOG, f4=f3 T
T is the operation of matrix transposition, the two-dimensional filtering operator that LOG is 5 × 5.
In one embodiment of the invention, the step (63) includes the following steps:
(631) rarefaction representation coefficient β is calculated according to the OMP algorithm, wherein
Wherein, y is the data after dimensionality reduction decompression, T0For given degree of rarefication, βiFor the daughter element in matrix β；
(632) the rarefaction representation coefficient β is multiplied with the high-resolution dictionary, obtains the high information quantity and rebuilds son
Region.
In one embodiment of the invention, the interpolation algorithm is bicubic interpolation algorithm.
The embodiment of the present invention has the advantages that,
1, the embodiment of the present invention proposes that the method for establishing image library training dictionary obtains image set off-line training dictionary library
To the mapping relations of high-definition picture and low-resolution image, optimizes image reconstruction step, effectively reduces method runing time,
With adaptivity and high efficiency.
2, the super-resolution rebuilding image that the present invention obtains can carry out the change of scale of arbitrary size.In engineer application
It can satisfy the demand in the case where difference, there is preferable applicability.
3, the threshold value in the present invention in method is not required to during place based on fixation is optimized after a large amount of comparative tests
It is changed, in the case where not influencing image reconstruction effect, effectively raises image reconstruction efficiency.
4, it is zoomed in and out again according to the image after super-resolution rebuilding, since algorithm is reconstructed high-frequency information, for scaling
Effect can be more preferable.
Detailed description of the invention
Fig. 1 is a kind of method for reconstructing flow diagram of stepless zooming super-resolution image provided in an embodiment of the present invention；
Fig. 2 is a kind of dictionary training schematic illustration provided in an embodiment of the present invention；
Fig. 3 is a kind of method for reconstructing schematic illustration of stepless zooming super-resolution image provided in an embodiment of the present invention.
Specific embodiment
Further detailed description is done to the present invention combined with specific embodiments below, but embodiments of the present invention are not limited to
This.
Embodiment one
Referring to Figure 1, Fig. 2 and Fig. 3, Fig. 1 are a kind of stepless zooming super-resolution image provided in an embodiment of the present invention
Method for reconstructing flow diagram；Fig. 2 is a kind of dictionary training schematic illustration provided in an embodiment of the present invention；Fig. 3 is the present invention
A kind of method for reconstructing schematic illustration for stepless zooming super-resolution image that embodiment provides.The stepless zooming super-resolution figure
The method for reconstructing of picture, includes the following steps:
Step 1, Fuzzy Processing and N times of down-sampling are carried out according to degradation model to high-resolution sample image, obtains low point
Resolution sample image；
Step 2, using K-SVD method, dictionary training is carried out using the low resolution sample image, obtains high-resolution word
Allusion quotation and low-resolution dictionary；
Wherein, the method that step 1 and step 2 establish image library training dictionary obtains image set off-line training dictionary library
The mapping relations of high-definition picture and low-resolution image optimize image reconstruction step, effectively reduce method runing time, have
There are adaptivity and high efficiency.
Step 3, low-resolution image to be reconstructed is pre-processed to obtain several low-resolution image blocks；
Step 4, fixed threshold is calculated using the low resolution sample image；
Step 5, judge whether the pixel value of each low-resolution image block is less than the fixed threshold；If so,
Then determine that the low-resolution image block is Poor information image block；If it is not, then determining that the low-resolution image block is
High information quantity image block；
Wherein, step 5 significantly improves reconstruction speed while guaranteeing reconstruction quality, is suitble to time performance requirement
Higher application.
Step 6, the high information quantity image block is used and rebuilds to obtain high letter based on dictionary learning and rarefaction representation algorithm
Breath amount rebuilds subregion；
Wherein, the rarefaction representation method for reconstructing based on study avoids the artificial selection of neighbour's number, and it is preferable to rebuild effect.
Step 7, the Poor information image block is rebuild to obtain Poor information reconstruction subregion using interpolation algorithm；
Step 8, the high information quantity subregion subregion progress image mosaic is rebuild with the Poor information to be surpassed
Resolution reconstruction image；
Step 9, using interpolation algorithm by the super-resolution rebuilding image progress M times of interpolation obtain N*M (N=2,3,4；M
> 0) super-resolution rebuilding image again.
Specifically, the step 2 includes the following steps:
(21) characteristics of image that the low resolution sample image is extracted by feature extracting method, obtains extraterrestrial target
High-resolution features information and low resolution characteristic information.
(22) utilize the K-SVD method, to the high-resolution features information and the low resolution characteristic information into
Row joint training obtains the high-resolution dictionary and the low-resolution dictionary.
Wherein, the step (22) includes the following steps:
(221) sparse K-SVD method training low-resolution dictionary is utilized；
(222) high-resolution dictionary is calculated.
Specifically, the step 3 includes the following steps:
(31) the first image is obtained to the low-resolution image denoising to be reconstructed；
(32) the first image deblurring is obtained into the second image；
(33) second image is split, preservation processing according to fixed length and width, forms the low-resolution image
Block.
Specifically, the step 4 includes the following steps:
(41) divide low resolution sample image progress block to obtain low resolution sample image block；
(42) marginal information that the low resolution sample image block is extracted using Boundary extracting algorithm, is counted each described low
The pixel value distribution situation of the pixel value of resolution ratio sample image block and all low resolution sample image blocks；
(43) according to the pixel value of each low resolution sample image block and all low resolution sample image blocks
Pixel value distribution situation choose X (X > 1) a candidate thresholds；
(44) each candidate thresholds are carried out using the rarefaction representation image super-resolution rebuilding algorithm based on study
Calculating forms multiple high-resolution candidate images；
(45) according to the operation time of the rarefaction representation image super-resolution rebuilding algorithm based on study and corresponding shape
At the high-resolution candidate image resolution ratio, the information content threshold value is selected from X candidate thresholds.
Preferably, the Boundary extracting algorithm is Canny operator edge detection algorithm；
Furthermore the step 6 includes the following steps:
(61) image filtering processing is carried out for the high information quantity image block, and carries out high-frequency characteristic and extracts acquisition high frequency
Information；
(62) KPCA dimensionality reduction is carried out to the high information quantity image block using the high-frequency information, realizes high dimensional data pressure
Contracting；
(63) the compressed data of dimensionality reduction are rebuild using OMP algorithm to obtain the high information quantity reconstruction subregion.
Further, described image filtering operation uses two-dimensional filtering operator filtering device group, and the filter group is f=
{f1,f2,f3,f4, wherein
f1=[1, -1], f2=f1 T
f3=LOG, f4=f3 T
T is the operation of matrix transposition, the two-dimensional filtering operator that LOG is 5 × 5.
Further, the step (63) includes the following steps:
(631) rarefaction representation coefficient β is calculated according to the OMP algorithm, wherein
Wherein, y is the data after dimensionality reduction decompression, T0For given degree of rarefication, βiFor the daughter element in matrix β.
(632) the rarefaction representation coefficient β is multiplied with the high-resolution dictionary, obtains the high information quantity and rebuilds son
Region.
Preferably, the interpolation algorithm is bicubic interpolation algorithm.Such as linear interpolation, closest member can also be used
Method, bilinear interpolation method etc..
The present embodiment has the advantages that
1, the embodiment of the present invention proposes that the method for establishing image library training dictionary obtains image set off-line training dictionary library
To the mapping relations of high-definition picture and low-resolution image, optimizes image reconstruction step, effectively reduces method runing time,
With adaptivity and high efficiency.
2, the super-resolution rebuilding image that the present invention obtains can carry out the change of scale of arbitrary size.In engineer application
It can satisfy the demand in the case where difference, there is preferable applicability.
3, the threshold value in the present invention in method is not required to during place based on fixation is optimized after a large amount of comparative tests
It is changed, in the case where not influencing image reconstruction effect, effectively raises image reconstruction efficiency.
Embodiment two
On the basis of the above embodiment 1, the present embodiment provides the reconstruction sides of another stepless zooming super-resolution image
Method includes dictionary training process, three image reconstruction process, dimensional variation process processing steps.Specifically comprise the following steps:
S1: dictionary reconstruction process.
S11: utilizing a large amount of high-resolution sample images, and high-definition picture is carried out mould according to revised degradation model
Paste and N times of down-sampling, obtain corresponding low resolution sample image.
S12: characteristics of image is extracted by feature extracting method to the low resolution sample image that step 1 obtains, obtains sky
Between target high-low resolution characteristic information, that is, XsAnd Ys。
S13: utilizing K-SVD method, carries out joint training to high-low resolution characteristic information, obtains high-low resolution word
Allusion quotation.
S13a: training low-resolution dictionary.Base dictionary Φ selected complete DCT dictionary, was asked using sparse K-SVD method
Solution:
Then low-resolution dictionary Dl=Φ W, W are an atom representing matrixes.Compared with parsing dictionary model, double sparse words
Allusion quotation model provides adaptivity by the modification to W.
S13b: high-resolution dictionary is calculated.Assuming that high-resolution-low-resolution image block is in the low resolution of high-resolution-
Rate dictionary rarefaction representation coefficient A having the same under can then be calculated by minimizing the approximate error in following formula
High-resolution dictionary Dh:
It is solved using pseudoinverse:
Dh=XsA+=XsAT(AAT)-1 (3)
Wherein, subscript "+" indicates pseudoinverse.
S2 image reconstruction process.
S21: pre-processing low-resolution image to be reconstructed, wherein mainly including image denoising, image deblurring
It is operated with sample piecemeal.Its processing step are as follows:
S21a: to low-resolution image denoising to be reconstructed；
S21b: the image deblurring that step S21a is obtained；
S21c: carrying out block division to the image that step S21b is obtained, entire image is split according to fixed length and width,
It saves；
S22: fixed threshold is calculated using the low resolution sample image obtained in step S11, specific processing step is as follows:
S22a: a large amount of low resolution sample images obtained in step S11 are subjected to block segmentation and obtain image block, image
The length and width of block are the same as step S21c；
S22b: using Boundary extracting algorithm extract image block marginal information, count each image block information content and all figures
As the information content distribution situation of block；
S22c: information content is highest in selection image block, and the pixel value for obtaining the image block is F1, takes f=F1/4, then f*
40%≤threshold value≤f*60%, take it is several within the scope of this represent threshold value, for example, volume, which can take, represents threshold value as follows: f*40%,
F*45%, f*50%, f*55%, f*60% calculate the corresponding rarefaction representation Image Super-resolution based on study of each threshold point
The resolution ratio of image after the reconstruction time of rate algorithm for reconstructing and reconstruction can judge according to subjective assessment and PSNR.Then root
Fixed threshold is determined in threshold value from several represent according to user's demand, if user's demand more focuses on the time, ratio is just turned up in that,
Time is with regard to fast；If opposite user focuses on rebuilding effect, that just turns down ratio, and bring is exactly to rebuild effect preferably but the time
It is longer.For example, being computed rear f*50% is best suitable for user's demand, then fixed threshold=f*50% is taken.
S22d: the input step S21 low-resolution image block to be reconstructed obtained extracts image using Boundary extracting algorithm
Block edge information, when the image block information content is no more than step S22c determining fixed threshold, the image block is Poor information figure
As block；It otherwise, is high information quantity image block.Preferably, the Boundary extracting algorithm is Canny operator edge detection algorithm.
S23: image filtering operations are carried out for high information quantity image block, high-frequency characteristic extraction are carried out, using two-dimensional filtering
Operator filtering device group, filter group used are f={ f1,f2,f3,f4, it is made of, is respectively as follows: four different filters
f1=[1, -1], f2=f1 T (4)
f3=LOG, f4=f3 T (5)
Wherein superscript T representing matrix transposition operates, and LOG represents the two-dimensional filtering operator of one kind 5 × 5.By high frequency spy
Image block high-frequency information is obtained after levying extraction operation, with xlIt indicates.
S24: KPCA dimensionality reduction is carried out to the image block that step S23 is obtained, realizes high dimensional data compression.Steps are as follows for dimensionality reduction:
S24a: high dimensional data set is expressed as X={ x1,x2,x3,...,xM},xi∈RD, KPCA method is by non-linear
Mapping function x → Φ (x) ∈ F, wherein F is feature space, just each data x can be mapped to a high dimensional feature sky in this way
Between.
S24b: kernel function will carry out the respective operations of point x a to F by Φ, and in thus obtained F data satisfaction
The condition of the heart, it may be assumed that
The then covariance matrix in feature space F are as follows:
S24c: eigenvalue λ >=0 and the feature vector of c are asked
V ∈ F { 0 }, Cv=λ v (8)
Then have
(Φ(xv) Cv)=λ (Φ (xv)·v) (9)
In view of all feature vectors are represented by Φ (x1),Φ(x2),...,Φ(xM) linear combination, it may be assumed that
Then have:
In formula, v=1,2,3 ..., M define M × M and tie up matrix Kμv
Kμv:=(Φ (xμ)·Φ(xv)) (12)
S24d: it solves above formula and obtains characteristic value and feature vector, for data acquisition system in characteristic vector space VkProjection
It can be write as:
So, data are projected to the feature vector V of covariance matrixkOn, projection result (the namely table of low-dimensional data
Showing can y) indicate are as follows:
S25: the rarefaction representation image super-resolution weight based on study is carried out to by the compressed data of step S24 dimensionality reduction
It builds, the specific steps are as follows:
S25a: the low-dimensional data y obtained using OMP algorithm to step S24d is in low-resolution dictionary DlUnder rarefaction representation
Factor beta solves following equation:
Wherein T0For given degree of rarefication, βiFor the daughter element in matrix β.
S25b: by the rarefaction representation coefficient β acquired and high-resolution dictionary DhIt is multiplied, the high information quantity rebuild is rebuild
Subregion, it may be assumed that
X=Dhβ
Wherein X is the super-resolution subregion acquired.
S26: the Poor information image block that step 22d is obtained is rebuild to obtain low information using bicubic interpolation algorithm
Amount rebuilds subregion；
S27: high information quantity subregion and the Poor information are rebuild into subregion progress image mosaic and obtain super-resolution
Reconstruction image；
S28: using bicubic interpolation algorithm by the super-resolution rebuilding image progress M times of interpolation obtain N*M (N=2,
3,4；M > 0) times super-resolution rebuilding image.
The above content is a further detailed description of the present invention in conjunction with specific preferred embodiments, and it cannot be said that
Specific implementation of the invention is only limited to these instructions.For those of ordinary skill in the art to which the present invention belongs, exist
Under the premise of not departing from present inventive concept, can also make it is several it is simple deduction or or replacement, all shall be regarded as belonging to the present invention
Protection scope.
Claims (9)
1. a kind of method for reconstructing of stepless zooming super-resolution image, which comprises the steps of:
Step 1, Fuzzy Processing and N times of down-sampling are carried out according to degradation model to high-resolution sample image, obtains low resolution
Sample image；
Step 2, using K-SVD method, carry out dictionary training using the low resolution sample image, obtain high-resolution dictionary with
Low-resolution dictionary；
Step 3, low-resolution image to be reconstructed is pre-processed to obtain several low-resolution image blocks；
Step 4, fixed threshold is calculated using the low resolution sample image；
The step 4 includes the following steps:
(41) divide low resolution sample image progress block to obtain low resolution sample image block；
(42) marginal information that the low resolution sample image block is extracted using Boundary extracting algorithm, counts each low resolution
The pixel value distribution situation of the pixel value of rate sample image block and all low resolution sample image blocks；
(43) according to the picture of the pixel value of each low resolution sample image block and all low resolution sample image blocks
Plain Distribution value situation chooses X candidate thresholds, X > 1；
(44) each candidate thresholds are calculated using the rarefaction representation image super-resolution rebuilding algorithm based on study
Form multiple high-resolution candidate images；
(45) it according to operation time of the rarefaction representation image super-resolution rebuilding algorithm based on study and is correspondingly formed
The resolution ratio of the high-resolution candidate image selects the information content threshold value from X candidate thresholds；
Step 5, judge whether the pixel value of each low-resolution image block is less than the fixed threshold；If so, sentencing
The fixed low-resolution image block is Poor information image block；If it is not, then determining the low-resolution image block for height letter
Breath amount image block；
Step 6, the high information quantity image block is used and rebuilds to obtain high information quantity based on dictionary learning and rarefaction representation algorithm
Rebuild subregion；
Step 7, the Poor information image block is rebuild to obtain Poor information reconstruction subregion using interpolation algorithm；
Step 8, the high information quantity subregion and the Poor information are rebuild into subregion progress image mosaic and obtains super-resolution
Rate reconstruction image；
Step 9, M times of interpolation of super-resolution rebuilding image progress is obtained by N*M times of Super-resolution reconstruction using interpolation algorithm
Build image, N=2,3,4；M>0.
2. the method for reconstructing of stepless zooming super-resolution image according to claim 1, which is characterized in that the step 2
Include the following steps:
(21) characteristics of image that the low resolution sample image is extracted by feature extracting method, obtains the high score of extraterrestrial target
Resolution characteristic information and low resolution characteristic information；
(22) the K-SVD method is utilized, the high-resolution features information and the low resolution characteristic information are joined
Training is closed, the high-resolution dictionary and the low-resolution dictionary are obtained.
3. the method for reconstructing of stepless zooming super-resolution image according to claim 2, which is characterized in that the step
(22) include the following steps:
(221) sparse K-SVD method training low-resolution dictionary is utilized；
(222) high-resolution dictionary is calculated.
4. the method for reconstructing of stepless zooming super-resolution image according to claim 1, which is characterized in that the step 3
Include the following steps:
(31) the first image is obtained to the low-resolution image denoising to be reconstructed；
(32) the first image deblurring is obtained into the second image；
(33) second image is split, preservation processing according to fixed length and width, forms the low-resolution image block.
5. the method for reconstructing of stepless zooming super-resolution image according to claim 4, which is characterized in that the edge mentions
Taking algorithm is Canny operator edge detection algorithm.
6. the method for reconstructing of stepless zooming super-resolution image according to claim 1, which is characterized in that the step 6
Include the following steps:
(61) image filtering processing is carried out for the high information quantity image block, and carries out high-frequency characteristic and extracts acquisition high frequency letter
Breath；
(62) KPCA dimensionality reduction is carried out to the high information quantity image block using the high-frequency information, realizes high dimensional data compression；
(63) the compressed data of dimensionality reduction are rebuild using OMP algorithm to obtain the high information quantity reconstruction subregion.
7. the method for reconstructing of stepless zooming super-resolution image according to claim 6, which is characterized in that described image filter
Wave operation uses two-dimensional filtering operator filtering device group, and the filter group is f={ f1,f2,f3,f4, wherein
f1=[1, -1], f2=f1 T
f3=LOG, f4=f3 T
T is the operation of matrix transposition, the two-dimensional filtering operator that LOG is 5 × 5.
8. the method for reconstructing of stepless zooming super-resolution image according to claim 6, which is characterized in that the step
(63) include the following steps:
(631) rarefaction representation coefficient β is calculated according to the OMP algorithm, wherein
Wherein, y is the data after dimensionality reduction decompression, T0For given degree of rarefication, βiFor the daughter element in matrix β, DlFor low resolution
Dictionary；
(632) the rarefaction representation coefficient β is multiplied with the high-resolution dictionary, obtains the high information quantity and rebuilds sub-district
Domain.
9. the method for reconstructing of stepless zooming super-resolution image according to claim 1, which is characterized in that the interpolation is calculated
Method is bicubic interpolation algorithm.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201710267753.2A CN107169925B (en) | 2017-04-21 | 2017-04-21 | The method for reconstructing of stepless zooming super-resolution image |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201710267753.2A CN107169925B (en) | 2017-04-21 | 2017-04-21 | The method for reconstructing of stepless zooming super-resolution image |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN107169925A CN107169925A (en) | 2017-09-15 |
| CN107169925B true CN107169925B (en) | 2019-10-22 |
Family
ID=59812837
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201710267753.2A Active CN107169925B (en) | 2017-04-21 | 2017-04-21 | The method for reconstructing of stepless zooming super-resolution image |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN107169925B (en) |
Families Citing this family (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN109064394B (en) * | 2018-06-11 | 2023-07-18 | 西安电子科技大学 | Image super-resolution reconstruction method based on convolutional neural network |
| CN111986078B (en) * | 2019-05-21 | 2023-02-10 | 四川大学 | Multi-scale core CT image fusion reconstruction method based on guide data |
Citations (5)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN104463801A (en) * | 2014-12-04 | 2015-03-25 | 河西学院 | Multi-sensing-information fusion method based on self-adaptation dictionary learning |
| WO2015180055A1 (en) * | 2014-05-28 | 2015-12-03 | 北京大学深圳研究生院 | Super-resolution image reconstruction method and apparatus based on classified dictionary database |
| CN105427243A (en) * | 2015-10-29 | 2016-03-23 | 中国科学院自动化研究所 | Video super-resolution reconstruction method based on adaptive interpolation kernel learning |
| US9466132B2 (en) * | 2014-12-08 | 2016-10-11 | General Electric Company | Systems and methods for motion mitigation determinations |
| CN106408550A (en) * | 2016-09-22 | 2017-02-15 | 天津工业大学 | Improved self-adaptive multi-dictionary learning image super-resolution reconstruction method |
-
2017
- 2017-04-21 CN CN201710267753.2A patent/CN107169925B/en active Active
Patent Citations (5)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| WO2015180055A1 (en) * | 2014-05-28 | 2015-12-03 | 北京大学深圳研究生院 | Super-resolution image reconstruction method and apparatus based on classified dictionary database |
| CN104463801A (en) * | 2014-12-04 | 2015-03-25 | 河西学院 | Multi-sensing-information fusion method based on self-adaptation dictionary learning |
| US9466132B2 (en) * | 2014-12-08 | 2016-10-11 | General Electric Company | Systems and methods for motion mitigation determinations |
| CN105427243A (en) * | 2015-10-29 | 2016-03-23 | 中国科学院自动化研究所 | Video super-resolution reconstruction method based on adaptive interpolation kernel learning |
| CN106408550A (en) * | 2016-09-22 | 2017-02-15 | 天津工业大学 | Improved self-adaptive multi-dictionary learning image super-resolution reconstruction method |
Non-Patent Citations (1)
| Title |
|---|
| 稀疏阈值的超分辨率图像重建;何阳 等;《中国光学》;20161015;第532-539页 * |
Also Published As
| Publication number | Publication date |
|---|---|
| CN107169925A (en) | 2017-09-15 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| Jin et al. | A flexible deep CNN framework for image restoration | |
| CN106952228B (en) | Super-resolution reconstruction method of single image based on image non-local self-similarity | |
| Gao et al. | Image super-resolution with sparse neighbor embedding | |
| Zhu et al. | Single image super-resolution using deformable patches | |
| Lau et al. | Restoration of atmospheric turbulence-distorted images via RPCA and quasiconformal maps | |
| Dong et al. | Nonlocally centralized sparse representation for image restoration | |
| Zhang et al. | Single image super-resolution with multiscale similarity learning | |
| Ren et al. | Single image super-resolution using local geometric duality and non-local similarity | |
| CN104574336B (en) | Super-resolution image reconstruction system based on adaptive sub- mould dictionary selection | |
| CN105631807B (en) | The single-frame image super-resolution reconstruction method chosen based on sparse domain | |
| CN106709875A (en) | Compressed low-resolution image restoration method based on combined deep network | |
| CN104657962B (en) | The Image Super-resolution Reconstruction method returned based on cascading linear | |
| CN103077511A (en) | Image super-resolution reconstruction method based on dictionary learning and structure similarity | |
| CN105550989B (en) | The image super-resolution method returned based on non local Gaussian process | |
| CN106920214A (en) | Spatial target images super resolution ratio reconstruction method | |
| CN113781308A (en) | Image super-resolution reconstruction method and device, storage medium and electronic equipment | |
| CN108830791A (en) | Image super-resolution method based on itself sample and rarefaction representation | |
| Mandal et al. | Edge preserving single image super resolution in sparse environment | |
| CN108460723B (en) | Bilateral total variation image super-resolution reconstruction method based on neighborhood similarity | |
| CN107169925B (en) | The method for reconstructing of stepless zooming super-resolution image | |
| Zhang et al. | Learning stacking regressors for single image super-resolution | |
| CN109559278B (en) | Super resolution image reconstruction method and system based on multiple features study | |
| CN108492252A (en) | Face image super-resolution reconstruction method based on secondary reconstruction | |
| CN107203968A (en) | Single image super resolution ratio reconstruction method based on improved subspace tracing algorithm | |
| Liu et al. | Image super-resolution via hierarchical and collaborative sparse representation |
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 |