CN105931181B - Super resolution image reconstruction method and system based on non-coupled mapping relations - Google Patents
Super resolution image reconstruction method and system based on non-coupled mapping relations Download PDFInfo
- Publication number
- CN105931181B CN105931181B CN201610231568.3A CN201610231568A CN105931181B CN 105931181 B CN105931181 B CN 105931181B CN 201610231568 A CN201610231568 A CN 201610231568A CN 105931181 B CN105931181 B CN 105931181B
- Authority
- CN
- China
- Prior art keywords
- image
- layer
- resolution
- dictionary
- pyramid
- 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 38
- 238000013507 mapping Methods 0.000 title claims abstract description 24
- 239000011159 matrix material Substances 0.000 claims abstract description 16
- 238000012549 training Methods 0.000 claims abstract description 13
- 238000004458 analytical method Methods 0.000 claims abstract description 11
- 238000005070 sampling Methods 0.000 claims description 25
- 238000001914 filtration Methods 0.000 claims description 24
- 230000000593 degrading effect Effects 0.000 claims description 18
- 238000000605 extraction Methods 0.000 claims description 9
- 238000000205 computational method Methods 0.000 claims description 5
- 238000006243 chemical reaction Methods 0.000 claims description 4
- 239000000126 substance Substances 0.000 claims description 4
- 238000005516 engineering process Methods 0.000 description 9
- 230000008878 coupling Effects 0.000 description 5
- 238000010168 coupling process Methods 0.000 description 5
- 238000005859 coupling reaction Methods 0.000 description 5
- 230000008569 process Effects 0.000 description 5
- 230000008901 benefit Effects 0.000 description 3
- 230000006870 function Effects 0.000 description 3
- 230000003044 adaptive effect Effects 0.000 description 2
- 230000000149 penetrating effect Effects 0.000 description 2
- 238000012545 processing Methods 0.000 description 2
- 238000007493 shaping process Methods 0.000 description 2
- 230000003321 amplification Effects 0.000 description 1
- 230000000295 complement effect Effects 0.000 description 1
- 238000000354 decomposition reaction Methods 0.000 description 1
- 238000011161 development Methods 0.000 description 1
- 238000006073 displacement reaction Methods 0.000 description 1
- 238000002474 experimental method Methods 0.000 description 1
- 238000003384 imaging method Methods 0.000 description 1
- 238000012986 modification Methods 0.000 description 1
- 230000004048 modification Effects 0.000 description 1
- 238000003199 nucleic acid amplification method Methods 0.000 description 1
- 238000012805 post-processing Methods 0.000 description 1
- 238000003672 processing method Methods 0.000 description 1
- 230000002194 synthesizing effect Effects 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
- G06T3/4053—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
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T2207/00—Indexing scheme for image analysis or image enhancement
- G06T2207/20—Special algorithmic details
- G06T2207/20016—Hierarchical, coarse-to-fine, multiscale or multiresolution image processing; Pyramid transform
Landscapes
- Physics & Mathematics (AREA)
- General Physics & Mathematics (AREA)
- Engineering & Computer Science (AREA)
- Theoretical Computer Science (AREA)
- Image Processing (AREA)
- Image Analysis (AREA)
Abstract
A kind of super resolution image reconstruction method and system based on non-coupled mapping relations, including training image and original low-resolution training image to be filtered using filter original high resolution respectively, obtain corresponding pyramid image, complete dictionary was all built respectively to each layer image, and be layered and carry out structural similarity analysis, a certain layer is unsatisfactory for the filter parameter that predetermined threshold value then adjusts this layer；Pending low resolution image is equally filtered, obtain corresponding pyramid image, signal matrix is configured to every layer image, sparse expression is carried out to the signal matrix using low-resolution dictionary, the high resolution image of respective layer is obtained by the high-resolution dictionary of sparse coefficient and respective layer, each layer of high resolution image is finally obtained into final reconstructed image by the operation of liftering.The present invention effectively excavates effective information of the raw video in different resolution, greatly strengthens the ability to express of constructed dictionary, improves reconstruction precision.
Description
Technical field
The present invention relates to super resolution image reconstruction technical fields, more particularly to one kind is by establishing different resolution image
Between non-coupled mapping relations complete image across the technical solution of space combination, completed between dictionary atom in the method for filtering
Non-coupled mapping relations, realize the super-resolution rebuilding of image.
Background technology
In practical applications, high resolution image is always what people were pursued.But due to by imaging sensor and light
The limitation of the process costs of device is learned, it is relatively difficult to obtain high resolution image in many cases.Therefore, develop super-resolution
Reconstruction technique, the spatial resolution that image is improved by way of post-processing have prodigious realistic meaning.
Currently, the development trend of super-resolution rebuilding technology mainly has three classes：
Super-resolution rebuilding technology based on interpolation：This method mainly passes through opposite between pre-estimating out sub- picture dot
Then the reconstruction of high resolution image is completed in displacement by nonuniform sampling.The advantages of this method is that computation complexity is low,
The disadvantage is that being easy fuzzy edge and texture information.
Super-resolution rebuilding technology based on reconstruction：This method is mainly the model that degrades according to image, and image is added
Prior information (for example, smoothness constraint condition etc.) solves high resolution image.The advantages of this method saves image
Edge and texture information, the disadvantage is that being generally up to 2 times to the amplification factor of image.
Super-resolution rebuilding technology based on study：This method is mainly first built by high-low resolution Image Database respectively
Vertical high-low resolution dictionary, while the correspondence between high-low resolution dictionary atom is obtained, finally by this correspondence
It is applied among pending image, completes the super-resolution rebuilding of image.
In the super-resolution rebuilding technology based on study, this correspondence is to map correspondingly under normal conditions
Relationship.However, in practical applications, this one-to-one mapping relations often limit its precision.For example, for complexity
In remote sensing image, if differences in resolution between image is huge (to differ 16 between such as Landsat images and MODIS images
Times), it is simple by the one-to-one relationship between dictionary atom it is difficult to ensure that its precision for rebuilding.
As it can be seen that not yet there is ideal super-resolution rebuilding technology.
Invention content
For the disadvantage in the existing super resolution image reconstruction based on study, the present invention proposes that one kind is reflected based on non-coupled
The super resolution image reconstruction method for penetrating relationship builds pyramid image by the method for filtering, passes through the coupling on every layer image
Conjunction relationship carrys out the final non-coupled relationship for synthesizing overall image, completes the reconstruction of final high resolution image.
Technical scheme of the present invention provides a kind of super resolution image reconstruction method based on non-coupled mapping relations, including
Following steps：
Step 1, image and original low-resolution training image is trained to be carried out using filter original high resolution respectively
Filtering, obtains corresponding pyramid image, is denoted as pyramid image A and pyramid image B respectively；
Step 2, be based on extraction image sampling block, each layer image of pyramid image A is all built respectively one it is excessively complete
Standby dictionary, obtains high-resolution dictionary, all builds an excessively complete dictionary respectively to each layer image of pyramid image B, obtains
It to low-resolution dictionary, and is layered and carries out structural similarity analysis, if be unsatisfactory for the structural similarity analysis of a certain layer pre-
If threshold value then adjusts the filter parameter of this layer, return to step 1 filters again adjusts the layer image, if meeting predetermined threshold value,
Then record current filter parameter, determine every layer all meet predetermined threshold value after enter step 3；
Step 3, according to the filter parameter recorded in step 2, pending low resolution image is similarly filtered
Wave operates, and obtains corresponding pyramid image, is denoted as pyramid image C, and image is extracted to each layer image of pyramid image C
Sampling block；
Step 4, signal matrix is configured to every layer image of pyramid image C respectively, utilizes the low resolution of respective layer
Dictionary carries out sparse expression to the signal matrix, obtains sparse coefficient；
Step 5, the high resolution image of respective layer is obtained by the high-resolution dictionary of sparse coefficient and respective layer, finally
Each layer of high resolution image is obtained into final reconstructed image by the operation of liftering.
Moreover, in step 2, the computational methods that structural similarity SSIM is carried out to any layer are as follows,
Wherein, the high resolution image after P and O respectively represents low resolution image and degrades, μPRepresent low resolution shadow
As mean value, μORepresent the high resolution image mean value after degrading, σPRepresent the variance of low resolution image, σOAfter representative degrades
The variance of high resolution image, σPOThe covariance between high resolution image after representing low resolution image and degrading, parameter
C1=K1× L, C2=K2× L, L are the max pixel value in grayscale image, K1And K2It is preset constant；Low resolution image is
A certain layer in pyramid image B, the high resolution image after degrading are the results after equivalent layer degrades in pyramid image A.
Moreover, in step 4, the realization method for carrying out sparse expression is, for each layer of image, by image sampling block by
One row chemical conversion column vector, is utilized respectively each column vector the low-resolution dictionary of respective layer, is carried out using Lasso algorithms sparse
Expression, solves corresponding sparse coefficient.
The present invention provides a kind of super resolution image reconstruction system based on non-coupled mapping relations, comprises the following modules：
First module, for training image and original low-resolution training image using filtering original high resolution respectively
Device is filtered, and obtains corresponding pyramid image, is denoted as pyramid image A and pyramid image B respectively；
Second module, for based on extraction image sampling block, one all to be built respectively to each layer image of pyramid image A
A excessively complete dictionary, obtains high-resolution dictionary, and an excessively complete word is all built respectively to each layer image of pyramid image B
Allusion quotation obtains low-resolution dictionary, and is layered and carries out structural similarity analysis, if analyzed the structural similarity of a certain layer discontented
Sufficient predetermined threshold value then adjusts the filter parameter of this layer, and filtering adjusts the layer image to the work of the first module of order again, if full
Sufficient predetermined threshold value then records current filter parameter, determines that every layer all meets the work of predetermined threshold value post command third module；
Third module, for according to the filter parameter recorded in the second module, to pending low resolution image into
The same filtering operation of row, obtains corresponding pyramid image, is denoted as pyramid image C, to each layer of shadow of pyramid image C
As extraction image sampling block；
4th module utilizes the low of respective layer for being configured to signal matrix to every layer image of pyramid image C respectively
Resolution ratio dictionary carries out sparse expression to the signal matrix, obtains sparse coefficient；
5th module, for obtaining the high-resolution shadow of respective layer by the high-resolution dictionary of sparse coefficient and respective layer
Each layer of high resolution image is finally obtained final reconstructed image by picture by the operation of liftering.
Moreover, in the second module, the computational methods that structural similarity SSIM is carried out to any layer are as follows,
Wherein, the high resolution image after P and O respectively represents low resolution image and degrades, μPRepresent low resolution shadow
As mean value, μORepresent the high resolution image mean value after degrading, σPRepresent the variance of low resolution image, σOAfter representative degrades
The variance of high resolution image, σPOThe covariance between high resolution image after representing low resolution image and degrading, parameter
C1=K1× L, C2=K2× L, L are the max pixel value in grayscale image, K1And K2It is preset constant；Low resolution image is
A certain layer in pyramid image B, the high resolution image after degrading are the results after equivalent layer degrades in pyramid image A.
Moreover, in the 4th module, the realization method for carrying out sparse expression is, for each layer of image, image to be sampled
Block arranges chemical conversion column vector one by one, and the low-resolution dictionary of respective layer is utilized respectively for each column vector, is carried out using Lasso algorithms
Sparse expression solves corresponding sparse coefficient.
Super resolution image reconstruction method proposed by the present invention based on non-coupled mapping relations, utilizes pyramid image structure
Building has the characteristics that the dictionary of layering, carries out layering super-resolution rebuilding to pending image, finally synthesizes high resolution image.It should
The filtering of method has extracted the information of raw video different resolution, while can also alleviate height to a certain extent
Differences in resolution between resolution image.Before carrying out sparse reconstruction to pending image, same pyramid shadow is carried out to it
As structure, progress image is matched with the similitude of dictionary atom in the different resolution level of image, is greatly strengthened and is waited locating
Manage the correlation between image and dictionary atom.Finally by the process of liftering, every layer of high resolution image is integrated into
Complete result.
Super resolution image reconstruction method proposed by the present invention based on non-coupled mapping relations, can effectively excavate
Effective information of the raw video in different resolution, greatly strengthens the ability to express of constructed dictionary, by treating place
It manages image and carries out pyramid filtering, enhance its correlation between dictionary atom, due to the height on each layer image point
Relationship between atom simultaneously differs, thus it is final integrate height out divide the relationship of dictionary atom present one it is non-thread
Property coupled relation, this relationship are actually more suitable for the mapping in real world, improve its final reconstruction precision.In video
The fields such as image, natural image, medical image, remote sensing image have great application value.Therefore, non-coupled mapping is based on to close
The super resolution image reconstruction method of system not only has very important learning value but also has important practical significance.
Description of the drawings
Fig. 1 is the embodiment of the present invention flow chart.
Specific implementation mode
In order to make the purpose , technical scheme and advantage of the present invention be clearer, below in conjunction with attached drawing, to according to this hair
The super resolution image reconstruction method of the non-coupled mapping relations of bright one embodiment is further described.It should be appreciated that this
The described specific embodiment in place is used only for explaining the present invention, is not intended to limit the present invention.
Present invention is generally directed to the mapping relations between dictionary atom, are dropped by the image simulated in high-low resolution space
Matter and non-linear relation obtain the mapping relations between final dictionary atom with the method for filtering, are finally reflected this
The relationship of penetrating is applied to reconstruction process.The present invention enhances its learning ability, makes by fully excavating existing potential information between image
It obtains raw information to be more reasonably utilized, improves the practical applicability of algorithm.
Technical solution of the present invention can be used computer software technology and realize automatic running flow.It is described in detail below in conjunction with Fig. 1 real
Apply a specific steps for super resolution image reconstruction method.
Step 1, image is trained to original high resolution by the method for filtering (such as adaptive wavelet filtering etc.) respectively
It is filtered with original low-resolution training image, obtains corresponding pyramid image, be denoted as pyramid image A and pyramid respectively
Image B.
The super resolution image reconstruction method based on non-coupled mapping relations proposed in the present invention, is by advance filter
Wave process builds image pyramid, and final Non-linear coupling is synthesized by the interatomic coupled relation of dictionary on each layer
Relationship, the specific filtering method of embodiment select existing adaptive wavelet filtering (ATWT), art technology when specific implementation
Personnel can also be voluntarily arranged using other filtering methods.Small echo in embodiment uses 5 × 5 gaussian kernel function, decomposition layer
Number is 3 layers.Filter in the present invention is simulated by Gaussian function, subsequently will be by adjusting the variance yields in Gaussian function
Filter is optimized with filtering size, size is filtered under normal circumstances and is less than image block size, it is right that variance yields passes through
Image to be filtered carries out an estimation.
The empirical value of filtering size is usually the half of image block size, and those skilled in the art can be when specific implementation
This empirical value is nearby chosen.The selection of variance is according to a statistics for carrying out variance to pending image, empirical value choosing
Take the peak value of the occurrence number in statistical result, those skilled in the art can carry out near this empirical value when specific implementation
It chooses.
Step 2, be based on extraction image sampling block, each layer image of pyramid image A is all built respectively one it is excessively complete
Standby dictionary, obtains high-resolution dictionary, all builds an excessively complete dictionary respectively to each layer image of pyramid image B, obtains
It to low-resolution dictionary, and is layered and carries out structural similarity analysis, if be unsatisfactory for the structural similarity analysis of a certain layer pre-
If threshold value then adjusts the filter parameter of this layer, return to step 1 filters again adjusts the layer image, if meeting predetermined threshold value,
Then record current filter parameter, determine every layer all meet predetermined threshold value after enter step 3.
Construct an excessively complete dictionary respectively using training sample all in each layer, each layer of excessively complete dictionary it
Between all meet coupled relation, but coupled relation between layers is different, what can be synthesized the non-linear coupling of an entirety from
Conjunction relationship.
Different resolution image is compared and analyzed later for convenience, it can be (such as double by simply up-sampling
Linear interpolation) it operates the size of low resolution image alignment to high resolution image.Image sampling block size is 7 in embodiment
× 7, sampling block overlapping region is 7 × 2, is instructed in 61504 training samples on the raw video of 500 × 500 sizes
Get the dictionary of only 512 49 dimension samples.After being filtered to it, the second layer, top layer image size be respectively 250 ×
250 and 125 × 125, the corresponding block size that samples becomes 5 × 5 and 3 × 3, and dictionary dimension is respectively 25 and 9.Think this three layers
Dictionary includes the information in the different resolution level of image.
The present invention is for each layer in pyramid image, by judging the high resolution image after degrading and original low
Correlation between resolution image decides whether to adjust the parameter of filter, so that each layer of high-low resolution word
Atom in allusion quotation ensures maximum correlation.
High-resolution dictionary after judging low-resolution dictionary atom by structural similarity (SSIM) index and degrade
Similitude between atom carries out shown in the computational methods such as formula (1) of SSIM any layer：
Wherein, the high-resolution after P and O respectively represents low resolution image (a certain layer in pyramid image B) and degrades
Image (result after equivalent layer degrades in pyramid image A), μPRepresent low resolution image mean value, μORepresent the height after degrading
Resolution image mean value, σPRepresent the variance of low resolution image, σORepresent the variance of the high resolution image after degrading, σPOGeneration
Covariance between table low resolution image and high resolution image after degrading, parameter C1=K1× L, C2=K2× L, here L
For the max pixel value in grayscale image, i.e. L=255, K1And K2It is a smaller constant, art technology when specific implementation
Personnel can voluntarily preset, K in embodiment1=0.01, K2=0.02.As can be seen that SSIM values are bigger from formula (1), indicate
Two images are more similar, it is more reasonable to illustrate that it decomposes raw video, error is smaller.A threshold value is arranged in embodiment, when certain
When the SSIM of layer is less than this threshold value, the parameter of filter is adjusted, step 1 is gone to, re-establishes the layer in pyramid, until
SSIM values reach threshold value, the current final filter parameter of record, determine every layer all meet advance threshold value after enter step 3.Tool
When body is implemented those skilled in the art can voluntarily predetermined threshold value value.Preferably, SSIM threshold values are between 0.75~0.85
It chooses, 0.80 is taken in embodiment.
Step 3, according to filter parameter final after being adjusted in step 2, the image for treating oversubscription is similarly filtered
Operation, the pyramid image inputted are denoted as pyramid image C.
Wait for the i.e. pending low resolution image of the image of oversubscription.In embodiment, to each layer image of pyramid image C
Extract image sampling block, sampling block size to ensure it is consistent with the dimension of corresponding dictionary, as embodiment sampling block be 7 ×
7, dictionary is 49 dimensions；And the parameter of filter is consistent with the parameter after being adjusted in step 2.
Step 4, signal matrix is configured to every layer image of pyramid image C respectively, utilizes the low resolution of respective layer
Dictionary carries out sparse expression to the signal matrix.
The sparse expression that signal matrix is carried out in embodiment is (general according to the sequence of sampling block for each layer of image
Since the upper left corner of image, according to first from left to right and then again sequence from top to bottom) row chemical conversion column vector one by one, for theA column vector is carried out sparse using the excessively complete dictionary (using low-resolution dictionary at this time) of parameter after the adjustment in step 2
Expression, as shown in formula (2).To each sampling block in each layer image, Lasso algorithms are respectively adopted and solve corresponding sparse coefficient
α。
Qi Zhong ||·||1Indicate 1 norm,Indicate that square of two norms, λ are regularization parameter, xlFor in signal matrix
Column vector, DlFor the low-resolution dictionary of respective layer.In embodiment, pyramid haves three layers image,
Step 5, the high resolution image of respective layer is obtained by the high-resolution dictionary of sparse coefficient and respective layer, finally
Each layer of high resolution image is obtained into final reconstructed image by the operation of liftering.
When specific operation, it is high-resolution according to corresponding the sparse coefficient α and respective layer of each sampling block in each layer image
Rate dictionary obtains the corresponding image blocks of high resolution image of respective layer.In order to make the pixel value between adjacent image blocks seamlessly transit
(avoiding generating artefact), at the initial stage of sampling, keeps certain overlapping between image blocks.Finally in processing high score image blocks
When, the operation that is averaged for overlapping region.
By carrying out same filtering layered shaping to pending low resolution image, according to it is same adopt block mode by
One carries out sparse solution；By sparse coefficient and corresponding high-resolution dictionary, the reconstruction of final high resolution image is completed.This
Invention is using the complementary information in the different resolution level of same image, in the nonlinear coupling relationship for simulating complete dictionary
While, the correlation between pending part and corresponding dictionary is also enhanced in the sparse solution stage, improves sparse coefficient
Accuracy.
In order to make image reconstruction process have robustness, when it is implemented, can in advance to pending low resolution image,
Original high resolution trains image and original low-resolution training image to carry out subtracting the processing of mean value respectively, and it is complete to execute the above flow
After reconstruction, the corresponding mean value of pending low resolution image again add-back is come, the high-resolution shadow of final reconstruction is obtained
Picture.
When it is implemented, method provided by the present invention, which can be based on software technology, realizes automatic running flow, mould can also be used
Block mode realizes corresponding system.
The present invention provides a kind of super resolution image reconstruction system based on non-coupled mapping relations, comprises the following modules：
First module, for training image and original low-resolution training image using filtering original high resolution respectively
Device is filtered, and obtains corresponding pyramid image, is denoted as pyramid image A and pyramid image B respectively；
Second module, for based on extraction image sampling block, one all to be built respectively to each layer image of pyramid image A
A excessively complete dictionary, obtains high-resolution dictionary, and an excessively complete word is all built respectively to each layer image of pyramid image B
Allusion quotation obtains low-resolution dictionary, and is layered and carries out structural similarity analysis, if analyzed the structural similarity of a certain layer discontented
Sufficient predetermined threshold value then adjusts the filter parameter of this layer, and filtering adjusts the layer image to the work of the first module of order again, if full
Sufficient predetermined threshold value then records current filter parameter, determines that every layer all meets the work of predetermined threshold value post command third module；
Third module, for according to the filter parameter recorded in the second module, to pending low resolution image into
The same filtering operation of row, obtains corresponding pyramid image, is denoted as pyramid image C, to each layer of shadow of pyramid image C
As extraction image sampling block；
4th module utilizes the low of respective layer for being configured to signal matrix to every layer image of pyramid image C respectively
Resolution ratio dictionary carries out sparse expression to the signal matrix, obtains sparse coefficient；
5th module, for obtaining the high-resolution shadow of respective layer by the high-resolution dictionary of sparse coefficient and respective layer
Each layer of high resolution image is finally obtained final reconstructed image by picture by the operation of liftering.
Each module specific implementation can be found in corresponding steps, and it will not go into details by the present invention.
It will appreciated by the skilled person that can not only be passed through during dictionary constructs using the present invention
The processing method of layering, by the different of mapping relations between layers to synthesize the nonlinear coupling relationship of an entirety,
Enhance ability to express of the dictionary to real world, can also enhance pending during the sparse coding of layered shaping image
What part and the correlation between corresponding dictionary, enhance the accuracy of its sparse coefficient from.
Using remote sensing images in embodiment in the present invention, but it is not limited to remote sensing image.For other images,
For example, video image, natural image, medical image etc. all have extensive versatility, by the less-restrictive of objective factor.By mould
Draft experiment actual test the result shows that, this method have higher precision, can effectively keep image edge information while,
Restore high-resolution information to the greatest extent.
It should be noted that and understanding, the feelings of the spirit and scope of the present invention required by not departing from appended claims
Under condition, various modifications and improvements can be made to the present invention of foregoing detailed description.It is therefore desirable to the model of the technical solution of protection
It encloses and is not limited by given any specific exemplary teachings.
Claims (6)
1. a kind of super resolution image reconstruction method based on non-coupled mapping relations, which is characterized in that include the following steps：
Step 1, image and original low-resolution training image is trained to be filtered using filter original high resolution respectively,
Corresponding pyramid image is obtained, is denoted as pyramid image A and pyramid image B respectively；
Step 2, it is based on extraction image sampling block, an excessively complete word is all built respectively to each layer image of pyramid image A
Allusion quotation obtains high-resolution dictionary, all builds an excessively complete dictionary respectively to each layer image of pyramid image B, obtains low
Resolution ratio dictionary, and be layered and carry out structural similarity analysis, if the structural similarity analysis to a certain layer is unsatisfactory for default threshold
Value then adjusts the filter parameter of this layer, and return to step 1 filters again to be adjusted the layer image and remember if meeting predetermined threshold value
Record current filter parameter, determine every layer all meet predetermined threshold value after enter step 3；
Step 3, according to the filter parameter recorded in step 2, same filtering behaviour is carried out to pending low resolution image
Make, obtains corresponding pyramid image, be denoted as pyramid image C, each layer image extraction image sampling to pyramid image C
Block；
Step 4, signal matrix is configured to every layer image of pyramid image C respectively, utilizes the low-resolution dictionary of respective layer
Sparse expression is carried out to the signal matrix, obtains sparse coefficient；
Step 5, the high resolution image of respective layer is obtained by the high-resolution dictionary of sparse coefficient and respective layer, it finally will be every
One layer of high resolution image obtains final reconstructed image by the operation of liftering.
2. the super resolution image reconstruction method based on non-coupled mapping relations according to claim 1, it is characterised in that：Step
In rapid 2, the computational methods that structural similarity SSIM is carried out to any layer are as follows,
Wherein, the high resolution image after P and O respectively represents low resolution image and degrades, μPIt is equal to represent low resolution image
Value, μORepresent the high resolution image mean value after degrading, σPRepresent the variance of low resolution image, σORepresent the high score after degrading
The variance of resolution image, σPOThe covariance between high resolution image after representing low resolution image and degrading, parameter C1=
K1× L, C2=K2× L, L are the max pixel value in grayscale image, K1And K2It is preset constant；Low resolution image is golden word
A certain layer in tower image B, the high resolution image after degrading are the results after equivalent layer degrades in pyramid image A.
3. the super resolution image reconstruction method according to claim 1 or claim 2 based on non-coupled mapping relations, feature exist
In：In step 4, the realization method for carrying out sparse expression is, for each layer of image, image sampling block is arranged to chemical conversion row one by one
Vector is utilized respectively each column vector the low-resolution dictionary of respective layer, carries out sparse expression using Lasso algorithms, solves
Corresponding sparse coefficient.
4. a kind of super resolution image reconstruction system based on non-coupled mapping relations, which is characterized in that comprise the following modules：
First module, for respectively to original high resolution train image and original low-resolution training image using filter into
Row filtering, obtains corresponding pyramid image, is denoted as pyramid image A and pyramid image B respectively；
Second module, for based on extraction image sampling block, a mistake all to be built respectively to each layer image of pyramid image A
Complete dictionary obtains high-resolution dictionary, and an excessively complete dictionary is all built respectively to each layer image of pyramid image B,
Low-resolution dictionary is obtained, and is layered and carries out structural similarity analysis, if the structural similarity analysis to a certain layer is unsatisfactory for
Predetermined threshold value then adjusts the filter parameter of this layer, and filtering adjusts the layer image to the work of the first module of order again, if met
Predetermined threshold value then records current filter parameter, determines that every layer all meets the work of predetermined threshold value post command third module；
Third module, for according to the filter parameter recorded in the second module, being carried out to pending low resolution image same
The filtering operation of sample obtains corresponding pyramid image, is denoted as pyramid image C, is carried to each layer image of pyramid image C
Take image sampling block；
4th module utilizes the low resolution of respective layer for being configured to signal matrix to every layer image of pyramid image C respectively
Rate dictionary carries out sparse expression to the signal matrix, obtains sparse coefficient；
5th module, for obtaining the high resolution image of respective layer by the high-resolution dictionary of sparse coefficient and respective layer,
Each layer of high resolution image is finally obtained into final reconstructed image by the operation of liftering.
5. the super resolution image reconstruction system based on non-coupled mapping relations according to claim 4, it is characterised in that：The
In two modules, the computational methods that structural similarity SSIM is carried out to any layer are as follows,
Wherein, the high resolution image after P and O respectively represents low resolution image and degrades, μPIt is equal to represent low resolution image
Value, μORepresent the high resolution image mean value after degrading, σPRepresent the variance of low resolution image, σORepresent the high score after degrading
The variance of resolution image, σPOThe covariance between high resolution image after representing low resolution image and degrading, parameter C1=
K1× L, C2=K2× L, L are the max pixel value in grayscale image, K1And K2It is preset constant；Low resolution image is golden word
A certain layer in tower image B, the high resolution image after degrading are the results after equivalent layer degrades in pyramid image A.
6. the super resolution image reconstruction system based on non-coupled mapping relations, feature exist according to claim 4 or 5
In：In 4th module, the realization method for carrying out sparse expression is, for each layer of image, by image sampling block arranging one by one
At column vector, it is utilized respectively the low-resolution dictionary of respective layer for each column vector, sparse expression is carried out using Lasso algorithms,
Solve corresponding sparse coefficient.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201610231568.3A CN105931181B (en) | 2016-04-14 | 2016-04-14 | Super resolution image reconstruction method and system based on non-coupled mapping relations |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201610231568.3A CN105931181B (en) | 2016-04-14 | 2016-04-14 | Super resolution image reconstruction method and system based on non-coupled mapping relations |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN105931181A CN105931181A (en) | 2016-09-07 |
| CN105931181B true CN105931181B (en) | 2018-10-26 |
Family
ID=56838973
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201610231568.3A Active CN105931181B (en) | 2016-04-14 | 2016-04-14 | Super resolution image reconstruction method and system based on non-coupled mapping relations |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN105931181B (en) |
Families Citing this family (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN106651772B (en) * | 2016-11-25 | 2020-06-16 | 宁波大学 | Super-resolution reconstruction method of satellite cloud picture |
| CN106952226A (en) * | 2017-03-06 | 2017-07-14 | 武汉大学 | A kind of F MSA super resolution ratio reconstruction methods |
| CN107025632B (en) * | 2017-04-13 | 2020-06-30 | 首都师范大学 | Image super-resolution reconstruction method and system |
| CN108092668A (en) * | 2017-12-04 | 2018-05-29 | 上海无线电设备研究所 | A kind of linear FM signal restructing algorithm based on minimum sparse ruler sampling |
Citations (5)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN102542549A (en) * | 2012-01-04 | 2012-07-04 | 西安电子科技大学 | Multi-spectral and panchromatic image super-resolution fusion method based on compressive sensing |
| CN103020897A (en) * | 2012-09-28 | 2013-04-03 | 香港应用科技研究院有限公司 | Device for reconstructing based on super-resolution of multi-block single-frame image, system and method thereof |
| CN103325104A (en) * | 2013-06-09 | 2013-09-25 | 中山大学 | Facial image super-resolution reestablishing method based on iterative sparse representation |
| CN105338357A (en) * | 2015-09-29 | 2016-02-17 | 湖北工业大学 | Distributed video compressed sensing coding technical method |
| CN105488759A (en) * | 2015-12-09 | 2016-04-13 | 南京邮电大学 | Image super-resolution reconstruction method based on local regression model |
-
2016
- 2016-04-14 CN CN201610231568.3A patent/CN105931181B/en active Active
Patent Citations (5)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN102542549A (en) * | 2012-01-04 | 2012-07-04 | 西安电子科技大学 | Multi-spectral and panchromatic image super-resolution fusion method based on compressive sensing |
| CN103020897A (en) * | 2012-09-28 | 2013-04-03 | 香港应用科技研究院有限公司 | Device for reconstructing based on super-resolution of multi-block single-frame image, system and method thereof |
| CN103325104A (en) * | 2013-06-09 | 2013-09-25 | 中山大学 | Facial image super-resolution reestablishing method based on iterative sparse representation |
| CN105338357A (en) * | 2015-09-29 | 2016-02-17 | 湖北工业大学 | Distributed video compressed sensing coding technical method |
| CN105488759A (en) * | 2015-12-09 | 2016-04-13 | 南京邮电大学 | Image super-resolution reconstruction method based on local regression model |
Also Published As
| Publication number | Publication date |
|---|---|
| CN105931181A (en) | 2016-09-07 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| Xiang et al. | Deep learning for image inpainting: A survey | |
| CN110119780B (en) | Hyper-spectral image super-resolution reconstruction method based on generation countermeasure network | |
| CN114140353B (en) | Swin-Transformer image denoising method and system based on channel attention | |
| Karami et al. | Persian sign language (PSL) recognition using wavelet transform and neural networks | |
| US10593021B1 (en) | Motion deblurring using neural network architectures | |
| CN104778671B (en) | A kind of image super-resolution method based on SAE and rarefaction representation | |
| CN104835130A (en) | Multi-exposure image fusion method | |
| CN105931181B (en) | Super resolution image reconstruction method and system based on non-coupled mapping relations | |
| WO2020133636A1 (en) | Method and system for intelligent envelope detection and warning in prostate surgery | |
| CN106157244A (en) | A kind of QR Code Image Super-resolution Reconstruction method based on rarefaction representation | |
| CN109166130A (en) | A kind of image processing method and image processing apparatus | |
| CN109214989B (en) | Single image super resolution ratio reconstruction method based on Orientation Features prediction priori | |
| CN103279933B (en) | A kind of single image super resolution ratio reconstruction method based on bilayer model | |
| CN105046672A (en) | Method for image super-resolution reconstruction | |
| CN104751162A (en) | Hyperspectral remote sensing data feature extraction method based on convolution neural network | |
| CN101950415B (en) | Shape semantic model constraint-based face super-resolution processing method | |
| CN106897987A (en) | Image interfusion method based on translation invariant shearing wave and stack own coding | |
| CN105335929A (en) | Depth map super-resolution method | |
| CN109559278B (en) | Super resolution image reconstruction method and system based on multiple features study | |
| CN111626912A (en) | Watermark removing method and device | |
| CN109903373A (en) | A kind of high quality human face generating method based on multiple dimensioned residual error network | |
| CN116416156A (en) | Swin transducer-based medical image denoising method | |
| CN112488971A (en) | Medical image fusion method for generating countermeasure network based on spatial attention mechanism and depth convolution | |
| CN115330643A (en) | Earthquake denoising method based on convolution neural network and visual transformation neural network | |
| CN116452930A (en) | Multispectral image fusion method and multispectral image fusion system based on frequency domain enhancement in degradation environment |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| C06 | Publication | ||
| PB01 | Publication | ||
| C10 | Entry into substantive examination | ||
| SE01 | Entry into force of request for substantive examination | ||
| GR01 | Patent grant | ||
| GR01 | Patent grant |