CN102222321A - Blind reconstruction method for video sequence - Google Patents
Blind reconstruction method for video sequence Download PDFInfo
- Publication number
- CN102222321A CN102222321A CN 201110134728 CN201110134728A CN102222321A CN 102222321 A CN102222321 A CN 102222321A CN 201110134728 CN201110134728 CN 201110134728 CN 201110134728 A CN201110134728 A CN 201110134728A CN 102222321 A CN102222321 A CN 102222321A
- Authority
- CN
- China
- Prior art keywords
- image
- resolution
- video sequence
- super
- formula
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Pending
Links
- 238000000034 method Methods 0.000 title claims abstract description 64
- 230000033001 locomotion Effects 0.000 claims abstract description 43
- PXFBZOLANLWPMH-UHFFFAOYSA-N 16-Epiaffinine Natural products C1C(C2=CC=CC=C2N2)=C2C(=O)CC2C(=CC)CN(C)C1C2CO PXFBZOLANLWPMH-UHFFFAOYSA-N 0.000 claims abstract description 20
- 239000011159 matrix material Substances 0.000 claims description 20
- 238000005516 engineering process Methods 0.000 claims description 12
- 230000009466 transformation Effects 0.000 claims description 12
- 239000004744 fabric Substances 0.000 claims description 11
- 238000005457 optimization Methods 0.000 claims description 8
- 238000005070 sampling Methods 0.000 claims description 7
- 238000004458 analytical method Methods 0.000 claims description 4
- 230000004304 visual acuity Effects 0.000 claims description 4
- 239000000203 mixture Substances 0.000 claims description 3
- 230000000694 effects Effects 0.000 abstract description 15
- 238000012545 processing Methods 0.000 abstract description 10
- 238000012544 monitoring process Methods 0.000 abstract description 2
- 230000006870 function Effects 0.000 description 30
- 230000003044 adaptive effect Effects 0.000 description 9
- 238000006731 degradation reaction Methods 0.000 description 6
- 230000015556 catabolic process Effects 0.000 description 5
- 238000011160 research Methods 0.000 description 5
- 230000003412 degenerative effect Effects 0.000 description 3
- 238000013461 design Methods 0.000 description 3
- 230000002146 bilateral effect Effects 0.000 description 2
- 238000006243 chemical reaction Methods 0.000 description 2
- 238000002790 cross-validation Methods 0.000 description 2
- 238000011161 development Methods 0.000 description 2
- 230000018109 developmental process Effects 0.000 description 2
- 238000009792 diffusion process Methods 0.000 description 2
- 230000002708 enhancing effect Effects 0.000 description 2
- 230000010354 integration Effects 0.000 description 2
- 208000004547 Hallucinations Diseases 0.000 description 1
- 230000006978 adaptation Effects 0.000 description 1
- 238000013459 approach Methods 0.000 description 1
- 238000012512 characterization method Methods 0.000 description 1
- 230000000295 complement effect Effects 0.000 description 1
- 238000007906 compression Methods 0.000 description 1
- 230000008878 coupling Effects 0.000 description 1
- 238000010168 coupling process Methods 0.000 description 1
- 238000005859 coupling reaction Methods 0.000 description 1
- 230000007812 deficiency Effects 0.000 description 1
- 238000002059 diagnostic imaging Methods 0.000 description 1
- 238000010586 diagram Methods 0.000 description 1
- 239000003623 enhancer Substances 0.000 description 1
- 238000001914 filtration Methods 0.000 description 1
- 238000003709 image segmentation Methods 0.000 description 1
- 238000003384 imaging method Methods 0.000 description 1
- 238000011835 investigation Methods 0.000 description 1
- 238000007619 statistical method Methods 0.000 description 1
- 238000012549 training Methods 0.000 description 1
- 230000000007 visual effect Effects 0.000 description 1
Images
Landscapes
- Image Processing (AREA)
Abstract
The invention discloses a blind reconstruction method for a video sequence. The method comprises the following steps of: (1) initializing an algorithm, including initializing an affine motion initial parameter, a fuzzy kernel function and a high-resolution video sequence image; (2) establishing an image enhancement observation model; (3) performing the following iterative algorithms: 1, identifying the fuzzy kernel function; 2, performing super-resolution reconstruction on the video sequence image to obtain a high-resolution image; 3, estimating an affine motion parameter; and 4, judging whether results obtained in the steps 2 and 3 satisfy an iteration ending condition, and otherwise, returning to the step 1 until the condition is satisfied; and (4) obtaining a final video sequence reconstruction image. According to the method, the quality of a reconstructed video can be enhanced effectively, blind reconstruction can be performed on any group of low-resolution video sequences according to the characteristics of the video sequences, and the image display effect is enhanced; and the method plays an important theoretical and practical role in processing random remote sensing images, processing medical videos, developing military safety monitoring systems and the like.
Description
Technical field
The invention belongs to digital picture enhancement techniques field, relate to the identification technique of fuzzy kernel function in the blind reconstruction of super-resolution of video sequence image super-resolution reconstruction technique and low-resolution image.
Background technology
In recent years, the super-resolution rebuilding technology has become a research focus of image processing field, is widely used in a plurality of fields such as remote sensing, medical imaging and military affairs.Yet current super resolution ratio reconstruction method is not considered different sequence of low resolution pictures characteristics separately, lacks adaptivity; And, do not carry out deep research to the fuzzy step in the reconstruction model, they often in process of reconstruction, configure the fuzzy kernel function in the image degradation process in advance, thereby do not reach blind completely reconstruction; Simultaneously, application is confined to several common LR (low resolution) images mostly, and for the wider LR video sequence of application prospect, the super-resolution rebuilding technology is not brought into play its due effect, reaches the super-resolution real-time reconstruction of video sequence.We can say that the blind reconstruction technique of the super-resolution rebuilding of video sequence and super-resolution is two big hot issues of current super-resolution technique research.The blind reconstruction of the super-resolution of video sequence is the extension and the development of super-resolution rebuilding technology, its main task is for given low-resolution video sequence, automatically it is converted to high-resolution video in real time in the short period of time, in this process,, can be applicable to many hot spot application fields such as security monitoring, HDTV without any need for artificial intervention and setting.Though the blind reconstruction technique of video sequence super-resolution has received certain concern, valuable achievement is actually rare, remains more to be furtherd investigate.
The super-resolution rebuilding technology of present image be based on mostly the image modeling theory regular super resolution rate reconstruction method and based on study the super-resolution rebuilding method.(Regularized SR Reconstruction RSRR) mainly is divided into based on the super-resolution rebuilding of maximum a posteriori probability (MAP) framework with based on the super-resolution rebuilding of regularization theory (regular space several picture model) based on the regular super resolution rate reconstruction method of image modeling theory.Under the MAP framework, super-resolution rebuilding is a statistical Inference based on MRF (Markov random field) statistics prior model.Document (IEEE Transactions on Image Processing, 1997,6 (12): 1646-1658) MAP is estimated add in the common factor of constraint set, for super-resolution rebuilding has been constructed a new convex set solution space as a convex set constraint.Document (IEEE Transactions on Image Processing, 2007,16 (2): 479-490) proposed a kind of image super-resolution rebuilding method based on the maximum a posteriori probability model, this method is estimated that image segmentation and image motion also incorporate wherein when carrying out super-resolution rebuilding.Super-resolution rebuilding basic framework based on the regularization theory is, obtains the regularization energy functional according to the degradation model of sequence of low resolution pictures and the regular terms structure of iconic model correspondence, obtains high-definition picture by the minimization of energy functional.With the MAP frame clsss seemingly, the appropriate design of regular space several picture model is directly determining the visual effect of super-resolution rebuilding image.Document (Proceedings of International Conference onPattern Recognition, 2000,1:600-605) propose based on continuous total variation model (Total Variation, text image Sequences Superresolution reconstruction algorithm TV).Document (IEEE Transactions on Image Processing, 2004,13 (10): 1327-1344) propose a kind of bilateral filtering and l
1The bilateral total variation model of norm coupling is used for super-resolution image reconstruction, and adopts the piece estimation approach to carry out motion estimation.Document (Journal of Electronic Imaging, 2003,12 (2): 244-251) propose to strengthen and anisotropy diffusion variation super-resolution rebuilding algorithm based on the edge of diffusion tensor.In addition, based on forward and oppositely partial differential equation (Partial Differential Equation, has based on the quick super-resolution reconstruction algorithm of adaptive wiener filter, based on the statistical analysis technique of lowest mean square etc. and to rebuild effect preferably image denoising PDE) and super-resolution enhancement algorithms.
The main thought of rebuilding research based on the super-resolution (example-based) of study is to obtain prior imformation and then improve resolution by learning existing HR (high resolving power) image sequence.Less and resolution improves under the bigger situation of multiple at the frame number of LR image sequence, and the complementary sample information that the LR image sequence can provide is relatively limited, and deficiency is so that the super-resolution rebuilding algorithm recovers more high-frequency information.Priori about image itself just seems extremely important in this case.Except traditional MRF statistics prior model and regular space several picture model can provide the prior imformation of image, another kind of important method is carried out learning training by nerual network technique exactly and is obtained.Baker etc. have proposed the face image super-resolution rebuilding algorithm of a kind of Hallucination of being called, by introducing the feature identification of LR image, obtain than the better effect of reconstruction algorithm such as super-resolution R under big resolution enhancer situation.
The current overwhelming majority's super-resolution rebuilding algorithm is often paid attention to the effect of reconstructed image, is not very high to the requirement of arithmetic speed.Video super-resolution is rebuild, for example high-definition television standard and synthetic video zoom, and not only requirement can improve the spatial resolution of each frame of video, and requirement can the corresponding high-resolution video sequence of Fast Reconstruction.Current technology has noted solving this class problem, for example, can carry out the HR image of Fast estimation present frame according to the present frame of estimation of former frame HR image and LR video sequence.The reconstruction of compressed video super resolution reconstruction and traditional super-resolution is different in addition, and compressed video is represented as the motion vector and the conversion coefficient of sequence.Compressed video super resolution reconstruction at first needs video is decoded usually, rebuilds after being recovered to the normal image sequence again.Therefore, quantizing noise, staircase effect and the ringing effect etc. that produce in the compressed video super resolution reconstruction process need consideration compression process.
At present, the degradation model that most of super-resolution rebuilding algorithms are all supposed the LR image sequence is known, especially supposes known fuzzy kernel function.But under a lot of actual conditions, the degenerative process of image is a parameter model unknown or that only know fuzzy kernel function.Therefore, need carry out identification to fuzzy kernel function, super-resolution rebuilding algorithm in this case is called the blind reconstruction of super-resolution.Document (IEEE Transactions on Image Processing, 2001,10 (9): 1299-1308) according to broad sense cross validation (Generalized Cross Validation, GCV) and the Gauss integration theory carry out the estimation of fuzzy parameter, obtain unknown parameter by the nonlinear optimization problem of finding the solution polytomy variable, and utilize the Gauss integration theory to estimate the GCV objective function accurately and effectively.Because the blind reconstruction the complex nature of the problem of super-resolution, the blind reconstruction achievement in research of super-resolution is relatively limited at present, so the blind reconstructed value of super-resolution gets special concern and further investigation.
Summary of the invention
Purpose of the present invention, be to provide a kind of video sequence blind method for reconstructing, it can effectively improve the quality of reconstruction video, reach for any one group of low-resolution video sequence, can both carry out blind reconstruction according to its characteristics, improve the display effect of image, this all has important theory and practical significance for aspects such as at random remote sensing image processing, medical science Video processing, military security Development of supervision system based on PLC.
In order to reach above-mentioned purpose, the technical solution adopted in the present invention is:
The blind method for reconstructing of a kind of video sequence comprises the steps:
(1) algorithm is carried out initialization, comprise initialization affine motion initial parameter, initialization fuzzy kernel function and initialization high-resolution video sequence image;
(2) set up figure image intensifying observation model;
(3) carry out following iterative algorithm:
1. carry out the fuzzy kernel function identification;
2. the super-resolution rebuilding of video sequence image draws high-definition picture;
3. carry out the affine motion parameter estimation;
The 2. 4. judge, 3. whether the result that draws of step satisfies stopping criterion for iteration, if do not satisfy, return 1., up to satisfying condition;
(4) draw final video sequence reconstructed image.
In the above-mentioned steps (1), select Gaussian Blur, high burnt fuzzy or linear fuzzy initial value for use as fuzzy kernel function.
In the above-mentioned steps (1), affine motion initial parameter α
(0)Choosing method be:
Order:
θ wherein, s, t represent the affine parameter of rotation, Pan and Zoom respectively, and k represents the LR observed image of k width of cloth m * n, and (1) formula of employing represents the HR image is obtained the process of LR image after affined transformation:
f
k(u,v,t)=f
k(c
1ku+c
2kv+c
3k,c
4ku+c
5kv+c
6k,t-1) (1)
Wherein, f
kBe low-resolution image;
If:
In the above-mentioned steps (1), high resolving power initial pictures z
(0)Pass through α
(0)Obtain, method is as follows:
z
(0)=(W
T(α
(0))W(α
(0))+I)
-1W
T(α
(0))y (4)
The detailed process of above-mentioned steps (2) is: order:
Wherein, y
kBe the vector of LR observed image N * 1 (N=mn) of composition after the dictionary ordering of k width of cloth m * n, if r
1And r
2Be respectively the down-sampling factor of level and vertical direction, so, z is that size is r
1M * r
2The r that the HR image of n forms after the dictionary ordering
1r
2The vector of N * 1, M
kFor size is r
1r
2N * r
1r
2The affine transformation matrix of N, B
kFor size is r
1r
2N * r
1r
2The fuzzy matrix of N, D are that size is N * r
1r
2N down-sampling matrix, n
kNoise vector for N * 1; In the following formula
Be the transformation matrix of image movement at the uniform velocity,
Be the transformation matrix of the non-movement at the uniform velocity of image, simultaneously entire image regarded as a complete motion field, and established m
k(x
U, v)=[m
K, u(x
U, v), m
K, v(x
U, v)] be the motion vector of image slices vegetarian refreshments, wherein x
U, v=[x
u, x
v] be the pixel of image, same, the motion vector of image also is divided into the non-movement at the uniform velocity vector of movement at the uniform velocity vector sum, makes it be:
In the above-mentioned steps (3), also the image reconstruction model is optimized, obtains:
Wherein, α
kBe the affine motion parameter, then the image reconstruction formula is:
Adopt again based on projecting method (10) formula is optimized of least square and find the solution.
After the above-mentioned steps (3), also comprise steps A, the process of described steps A is:
[1] according to the maximum a posteriori probability technology super-resolution rebuilding model of deriving, as follows:
(11) in the formula,
Be the estimated value of HR image, Δ z
(i)=z
(i)-z
(i-1), z
(i)It is resulting HR image after the i time iteration;
[2] (11) formula is carried out analysis optimization, draw the super-resolution rebuilding computation model, as follows:
Wherein, λ
1, λ
2For adjusting parameter, Q
1Be stable matrix;
[3] the different low-resolution images of difference are to the influence of super-resolution rebuilding image, shown in (13) formula:
(13) in the formula, l
kBe the importance weight of k width of cloth LR image to SR influence that reconstructed image constitutes;
[4] solve (3) formula by the fastest gradient method, draw final reconstructed image.
After adopting such scheme, the present invention sets up figure image intensifying observation model according to the video sequence characteristics, on this basis video sequence image is carried out blind reconstruction.The step of blind reconstruction is divided into: steps such as fuzzy kernel function identification, parameter optimization.For the effect of further boosting algorithm, the present invention also incorporates the adaptivity technology in algorithm, and making can be according to the timely corrected parameter of the characteristics of input picture when video sequence is rebuild.
Description of drawings
Fig. 1 is a process flow diagram of the present invention.
Specific implementation method
The present invention relates to the optimization at video sequence of the automatic selection of super-resolution rebuilding algorithm, fuzzy kernel function of digital picture and whole algorithm, its specific implementation method may further comprise the steps:
1, the initialization of algorithm
Initialization problem in the algorithm promptly at first will provide initial value before algorithm carries out
z
(0)And initial fuzzy kernel function, the concrete steps that the present invention proposes are:
[1] selects for use Gaussian Blur (can adopt other fuzzy kernel function as the initial value of fuzzy kernel function?);
[2] affine motion initial parameter α
(0)Choosing method as follows:
Order:
θ wherein, s, t represent the affine parameter of rotation, Pan and Zoom respectively, and the process that the HR image is obtained the LR image after affined transformation can be represented with (1) formula:
f
k(u,v,t)=f
k(c
1ku?+c
2kv?+c
3k,c
4ku?+c
5kv?+c
6k,t-1) (1)
Wherein, f
kBe low-resolution image.
If:
[3] high resolving power initial pictures z
(0)Pass through α
(0)Obtain, method is as follows:
z
(0)=(W
T(α
(0))W(α
(0))+I)
-1W
T(α
(0))y (4)
2, set up figure image intensifying observation model
(1) foundation of figure image intensifying observation model
One width of cloth HR image is added white Gaussian noise through processing such as fuzzy, affine motion conversion, sampling, has just become a width of cloth LR image, the degenerative process of Here it is image, and just the LR image carries out SR image observation model commonly used when rebuilding.If: there is p width of cloth size to be the LR observed image of m * n
So, the degradation model according to image has
y
k=DB
kM
kz+n
k,1≤k≤p (5)
Here, y
kBe the vector of LR observed image N * 1 (N=mn) of composition after the dictionary ordering of k width of cloth m * n, if r
1And r
2Be respectively the down-sampling factor of level and vertical direction, so, z is that size is r
1M * r
2The r that the HR image of n forms after the dictionary ordering
1r
2The vector of N * 1, M
kFor size is r
1r
2N * r
1r
2The affine transformation matrix of N, B
kFor size is r
1r
2N * r
1r
2The fuzzy matrix of N, D are that size is N * r
1r
2N down-sampling matrix, n
kNoise vector for N * 1.
Though above-mentioned observation model is used extensively,, it also is not suitable for the non-movement at the uniform velocity of image, therefore the consecutive frame when video sequence image not only has the 2D plane motion, and in 3D when motion, arranged, this model is the motion of token image accurately, thereby influences the effect of image reconstruction.
The present invention is based on above-mentioned analysis and consider, draw a kind of enhancing image observation model of suitable sequence of video images super-resolution rebuilding, specific as follows, order:
1≤k≤p (6)
In the following formula
Be the transformation matrix of image movement at the uniform velocity,
Be the transformation matrix of the non-movement at the uniform velocity of image, simultaneously entire image regarded as a complete motion field, and established m
k(x
U, v)=[m
K, u(x
U, v), m
K, v(x
U, v)] be the motion vector of image slices vegetarian refreshments, wherein x
U, v=[x
u, x
v] be the pixel of image, same, the motion vector of image also is divided into the non-movement at the uniform velocity vector of movement at the uniform velocity vector sum, makes it be:
The image reconstruction model that the present invention draws is characterization system more accurately, and can draw the global motion territory of image, promptly in image super-resolution rebuilding, carries out the estimation of image.
(2) at the super-resolution algorithms optimization of video sequence
The present invention is optimized the super-resolution rebuilding algorithm at video sequence, as follows in detail:
When in the blind reconstruction of super-resolution, setting up figure image intensifying observation model, image registration is taken into account, when carrying out super-resolution rebuilding, carry out image registration simultaneously, to effectively reduce operation time, effectively strengthen the real-time that SR rebuilds, like this, the enhancing image observation model of (6) formula correspondingly becomes:
(9) in the formula, α
kBe the affine motion parameter.The image reconstruction formula is:
When video sequence image SR rebuilds, adopt certain optimization method to calculate the affine motion parameter alpha simultaneously
kWith HR image z, minimizing computing time that can be by a relatively large margin.The present invention adopts and finds the solution based on projecting method (10) formula is optimized of least square.
3, the identification of fuzzy kernel function in the blind reconstruction of SR
In the super-resolution rebuilding process,, how effectively to determine fuzzy matrix B for the image degradation model of (5) formula
k, promptly identification fuzzy kernel function how is the difficult point of blind reconstruction technique always.To this, the method that the present invention provides a kind of fuzzy decision solves this problem, and concrete steps are as follows:
[1] the present invention puts in order and analyzes all fuzzy kernel functions, sets up fuzzy kernel function storehouse Θ commonly used, and is as follows:
Having comprised 3 kinds of fuzzy kernel functions commonly used in the fuzzy kernel function storehouse of following formula, is respectively that out of focus is fuzzy, Gaussian Blur and linear fuzzy.Kind in this fuzzy kernel function storehouse is not fixed, and can increase other fuzzy kernel function according to the accuracy requirement of identification in the storehouse.
[2] key of the blind reconstruction of super-resolution is exactly how accurately to select the most near the fuzzy kernel function of truth in the Θ of fuzzy kernel function storehouse, and the present invention designs a kind of ergodic algorithm, and is as follows:
For?all?k?do
For?a1l?i?do
end?For
end?For
Make the smear out effect matrix B
kWith the Laplce of kernel function under certain parameter in the fuzzy kernel function storehouse apart from minimum, this kernel function can be appointed as current SR rebuild in the fuzzy kernel function of degradation model, formula is as follows:
In the following formula, i represents fuzzy kind in the fuzzy storehouse, and θ represents the fuzzy parameter of fuzzy kernel function.
4, the selection of adaptive weight
The present invention has introduced adaptive thought when super-resolution rebuilding, strengthened the effect of reconstruction algorithm greatly, and still, the system of selection of the adaptive weight in the adaptive algorithm has bigger influence to the speed of algorithm, the adaptive weight l in (7) formula
kHow to select, will directly influence the effect of video sequence SR reconstruction and the real-time of algorithm, the step that the present invention designs weights is as follows:
Super-resolution rebuilding adaptive algorithm concrete steps are as follows:
[1] according to the maximum a posteriori probability technology super-resolution rebuilding model of deriving, as follows:
(11) in the formula,
Be the estimated value of HR image, Δ z
(i)=z
(i)-z
(i-1), z
(i)It is resulting HR image after the i time iteration.
[2] (11) formula is carried out analysis optimization, draw the super-resolution rebuilding computation model, as follows:
Wherein, λ
1, λ
2For adjusting parameter, Q
1Be stable matrix.
[3] introduce adaptive weighted thought, distinguish of the influence of different low-resolution images to the super-resolution rebuilding image, make the super-resolution rebuilding algorithm have adaptivity, token image reconstruction model so more accurately, algorithm also can reach convergence faster and better effect, shown in (13) formula:
(13) in the formula, l
kBe the importance weight of k width of cloth LR image to SR influence that reconstructed image constitutes.
[4] solve (3) formula by the fastest gradient method, draw final reconstructed image.
In sum, the blind reconstruction technology of the image super-resolution that the present invention relates to is the characteristics according to the different sequence of low resolution pictures that observe, the adaptively fuzzy kernel function in the recognisable image degenerative process; Then a kind of fast and effectively method for registering images and the blind reconstruction of image super-resolution are organically blended; In the blind reconstruction process of super-resolution, consider different low-resolution images to rebuilding the impact of effect, the video sequence of carrying out that can self adaptation is rebuild, and improves the video image display performance. The image quality can be effectively promoted based on the blind reconstruction technology of given video sequence image super-resolution, the exploitation of civilian video image process software can be instructed.
Claims (7)
1. the blind method for reconstructing of video sequence is characterized in that comprising the steps:
(1) algorithm is carried out initialization, comprise initialization affine motion initial parameter, initialization fuzzy kernel function and initialization high-resolution video sequence image;
(2) set up figure image intensifying observation model;
(3) carry out following iterative algorithm:
1. carry out the fuzzy kernel function identification;
2. the super-resolution rebuilding of video sequence image draws high-definition picture;
3. carry out the affine motion parameter estimation;
The 2. 4. judge, 3. whether the result that draws of step satisfies stopping criterion for iteration, if do not satisfy, return 1., up to satisfying condition;
(4) draw final video sequence reconstructed image.
2. the blind method for reconstructing of a kind of video sequence as claimed in claim 1 is characterized in that: in the described step (1), select Gaussian Blur, high burnt fuzzy or linear fuzzy initial value as fuzzy kernel function for use.
3. the blind method for reconstructing of a kind of video sequence as claimed in claim 1 or 2 is characterized in that in the described step (1) affine motion initial parameter α
(0)Choosing method be:
Order:
θ wherein, s, t represent the affine parameter of rotation, Pan and Zoom respectively, and k represents the LR observed image of k width of cloth m * n, and (1) formula of employing represents the HR image is obtained the process of LR image after affined transformation:
f
k(u,v,t)=f
k(c
1ku+c
2kv+c
3k,c
4ku+c
5kv+c
6k,t-1) (1)
Wherein, f
kBe low-resolution image;
If:
4. as claim 1, the blind method for reconstructing of 2 or 3 described a kind of video sequences, it is characterized in that in the described step (1) high resolving power initial pictures z
(0)Pass through α
(0)Obtain, method is as follows:
z
(0)=(W
T(α
(0))W(α
(0))+I)
-1W
T(α
(0))y (4)
Wherein,
5. the blind method for reconstructing of a kind of video sequence as claimed in claim 1 is characterized in that the detailed process of described step (2) is: order:
1≤k≤p (6)
Wherein, y
kBe the vector of LR observed image N * 1 (N=mn) of composition after the dictionary ordering of k width of cloth m * n, if r
1And r
2Be respectively the down-sampling factor of level and vertical direction, so, z is that size is r
1M * r
2The r that the HR image of n forms after the dictionary ordering
1r
2The vector of N * 1, M
kFor size is r
1r
2N * r
1r
2The affine transformation matrix of N, B
kFor size is r
1r
2N * r
1r
2The fuzzy matrix of N, D are that size is N * r
1r
2N down-sampling matrix, n
kNoise vector for N * 1; In the following formula
Be the transformation matrix of image movement at the uniform velocity,
Be the transformation matrix of the non-movement at the uniform velocity of image, simultaneously entire image regarded as a complete motion field, and established m
k(x
U, v)=[m
K, u(x
U, v), m
K, v(x
U, v)] be the motion vector of image slices vegetarian refreshments, wherein x
U, v=[x
u, x
v] be the pixel of image, same, the motion vector of image also is divided into the non-movement at the uniform velocity vector of movement at the uniform velocity vector sum, makes it be:
6. the blind method for reconstructing of a kind of video sequence as claimed in claim 5 is characterized in that also the image reconstruction model being optimized in the described step (3), obtains:
Wherein, α
kBe the affine motion parameter, then the image reconstruction formula is:
Adopt again based on projecting method (10) formula is optimized of least square and find the solution.
7. the blind method for reconstructing of a kind of video sequence as claimed in claim 6 is characterized in that: after the described step (3), also comprise steps A, the process of described steps A is:
[1] according to the maximum a posteriori probability technology super-resolution rebuilding model of deriving, as follows:
(11) in the formula,
Be the estimated value of HR image, Δ z
(i)=z
(i)-z
(i-1), z
(i)It is resulting HR image after the i time iteration;
[2] (11) formula is carried out analysis optimization, draw the super-resolution rebuilding computation model, as follows:
Wherein, λ
1, λ
2For adjusting parameter, Q
1Be stable matrix;
[3] the different low-resolution images of difference are to the influence of super-resolution rebuilding image, shown in (13) formula:
(13) in the formula, l
kBe the importance weight of k width of cloth LR image to SR influence that reconstructed image constitutes;
[4] solve (3) formula by the fastest gradient method, draw final reconstructed image.
Priority Applications (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN 201110134728 CN102222321A (en) | 2011-05-24 | 2011-05-24 | Blind reconstruction method for video sequence |
Applications Claiming Priority (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN 201110134728 CN102222321A (en) | 2011-05-24 | 2011-05-24 | Blind reconstruction method for video sequence |
Publications (1)
Publication Number | Publication Date |
---|---|
CN102222321A true CN102222321A (en) | 2011-10-19 |
Family
ID=44778866
Family Applications (1)
Application Number | Title | Priority Date | Filing Date |
---|---|---|---|
CN 201110134728 Pending CN102222321A (en) | 2011-05-24 | 2011-05-24 | Blind reconstruction method for video sequence |
Country Status (1)
Country | Link |
---|---|
CN (1) | CN102222321A (en) |
Cited By (6)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN103606130A (en) * | 2013-10-22 | 2014-02-26 | 中国电子科技集团公司第二十八研究所 | Infrared degraded image adaptive restoration method |
CN104574338A (en) * | 2015-01-26 | 2015-04-29 | 西安交通大学 | Remote sensing image super-resolution reconstruction method based on multi-angle linear array CCD sensors |
CN106791273A (en) * | 2016-12-07 | 2017-05-31 | 重庆大学 | A kind of video blind restoration method of combination inter-frame information |
CN109543548A (en) * | 2018-10-26 | 2019-03-29 | 桂林电子科技大学 | A kind of face identification method, device and storage medium |
CN109819321A (en) * | 2019-03-13 | 2019-05-28 | 中国科学技术大学 | A kind of video super-resolution Enhancement Method |
CN110532853A (en) * | 2019-07-09 | 2019-12-03 | 中国空间技术研究院 | The classification method and device of remote sensing time-out phase data |
Citations (2)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
US6470097B1 (en) * | 1999-01-22 | 2002-10-22 | Siemens Corporation Research, Inc. | Total variational blind image restoration from image sequences |
CN101441765A (en) * | 2008-11-19 | 2009-05-27 | 西安电子科技大学 | Self-adapting regular super resolution image reconstruction method for maintaining edge clear |
-
2011
- 2011-05-24 CN CN 201110134728 patent/CN102222321A/en active Pending
Patent Citations (2)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
US6470097B1 (en) * | 1999-01-22 | 2002-10-22 | Siemens Corporation Research, Inc. | Total variational blind image restoration from image sequences |
CN101441765A (en) * | 2008-11-19 | 2009-05-27 | 西安电子科技大学 | Self-adapting regular super resolution image reconstruction method for maintaining edge clear |
Non-Patent Citations (2)
Title |
---|
《中国图象图形学报》 20101231 杨欣 等 基于非线性最小二乘的图像自适应SR重建以及运动估计 1714-1719 第15卷, 第12期 * |
《宇航学报》 20100131 杨欣 等 基于最大后验概率的SAR图像自适应超分辨率盲重建 217-221 第31卷, 第1期 * |
Cited By (10)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN103606130A (en) * | 2013-10-22 | 2014-02-26 | 中国电子科技集团公司第二十八研究所 | Infrared degraded image adaptive restoration method |
CN104574338A (en) * | 2015-01-26 | 2015-04-29 | 西安交通大学 | Remote sensing image super-resolution reconstruction method based on multi-angle linear array CCD sensors |
CN104574338B (en) * | 2015-01-26 | 2017-04-26 | 西安交通大学 | Remote sensing image super-resolution reconstruction method based on multi-angle linear array CCD sensors |
CN106791273A (en) * | 2016-12-07 | 2017-05-31 | 重庆大学 | A kind of video blind restoration method of combination inter-frame information |
CN106791273B (en) * | 2016-12-07 | 2019-08-20 | 重庆大学 | A kind of video blind restoration method of combination inter-frame information |
CN109543548A (en) * | 2018-10-26 | 2019-03-29 | 桂林电子科技大学 | A kind of face identification method, device and storage medium |
CN109819321A (en) * | 2019-03-13 | 2019-05-28 | 中国科学技术大学 | A kind of video super-resolution Enhancement Method |
CN109819321B (en) * | 2019-03-13 | 2020-06-26 | 中国科学技术大学 | Video super-resolution enhancement method |
CN110532853A (en) * | 2019-07-09 | 2019-12-03 | 中国空间技术研究院 | The classification method and device of remote sensing time-out phase data |
CN110532853B (en) * | 2019-07-09 | 2021-10-15 | 中国空间技术研究院 | Remote sensing time-exceeding phase data classification method and device |
Similar Documents
Publication | Publication Date | Title |
---|---|---|
Chen et al. | Real-world single image super-resolution: A brief review | |
Li et al. | Video superresolution via motion compensation and deep residual learning | |
CN113362223B (en) | Image super-resolution reconstruction method based on attention mechanism and two-channel network | |
CN106952228A (en) | The super resolution ratio reconstruction method of single image based on the non local self-similarity of image | |
CN109523470A (en) | A kind of depth image super resolution ratio reconstruction method and system | |
CN110136060B (en) | Image super-resolution reconstruction method based on shallow dense connection network | |
Ju et al. | BDPK: Bayesian dehazing using prior knowledge | |
CN102222321A (en) | Blind reconstruction method for video sequence | |
CN110689482A (en) | Face super-resolution method based on supervised pixel-by-pixel generation countermeasure network | |
CN103049885A (en) | Super-resolution image reconstruction method using analysis sparse representation | |
CN102243711A (en) | Neighbor embedding-based image super-resolution reconstruction method | |
Lohdefink et al. | Self-supervised domain mismatch estimation for autonomous perception | |
CN108280804A (en) | A kind of multi-frame image super-resolution reconstruction method | |
CN105631890B (en) | Picture quality evaluation method out of focus based on image gradient and phase equalization | |
Pu et al. | Removing rain based on a cycle generative adversarial network | |
Ziwei et al. | Overview on image super resolution reconstruction | |
CN105931189A (en) | Video ultra-resolution method and apparatus based on improved ultra-resolution parameterized model | |
CN109615576B (en) | Single-frame image super-resolution reconstruction method based on cascade regression basis learning | |
Wang et al. | Single image haze removal via attention-based transmission estimation and classification fusion network | |
Deng et al. | Efficient test-time adaptation for super-resolution with second-order degradation and reconstruction | |
Yu et al. | Scale-aware frequency attention network for super-resolution | |
Zheng et al. | Depth image super-resolution using multi-dictionary sparse representation | |
Wang et al. | Asymmetric dual-direction quasi-recursive network for single hyperspectral image super-resolution | |
CN104182931A (en) | Super resolution method and device | |
Cao et al. | Super-resolution algorithm through neighbor embedding with new feature selection and example training |
Legal Events
Date | Code | Title | Description |
---|---|---|---|
C06 | Publication | ||
PB01 | Publication | ||
C10 | Entry into substantive examination | ||
SE01 | Entry into force of request for substantive examination | ||
C02 | Deemed withdrawal of patent application after publication (patent law 2001) | ||
WD01 | Invention patent application deemed withdrawn after publication |
Application publication date: 20111019 |