CN109658361A - A kind of moving scene super resolution ratio reconstruction method for taking motion estimation error into account - Google Patents
A kind of moving scene super resolution ratio reconstruction method for taking motion estimation error into account Download PDFInfo
- Publication number
- CN109658361A CN109658361A CN201811612588.0A CN201811612588A CN109658361A CN 109658361 A CN109658361 A CN 109658361A CN 201811612588 A CN201811612588 A CN 201811612588A CN 109658361 A CN109658361 A CN 109658361A
- Authority
- CN
- China
- Prior art keywords
- image
- resolution
- super
- motion estimation
- estimation error
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Granted
Links
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T5/00—Image enhancement or restoration
-
- 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
- G06T2207/10—Image acquisition modality
- G06T2207/10016—Video; Image sequence
-
- 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
- G06T2207/20024—Filtering details
- G06T2207/20028—Bilateral filtering
Landscapes
- Physics & Mathematics (AREA)
- General Physics & Mathematics (AREA)
- Engineering & Computer Science (AREA)
- Theoretical Computer Science (AREA)
- Image Analysis (AREA)
Abstract
The present invention provides a kind of moving scene super resolution ratio reconstruction methods for taking motion estimation error into account, including estimation and the initialization of high score image, super-resolution rebuilding model construction, calculating adaptive threshold parameter, computation model parameter, calculating high resolution image, output high resolution image.The present invention is first from regularization Super-resolution reconstruction established model, under the premise of study movement evaluated error is existing, introduce robust iterative, low resolution image contribution amount in reconstruction process of the adaptive reduction containing motion estimation error, to weaken influence of the motion estimation error to reconstructed results；From the angle analysis of robust iterative robustness of the fidelity term based on L1 norm and L2 norm for motion estimation error, it is proposed the bilateral filtering super resolution ratio reconstruction method based on robust iterative, realize steady super-resolution rebuilding in the presence of motion estimation error, and the experimental verification validity of method, therefore can realize high-precision image in moving scene.
Description
Technical field
The present invention relates to the technical fields of moving scene super-resolution rebuilding more particularly to one kind to take motion estimation error into account
Moving scene super resolution ratio reconstruction method.
Background technique
Recently as the continuous development of field of aerospace, satellite video data initially enter the visual field of people, video
Satellite becomes the another hot spot of recent domestic research.Video satellite can shoot the dynamic of atural object on the earth in the form of video
State information, the monitoring suitable for various dynamic objects.But the resolution ratio of current video satellite is not high, generally uses Super-resolution reconstruction
The method built improves video quality, and rebuilding estimation for multiframe oversubscription is a critical issue, and estimation will definitely not be led
It causes the moving object after rebuilding to there is phenomena such as " ring " or " hangover ", influences the resolution ratio of dynamic object, therefore, it is necessary to invent
The good super resolution ratio reconstruction method of a kind of pair of motion estimation error robustness.
Under the premise of method for estimating precision determines, in current oversubscription reconstruction model, improves estimation and miss
The main solution of poor robustness has, 1. directly improve method for estimating improve precision (reference: Baboulaz L,
Dragotti P L.Exact feature extraction using finite rate of innovation
principles with an application to image super-resolution.IEEE Transactions on
Image Processing, 2009,18 (2): 281-298.), but this method complexity is higher；2. estimation and oversubscription are rebuild
(the reference: Shen H F, Zhang L P, Huang B, Li P X.A map approach or joint of Combined estimator method
motion estimation,segmentation,and super resolution.IEEE Transactions on
Image Processing, 2007,16 (2): 479-490.), it solves in iterative process in reconstruction model simultaneously to estimation
As a result it is updated, the advantage with iteration adjustment, but inefficient；3. piecemeal estimation and the method rebuild (reference: Su
H,Tang L,Wu Y,Tretter D,Zhou J.Spatially adaptive block-based super-
Resolution.IEEE Transactions on Image Processing, 2012,21 (3): 1031-1045.), it is first right
High-definition picture plane carries out adaptive piecemeal, using information such as motion match precision, image texture situations as Block Characteristic,
Method using machine learning is the different traditional super-resolution algorithms of different regional choices, and algorithm is complex, moves width
Spend the atural object less effective that differs greatly.In these current Super-resolution reconstruction established models, MAP estimation method (Maximum
A posterior, MAP) (Schultz R R, Stevenson R L.Extraction of high-resolution
frames from video sequences[J].Image Processing,IEEE Transactions on,1996,5
(6): 996-1011.) effect rebuild can be improved by the way that prior-constrained item is added, but this method fidelity term is fixed, it cannot be certainly
The selection fidelity term of adaptation.
The above method is summarized, at present under the premise of containing motion estimation error, general method for reconstructing all has " vibration
Bell " or " hangover " phenomenon, it is sensitive to estimation, although prior-constrained improvement reconstruction quality can be added in MAP estimation, not
Selection fidelity term that can be adaptive according to picture material.Therefore, there is an urgent need to one to be further processed in reconstruction model
The super-resolution rebuilding algorithm of estimation residual error.
Summary of the invention
In view of the above technical problems, it is super that the purpose of the present invention is to provide a kind of moving scenes for taking motion estimation error into account
Resolution reconstruction method, the bilateral filtering super resolution ratio reconstruction method based on robust iterative, realizes and deposits in motion estimation error
Steady super-resolution rebuilding in case.
To achieve the above object, the present invention is a kind of provides the moving scene super-resolution rebuilding for taking motion estimation error into account
Method includes the following steps:
S1: a frame is selected to estimate adjacent interframe with global motion estimating method as reference frame in image sequence
Motion vectorIndicate x, the light stream vector on the direction y, in formulaWith the fortune of obtained adjacent interframe
Dynamic vectorCalculate motion vector of each frame relative to reference frameWherein i represents benchmark frame sequence；F is represented
Estimate the sequence of consecutive frame；R represents the sequence that estimate overall movement vector；With the method for interpolation sports groundWithIt inserts
It is worth in high-resolution grid, and initializes high resolution image with the method for interpolation
S2: BTV is selected to construct Super-resolution reconstruction established model using M- estimation function as fidelity term as regular terms；
S3: adaptive thresholding value parameter a is calculatedk, in order to reduce the influence of estimation residual error and PSF evaluated error,
Adaptive thresholding value parameter a is determined according to the reliability of each framekValue, make the low resolution image for having larger residual error
akIt is smaller, reduce its contribution amount to high resolution image；
S4: computation model parameter Wk,n, Wk,nIt is the diagonal matrix for measuring residual error power；
S5: to the obtained data of step S2, S3 and S4, corresponding high resolution image is found out, according to steepest descent method
Whether the number of iterations n reaches times N, reaches and just terminates；
S6: the number of iterations n reaches corresponding times N output, otherwise continues cycling through；
S7: output high resolution image.
Preferably, the super-resolution rebuilding objective function in step S2, super-resolution rebuilding objective function:
Wherein akIt is adaptive thresholding value parameter, ek,mIt is residual error, λ,It is weight parameter, X is high resolution image,
M is the number of image pixel, and 2*P+1 is the size of one-dimensional two-sided filter core, and P is filter kernel size,Generation respectively
Table horizontal direction translates l pixel, and vertical direction translates h pixel.
Further, a in the step S3kSolution formulaWherein t > 0 controls quadratic function
Decaying,It represents between every frame simulation low resolution image and observation low resolution image
Average residual error, that D is represented is down-sampling matrix, BKThat represent is fuzzy matrix, MKWhat is represented is deformation matrix,It is just
The high resolution image of beginning, YkRepresent low resolution image；What M was represented is the number of image pixel, and t, r are respectively by formulaWithIt obtains, wherein aminTake 0.1, amaxIt takes in all low resolution images most
Big mean residual, i.e. amax=Emax, EminIt is minimum residual error.
Further, it includes: to judge n < N that the number of iterations n, which reaches corresponding times N output, in the step S6, is, then
Return step S3, it is no, then enter step S7；
By upper, the moving scene super resolution ratio reconstruction method for taking motion estimation error into account of the invention is super from regularization first
Resolution reconstruction model sets out, and under the premise of study movement evaluated error is existing, introduces robust iterative, adaptive reduction contains
Contribution amount of the low resolution image of motion estimation error in reconstruction process, to weaken motion estimation error to reconstructed results
Influence；From the angle analysis of robust iterative robust of the fidelity term based on L1 norm and L2 norm for motion estimation error
Property, the bilateral filtering super resolution ratio reconstruction method based on robust iterative is proposed, the feelings existing for motion estimation error are realized
Steady super-resolution rebuilding under condition, and the experimental verification validity of method.The no approximation of the treatment process of the method for the present invention,
And primary operational carries out in the oversubscription of moving target scene is rebuild, therefore can realize high-precision image in moving scene.
Detailed description of the invention
Fig. 1 is the flow chart of the moving scene super resolution ratio reconstruction method for taking motion estimation error into account of the invention；
Fig. 2 is that different regular terms reconstructed results compare figure；
Fig. 3 is the reconstructed results comparison diagram of SkyBox satellite video；
Fig. 4 is the reconstructed results comparison diagram of one number of Jilin.
Specific embodiment
Referring to FIG. 1 to FIG. 4 to the moving scene super-resolution rebuilding side of the present invention for taking motion estimation error into account
Method is described in detail.
As shown in Figure 1, in order to carry out super-resolution rebuilding to image, the present invention takes the moving scene of motion estimation error into account
Super resolution ratio reconstruction method comprises the following steps that
Step S1: selecting a frame as reference frame in image sequence, adjacent to estimate with global motion estimating method
The motion vector of interframeIndicate x, the light stream vector on the direction y, in formulaWith obtained consecutive frame
Between motion vectorCalculate motion vector of each frame relative to reference frameWherein i represents reference frame sequence
Column；F represents the sequence (when estimation consecutive frame) that estimate frame, is relative to the motion vector of first frame；R representative will estimate frame
Sequence (estimation overall movement vector), be exactly that the three of consecutive frame is added；With the method for interpolation sports groundWithInterpolation
High resolution image is initialized onto high-resolution grid, and with the method for interpolation
Step S2: regular terms selects BTV to construct super-resolution using M- estimation function as fidelity term as regular terms
Reconstruction model, super-resolution rebuilding objective function such as following formula:
Wherein M is the number of image pixel, M=N1*N2, Represent r-th of high score shadow
As upper pixel xrThe process of m pixel on to low point of image of kth frame, including move, obscure and down-sampling.X is high-resolution
Rate image, yk,mRepresent m-th of pixel of kth frame observation low resolution image.Actually ek,mIndicate simulation low resolution image
With the residual error of low point of image of observation.akIt is adaptive thresholding value parameter, λ,It is weight parameter, 2*P+1 is one-dimensional bilateral filter
The size of wave device core, P filter kernel size,It respectively represents horizontal direction and translates l pixel, vertical direction translates h
Pixel.
Step S3: adaptive thresholding value parameter is calculatedThe decaying of quadratic function is controlled,It represents average residual between every frame simulation low resolution image and observation low resolution image
Remaining error, that D is represented is down-sampling matrix, size N1N2×r2N1N2, BKWhat is represented is fuzzy matrix, and size is
r2N1N2×r2N1N2；MKThat represent is deformation matrix, size r2N1N2×r2N1N2；It is initial high resolution image,
Generally obtained by reference low resolution image interpolation；YkRepresent low resolution image；That M is represented is the number of image pixel, M=
N1×N2.T, r is respectively by formulaWithIt obtains, wherein aminTake 0.1, amaxTake institute
There are maximum mean residual, i.e. a in low resolution imagemax=Emax, EminIt is minimum residual error, generally takes 10-6.In order to reduce
The influence of estimation residual error and PSF evaluated error determines adaptive thresholding value parameter a according to the reliability of each framek
Value, make a for the low resolution image for having larger residual errorkIt is smaller, reduce its contribution amount to high resolution image.
Step S4: iterative solution formula to the end is obtained according to steepest descent method formula:Wherein β
For iteration step length, computation model parameter Wk,n, Wk,nIt is the diagonal matrix for measuring residual error power, passes throughIt calculates,It is i-th of element of residual vector, calculation formula:
Step S5: to the obtained data of step S2, S3 and S4, corresponding high resolution image is found out, according under steepest
Whether drop method the number of iterations n reaches times N, reaches and just terminates.
Step S6: the number of iterations n reaches corresponding times N output: judging n < N, is, then return step S3；It is no, then into
Enter step S7.
Step S7: output high resolution image.
As shown in Fig. 2, this experiment is the performance of more different regular terms, using the steady fidelity term of the present invention, use respectively
Tivhonov, TV, BTV rebuild low-resolution video sequence (such as Fig. 2 (a)), there are moulds in original image as regular terms
Paste, noise situations.Fig. 2 (b) and (c) are Tivhonov and TV regular terms reconstructed results respectively, the inhibition situation of noise in image
Generally, some are fuzzy for writing.Fig. 2 (d) is BTV regular terms reconstructed results, and noise is inhibited, and edge is kept preferably, and details is clear
It is clear.
As shown in figure 3, dividing using BTV as bound term for verifying the method for the present invention the Lu Bangxing of motion estimation error
Not Cai Yong L1 norm, L2 norm, robust iterative of the present invention as fidelity term (abbreviation method 1, method 2, the method for the present invention), altogether into
The satellite video data of SkyBox and Jilin No.1 are respectively adopted in two groups of comparative experimentss of row, experimental data, and Experimental Area is respectively
Caliph tower area and Mexico Du Lange, it is 5 frames that frame number is rebuild in experiment, and rebuilding multiple is 2 times.It is big in image data sequence
Part atural object is static background atural object, and dynamic atural object is the aircraft of movement, vehicle.Overall motion estimation side is used in experiment
Method, this method are preferable to static atural object estimated result, but when clapping has moving object in scene, the movement ginseng of moving object
Number evaluated error is very big, can clearly react algorithm for reconstructing to the susceptibility of kinematic error.
Fig. 3 (a) is the raw video of SkyBox satellite video, static background atural object and is quickly moved winged in Experimental Area
Machine motion state is different, and the ratio very little that aircraft occupies in entire image, can by the principle of overall motion estimation
To know, the estimation of kinematic parameter is based on the pixel of background parts, so static atural object estimation is relatively accurate, the movement of aircraft
Estimation be it is inaccurate, i.e., kinematic parameter error is mainly reflected in aircraft portions, and this fractional error also will have a direct impact on
The result of method for reconstructing.The image and details (such as Fig. 3) of observation reconstruction front and back can be seen that the static atural object after rebuilding in image
Details is more richer than original image, and the edge contour in river and building in atural object is all more clear than original image.But it is transported in image
Dynamic aircraft is widely different after rebuilding, and the result of method 2 such as Fig. 3 (b) and (f), static atural object reconstructed results are fine, sport plane
Smudgy, the form of aircraft obscures visible.The result of method 1 is better than method 2, but aircraft weight in the movement in the method for the present invention
It is apparent to build rear profile edge details, fuselage interior Pixel Information is more evenly consistent, and Global Information is kept in original image just
The consistency of body movement, and increase local message.Therefore, description of test the method for the present invention is unwise to action reference variable error
Sense, it is preferable to rebuild statically and dynamically object in the case where certain movement error can be contained in kinematic matrix.
As shown in figure 4, Fig. 4 is the experimental result of one number of Jilin, it mainly include static atural object and fortune in group experiment
Vehicle in dynamic.Since vehicle and static building etc. are there are the difference of motion state, motion estimation error is mainly reflected in
Vehicle sections.The image of front and back is rebuild in comparison it can be found that the static background information after several method is rebuild is all preferable, such as Fig. 4
(e) the corresponding building information of big frame portion point in, edge and internal information are all more clear than original image, but the reconstructed results of moving vehicle
It differs greatly.Vehicle trailing phenomenon after rebuilding such as method 2 (small frame portion point corresponding (b) and (f) in Fig. 4 (e)) is serious, both
See that vehicle is longer than original image on the whole, in detail view the Pixel Information of vehicle interior there are the alternate grid phenomenon of monochrome pixels, this
It is to cause algorithm calculating also to generate deviation since there are errors for vehicle sections estimation, make information of vehicles in length and inside
Pixel portion generates mistake.Method 1 (such as Fig. 4 (c) and (g)) although in vehicle do not trail, vehicle and road junction
Pixel Information distortion, have black outline border (such as Fig. 4 (g)), vehicle internal information the phenomenon that there is also black and white grid.Side of the present invention
Vehicle after method (such as Fig. 4 (d)) reconstruction, not only maintains original vehicle information, but also profile and internal information are all apparent, no
In the presence of hangover and black grid phenomenon, detailed information is saved more intact.
Claims (4)
1. a kind of moving scene super resolution ratio reconstruction method for taking motion estimation error into account, which comprises the steps of:
S1: a frame is selected to estimate the movement of adjacent interframe with global motion estimating method as reference frame in image sequence
VectorIndicate x, the light stream vector on the direction y, in formulaIt is sweared with the movement of obtained adjacent interframe
AmountCalculate motion vector of each frame relative to reference frameWherein i represents benchmark frame sequence；F representative will be estimated
Count the sequence of consecutive frame；R represents the sequence that estimate overall movement vector；With the method for interpolation sports groundWithIt is interpolated into
In high-resolution grid, and high resolution image is initialized with the method for interpolation
S2: BTV is selected to construct Super-resolution reconstruction established model using M- estimation function as fidelity term as regular terms；
S3: adaptive thresholding value parameter a is calculatedk；
S4: computation model parameter Wk,n, Wk,nIt is the diagonal matrix for measuring residual error power；
S5: to the obtained data of step S2, S3 and S4, corresponding high resolution image is found out, according to steepest descent method iteration
Whether frequency n reaches times N, reaches and just terminates；
S6: the number of iterations n reaches corresponding times N output, otherwise continues cycling through；
S7: output high resolution image.
2. the moving scene super resolution ratio reconstruction method according to claim 1 for taking motion estimation error into account, feature exist
In super-resolution rebuilding objective function in the step S2:
Wherein akIt is adaptive thresholding value parameter, ek,mIt is residual error, λ,It is weight parameter, X is high resolution image, and M is
The number of image pixel, 2*P+1 are the sizes of one-dimensional two-sided filter core, and P is filter kernel size, Respectively represent water
Square to translation l pixel, vertical direction translation h pixel.
3. the moving scene super resolution ratio reconstruction method according to claim 1 for taking motion estimation error into account, feature exist
In a in the step S3kSolution formulaWherein t > 0 controls the decaying of quadratic function,It represents average residual between every frame simulation low resolution image and observation low resolution image
Remaining error, that D is represented is down-sampling matrix, BKThat represent is fuzzy matrix, MKWhat is represented is deformation matrix,It is initial height
Resolution image, YkRepresent low resolution image；What M was represented is the number of image pixel, and t, r are respectively by formulaWithIt obtains, wherein aminTake 0.1, amaxIt takes in all low resolution images most
Big mean residual, i.e. amax=Emax, EminIt is minimum residual error.
4. the moving scene super resolution ratio reconstruction method according to claim 1 for taking motion estimation error into account, feature exist
In it includes: to judge n < N that the number of iterations n, which reaches corresponding times N output, in the step S6, is, then return step S3；It is no,
Then enter step S7.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201811612588.0A CN109658361B (en) | 2018-12-27 | 2018-12-27 | Motion scene super-resolution reconstruction method considering motion estimation errors |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201811612588.0A CN109658361B (en) | 2018-12-27 | 2018-12-27 | Motion scene super-resolution reconstruction method considering motion estimation errors |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN109658361A true CN109658361A (en) | 2019-04-19 |
| CN109658361B CN109658361B (en) | 2022-12-06 |
Family
ID=66117710
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201811612588.0A Active CN109658361B (en) | 2018-12-27 | 2018-12-27 | Motion scene super-resolution reconstruction method considering motion estimation errors |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN109658361B (en) |
Cited By (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN110211193A (en) * | 2019-05-17 | 2019-09-06 | 山东财经大学 | Three dimensional CT interlayer image interpolation reparation and super-resolution processing method and device |
| CN110263699A (en) * | 2019-06-17 | 2019-09-20 | 睿魔智能科技（深圳）有限公司 | Method of video image processing, device, equipment and storage medium |
| CN112184549A (en) * | 2020-09-14 | 2021-01-05 | 阿坝师范学院 | Super-resolution image reconstruction method based on space-time transformation technology |
| CN112270697A (en) * | 2020-10-13 | 2021-01-26 | 清华大学 | Satellite sequence image moving target detection method combined with super-resolution reconstruction |
Citations (7)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN102122387A (en) * | 2010-12-18 | 2011-07-13 | 浙江大学 | Super-resolution image reconstruction method for robust |
| CN102194222A (en) * | 2011-04-26 | 2011-09-21 | 浙江大学 | Image reconstruction method based on combination of motion estimation and super-resolution reconstruction |
| CN102236889A (en) * | 2010-05-18 | 2011-11-09 | 王洪剑 | Super-resolution reconfiguration method based on multiframe motion estimation and merging |
| US20120300122A1 (en) * | 2011-05-26 | 2012-11-29 | Microsoft Corporation | Adaptive super resolution for video enhancement |
| KR20130077646A (en) * | 2011-12-29 | 2013-07-09 | 광운대학교 산학협력단 | A super-resolution method by motion estimation with sub-pixel accuracy using 6-tap fir filter |
| WO2015180053A1 (en) * | 2014-05-28 | 2015-12-03 | 北京大学深圳研究生院 | Method and apparatus for rapidly reconstructing super-resolution image |
| CN106056540A (en) * | 2016-07-08 | 2016-10-26 | 北京邮电大学 | Video time-space super-resolution reconstruction method based on robust optical flow and Zernike invariant moment |
-
2018
- 2018-12-27 CN CN201811612588.0A patent/CN109658361B/en active Active
Patent Citations (7)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN102236889A (en) * | 2010-05-18 | 2011-11-09 | 王洪剑 | Super-resolution reconfiguration method based on multiframe motion estimation and merging |
| CN102122387A (en) * | 2010-12-18 | 2011-07-13 | 浙江大学 | Super-resolution image reconstruction method for robust |
| CN102194222A (en) * | 2011-04-26 | 2011-09-21 | 浙江大学 | Image reconstruction method based on combination of motion estimation and super-resolution reconstruction |
| US20120300122A1 (en) * | 2011-05-26 | 2012-11-29 | Microsoft Corporation | Adaptive super resolution for video enhancement |
| KR20130077646A (en) * | 2011-12-29 | 2013-07-09 | 광운대학교 산학협력단 | A super-resolution method by motion estimation with sub-pixel accuracy using 6-tap fir filter |
| WO2015180053A1 (en) * | 2014-05-28 | 2015-12-03 | 北京大学深圳研究生院 | Method and apparatus for rapidly reconstructing super-resolution image |
| CN106056540A (en) * | 2016-07-08 | 2016-10-26 | 北京邮电大学 | Video time-space super-resolution reconstruction method based on robust optical flow and Zernike invariant moment |
Non-Patent Citations (4)
| Title |
|---|
| 刘淼，曹汉强，李旭涛等: ""基于运动估计误差和边缘约束的超分辨率重构"", 《中国图象图形学报》 * |
| 卜丽静，郑新杰，张正鹏: ""吉林一号卫星视频影像超分辨率重建"", 《国土资源遥感》 * |
| 王伯阳,韩晓,张文奇: ""图像超分辨率重建算法比较研究"", 《沈阳大学学报》 * |
| 韩玉兵，陈小蔷，吴乐南: ""一种视频序列的超分辨率重建算法"", 《电子学报》 * |
Cited By (7)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN110211193A (en) * | 2019-05-17 | 2019-09-06 | 山东财经大学 | Three dimensional CT interlayer image interpolation reparation and super-resolution processing method and device |
| CN110263699A (en) * | 2019-06-17 | 2019-09-20 | 睿魔智能科技（深圳）有限公司 | Method of video image processing, device, equipment and storage medium |
| CN110263699B (en) * | 2019-06-17 | 2021-10-22 | 睿魔智能科技（深圳）有限公司 | Video image processing method, device, equipment and storage medium |
| CN112184549A (en) * | 2020-09-14 | 2021-01-05 | 阿坝师范学院 | Super-resolution image reconstruction method based on space-time transformation technology |
| CN112184549B (en) * | 2020-09-14 | 2023-06-23 | 阿坝师范学院 | Super-resolution image reconstruction method based on space-time transformation technology |
| CN112270697A (en) * | 2020-10-13 | 2021-01-26 | 清华大学 | Satellite sequence image moving target detection method combined with super-resolution reconstruction |
| CN112270697B (en) * | 2020-10-13 | 2022-11-18 | 清华大学 | Satellite sequence image moving target detection method combined with super-resolution reconstruction |
Also Published As
| Publication number | Publication date |
|---|---|
| CN109658361B (en) | 2022-12-06 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| Ancuti et al. | NTIRE 2021 nonhomogeneous dehazing challenge report | |
| CN107025632B (en) | Image super-resolution reconstruction method and system | |
| CN109658361A (en) | A kind of moving scene super resolution ratio reconstruction method for taking motion estimation error into account | |
| CN102360498B (en) | Reconstruction method for image super-resolution | |
| CN107274347A (en) | A kind of video super-resolution method for reconstructing based on depth residual error network | |
| CN109993095B (en) | Frame level feature aggregation method for video target detection | |
| CN105809712B (en) | A kind of efficient big displacement light stream method of estimation | |
| CN101976435B (en) | Combination learning super-resolution method based on dual constraint | |
| CN106408524A (en) | Two-dimensional image-assisted depth image enhancement method | |
| CN103034982B (en) | Image super-resolution rebuilding method based on variable focal length video sequence | |
| CN103514608B (en) | Moving object detection based on movement attention fusion model and extracting method | |
| CN104867111A (en) | Block-blur-kernel-set-based heterogeneous video blind deblurring method | |
| CN112837245A (en) | Dynamic scene deblurring method based on multi-mode fusion | |
| CN106780338A (en) | Based on anisotropic quick super-resolution method for reconstructing | |
| Su et al. | Super-resolution without dense flow | |
| CN103971354A (en) | Method for reconstructing low-resolution infrared image into high-resolution infrared image | |
| CN114463237A (en) | Real-time video rain removing method based on global motion compensation and inter-frame time domain correlation | |
| CN102222321A (en) | Blind reconstruction method for video sequence | |
| Ye et al. | SNR-Prior Guided Trajectory-Aware Transformer for Low-Light Video Enhancement | |
| Xia et al. | SVFI: spiking-based video frame interpolation for high-speed motion | |
| Liu et al. | Fast Generalizable Gaussian Splatting Reconstruction from Multi-View Stereo | |
| CN117689541A (en) | Multi-region classification video super-resolution reconstruction method with optimized time redundancy | |
| CN108399629B (en) | Image array optical flow estimation method of artificial compound eye camera | |
| CN116523743A (en) | Game super-resolution method based on cyclic neural network | |
| Amiri et al. | A fast video super resolution for facial image |
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 |