CN109255822A - A kind of multiple dimensioned resolution compression perception method for reconstructing between encoding the time-out multiple constraint - Google Patents
A kind of multiple dimensioned resolution compression perception method for reconstructing between encoding the time-out multiple constraint Download PDFInfo
- Publication number
- CN109255822A CN109255822A CN201810767498.2A CN201810767498A CN109255822A CN 109255822 A CN109255822 A CN 109255822A CN 201810767498 A CN201810767498 A CN 201810767498A CN 109255822 A CN109255822 A CN 109255822A
- Authority
- CN
- China
- Prior art keywords
- matrix
- algorithm
- image
- reconstructing
- resolution
- 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
- G06T9/00—Image coding
Abstract
The invention discloses a kind of times constrained based on multiple dimensioned code aperture and multiple canonical to compress method for reconstructing, using the kinetic characteristic of target scene as fusion foundation, the method rebuild using multiple dimensioned code aperture and multiple constraint canonical realizes that the super temporal resolution of the compressed sensing video sequence image of fast motion scenes is restored.This method can guarantee the clarity of target sport foreground and static background simultaneously, effectively improve the sport efficiency of algorithm for reconstructing；Algorithm uses multiple dimensioned observing matrix to realize the coding twice in aperture according to the sparse characteristic of target scene, and the quick reconstruction of CACTI may be implemented；It is constructed using the transform domain sparse characteristic of target scene as priori knowledge and rebuilds canonical bound term, the problem of more canonicals constrain is solved using ADMM algorithm, compared to existing algorithm for reconstructing, the algorithm has stronger robustness to noise and motion blur, and the algorithm reconstruction effect that video compress perceives under the conditions of strong noise or high frame frequency can be improved.
Description
Technical field
The present invention relates to light field modulation and calculate technical field of imaging, and in particular to one kind based on multiple dimensioned code aperture and
The time of multiple canonical constraint compresses method for reconstructing.
Background technique
Tracking imaging system towards high-speed moving object is limited to the pixel dimension and reading circuit of sensor, easily occurs
Due to the time for exposure it is too long caused by motion blur or camera frame frequency it is too low caused by time lack sampling, it is mixed so as to cause movement
It is folded.The CACTI time compresses aperture coded imaging method (CACTI, Coded Aperture Compressive Temporal
Imager proposition) effectively raises the capture ability of high-speed moving object, has provided for the record analysis of critical event
Effect means, this method have become an important research direction in compressed sensing field.
CACTI is the fast imaging mode based on aperture coding principle, and the process employs an active encoder elements pair
Time carries out mechanical transformation, by the compression image of multiple low spatial and temporal resolutions of acquisition, carries out by broad sense alternating projection method
It rebuilds, 8~148 times of temporal resolution of video can be improved.With Hybrid camera, camera array, coding exposure image and DLP
Technologies such as (Digital Light Processing) are compared, and the technology is in the performance indicator and integral for not changing image device
Between under the premise of, without high-precision and high speed spectrum assignment, so that it may the image sequence for obtaining high frame frequency, breach it is existing at
Limitation as device to optical system imaging performance.
The imaging performance of CACTI is influenced significant, and the degree of rarefication of restructing algorithm and observing matrix, target by restructing algorithm
It is closely related with compression ratio.Existing restructing algorithm mainly includes three classes:
The first kind, the restructing algorithm based on transform domain, typical algorithm include GAP and its innovatory algorithm.The algorithm, which utilizes, to be added
Weigh l2,1Norm substitutes l1Norm solves convex optimization problem, in the case where meeting the equidistant condition of RIP, can converge to ideal solution,
Compressed sensing suitable for natural image and video.The algorithm has fast convergence rate, operation efficiency height, adaptable spy
Point.
The shortcomings that GAP algorithm, is, domain of variation and weighting l are used in algorithm2,1Norm increases the complexity of calculating,
Simultaneously for the image of different degree of rarefications need that suitable weighting coefficient and domain of variation is selected to can be only achieved ideal effect.
Second class, the restructing algorithm based on reconstruction, typical algorithm such as OMP and TV algorithm.Wherein the operation of OMP algorithm algorithm is imitated
Rate is lower；TV algorithm passes through TV regular terms replacement weighting l2,1Norm constraint item can achieve the performance close to GAP algorithm, simplify
The operation of algorithm.But the information of single-frame images is utilized in the algorithm, is suitable for the lower natural scene of target sparse degree, it is right
In the application such as document recording and target scene monitoring, it still needs further improvement for algorithm performance.
Third class, the restructing algorithm based on study, typical algorithm is as being based on GMM (Gaussian Mixture Model)
Inversion algorithm.The algorithm has centainly in the way of online or offline dictionary learning for particular video sequence quality reconstruction
It is promoted.Wherein the GMM algorithm of online updating has robustness, especially simple fortune for different video and Target Motion Character
Emotionally condition, quality reconstruction are promoted obvious.For quick, complicated motion conditions, the GMM algorithm updated offline is more practical.But
It is that the GAP algorithm that it compares is not optimized for special scenes, especially weighting coefficient uses default value, therefore, needs
Algorithms selection is carried out according to actual use scene.In addition, the GMM algorithm arithmetic speed using parallel computation is better than GAP algorithm,
But for serial computing occasion, speed is still slower than GAP algorithm.
Comprehensive analysis, CACTI imaging method has the following problems at present: first, due to using motion encoded aperture mode,
System needs to increase high resolution mask or spatial light modulator, reduces the luminous flux of imaging system, so that imaging results pair
Noise is more sensitive.Second, existing algorithm restructuring procedure mostly utilizes frame information, and the redundancy of interframe is not made full use of to believe
Breath.
Summary of the invention
In view of this, the present invention provides a kind of times constrained based on multiple dimensioned code aperture and multiple canonical to compress weight
Construction method realizes time super-resolution using transform domain and the blending algorithm of TV canonical using the kinetic characteristic of target scene as foundation
Rate compressed sensing and reconstruction, algorithm can improve the clarity of moving target and static background simultaneously, and have to noise relatively strong
Robustness, suitable for the video compress of targeted surveillance and document recording equipment perceive.
It is a kind of time-out between resolution compression perceive method for reconstructing, include the following steps:
Step 1, low resolution encoder matrix is up-sampled, obtains high resolution matrix；By the high resolution matrix
It is multiplied with same scale random matrix, obtains multiple dimensioned encoder matrix；CACTI imaging system is based on the multiple dimensioned encoder matrix and obtains
Obtain the coding aliased image of moving scene；
Step 2, coding aliased image is perceived based on the video compress of TV canonical using the low resolution matrix in step 1
Image reconstruction is carried out, reconstructed image sequence is obtained:
Step 3, reconstructed image step 2 obtained up-samples, and obtains the initial estimation of reconstructed image sequence, is based on
The initial estimation of the reconstructed image sequence seeks interframe movement vector；
Step 4, using interframe movement vector, the video compress sensing reconstructing based on multiple constraint is carried out to aliased image,
Wherein, using ADMM algorithm reconstructed image, that is, the restricted problem of following formula is solved, reconstructed image x to be solved is obtained:
Wherein, y is aliased image, and Φ is multiple dimensioned encoder matrix, and W is interframe movement vector, and x is original image to be estimated
Sequence, x are single-frame images to be estimated, x(t)Indicate the t times iterative estimate result of image sequence；ω is the corresponding system of transform domain
Number,Indicate l2,1The Weighted Group of norm, is defined asN is that single observes the restructural figure of aliased image
Frame number as sequence comprising image, k indicate the kth frame in this sequence；θ indicates the division variable in alternately Multiplier Algorithm, herein
Indicate the interframe movement vector obtained after interframe is updated with x iteration；θ1It indicates between current image to be estimated and a later frame image
Motion vector θkIndicate the motion vector between kth frame and k+1 frame； Table respectively
Show gradient vertically and horizontally.
Preferably, low resolution matrix is using local hadamard matrix or block circulant matrix.
Preferably, random matrix can choose gaussian random matrix or Bernoulli Jacob's matrix.
Preferably, being integral multiple relation, the scale of high resolution matrix between low resolution matrix and high resolution matrix
It is 4-32 times of low resolution matrix.
Preferably, obtaining interframe movement vector method in the step 3 are as follows:
By its SIFT value of the reconstructed image sequence node-by-node algorithm of initial estimation, SIFT density image f is obtainedi, interframe movement
Vector w is indicated are as follows:
Wherein, fiAnd fi+1Indicate the SIFT density image of consecutive frame；W (p) indicates the motion vector at p point；T ' is truncation
Threshold value, for accelerating operation；su,svRespectively indicate SIFT stream, the wavelet coefficient under Haar wavelet transform base；λ1、λ2Indicate weighting system
Number；U (p) and v (p) respectively indicates the vertically and horizontally component of motion vector w (p)；α, d respectively indicate truncation funcation weighting coefficient
And threshold value.
Preferably, the multiple constraint weight coefficient is obtained using following formula:
Wherein, u, v are respectively the vertical and horizontal movement component of current pixel point, and T is the threshold value of motion vector；η1,η2Point
It Wei not transform domain in multiple constraintWith | | TV (x) | | weight coefficient, γ be normalization operator, make
Preferably, being solved using alternative projection algorithm to the restricted problem in step 4.
The invention has the following beneficial effects:
(1) time compressed sensing method for reconstructing proposed by the present invention, using the kinetic characteristic of target scene as fusion according to
According to the method rebuild using multiple dimensioned code aperture and multiple constraint canonical realizes the compressed sensing video sequence of fast motion scenes
The super temporal resolution of column image is restored.This method can guarantee the clarity of target sport foreground and static background simultaneously, have
Effect improves the sport efficiency of algorithm for reconstructing.
(2) present invention proposes a kind of multiple dimensioned code aperture method, and algorithm uses more according to the sparse characteristic of target scene
The observing matrix of scale realizes the coding twice in aperture, and the quick reconstruction of CACTI may be implemented.
(3) present invention proposes a kind of more regularization methods, is known using the transform domain sparse characteristic of target scene as priori
Know building and rebuild canonical bound term, solves the problem of more canonicals constrain using ADMM algorithm, compare existing algorithm for reconstructing, the calculation
Method has stronger robustness to noise and motion blur, and algorithm video compress sense under the conditions of strong noise or high frame frequency can be improved
The reconstruction effect known.
Detailed description of the invention
Fig. 1 is the multiple constraint time compressed sensing method for reconstructing flow chart based on kinetic characteristic.
Specific embodiment
The present invention will now be described in detail with reference to the accompanying drawings and examples.
The present invention provides a kind of time compressed sensing reconstruction sides constrained based on multiple dimensioned code aperture and multiple canonical
Method can realize the rapid time super-resolution rebuilding of video compress perception under the conditions of strong noise and high compression rate, rebuild knot
Fruit takes into account the clarity and texture information of moving target and static background.
The basic principle of this method is the Super-resolution reconstruction constrained based on multiple dimensioned code aperture compressed sensing and more canonicals
Construction method is utilized using different transform domains and full variational methods to the recovery effect difference of the target scene of different motion characteristic
Sparse characteristic constructs multiple canonical bound terms as priori knowledge, combines and reconstructs optimal result.The committed step of algorithm: first
First, it combines to form multiple dimensioned observing matrix using low resolution and high resolution observations matrix, control DMD is realized to moving scene
Aperture coding；Low-resolution video sequence is quickly reconstructed by improved ADMM+TV algorithm；Secondly, utilizing SIFT Flow
Algorithm seeks interframe movement vector sum multiple constraint weighted value；Finally, going out the optimal of video sequence using more canonical constraint reestablishings
Estimation, the imaging system images specific steps that the multiple constraint based on kinetic characteristic is rebuild:
Step 1, low resolution aperture encoder matrix is up-sampled, obtains high resolution matrix；By the high-resolution
Matrix is multiplied with same scale random matrix, obtains multiple dimensioned encoder matrix；Wherein, local hada can be used in low resolution matrix
Ma matrix or block circulant matrix；Random matrix can choose gaussian random matrix or Bernoulli Jacob's matrix；Low resolution aperture encodes square
Battle array refers to the low aperture encoder matrix of the scale of the aperture encoder matrix generally used than CACTI imaging system；For the ease of
It realizes, there are integral multiple relations between low resolution matrix and high resolution matrix, usually choose 4-32 times；CACTI imaging system
System obtains the coding aliased image of moving scene using multiple dimensioned encoder matrix, specifically: multiple dimensioned encoder matrix is by low resolution
Rate matrix and the matrix multiple of high resolution matrix obtain, and the element of middle high-resolution encoder matrix corresponds to encoder element most
Small physical size, the element of low resolution matrix are the integral multiple of high resolution matrix element.The original that CACTI is compressed using coding
Reason realizes the compression measurement of video sequence using active coding unit in aperture location.Encoder element is placed at aperture,
Coding mode is controlled by multiple dimensioned encoder matrix.The above process is expressed as linear matrix form:
Y=Φ x+n (1)
Wherein n is system imaging noise, and Φ is the transmission function of time-varying, expression formula are as follows:
Step 2, low resolution matrix helps to reduce the dimension of matrix operation, improves operation efficiency, in order to obtain movement
Motion information between scene realizes High precision reconstruction as prior-constrained knowledge, while reducing the operation time of algorithm, therefore
Figure is carried out using the video compress perception based on TV canonical to coding aliased image first with the low resolution matrix in step 1
As reconstruct:
Aliased image y can be encoded by a frame according to imaging model and reconstruct NFFrame image, restructing algorithm need to solve public
Formula (1) is solved using the TV regularization algorithm of ADMM,
Wherein, x indicates image to be solved；x(t)Indicate the image x to be solved that the t times iteration obtains；
When given intermediate quantity z, solution are as follows:
x(t+1)=z(t)+ΦT(ΦΦT)-1(y-Φz(t)) (4)
When given x, noise is taken into account into solution procedure, updating z indicates are as follows:
ρ(t+1):=ρ(t)+(θ(t+1)-b(t+1)) (7)
Wherein, Respectively indicate gradient vertically and horizontally.
Step 3, the estimation of interframe movement vector is calculated with multiple constraint weight coefficient:
For the ease of matrix operation, keep motion vector matrix consistent with super-resolution reconstruction image array scale, by step 2
Obtained sequence of low resolution pictures is up-sampled, and is obtained the initial estimation of high-definition picture sequence, is sought interframe movement
Vector.
Wherein, the method for interframe movement vector is sought are as follows: by its SIFT of the reconstructed image sequence node-by-node algorithm of initial estimation
Value, obtains SIFT density image fi, interframe movement vector w can indicate are as follows:
Wherein, fiAnd fi+1Indicate the SIFT density image of consecutive frame；W (p) indicates the motion vector at p point；T ' is truncation
Threshold value, for accelerating operation；su,svRespectively indicate SIFT stream, the wavelet coefficient under Haar wavelet transform base；λ1、λ2Indicate weighting system
Number；U (p) and v (p) respectively indicates the vertically and horizontally component of motion vector w (p)；α, d respectively indicate truncation funcation weighting coefficient
And threshold value.
Traditional SIFT stream calculation formula includes SIFT descriptor shape constancy bound term, moves vertically and horizontally the one of component
Rank norm constraint item and Movement consistency bound term, due to including l in SIFT stream calculation cost function1The moving displacement of norm
, so search range is restricted, when leading to larger displacement, there are large errors.Small echo sparse constraint proposed by the present invention
SIFT flows algorithm for estimating and compound movement scene may be implemented using wavelet transformed domain sparsity and rotation and translation invariance
Inter frame motion estimation.
Since SIFT descriptor is based on the pixel in pixel and its neighborhood system, carried out using low resolution matrix
The estimation of initial pictures sequence, and estimate that the result of interframe movement vector can really reflect velocity field variation tendency.
Multiple constraint weight coefficient is obtained using following formula:
Wherein, η1,η2Respectively transform domain in multiple constraintWith | | TV (x) | | weight coefficient, u, v be respectively work as
The vertical and horizontal movement component of preceding pixel point, T are the threshold value of motion vector；γ is normalization operator, is made
Step 4, using interframe movement vector, the video compress sensing reconstructing based on multiple constraint is carried out to aliased image:
Sparse optimization algorithm based on alternating direction multipliers method (ADMM) has very the minimum model for solving belt restraining
Big advantage, its essence of the algorithm for reconstructing based on ADMM+TV are still to rely on the TV canonical algorithm of present frame, static state in reconstructed results
The reconstruction effect of background is poor；Although using Weighted Convex cluster based on GAP+Wavelet/DCT algorithm, single observation is utilized
Reconstructed frame between redundancy, but its reconstructed results need further to be promoted.Therefore present invention comprehensive utilization image
The Weighted Convex cluster of transform domain and full variation are as Prior Knowledge Constraints restructuring procedure, it is ensured that the clarity of entire visual field,
Improve reconstruction accuracy.
Using ADMM algorithm reconstructed image, it is equivalent to following restricted problem:
Wherein, y is aliased image, and Φ is multiple dimensioned encoder matrix, and W is interframe movement vector, and x is original image to be estimated
Sequence, x are single-frame images to be estimated, x(t)Indicate the t times iterative estimate result of image sequence；ω is the corresponding system of transform domain
Number,Indicate l2,1The Weighted Group of norm, is defined asN is that single observes the restructural figure of aliased image
Frame number as sequence comprising image, k indicate the kth frame in this sequence；θ indicates the division variable in alternately Multiplier Algorithm, herein
Indicate the interframe movement vector obtained after interframe is updated with x iteration；θ1It indicates between current image to be estimated and a later frame image
Motion vector θkIndicate the motion vector between kth frame and k+1 frame； Table respectively
Show gradient vertically and horizontally.
Formula (11) is solved using alternative projection algorithm:
Wherein λi, i=1,2,3 is regular coefficient；T is the number of iterations；M is that square is converted in the corresponding space of interframe movement vector
Battle array.Above formula can update θ and x using alternating iteration method.The super-resolution rebuilding of entire video sequence is considered as one by above formula
Optimization problem carries out joint solution.θi (t)Indicate the t times iteration result of the motion vector between t frame and t+1 frame.
Image quality evaluation: the image quality of reconstructed image sequence in step 4 is calculated, using subjective assessment and objectively evaluates index
The mode combined objectively evaluates index using fusion rules index.
Wherein, piIndicate the degree or probability of each image distortion；qiIndicate the evaluation score of corresponding different evaluation index,
Including similarity SSIM, signal-to-noise ratio PSNR, reconstructed error MSE and blind evaluation index BIQI；λiFor normalization coefficient, for SSIM
Its value is positive with PSNR, and for MSE and BIQI, its value is negative；S indicates time compression ratio,N indicates evaluation index number
Amount.If the requirements are not met, resets kinetic characteristic discrimination threshold, return step 3.If satisfied the use demand, export
Reconstruction result.
In conclusion the above is merely preferred embodiments of the present invention, being not intended to limit the scope of the present invention.
All within the spirits and principles of the present invention, any modification, equivalent replacement, improvement and so on should be included in of the invention
Within protection scope.
Claims (7)
1. resolution compression perceives method for reconstructing between a kind of time-out, which comprises the steps of:
Step 1, low resolution encoder matrix is up-sampled, obtains high resolution matrix；By the high resolution matrix and together
Equal scales random matrix is multiplied, and obtains multiple dimensioned encoder matrix；CACTI imaging system is based on the multiple dimensioned encoder matrix and is transported
The coding aliased image of dynamic scene；
Step 2, coding aliased image is perceived using the low resolution matrix in step 1 based on the video compress of TV canonical and is carried out
Image reconstruction obtains reconstructed image sequence:
Step 3, reconstructed image step 2 obtained up-samples, and obtains the initial estimation of reconstructed image sequence, heavy based on this
The initial estimation of structure image sequence seeks interframe movement vector；
Step 4, using interframe movement vector, the video compress sensing reconstructing based on multiple constraint is carried out to aliased image, wherein
Using ADMM algorithm reconstructed image, that is, the restricted problem of following formula is solved, reconstructed image x to be solved is obtained:
Wherein, y is aliased image, and Φ is multiple dimensioned encoder matrix, and W is interframe movement vector, and x is original image sequence to be estimated
Column, x are single-frame images to be estimated, x(t)Indicate the t times iterative estimate result of image sequence；ω is the corresponding coefficient of transform domain,Indicate l2,1The Weighted Group of norm, is defined asN is that single observes the restructural image of aliased image
Sequence includes the frame number of image, and k indicates the kth frame in this sequence；θ indicates the division variable in alternately Multiplier Algorithm, herein table
Show the interframe movement vector obtained after interframe is updated with x iteration；θ1Indicate the fortune between current image to be estimated and a later frame image
Dynamic vector θkIndicate the motion vector between kth frame and k+1 frame； It respectively indicates
Gradient vertically and horizontally.
2. resolution compression perceives method for reconstructing between a kind of time-out as described in claim 1, which is characterized in that wherein, low point
Resolution matrix is using local hadamard matrix or block circulant matrix.
3. resolution compression perceives method for reconstructing between a kind of time-out as claimed in claim 1 or 2, which is characterized in that random square
Battle array can choose gaussian random matrix or Bernoulli Jacob's matrix.
4. resolution compression perceives method for reconstructing between a kind of time-out as claimed in claim 1 or 2, which is characterized in that low resolution
It is integral multiple relation between rate matrix and high resolution matrix, the scale of high resolution matrix is the 4-32 of low resolution matrix
Times.
5. resolution compression perceives method for reconstructing between a kind of time-out as described in claim 1, which is characterized in that the step 3
Middle acquisition interframe movement vector method are as follows:
By its SIFT value of the reconstructed image sequence node-by-node algorithm of initial estimation, SIFT density image f is obtainedi, interframe movement vector w
It indicates are as follows:
Wherein, fiAnd fi+1Indicate the SIFT density image of consecutive frame；W (p) indicates the motion vector at p point；t' it is interceptive value,
For accelerating operation；su,svRespectively indicate SIFT stream, the wavelet coefficient under Haar wavelet transform base；λ1、λ2Indicate weighting coefficient；u
(p) the vertically and horizontally component of motion vector w (p) is respectively indicated with v (p)；α, d respectively indicate truncation funcation weighting coefficient and threshold
Value.
6. resolution compression perceives method for reconstructing between a kind of time-out as claimed in claim 1 or 5, which is characterized in that described more
Beam weight coefficient is weighed about to obtain using following formula:
Wherein, u, v are respectively the vertical and horizontal movement component of current pixel point, and T is the threshold value of motion vector；η1,η2Respectively
Transform domain in multiple constraintWith | | TV (x) | | weight coefficient, γ be normalization operator, make
7. resolution compression perceives method for reconstructing between a kind of time-out as described in claim 1, which is characterized in that thrown using alternating
Shadow algorithm solves the restricted problem in step 4.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201810767498.2A CN109255822B (en) | 2018-07-13 | 2018-07-13 | Multi-scale coding and multi-constraint compression sensing reconstruction method for resolution ratio between times out |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201810767498.2A CN109255822B (en) | 2018-07-13 | 2018-07-13 | Multi-scale coding and multi-constraint compression sensing reconstruction method for resolution ratio between times out |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN109255822A true CN109255822A (en) | 2019-01-22 |
| CN109255822B CN109255822B (en) | 2023-02-24 |
Family
ID=65049066
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201810767498.2A Active CN109255822B (en) | 2018-07-13 | 2018-07-13 | Multi-scale coding and multi-constraint compression sensing reconstruction method for resolution ratio between times out |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN109255822B (en) |
Cited By (9)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN110223238A (en) * | 2019-04-30 | 2019-09-10 | 北京理工大学 | Encoded light shines image reconstruction 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 |
| CN110334676A (en) * | 2019-07-11 | 2019-10-15 | 山东领能电子科技有限公司 | A kind of small target detecting method compared on the blurred picture locally restored based on Feature Descriptor |
| CN110827198A (en) * | 2019-10-14 | 2020-02-21 | 唐山学院 | Multi-camera panoramic image construction method based on compressed sensing and super-resolution reconstruction |
| CN111756602A (en) * | 2020-06-29 | 2020-10-09 | 上海商汤智能科技有限公司 | Communication timeout detection method in neural network model training and related product |
| 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 |
| CN112420214A (en) * | 2019-08-21 | 2021-02-26 | 无锡祥生医疗科技股份有限公司 | Ultrasonic consultation remote sharing method, platform and system |
| CN113516689A (en) * | 2021-09-10 | 2021-10-19 | 北京理工大学 | Schlieren characteristic visualization enhancement method based on associated frame constraint |
| WO2023133888A1 (en) * | 2022-01-17 | 2023-07-20 | 深圳市大疆创新科技有限公司 | Image processing method and apparatus, remote control device, system, and storage medium |
Citations (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN102438102A (en) * | 2011-10-26 | 2012-05-02 | 西安电子科技大学 | Super-resolution imaging system based on compression coding aperture and imaging method thereof |
| US20160173771A1 (en) * | 2014-12-11 | 2016-06-16 | Xerox Corporation | High-resolution imaging devices using low-resolution sensors and compressive sensing exploiting joint sparsity |
| CN107025632A (en) * | 2017-04-13 | 2017-08-08 | 首都师范大学 | A kind of image super-resolution rebuilding method and system |
-
2018
- 2018-07-13 CN CN201810767498.2A patent/CN109255822B/en active Active
Patent Citations (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN102438102A (en) * | 2011-10-26 | 2012-05-02 | 西安电子科技大学 | Super-resolution imaging system based on compression coding aperture and imaging method thereof |
| US20160173771A1 (en) * | 2014-12-11 | 2016-06-16 | Xerox Corporation | High-resolution imaging devices using low-resolution sensors and compressive sensing exploiting joint sparsity |
| CN107025632A (en) * | 2017-04-13 | 2017-08-08 | 首都师范大学 | A kind of image super-resolution rebuilding method and system |
Non-Patent Citations (1)
| Title |
|---|
| 胡彦婷等: "超分辨率重建技术研究进展", 《信息技术》 * |
Cited By (13)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN110223238A (en) * | 2019-04-30 | 2019-09-10 | 北京理工大学 | Encoded light shines image reconstruction 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 |
| CN110334676A (en) * | 2019-07-11 | 2019-10-15 | 山东领能电子科技有限公司 | A kind of small target detecting method compared on the blurred picture locally restored based on Feature Descriptor |
| CN110334676B (en) * | 2019-07-11 | 2023-06-30 | 山东领能电子科技有限公司 | Small target detection method based on feature descriptor comparison on locally restored fuzzy picture |
| CN112420214A (en) * | 2019-08-21 | 2021-02-26 | 无锡祥生医疗科技股份有限公司 | Ultrasonic consultation remote sharing method, platform and system |
| CN112420214B (en) * | 2019-08-21 | 2023-11-21 | 无锡祥生医疗科技股份有限公司 | Remote sharing method, platform and system for ultrasonic consultation |
| CN110827198A (en) * | 2019-10-14 | 2020-02-21 | 唐山学院 | Multi-camera panoramic image construction method based on compressed sensing and super-resolution reconstruction |
| CN110827198B (en) * | 2019-10-14 | 2023-11-21 | 唐山学院 | Multi-camera panoramic image construction method based on compressed sensing and super-resolution reconstruction |
| CN111756602A (en) * | 2020-06-29 | 2020-10-09 | 上海商汤智能科技有限公司 | Communication timeout detection method in neural network model training and related product |
| 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 |
| CN113516689A (en) * | 2021-09-10 | 2021-10-19 | 北京理工大学 | Schlieren characteristic visualization enhancement method based on associated frame constraint |
| WO2023133888A1 (en) * | 2022-01-17 | 2023-07-20 | 深圳市大疆创新科技有限公司 | Image processing method and apparatus, remote control device, system, and storage medium |
Also Published As
| Publication number | Publication date |
|---|---|
| CN109255822B (en) | 2023-02-24 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN109255822A (en) | A kind of multiple dimensioned resolution compression perception method for reconstructing between encoding the time-out multiple constraint | |
| Liu et al. | Video super-resolution based on deep learning: a comprehensive survey | |
| Liu et al. | Depth reconstruction from sparse samples: Representation, algorithm, and sampling | |
| Sankaranarayanan et al. | CS-MUVI: Video compressive sensing for spatial-multiplexing cameras | |
| Farrugia et al. | Light field super-resolution using a low-rank prior and deep convolutional neural networks | |
| Sankaranarayanan et al. | Video compressive sensing for spatial multiplexing cameras using motion-flow models | |
| CN102970529B (en) | A kind of object-based multi-view point video fractal image compression & decompression method | |
| CN111915484A (en) | Reference image guiding super-resolution method based on dense matching and self-adaptive fusion | |
| US20140063314A1 (en) | System And Method Of Video Compressive Sensing For Spatial-Multiplexing Cameras | |
| WO2022016350A1 (en) | Light field image processing method, light field image encoder and decoder, and storage medium | |
| CN112102163B (en) | Continuous multi-frame image super-resolution reconstruction method based on multi-scale motion compensation framework and recursive learning | |
| Cheng et al. | A dual camera system for high spatiotemporal resolution video acquisition | |
| Jia et al. | Convolutional neural network-based occupancy map accuracy improvement for video-based point cloud compression | |
| Hu et al. | An adaptive two-layer light field compression scheme using GNN-based reconstruction | |
| CN116205962B (en) | Monocular depth estimation method and system based on complete context information | |
| Shuai et al. | Mobiup: An upsampling-based system architecture for high-quality video streaming on mobile devices | |
| CN105915892A (en) | Panoramic video quality determination method and system | |
| CN116542889A (en) | Panoramic video enhancement method with stable view point | |
| CN108460723A (en) | Bilateral full variation image super-resolution rebuilding method based on neighborhood similarity | |
| Chandramouli et al. | A generative model for generic light field reconstruction | |
| CN114359041A (en) | Light field image space super-resolution reconstruction method | |
| Zhao et al. | A hybrid-3D convolutional network for video compressive sensing | |
| CN116957931A (en) | Method for improving image quality of camera image based on nerve radiation field | |
| Choudhury et al. | Low bit-rate compression of video and light-field data using coded snapshots and learned dictionaries | |
| Yan et al. | Compressive sampling for array cameras |
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 |