CN106920220B - The turbulent flow method for blindly restoring image optimized based on dark primary and alternating direction multipliers method - Google Patents
The turbulent flow method for blindly restoring image optimized based on dark primary and alternating direction multipliers method Download PDFInfo
- Publication number
- CN106920220B CN106920220B CN201710101304.0A CN201710101304A CN106920220B CN 106920220 B CN106920220 B CN 106920220B CN 201710101304 A CN201710101304 A CN 201710101304A CN 106920220 B CN106920220 B CN 106920220B
- Authority
- CN
- China
- Prior art keywords
- image
- estimated
- gradient
- scale
- target image
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Active
Links
- 238000000034 method Methods 0.000 title claims abstract description 45
- 238000004422 calculation algorithm Methods 0.000 claims abstract description 40
- 230000006870 function Effects 0.000 claims description 17
- 239000011159 matrix material Substances 0.000 claims description 14
- 238000005457 optimization Methods 0.000 claims description 10
- 238000011480 coordinate descent method Methods 0.000 claims description 5
- 230000003190 augmentative effect Effects 0.000 claims description 3
- 230000009977 dual effect Effects 0.000 claims description 3
- 238000000926 separation method Methods 0.000 claims description 3
- 230000000694 effects Effects 0.000 abstract description 9
- 238000011084 recovery Methods 0.000 abstract description 3
- 238000012804 iterative process Methods 0.000 abstract 1
- 230000015556 catabolic process Effects 0.000 description 8
- 238000006731 degradation reaction Methods 0.000 description 8
- 238000011156 evaluation Methods 0.000 description 4
- 238000012545 processing Methods 0.000 description 4
- 230000008569 process Effects 0.000 description 3
- 238000004364 calculation method Methods 0.000 description 2
- 230000007547 defect Effects 0.000 description 2
- 238000013461 design Methods 0.000 description 2
- 238000001514 detection method Methods 0.000 description 2
- 238000002474 experimental method Methods 0.000 description 2
- 238000004088 simulation Methods 0.000 description 2
- 241000208140 Acer Species 0.000 description 1
- 230000009286 beneficial effect Effects 0.000 description 1
- 230000008859 change Effects 0.000 description 1
- 125000004122 cyclic group Chemical group 0.000 description 1
- 238000001914 filtration Methods 0.000 description 1
- 238000005286 illumination Methods 0.000 description 1
- 238000003384 imaging method Methods 0.000 description 1
- 230000003993 interaction Effects 0.000 description 1
- 238000005259 measurement Methods 0.000 description 1
- 239000000203 mixture Substances 0.000 description 1
- 230000003287 optical effect Effects 0.000 description 1
- 230000001575 pathological effect Effects 0.000 description 1
- 238000004321 preservation Methods 0.000 description 1
- 238000003672 processing method Methods 0.000 description 1
- 238000011160 research Methods 0.000 description 1
- 230000035945 sensitivity Effects 0.000 description 1
- 230000003595 spectral effect Effects 0.000 description 1
- 238000012360 testing method Methods 0.000 description 1
- 238000004800 variational method Methods 0.000 description 1
- 230000000007 visual effect Effects 0.000 description 1
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T5/00—Image enhancement or restoration
- G06T5/73—Deblurring; Sharpening
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T5/00—Image enhancement or restoration
- G06T5/70—Denoising; Smoothing
Landscapes
- Physics & Mathematics (AREA)
- General Physics & Mathematics (AREA)
- Engineering & Computer Science (AREA)
- Theoretical Computer Science (AREA)
- Image Processing (AREA)
Abstract
The present invention relates to a kind of turbulent flow method for blindly restoring image optimized based on dark primary and alternating direction multipliers method, first, based on multiple dimensioned thought, apply that dark primary is prior-constrained, applies sparse constraint and energy constraint to point spread function to image on each layer of scale, then, the fuzzy core and image under current scale are estimated using coordinate descent alternating iteration.When reaching out to out, the fuzzy core finally estimated, finally, realizing the fast quick-recovery of image detail using derivative alternating direction multipliers method in conjunction with total variation model.The method of the present invention is using the dark primary prior information of clear image as bound term, be conducive to cost function and converge to clear solution in an iterative process, it solves the problems, such as that blind restoration algorithm is constrained under maximum a posteriori probability frame using gradient prior information and easily acquires fuzzy solution, therefore restoration result visually, more image details can be recovered, ringing effect is less, effectively improves recovering quality.
Description
Technical Field
The invention belongs to a digital image processing method, relates to a new method for restoring a single-frame atmospheric turbulence degraded image, and particularly relates to a turbulence image blind restoration method based on optimization of a dark primary color and an alternating direction multiplier method.
Background
When the aircraft flies at supersonic speed in the atmosphere, the aircraft and the atmosphere have violent interaction to form a complex high-temperature turbulent flow field, and the turbulent flow effect can cause target images received by an optical system of the aircraft to shift, shake, blur and the like, so that the target detection, identification and tracking capabilities of the aircraft are seriously influenced, and even the identification target cannot be detected in serious cases. Therefore, effectively restoring the original target image from the turbulence degradation image is one of the key problems that must be solved for realizing the ultrasonic cruise imaging detection and accurate guidance.
The greatest feature of turbulence is strong randomness, which makes it difficult to describe and measure the specific form of the Point Spread Function (PSF) when modeling, and degradation information must be estimated from the observed image in some way. For such serious pathological problems, blind restoration algorithms are mostly adopted for processing at present, that is, under the framework of Maximum Posterior Probability (MAP), a target image is restored by using an effective estimation criterion according to a fuzzy kernel and the prior knowledge of a clear image. The key to solve the deblurring problem by using a blind restoration algorithm is how to design effective priori knowledge of a clear image to guide image restoration, most of the priori knowledge of the clear image commonly used at present is concentrated on statistical research on gradient distribution of a natural image, for example, Fergus et al adopts a Gaussian mixture model to depict the gradient distribution of the image, Shan et al constructs a piecewise function to fit the gradient distribution of the image, Krishnan et al adopts a super Laplace distribution to approximate the gradient distribution of the image, but the rules of the priori knowledge only conform to the image in a specific scene, and the priori knowledge is directly applied under an MAP framework, so that the blind restoration algorithm can easily obtain a fuzzy solution, and a contradiction is generated. In order to avoid the problem that a blind restoration algorithm is restrained by using image gradient distribution prior information under an MAP framework and a local optimal solution is easy to obtain, Levin does not adopt a mode of jointly estimating a target image and a fuzzy kernel, directly estimates the fuzzy kernel from the fuzzy image, and then estimates the target image by adopting a non-blind restoration algorithm; fergus replaces an image by a variational Bayes method, and the two algorithm processes have the advantages that the optimal solution can be converged to the global optimal solution most probably theoretically, but the algorithm is high in calculation cost and time-consuming, and relatively, an MAP framework is simple and clear and is easy to solve.
Under the MAP framework, in order to make the blind restoration algorithm converge to a clear solution as much as possible, two processing ideas are common, one is to find more appropriate prior information of a clear image, for example, a sparse ratio l is adopted by Krishnan1/l2Norm is used as a constraint term, and Michaeli uses image block cyclic scale invariance as a priori information constraint term, but the constraints are limited to images in a specific scene. Pan applies the dark channel prior theory in image defogging to image deblurring for the first time, and obtains good effect in processing a motion blurred image, a low-illumination blurred image and a non-uniform blurred image, but the algorithm has the defect of sensitivity to noise, and when the blurred image has larger noise, the processing result of the algorithm has ringing effect; another idea is to introduce edge selection and recover an image by using strong edges, but this kind of method involves complicated edge selection, how to design a rule of "large gradient preservation and small gradient rejection" is a problem, and when the saliency of the image is not very strong, the algorithm cannot select a proper edge to estimate the blur kernel.
Therefore, in the traditional atmospheric turbulence image blind restoration method, under the condition of relatively serious noise or fuzzy degree, the image restoration visual quality is poor, artifacts are easily generated, and the method is sensitive to noise.
Disclosure of Invention
Technical problem to be solved
In order to avoid the defects of the prior art, the invention provides a turbulence image blind restoration method based on dark primary color and alternating direction multiplier optimization, which improves the restoration quality of turbulence degraded images and improves the anti-noise performance.
The basic idea of the invention is: under the MAP framework, based on the multi-scale thought, firstly, a fuzzy kernel is estimated for an input observation image by adopting a coarse-to-fine image pyramid method, under the dark primary color constraint and the image gradient constraint, a clear image and a fuzzy kernel under each scale are immediately estimated, when the scale reaches the maximum, the finally estimated fuzzy kernel is obtained, then, under a total variation model, an ADMM is adopted to optimize and solve a target image, noise is removed through an anisotropic filter, and finally, when the algorithm meets the convergence condition, the finally estimated image is obtained.
Technical scheme
A turbulence image blind restoration method based on dark primary color and alternating direction multiplier optimization is characterized by comprising the following steps:
step 1, estimating a fuzzy core and a target image on each layer of scale: based on the multi-scale thought, the original target image x and the fuzzy kernel k are estimated for the target observation image y by adopting the following cost function:
alternately solving x and k by adopting a coordinate descent method; in the above formula, y is the target observed image, x is the original target image,representing convolution, k is the blur kernel, γ, μ and γ are the parametric weights,representing the gradient of the image, d (x) representing the dark primary constraint of the image;
step 2, optimizing by an alternative direction multiplier method: and according to the estimated fuzzy kernel k, optimizing and solving the finally estimated original target image x by adopting an alternating direction multiplier method, wherein the method comprises the following steps:
the total variation model of image restoration is as follows:
where D is the discrete gradient operator,tau is the parameter weight;
the image restoration total variation model based on the derivative space is as follows:
wherein,d is Dx and μ is the parameter weight. Adding an auxiliary variable f, and enabling f to be d;
defining the augmented Lagrangian function as:
updating the parameters f, d and q by adopting an alternating direction multiplier method, and giving the parameters f and q, d to be solved by the following formula:
when d is updated and solved, the Lagrangian dual function of d is constructed as follows:
the KKT condition is:
the corresponding solution is:
b ═ k in the above formulaTk + delta I, formula (6) is put in a Fourier domain to be solved, an ADMM optimization algorithm is adopted to solve, and eps is set to be 10-4The convergence condition is as follows:when the algorithm converges, the final estimated original target image x is obtained.
The coordinate descent method for alternately solving x and k comprises the following steps:
step a: equation (1) is split into the following two cost functions:
step b, estimating a target image x on each layer of scale:
solving for L in equation (9) using semi-quadratic variable separation0Minimizing the problem, introducing an auxiliary variable u for D (·), and introducing g ═ g (g) for the horizontal gradient and the vertical gradient of the image respectivelyh,gv) Rewriting formula (9) is:
solving for any one of the variables x, u, and g in equation (11) by iteratively minimizing x, u, and g in alternation;
when solving for the variable x, the part of equation (11) that finds x is extracted:
where the non-linear operator d (i) in equation (12) is equal to the selection matrix M multiplied by the vector x:
D(x)=Mx (13)
the non-linear operator selection matrix M satisfies:
wherein z is an element of an ith row in the selection matrix M, and j is a position corresponding to a minimum pixel value of the image I under a given image block;
estimating a target image x at the current scale by the following formula:
in the above formula, TkThe method is characterized in that a Toeplitz matrix of a fuzzy kernel k is obtained by Fast Fourier Transform (FFT) under the vector form of y, g and u;
after the target image x estimated under the current scale is solved, u and g are solved by the following formula:
step c, estimating a target image blur kernel k on each layer scale: after obtaining the estimated image x, equation (10) becomes the least squares problem, and is solved by a gradient-based method, rewriting equation (10):
wherein,representing the gradient of the original target image,representing the gradient of the blurred image, k being a blur kernel and gamma being a parameter weight;
after the fuzzy kernel under the current scale is estimated in each iteration, applying non-negativity constraint and energy constraint to the fuzzy kernel:
when the scale reaches the maximum, the final estimated blur kernel k is obtained.
Advantageous effects
The invention provides a turbulence image blind restoration method based on dark primary color and alternating direction multiplier optimization. And when the maximum scale is reached, obtaining a finally estimated fuzzy kernel, and finally, combining a total variation model and adopting a derivative alternating direction multiplier method to realize the rapid recovery of image details. The prior information constraint used by the algorithm is beneficial to obtaining a clear solution, the problem that the fuzzy solution is easy to obtain by using image gradient distribution prior information under an MAP framework in the traditional blind restoration algorithm is solved, meanwhile, the algorithm can be converged to a global optimal solution under a total variation model, artifacts generated in the image restoration process can be effectively inhibited, better target image details are recovered, and finally the solution has a better restoration effect.
The main advantages of the invention include the following aspects: firstly, dark primary color prior information is adopted for constraint, the cost function has low overall energy in the iteration process, clear solution is more favorably obtained, and the contradiction problem that the blind recovery algorithm is easy to obtain the fuzzy solution by using gradient distribution prior information constraint under an MAP framework is solved; secondly, because the variational Method can be converged to a global optimal solution approximately, after an estimated fuzzy core is obtained, an image restoration total variational model is constructed, a derivative Alternating Direction Multiplier Method (ADMM) is adopted to optimally solve a target image, only four steps of FFT (fast Fourier transform) are needed in each iteration, and the speed is high; thirdly, aiming at the problem of poor anti-noise performance of the traditional atmospheric turbulence image blind restoration algorithm, anisotropic filtering denoising is adopted, more image details can be recovered while denoising is carried out, and a satisfactory restoration effect is obtained.
Drawings
FIG. 1: method for restoring image
FIG. 2: blind restoration result of simulated turbulence degradation image by adopting different restoration algorithms
(a) A source image; (b) a turbulence degradation image; (c) recovering the result of the IBD algorithm; (d) the algorithm proposed by Zhu [2013] recovers the result; (e) the algorithm provided by the plum glory [2015] restores the result; (f) restoring the result of the algorithm provided by Pan [2016 ]; (g) the algorithm provided by the invention recovers the result;
FIG. 3: blind restoration result of actual turbulence degraded image by adopting different restoration algorithms
(a) A turbulence degradation image; (b) recovering the result of the IBD algorithm; (c) the algorithm proposed by Zhu [2013] recovers the result; (d) the algorithm provided by the plum glory [2015] restores the result; (e) restoring the result of the algorithm provided by Pan [2016 ]; (f) the algorithm provided by the invention recovers the result;
Detailed Description
The invention will now be further described with reference to the following examples and drawings:
the hardware environment for the experimental implementation herein is: acer V3-572G-59TB computer, 4G memory, 840M is unique, Kurui i5-4210U, software environment is 64 bits running in Windows7 flagship edition, MATLAB software is R2013 b. Two sets of types of experiments were performed, one set being simulated data and the other being measured data. Simulation data the phase of atmospheric turbulence is simulated by a spectral inversion method by using a maritime satellite image of 256piexls multiplied by 256piexlsA position screen for performing turbulence degradation fuzzy simulation experiment on the satellite image, wherein the experiment sets the atmospheric coherence length r0The diameter D of the telescope is 0.05m and 1.0 m. Zhu [2013] is adopted as actually measured image data]A given atmospheric turbulence image test database.
The invention is implemented as follows:
step 1, estimating a fuzzy kernel and a target image on each layer of scale: based on the multi-scale thought, the original target image x and the fuzzy kernel k are estimated for the target observation image y by adopting the following cost function:
wherein y is the target observation image, x is the original target image,representing convolution, k is the blur kernel, γ, μ and γ are the parametric weights,representing the gradient of the image, and d (x) representing the dark primary constraint of the image. In the cost function, the first term is a data fidelity term and measures the similarity between the estimated image and the original target image, the second term is a constraint term of a point spread function, the third term is a gradient constraint on the target image so as to ensure that the estimated image can keep a larger gradient, and the last term is a dark primary color prior information constraint term of the target image.
And (3) solving x and k alternately by adopting a coordinate descent method, and splitting the formula (20) into the following two cost functions:
(a) estimation of target image x at each layer scale
Solving for L in equation (21) using semi-quadratic variable separation0Minimizing the problem, introducing an auxiliary variable u for D (·), and introducing g ═ g (g) for the horizontal gradient and the vertical gradient of the image respectivelyh,gv) Thus, equation (21) can be rewritten as:
solving any one variable in the formula (23) by alternately and iteratively minimizing x, u and g, the part of the formula (23) for x is extracted as follows:
the non-linear operator d (i) is equal to the selection matrix M multiplied by the vector x:
D(x)=Mx (25)
the non-linear operator selection matrix M satisfies:
where z is an element in the ith row in the selection matrix M, and j is a position of the image I corresponding to the minimum pixel value under a given image block. By means of the selection matrix M, the target image x is estimated instantaneously by the formula (27):
in the above formula, TkIs the Toeplitz matrix of the blur kernel k, and y, g and u represent vector forms of y, g and u, respectively. The vector matrix of the above equation is solved by Fast Fourier Transform (FFT).
After the target image x estimated in real time is solved, u and g are solved by the following two equations:
(b) estimation of target image blur kernel k at each layer scale
After the instantaneous estimated image x is obtained from step (a) in step 1, equation (22) becomes the least squares problem, and the gradient-based method is used to solve, and equation (3) is rewritten as follows:
wherein,representing the gradient of the original target image,representing the gradient of the blurred image, k being the blur kernel and gamma being the parameter weight.
After the fuzzy kernel at the current scale is estimated in each iteration, applying non-negativity constraint and energy constraint to the fuzzy kernel as follows:
when the scale reaches the maximum, the final estimated blur kernel k is obtained.
Step 2, optimizing by using an alternating direction multiplier method: and (3) after the finally estimated fuzzy kernel k is obtained in the step (1), an alternative direction multiplier method is adopted to optimally solve the finally estimated original target image x.
The total variation model for image restoration can be generally expressed as follows:
where D is the discrete gradient operator,τ is the parameter weight.
The image restoration total variation model based on the derivative space is as follows:
wherein,d is Dx and μ is the parameter weight. An auxiliary variable f is added, let f be d. Defining the augmented Lagrangian function as:
updating parameters f, d and q by adopting an alternating direction multiplier method, wherein the key is to solve d, and given parameters f and q, d can be solved by the following formula:
when d is updated and solved, the Lagrangian dual function of d is constructed as follows:
the KKT condition is:
the corresponding solution is:
in the above formula, B ═ kTk + δ I, equation (38) is solved in fourier domain by using ADMM optimization algorithm, and eps is set to 10-4The convergence condition is as follows:when the algorithm converges, the final estimated original target image x can be obtained.
The Peak Signal-to-Noise Ratio (PSNR) of the image is selected as an objective evaluation index for simulating the turbulence degradation image data, and different images are quantitatively evaluated by adopting different algorithms for restoration effects. The objective evaluation results are shown in table 1. Wherein the peak signal-to-noise ratio is defined as shown in formula (39).
PSNR=10*log(2552/MSE) (39)
In formula (39), the multiplication is represented, and the MSE represents the mean square error, which is defined as formula (40).
In the formula (40), m and n represent the length and width of the image, f (i, j) andrespectively representing the gray values of the ideal image and the evaluated image at the pixel points (i, j).
TABLE 1 comparison of algorithms for simulating turbulent images
Algorithm | PSNR(dB) |
IBD | 13.5881 |
Pan[2016] | 17.5771 |
Li Hui (2015)] | 17.3317 |
In this context | 17.8309 |
The Gray Mean Gradients (GMG) and laplacian gradient (LS) of the image are selected to measure the restoration effect of the actually measured degraded image. GMG and LS belong to non-reference evaluation, can effectively reflect the contrast and texture change characteristics of the image, and the larger the value of the GMG and the LS, the more the image details are, the sharper the edge is, and the better the image quality is. The calculation formulas of GMG and LS are respectively as follows:
where g denotes the evaluated image of size M × N, and i and j denote the row and column coordinates, respectively, of the image g.
The objective evaluation results are shown in table 2.
TABLE 2 actual measurement turbulence degradation image algorithm comparison
Claims (2)
1. A turbulence image blind restoration method based on dark primary color and alternating direction multiplier optimization is characterized by comprising the following steps:
step 1, estimating a fuzzy core and a target image on each layer of scale: based on the multi-scale thought, the original target image x and the fuzzy kernel k are estimated for the target observation image y by adopting the following cost function:
alternately solving x and k by adopting a coordinate descent method; in the above formula, y is the target observed image, x is the original target image,representing convolution, k being a blur kernel, γ, μ and γ being parametric weights, ▽ x representing the gradient of the image, d (x) representing the dark primary constraint of the image;
step 2, optimizing by an alternative direction multiplier method: and according to the estimated fuzzy kernel k, optimizing and solving the finally estimated original target image x by adopting an alternating direction multiplier method, wherein the method comprises the following steps:
the total variation model of image restoration is as follows:
where D is the discrete gradient operator,tau is the parameter weight;
the image restoration total variation model based on the derivative space is as follows:
wherein,d is Dx and mu is the parameter weight; adding an auxiliary variable f, and enabling f to be d;
defining the augmented Lagrangian function as:
updating the parameters f, d and q by adopting an alternating direction multiplier method, and giving the parameters f and q, d to be solved by the following formula:
when d is updated and solved, the Lagrangian dual function of d is constructed as follows:
the KKT condition is:
the corresponding solution is:
b ═ k in the above formulaTk + delta I, formula (6) is put in a Fourier domain to be solved, an ADMM optimization algorithm is adopted to solve, and eps is set to be 10-4The convergence condition is as follows:when the algorithm converges, the final estimated original target image x is obtained.
2. The blind restoration method for turbulent images based on the optimization of the dark primaries and the alternating direction multiplier method as claimed in claim 1, wherein: the coordinate descent method for alternately solving x and k comprises the following steps:
step a: equation (1) is split into the following two cost functions:
step b, estimating a target image x on each layer of scale:
solving a formula (9) by adopting semi-quadratic variable separation, introducing an auxiliary variable u into D (·), and respectively introducing g ═ g into the horizontal gradient and the vertical gradient of the imageh,gv) Rewriting formula (9) is:
solving for any one of the variables x, u, and g in equation (11) by iteratively minimizing x, u, and g in alternation;
when solving for the variable x, the part of equation (11) that finds x is extracted:
where the non-linear operator d (x) in equation (12) is equal to the selection matrix M multiplied by the vector x:
D(x)=Mx (13)
the non-linear operator selection matrix M satisfies:
wherein z is an element of an ith row in the selection matrix M, and j is a position corresponding to a minimum pixel value of the image I under a given image block;
estimating a target image x at the current scale by the following formula:
in the above formula, TkThe method is characterized in that a Toeplitz matrix of a fuzzy kernel k is obtained by Fast Fourier Transform (FFT) under the vector form of y, g and u;
after the target image x estimated under the current scale is solved, u and g are solved by the following formula:
step c, estimating a target image blur kernel k on each layer scale: after obtaining the estimated image x, equation (10) becomes the least squares problem, and is solved by a gradient-based method, rewriting equation (10):
wherein,representing the gradient of the original target image,representing the gradient of the blurred image, k being a blur kernel and gamma being a parameter weight;
after the fuzzy kernel under the current scale is estimated in each iteration, applying non-negativity constraint and energy constraint to the fuzzy kernel:
when the scale reaches the maximum, the final estimated blur kernel k is obtained.
Priority Applications (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201710101304.0A CN106920220B (en) | 2017-02-24 | 2017-02-24 | The turbulent flow method for blindly restoring image optimized based on dark primary and alternating direction multipliers method |
Applications Claiming Priority (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201710101304.0A CN106920220B (en) | 2017-02-24 | 2017-02-24 | The turbulent flow method for blindly restoring image optimized based on dark primary and alternating direction multipliers method |
Publications (2)
Publication Number | Publication Date |
---|---|
CN106920220A CN106920220A (en) | 2017-07-04 |
CN106920220B true CN106920220B (en) | 2019-07-19 |
Family
ID=59454169
Family Applications (1)
Application Number | Title | Priority Date | Filing Date |
---|---|---|---|
CN201710101304.0A Active CN106920220B (en) | 2017-02-24 | 2017-02-24 | The turbulent flow method for blindly restoring image optimized based on dark primary and alternating direction multipliers method |
Country Status (1)
Country | Link |
---|---|
CN (1) | CN106920220B (en) |
Cited By (1)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN108332685A (en) * | 2018-02-05 | 2018-07-27 | 哈尔滨理工大学 | A kind of coding structural light three-dimensional measurement method |
Families Citing this family (14)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN108154482A (en) * | 2017-10-24 | 2018-06-12 | 天津大学 | Image Blind deblurring method based on dark channel prior and multi-direction Weighted T V |
CN107784361B (en) * | 2017-11-20 | 2020-06-26 | 北京大学 | Image recognition method for neural network optimization |
CN108830802B (en) * | 2018-05-16 | 2022-07-08 | 浙江大学 | Image blur kernel estimation method based on short exposure image gradient guidance |
CN110796616B (en) * | 2019-10-23 | 2022-05-10 | 武汉工程大学 | Turbulence degradation image recovery method based on norm constraint and self-adaptive weighted gradient |
CN111047544B (en) * | 2020-01-08 | 2022-09-23 | 华中科技大学 | Saturated image deblurring method based on nonlinear degradation model |
CN112184567A (en) * | 2020-08-28 | 2021-01-05 | 江苏海洋大学 | Multi-channel blind identification adaptive optical image restoration method based on alternate minimization |
CN112365420B (en) * | 2020-11-12 | 2024-03-15 | 重庆邮电大学 | Fuzzy image blind restoration method based on non-local window gradient |
CN112700370B (en) * | 2020-12-29 | 2023-06-13 | 南京理工大学 | Hyper-spectral image super-resolution method for super-parameter fidelity and depth priori combined learning |
CN112597433B (en) * | 2021-01-11 | 2024-01-02 | 中国人民解放军国防科技大学 | Fourier phase recovery method and system based on plug-and-play neural network |
CN113379647B (en) * | 2021-07-08 | 2024-07-02 | 湘潭大学 | Multi-feature image restoration method for optimizing PSF estimation |
CN113538374B (en) * | 2021-07-15 | 2023-11-28 | 中国科学院上海技术物理研究所 | Infrared image blurring correction method for high-speed moving object |
CN114897734B (en) * | 2022-05-18 | 2024-05-28 | 北京化工大学 | Measured target image restoration method based on gradient direction priori |
CN116309190B (en) * | 2023-05-17 | 2023-08-15 | 武汉工程大学 | Turbulence degradation image recovery method based on optimal region median prior |
CN117078538B (en) * | 2023-07-19 | 2024-02-13 | 华中科技大学 | Correction method of remote atmospheric turbulence image based on pixel motion statistics |
Citations (3)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN104091314A (en) * | 2014-07-22 | 2014-10-08 | 西北工业大学 | Turbulence-degraded image blind restoration method based on edge prediction and sparse ratio regular constraints |
CN105913394A (en) * | 2016-04-08 | 2016-08-31 | 南京理工大学 | Degraded image sequence-based blind image restoration method |
CN105957043A (en) * | 2016-06-22 | 2016-09-21 | 西北工业大学 | Gradient automatic activation-based blurred image blind restoration method |
Family Cites Families (2)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
JP5779724B2 (en) * | 2012-09-26 | 2015-09-16 | 富士フイルム株式会社 | Image processing apparatus, imaging apparatus, computer, and program |
US10048338B2 (en) * | 2014-04-25 | 2018-08-14 | Mayo Foundation For Medical Education And Research | Systems and methods for efficiently generating magnetic resonance images from incomplete data |
-
2017
- 2017-02-24 CN CN201710101304.0A patent/CN106920220B/en active Active
Patent Citations (3)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN104091314A (en) * | 2014-07-22 | 2014-10-08 | 西北工业大学 | Turbulence-degraded image blind restoration method based on edge prediction and sparse ratio regular constraints |
CN105913394A (en) * | 2016-04-08 | 2016-08-31 | 南京理工大学 | Degraded image sequence-based blind image restoration method |
CN105957043A (en) * | 2016-06-22 | 2016-09-21 | 西北工业大学 | Gradient automatic activation-based blurred image blind restoration method |
Non-Patent Citations (2)
Title |
---|
Multichannel Blind Iterative Image Restoration;Filip ˘Sroubek等;《IEEE Transactions on Image Processing》;20030818;第12卷(第9期);第1094-1106页 |
基于Cosparse模型的图像复原算法研究;姬社平;《中国优秀硕士学位论文全文数据库信息科技辑》;20151215(第12期);全文 |
Cited By (2)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN108332685A (en) * | 2018-02-05 | 2018-07-27 | 哈尔滨理工大学 | A kind of coding structural light three-dimensional measurement method |
CN108332685B (en) * | 2018-02-05 | 2019-11-29 | 哈尔滨理工大学 | A kind of coding structural light three-dimensional measurement method |
Also Published As
Publication number | Publication date |
---|---|
CN106920220A (en) | 2017-07-04 |
Similar Documents
Publication | Publication Date | Title |
---|---|---|
CN106920220B (en) | The turbulent flow method for blindly restoring image optimized based on dark primary and alternating direction multipliers method | |
CN105493140B (en) | Image deblurring method and system | |
CN110675347B (en) | Image blind restoration method based on group sparse representation | |
Shi et al. | Total variation image restoration using hyper-Laplacian prior with overlapping group sparsity | |
CN103761710B (en) | The blind deblurring method of efficient image based on edge self-adaption | |
CN104103050B (en) | A kind of real video restored method based on local policy | |
CN107133923B (en) | Fuzzy image non-blind deblurring method based on adaptive gradient sparse model | |
CN106127688B (en) | A kind of super-resolution image reconstruction method and its system | |
CN110796616B (en) | Turbulence degradation image recovery method based on norm constraint and self-adaptive weighted gradient | |
CN104282003B (en) | Digital blurred image blind restoration method based on gradient screening | |
CN106097256B (en) | A kind of video image fuzziness detection method based on Image Blind deblurring | |
Xia et al. | A discrete-time learning algorithm for image restoration using a novel L2-norm noise constrained estimation | |
CN109671029A (en) | Image denoising algorithm based on gamma norm minimum | |
CN113538374B (en) | Infrared image blurring correction method for high-speed moving object | |
CN113962908B (en) | Pneumatic optical effect large-visual-field degraded image point-by-point correction restoration method and system | |
CN112215773A (en) | Local motion deblurring method and device based on visual saliency and storage medium | |
CN112233046B (en) | Image restoration method under Cauchy noise and application thereof | |
US9008453B2 (en) | Blur-kernel estimation from spectral irregularities | |
CN112184567A (en) | Multi-channel blind identification adaptive optical image restoration method based on alternate minimization | |
CN112862729B (en) | Remote sensing image denoising method based on characteristic curve guidance | |
CN114037636A (en) | Multi-frame blind restoration method for correcting image by adaptive optical system | |
Zeng et al. | Single image motion deblurring based on modified denseNet | |
Santhosh et al. | A novel approach for denoising coloured remote sensing image using Legendre Fenchel transformation | |
Nunes et al. | A novel approach for mitigating atmospheric turbulence using weighted average Sobolev gradient and Laplacian | |
Wen et al. | Patch-wise blind image deblurring via Michelson channel prior |
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 |