CN107730468B - Method for recovering sharp image under unmanned aerial vehicle fuzzy noise image - Google Patents

Method for recovering sharp image under unmanned aerial vehicle fuzzy noise image Download PDF

Info

Publication number
CN107730468B
CN107730468B CN201710990638.8A CN201710990638A CN107730468B CN 107730468 B CN107730468 B CN 107730468B CN 201710990638 A CN201710990638 A CN 201710990638A CN 107730468 B CN107730468 B CN 107730468B
Authority
CN
China
Prior art keywords
image
matrix
representing
equation
fuzzy
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
Application number
CN201710990638.8A
Other languages
Chinese (zh)
Other versions
CN107730468A (en
Inventor
张肇健
刘宏清
Current Assignee (The listed assignees may be inaccurate. Google has not performed a legal analysis and makes no representation or warranty as to the accuracy of the list.)
Yingaisi Technology Shenzhen Co ltd
Original Assignee
Yingaisi Technology Shenzhen Co ltd
Priority date (The priority date 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 date listed.)
Filing date
Publication date
Application filed by Yingaisi Technology Shenzhen Co ltd filed Critical Yingaisi Technology Shenzhen Co ltd
Priority to CN201710990638.8A priority Critical patent/CN107730468B/en
Publication of CN107730468A publication Critical patent/CN107730468A/en
Application granted granted Critical
Publication of CN107730468B publication Critical patent/CN107730468B/en
Active legal-status Critical Current
Anticipated expiration legal-status Critical

Links

Images

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

Landscapes

  • Physics & Mathematics (AREA)
  • General Physics & Mathematics (AREA)
  • Engineering & Computer Science (AREA)
  • Theoretical Computer Science (AREA)
  • Image Processing (AREA)

Abstract

The invention relates to an image restoration technology, in particular to a method for restoring a clear image under a fuzzy noise image of an unmanned aerial vehicle, which comprises the following steps: acquiring a fuzzy image of the unmanned aerial vehicle, and constructing a first optimization equation; substituting the group sparse domain of the clear image constructed by the dictionary learning method and the group sparse domain of the fuzzy kernel matrix into the first optimization equation to obtain a second optimization equation; substituting the second optimal equation into a separated Brazilian iteration SBI algorithm to obtain a fuzzy kernel matrix to obtain a restored image; the method constructs a joint estimation optimization equation by using the sparse characteristics of the fuzzy kernel and the original image in the group sparse domain, obtains the estimation of the fuzzy kernel while eliminating the fuzzy, obtains the most accurate solution through an iterative algorithm, and restores the fuzzy image under the condition of fully preserving the image details.

Description

Method for recovering sharp image under unmanned aerial vehicle fuzzy noise image
Technical Field
The invention relates to the technical field of image restoration, in particular to a method for restoring a clear image under a fuzzy noise image of an unmanned aerial vehicle.
Background
In 2006, Fergus et al proposed a spatial domain prior model, which can effectively estimate a Point Spread Function (PSF) and remove a complicated camera shake blur, but only an RL deconvolution is used to reconstruct an image, and the ringing effect in the recovered result is significant. In 2007, jiaya proposed a new method for estimating PSF using image transparency map. Shan in 2008 proposes a blind restoration method in which PSF estimation and image restoration are performed simultaneously, and the effect is good. Joshi, Cho, Lee and the like adopt a simplified Gaussian prior model to estimate a fuzzy core by predicting a clear boundary of an image, so that rapid solution can be realized, but the noise influence is overlarge. Levin summarizes the existing deconvolution deblurring algorithm, obtains the conclusion that the solution of the problem of the blur kernel and the hidden image by using a Maximum A Posteriori (MAP) probability method is unreliable, and provides a method for independently estimating the blur kernel by using the Maximum a posteriori probability method. Li Xu et al proposed a fuzzy core estimation algorithm based on selection of an algorithm with informative boundaries and ISD, which separately performed initialization of a fuzzy core and extraction of detailed structures of the fuzzy core, and extracted a relatively real fuzzy core. Hui Ji et al propose a more robust image deblurring method under the condition that the blur kernel has errors, and simultaneously estimate the blur kernel and restore the image, thereby obtaining better effect. Yuan et al, 2008, proposed a method of deconvolution of non-blind images between and within scales called progressive. They estimated the PSF of the blurred image using the method of Fergus in 2006, and designed a multi-scale iterative non-blind deconvolution method that can effectively reduce ringing distortion in deblurred images. In 2009 Joshi et al combined with local two-color priors to remove noise, this method uses an iterative weighted least squares approach to solve the nonlinear most effective problem. Uwe Schmidt in 2011 provides an algorithm for non-blind deconvolution noise estimation and hidden image estimation based on Bayes minimum mean square error sampling, and a good effect is achieved. In recent years, sparse representation is becoming more popular in the field of image processing, and has a non-trivial position in the fields of image restoration, image denoising, super-resolution reconstruction, pattern recognition, and the like. In 2014, Zhang Jian and the like propose a method for solving the problem of image restoration by establishing a sparse representation model of a structure group by using the structure group as a basic unit for image sparse representation, and the like, and a good effect is achieved.
Image deblurring is a classic problem in image restoration technology, and a classic model of image restoration technology is shown in fig. 2, and a plurality of relevant technical researches are widely carried out at home and abroad aiming at the problem. The diversity of natural images makes the image deblurring process relatively difficult, and due to the complexity of the blurred image degradation process, an image degradation model is usually built according to objective hypothesis constraints. Currently, a widely recognized basic image degradation model can be expressed as the convolution of a sharp image with a point spread function plus the effect of noise. The image deblurring process can be viewed as the inverse of this operation. According to the acquisition condition of the prior knowledge, the image deblurring processing can be divided into non-blind processing and blind processing. The non-blind processing is that under the condition that fuzzy kernels or image degradation model prior knowledge is known, an original image needs to be restored from an observed fuzzy image, common methods include wiener filtering and Richardson-Lucy regularization algorithms, but the restoration result can generate a serious ringing effect. However, in the actual restoration, it is not possible to know in advance so much knowledge about the image degradation model that the algorithm can obtain, and therefore, the restored image effect is not satisfactory. In actual life, what people really need is a process of restoring an original image by combining an observed blurred image with the existing image characteristic knowledge and fully utilizing the prior knowledge of a blur kernel and noise, wherein the process is the basic idea and the essential significance of image blind processing.
Disclosure of Invention
In view of the above technical problems, the present invention provides a method for recovering a sharp image under a blurred noise image of an unmanned aerial vehicle, as shown in fig. 1, the method includes:
s1, acquiring a fuzzy image of the unmanned aerial vehicle, and constructing a first optimization equation;
s2, substituting the group sparse domain of the clear image and the group sparse domain of the fuzzy kernel matrix obtained by the dictionary learning method into the first optimization equation to obtain a second optimization equation;
and S3, substituting the second optimization equation into a separated Brazilian iteration SBI algorithm to obtain a fuzzy kernel matrix and obtain a restored image.
Preferably, constructing the first optimization equation according to obtaining the blurred image of the drone includes:
the fuzzy image of the unmanned aerial vehicle is as follows:
y＝h*x+n
the vector is represented as:
Y＝HX+N
the first optimization equation is constructed as follows:
Figure GDA0002672288830000031
wherein | · | purple sweet1Represents the norm of · L1,
Figure GDA0002672288830000032
euclidean distance of expression, axSparse representation representing sharp images, aHRepresenting a sparse representation of a blur kernel matrix, λ, β representing constraint coefficients of the norm L1, Y representing the captured marred image matrix, H representing a blur kernel matrix, N representing gaussian noise, N representing a gaussian noise matrix, and x representing a convolution.
Preferably, the second optimization equation is expressed as:
Figure GDA0002672288830000033
wherein,
Figure GDA0002672288830000036
representing group sparse forms, DxGroup sparse domain, D, representing sharp imagesHA set sparse domain representing a blur kernel matrix.
Preferably, the group sparse domain is composed of similarity modules, wherein the similarity modules are determined by euclidean distance, and in order to increase the robustness of matching block selection, Lp norm is introduced as distance measure, and then the influence of noise is considered, and noise variance is introduced, and is expressed as:
Figure GDA0002672288830000035
wherein | · | purple sweetpLp norm, x of expression1iDenotes the ith target block, y1iDenotes the ith known block, N denotes the number of target blocks, and S denotes the standard deviation of the sample sequence.
Preferably, substituting the second optimization equation into the separable bragman iteration SBI algorithm to solve the fuzzy kernel matrix, and obtaining the restored image includes:
designing a two-step iteration method to solve the first fuzzy kernel estimation matrix, and solving the first clear image estimation matrix according to the solution of the first fuzzy kernel estimation matrix;
rewriting the first fuzzy kernel estimation matrix and the first clear image estimation matrix into a second fuzzy kernel estimation matrix and a second clear image estimation matrix which are suitable for an SBI algorithm;
introducing a problem equation, solving the problem equation by using an SBI algorithm, and solving sparse representation of a clear image through multiple iterations;
obtaining error residuals of multiple iterations through a soft decision threshold of a component mode, solving to obtain a fuzzy kernel matrix, and obtaining a restored image;
where u and v represent variables, f (u) represents a norm equation for the variable u, G (v) represents a norm equation for the variable v, and G represents a relation between the variables u, v.
Preferably, the two-step iterative method comprises:
the first step is as follows: solving a first fuzzy core estimation matrix according to the following fuzzy core equation:
Figure GDA0002672288830000041
the second step is that: solving the following sharp image equation according to the first blur kernel estimation matrix:
Figure GDA0002672288830000042
wherein | · | purple sweet1Represents the norm of · L1,
Figure GDA0002672288830000043
the euclidean distance of the expression is,
Figure GDA0002672288830000044
representation group sparse form axSparse representation representing sharp images, aHRepresenting a sparse representation of the blur kernel matrix, λ, β represent the constraint coefficients of the L1 norm, Y represents the captured impaired image matrix, and H represents the blur kernel matrix.
Preferably, the first blur kernel estimation matrix and the second sharp image estimation matrix are respectively:
first blur kernel estimation matrix:
Figure GDA0002672288830000045
Figure GDA0002672288830000046
second sharp image estimation matrix:
Figure GDA0002672288830000051
Figure GDA0002672288830000052
wherein | · | purple sweet1Represents the norm of · L1,
Figure GDA0002672288830000053
the euclidean distance of the expression is,
Figure GDA0002672288830000054
representing a group sparse form, axSparse representation of sharp images and aHRepresenting a sparse representation of a blur kernel matrix, λ, β representing the constraint coefficients of the L1 norm, Y representing the captured impaired image matrix, H representing the blur kernel matrix, DxGroup sparse domain, D, representing sharp imagesHA set sparse domain representing a blur kernel matrix.
Preferably, the problem equation comprises:
Figure GDA0002672288830000055
s.t.u＝Gv
the problem equation writes extended lagrangian functions of f (u) and g (v) by solving variables u, v in a loop iteration mode, wherein the lagrangian extension function of f (u) is as follows:
Figure GDA0002672288830000056
the lagrange expansion function of g (v) is:
Figure GDA0002672288830000057
wherein μ denotes a penalty factor, b denotes an iterative error residue, u and v denote variables, f (u) denotes a norm equation for the variable u, G (v) denotes a norm equation for the variable v, G denotes a relation between the variables u, v,
Figure GDA0002672288830000058
representing the euclidean distance of.
On the basis of the prior art, the invention constructs a joint estimation optimization equation by using the sparse characteristics of the fuzzy kernel and the original image in the group sparse domain, obtains the estimation of the fuzzy kernel while eliminating the fuzzy, obtains the most accurate solution by an iterative algorithm, and recovers the fuzzy image under the condition of fully preserving the image details so as to obtain better recovery effect.
Drawings
FIG. 1 is a flowchart of a method for recovering a sharp image under a blurred noise image of an unmanned aerial vehicle according to a preferred embodiment of the invention;
FIG. 2 is an image degradation-restoration model of the present invention;
FIG. 3 is a graph illustrating the effect of different values of the Lp norm on the selection of a matching block in the proposed algorithm of the preferred embodiment of the present invention;
FIG. 4 is a comparison of an image of an automobile taken by the UAV and a repaired image of the invention, with the image of the automobile taken by the UAV on the left and the image of the repaired image of the invention on the right;
fig. 5 is a comparison graph of a human body image shot by the unmanned aerial vehicle and a human body image restored by the present invention, wherein the left side is the human body image shot by the unmanned aerial vehicle, and the right side is the human body image restored by the present invention.
Detailed Description
In order to make the objects, technical solutions and advantages of the present invention more clearly and completely apparent, the technical solutions in the embodiments of the present invention are described below with reference to the accompanying drawings, and it is obvious that the described embodiments are only a part of the embodiments of the present invention, and not all of the embodiments.
The invention discloses a method for recovering a clear image under a fuzzy noise image of an unmanned aerial vehicle, which comprises the following steps:
the known signal model is as follows:
y＝h*x+n
where y represents the corrupted image, h represents the blur kernel, n represents gaussian noise, and x represents the sharp image.
Expressed in column vector form as:
Y＝HX+N
wherein Y, X and N respectively represent column vector forms of Y, X and N, Y represents a damaged image matrix, H represents a fuzzy kernel matrix, N represents a Gaussian noise matrix, and X represents a clear image matrix.
Constructing a first optimization equation according to the damaged image:
Figure GDA0002672288830000061
wherein, ax，aHRepresenting a sharp image and a sparse representation of the blur kernel matrix, the problem of image restoration then becomes a solution to ax，aHTo a problem of (a).
The invention | · | non-conducting phosphor1L1 norm of expression，
Figure GDA0002672288830000071
The euclidean distance of the expression is,
Figure GDA0002672288830000072
the influence of the matching block in the proposed algorithm of the preferred embodiment of the present invention when selecting norms of different p values in the proposed algorithm is shown in fig. 3, wherein when the p values are the same, the Peak Signal to Noise Ratio (PSNR) increases as the Signal-to-Noise Ratio (SNR) increases, and when the p values are the same, the smaller the p value, the larger the output Signal-to-Noise Ratio (PSNR) is. As can be seen from the figure, PSNR is maximum when p is 0.5. Therefore, in this patent, p is preferably 0.5 as the value of p for the similar block search.
According to the signal reconstruction rule, to obtain a sparse representation, the transform domain needs to be known, and thus the matrix Dx，DHThe set of sparse domains used to represent the sharp image and the blur kernel, respectively, may result in a second optimization equation:
Figure GDA0002672288830000073
wherein,
Figure GDA0002672288830000074
sparse representation a for representing group sparse form in order to solve for sharp imagesxSparse representation of a fuzzy kernel matrix aHA two-step iteration solving method is provided:
the first step is as follows: initializing the solution of the fuzzy core:
Figure GDA0002672288830000075
the second step is that: solving using a first fuzzy kernel estimation matrix solved in the first section:
Figure GDA0002672288830000076
circularly iterating between the first step and the second step until a stop condition is met; wherein,
Figure GDA0002672288830000077
a first blur kernel estimation matrix is represented,
Figure GDA0002672288830000078
representing a first sharp image estimate matrix.
The group sparse domain used in the invention provides a method for exploring local and non-local information of an image, the group sparse domain performs sufficient learning by taking a group as a unit, and each group is formed by combining a plurality of similar blocks together. Therefore, in group sparseness, the search for a similar block that matches the target block is important.
For simplicity, the similarity block is determined by the euclidean distance, expressed as:
Figure GDA0002672288830000079
wherein x is1Representing the target block, y1Is an image block to be determined. In the present invention, in order to increase the robustness of matching block selection, Lp norm is introduced as distance measure, which is expressed as:
Figure GDA0002672288830000081
obviously, when p is 2, it is reduced to euclidean distance. In addition, considering that the influence of noise may seriously destroy the result of the matching block search, in order to eliminate the influence of noise, noise variance is introduced, and then Lp norm distance is modified as:
Figure GDA0002672288830000082
when p is 2, it becomes simplified to a mahalanobis distance, where x is1iDenotes the ith target block, y1iRepresenting the ith image block to be determined, s representing x in the sample sequence1i，y1iIs calculated by the following formula:
Figure GDA0002672288830000083
wherein, N represents the number of samples,
Figure GDA0002672288830000084
the sample mean is indicated.
With the determination of the weight coefficients and the selection of the matching blocks, the solution of the second optimization equation follows, and in order to effectively solve the second optimization equation, an SBI algorithm is introduced, which is used to solve the following problem:
Figure GDA0002672288830000085
s.t.u＝Gv
SBI is implemented by solving in a loop iterative manner the variables u, v, f (u) therein to represent a norm equation for the variable u, G (v) to represent a norm equation for the variable v, G to represent a relationship between the variables u, v, RNRepresenting an N-dimensional vector.
The SBI solves the variables u and v in the SBI in a loop iteration mode, and the specific solving process is as follows:
Figure GDA0002672288830000091
wherein,
Figure GDA0002672288830000092
representing two extended Lagrangian functions, k-tablesThe number of iterations is shown,
Figure GDA0002672288830000093
when the minimum value is taken
Figure GDA0002672288830000094
Value of (a), bk+1Representing the residual error of the iteration of k +1, b0、u0、v0Initial values of the residual error, variables u, v of the iteration are represented, respectively.
In order to cooperate with the SBI algorithm, the first blur kernel estimation matrix and the first sharp image estimation matrix are rewritten into a second blur kernel estimation matrix and a second sharp image estimation matrix suitable for the SBI algorithm:
second blur kernel estimation matrix:
Figure GDA0002672288830000095
Figure GDA0002672288830000096
second sharp image estimation matrix:
Figure GDA0002672288830000097
Figure GDA0002672288830000098
substituting the second blur kernel estimation matrix and the second sharp image estimation matrix into an SBI algorithm to obtain a first sharp image equation set for the sharp image:
Figure GDA0002672288830000101
a first fuzzy kernel equation set for each fuzzy kernel:
Figure GDA0002672288830000102
for the u-subproblem in the first sharp image equation set to be actually a least square optimization problem under the constraint of L2 norm, the approximate form of the solution can be easily found as follows:
Figure GDA0002672288830000103
wherein,
Figure GDA0002672288830000104
it is noted that, to simplify the identification of iterations in the above equation is omitted,
Figure GDA0002672288830000105
is composed of
Figure GDA0002672288830000106
Mu is a penalty factor.
Equation a for the first sharp imagexSub-problem, if DXThe optimization problem can be well solved by simple threshold judgment when the system is a traditional sparse domain, such as a wavelet domain and a tight wavelet domain, and the group sparse domain D under the group sparse representationXIs complex, multiple, and cannot apply the threshold decision directly to all groups, however in each iteration the following equation holds:
Figure GDA0002672288830000107
wherein,
Figure GDA0002672288830000109
denotes the G thiGroup sparse domain of groups, on the basis of which, for axThe solution of (a) can be obtained by an independent solution for each group,namely:
Figure GDA0002672288830000108
the solution of the above equation can be obtained by component-wise soft decision threshold, that is:
Figure GDA0002672288830000111
wherein, the soft decision threshold T is defined as:
Figure GDA0002672288830000112
the optimization problem in soft decision threshold is applied independently to all groups, and axBy combining all
Figure GDA0002672288830000116
Is obtained in which
Figure GDA0002672288830000117
Denotes the G thiSparse representation of the clear image of the group.
The first sharp image equation set and the first blur kernel equation set have similar structures, and the first sharp image equation set and the first blur kernel equation set can be obtained according to the solution of the first sharp image equation set:
Figure GDA0002672288830000113
Figure GDA0002672288830000114
wherein,
Figure GDA0002672288830000118
denotes the G thiGroup sparse domain of groups, aHBy combining all
Figure GDA0002672288830000119
The method comprises the steps of (1) obtaining,
Figure GDA00026722888300001110
denotes the G thiSparse representation of the fuzzy kernels of the group.
The algorithm structure of the solution is as follows:
TABLE II:The proposed algorithm.
Figure GDA0002672288830000115
in summary, according to the embodiments of the present invention, the set-up sparse-group-based image sparse representation model replaces a conventional image block with an image group structure as a basic unit, learns and obtains an adaptive dictionary for each image group structure, constructs a joint estimation optimization equation by using the sparse characteristics of a fuzzy kernel and an original image in a group sparse domain, obtains an estimation on the fuzzy kernel while eliminating the blur, obtains a most accurate solution by an iterative algorithm, and restores the fuzzy image under the condition of fully preserving image details; as shown in fig. 4, the image repaired by the method is clearer than the image shot by the unmanned aerial vehicle, and the license plate number in the image is compared to be known to be fuzzy, so that the license plate number can be clearly seen after the image is repaired by the method; as shown in fig. 5, the picture that unmanned aerial vehicle shot is whole fuzzy, seriously blurring, and in the picture the face portrait blends with the background, and the prosthetic image detail position of this patent is clear, and the lines are obvious.
Those skilled in the art will appreciate that all or part of the steps in the methods of the above embodiments may be implemented by associated hardware instructed by a program, which may be stored in a computer-readable storage medium, and the storage medium may include: ROM, RAM, magnetic or optical disks, and the like.
Furthermore, the terms "first", "second", "third", "fourth" are used for descriptive purposes only and are not to be construed as indicating or implying a relative importance or implicitly indicating the number of technical features indicated, whereby the features defined as "first", "second", "third", "fourth" may explicitly or implicitly include at least one such feature and are not to be construed as limiting the invention.
The above-mentioned embodiments, which further illustrate the objects, technical solutions and advantages of the present invention, should be understood that the above-mentioned embodiments are only preferred embodiments of the present invention, and should not be construed as limiting the present invention, and any modifications, equivalents, improvements, etc. made within the spirit and principle of the present invention should be included in the protection scope of the present invention.

Claims (5)

1. A method for recovering a clear image under an unmanned aerial vehicle fuzzy noise image is characterized by comprising the following steps:
s1, acquiring a blurred image of the drone, where Y is h x + N, and a vector of the blurred image is Y HX + N, and constructing a first optimization equation, where the first optimization equation is expressed as:
Figure FDA0002681607660000011
wherein | · | purple sweet1The L1 norm representing a · is,
Figure FDA0002681607660000012
euclidean distance of expression, X represents clear image, X represents matrix of clear image, axSparse representation representing sharp images, aHRepresenting a sparse representation of a blur kernel matrix, λ, β representing constraint coefficients of the norm L1, Y representing the captured marred image, Y representing the captured marred image matrix, H representing a blur kernel, H representing a blur kernel matrix, N representing Gaussian noise, N representing a Gaussian noise matrix, and x representing a convolution
S2, substituting the group sparse domain of the clear image and the group sparse domain of the fuzzy kernel matrix obtained by the dictionary learning method into the first optimization equation to obtain a second optimization equation, wherein the second optimization equation is expressed as:
Figure FDA0002681607660000013
wherein,
Figure FDA0002681607660000014
representing group sparse forms, DxGroup sparse domain, D, representing sharp imagesHA set sparse domain representing a fuzzy kernel matrix;
s3, substituting the second optimization equation into a separating Brazilian iteration SBI algorithm to obtain a fuzzy kernel matrix and obtain a restored image, wherein the method specifically comprises the following steps:
designing a two-step iteration method to solve the first fuzzy kernel estimation matrix, and solving the first clear image estimation matrix according to the solution of the first fuzzy kernel estimation matrix;
rewriting the first fuzzy kernel estimation matrix and the first clear image estimation matrix into a second fuzzy kernel estimation matrix and a second clear image estimation matrix which are suitable for an SBI algorithm;
introducing a problem equation, solving the problem equation by using an SBI algorithm, and solving sparse representation of a clear image through multiple iterations;
and obtaining error residuals of multiple iterations through a soft decision threshold in a component mode, solving to obtain sparse representation of a fuzzy kernel, and obtaining a restored image.
2. The method for restoring a sharp image under an unmanned aerial vehicle blurred noise image as claimed in claim 1, wherein the group of sparse domains is composed of similarity modules, the similarity modules are determined by euclidean distance, Lp norm is introduced as distance measure for increasing robustness of matching block selection, influence of noise is considered, and noise variance is introduced and expressed as:
Figure FDA0002681607660000021
wherein | · | purple sweetpLp norm of expression, p is a constant, x1iDenotes the ith target block, y1iDenotes the ith known block, M denotes the number of target blocks, and S denotes the standard deviation of the sample sequence.
3. The method for recovering the sharp image under the unmanned aerial vehicle blurred noise image as claimed in claim 1, wherein the two-step iteration method comprises the following steps:
the first step is as follows: solving for the sparse representation a of the sharp image according to the following sharp image equationx：
Figure FDA0002681607660000022
The second step is that: sparse representation a of sharp images from the first stepxObtaining a first sharp image estimation matrix
Figure FDA0002681607660000023
Estimating a matrix from the first sharp image
Figure FDA0002681607660000024
Solving the following fuzzy kernel equation:
Figure FDA0002681607660000025
wherein,
Figure FDA0002681607660000026
a first blur kernel estimation matrix is represented,
Figure FDA0002681607660000027
representing a first sharp image estimate matrix.
4. The method for recovering sharp images under unmanned aerial vehicle blurred noise images, according to claim 3, wherein the second blur kernel estimation matrix and the second sharp image estimation matrix are respectively:
second blur kernel estimation matrix:
Figure FDA0002681607660000028
Figure FDA0002681607660000029
second sharp image estimation matrix:
Figure FDA0002681607660000031
Figure FDA0002681607660000032
wherein,
Figure FDA0002681607660000033
to represent an estimate of image x.
5. The method for recovering the sharp image under the unmanned aerial vehicle blurred noise image as claimed in claim 3, wherein the problem equation is expressed as:
Figure FDA0002681607660000034
s.t.u＝Gv
wherein u ∈ RK、v∈RK，RKExpressing a real vector of K dimension, solving variables u and v in the problem equation through loop iteration based on an SBI algorithm, and writing extended Lagrangian functions of f (u) and g (v)A number, wherein the Lagrangian extension function of f (u) is:
Figure FDA0002681607660000035
the lagrange expansion function of g (v) is:
Figure FDA0002681607660000036
where μ denotes a penalty factor, b denotes an iterative error residue, u and v denote variables, f (u) denotes a norm equation for the variable u, G (v) denotes a norm equation for the variable v, and G denotes a relation between the variables u, v.
CN201710990638.8A 2017-10-23 2017-10-23 Method for recovering sharp image under unmanned aerial vehicle fuzzy noise image Active CN107730468B (en)

Priority Applications (1)

Application Number Priority Date Filing Date Title
CN201710990638.8A CN107730468B (en) 2017-10-23 2017-10-23 Method for recovering sharp image under unmanned aerial vehicle fuzzy noise image

Applications Claiming Priority (1)

Application Number Priority Date Filing Date Title
CN201710990638.8A CN107730468B (en) 2017-10-23 2017-10-23 Method for recovering sharp image under unmanned aerial vehicle fuzzy noise image

Publications (2)

Publication Number Publication Date
CN107730468A CN107730468A (en) 2018-02-23
CN107730468B true CN107730468B (en) 2020-11-10

Family

ID=61212387

Family Applications (1)

Application Number Title Priority Date Filing Date
CN201710990638.8A Active CN107730468B (en) 2017-10-23 2017-10-23 Method for recovering sharp image under unmanned aerial vehicle fuzzy noise image

Country Status (1)

Country Link
CN (1) CN107730468B (en)

Families Citing this family (4)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN108921813B (en) * 2018-05-16 2020-08-18 中设设计集团股份有限公司 Unmanned aerial vehicle detection bridge structure crack identification method based on machine vision
CN112381732B (en) * 2020-11-13 2023-09-05 广东工业大学 Image recovery method and system based on multi-scale random proximity algorithm
CN113379647B (en) * 2021-07-08 2024-07-02 湘潭大学 Multi-feature image restoration method for optimizing PSF estimation
CN114202459B (en) * 2021-10-26 2024-03-22 北京工业大学 Blind image super-resolution method based on depth priori

Citations (4)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN104134196A (en) * 2014-08-08 2014-11-05 重庆大学 Split Bregman weight iteration image blind restoration method based on non-convex higher-order total variation model
CN105469363A (en) * 2014-09-05 2016-04-06 华为技术有限公司 Method and device for generating sharp image based on blurred image
CN106204482A (en) * 2016-07-08 2016-12-07 桂林电子科技大学 Based on the mixed noise minimizing technology that weighting is sparse
CN106651974A (en) * 2016-11-03 2017-05-10 中南民族大学 Image compressive sensing reconstruction system and method utilizing weighted structural group sparse regulation

Family Cites Families (1)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
US9286662B2 (en) * 2013-09-26 2016-03-15 Siemens Aktiengesellschaft Single image super resolution and denoising using multiple wavelet domain sparsity

Patent Citations (4)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN104134196A (en) * 2014-08-08 2014-11-05 重庆大学 Split Bregman weight iteration image blind restoration method based on non-convex higher-order total variation model
CN105469363A (en) * 2014-09-05 2016-04-06 华为技术有限公司 Method and device for generating sharp image based on blurred image
CN106204482A (en) * 2016-07-08 2016-12-07 桂林电子科技大学 Based on the mixed noise minimizing technology that weighting is sparse
CN106651974A (en) * 2016-11-03 2017-05-10 中南民族大学 Image compressive sensing reconstruction system and method utilizing weighted structural group sparse regulation

Non-Patent Citations (6)

* Cited by examiner, † Cited by third party
Title
Group-Based Sparse Representation for Image Restoration;Jian Zhang 等;《IEEE TRANSACTIONS ON IMAGE PROCESSING》;20140831;第23卷(第8期);3336-3351 *
Image Deblurring with Blur Kernel Estimation in RGB Channels;Xianqiu Xu 等;《2016 IEEE International Conference on Digital Signal Processing (DSP)》;20161018;681-684 *
SPARSE REPRESENTATION BASED BLIND IMAGE DEBLURRING;Haichao Zhang等;《2011 IEEE International Conference on Multimedia and Expo》;20110906;第2节 *
一种改进的组稀疏表示图像去噪方法;马杰 等;《微电子学与计算机》;20170630;第34卷(第6期);99-103 *
基于稀疏表示模型的图像复原技术研究;张健;《中国博士学位论文全文数据库信息科技辑》;20141215;第2014年卷(第12期);第1.3节，第2.2.3节，第4.3-4.5节 *
张健.基于稀疏表示模型的图像复原技术研究.《中国博士学位论文全文数据库信息科技辑》.2014,第2014年卷(第12期), *

Also Published As

Publication number Publication date
CN107730468A (en) 2018-02-23

Similar Documents

Publication Publication Date Title
CN107730468B (en) Method for recovering sharp image under unmanned aerial vehicle fuzzy noise image
Gopalan et al. A blur-robust descriptor with applications to face recognition
CN110675347B (en) Image blind restoration method based on group sparse representation
CN110782399A (en) Image deblurring method based on multitask CNN
Wen et al. Joint adaptive sparsity and low-rankness on the fly: An online tensor reconstruction scheme for video denoising
CN111028163A (en) Convolution neural network-based combined image denoising and weak light enhancement method
CN112164011B (en) Motion image deblurring method based on self-adaptive residual error and recursive cross attention
CN111028177A (en) Edge-based deep learning image motion blur removing method
CN107451961B (en) Method for recovering sharp image under multiple fuzzy noise images
CN114331886B (en) Image deblurring method based on depth features
CN113763268B (en) Blind restoration method and system for face image
CN115345791A (en) Infrared image deblurring algorithm based on attention mechanism residual error network model
CN106033595B (en) Image blind deblurring method based on local constraint
CN115272113A (en) Image deblurring method based on multi-scale frequency separation network
CN114764750B (en) Image denoising method based on self-adaptive consistency priori depth network
CN113902647A (en) Image deblurring method based on double closed-loop network
Li et al. Distribution-transformed network for impulse noise removal
CN110266318B (en) Measurement matrix optimization method based on gradient projection algorithm in compressed sensing signal reconstruction
Nawaz et al. Image super resolution by sparse linear regression and iterative back projection
Kim et al. MetaWeather: Few-Shot Weather-Degraded Image Restoration via Degradation Pattern Matching
Velayudhan et al. Two-phase approach for recovering images corrupted by Gaussian-plus-impulse noise
CN110717873A (en) Traffic sign deblurring detection recognition algorithm based on multi-scale residual error
David et al. Foreground/background segmentation with learned dictionary
Revanth et al. Non-Homogeneous Haze Image Formation Model Based Single Image Dehazing
CN113962899B (en) Fundus image processing method, fundus image processing device, electronic apparatus, and storage medium

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