CN108765325A - A kind of small drone Restoration method of blurred image - Google Patents
A kind of small drone Restoration method of blurred image Download PDFInfo
- Publication number
- CN108765325A CN108765325A CN201810471510.5A CN201810471510A CN108765325A CN 108765325 A CN108765325 A CN 108765325A CN 201810471510 A CN201810471510 A CN 201810471510A CN 108765325 A CN108765325 A CN 108765325A
- Authority
- CN
- China
- Prior art keywords
- image
- restored
- obtains
- fuzzy
- small drone
- 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
- 238000000034 method Methods 0.000 title claims abstract description 53
- 238000001228 spectrum Methods 0.000 claims abstract description 15
- 238000001914 filtration Methods 0.000 claims abstract description 13
- 238000002156 mixing Methods 0.000 claims abstract description 11
- 230000009466 transformation Effects 0.000 claims abstract description 10
- 238000013527 convolutional neural network Methods 0.000 claims abstract description 8
- 238000011084 recovery Methods 0.000 claims description 11
- 238000000605 extraction Methods 0.000 claims description 10
- 238000010586 diagram Methods 0.000 claims description 8
- 230000008569 process Effects 0.000 claims description 8
- 238000003708 edge detection Methods 0.000 claims description 7
- 230000035939 shock Effects 0.000 claims description 5
- 238000002834 transmittance Methods 0.000 claims description 5
- 238000006243 chemical reaction Methods 0.000 claims description 3
- 230000006872 improvement Effects 0.000 claims description 3
- 230000000877 morphologic effect Effects 0.000 claims description 3
- 230000011218 segmentation Effects 0.000 claims description 3
- 230000007704 transition Effects 0.000 claims description 3
- 230000003014 reinforcing effect Effects 0.000 abstract description 2
- 230000000694 effects Effects 0.000 description 10
- 238000005516 engineering process Methods 0.000 description 4
- 238000003384 imaging method Methods 0.000 description 4
- 238000012545 processing Methods 0.000 description 3
- 230000007850 degeneration Effects 0.000 description 2
- 239000011159 matrix material Substances 0.000 description 2
- 239000003595 mist Substances 0.000 description 2
- 230000003595 spectral effect Effects 0.000 description 2
- 238000013459 approach Methods 0.000 description 1
- 230000000295 complement effect Effects 0.000 description 1
- 230000008878 coupling Effects 0.000 description 1
- 238000010168 coupling process Methods 0.000 description 1
- 238000005859 coupling reaction Methods 0.000 description 1
- 230000006866 deterioration Effects 0.000 description 1
- 238000011156 evaluation Methods 0.000 description 1
- 230000036039 immunity Effects 0.000 description 1
- 230000007246 mechanism Effects 0.000 description 1
- 239000000203 mixture Substances 0.000 description 1
- 230000004048 modification Effects 0.000 description 1
- 238000012986 modification Methods 0.000 description 1
- 238000005457 optimization Methods 0.000 description 1
- 238000012805 post-processing Methods 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
- 238000010183 spectrum analysis Methods 0.000 description 1
- 239000004575 stone Substances 0.000 description 1
- 238000012546 transfer Methods 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
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T2207/00—Indexing scheme for image analysis or image enhancement
- G06T2207/10—Image acquisition modality
- G06T2207/10004—Still image; Photographic image
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T2207/00—Indexing scheme for image analysis or image enhancement
- G06T2207/20—Special algorithmic details
- G06T2207/20048—Transform domain processing
- G06T2207/20061—Hough transform
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T2207/00—Indexing scheme for image analysis or image enhancement
- G06T2207/20—Special algorithmic details
- G06T2207/20084—Artificial neural networks [ANN]
Landscapes
- Physics & Mathematics (AREA)
- General Physics & Mathematics (AREA)
- Engineering & Computer Science (AREA)
- Theoretical Computer Science (AREA)
- Image Processing (AREA)
- Image Analysis (AREA)
Abstract
The invention discloses a kind of small drone Restoration method of blurred image, are related to technical field of computer vision.The practicability and robustness for focusing on reinforcing small drone blur image restoration of this method.The present invention comprises the steps of:(1) it by small drone image, is identified using the unmanned plane image vague category identifier recognizer based on convolutional neural networks, obtains vague category identifier;(2) it by the image that vague category identifier is motion blur, is restored using the blind restoration algorithm of the motion blur image of mixed characteristic regularization constraint, obtains clear restored image;(3) it by the image that vague category identifier is atmosphere fuzzy, is restored using the image defogging algorithm based on mixing priori and guiding filtering, obtains clear restored image;(4) it by the image that vague category identifier is defocus blur, is restored with the blind restoration algorithm of defocus blur for improving Hough transformation using based on frequency spectrum pretreatment, obtains clear restored image.
Description
Technical field
The present invention relates to technical field of computer vision, particularly relate to a kind of small drone Restoration method of blurred image.
Background technology
In recent years, small drone becomes near-earth remote sensing information with high maneuverability, high performance-price ratio and lower operation difficulty
The important way of acquisition, how each field tool such as photography, near-earth exploration, intelligent transportation, fire-fighting anti-terrorism and military surveillance in low latitude
Have wide practical use.Small drone is vulnerable to bad weather, holder shake, relative motion, imaging in imaging process
The factors such as the system failure influence, and cause image is fuzzy to degrade, and atmosphere fuzzy, motion blur and defocus blur are its common three kinds
Vague category identifier.Image obscures the timely accurate resolution held with policymaker of direct interference information, therefore is gone to image
Fuzzy Processing becomes the key for improving small drone acquisition of information quality.Blur image restoration be according to image deterioration mechanism,
Image degeneration physical model is established using known priori conditions, targetedly restores the process of clear image.It is mainly solved
Certainly vague category identifier identification and all types of blur image restoration problems.Traditional deblurring method based on image enhancement is compared, is obscured
Image recovery method specific aim is stronger, and deblurring effect is more preferable, and information preservation is more complete.In conclusion research is a kind of small-sized
UAV Fuzzy image recovery method has important practical significance.
In terms of blur image restoration, scholar does a lot of work:Stone jewel et al. is proposed for movement imaging hybrid guided mode
The full Variational Image Restoration method of paste.It is first depending on the spectral characteristic qualitative recognition vague category identifier of blurred picture, then using falling
The point spread function of spectrum analysis standard measure ambiguous estimation model, is eventually adding Coupling Gradient fidelity term, improves full Variational Restoration and calculates
Method restores blurred picture.This method preferably solves the problems, such as the movement blind recovery of mixed image, but Fuzzy Processing type is excessively limited to,
Application range is smaller.Xu Zongqi proposes a kind of blind restoration processing method of usable image.Blurred picture is divided first with Cepstrum Method
For movement, defocus and other fuzzy three classes, then it is restored using improved smoothness constraint biregular blindly restoring image algorithm
Its fuzzy graph, using parametric method restoring movement and defocus blur image, finally by ringing effect post-processing algorithm Optimization restoration
As a result, obtaining restoring clear image.This method application range is wider, but vague category identifier accuracy of identification is poor.Han little Fang et al. is carried
Go out a kind of move and defocus blur image restored method.Blurred picture is pre-processed first to obtain log spectrum binary map, to it
Hough transformation is carried out, judges image vague category identifier by comparing the bright spot number in transformation matrix, is then directed to motion blur image,
Using directional differential ambiguous estimation direction twice, blurred picture is obtained followed by improved Prewiit operators and the Fermi function
Edge function, finally utilize Wiener filtering algorithm, restore to obtain clear image in conjunction with modulation transfer function.This method obscures class
Type accuracy of identification is higher, but recovery effect is poor, and ringing is apparent, and practicability is not strong.The boundless and indistinct proposition remote sensing images of Cui Guang
Increased quality and evaluation method.The remote sensing image that analyzes of system first is imaged link model, then respectively according to imaging
Various degeneration factors in link propose restored method, finally assess restored image quality.This method is highly practical,
But it is not particularly suited for unmanned plane image procossing.Chou Xiang proposes the restored method for unmanned aerial vehicle remote sensing image.By conclusion nobody
The common vague category identifier of machine image proposes to be utilized respectively the blind restoration algorithm based on L0 sparse priors, eliminates the full of camera exceptional value
Unmanned aerial vehicle remote sensing Atmospheric Degraded Image restoration algorithm with the blind restoration algorithm of blurred picture, based on Multiple Scattering APSF estimations is multiple
Original, targetedly restoring movement is fuzzy, exceptional value is interfered, atmosphere fuzzy image.This method specific aim is stronger, but recovery effect
Need to be further enhanced, and do not consider that vague category identifier identifies, the type identification difficulty that degrades is larger.
Invention content
In view of this, it is an object of the invention to propose a kind of small drone Restoration method of blurred image, this method energy
Enough simplify blur image restoration process, improves image restoration quality.
Based on above-mentioned purpose, technical solution provided by the invention is:
A kind of small drone Restoration method of blurred image is applied to small drone image, and this method includes following
Step:
Step 1:By small drone blurred picture, the unmanned plane image vague category identifier based on convolutional neural networks is utilized
Recognizer is identified, and obtains vague category identifier, including motion blur image, atmosphere fuzzy image and defocus blur image,
Motion blur image is executed into step 2, atmosphere fuzzy image executes step 3, and defocus blur image executes step 4;
Step 2:By the image that vague category identifier is motion blur, the motion blur figure of mixed characteristic regularization constraint is utilized
As blind restoration algorithm is restored, clear restored image is obtained;
Step 3:By the image that vague category identifier is atmosphere fuzzy, gone using the image based on mixing priori and guiding filtering
Mist algorithm is restored, and clear restored image is obtained;
Step 4:By the image that vague category identifier is defocus blur, pre-processes using based on frequency spectrum and improve Hough transformation
The blind restoration algorithm of defocus blur is restored, and clear restored image is obtained;
Complete the recovery of small drone blurred picture.
Wherein, step 1 specifically includes following steps:
(101) small drone blurred picture is utilized into Fast Fourier Transform (FFT), obtains the spectrogram of blurred picture;
(102) spectrogram of blurred picture is subjected to feature extraction using the convolutional neural networks model trained, obtained
Characteristic pattern;
(103) characteristic pattern is inputted into grader, obtains vague category identifier.
Wherein, step 2 specifically includes following steps:
(201) motion blur image is utilized into the edge detection algorithm and impact filtering that phase equalization is drilled based on conformal list
Device carries out edge extraction and sharpens, the edge details sharpened;
(202) building fuzzy nuclear model is:
Wherein, L is clear image, and B is motion blur image,For the single order local derviation of motion blur image, k is fuzzy
Core,For the edge details of sharpening, | | | |1With | | | |2L is indicated respectively1、L2Norm,Table respectively
Show the single order local derviation and second order local derviation of fuzzy core,For sparse regular terms,For smooth regular terms, λk1、
λk2、λk3The parameter of respectively each regular terms;
(203) final obscure is obtained to fuzzy core amendment according to the pixel nonnegativity of fuzzy core and conservation of energy characteristic
Nuclear model:
∫ kdxdy=1
Wherein, (i, j) be fuzzy core coordinate, max (k (:)) it is fuzzy core max pixel value, μ is threshold coefficient, ∫ kdxdy
=1 indicates that fuzzy core is normalized;
(204) structure restored image model is:
Wherein,For super Laplace prior item,To protect side regular terms, SF (L*)
For the clear image Jing Guo Shock filter process;
(205) non-negative characteristic revision is carried out to restored image, obtains final restored image model:
(206) linear multi-Scale Pyramid is built to small drone motion blur image, half is utilized in each layers of resolution
The variable division strategy of quadratic form optimizes solution to final fuzzy nuclear model and restored image model, until meeting iteration
Number obtains the fuzzy core of each layers of resolution and the solution of restored image, according to the fuzzy core of each layers of resolution and restored image
Solution obtains the clear restored image of motion blur image.
Wherein, step 3 specifically includes following steps:
(301) the dark channel diagram I of atmosphere fuzzy image is soughtdark(x) and depth map dr(x),
Wherein, IcA Color Channel for being atmosphere fuzzy image I in RGB color, Ω (x) are centered on x
A region, IvAnd IsThe respectively luminance channel in hsv color space and saturation degree channel, ε are the depth that stochastic variable represents
Spend image random error, θ0、θ1、θ2For linear coefficient;
(302) pixel region is redefined using the method for double constraint segmentations, marks off four
Pixel region is described in detail below:
Wherein, I1For high brightness thick fog region, I2For non-high brightness thick fog region, I3For the non-thick fog region of high brightness, I4For
The non-non- thick fog region of high brightness, α and β are region division threshold value;
(303) the pixel coordinate point position for extracting in dark channel diagram and depth map before brightness 0.1% respectively, then compares two
The pixel coordinate point position that width image zooming-out arrives retains if pixel coordinate point position exists simultaneously in two width figures, otherwise picks
It removes, the value with maximum brightness corresponding to the pixel coordinate point remained is finally found in atmosphere fuzzy image as air
Light value A;
(304) coarse gas transmittance figure is built using mixing prior model:
T (x)=mtdark(x)+ntcolor(x)
N=1-m
Wherein, tdarkFor the thick transmissivity that dark primary priori obtains, tcolorDecay the obtained thick transmissivity of priori for color,
M and n is mixing priori coefficient,
tcolor(x, y)=e-ηd(x,y)
Wherein, ω is fidelity coefficient, and η is atmospheric scattering coefficient;
(305) coarse gas transmittance figure is optimized using guiding filtering algorithm, obtains smart atmospheric transmissivity figure t'(x,
y);
(306) air light value and smart atmospheric transmissivity figure are substituted into atmospherical scattering model, obtains the clear of atmosphere fuzzy image
Clear restored image:
Wherein, t0And t1For the parameter of introducing.
Wherein, step 4 specifically includes following steps:
(401) defocus blur image is utilized into fast two-dimensional Fourier transformation, obtains the spectrogram of defocus blur image;
(402) pixel transition region is estimated according to the gray value curve of spectrogram, and by 45 °, 135 °, -45 °, -135v four
The transitional region pixel mean value computation binary-state threshold in a direction;
(403) binary conversion treatment is carried out to spectrogram and morphologic filtering obtains frequency spectrum binary map;
(404) marginal information for utilizing Canny edge detection algorithms extraction frequency spectrum binary map, obtains the side of frequency spectrum binary map
Edge details;
(405) all marginal points are calculated to the distance of central point, are stored in set D;
(406) using central point as the center of circle, justified as radius work using any distance chosen in D, obtain candidate circle;
(407) judge whether the quantity of the marginal point on candidate circle is more than threshold value, if being then that zero is justified, corresponding to reservation
Radius;
(408) circulation step (406) is pre-set to step (407) until the number of traversal set D or zero circle reach
Maximum value;
(409) the two neighboring zero detected is justified, the mould of defocus blur image is estimated using adjacent zero ratio method
Paste core;
(410) it utilizes the restored image model in step (204) to be iterated recovery to defocus blur image and obtains defocus
The clear restored image of blurred picture.
The present invention is compared to the advantages of background technology:
The practicability and robustness for focusing on reinforcing small drone blur image restoration of this method.Experimental result table
Bright, the unmanned plane image vague category identifier recognizer based on convolutional neural networks compares traditional recognition method accuracy of identification higher,
The blind restoration algorithm of motion blur image of mixed characteristic regularization constraint to small drone motion blur image restoration effect compared with
Good, the image defogging algorithm based on mixing priori and guiding filtering is preferable to small drone atmosphere fuzzy image restoration effect,
It is pre-processed based on frequency spectrum and small drone defocus blur image is restored with the defocus blur blind restoration algorithm for improving Hough transformation
Effect is preferable, the strong robustness of small drone Restoration method of blurred image, and has preferable practicability.Small drone mould
Paste image recovery method can realize practicability and the stronger blur image restoration of robustness, be changed to the important of the prior art
Into.
Description of the drawings
In order to more clearly explain the embodiment of the invention or the technical proposal in the existing technology, to embodiment or will show below
There is attached drawing needed in technology description to be briefly described, it should be apparent that, the accompanying drawings in the following description is only this
Some embodiments of invention for those of ordinary skill in the art without creative efforts, can be with
Obtain other attached drawings according to these attached drawings.
Fig. 1 is a method flow diagram of the embodiment of the present invention.
Specific implementation mode
To make the objectives, technical solutions, and advantages of the present invention clearer, below in conjunction with specific embodiment, and reference
Attached drawing, the present invention is described in further detail.
Present embodiment elaborates the principle of small drone image restoration, according to the identification of image vague category identifier, movement mould
Paste restore, atmosphere fuzzy restore, defocus blur restore thinking calculated, emphatically to small drone image restoration flow into
Row Optimal improvements.It is as follows:
Step 1:By small drone blurred picture, the unmanned plane image vague category identifier based on convolutional neural networks is utilized
Recognizer is identified, and obtains vague category identifier, including motion blur image, atmosphere fuzzy image and defocus blur image,
Motion blur image is executed into step 2, atmosphere fuzzy image executes step 3, and defocus blur image executes step 4;
(101) small drone blurred picture is utilized into Fast Fourier Transform (FFT), obtains the spectrogram of blurred picture;
(102) spectrogram of blurred picture is subjected to feature extraction using the convolutional neural networks model trained, obtained
Characteristic pattern;
(103) characteristic pattern is inputted into Softmax graders, obtains blurred picture type.
Step 2:By the image that vague category identifier is motion blur, the motion blur figure of mixed characteristic regularization constraint is utilized
As blind restoration algorithm is restored, clear restored image is obtained;
(201) small drone motion blur image is utilized into the edge detection algorithm that phase equalization is drilled based on conformal list
It carries out edge extraction with shock filter and sharpens, the edge details sharpened
(202) building fuzzy nuclear model is:
Wherein, L is clear image, and B is motion blur image,For the single order local derviation of motion blur image, k is fuzzy
Core, | | | |1With | | | |2L is indicated respectively1、L2Norm,The single order local derviation and Second Order Partial of fuzzy core are indicated respectively
It leads,For sparse regular terms,For smooth regular terms, λk1、λk2、λk3The parameter of respectively each regularization term;
(203) final obscure is obtained to fuzzy core amendment according to the pixel nonnegativity of fuzzy core and conservation of energy characteristic
Nuclear model:
∫ kdxdy=1
Wherein, (i, j) be fuzzy core coordinate, max (k (:)) it is fuzzy core max pixel value, μ is threshold coefficient, ∫ kdxdy
=1 indicates that fuzzy core is normalized;
(204) structure restored image model is:
Wherein,For super Laplace prior item,To protect side regular terms, SF (L*)
For the clear image Jing Guo Shock filter process;
(205) non-negative characteristic revision is carried out to restored image, obtains final restored image model:
(206) linear multi-Scale Pyramid is built to small drone motion blur image, half is utilized in each layers of resolution
The variable division strategy of quadratic form optimizes solution to final fuzzy nuclear model and restored image model, until meeting iteration
Number obtains the fuzzy core of each layers of resolution and the solution of restored image, according to the fuzzy core of each layers of resolution and restored image
Solution obtains the clear restored image of motion blur image.
Strategy is sharpened compared to other Edge extractions, the edge detection algorithm Shandong of phase equalization is drilled based on conformal list
Stick is stronger, and speed faster, and is more suitable for obscuring the fidelity term estimation in nuclear model.L1 norms are the optimal convex close of L0 norms
Seemingly, the matrix of L0, L1 norm specification all has good sparse effect, but since the operation of L0 norms is complicated, uses L1 norms
As the Sparse rules operator of fuzzy nuclear model, algorithm arithmetic speed is improved.L2 norms compare L1 norms and can effectively improve mould
The generalization ability of type prevents model over-fitting.In conjunction with image single order, second order local derviation characteristic, fuzzy core image single order, second order are used
The L2 norms composition polyhybird regularization term of local derviation carries out smoothness constraint, is further suppressed while improving smoothness constraint effect
Exceptional value in fuzzy core.Compared with the biregularization of the sparse smoothness properties proposed in other algorithms obscures nuclear model, herein
The smooth and fidelity regular terms for improving fuzzy nuclear model enhances the noise immunity of fuzzy core while ensureing accurate estimation
Energy.
Compared with Gaussian Profile, laplacian distribution, surpass Laplce's fitting of distribution best results, so in restored image
Model estimation stages construct image regular terms using super Laplace prior, and the edge details for the restored image that can make are more
It is abundant.The blurring effect occurred in total variation model can preferably be compensated by protecting side regular terms, effectively overcome existing method
Present in cannot restore high quality clear image problem.Guarantor's side regular terms is smaller, shows that the restored image edge generated is got over
Close to the clear image edge sharpened.
Step 3:By the image that vague category identifier is atmosphere fuzzy, gone using the image based on mixing priori and guiding filtering
Mist algorithm is restored, and clear restored image is obtained;
(301) the dark channel diagram I of small drone atmosphere fuzzy image is soughtdark(x) and depth map dr(x),
Wherein, IcA Color Channel for being blurred picture I in RGB color, Ω (x) are one centered on x
A region, IvAnd IsThe respectively luminance channel in hsv color space and saturation degree channel, ε are the depth map that stochastic variable represents
As random error, θ0、θ1、θ2For linear coefficient;
(302) pixel region is redefined using the method for double constraint segmentations, marks off four
Pixel region is described in detail below:
Wherein, I1For high brightness thick fog region, I2For non-high brightness thick fog region, I3The non-thick fog region of high brightness, I4Non- height
The non-thick fog region of brightness, α and β are region division threshold value;
(303) a kind of new air light value method of estimation is proposed.Extract brightness in dark channel diagram and depth map respectively first
Then preceding 0.1% location of pixels compares the coordinate points position that two images are extracted, if coordinate points position exists simultaneously two width
Then retain in figure, it is on the contrary then reject, it is right that the coordinate points institute remained is finally found in small drone atmosphere fuzzy image
There should be the value of maximum brightness as air light value A;
(304) it proposes that a kind of mixing priori strategy estimates atmospheric transmissivity, is built using mixing prior model thick
Atmospheric transmissivity figure:
T (x)=mtdark(x)+ntcolor(x)
N=1-m
Wherein, tdarkFor the thick transmissivity that dark primary priori obtains, tcolorDecay the obtained thick transmissivity of priori for color,
M and n is mixing priori coefficient.
tcolor(x, y)=e-ηd(x,y)
Wherein, ω is fidelity coefficient, and η is atmospheric scattering coefficient;
(305) coarse gas transmittance figure is optimized using guiding filtering algorithm, obtains smart atmospheric transmissivity figure t'(x,
y);
(306) air light value and smart atmospheric transmissivity figure are substituted into atmospherical scattering model, obtains atmosphere fuzzy and clearly restores
Image:
Wherein, in order to avoid t is excessive or too small caused image fault, parameter t is introduced0And t1It is limited.
Dark primary priori is not suitable for highlight regions, but compares other first checking methods, preferable to the processing in thick fog region,
And color decaying priori is not suitable for thick fog region, but can preferably solve the problems, such as that highlight regions restore distortion, it is complementary compared with
By force;Dark primary priori and the priori characteristic of color decaying priori are more similar, and algorithm realization approach is essentially identical, it can be achieved that property
It is higher.
The size of dark channel value can in approximate description image fog concentration size, the concentration of value more dense fog is bigger;It is bright
Degree channel value can directly indicate the brightness of image, and value is bigger, and brightness is bigger;It is formed using dark channel diagram and luminance channel figure
Double constraintss divide the highlight regions of image and thick fog region, have certain theoretical foundation.
Step 4:By the image that vague category identifier is defocus blur, pre-processes using based on frequency spectrum and improve Hough transformation
The blind restoration algorithm of defocus blur is restored, and clear restored image is obtained;
(401) small drone defocus blur image is utilized into fast two-dimensional Fourier transformation, obtains the frequency of blurred picture
Spectrogram;
(402) pixel transition region is estimated according to the gray value curve of spectrogram, and by 45 °, 135 °, -45 °, -135 ° four
The transitional region pixel mean value computation binary-state threshold in a direction;
(403) binary conversion treatment is carried out to spectral image and morphologic filtering obtains frequency spectrum binary map;
(404) marginal information for utilizing Canny edge detection algorithms extraction frequency spectrum binary map, obtains the side of frequency spectrum binary map
Edge details;
(405) all marginal points are calculated to the distance of central point, are stored in set D;
(406) using central point as the center of circle, justified as radius work using any distance chosen in D, obtain candidate circle;
(407) judge whether the quantity of the marginal point on candidate circle is more than threshold value, if being then that zero is justified, corresponding to reservation
Radius;
(408) circulation step (406) is pre-set to step (407) until the number of traversal set D or zero circle reach
Maximum value;
(409) the two neighboring zero detected is justified, the mould of defocus blur image is estimated using adjacent zero ratio method
Paste core;
(410) it utilizes the restored image model in step (204) to be iterated recovery to defocus blur image and obtains defocus
The clear restored image of blurred picture.
Complete small drone blur image restoration.
Those of ordinary skills in the art should understand that:The discussion of any of the above embodiment is exemplary only, not
It is intended to imply that the scope of the present disclosure (including claim) is limited to these examples.All within the spirits and principles of the present invention,
Any omission made to the above embodiment, modification, equivalent replacement, improvement etc., should be included in protection scope of the present invention it
It is interior.
Claims (5)
1. a kind of small drone Restoration method of blurred image is applied to small drone image, which is characterized in that this method
Include the following steps:
Step 1:By small drone blurred picture, identified using the unmanned plane image vague category identifier based on convolutional neural networks
Algorithm is identified, and obtains vague category identifier, including motion blur image, atmosphere fuzzy image and defocus blur image, will transport
Dynamic blurred picture executes step 2, and atmosphere fuzzy image executes step 3, and defocus blur image executes step 4;
Step 2:By the image that vague category identifier is motion blur, the motion blur image using mixed characteristic regularization constraint is blind
Restoration algorithm is restored, and clear restored image is obtained;
Step 3:By the image that vague category identifier is atmosphere fuzzy, calculated using the image defogging based on mixing priori and guiding filtering
Method is restored, and clear restored image is obtained;
Step 4:By the image that vague category identifier is defocus blur, the defocus pre-processed based on frequency spectrum with improvement Hough transformation is utilized
It obscures blind restoration algorithm to be restored, obtains clear restored image;
Complete the recovery of small drone blurred picture.
2. small drone Restoration method of blurred image according to claim 1, which is characterized in that step 1 specifically includes
Following steps:
(101) small drone blurred picture is utilized into Fast Fourier Transform (FFT), obtains the spectrogram of blurred picture;
(102) spectrogram of blurred picture is subjected to feature extraction using the convolutional neural networks model trained, obtains feature
Figure;
(103) characteristic pattern is inputted into grader, obtains vague category identifier.
3. small drone Restoration method of blurred image according to claim 1, which is characterized in that step 2 specifically includes
Following steps:
(201) by motion blur image utilize based on conformal list drill phase equalization edge detection algorithm and shock filter into
Row edge extraction is with sharpening, the edge details sharpened;
(202) building fuzzy nuclear model is:
Wherein, L is clear image, and B is motion blur image, and ▽ B are the single order local derviation of motion blur image, and k is fuzzy core, ▽
SF(LCMPS) it is the edge details sharpened, | | | |1With | | | |2L is indicated respectively1、L2Norm, ▽ k, ▽2K indicates fuzzy core respectively
Single order local derviation and second order local derviation,For sparse regular terms,For smooth regular terms, λk1、λk2、λk3Respectively
The parameter of each regular terms;
(203) final fuzzy core mould is obtained to fuzzy core amendment according to the pixel nonnegativity of fuzzy core and conservation of energy characteristic
Type:
∫ kdxdy=1
Wherein, (i, j) be fuzzy core coordinate, max (k (:)) it is fuzzy core max pixel value, μ is threshold coefficient, ∫ kdxdy=1
Fuzzy core is normalized in expression;
(204) structure restored image model is:
Wherein,For super Laplace prior item,To protect side regular terms, SF (L*) it is warp
Cross the clear image of Shock filter process;
(205) non-negative characteristic revision is carried out to restored image, obtains final restored image model:
(206) linear multi-Scale Pyramid is built to small drone motion blur image, it is secondary using half in each layers of resolution
The variable division strategy of type optimizes solution to final fuzzy nuclear model and restored image model, until meeting iteration time
Number, obtains the fuzzy core of each layers of resolution and the solution of restored image, according to the solution of the fuzzy core of each layers of resolution and restored image
Obtain the clear restored image of motion blur image.
4. small drone Restoration method of blurred image according to claim 1, which is characterized in that step 3 specifically includes
Following steps:
(301) the dark channel diagram I of atmosphere fuzzy image is soughtdark(x) and depth map dr(x),
Wherein, IcA Color Channel for being atmosphere fuzzy image I in RGB color, Ω (x) are one centered on x
Region, IvAnd IsThe respectively luminance channel in hsv color space and saturation degree channel, ε are the depth image that stochastic variable represents
Random error, θ0、θ1、θ2For linear coefficient;
(302) pixel region is redefined using the method for double constraint segmentations, marks off four pixels
Region is described in detail below:
Wherein, I1For high brightness thick fog region, I2For non-high brightness thick fog region, I3For the non-thick fog region of high brightness, I4For non-height
The non-thick fog region of brightness, α and β are region division threshold value;
(303) the pixel coordinate point position for extracting in dark channel diagram and depth map before brightness 0.1% respectively, then compares two width figures
As the pixel coordinate point position extracted, retain if pixel coordinate point position exists simultaneously in two width figures, otherwise reject, most
The value with maximum brightness corresponding to the pixel coordinate point remained is found in atmosphere fuzzy image afterwards as air light value
A;
(304) coarse gas transmittance figure is built using mixing prior model:
T (x)=mtdark(x)+ntcolor(x)
N=1-m
Wherein, tdarkFor the thick transmissivity that dark primary priori obtains, tcolorFor the thick transmissivity that color decaying priori obtains, m and n
To mix priori coefficient,
tcolor(x, y)=e-ηd(x,y)
Wherein, ω is fidelity coefficient, and η is atmospheric scattering coefficient;
(305) coarse gas transmittance figure is optimized using guiding filtering algorithm, obtains smart atmospheric transmissivity figure t'(x, y);
(306) air light value and smart atmospheric transmissivity figure are substituted into atmospherical scattering model, obtains the clear multiple of atmosphere fuzzy image
Original image:
Wherein, t0And t1For the parameter of introducing.
5. small drone Restoration method of blurred image according to claim 3, it is characterised in that:Step 4 specifically includes
Following steps:
(401) defocus blur image is utilized into fast two-dimensional Fourier transformation, obtains the spectrogram of defocus blur image;
(402) pixel transition region is estimated according to the gray value curve of spectrogram, and by 45 °, 135 °, -45 °, -135 ° of four sides
To transitional region pixel mean value computation binary-state threshold;
(403) binary conversion treatment is carried out to spectrogram and morphologic filtering obtains frequency spectrum binary map;
(404) utilize the marginal information of Canny edge detection algorithms extraction frequency spectrum binary map, the edge for obtaining frequency spectrum binary map thin
Section;
(405) all marginal points are calculated to the distance of central point, are stored in set D;
(406) using central point as the center of circle, justified as radius work using any distance chosen in D, obtain candidate circle;
(407) judge whether the quantity of the marginal point on candidate circle is more than threshold value, if being then that zero is justified, retain corresponding half
Diameter;
(408) circulation step (406) is to step (407), until number that traversal set D or zero are justified reach it is pre-set most
Big value;
(409) the two neighboring zero detected is justified, the fuzzy core of defocus blur image is estimated using adjacent zero ratio method;
(410) it utilizes the restored image model in step (204) to be iterated recovery to defocus blur image and obtains defocus blur
The clear restored image of image.
Priority Applications (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201810471510.5A CN108765325B (en) | 2018-05-17 | 2018-05-17 | Small unmanned aerial vehicle blurred image restoration method |
Applications Claiming Priority (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201810471510.5A CN108765325B (en) | 2018-05-17 | 2018-05-17 | Small unmanned aerial vehicle blurred image restoration method |
Publications (2)
Publication Number | Publication Date |
---|---|
CN108765325A true CN108765325A (en) | 2018-11-06 |
CN108765325B CN108765325B (en) | 2021-06-29 |
Family
ID=64006814
Family Applications (1)
Application Number | Title | Priority Date | Filing Date |
---|---|---|---|
CN201810471510.5A Active CN108765325B (en) | 2018-05-17 | 2018-05-17 | Small unmanned aerial vehicle blurred image restoration method |
Country Status (1)
Country | Link |
---|---|
CN (1) | CN108765325B (en) |
Cited By (15)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN109521556A (en) * | 2018-12-07 | 2019-03-26 | 歌尔科技有限公司 | A kind of electron microscopic wearable device |
CN109584186A (en) * | 2018-12-25 | 2019-04-05 | 西北工业大学 | A kind of unmanned aerial vehicle onboard image defogging method and device |
CN110097521A (en) * | 2019-05-08 | 2019-08-06 | 华南理工大学 | A kind of convolutional neural networks image recovery method towards reflecting metal vision-based detection |
CN110288550A (en) * | 2019-06-28 | 2019-09-27 | 中国人民解放***箭军工程大学 | The single image defogging method of confrontation network is generated based on priori knowledge guiding conditions |
CN110400312A (en) * | 2019-07-31 | 2019-11-01 | 北京金山云网络技术有限公司 | Determine the method, apparatus and server of image vague category identifier |
CN110648291A (en) * | 2019-09-10 | 2020-01-03 | 武汉科技大学 | Unmanned aerial vehicle motion blurred image restoration method based on deep learning |
CN110676753A (en) * | 2019-10-14 | 2020-01-10 | 宁夏百川电力股份有限公司 | Intelligent inspection robot for power transmission line |
CN110874826A (en) * | 2019-11-18 | 2020-03-10 | 北京邮电大学 | Workpiece image defogging method and device applied to ion beam precise film coating |
CN110895141A (en) * | 2019-11-28 | 2020-03-20 | 梁彦云 | Residential space crowding degree analysis platform |
CN111717406A (en) * | 2020-06-17 | 2020-09-29 | 中国人民解放军陆军工程大学 | Unmanned aerial vehicle image acquisition system |
CN111881982A (en) * | 2020-07-30 | 2020-11-03 | 北京环境特性研究所 | Unmanned aerial vehicle target identification method |
CN112465777A (en) * | 2020-11-26 | 2021-03-09 | 华能通辽风力发电有限公司 | Fan blade surface defect identification technology based on video stream |
CN113191982A (en) * | 2021-05-14 | 2021-07-30 | 北京工业大学 | Single image defogging method based on morphological reconstruction and saturation compensation |
CN113822823A (en) * | 2021-11-17 | 2021-12-21 | 武汉工程大学 | Point neighbor restoration method and system for aerodynamic optical effect image space-variant fuzzy core |
CN114842366A (en) * | 2022-07-05 | 2022-08-02 | 山东中宇航空科技发展有限公司 | Stability identification method for agricultural plant protection unmanned aerial vehicle |
Citations (4)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
WO2009112710A2 (en) * | 2008-02-11 | 2009-09-17 | Realeyes3D | Method of restoring a blurred image acquired by means of a camera fitted to a communication terminal |
CN104331871A (en) * | 2014-12-02 | 2015-02-04 | 苏州大学 | Image de-blurring method and image de-blurring device |
CN106251301A (en) * | 2016-07-26 | 2016-12-21 | 北京工业大学 | A kind of single image defogging method based on dark primary priori |
CN107369134A (en) * | 2017-06-12 | 2017-11-21 | 上海斐讯数据通信技术有限公司 | A kind of image recovery method of blurred picture |
-
2018
- 2018-05-17 CN CN201810471510.5A patent/CN108765325B/en active Active
Patent Citations (4)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
WO2009112710A2 (en) * | 2008-02-11 | 2009-09-17 | Realeyes3D | Method of restoring a blurred image acquired by means of a camera fitted to a communication terminal |
CN104331871A (en) * | 2014-12-02 | 2015-02-04 | 苏州大学 | Image de-blurring method and image de-blurring device |
CN106251301A (en) * | 2016-07-26 | 2016-12-21 | 北京工业大学 | A kind of single image defogging method based on dark primary priori |
CN107369134A (en) * | 2017-06-12 | 2017-11-21 | 上海斐讯数据通信技术有限公司 | A kind of image recovery method of blurred picture |
Non-Patent Citations (1)
Title |
---|
仇翔等: "基于LO稀疏先验的相机抖动模糊图像盲复原", 《光学精密工程》 * |
Cited By (23)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN109521556A (en) * | 2018-12-07 | 2019-03-26 | 歌尔科技有限公司 | A kind of electron microscopic wearable device |
CN109584186A (en) * | 2018-12-25 | 2019-04-05 | 西北工业大学 | A kind of unmanned aerial vehicle onboard image defogging method and device |
CN110097521A (en) * | 2019-05-08 | 2019-08-06 | 华南理工大学 | A kind of convolutional neural networks image recovery method towards reflecting metal vision-based detection |
CN110097521B (en) * | 2019-05-08 | 2023-02-28 | 华南理工大学 | Convolution neural network image restoration method for reflective metal visual detection |
CN110288550B (en) * | 2019-06-28 | 2020-04-24 | 中国人民解放***箭军工程大学 | Single-image defogging method for generating countermeasure network based on priori knowledge guiding condition |
CN110288550A (en) * | 2019-06-28 | 2019-09-27 | 中国人民解放***箭军工程大学 | The single image defogging method of confrontation network is generated based on priori knowledge guiding conditions |
CN110400312A (en) * | 2019-07-31 | 2019-11-01 | 北京金山云网络技术有限公司 | Determine the method, apparatus and server of image vague category identifier |
CN110648291A (en) * | 2019-09-10 | 2020-01-03 | 武汉科技大学 | Unmanned aerial vehicle motion blurred image restoration method based on deep learning |
CN110648291B (en) * | 2019-09-10 | 2023-03-03 | 武汉科技大学 | Unmanned aerial vehicle motion blurred image restoration method based on deep learning |
CN110676753B (en) * | 2019-10-14 | 2020-06-23 | 宁夏百川电力股份有限公司 | Intelligent inspection robot for power transmission line |
CN110676753A (en) * | 2019-10-14 | 2020-01-10 | 宁夏百川电力股份有限公司 | Intelligent inspection robot for power transmission line |
CN110874826B (en) * | 2019-11-18 | 2020-07-31 | 北京邮电大学 | Workpiece image defogging method and device applied to ion beam precise film coating |
CN110874826A (en) * | 2019-11-18 | 2020-03-10 | 北京邮电大学 | Workpiece image defogging method and device applied to ion beam precise film coating |
CN110895141A (en) * | 2019-11-28 | 2020-03-20 | 梁彦云 | Residential space crowding degree analysis platform |
CN111717406B (en) * | 2020-06-17 | 2021-10-01 | 中国人民解放军陆军工程大学 | Unmanned aerial vehicle image acquisition system |
CN111717406A (en) * | 2020-06-17 | 2020-09-29 | 中国人民解放军陆军工程大学 | Unmanned aerial vehicle image acquisition system |
CN111881982A (en) * | 2020-07-30 | 2020-11-03 | 北京环境特性研究所 | Unmanned aerial vehicle target identification method |
CN112465777A (en) * | 2020-11-26 | 2021-03-09 | 华能通辽风力发电有限公司 | Fan blade surface defect identification technology based on video stream |
CN113191982A (en) * | 2021-05-14 | 2021-07-30 | 北京工业大学 | Single image defogging method based on morphological reconstruction and saturation compensation |
CN113191982B (en) * | 2021-05-14 | 2024-05-28 | 北京工业大学 | Single image defogging method based on morphological reconstruction and saturation compensation |
CN113822823A (en) * | 2021-11-17 | 2021-12-21 | 武汉工程大学 | Point neighbor restoration method and system for aerodynamic optical effect image space-variant fuzzy core |
CN114842366A (en) * | 2022-07-05 | 2022-08-02 | 山东中宇航空科技发展有限公司 | Stability identification method for agricultural plant protection unmanned aerial vehicle |
CN114842366B (en) * | 2022-07-05 | 2022-09-16 | 山东中宇航空科技发展有限公司 | Stability identification method for agricultural plant protection unmanned aerial vehicle |
Also Published As
Publication number | Publication date |
---|---|
CN108765325B (en) | 2021-06-29 |
Similar Documents
Publication | Publication Date | Title |
---|---|---|
CN108765325A (en) | A kind of small drone Restoration method of blurred image | |
CN104392205B (en) | A kind of recognition methods and system of improper license plate | |
CN111079556A (en) | Multi-temporal unmanned aerial vehicle video image change area detection and classification method | |
CN105701483B (en) | A kind of urban border extracting method of fusion Multi-spectral Remote Sensing Data and night lights remotely-sensed data | |
CN107103317A (en) | Fuzzy license plate image recognition algorithm based on image co-registration and blind deconvolution | |
CN108596166A (en) | A kind of container number identification method based on convolutional neural networks classification | |
CN110378332A (en) | A kind of container terminal case number (CN) and Train number recognition method and system | |
WO2022027931A1 (en) | Video image-based foreground detection method for vehicle in motion | |
CN110866879B (en) | Image rain removing method based on multi-density rain print perception | |
CN106815583B (en) | Method for positioning license plate of vehicle at night based on combination of MSER and SWT | |
CN110766020A (en) | System and method for detecting and identifying multi-language natural scene text | |
CN103578083B (en) | Single image defogging method based on associating average drifting | |
CN105005778A (en) | Expressway vehicle detection method based on visual background extraction | |
CN110310241A (en) | A kind of more air light value traffic image defogging methods of fusion depth areas segmentation | |
CN107358585A (en) | Misty Image Enhancement Method based on fractional order differential and dark primary priori | |
CN104766344B (en) | Vehicle checking method based on movement edge extractor | |
Patil et al. | Motion saliency based generative adversarial network for underwater moving object segmentation | |
CN108319958A (en) | A kind of matched driving license of feature based fusion detects and recognition methods | |
Yamashita et al. | Multiple skip connections of dilated convolution network for semantic segmentation | |
CN105405138A (en) | Water surface target tracking method based on saliency detection | |
CN114387592B (en) | Character positioning and identifying method under complex background | |
CN113673384A (en) | Oracle character detection method for guiding texture feature autonomous learning by LM filter bank | |
CN112613392A (en) | Lane line detection method, device and system based on semantic segmentation and storage medium | |
CN111915634A (en) | Target object edge detection method and system based on fusion strategy | |
CN105574860B (en) | A kind of rudder piece deflection polarity recognition methods |
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 |