CN107680040A - A kind of blind convolution super-resolution reconstruction method of multiframe based on bayesian criterion and device - Google Patents
A kind of blind convolution super-resolution reconstruction method of multiframe based on bayesian criterion and device Download PDFInfo
- Publication number
- CN107680040A CN107680040A CN201710879520.8A CN201710879520A CN107680040A CN 107680040 A CN107680040 A CN 107680040A CN 201710879520 A CN201710879520 A CN 201710879520A CN 107680040 A CN107680040 A CN 107680040A
- Authority
- CN
- China
- Prior art keywords
- image
- super
- algorithm
- resolution
- area
- 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.)
- Pending
Links
- 238000000034 method Methods 0.000 title claims abstract description 70
- 238000013441 quality evaluation Methods 0.000 claims abstract description 36
- 238000011084 recovery Methods 0.000 claims abstract description 9
- 238000004364 calculation method Methods 0.000 claims description 15
- 230000005855 radiation Effects 0.000 claims description 14
- 238000001514 detection method Methods 0.000 claims description 3
- 238000002939 conjugate gradient method Methods 0.000 description 7
- PXFBZOLANLWPMH-UHFFFAOYSA-N 16-Epiaffinine Natural products C1C(C2=CC=CC=C2N2)=C2C(=O)CC2C(=CC)CN(C)C1C2CO PXFBZOLANLWPMH-UHFFFAOYSA-N 0.000 description 6
- 238000005457 optimization Methods 0.000 description 6
- 230000009466 transformation Effects 0.000 description 6
- 239000013598 vector Substances 0.000 description 6
- 238000010586 diagram Methods 0.000 description 5
- 238000003384 imaging method Methods 0.000 description 5
- 238000005516 engineering process Methods 0.000 description 4
- 238000007476 Maximum Likelihood Methods 0.000 description 3
- 230000008901 benefit Effects 0.000 description 3
- 241000208340 Araliaceae Species 0.000 description 2
- 235000005035 Panax pseudoginseng ssp. pseudoginseng Nutrition 0.000 description 2
- 235000003140 Panax quinquefolius Nutrition 0.000 description 2
- 241000287181 Sturnus vulgaris Species 0.000 description 2
- 230000002159 abnormal effect Effects 0.000 description 2
- 230000007423 decrease Effects 0.000 description 2
- 230000007812 deficiency Effects 0.000 description 2
- 238000011156 evaluation Methods 0.000 description 2
- 235000008434 ginseng Nutrition 0.000 description 2
- 239000011159 matrix material Substances 0.000 description 2
- 238000012634 optical imaging Methods 0.000 description 2
- 230000000149 penetrating effect Effects 0.000 description 2
- 238000013016 damping Methods 0.000 description 1
- 230000003247 decreasing effect Effects 0.000 description 1
- 238000009792 diffusion process Methods 0.000 description 1
- 238000002592 echocardiography Methods 0.000 description 1
- 238000003708 edge detection Methods 0.000 description 1
- 238000011478 gradient descent method Methods 0.000 description 1
- 238000010606 normalization Methods 0.000 description 1
- 230000000474 nursing effect Effects 0.000 description 1
- 230000003287 optical effect Effects 0.000 description 1
- 238000012805 post-processing Methods 0.000 description 1
- 230000001131 transforming effect Effects 0.000 description 1
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T3/00—Geometric image transformation in the plane of the image
- G06T3/40—Scaling the whole image or part thereof
- G06T3/4053—Super resolution, i.e. output image resolution higher than sensor resolution
-
- G06T3/147—
-
- G06T5/80—
-
- 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]
Abstract
The invention discloses a kind of blind convolution super-resolution reconstruction method of multiframe based on bayesian criterion and device.Wherein, this method comprises the following steps:Algorithm is selected to obtain the matching area of the target image after the area-of-interest and radiant correction of reference picture by the image quality evaluation of Same Scene sequence image and frame;The matching area of target image after radiant correction obtains radiancy and accurately geometric distortion parameter by image registration algorithm;Accurately geometric distortion parameter is obtained into the point spread function of Image Super-resolution recovery by multiframe blind deconvolution Image Restoration Algorithm;The point spread function that radiancy and Image Super-resolution are restored is obtained into Super-resolution Reconstruction image by maximum a posteriori super-resolution rebuilding algorithm.The present invention solves the problems, such as that traditional general algorithm is to point spread function, motion blur and insufficient to image structure information and the consideration such as openness, system point spread function and multiple image registration parameter is estimated automatically, the image resolution ratio improved.
Description
Technical field
The invention belongs to super-resolution rebuilding technical field, more particularly to a kind of blind convolution of multiframe based on bayesian criterion
Super-resolution reconstruction method and device.
Background technology
Super-resolution rebuilding technology (Super-Resolution, SR) is a kind of Image Post-processing Techniques based on software,
Several low-resolution images of Same Scene can be utilized to reconstruct a panel height image in different resolution, be to improve image spatial resolution
Another effective way.
A variety of classical super-resolution algorithms international at present include:Iterative backprojection method (Iterative Back
Projection, IBP), projections onto convex sets (Projection Onto Convex, POCS), maximum likelihood estimate
(Maximum Likelihood, ML), MAP estimation method (Maximum A Posterior, MAP) and mixing ML/MAP/
POCS methods etc..The observing and nursing of these methods is mostly the shift characteristics based on Fourier transformation, though calculating simply, is not had in model
There is the point spread function (PSF) for considering optical system, also do not account for motion blur and observation noise, can only be confined to global flat
Shifting movement and the constant model that degrades of linear space, and super-resolution rebuilding process can not be effectively introduced into the constraint of priori canonical.Separately
Outer much substantially hyperspherical global optimizations of least square Euclidean distance, what is asked is not sparse solution, so as to picture structure
Information and openness consideration deficiency, sequence image reconstruction precision are restricted by registration error.
The content of the invention
Present invention solves the technical problem that it is:Overcome the deficiencies in the prior art, there is provided a kind of based on bayesian criterion
The blind convolution super-resolution reconstruction method of multiframe and device, solves traditional general algorithm to point spread function, motion blur and to figure
As structural information and it is openness the problem of, system point spread function and multiple image registration parameter are estimated automatically, raising
Image resolution ratio.
The object of the invention is achieved by the following technical programs:The invention provides a kind of based on the more of bayesian criterion
The blind convolution super-resolution reconstruction method of frame, the described method comprises the following steps:The image quality evaluation of Same Scene sequence image selects with frame
Step:Algorithm is selected to obtain area-of-interest and the radiation school of reference picture by the image quality evaluation and frame of Same Scene sequence image
The matching area of target image after just;Image registration step:The matching area of target image after radiant correction passes through image
Registration Algorithm obtains radiancy and accurately geometric distortion parameter;Multiframe blind deconvolution image restoration step:To accurately geometry
Distortion parameter obtains the point spread function of Image Super-resolution recovery by multiframe blind deconvolution Image Restoration Algorithm;Maximum a posteriori surpasses
Resolution reconstruction process:The point spread function that radiancy and Image Super-resolution are restored is calculated by maximum a posteriori super-resolution rebuilding
Method obtains Super-resolution Reconstruction image.
In the blind convolution super-resolution reconstruction method of the above-mentioned multiframe based on bayesian criterion, the picture matter of Same Scene sequence image
Evaluation selects step to further comprise with frame:Area-of-interest selecting step:Pass through area-of-interest algorithm picks reference picture
The matching area of area-of-interest and target image;Image quality evaluation selects step with frame:The matching area of target image is led to
Cross image quality evaluation and the image of the edge gradient of matching area that frame selects algorithm to obtain target image and maximum;Radiation calibration
Step:After edge gradient and maximum image to the matching area of target image obtain radiant correction by radiation calibration formula
Target image matching area.
In the blind convolution super-resolution reconstruction method of the above-mentioned multiframe based on bayesian criterion, area-of-interest algorithm includes following
Step:(1) area-of-interest in reference picture is chosen;(2) area-of-interest in the reference picture in calculation procedure (1)
SURF features;(3) in target image identical geographical location, twice of size area is chosen as region to be matched;(4) calculate
The SURF features in the region to be matched in step (3);(5) it is the SURF of the area-of-interest in the reference picture of step (2) is special
The SURF features in the region to be matched of step of seeking peace (4) are matched, if the match is successful, are gone to step (6);Otherwise expand and treat
Matching area, go to step (4);(6) matching area of the area-of-interest and target image in reference picture is determined.
In the blind convolution super-resolution reconstruction method of the above-mentioned multiframe based on bayesian criterion, image quality evaluation selects algorithm with frame
Comprise the following steps:With the edge of the matching area of Sobel operators detection target image;By the matching area decile of target image
For several regions, the maximum edge of Sobel gradients in each region is chosen;Along the Sobel ladders in each region chosen
Maximum edge-perpendicular direction bi-cubic interpolation is spent, obtains the image brightness distribution in the edge-perpendicular direction in each part, its
In, each region includes multiple parts;It is determined that the maximum of points of the image brightness distribution in edge-perpendicular direction in each part
And minimum point, the slope for the straight line being fitted according to maximum of points and minimum point determine the gradient at edge;According to each part
The absolute value of the gradient at interior edge sum to obtain regional edge gradient and;According to the edge gradient of regional and choosing
Take the image corresponding to the region of edge gradient and maximum.
In the blind convolution super-resolution reconstruction method of the above-mentioned multiframe based on bayesian criterion, image registration step is further wrapped
Include:Geometric distortion parametric estimation step based on SURF characteristic points:The matching area of target image after radiant correction is passed through
The image registration algorithm then put based on SURF spy obtains the preliminary geometric correction parameter of image;Radiancy estimating step:To radiating school
Radiancy of the matching area of target image after just after radiancy parameter estimation algorithm obtains radiant correction;Based on image
The geometric distortion parametric estimation step of intensity:Base is passed through according to the radiancy after radiant correction and the preliminary geometric correction parameter of image
Accurately geometric distortion parameter is obtained in the geometric distortion parameter estimation algorithm of image intensity.
In the blind convolution super-resolution reconstruction method of the above-mentioned multiframe based on bayesian criterion, in multiframe blind deconvolution image restoration
In step, multiframe blind deconvolution Image Restoration Algorithm comprises the following steps:Step 1:Accurately geometric distortion parameter it will be used as just
Beginning variable;Step 2:The random perturbation variable of Bernoulli Jacob's distribution is obeyed in generation;Step 3:Random perturbation variable is superimposed upon just
Beginning variable;Step 4:The validity function of calculation procedure threeStep 5:Random perturbation variable is negated and is superimposed upon initially
Variable;Step 6:The validity function of calculation procedure fiveStep 7:Calculate the variable quantity of validity functionStep 8:The variable quantity of validity function and initializaing variable are updated into linear difference formula to obtain
New variable, using the new variable as initializaing variable, repeat step two to step 7, until obtaining what Image Super-resolution restored
Point spread function.
In the blind convolution super-resolution reconstruction method of the above-mentioned multiframe based on bayesian criterion, maximum a posteriori super-resolution rebuilding step
Suddenly following steps are further comprised:Same Scene sequence image is corrected to obtain by the radiancy in image registration step
Image after radiant correction;The point spread function that image after radiant correction restores with Image Super-resolution passes through maximum a posteriori oversubscription
Resolution algorithm for reconstructing obtains Super-resolution Reconstruction image.
On the other hand, present invention also offers a kind of blind convolution Super-resolution Reconstruction device of multiframe based on bayesian criterion,
Including:The image quality evaluation of Same Scene sequence image and frame modeling block, for the image quality evaluation by Same Scene sequence image
Algorithm is selected to obtain the matching area of the target image after the area-of-interest and radiant correction of reference picture with frame;Image registration mould
Block, the matching area for the target image after radiant correction obtains radiancy by image registration algorithm and accurately geometry is abnormal
Variable element;Multiframe blind deconvolution image restoration module, for accurately geometric distortion parameter multiframe blind deconvolution image will to be passed through
Restoration algorithm obtains the point spread function of Image Super-resolution recovery;Maximum a posteriori super-resolution rebuilding module, for by radiancy
The point spread function restored with Image Super-resolution obtains Super-resolution Reconstruction image by maximum a posteriori super-resolution rebuilding algorithm.
In the blind convolution Super-resolution Reconstruction device of the above-mentioned multiframe based on bayesian criterion, the Same Scene sequence image
Image quality evaluation includes with frame modeling block:Area-of-interest chooses module, for passing through area-of-interest algorithm picks reference picture
Area-of-interest and target image matching area;Image quality evaluation and frame modeling block, for the matching to target image
Algorithm is selected to obtain the edge gradient of matching area and the image of maximum of target image by image quality evaluation and frame in region;Spoke
Scaling module is penetrated, is obtained for the edge gradient of the matching area to target image and maximum image by radiation calibration formula
The matching area of target image after radiant correction.
In the blind convolution Super-resolution Reconstruction device of the above-mentioned multiframe based on bayesian criterion, described image registration module includes:
Geometric distortion parameter estimation module based on SURF characteristic points, for the matching area process to the target image after radiant correction
The image registration algorithm then put based on SURF spy obtains the preliminary geometric correction parameter of image;Radiancy estimation module, for spoke
Radiancy of the matching area for the target image penetrated after correction after radiancy parameter estimation algorithm obtains radiant correction;It is based on
The geometric distortion parameter estimation module of image intensity, for being joined according to the radiancy after radiant correction and the preliminary geometric correction of image
Number obtains accurately geometric distortion parameter by the geometric distortion parameter estimation algorithm based on image intensity.
The present invention has the advantages that compared with prior art:
(1) present invention stares the imaging characteristicses of Optical Imaging Satellite for stationary orbit, for the multiframe sequence of Same Scene
Row image, by the Super-resolution reconstruction that the multiframe blind deconvolution image restoration method based on bayesian criterion is introduced to remote sensing images
In building, solve to estimate while multiple image registration parameter, system point spread function and super-resolution image under Bayesian frame
Problem, complete image super-resolution rebuilding flow is devised, image can be lifted using satellite multiframe sequence image
Spatial resolution;
(2) method for registering of the invention by using feature based, by image registration techniques and base based on SURF features
It is used in combination in the registration technique of image intensity, first carries out the estimation of geometric distortion parameter, then estimate radiancy parameter again,
Pass through radiometric calibration amendment after obtaining radiancy parameter, it is possible to use the higher image based on image intensity of registration accuracy
Method for registering, solve between Same Scene multiple image due to imaging time span is longer and there is brightness change to cause
The problem of image registration accuracy declines;
(3) requirement of the image super-resolution rebuilding to image registration is very high, only carries out single image registration accuracy and is difficult
Reach requirement, but if directly carrying out geometric distortion parameter Estimation, calculating process will be caused time-consuming very long, the present invention is by adopting
Geometric distortion parameter is obtained with the method for registering based on image intensity, makes high-definition picture and geometric distortion parameter alternating iteration
Renewal, arithmetic speed can be accelerated while ensureing that image registration accuracy does not decline, algorithm is reduced and take;
(4) present invention is restored for estimating point spread function by using multiframe blind deconvolution, and uses Gaussian function fitting
Point spread function, by the use of the standard deviation of Gaussian function as the method for point spread function parameter, affine transformation image can be completed
On the basis of accuracy registration, the multiframe blind deconvolution super-resolution rebuilding technology based on bayesian criterion is realized, improves image space
Resolution ratio, and optimized algorithm, lift calculating speed.
Brief description of the drawings
By reading the detailed description of hereafter preferred embodiment, it is various other the advantages of and benefit it is common for this area
Technical staff will be clear understanding.Accompanying drawing is only used for showing the purpose of preferred embodiment, and is not considered as to the present invention
Limitation.And in whole accompanying drawing, identical part is denoted by the same reference numerals.In the accompanying drawings:
Fig. 1 is the flow of the blind convolution super-resolution reconstruction method of the multiframe provided in an embodiment of the present invention based on bayesian criterion
Figure;
Fig. 2 is the schematic diagram of the interesting image regions selection provided in an embodiment of the present invention based on SURF;
Fig. 3 (a) is the schematic diagram at remote sensing images Sobel edges provided in an embodiment of the present invention;
Fig. 3 (b) is another schematic diagram at remote sensing images Sobel edges provided in an embodiment of the present invention;
Fig. 4 is the schematic diagram that bi-cubic interpolation provided in an embodiment of the present invention seeks edge-perpendicular direction Luminance Distribution;
Fig. 5 is the schematic diagram that edge-perpendicular direction provided in an embodiment of the present invention Luminance Distribution solves with gradient;
Fig. 6 is the flow chart of multiframe blind deconvolution image restoration provided in an embodiment of the present invention.
Embodiment
The exemplary embodiment of the disclosure is more fully described below with reference to accompanying drawings.Although the disclosure is shown in accompanying drawing
Exemplary embodiment, it being understood, however, that may be realized in various forms the disclosure without should be by embodiments set forth here
Limited.On the contrary, these embodiments are provided to facilitate a more thoroughly understanding of the present invention, and can be by the scope of the present disclosure
Completely it is communicated to those skilled in the art.It should be noted that in the case where not conflicting, embodiment in the present invention and
Feature in embodiment can be mutually combined.Describe the present invention in detail below with reference to the accompanying drawings and in conjunction with the embodiments.
Embodiment of the method:
Fig. 1 is the flow of the blind convolution super-resolution reconstruction method of the multiframe provided in an embodiment of the present invention based on bayesian criterion
Figure.With reference to figure 1, it is somebody's turn to do the blind convolution super-resolution reconstruction method of multiframe based on bayesian criterion and comprises the following steps:
The image quality evaluation of Same Scene sequence image selects step with frame:By the image quality evaluation of Same Scene sequence image with
Frame selects algorithm to obtain the matching area of the target image after the area-of-interest and radiant correction of reference picture;
Image registration step:The matching area of target image after radiant correction obtains radiancy by image registration algorithm
Accurately geometric distortion parameter;
Multiframe blind deconvolution image restoration step:Accurately geometric distortion parameter it will pass through multiframe blind deconvolution image restoration
Algorithm obtains the point spread function of Image Super-resolution recovery;
Maximum a posteriori super-resolution rebuilding step:The point spread function that radiancy and Image Super-resolution are restored is passed through into maximum
Posteriority super-resolution rebuilding algorithm obtains Super-resolution Reconstruction image.
In above-described embodiment, image quality evaluation and the frame of Same Scene sequence image select step to further comprise:
Area-of-interest selecting step:Pass through the area-of-interest and target figure of area-of-interest algorithm picks reference picture
The matching area of picture.
Further, area-of-interest algorithm comprises the following steps:
(1) area-of-interest in reference picture is chosen;
(2) the SURF features of the area-of-interest in the reference picture in calculation procedure (1);
(3) in target image identical geographical location, twice of size area is chosen as region to be matched;
(4) the SURF features in the region to be matched in calculation procedure (3);
(5) by the region to be matched of the SURF features and step (4) of the area-of-interest in the reference picture of step (2)
SURF features are matched, if the match is successful, are gone to step (6);Otherwise expand region to be matched, go to step (4);
(6) matching area of the area-of-interest and target image in reference picture is determined.
Specifically, a width remote sensing images area-of-interest is chosen manually as template, the geometric properties of calculation template;Then
According to the geographical coordinate of area-of-interest, in other remote sensing images correspondence position near zone computational geometry features, to two width figures
The geometric properties of picture are matched, and can also obtain the area-of-interest in other images.
As shown in Fig. 2 left figure is reference picture, right figure is target image, affine transformation between the two be present.In left figure
Area-of-interest (left frame region) is chosen, calculates its SURF feature;SURF features (figure is also calculated in large area in right figure
In only draw the feature of matching), characteristic matching is then carried out, so as to obtain the area-of-interest in right figure (left frame region).
Due to affine transformation relationship between two width figures be present, the true geographic range of the region of interest domain representation in two images is incomplete
It is identical.
Image quality evaluation selects step with frame:Algorithm is selected by image quality evaluation and frame to the matching area of target image
Obtain the edge gradient of the matching area of target image and maximum image.
Further, image quality evaluation selects algorithm to comprise the following steps with frame:(1) with Sobel operators detection target figure
The edge of the matching area of picture;(2) matching area of target image is divided into several regions, chosen in each region
The maximum edge of Sobel gradients;(3) along the maximum edge-perpendicular direction pair cube of the Sobel gradients in each region chosen
Interpolation, the image brightness distribution in the edge-perpendicular direction in each part is obtained, wherein, each region includes multiple parts;Such as
Shown in Fig. 4, along the edge-perpendicular direction bi-cubic interpolation chosen, the image brightness distribution in acquisition edge-perpendicular direction.(4) determine
The maximum of points and minimum point of the image brightness distribution in the edge-perpendicular direction in each part, according to maximum of points and minimum
The slope of the straight line of value point fitting determines the gradient at edge, as shown in figure 5, the point of black is the interpolation point of gradient direction, straight line
Line is maximum of points and minimum point fitting, and its slope is the gradient at edge.(5) according to the gradient at the edge in each part
Absolute value sum to obtain regional edge gradient and;(6) according to the edge gradient of regional and selection edge gradient
With the image corresponding to the region of maximum.Specifically, edge gradient and bigger, image fog-level is smaller, and picture quality is better;
Edge gradient and smaller, image fog-level is bigger, and picture quality is poorer.The purpose of step (6) be reject fog-level compared with
Big image.For less image, because the edge feature of image is insufficient, image quality evaluation selects algorithm can not be very with frame
Evaluation image quality well.For this situation, directly the Sobel gradients of image can be added, reject Sobel gradients compared with
Small image.Because the radiancy of image can influence Sobel gradients, if necessary to carry out registration to radiancy, picture frame choosing will
It is placed on after radiancy parameter Estimation.
Specifically, using carry out field convolution is each put in the direction template and image of two 3 × 3, respectively in level side
Edge is detected to vertical direction.Edge detection template horizontally and vertically is respectively:
Sobel gradients are defined as:
Hx(x,y)、Hy(x, y) is horizontally oriented the gradient with vertical direction respectively.Sobel gradients are more than some threshold value
Pixel be set to 1, remaining is set to 0, can obtain the edge of image, as shown in Fig. 3 (a) and Fig. 3 (b).The gradient direction at edge
It is:
Radiation calibration step:Edge gradient and maximum image to the matching area of target image is public by radiation calibration
Formula obtains the matching area of the target image after radiant correction.
Specifically, the formula of calibration is as follows:
L=Gain*DN+Offset (4)
Wherein, the metadata of remote sensing image includes gain G ain and biases Offset value.DN is the brightness of original image
Quantized value, L are target emanation brightness values.
In above-described embodiment, image registration step further comprises:
Geometric distortion parametric estimation step based on SURF characteristic points:To the matching area of the target image after radiant correction
The preliminary geometric correction parameter of image is obtained by the image registration algorithm then put based on SURF spy.
Specifically, SURF features are the vectors of one 64 dimension, describe certain point and the intensity around it, graded are closed
System.The SURF feature Euclidean distances of two points are smaller, then two points represent that the possibility of same target is higher.SURF is to image
Brightness change, rotation, scaling, or even affine transformation all keep certain consistency.
The image registration algorithm then put based on SURF spy is comprised the following steps:School is radiated using different size of template generation
Multiple different proportion images of target image after just;The Hessen matrix of multiple different proportion images is built, it is determined that each ratio
The extreme value (including maximum and minimum) of image;The extreme point of some scaled image is chosen, by the extreme point of the scaled image
Scaled image extreme point adjacent thereto compares, if being still extreme value, using the extreme point of the scaled image as preliminary special
Sign point;Further, multiple preliminary characteristic points are chosen according to this step;The sub-pix position of preliminary characteristic point is determined with cubic interpolation
Put, and the less preliminary characteristic point of gradient is removed, retain the big preliminary characteristic point of Gradient;With the big preliminary characteristic point of gradient
Centered on, 6s (s is graphical rule) is that the Harr small echos of 60 degree of sector regions of radius calculation respond sum, should and be an arrow
Amount, direction vector corresponding to vector maximum is set to the principal direction of characteristic point;Centered on the preliminary characteristic point big by gradient, side is taken
The square area (square direction is identical with principal direction) of a length of 20s (s is graphical rule), square is divided into 4 × 4 small
Square, calculate each square Harr wavelet characters (including x directions sum, y directions sum, x directions absolute value sum, y
Direction absolute value sum), obtain 64 n dimensional vector ns of the big preliminary characteristic point of the gradient;Calculate between 64 n dimensional vector ns it is European away from
From distance minimum is defined as match point.The preliminary geometric correction parameter of image is determined according to match point.
Radiancy estimating step:Radiancy parameter estimation algorithm is passed through to the matching area of the target image after radiant correction
Obtain the radiancy after radiant correction.
Specifically, when geometrical registration parameter and radiancy registration parameter estimation it is more accurate, registering image and benchmark image
Difference should be smaller.What optimized-type parameter Estimation referred to is exactly to minimize following formula by the method optimized,
Wherein,It is l width images ylWith kth width image ykImage after registration, θlBe registration parameter form to
Amount.Assuming that geometric distortion meets affine Transform Model,Mathematic(al) representation be:
Geometric distortion parameterBeing solved by the method for registering based on SURF to obtain, and radiancy parameterSolved using the method for optimization.
It is conjugate gradient method to solve the optimization method that radiancy parameter uses, and conjugate gradient method needs solution formula (5) right
The derivative of radiancy parameter.Formula (5) is rightWithDerivative be respectively
⊙ is Hadamard (Hadamard) operator in formula, represents that corresponding element is multiplied, summation ∑ is then represented to image institute
There are pixel summation, y 'lIt is l width images ylWith kth width image ykImage after geometrical registration.Conjugate gradient method is a kind of conventional
Solution unconstrained problem optimal method, algorithm structure is simple, has super-linear convergence.
Geometric distortion parametric estimation step based on image intensity:It is tentatively several according to the radiancy after radiant correction and image
What correction parameter obtains accurately geometric distortion parameter by the geometric distortion parameter estimation algorithm based on image intensity.
Although unrelated with the change of image radiation degree based on SURF image registration algorithm, registration accuracy is general, also needs
It is further smart registering.We employ the registration Algorithm based on image intensity, and in order to strengthen the stability of algorithm, first build
Vertical image pyramid, take by the thick registration strategies to essence.
Geometric distortion parameter estimation algorithm based on image intensity comprises the following steps:(1) according to the spoke after radiant correction
Degree of penetrating, adjust the brightness of target image;(2) L of reference picture and target image is established2Type multinomial pyramid;(3) order is based on
The preliminary geometric correction parameter of image that SURF method for registering images obtains is as initial value;(4) Marquardt-Levenberg is used
(ML) algorithm registration large scale reference picture and target image obtain large scale geometric correction of imagery parameter;(5) ratio is reduced, with
The large scale geometric correction of imagery parameter that (4) step obtains makees initial value, repeat step (4), until minimum scale image is matched somebody with somebody
It is accurate.
In step (4), ML algorithms are a kind of optimized algorithms of solving non-linear least square problem, it is necessary to the generation of optimization
Valency function and formula (5) it is identical, it is necessary to solve variable be
In order to describe conveniently, ML algorithms are illustrated with an example.The purpose of ML algorithms is to give one group of data (xi,yi) and
Function f (), finding optimal parameter p makesIt is minimum.
In ML algorithms, each iteration is to find a suitable damping factor λ, when λ very littles, δkIt is Gauss-
The optimal step size of Newton methods, when λ is very big, deteriorate to the optimal step size calculating formula of gradient descent method.
, it is necessary to which registration parameter is transferred into next yardstick after large scale image obtains registration, the relation of parameter is as follows:
a1'=a1,a2'=a2,a3'=a3,a4'=a4
p1'=sp1,p2'=sp2
Wherein band ' the next yardstick of symbology parameter, s is the ratio between two yardsticks.
In multiframe blind deconvolution image restoration step, multiframe blind deconvolution Image Restoration Algorithm comprises the following steps:Step
Rapid one:Will accurately geometric distortion parameter as initializaing variable;Step 2:The random perturbation that Bernoulli Jacob's distribution is obeyed in generation becomes
Amount;Step 3:Random perturbation variable is superimposed upon initializaing variable;Step 4:The validity function of calculation procedure threeStep
Rapid five:Random perturbation variable is negated and is superimposed upon initializaing variable;Step 6:The validity function of calculation procedure fiveStep
Seven:Calculate the variable quantity of validity functionStep 8:By the variable quantity of validity function and initially
Variable is updated to linear difference formula and obtains new variable, using the new variable as initializaing variable, repeat step two to step
Seven, until obtaining the point spread function that Image Super-resolution restores.
Specifically, multiple image restoration algorithm is based on the maximum-likelihood criterion Jing Guo regularization, object function:
Wherein, f be restore image, hiIt is the i-th width observed image yiCorresponding point spread function,Image yiNoise
Variance, it with image-related regular terms, α is regularization parameter that T (f), which is,.Conventional regular terms has Tikhonov and total variation.
It is not directly to minimize formula (9) with conjugate gradient method when estimation image and point spread function, but takes alternating
The strategy of iteration, as shown in Figure 6.Alternating iteration is exactly under conditions of assuming that restored image is constant, minimizes formula (9) estimation point
Spread function;It is then assumed that point spread function is constant, formula (9) estimation restored image is minimized.So move in circles, make a diffusion
Function and restored image are progressively close to true value.F in figure0It is superimposed image of the image after geometry and radiancy registration, h0It is mark
The Gaussian function that quasi- difference is 0.7, its mathematic(al) representation are:
Wherein, σ is the standard deviation of Gaussian function.
After the estimation of point spread function is calculated, point spread function is projected into fine-resolution meshes, is directly used in figure
As super-resolution rebuilding.But the amount of calculation of super-resolution rebuilding in itself is very big, it should reduces unknown quantity as far as possible.In order to reduce
The amount of calculation of image super-resolution rebuilding, with Gaussian function fitting point spread function, expanded by the use of the standard deviation of Gaussian function as point
Dissipate the parameter of function.Fitting employs SPGD algorithms, and its basic thought is described as follows:If current drive controller voltage (is treated
Seek variable) it is u(k)=(u1,u2,...,uN), cost function to be optimized is J (u), small random perturbation vector during kth time iteration
{δujPositively and negatively it is applied in parallel to N number of driver of wave-front corrector.Now mutually strained caused by system performance index
Changing δ J is
δJ+=J (u1+δu1,u2+δu2,...,uN+δuN)
δJ-=J (u1-δu1,u2-δu2,...,uN-δuN) (11)
δ J=δ J+-δJ-
Then the variable quantity δ J of property data and random perturbation { δ ujCarry out the gradient estimation of kth time iteration.In ladder
The direction that degree declines is controlled the iteration of parameter:
u(k+1)={ uj (k+1)}={ uj (k)+γδJδuj} (12)
In formula, parameter γ take on the occasion of when towards the optimization of performance indications great direction;Towards performance indications when γ takes negative value
Minimum direction optimization.
The cost function of fitting is
WhereinIt is average point spread function, i.e.,
N is the number of point spread function.
The point spread function standard deviation that above-mentioned algorithm is fitted to obtain obtains under low-resolution image.According to imaging mould
Type, point spread function should be projected in fine-resolution meshes.For image super-resolution restore point spread function ginseng
Number is should be
σs=s × σ (15)
Wherein s is the multiple of super-resolution.
In above-described embodiment, maximum a posteriori super-resolution rebuilding step further comprises following steps:Pass through image registration
Radiancy in step is corrected to obtain the image after radiant correction to Same Scene sequence image;Image after radiant correction
The point spread function restored with Image Super-resolution obtains Super-resolution Reconstruction image by maximum a posteriori super-resolution rebuilding algorithm.
Specifically, by the geometric distortion parameter estimated, the point spread function that image blind deconvolution is estimated to obtain is projected to
On high-definition picture grid, and think that the point spread function is more accurate, influence the main of high resolution image reconstruction quality
It is the evaluated error of geometrical registration error, i.e. geometric distortion parameter.Under above-mentioned hypothesis, maximum a posteriori probability is:
Assume that geometric distortion parameter r is obeyed in above formula to be uniformly distributed.Section 1 is likelihood item on the right of equation, its expression formula
For:
And p (x) represent high-definition picture priori item, this programme using stop primary markov random file priori,
I.e.
Wherein G is gradient operator, and α is the parameter of not primary potential function, and v is the parameter for characterizing priori intensity, and Z is normalization ginseng
Number.Natural logrithm is taken to formula (16) both sides, and takes opposite number, after neglecting constant term, obtaining cost function is:
With image blind deconvolution similarly, take alternating iteration strategy minimize above formula come estimate geometric distortion parameter and
High-definition picture.Image estimation minimizes formula (21) using conjugate gradient method, and the derivative that conjugate gradient method is used is as follows,
However, cost function is not easy to try to achieve relative to the derivative of geometric distortion parameter, it is impossible to directly uses conjugate gradient method
Solve.This programme realizes the minimum and geometric distortion parameter of cost function using the method for registering based on image intensity indirectly
Solve.Reference picture is that super-resolution image passes through the low resolution image to degrade again, and reference picture needs and all observation
Gained low-resolution image is registering, and the parameter after registration is transforming to high-resolution coordinate system.
The initial value x of high-definition picture0It can be calculated by following formula,
Wherein W, y, Λα、ΛβIt is all W respectivelyk、yk、Stacking, S is a diagonal matrix, diagonal element
The sum for each row of W that element is.
The present embodiment stares the imaging characteristicses of Optical Imaging Satellite for stationary orbit, for more frame sequences of Same Scene
Image, by the super-resolution rebuilding that the multiframe blind deconvolution image restoration method based on bayesian criterion is introduced to remote sensing images
In, estimate to ask while solving multiple image registration parameter, system point spread function and super-resolution image under Bayesian frame
Topic, is devised complete image super-resolution rebuilding flow, the sky of image can be lifted using satellite multiframe sequence image
Between resolution ratio;By using the method for registering of feature based, by image registration techniques based on SURF features and strong based on image
The registration technique of degree is used in combination, and first carries out the estimation of geometric distortion parameter, then estimates radiancy parameter again, radiated
Pass through radiometric calibration amendment after degree parameter, it is possible to use the higher image registration side based on image intensity of registration accuracy
Method, solve between Same Scene multiple image due to imaging time span is longer and there is brightness change to cause image to be matched somebody with somebody
The problem of quasi- precise decreasing;Requirement of the image super-resolution rebuilding to image registration is very high, only carries out single image registration essence
Degree is extremely difficult to require, but if directly carrying out geometric distortion parameter Estimation, calculating process will be caused to take very long, this implementation
Example obtains geometric distortion parameter by using the method for registering based on image intensity, makes high-definition picture and geometric distortion parameter
Alternating iteration updates, and can accelerate arithmetic speed while ensureing that image registration accuracy does not decline, reduce algorithm and take;This
Embodiment is restored for estimating point spread function by using multiframe blind deconvolution, and with Gaussian function fitting point spread function,
By the use of the standard deviation of Gaussian function as the method for point spread function parameter, affine transformation image accuracy registration basis can be completed
On, the multiframe blind deconvolution super-resolution rebuilding technology based on bayesian criterion is realized, improves image spatial resolution, and optimize
Algorithm, lift calculating speed.
Device embodiment:
The present embodiment additionally provides a kind of blind convolution Super-resolution Reconstruction device of multiframe based on bayesian criterion, and its feature exists
In including:The image quality evaluation of Same Scene sequence image and frame modeling block, image registration module, multiframe blind deconvolution image restoration
Module and maximum a posteriori super-resolution rebuilding module;Wherein, the image quality evaluation of Same Scene sequence image and frame modeling block, are used for
After selecting algorithm to obtain the area-of-interest and radiant correction of reference picture with frame by the image quality evaluation of Same Scene sequence image
Target image matching area;Image registration module, the matching area for the target image after radiant correction pass through image
Registration Algorithm obtains radiancy and accurately geometric distortion parameter;Multiframe blind deconvolution image restoration module, for will accurately
Geometric distortion parameter obtains the point spread function of Image Super-resolution recovery by multiframe blind deconvolution Image Restoration Algorithm;After maximum
Super-resolution rebuilding module is tested, for the point spread function of radiancy and Image Super-resolution recovery to be passed through into maximum a posteriori super-resolution
Rate algorithm for reconstructing obtains Super-resolution Reconstruction image.
In above-described embodiment, image quality evaluation and the frame modeling block of Same Scene sequence image include:Area-of-interest is chosen
Module, image quality evaluation and frame modeling block and radiation calibration module;Wherein, area-of-interest chooses module, for passing through sense
The area-of-interest of interest region algorithm picks reference picture and the matching area of target image;Image quality evaluation and frame modeling
Block, algorithm is selected to obtain the matching area of target image by image quality evaluation and frame for the matching area to target image
Edge gradient and maximum image;Radiation calibration module, edge gradient and maximum for the matching area to target image
Image obtains the matching area of the target image after radiant correction by radiation calibration formula.
In above-described embodiment, image registration module includes:Geometric distortion parameter estimation module, spoke based on SURF characteristic points
Degree of penetrating estimation module and the geometric distortion parameter estimation module based on image intensity;Wherein, the geometry based on SURF characteristic points is abnormal
Variable element estimation module, match somebody with somebody for passing through the image then put based on SURF spy to the matching area of the target image after radiant correction
Quasi- algorithm obtains the preliminary geometric correction parameter of image;Radiancy estimation module, for the target image after radiant correction
Radiancy with region after radiancy parameter estimation algorithm obtains radiant correction;Geometric distortion parameter based on image intensity
Estimation module, for being passed through according to the radiancy after radiant correction and the preliminary geometric correction parameter of image based on the several of image intensity
What distortion parameter algorithm for estimating obtains accurately geometric distortion parameter.
Embodiment described above is the present invention more preferably embodiment, and those skilled in the art is in this hair
The usual variations and alternatives carried out in the range of bright technical scheme should all include within the scope of the present invention.
Claims (10)
- A kind of 1. blind convolution super-resolution reconstruction method of multiframe based on bayesian criterion, it is characterised in that methods described include with Lower step:The image quality evaluation of Same Scene sequence image selects step with frame:Selected by the image quality evaluation and frame of Same Scene sequence image Algorithm obtains the matching area of the target image after the area-of-interest and radiant correction of reference picture;Image registration step:The matching area of target image after radiant correction obtains radiancy and essence by image registration algorithm Accurate geometric distortion parameter;Multiframe blind deconvolution image restoration step:Accurately geometric distortion parameter it will pass through multiframe blind deconvolution Image Restoration Algorithm Obtain the point spread function of Image Super-resolution recovery;Maximum a posteriori super-resolution rebuilding step:The point spread function that radiancy and Image Super-resolution are restored is passed through into maximum a posteriori Super-resolution rebuilding algorithm obtains Super-resolution Reconstruction image.
- 2. the blind convolution super-resolution reconstruction method of the multiframe according to claim 1 based on bayesian criterion, it is characterised in that: The image quality evaluation of Same Scene sequence image selects step to further comprise with frame:Area-of-interest selecting step:Pass through the area-of-interest of area-of-interest algorithm picks reference picture and target image Matching area;Image quality evaluation selects step with frame:Algorithm is selected to obtain by image quality evaluation and frame the matching area of target image The edge gradient of the matching area of target image and maximum image;Radiation calibration step:Edge gradient and maximum image to the matching area of target image are obtained by radiation calibration formula The matching area of target image after to radiant correction.
- 3. the blind convolution super-resolution reconstruction method of the multiframe according to claim 2 based on bayesian criterion, it is characterised in that: Area-of-interest algorithm comprises the following steps:(1) area-of-interest in reference picture is chosen;(2) the SURF features of the area-of-interest in the reference picture in calculation procedure (1);(3) in target image identical geographical location, twice of size area is chosen as region to be matched;(4) the SURF features in the region to be matched in calculation procedure (3);(5) by the SURF in the region to be matched of the SURF features and step (4) of the area-of-interest in the reference picture of step (2) Feature is matched, if the match is successful, is gone to step (6);Otherwise expand region to be matched, go to step (4);(6) matching area of the area-of-interest and target image in reference picture is determined.
- 4. the blind convolution super-resolution reconstruction method of the multiframe according to claim 3 based on bayesian criterion, it is characterised in that: Image quality evaluation selects algorithm to comprise the following steps with frame:With the edge of the matching area of Sobel operators detection target image;The matching area of target image is divided into several regions, chooses the edge of the Sobel gradients maximum in each region;The edge-perpendicular direction bi-cubic interpolation maximum along the Sobel gradients in each region chosen, is obtained in each part Edge-perpendicular direction image brightness distribution, wherein, each region includes multiple parts;It is determined that the maximum of points and minimum point of the image brightness distribution in edge-perpendicular direction in each part, according to maximum The slope of point and the straight line of minimum point fitting determines the gradient at edge;According to the absolute value of the gradient at the edge in each part sum to obtain regional edge gradient and;According to the image corresponding to the edge gradient of regional and selection edge gradient and maximum region.
- 5. the blind convolution super-resolution reconstruction method of the multiframe according to claim 2 based on bayesian criterion, it is characterised in that: Image registration step further comprises:Geometric distortion parametric estimation step based on SURF characteristic points:The matching area of target image after radiant correction is passed through The image registration algorithm then put based on SURF spy obtains the preliminary geometric correction parameter of image;Radiancy estimating step:The matching area of target image after radiant correction is obtained by radiancy parameter estimation algorithm Radiancy after radiant correction;Geometric distortion parametric estimation step based on image intensity:According to the radiancy after radiant correction and the preliminary geometry school of image Positive parameter obtains accurately geometric distortion parameter by the geometric distortion parameter estimation algorithm based on image intensity.
- 6. the blind convolution super-resolution reconstruction method of the multiframe according to claim 5 based on bayesian criterion, it is characterised in that: In multiframe blind deconvolution image restoration step, multiframe blind deconvolution Image Restoration Algorithm comprises the following steps:Step 1:Will accurately geometric distortion parameter as initializaing variable;Step 2:The random perturbation variable of Bernoulli Jacob's distribution is obeyed in generation;Step 3:Random perturbation variable is superimposed upon initializaing variable;Step 4:The validity function of calculation procedure threeStep 5:Random perturbation variable is negated and is superimposed upon initializaing variable;Step 6:The validity function of calculation procedure fiveStep 7:Calculate the variable quantity of validity functionStep 8:The variable quantity of validity function and initializaing variable are updated to linear difference formula and obtain new variable, will The new variable is as initializaing variable, repeat step two to step 7, until obtaining the point spread function that Image Super-resolution restores.
- 7. the blind convolution super-resolution reconstruction method of the multiframe according to claim 6 based on bayesian criterion, it is characterised in that: Maximum a posteriori super-resolution rebuilding step further comprises following steps:It is corrected to obtain the image after radiant correction to Same Scene sequence image by the radiancy in image registration step;The point spread function that image after radiant correction restores with Image Super-resolution passes through maximum a posteriori super-resolution rebuilding algorithm Obtain Super-resolution Reconstruction image.
- A kind of 8. blind convolution Super-resolution Reconstruction device of multiframe based on bayesian criterion, it is characterised in that including:The image quality evaluation of Same Scene sequence image and frame modeling block, for the image quality evaluation by Same Scene sequence image with Frame selects algorithm to obtain the matching area of the target image after the area-of-interest and radiant correction of reference picture;Image registration module, the matching area for the target image after radiant correction obtain radiancy by image registration algorithm Accurately geometric distortion parameter;Multiframe blind deconvolution image restoration module, for accurately geometric distortion parameter multiframe blind deconvolution image restoration will to be passed through Algorithm obtains the point spread function of Image Super-resolution recovery;Maximum a posteriori super-resolution rebuilding module, for the point spread function of radiancy and Image Super-resolution recovery to be passed through into maximum Posteriority super-resolution rebuilding algorithm obtains Super-resolution Reconstruction image.
- 9. the blind convolution Super-resolution Reconstruction device of the multiframe according to claim 8 based on bayesian criterion, it is characterised in that The image quality evaluation of the Same Scene sequence image includes with frame modeling block:Area-of-interest chooses module, for the area-of-interest and target figure by area-of-interest algorithm picks reference picture The matching area of picture;Image quality evaluation and frame modeling block, algorithm is selected by image quality evaluation and frame for the matching area to target image Obtain the edge gradient of the matching area of target image and maximum image;Radiation calibration module, it is public by radiation calibration for the edge gradient of the matching area to target image and maximum image Formula obtains the matching area of the target image after radiant correction.
- 10. the blind convolution Super-resolution Reconstruction device of the multiframe according to claim 8 based on bayesian criterion, its feature exist In described image registration module includes:Geometric distortion parameter estimation module based on SURF characteristic points, for the matching area to the target image after radiant correction The preliminary geometric correction parameter of image is obtained by the image registration algorithm then put based on SURF spy;Radiancy estimation module, for passing through radiancy parameter estimation algorithm to the matching area of the target image after radiant correction Obtain the radiancy after radiant correction;Geometric distortion parameter estimation module based on image intensity, for tentatively several according to the radiancy after radiant correction and image What correction parameter obtains accurately geometric distortion parameter by the geometric distortion parameter estimation algorithm based on image intensity.
Priority Applications (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201710879520.8A CN107680040A (en) | 2017-09-26 | 2017-09-26 | A kind of blind convolution super-resolution reconstruction method of multiframe based on bayesian criterion and device |
Applications Claiming Priority (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201710879520.8A CN107680040A (en) | 2017-09-26 | 2017-09-26 | A kind of blind convolution super-resolution reconstruction method of multiframe based on bayesian criterion and device |
Publications (1)
Publication Number | Publication Date |
---|---|
CN107680040A true CN107680040A (en) | 2018-02-09 |
Family
ID=61137389
Family Applications (1)
Application Number | Title | Priority Date | Filing Date |
---|---|---|---|
CN201710879520.8A Pending CN107680040A (en) | 2017-09-26 | 2017-09-26 | A kind of blind convolution super-resolution reconstruction method of multiframe based on bayesian criterion and device |
Country Status (1)
Country | Link |
---|---|
CN (1) | CN107680040A (en) |
Cited By (12)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN108805220A (en) * | 2018-07-25 | 2018-11-13 | 广东奥普特科技股份有限公司 | A kind of Fast template matching algorithm based on gradient integral |
CN109064504A (en) * | 2018-08-24 | 2018-12-21 | 深圳市商汤科技有限公司 | Image processing method, device and computer storage medium |
CN109146941A (en) * | 2018-06-04 | 2019-01-04 | 成都通甲优博科技有限责任公司 | A kind of depth image optimization method and system based on net region division |
CN109767389A (en) * | 2019-01-15 | 2019-05-17 | 四川大学 | Adaptive weighted double blind super-resolution reconstruction methods of norm remote sensing images based on local and non local joint priori |
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 |
CN110415242A (en) * | 2019-08-02 | 2019-11-05 | 中国人民解放军军事科学院国防科技创新研究院 | A kind of super-resolution multiplying power evaluation method based on reference picture |
CN111176832A (en) * | 2019-12-06 | 2020-05-19 | 重庆邮电大学 | Performance optimization and parameter configuration method based on memory computing framework Spark |
CN112261315A (en) * | 2020-09-07 | 2021-01-22 | 清华大学 | High-resolution calculation imaging system and method based on camera array aperture synthesis |
CN112508788A (en) * | 2020-12-15 | 2021-03-16 | 华中科技大学 | Spatial neighborhood group target super-resolution method based on multi-frame observation information |
CN113496463A (en) * | 2020-03-19 | 2021-10-12 | 深圳华大生命科学研究院 | Image reconstruction method, system, electronic device and storage medium |
CN114897717A (en) * | 2022-04-25 | 2022-08-12 | 电子科技大学 | Multi-frame correction blind deconvolution method for solar image restoration |
CN117115025A (en) * | 2023-08-25 | 2023-11-24 | 中国人民解放军国防科技大学 | Image rapid restoration method, device and medium based on multiframe blind deconvolution |
Citations (2)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN102360498A (en) * | 2011-10-27 | 2012-02-22 | 江苏省邮电规划设计院有限责任公司 | Reconstruction method for image super-resolution |
CN105205783A (en) * | 2015-09-14 | 2015-12-30 | 河海大学 | SAR image blind super-resolution reestablishment method in combination with priori estimation |
-
2017
- 2017-09-26 CN CN201710879520.8A patent/CN107680040A/en active Pending
Patent Citations (2)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN102360498A (en) * | 2011-10-27 | 2012-02-22 | 江苏省邮电规划设计院有限责任公司 | Reconstruction method for image super-resolution |
CN105205783A (en) * | 2015-09-14 | 2015-12-30 | 河海大学 | SAR image blind super-resolution reestablishment method in combination with priori estimation |
Non-Patent Citations (2)
Title |
---|
JIEPING XU等: "Multi-Frame Super-Resolution of Gaofen-4 Remote Sensing Images", 《SENSOR》 * |
许洁平: "空间目标图像盲解卷积技术研究", 《中国优秀硕士学位论文全文数据库-信息科技辑》 * |
Cited By (17)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN109146941A (en) * | 2018-06-04 | 2019-01-04 | 成都通甲优博科技有限责任公司 | A kind of depth image optimization method and system based on net region division |
CN108805220A (en) * | 2018-07-25 | 2018-11-13 | 广东奥普特科技股份有限公司 | A kind of Fast template matching algorithm based on gradient integral |
CN109064504A (en) * | 2018-08-24 | 2018-12-21 | 深圳市商汤科技有限公司 | Image processing method, device and computer storage medium |
CN109767389A (en) * | 2019-01-15 | 2019-05-17 | 四川大学 | Adaptive weighted double blind super-resolution reconstruction methods of norm remote sensing images based on local and non local joint priori |
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 |
CN110415242A (en) * | 2019-08-02 | 2019-11-05 | 中国人民解放军军事科学院国防科技创新研究院 | A kind of super-resolution multiplying power evaluation method based on reference picture |
CN110415242B (en) * | 2019-08-02 | 2020-05-19 | 中国人民解放军军事科学院国防科技创新研究院 | Super-resolution magnification evaluation method based on reference image |
CN111176832B (en) * | 2019-12-06 | 2022-07-01 | 重庆邮电大学 | Performance optimization and parameter configuration method based on memory computing framework Spark |
CN111176832A (en) * | 2019-12-06 | 2020-05-19 | 重庆邮电大学 | Performance optimization and parameter configuration method based on memory computing framework Spark |
CN113496463A (en) * | 2020-03-19 | 2021-10-12 | 深圳华大生命科学研究院 | Image reconstruction method, system, electronic device and storage medium |
CN112261315A (en) * | 2020-09-07 | 2021-01-22 | 清华大学 | High-resolution calculation imaging system and method based on camera array aperture synthesis |
CN112508788A (en) * | 2020-12-15 | 2021-03-16 | 华中科技大学 | Spatial neighborhood group target super-resolution method based on multi-frame observation information |
CN114897717A (en) * | 2022-04-25 | 2022-08-12 | 电子科技大学 | Multi-frame correction blind deconvolution method for solar image restoration |
CN114897717B (en) * | 2022-04-25 | 2023-06-27 | 电子科技大学 | Multiframe correction blind deconvolution method for sun image restoration |
CN117115025A (en) * | 2023-08-25 | 2023-11-24 | 中国人民解放军国防科技大学 | Image rapid restoration method, device and medium based on multiframe blind deconvolution |
CN117115025B (en) * | 2023-08-25 | 2024-04-05 | 中国人民解放军国防科技大学 | Image rapid restoration method, device and medium based on multiframe blind deconvolution |
Similar Documents
Publication | Publication Date | Title |
---|---|---|
CN107680040A (en) | A kind of blind convolution super-resolution reconstruction method of multiframe based on bayesian criterion and device | |
Wang et al. | Fast image upsampling via the displacement field | |
CN107301661A (en) | High-resolution remote sensing image method for registering based on edge point feature | |
CN106204447A (en) | The super resolution ratio reconstruction method with convolutional neural networks is divided based on total variance | |
CN108961180B (en) | Infrared image enhancement method and system | |
Jalobeanu et al. | An adaptive Gaussian model for satellite image deblurring | |
CN103136734A (en) | Restraining method on edge Halo effects during process of resetting projections onto convex sets (POCS) super-resolution image | |
CN109801343A (en) | Based on annular artifact bearing calibration, the CT control system for rebuilding front and back image | |
Tao et al. | Spectrum-to-kernel translation for accurate blind image super-resolution | |
CN103632345A (en) | MRI image inhomogeneity correction method based on regularization | |
CN103020905B (en) | For the sparse constraint adaptive N LM super resolution ratio reconstruction method of character image | |
Zheng et al. | Automatic correction of intensity nonuniformity from sparseness of gradient distribution in medical images | |
CN110211193A (en) | Three dimensional CT interlayer image interpolation reparation and super-resolution processing method and device | |
EP2081150A1 (en) | Method of improving the resolution of a moving object in a digital image sequence | |
CN102831599B (en) | Registering method of medical image with uneven light and shade | |
CN106600539B (en) | A kind of single image super-resolution method based on part across scale autoregression model | |
Zhang et al. | Group-based sparse representation for Fourier ptychography microscopy | |
CN103020936A (en) | Super-resolution reconstruction method of facial image | |
CN112164006A (en) | Image color homogenizing method and device, electronic equipment and storage medium | |
Chan et al. | A comparison of superresolution reconstruction methods for multi-angle CHRIS/Proba images | |
CN107767342B (en) | Wavelet transform super-resolution image reconstruction method based on integral adjustment model | |
CN113506212B (en) | Improved hyperspectral image super-resolution reconstruction method based on POCS | |
CN113160271B (en) | High-precision infrared target tracking method integrating correlation filtering and particle filtering | |
CN112365420B (en) | Fuzzy image blind restoration method based on non-local window gradient | |
Jarolim et al. | Instrument-To-Instrument translation: Instrumental advances drive restoration of solar observation series via deep learning |
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 | ||
RJ01 | Rejection of invention patent application after publication | ||
RJ01 | Rejection of invention patent application after publication |
Application publication date: 20180209 |