CN108961163A - A kind of high-resolution satellite image super-resolution reconstruction method - Google Patents
A kind of high-resolution satellite image super-resolution reconstruction method Download PDFInfo
- Publication number
- CN108961163A CN108961163A CN201810686158.7A CN201810686158A CN108961163A CN 108961163 A CN108961163 A CN 108961163A CN 201810686158 A CN201810686158 A CN 201810686158A CN 108961163 A CN108961163 A CN 108961163A
- Authority
- CN
- China
- Prior art keywords
- image
- resolution
- point
- super
- reconstruction
- 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 100
- 239000011159 matrix material Substances 0.000 claims abstract description 46
- 230000006866 deterioration Effects 0.000 claims abstract description 9
- 238000004458 analytical method Methods 0.000 claims abstract description 8
- 230000008569 process Effects 0.000 claims description 21
- 238000003708 edge detection Methods 0.000 claims description 8
- 238000007476 Maximum Likelihood Methods 0.000 claims description 7
- 238000012935 Averaging Methods 0.000 claims description 6
- 230000009466 transformation Effects 0.000 claims description 6
- 238000004364 calculation method Methods 0.000 claims description 3
- 238000013461 design Methods 0.000 claims description 3
- 238000000605 extraction Methods 0.000 claims description 2
- 238000005457 optimization Methods 0.000 claims description 2
- 238000012545 processing Methods 0.000 abstract description 11
- 238000005070 sampling Methods 0.000 abstract description 11
- 238000005516 engineering process Methods 0.000 abstract description 6
- 230000000295 complement effect Effects 0.000 abstract description 4
- 230000015572 biosynthetic process Effects 0.000 abstract description 2
- 238000003786 synthesis reaction Methods 0.000 abstract description 2
- 238000003384 imaging method Methods 0.000 description 15
- 238000004422 calculation algorithm Methods 0.000 description 12
- 230000000694 effects Effects 0.000 description 10
- 238000010586 diagram Methods 0.000 description 6
- 238000013507 mapping Methods 0.000 description 6
- 230000003287 optical effect Effects 0.000 description 6
- 230000001133 acceleration Effects 0.000 description 4
- 230000010354 integration Effects 0.000 description 3
- 230000008901 benefit Effects 0.000 description 2
- 238000007796 conventional method Methods 0.000 description 2
- 230000003412 degenerative effect Effects 0.000 description 2
- 238000001914 filtration Methods 0.000 description 2
- 238000011084 recovery Methods 0.000 description 2
- 239000000654 additive Substances 0.000 description 1
- 230000000996 additive effect Effects 0.000 description 1
- 230000003321 amplification Effects 0.000 description 1
- 235000019606 astringent taste Nutrition 0.000 description 1
- 238000005452 bending Methods 0.000 description 1
- 230000009286 beneficial effect Effects 0.000 description 1
- 230000008859 change Effects 0.000 description 1
- 238000010276 construction Methods 0.000 description 1
- 238000006731 degradation reaction Methods 0.000 description 1
- 238000000151 deposition Methods 0.000 description 1
- 238000011161 development Methods 0.000 description 1
- 238000006073 displacement reaction Methods 0.000 description 1
- 238000009826 distribution Methods 0.000 description 1
- 230000002708 enhancing effect Effects 0.000 description 1
- 238000002474 experimental method Methods 0.000 description 1
- 230000006872 improvement Effects 0.000 description 1
- 238000004519 manufacturing process Methods 0.000 description 1
- 238000012986 modification Methods 0.000 description 1
- 230000004048 modification Effects 0.000 description 1
- 238000003199 nucleic acid amplification method Methods 0.000 description 1
- 238000012634 optical imaging Methods 0.000 description 1
- 238000003909 pattern recognition Methods 0.000 description 1
- 230000003014 reinforcing effect Effects 0.000 description 1
- 238000011160 research Methods 0.000 description 1
- 230000002441 reversible effect Effects 0.000 description 1
- 238000004088 simulation Methods 0.000 description 1
- 238000012360 testing method Methods 0.000 description 1
- 208000013021 vision distortion Diseases 0.000 description 1
- 230000000007 visual effect Effects 0.000 description 1
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T3/00—Geometric image transformations in the plane of the image
- G06T3/40—Scaling of whole images or parts thereof, e.g. expanding or contracting
- G06T3/4053—Scaling of whole images or parts thereof, e.g. expanding or contracting based on super-resolution, i.e. the output image resolution being higher than the sensor resolution
- G06T3/4076—Scaling of whole images or parts thereof, e.g. expanding or contracting based on super-resolution, i.e. the output image resolution being higher than the sensor resolution using the original low-resolution images to iteratively correct the high-resolution images
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T3/00—Geometric image transformations in the plane of the image
- G06T3/40—Scaling of whole images or parts thereof, e.g. expanding or contracting
- G06T3/4023—Scaling of whole images or parts thereof, e.g. expanding or contracting based on decimating pixels or lines of pixels; based on inserting pixels or lines of pixels
Landscapes
- Physics & Mathematics (AREA)
- General Physics & Mathematics (AREA)
- Engineering & Computer Science (AREA)
- Theoretical Computer Science (AREA)
- Image Processing (AREA)
Abstract
A kind of high-resolution satellite image super-resolution reconstruction method, it is related to high-definition picture processing technology field, solving existing image reconstruction method, there are image blur, distortion, lack sampling and noises etc., in addition, the problems such as in the presence of artificial processing error and the complementary information between multiframe image cannot be made full use of, a kind of high-resolution satellite image super-resolution reconstruction method is provided, including, 8 frame sequence low-resolution images are inputted, and original reconstruction image is constructed to the first frame of image;The fuzzy matrix of other frames and the geometric deformation matrix of the opposite first frame of other frames are calculated using Active view image deterioration analysis model, and down-sampled matrix is determined compared with the magnification ratio of first frame according to original reconstruction image;Using improved POCS super-resolution reconstruction method, residual computations formula, residual error constraint set and point projection operator are designed, amendment is iterated to original reconstruction image.The present invention improves image detail information significantly, and image synthesis quality is made to be greatly improved.
Description
Technical field
The present invention relates to high-definition picture processing technology fields, and in particular to a kind of high-resolution satellite image super-resolution
Method for reconstructing.
Background technique
Satellite image super-resolution rebuilding technology, which refers to push away a width low resolution using computer, sweeps image or video figure
As sequence is handled, a kind of image processing techniques of high-definition picture is recovered, high-definition picture has high pixel close
Degree, can provide more details, these details often play a key effect in subsequent application and development, such as target identification.
In great amount of images application field, people often it is expected to obtain high-definition picture, and high-definition picture means figure
The pixel density of picture is high, is capable of providing more details, and these details are indispensable in numerous applications, for remote sensing images
It also is in this way, allowing computer to regard for example, be just easy to distinguish similar image from homologue using high-resolution satellite image
Pattern-recognition performance in feel greatly improves.Since 1970s, charge-coupled device (CCD), cmos image are passed
Sensor has just been widely used for capturing digital picture, although these sensors are suitable for the application of most of image
, but still there are some applications to need higher resolution, image more high-definition.Due to the limitation of hardware, in imaging process
Imaging sensor pixel dimension can not ad infinitum reduce, such as charge-coupled device (CCD) pixel of -5 satellite of French SPOT
Size highest has had reached 12.5 μm, and the ground resolution of panchromatic wave-band is 5m.The improvement of remote sensor image system hardware
Demand of the remote sensing application to image spatial resolution is not always caught up with, increase sensor manufacturing cost cost simply is excessive, therefore
Seek low cost enhancing current resolution level method be it is very necessary, super resolution ratio reconstruction method comes into being.
Super-resolution rebuilding is handled single width or several low resolution images with complementary information, obtains one
The technology of width or several high resolution images, existing super resolution ratio reconstruction method are broadly divided into based on frequency domain and based on airspace
Super-resolution Reconstruction.It is usually the super-resolution rebuilding of Fourier transformation and wavelet transformation based on frequency domain super-resolution rate method for reconstructing
Method, such method advantage are that the speed of service is very fast, it is relatively easy but constant because being confined to linear space to calculate, and only include
The priori knowledge of limit could not become the mainstream research method of Super-resolution Reconstruction technology, the super resolution ratio reconstruction method based on airspace
Specifically include that interpolation method, regularization method (including L1 norm, L2 norm and TV regular terms), maximum a posteriori probability (maximum
A posterior MAP) method and method for reconstructing based on study, etc..Interpolation method makes because not increasing new high-frequency information
It obtains reconstructed results and the phenomenon that vision distortions such as sawtooth, fuzzy occurs;How properly to be provided in regularization method and keeps image side
The priori conditions of edge structure are still difficult point;Probability theory method improve reconstructed results cost be complexity it is higher, calculate cost compared with
Greatly;Method based on study is often handled single width remote sensing images, limits answering for different information between sequence remote sensing images
With.Conventional super-resolution method for reconstructing thinks that the first step of image deterioration process is geometric deformation, and using scene as coordinate system, scene becomes
It is dynamic to determine that geometric deformation, high-definition picture successively pass through geometric deformation, it obscures, lack sampling and a series of noise this process that degrade
Low-resolution image is finally obtained, this viewpoint is also referred to as Passive view.However the universal imaging pattern of video satellite is solidifying
Depending on imaging, refer to the movement with satellite, optical imaging system keeps a close watch on a certain target area always, can be observed continuously in visual field
Variation, Passive view is readily incorporated calculating error, main 3 reasons, first is that the satellite phase of staring imaging for staring imaging
The variation of machine posture is greater than the variation of scene, solves the geometric form that the geometric deformation matrix of camera posture is changed than solving scene
Bending moment battle array is accurate.Second is that Passive view solves geometric deformation matrix by the registration between high-definition picture, and high resolution graphics
As after original low-resolution image rises sampling, deblurring, there is artificial processing error.Third is that information utilization is poor, because by
The artificial high-definition picture of in-motion viewing point carries out convolution, but its usual right and wrong after geometric deformation after geometric deformation with fuzzy matrix
Integer displacement.
The method of Remote Sensing Image Quality is improved other than Super-resolution Reconstruction, there are also optical remote sensing image restored methods.Star
It carries, aerial remote sensing imaging system is by factors such as optical diffraction, atmospheric turbulance, CCD lack sampling, Platform Vibration, system noises
It influences, the detail of the high frequency more than imaging system cutoff frequency is lost, and image blur, aliasing, spatial resolution is caused to reduce.
Typical image restoration method has Wiener filtering, Richardson-Lucy filtering, the deconvolution of maximum likelihood iteration, least commitment
Two multiply recovery, maximum entropy restoration etc..The advantages of image recovery method is that imaging system priori knowledge can be made full use of, and super
Resolved reconstruction is compared, fast speed, and disadvantage is that the complementary information between multiframe image cannot be made full use of.
Summary of the invention
The present invention is that there are image blur, distortion, lack sampling and noises etc. for the existing image reconstruction method of solution, in addition, depositing
In artificially processing error and the problems such as the complementary information between multiframe image cannot be made full use of, a kind of high-resolution is provided and is defended
Star image super-resolution reconstruction method.
A kind of high-resolution satellite image super-resolution reconstruction method, this method are realized by following steps:
Step 1: 8 frame sequence low-resolution images of input, and original reconstruction image is constructed to the first frame of described image;
Step 2: calculating the fuzzy matrix and other frame phases of other frames using Active view image deterioration analysis model
Down-sampled matrix is determined compared with the magnification ratio of first frame to the geometric deformation matrix of first frame, and according to original reconstruction image;
Calculate the process of fuzzy matrix are as follows:
Fuzzy matrix is solved using inclination recognition status, primarily determines the marginal point in every a line according to edge detection results,
It is corresponding to every a line ESF data point as ESF sampled point by row sampled edge point two sides data point according to the marginal point acquired
Position is averaging, and obtains final ESF, calculus of finite differences obtains LSF data point, according to LSF data point calculation two-dimensional points spread function
Image, i.e. fuzzy matrix;
Step 3: design residual computations formula, residual error constraint set and point are thrown using improved POCS super-resolution reconstruction method
Shadow operator is iterated amendment to original reconstruction image, realizes the Super-resolution Reconstruction of high-resolution satellite image.
Beneficial effects of the present invention: the present invention analyzes image degradation process with Active view, and non-traditional Passive view divides
Analysis establishes the image deterioration model based on staring imaging feature, to traditional POCS method from the angle of optical satellite camera posture
It improves, furthermore traditional POCS (projections onto convex sets) method original reconstruction image compensation rule is improved, and with
Deconvolution method combines, and details reinforcing effect becomes apparent from compared with original method, using GPU (Graphics Processing
Unit, image processor) and hardware multithreading acceleration technique, greatly shorten the processing time.It is additionally real using GPU acceleration technique
Existing individual, which pushes away, sweeps image or video single frames image IBP (iterative backprojection) super-resolution reconstruction method.The present invention improves significantly
Image detail information makes the indexs such as image edge energy, comentropy, mtf value be significantly improved, maintains peak value
Signal-to-noise ratio makes image synthesis quality be greatly improved.
For traditional POCS method Passive view degrade analysis model be readily incorporated calculate error the shortcomings that, propose actively see
Point image degrades analysis model, from the angle of optical camera posture, establishes the image deterioration model based on staring imaging feature, this
One model that degrades has an important influence multiframe Super-resolution Reconstruction effect.
For the bad disadvantage of traditional POCS Super-resolution Reconstruction algorithm original reconstruction image estimation effect, maximum likelihood is changed
It is combined for deconvolution method with traditional interpolation method, substantially increases original reconstruction image estimation quality.
For higher, the time-consuming disadvantage of traditional POCS algorithm computation complexity, changed using GPU acceleration technique to maximum likelihood
For deconvolution, interframe registration, time-consuming calculate such as iterative compensation is optimized, and is greatly shortened and is calculated the time.
Detailed description of the invention
Fig. 1 is high-resolution satellite image super-resolution reconstruction method overall flow figure of the present invention;
Fig. 2 is the stream that original reconstruction image is constructed in high-resolution satellite image super-resolution reconstruction method of the present invention
Cheng Tu;
Fig. 3 is sword edge image effect picture in high-resolution satellite image super-resolution reconstruction method of the present invention;
Fig. 4 is to carry out side using Canny operator in high-resolution satellite image super-resolution reconstruction method of the present invention
The effect picture that edge detects;
Fig. 5 is ESF schematic diagram;
Fig. 6 is LSF schematic diagram;
Fig. 7 is LSF effect picture;
Fig. 8 a and Fig. 8 b are respectively to establish in high-resolution satellite image super-resolution reconstruction method of the present invention in Fig. 8
The flow chart and simulation drawing of residual error constraint set;
Fig. 9 a and Fig. 9 b are respectively to use high-resolution satellite image super-resolution reconstruction method weight of the present invention in Fig. 9
The front and back comparison diagram of image is built, Fig. 9 c and Fig. 9 d are respectively to use high-resolution satellite image Super-resolution Reconstruction of the present invention
The front and back comparison diagram of method reconstruction image.
Specific embodiment
Specific embodiment one illustrates present embodiment, high-resolution satellite image Super-resolution Reconstruction in conjunction with Fig. 1 to Fig. 9
Method, this method is that bilinear interpolation and deconvolution method is combined to construct original reconstruction image first, then in Active view figure
On the basis of the analysis model that degrades, geometric deformation matrix of other frames with respect to first frame is calculatedFuzzy matrix Bk, down-sampled
Matrix D.Then using POCS super-resolution reconstruction method is improved, residual computations formula is designed, residual error constraint set puts projection operator
Deng being iterated amendment to original reconstruction image according to formula.
Detailed process are as follows:
Step 1: 8 frame sequence low-resolution images of input, and original reconstruction image is constructed to the first frame of described image;
Step 2: calculating the fuzzy matrix and other frame phases of other frames using Active view image deterioration analysis model
Down-sampled matrix is determined compared with the magnification ratio of first frame to the geometric deformation matrix of first frame, and according to original reconstruction image;
Calculate the process of fuzzy matrix are as follows:
Fuzzy matrix is solved using inclination recognition status, primarily determines the marginal point in every a line according to edge detection results,
It is corresponding to every a line ESF data point as ESF sampled point by row sampled edge point two sides data point according to the marginal point acquired
Position is averaging, and obtains final ESF, calculus of finite differences obtains LSF data point, according to LSF data point calculation two-dimensional points spread function
Image, i.e. fuzzy matrix;
Step 3: design residual computations formula, residual error constraint set and point are thrown using improved POCS super-resolution reconstruction method
Shadow operator is iterated amendment to original reconstruction image, realizes the Super-resolution Reconstruction of high-resolution satellite image.
In present embodiment, degraded model according to Active view, satellite image imaging degrades, and to can be described as HR (ideal for process
High-definition picture) successively to pass through blurring processes, the sensor pixels such as atmosphere fuzzy, optical dimming, sensor be fuzzy close for image
Spend the lack sampling process determined, and the geometric deformation process determined by camera posture.Since, this paper satellite image is without obviously making an uproar
Sound, therefore do not consider additive noise.Finally, LR image sequence is obtained by sensor.This process that degrades can be formulated:
yk=MkDBkx(1≤k≤p) (1)
Wherein MkIt is the geometric deformation matrix of kth frame image;BkIt is the fuzzy matrix of kth frame image;D is lack sampling matrix;
P is the total number of LR image.Super-resolution rebuilding process is the inverse process of this processKnown LR image
Sequence is degraded Inverse Model based on video satellite staring imaging, is passed through algorithm and is rebuild ideal HR image.
In original reconstruction Image construction, uncoiling integration method, the image point spread function that deconvolution method is used joined
It is acquired using recognition status.Choose contrast in the picture first compared with strong, the lesser sword border region of surface noise, as shown in Figure 2.It adopts
Edge detection, which is carried out, with Canny operator obtains Fig. 3.
According to edge detection results, the marginal point in every a line is primarily determined, according to the marginal point acquired, by row sampling side
Edge point two sides data point, as ESF (edge-spread function) sampled point, ESF sampled point schematic diagram is as shown in Figure 4.
Every a line ESF data point corresponding position is averaging, final ESF is obtained.Next LSF is obtained using calculus of finite differences
(line spread function) data point, LSF schematic diagram are as shown in Figure 5.
According to the image point spread function acquired using the deconvolution of maximum likelihood iteration method to original reconstruction image into
Row restores.For a linear spatially invariant system, the degenerative process provided in the time domain can be provided by following formula:
G (x, y)=h (x, y) * f (x, y)+η (x, y)
Wherein, h (x, y) is expression of the degenrate function under time domain, and operator * indicates convolution.It can by convolution theorem
Know, the convolution in time domain is equal to the product on frequency domain, institute's above formula being expressed as follows in a frequency domain:
G (u, v)=H (u, v) F (u, v)+N (u, v)
Capitalization item therein be before in formula respective items Fourier transformation.Degenrate function H typically refers to obscure,
Shake etc. influences.If it is considered to the influence very little of noise, then formula above can be write as G (u, v)=H (u, v) F (u,
V), it is assumed that H (u, v) is reversible, that can obtain F (u, v)=G (u, v)/H (u, v).But actually noise is difficult in institute
Exempt from, thus can only try to find out the estimated value of F (u, v), replaces G (u, v) with H (u, v) F (u, v)+N (u, v) at this time, then
Have: this namely deconvolution of F^ (u, v)=(H (u, v) F (u, v)+N (u, v))/H (u, v)=F (u, v)+N (u, v)/H (u, v)
Basic principle.Maximum likelihood iteration deconvolution is that the process of f and h is solved from observed image g, by Bayes theory, if image
Noise obeys Poisson distribution, and available likelihood function is as follows:
Wherein WithIt is the estimation of PSF and image respectively.Maximum likelihood function L can be obtained byWith
It enablesImage and PSF estimation can be expressed from the next:
Uncoiling integration method, which is added, can obtain higher-quality original reconstruction image, provide base for successive iterations super-resolution
Plinth guarantees entire algorithm process result high quality.Uncoiling integration method is optimized using GPU and hardware multithread technology, is contracted
Short program calculates the time.
Geometric deformation matrix in present embodimentImage registration based on SIFT is broadly divided into three steps, point
It is not: the extraction of image characteristic point, the matching of characteristic point, the solution of the transformation matrix between image.
Fuzzy matrix B described in present embodimentkFuzzy matrix B is solved using inclination recognition statusk.Tilt sword edge image such as
Fig. 3 primarily determines the marginal point in every a line according to edge detection results Fig. 4, according to the marginal point acquired, by row sampling side
Edge point two sides data point is averaging every a line ESF data point corresponding position, obtains as ESF (edge-spread function) sampled point
To final ESF, as shown in Figure 5.LSF (line spread function) data point is obtained using calculus of finite differences, as shown in fig. 6, according to LSF number
According to calculating two-dimensional points spread function image, i.e. fuzzy matrix Bk;
Down-sampled matrix D described in present embodiment: the times magnification of this paper high-definition picture phase lower resolution image
Number is 2 times, therefore down-sampled matrix is D,
Improved POCS Super-resolution Reconstruction algorithm is used in present embodiment step 3, POCS algorithm is designed into convex set reason
By Convex Set Theory is that a kind of important signal reconstruction is theoretical, is the mathematical theory basis of a major class iterative algorithm.The present invention passes through
Mapping relations between LR image sequence and HR image establish residual error constraint set and respective point projection operator based on the process that degrades,
Rebuild video satellite image.
Residual error constraint set defined in present embodiment is referred to BkIt is pilot process, wherein B between DkIt is kth frame
The fuzzy matrix of image, D are lack sampling matrix, it would be desirable to which gray value of the HR (high-definition picture) after fuzzy matrix is regarded as
The analogue value, the gray value approximation obtained during HR mesh mapping is to LR grid using interpolation method regard true value as, between
Difference is limited within the scope of some constant δ, establishes residual error constraint set.
Firstly, calculating the residual error analogue value.Ideal HR image becomes B by fuzzy matrixkx。
Secondly, calculating residual error true value.Establish HR grid xgrid=[n1, n2]TWith each frame LR gridBetween mapping, wherein [n1, n2] it is high-definition picture coordinate,For the low resolution of kth frame
Rate image coordinate.True value is solved by mapping relations, this mapping relations can be formulated as:
yk-grid=MkDxgrid (5)
The geometric transformation between LR image that image registration acquires can be passed through in the actual processAnd amplification factor
Liter sampling matrix D-1, LR mesh coordinate xgridBeing mapped to kth frame LR mesh coordinate is yk-grid:
Because the point on LR grid is all true value, HR mesh mapping can be approximately considered to LR grid and use bilinear interpolation
Obtained gray value is also true value, and value can be formulated as:
Finally, new algorithm residual error is as follows:
New algorithm residual error constraint set can indicate as follows:
Wherein, the determination of constant δ is highly important, and selection is excessive to will lead to constraint set poor astringency, too small to also result in
Numerical problem makes method for reconstructing unstable.We have carried out convergence test to the estimation of δ value in an experiment, after choosing convergence
δ, rebuild effect show we choose δ have certain portability, constant δ is selected as 1.5 in invention.
Projection operator corresponding with constraint set is established, changes the value of high-definition picture, is limited in convex set, no
Disconnected iteration finds optimal solution.
The projection operator takes a mode for projection, by the value of each point of amendment high-definition picture, is limited
System is on residual error constraint set, to obtain the high-definition picture of resolution ratio raising.High-definition picture x [n1, n2] project to residual error
Constraint setOn the process of new high-definition picture can be expressed as follows with formula:
Wherein, (p, q) is fuzzy matrix size, by calculating the difference of residual error and constraint with the fuzzy matrix weight
Product corrects the point outside residual error constraint set, and each frame is all passed through projection operator and is projected on the constraint set, constantly updates high score
The value of resolution image, to the last a frame.Block projection is much carried out to the region (p, q) based on Convex Set Theory algorithm, so once
Projection modification inaccuracy, just will affect entire block region, cause mosaic effect, overlapping artifact phenomenon.Present invention overcomes
This disadvantage.
In present embodiment, in terms of single frames super-resolution, the present invention has carried out GPU optimization to traditional IBP algorithm, for dividing
Resolution is that the super large remote sensing images of 12K*5K shortened to 4.5 minutes from 50 minutes processing time.For multiframe super-resolution, for biography
Deconvolution technique is added in the bad disadvantage of system method original reconstruction image estimation effect, and it is excellent to carry out GPU to deconvolution method
Change, original reconstruction image is restored using the image point spread function acquired, so that original reconstruction image estimation effect is remote
It is better than conventional method, provides good basis for subsequent processing.Image deterioration model is analyzed using Passive view for conventional method,
For the satellite video of staring imaging mode, it is readily incorporated the shortcomings that calculating error, image is analyzed with Active view for the first time
Degenerative process establishes the image deterioration model based on staring imaging feature, according to the model that degrades from the angle of optical camera posture
Push over residual computations formula, residual error constraint set, point projection operator etc..It is longer that the time is calculated for traditional multiframe super resolution algorithm
Disadvantage is optimized using multithreading and GPU acceleration technique, for super large video single frames that resolution ratio is 12K*5K from average
450 minutes processing time shortened to 41 minutes.
Claims (3)
1. a kind of high-resolution satellite image super-resolution reconstruction method, characterized in that this method is realized by following steps:
Step 1: 8 frame sequence low-resolution images of input, and original reconstruction image is constructed to the first frame of described image;
Step 2: calculating the fuzzy matrix and other frames opposite the of other frames using Active view image deterioration analysis model
The geometric deformation matrix of one frame, and down-sampled matrix is determined compared with the magnification ratio of first frame according to original reconstruction image;
Calculate the process of fuzzy matrix are as follows:
Fuzzy matrix is solved using inclination recognition status, primarily determines the marginal point in every a line according to edge detection results, according to
The marginal point acquired, by row sampled edge point two sides data point, as ESF sampled point, to every a line ESF data point corresponding position
It being averaging, obtains final ESF, calculus of finite differences obtains LSF data point, according to LSF data point calculation two-dimensional points spread function image,
That is fuzzy matrix;
Step 3: design residual computations formula, residual error constraint set and point projection are calculated using improved POCS super-resolution reconstruction method
Son is iterated amendment to original reconstruction image, realizes the Super-resolution Reconstruction of high-resolution satellite image.
2. a kind of high-resolution satellite image super-resolution reconstruction method according to claim 1, which is characterized in that step 1
In, using bilinear interpolation and deconvolution method and GPU optimization is carried out, using the image point spread function acquired to original reconstruction
Image is restored;
The image point spread function that the deconvolution method is used is acquired using recognition status;
It chooses sword border region in the picture first, edge detection is carried out using Canny operator, according to edge detection results, tentatively
The marginal point in every a line is determined, according to the marginal point acquired, by row sampled edge point two sides data point, as ESF sampled point;
Every a line ESF data point corresponding position is averaging, final ESF is obtained;LSF data point is obtained using calculus of finite differences, according to asking
The image point spread function obtained restores original reconstruction image using the method for maximum likelihood iteration deconvolution.
3. a kind of high-resolution satellite image super-resolution reconstruction method according to claim 1, which is characterized in that described several
The calculating of what deformation matrix is based on the image registration of SIFT, the extraction including image characteristic point, the matching of characteristic point, between image
The solution of transformation matrix.
Priority Applications (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201810686158.7A CN108961163A (en) | 2018-06-28 | 2018-06-28 | A kind of high-resolution satellite image super-resolution reconstruction method |
Applications Claiming Priority (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201810686158.7A CN108961163A (en) | 2018-06-28 | 2018-06-28 | A kind of high-resolution satellite image super-resolution reconstruction method |
Publications (1)
Publication Number | Publication Date |
---|---|
CN108961163A true CN108961163A (en) | 2018-12-07 |
Family
ID=64487347
Family Applications (1)
Application Number | Title | Priority Date | Filing Date |
---|---|---|---|
CN201810686158.7A Pending CN108961163A (en) | 2018-06-28 | 2018-06-28 | A kind of high-resolution satellite image super-resolution reconstruction method |
Country Status (1)
Country | Link |
---|---|
CN (1) | CN108961163A (en) |
Cited By (7)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN110113545A (en) * | 2019-04-25 | 2019-08-09 | 中国人民解放军61646部队 | A kind of imaging method of MM-CCD |
CN110717859A (en) * | 2019-10-09 | 2020-01-21 | 哈尔滨工业大学 | Super-resolution reconstruction method based on two-way video |
CN111553958A (en) * | 2019-02-08 | 2020-08-18 | 西门子医疗有限公司 | Calculation of image matrix size |
CN111667437A (en) * | 2019-03-06 | 2020-09-15 | 中国科学院微电子研究所 | Remote sensing image space-time fusion method based on Fourier domain unmixing |
CN112037136A (en) * | 2020-09-18 | 2020-12-04 | 中国科学院国家天文台南京天文光学技术研究所 | Super-resolution imaging method based on aperture modulation |
CN112488920A (en) * | 2020-12-02 | 2021-03-12 | 浙江工业大学 | Image regularization super-resolution reconstruction method based on Gaussian-like fuzzy kernel |
CN113962897A (en) * | 2021-11-02 | 2022-01-21 | 中国空间技术研究院 | Modulation transfer function compensation method and device based on sequence remote sensing image |
Citations (5)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN101980289A (en) * | 2010-10-25 | 2011-02-23 | 上海大学 | Frequency domain registration and convex set projection-based multi-frame image super-resolution reconstruction method |
CN102800071A (en) * | 2012-06-20 | 2012-11-28 | 南京航空航天大学 | Method for reconstructing super resolution of sequence image POCS |
CN103679652A (en) * | 2013-11-29 | 2014-03-26 | 北京空间机电研究所 | Image restoration system capable of improving imaging quality greatly |
CN104732578A (en) * | 2015-03-10 | 2015-06-24 | 山东科技大学 | Building texture optimization method based on oblique photograph technology |
US20170102442A1 (en) * | 2015-10-08 | 2017-04-13 | The General Hospital Corporation | Systems and methods for generalized slice dithered enhanced resolution magnetic resonance imaging |
-
2018
- 2018-06-28 CN CN201810686158.7A patent/CN108961163A/en active Pending
Patent Citations (5)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN101980289A (en) * | 2010-10-25 | 2011-02-23 | 上海大学 | Frequency domain registration and convex set projection-based multi-frame image super-resolution reconstruction method |
CN102800071A (en) * | 2012-06-20 | 2012-11-28 | 南京航空航天大学 | Method for reconstructing super resolution of sequence image POCS |
CN103679652A (en) * | 2013-11-29 | 2014-03-26 | 北京空间机电研究所 | Image restoration system capable of improving imaging quality greatly |
CN104732578A (en) * | 2015-03-10 | 2015-06-24 | 山东科技大学 | Building texture optimization method based on oblique photograph technology |
US20170102442A1 (en) * | 2015-10-08 | 2017-04-13 | The General Hospital Corporation | Systems and methods for generalized slice dithered enhanced resolution magnetic resonance imaging |
Non-Patent Citations (4)
Title |
---|
姚烨 等: "凝视成像降质模型的超分辨率重建", 《光学学报》 * |
师雪艳: "模糊图像自适应复原算法研究", 《中国优秀硕士学位论文全文数据库 信息科技集》 * |
李铁成: "基于倾斜刃边法的遥感图像调制传递函数计算及图像复原技术研究", 《中国博士学位论文全文数据库 信息科技辑》 * |
聂涛: "基于最大后验概率的视频超分辨率算法研究", 《中国优秀硕士学位论文全文数据库 信息科技集》 * |
Cited By (11)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN111553958A (en) * | 2019-02-08 | 2020-08-18 | 西门子医疗有限公司 | Calculation of image matrix size |
CN111553958B (en) * | 2019-02-08 | 2023-11-03 | 西门子医疗有限公司 | Calculation of image matrix size |
CN111667437A (en) * | 2019-03-06 | 2020-09-15 | 中国科学院微电子研究所 | Remote sensing image space-time fusion method based on Fourier domain unmixing |
CN110113545A (en) * | 2019-04-25 | 2019-08-09 | 中国人民解放军61646部队 | A kind of imaging method of MM-CCD |
CN110113545B (en) * | 2019-04-25 | 2021-05-11 | 中国人民解放军61646部队 | MM-CCD imaging method |
CN110717859A (en) * | 2019-10-09 | 2020-01-21 | 哈尔滨工业大学 | Super-resolution reconstruction method based on two-way video |
CN112037136A (en) * | 2020-09-18 | 2020-12-04 | 中国科学院国家天文台南京天文光学技术研究所 | Super-resolution imaging method based on aperture modulation |
CN112037136B (en) * | 2020-09-18 | 2023-12-26 | 中国科学院国家天文台南京天文光学技术研究所 | Super-resolution imaging method based on aperture modulation |
CN112488920A (en) * | 2020-12-02 | 2021-03-12 | 浙江工业大学 | Image regularization super-resolution reconstruction method based on Gaussian-like fuzzy kernel |
CN112488920B (en) * | 2020-12-02 | 2024-05-10 | 浙江工业大学 | Image regularization super-resolution reconstruction method based on Gaussian-like fuzzy core |
CN113962897A (en) * | 2021-11-02 | 2022-01-21 | 中国空间技术研究院 | Modulation transfer function compensation method and device based on sequence remote sensing image |
Similar Documents
Publication | Publication Date | Title |
---|---|---|
CN108961163A (en) | A kind of high-resolution satellite image super-resolution reconstruction method | |
CN105550993B (en) | Super resolution ratio reconstruction method based on multiple transform domain | |
JP2000188680A (en) | Method and system for producing high resolution image | |
CN108288256B (en) | Multispectral mosaic image restoration method | |
CN108090869A (en) | Super resolution ratio reconstruction method on a kind of star based on face battle array CMOS optical cameras | |
CN106846250B (en) | Super-resolution reconstruction method based on multi-scale filtering | |
Anger et al. | Fast and accurate multi-frame super-resolution of satellite images | |
CN104574338A (en) | Remote sensing image super-resolution reconstruction method based on multi-angle linear array CCD sensors | |
CN106920213B (en) | Method and system for acquiring high-resolution image | |
CN107194874B (en) | Super-resolution imaging system and method based on bias image stabilization | |
CN116883235A (en) | Distributed photoelectric oriented image stitching method and device | |
Angelopoulou et al. | FPGA-based real-time super-resolution on an adaptive image sensor | |
CN107767342B (en) | Wavelet transform super-resolution image reconstruction method based on integral adjustment model | |
Malczewski et al. | Super resolution for multimedia, image, and video processing applications | |
Zhang et al. | Video superresolution reconstruction using iterative back projection with critical-point filters based image matching | |
Mohan | Adaptive super-resolution image reconstruction with lorentzian error norm | |
Molini et al. | Deep learning for super-resolution of unregistered multi-temporal satellite images | |
JP2013069012A (en) | Multi-lens imaging apparatus | |
Seema et al. | Multi-frame image super-resolution by interpolation and iterative backward projection | |
CN112488920A (en) | Image regularization super-resolution reconstruction method based on Gaussian-like fuzzy kernel | |
Dellaert et al. | Mosaicing a large number of widely dispersed, noisy, and distorted images: A bayesian approach | |
Gunturk | Super-resolution imaging | |
CN108681988A (en) | A kind of image resolution ratio Enhancement Method of the robust based on multiple image | |
Patel Shreyas | Novel Iterative Back Projection Approach | |
Shi et al. | Dark Channel Based Multiframe Super-Resolution Reconstruction |
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: 20181207 |