Research Article Research on Adaptive Optics...
Transcript of Research Article Research on Adaptive Optics...
Research ArticleResearch on Adaptive Optics Image Restoration Algorithm byImproved Expectation Maximization Method
Lijuan Zhang12 Dongming Li3 Wei Su4 Jinhua Yang1 and Yutong Jiang15
1 School of Opto-Electronic Engineering Changchun University of Science and Technology Changchun 130022 China2 College of Computer Science and Engineering Changchun University of Technology Changchun 130012 China3 School of Information Technology Jilin Agriculture University Changchun 130118 China4 Informatization Center Changchun University of Science and Technology Changchun 130022 China5 College of Electrical and Mechanical Engineering Beijing Institute of Technology Beijing 100081 China
Correspondence should be addressed to Wei Su swcusteducn
Received 18 April 2014 Revised 5 June 2014 Accepted 6 June 2014 Published 6 July 2014
Academic Editor Fuding Xie
Copyright copy 2014 Lijuan Zhang et al This is an open access article distributed under the Creative Commons Attribution Licensewhich permits unrestricted use distribution and reproduction in any medium provided the original work is properly cited
To improve the effect of adaptive optics imagesrsquo restoration we put forward a deconvolution algorithm improved by the EMalgorithm which joints multiframe adaptive optics images based on expectation-maximization theory Firstly we need to makea mathematical model for the degenerate multiframe adaptive optics images The function model is deduced for the points thatspread with time based on phase errorThe AO images are denoised using the image power spectral density and support constraintSecondly the EM algorithm is improved by combining the AO imaging system parameters and regularization technique A costfunction for the joint-deconvolution multiframe AO images is given and the optimization model for their parameter estimationsis built Lastly the image-restoration experiments on both analog images and the real AO are performed to verify the recoveryeffect of our algorithm The experimental results show that comparing with the Wiener-IBD or RL-IBD algorithm our iterationsdecrease 143 and well improve the estimation accuracy The model distinguishes the PSF of the AO images and recovers theobserved target images clearly
1 Introduction
Adaptive optics (abbreviated as AO) is developed in orderto overcome the interference of atmospheric turbulence onthe telescope imaging and it can measure and correct theoptical wavefront phase of atmospheric turbulence in real-time The technology matures its application is extendingto civilian fields besides the fields of large telescopes andlaser engineering [1] In recent years with the development ofadaptive optics technology many large ground-based opticaltelescopes are equipped with adaptive optics system [2] in theworld
When surface-to-air object imaging (eg astral target) isobserved by ground-based optical telescopes adaptive opticssystem can only realize part of the wavefront error correctionand the information of high frequency on target suppressionand attenuation seriously [3 4] There are many factors such
as atmospheric turbulence the AO system error and photonnoise on imaging path Therefore we acquire more clearimages carrying on image restoration before adaptive opticscorrection At present there aremany image restoration algo-rithms to overcome the influence of atmospheric turbulencesuch as iterative blind deconvolution algorithm (abbreviatedas IBD) [4 5] IBD based on Richardson-Lucy iteration(abbreviated as RL-IBD) IBD based on theWiener algorithm(abbreviated as Wiener-IBD) and the algorithm based onthe expectationmaximization algorithm (abbreviated as EM)[6 7]
This paper proposes a blind image restoration method bymultiframe iteration with improved adaptive optical imagesbased on expectation maximization algorithm Section 2introduces the method of adaptive optics image restorationin detail including constructing adaptive optics model ofimaging process deducingmodel of point spread function on
Hindawi Publishing CorporationAbstract and Applied AnalysisVolume 2014 Article ID 781607 10 pageshttpdxdoiorg1011552014781607
2 Abstract and Applied Analysis
Atmospheric turbulence
DM
WFS
Controller
Separating slide
Corrected wavefront
Scientific camera
Flat wavefrontDeformed wavefront
Telescopersquos pupil
Figure 1 Works of adaptive optical system for imaging compensa-tion
AO image and researching the denoise processingmethod ofAO image Finally it is given a deconvolution method usingmultiframe AO images by combining AO imaging systemparameters with the regularization technique on the basis ofexpectation maximization algorithm Section 3 verifies thevalidity of the algorithm in this paper by using simulate AOimages actual observation binary AO images and stellar AOimages Section 4 concludes the paper
2 Restoration Algorithm ofAdaptive Optics Images
Adaptive optics system can measure control and correct theerror of optical wavefront in real-time Figure 1 shows theprinciple diagram of an adaptive optical system for imagingcompensationThe system includes wavefront sensor (abbre-viated as WFS) the deformable mirror (abbreviated as DM)high speed processor for wavefront and other parts Thesystem can measure the distortion of optical wavefront inreal-time which is caused by the influence of atmosphericturbulence and convert it to corresponding controlling signaladding to the wavefront correction element which cancompensate distortion of wavefront in real-time and theoptical telescope would be able to get close to target imagingfor diffraction AO images are corrected by AO system butits compensation or correction is not sufficient in order toget clearer target images AO images restoration algorithm isresearched
21 The Imaging Process of Adaptive Optics System Con-sidering the normal optical imaging model first of all theproper adaptive optics imagingmodel is set up which lays the
foundation of restoring high-quality adaptive optical imagesThe acquisition process of adaptive optics image model canconsist of a degradation function and an additive noise[5] the imaging process of AO in spatial frequency can beexpressed as follows
119892 (119909) = (119891 lowast ℎ) (119909) + 119899 (119909)
= intΩ
119891 (119910) ℎ (119909 minus 119910) 119889119910 + 119899 (119909) 119909 119910 isin Ω
(1)
where 119891(119909) is the original image for energy concentrationℎ(119909) is the point spread function (abbreviated as PSF) orimpulse response for image system 119892(119909) is output observa-tion images for the AO system 119899(119909) is the additive noise andΩ is the image area
In application we use multiple frames of short exposureimages 119892
119898119872
119898=1for one target to restore the target image 119891
119898 is the number of frames The model of multiframe shortexposure image with degradation can be written as
119892119898 (119909) = 119891 (119909) lowast ℎ119898 (119909) + 119899 (119909) 1 lt 119898 le 119872 (2)
where 119892119898(119909) and ℎ
119898(119909) represent the 119898th frame of short
exposure AO image with degradation for the target andits corresponding PSF respectively This paper adopts theselection techniques of frames referred to in [6] as the imagepreprocessing which picks up the images with better qualityfrom image sequence of short exposure images in order toimprove the solution of blind convolution and convergenceof iteration
22 Model of Point Spread Function on AO Images In adap-tive optics system the residual error of wavefront correctionmainly consists of incomplete correction error by turbulenceand noise of closed-loop system
When AO system works in closed-loop the transfererror function is the transfer function of wavefront phaseperturbation caused by atmospheric turbulence and thenoise of system introduced bywavefront sensor is closed-looptransfer function [4] Theoretically the mathematical modelof the PSF after closed-loop correction of adaptive opticssystem is shown as follows
ℎ (119910) =
1003816100381610038161003816100381610038161003816int 119875(119906)119890
minus119895sdot(2120587120582119897)119910sdot119906119889119906
1003816100381610038161003816100381610038161003816
2
(3)
In the formula
119875 (119906) = 1 within the lens aperture0 others
(4)
is the pupil function of telescope 119906 is the coordinates for thepupil plane 120582 is the central wavelength of imaging beam 119897 isthe focal length of the optical system 119910 is the coordinates oftwo-dimensional focal plane
Due to the error of focal length and optical deviationin optical system the pupil function should be rectified asfollows
119901 (119906) = 119875 (119906) 119890119895120579(119906) (5)
Abstract and Applied Analysis 3
where 120579(sdot) is the wavefront phase error caused by the error offocal length or optical deviation
So the mathematical model of PSF can be written as
ℎ (119910 120579) =
1003816100381610038161003816100381610038161003816int 119901(119906)119890
minus119895sdot(2120587120582119897)sdot119910sdot119906119889119906
1003816100381610038161003816100381610038161003816
2
=
1003816100381610038161003816100381610038161003816int 119875(119906)119890
119895120579(119906)119890minus119895sdot(2120587120582119897)sdot119910sdot119906
119889119906
1003816100381610038161003816100381610038161003816
2
(6)
The wavefront phase error is caused by the focal lengtherror or optical deviation due to the influence of atmosphericturbulence the PSFmodel changes with the focal length erroror optical deviation over time as follows
ℎ (119910 120579119905) =
1003816100381610038161003816100381610038161003816int 119875(119906)119890
119895120579119905(119906)119890minus119895sdot(2120587120582119897)sdot119910sdot119906
119889119906
1003816100381610038161003816100381610038161003816
2
(7)
In the formula 120579119905(sdot) is the phase error caused by atmospheric
turbulence changing over timeIn terms of AO system the exposure time of observing
images is shorter than turbulent fluctuation If the intensity ofthe target is kept unchanged during exposure time themodelfor series images is
119892 (119910 120579119905119898 119891) = int ℎ (119910 minus 119909 120579
119905119898) 119891 (119909) 119889119909 (8)
In the formula 120579119905119898
is the phase error of turbulence for the119898th frame short exposure AO image
23 The Denoising Processing of AO Images In order toreduce the noise of AO images Chinese researchers putforward themethod to suppress noise based on total variation[8 9] This paper proposes a method to suppress noise that isbased on the power spectral density (abbreviated as PSD) ofimage and the supporting domain field of constraints imagesReference [10] analyzed the method to minimize the noisein symmetrical support domain field of AO images Theanalysis for this method is based on the variance ratio of realimages In this paper the assessment about denoise resultusing noise changed function The denoising process of AOimage is actual a kind of image noise corrected process inFourier spectral domain If theAO image has the symmetricalsupport field its noise mainly consist by CCD read-in noiseand AO noise and then noise power spectral density can beresearched Therefore suppress noise for observation images119892(119909) and 119892
119888(119909) is constrained by field of support domain
119892119888 (119909) = 119892 (119909)119908 (119909) (9)
In the formula 119908(119909) is weighting function in support fieldthe image spectrum 119866
119888(119906) is
119866119888 (119906) = int119866 (V)119882 (119906 minus V) 119889119906 (10)
where 119906 is the spatial frequency of image119866(119906) = 119878(119906)lowast119873(119906)119878(119906) is the signal frequency spectrum function 119873(119906) is thenoise spectrum function and bandwidth function 119882(119906) =119866(119906) lowast 120575 In the limited support domain field119882(119906) is verynarrow which nearly equals the 120575 times of the width of 119866(119906)
According to the definition of J M Conan the PSD ofan image is the Fourier transformation of its covariancedefining the power spectrum of noise and power spectrumof distortion image [11]
PSD(119906)119899119899 = ⟨10038161003816100381610038161003816Re 119866 (119906) minus Re 119866 (119906)10038161003816100381610038161003816
2
⟩
PSD(119906)119891119891 = ⟨10038161003816100381610038161003816Im 119866 (119906) minus Im 119866 (119906)10038161003816100381610038161003816
2
⟩
(11)
In the formula 119866(119906) is the Fourier transformation of image119892(119909) 119866(119906) is the mean value of sample 119866(119906)
Regard the circle as target supporting domain field itsdiameter is from 8 pixels to 185 pixels The definition of thechanging image noise function 119865
119899(119866) is
119865119899 (119866) =
int (PSD(119906)119888119899119899minus PSD(119906)119888
119899119899) 119889119906
intPSD(119906)119888119899119899119889119906
(12)
In the formula PSD(119906)119888119899119899is the power spectral density of the
image with constraints of support domain field and PSD(119906)119888119899119899
is the power spectral density of the image without constraintsof support domain field
24 Improved Expectation Maximization Algorithm Refer-ence [12] deduced the point spread function of multiframedegraded images with turbulence and the iteration formulaof target image based on maximum likelihood criterion butusing thismethod leads to the larger solution space Howeverwe put forward the technique of regularization to modify themaximum likelihood estimation algorithm introduce con-straints in accordance with AO image in physical reality andadopt the improved EM algorithm onmultiframe iteration ofAO images to acquire ideal 119891 and ℎ
119896
241 EM Algorithm by Combining Regularization TechniqueEM algorithm is put forward by Dempster et al [13] forparameters of maximum likelihood estimate of the iterativealgorithm which is iteratively going on two basic stepscalculation steps E (Expectation) and M (Maximization) Inthis paper a new regularizationmethod combined with priorknowledge is applied to the EMalgorithm realizing amethodof multiframe AO image restoration
Based on prior knowledge of target image and PSF jointprobability distribution function defining the maximizationfunction will get target image estimation 119891 and PSF estima-tion ℎ that is
[119891 ℎ] = max119891ℎ
(119901 (119892 | 119891 ℎ 119886)) = max119891ℎ
(exp(minus119869 (119891 ℎ))
119911)
(13)
In the formula 119911 is the normalization factor The costfunction is
[119891 ℎ] = max119891ℎ
(119901 (119892 | 119891 ℎ 119886)) = max119891ℎ
(exp(minus119869 (119891 ℎ))
119911)
(14)
4 Abstract and Applied Analysis
So the problem is converted to minimize cost function119869(119891 ℎ) that is [119891 ℎ] = min
119891ℎ(119869(119891 ℎ)) The cost function
of AO images and the optimization model of its parameterestimation are shown as follows
119869 (119891 ℎ) = 119869119899(119892 | 119891 ℎ 119886) + 119869
ℎ (ℎ | 119886) + 119869119891 (119891 | 119886)
st 119869119899(119892 | 119891 ℎ 119886) = sum
119894119895
1
1205902119899
[119892 (119894 119895) minus (ℎ lowast 119891) (119894 119895)]2
119869ℎ (ℎ | 119886) =
120579119905
2sum
119906
10038161003816100381610038161003816(119906) minus 119867(119906)
10038161003816100381610038161003816
2
PSDℎ (119906)
119869119891(119891 | 119886) = 120572sum
119894119895
Φ(119903 (119891 120574 (119894 119895)))
(15)
Symbol explanation the first item represents the noise costfunction the second item represents the constraints costfunction of PSF and the third item is the cost functionswith target edge constraint 119886 is prior information 1205902
119899is the
variance of mixed noise 120579119905is the wavefront phase error
the power spectral density of the point spread function isPSD(119906)
ℎ= ⟨|(119906) minus 119867(119906)|
2
⟩ (119906) is the Fourier transfor-mation of ℎ(119909) 119867(119906) is the Fourier transform of averagePSF the estimation of PSD
ℎ(119906) can be estimated through the
AO system closed-loop control data and 120572 and 120574(nabla119909) areregularization adjustment parameters using the target edgeconstraint theory of Mugnier isotropic [14] as
Φ(119901) = 119901 minus ln (1 + 119901)
119901 (119891 120574 (nabla119909)) =
10038171003817100381710038171003817nabla119891 (119894 119895)
10038171003817100381710038171003817
120574 (nabla119909)
10038171003817100381710038171003817nabla119891 (119894 119895)
10038171003817100381710038171003817= ((nabla119891
119909(119894 119895))
2+ (nabla119891
119910(119894 119895))
2
)
12
(16)
where 120572 is a constant coefficient 120574(sdot) is a regularized functionthat has variable regularization coefficient associated with thegradient of each point 120574(nabla119909) = expminus|nabla119909|
221205852
is based on [15]0 lt 120574(nabla119909) le 1 parameter 120585 is the controlled smoothingcoefficient for selection and nabla119909 is the gradient amplitudein four directions for the point 119909(119894 119895) (where 119894 119895 are pixelcoordinates and 119909 is gray value of pixel)
This paper will apply all the prior information on theestimation of target and PSF in order to get the optimalestimation of target and the point spread function (PSF)Combine formulas (2) (8) and (15) to establish the costfunction for joint deconvolutionwithmultiframeAO images
119869multi (119891 ℎ) = 119869119899 (119892119898 | 119891 ℎ119898 119886) + 119869ℎ (ℎ119898 | 119886) + 119869119891 (119891 | 119886)
(17)
where119898 is the number of frames
242 Implement of Image Restoration Algorithm This paperis combining with the basic idea of the EM algorithm [13]through limited times of cycle calculations find the mostsuitable PSF estimation on the purpose of restoring target
image drastically modify the cost function (17) and make itcorrespond to full data [15] if 119892
119896(119910 | 119909) is full data set of 119896
times observation results 119892119896(119910) of observed image 119892
119896(119910) and
the expectation of cost function are as follows
119864 (119892119896(119910)) = 119864(sum
119909
119892119896(119910 | 119909))
= sum
119909
ℎ119896(119910 minus 119909) 119891 (119909)
119864 (119869multi (119891 ℎ)) = 119864(119869119899 (sum119909
119892119896| 119891 ℎ119898 119886))
+ 119864 (119869ℎ(ℎ119898| 119886)) + 119864 (119869
119891(119891 | 119886))
(18)
In the iteration process it is supposed that the 119899 minus1 iteration result is known 119864119899(119869multi(119891 ℎ)) represents thecurrent expectation of the cost function In order tominimizethe expectation of the cost function formula (18) can be dif-ferentiated by ℎ119899
119896(119909) and 119891119899(119909) then make the differentiation
equal to zero that is
120597119864119899(119869multi (119891 ℎ))
120597ℎ119899
119896(119909)
= 0120597119864119899(119869multi (119891 ℎ))
120597119891119899 (119909)= 0 (19)
The derivation process in detail is omitted so the finalsolutions are
119891119899+1
119894(119909) = 119891
119899
119894(119909)
+120579119905
120572 lowast Φ (119903 (119891119899
119894(119909) 120574 (119909)))
times sum
119910isin119884
sum
119898
ℎ119899
119894(119910 minus 119909)
times (119892119898minus sum119911isin119883ℎ119899
119894(119910 minus 119911) 119891
119899
119894(119911)
1205902119894
)
(20)
ℎ119899+1
119894(119909) = ℎ
119899
119894(119909)
+120579119905
2sum
119906
10038161003816100381610038161003816119894(119906) minus 119867
119894(119906)10038161003816100381610038161003816
2
119875119878119863ℎ 119894 (119906)
times sum
119910isin119884
119891119899+1
119894(119910 minus 119909)
times (119892119898minus sum119911isin119883ℎ119899
119894(119910 minus 119911) 119891
119899
119894(119911)
1205902119894
)
(21)
In this paper the specific steps of image restorationalgorithm are as follows
Step 1 (initialization operation)
(1) According to the method of Section 23 constrainthe support domain field of observed image and dealwith the noise and so forth regard the result of onetime Wiener filtering as the initial value of targetestimation119891
0
Abstract and Applied Analysis 5
(a) Original image (b) Degraded image frame 1 (c) Degraded image frame 2 (d) Degraded image frame 3
(e) Restored image by Wiener-IBD
(f) Restored image by RL-IBD (g) Restored image by our algo-rithm
Figure 2 Degraded images in experiment and restoration comparison of three algorithms
(a) Degraded image frame 1 (b) Degraded image frame 2 (c) Degraded image frame 3 (d) Degraded image frame 4
(e) Estimated PSF by ouralgorithm
(f) Restored image by our algo-rithm
(g) Restored image by RL-IBDalgorithm
(h) Restored image by Weiner-IBD algorithm
Figure 3 Multiframe original images and restoration comparison of three algorithms
(2) According to the method of Section 22 for the initialPSF estimation ℎ
0 normalized processing should be
taken at same time
(3) Set theMax iteration of extrinsic cycles
(4) SetMax count of target and PSF estimation
(5) Set the inner loop counter variable h count by PSFestimation variable f count of target inner cyclecount
Step 2 (the calculation of parameter values)
(1) According to the formulas of Section 23 the variance1205902
119899of mixed noise of each degradation image is
estimated
(2) According to the method and calculation formulasfrom Section 22 calculate the power spectral densityPSDℎ(119906) of PSF and estimate the wavefront phase
error 120579119905
6 Abstract and Applied Analysis
0
50
100
150
200
0
50
100
150
0
50
100
150
Y (pixel)X (pixel)
Gra
y
(a) Energy spectrum of observed image in Figure 3(a)
0
50
100
150
0
50
100
150
200
250
0
50
100
150
Y (pixel) X(pixel)
Gra
y
(b) Energy spectrum of our algorithm restored image in Figure 3(f)
0
50
100
150
200
250
0
50
100
150
0
50
100
150
Y (pixel)X (pixel)
Gra
y
(c) Energy spectrum of RL-IBD algorithm restored image in Figure 3(g)
0
100
100
150
00
50
50
100
150
200
250
2040
6080
Y (pixel)
X(pixel)
Gra
y
120
(d) Energy spectrum of Wiener-IBD algorithm restored image inFigure 3(h)
Figure 4 Energy spectrum of observed image and restoration comparison of three algorithms
(3) According to the scheme and formulas ofSection 241 estimate the regularization parameters120572 and 120574(nabla119909) of target edge
Step 3 (iterate calculation 119894 = 0 1 119872119886119909 119894119905119890119903119886119905119894119900119899)
(1) The inner loop count variable of PSF h count = 0(2) The iteration process of PSF estimation 119895 = 0 1
119872119886119909 119888119900119906119899119905
(a) Complete the PSF estimation ℎ(119894) with the con-jugate gradient method using optimized for-mula (20)
(b) Consider h count++ j++(c) Judge the value of the loop variable 119895 if 119895 lt119872119886119909 119888119900119906119899119905 continue otherwise turn to (3)
(3) The inner loop counter variable of target estimation119891 119888119900119906119899119905 = 0
(4) The iteration process of target estimation 119896 =
0 1 119872119886119909 119888119900119906119899119905
(a) The conjugate gradient method was used tooptimize formula (21) to complete target imageestimation 119891(119894)
(b) consider f count++ 119896 + +(c) judge loop variable 119896 if 119896 lt 119872119886119909 119888119900119906119899119905 con-
tinue otherwise turn to (5)
(5) Judge whether the extrinsic loop is over if it is 119894 gt119898119886119909 119894119905119890119903119886119905119894119900119899 then turn to Step 4
(6) 119894 = 119894 + 1 return to (1) and continue to loop
Step 4 (the algorithm is in the end) The cycle does not stopuntil the algorithm converges and the output target imageestimation is 119891 that is over
Abstract and Applied Analysis 7
(a) Degraded image frame 1 (b) Degraded image frame 2 (c) Degraded image frame 3 (d) Degraded image frame 4
(e) Restored image by Weiner-IBD algorithm
(f) Restored image by RL-IBDalgorithm
(g) Estimated PSF by ouralgorithm
(h) Restored image by our algo-rithm
0
0
2
4
6
8
100
100
200200
300
400
150
250
times104
50
0
Y (pixel)X
(pixel)
Gra
y
(i) Energy spectrum of observed image in Figure 5(a)
0
0
0
20
40
60
80
100
50
100
150
2
4
6
8
times104
Y (pixel)
X (pixel)
Gra
y
(j) Energy spectrum of our algorithm restored image in Figure 5(h)
01
02
03
04
05
06
07
08
Inte
nsity
Our algorithm
Wiener-IBDRL-IBD
(k) A line plot taken access the horizontal axis of image restored byWeiner-IBD RL-IBD and our algorithm
Figure 5 The multiframe observed stellar images and restoration comparison of three algorithms
8 Abstract and Applied Analysis
3 Results and Analysis of AO ImageRestoration Experiments
31 The Restoration Experiment on Simulate AO Images Thealgorithm proposed in this paper is used to do the simulationexperiment on degradation spot images the evaluation ofimage restoration quality is judged by using rootmean squareerror (RMSE) and improved ΔSNR Consider
RMSE = radic1
11987311198732sum
119894119895
(119891 (119894 119895) minus 119891 (119894 119895))2
ΔSNR = 100 log10
1003817100381710038171003817119892 minus 11989110038171003817100381710038172
10038171003817100381710038171003817119891 minus 119891
10038171003817100381710038171003817
2
(22)
The evaluation of estimation accuracy is defined as fol-lows
120576estimate =
radicsum119894119895isin119878ℎ
10038161003816100381610038161003816ℎ(119894 119895) minus ℎ(119894 119895)
10038161003816100381610038161003816
2
radicsum119894119895isin119878ℎ
1003816100381610038161003816ℎ(119894 119895)10038161003816100381610038162
(23)
In the formula 1198731 and 1198732 represent the total number ofpixels on 119909-axis and 119910-axis on the image respectively (119894 119895)represents a pixel point on image and 119891(119894 119895) represents thegray value of the point (119894 119895) on ideal image
In the simulation experiments of this paper the originalimage is infrared simulation imaging with spots for multipletargets with long distance 10 frames of simulation degradedimage series are generated by image degradation softwarefrom the key optical laboratory of Chinese Academy ofSciences and then add Gaussian white noise and make thesignal-to-noise ratio SNR of image for 20 dB Set parametersthe same as adaptive optical telescope of 12m on observatoryin YunnanThemain parameters of telescope imaging systemare the atmospheric coherence length 119903
0= 13 cm diameter
of telescope 119863 = 108m comprehensive focal length ofoptical system 119897 = 2242m the wavelength in center 120582 =700 nm and the pixel size of CCD is 74 120583m Experiments onIBD algorithm based on RL iterative [16] (RL-IBD) iterativeblind restoration algorithm based on theWiener filtering [17](Wiener-IBD) and the proposed algorithm are compared
Figure 2 is the compared results of the original imagesimulated multiframe degradation images and restorationimages based on three kinds of algorithm the image size is217 times 218 Figure 2(a) is the original target image and tenframes turbulence degraded image sequences are generatedby using image degradation simulation software as shownin Figures 2(b) 2(c) and 2(d) (to save space we show onlythree frames the rest is omitted) Figure 2(e) is the Wiener-IBD restored image RMSE = 1635 Figure 2(f) is the RL-IBD restored image the number of iterations is 800 timesRMSE = 1031 Figure 2(g) is restored image used methodin this paper the variance of wavefront phase error 120579
119905(sdot) is
614 times 10minus3 ℎ0= 826 times 10
minus3 and 1205902119899= 11 times 10
minus5 that theregularization parameters in the experiments are 120572 = 125the parameter of regularization function is 120585 = 134 extrinsiciterations for 100 times inner loop of the PSF for 4 image
Table 1 Comparison of ΔSNR and 120576estimate values among threealgorithms
Algorithmname ΔSNR (dB) 120576estimate Iteration Time-consuming
Wiener-IBD 1982 362 800 20645 sRL-IBD 3046 449 800 17236 sOur algorithm 4164 510 600 10923 s
cycles for 2 namely the total times of cycling is 600 RMSE =1005 Comparing with ΔSNR and precision from Figures2(e) 2(f) and 2(g) are shown in Table 1 Comparison resultsshow that the proposed algorithm in this paper is better thanWiener-IBD algorithm and RL-IBD algorithm and needs lesstime of iterations
32 Restoration Experiments about Binary AO Images With-out the reference of ideal image using the objective qualityevaluation with no reference image the Full Width HalfMaximum (abbreviated as FWHM) [18] is used as objectiveevaluation for image restoration in this paper FWHM refersto imaging peak pixels is two times of half peak includingthe FWHM on 119909 direction and 119910 direction for the betterevaluation in astronomical observation areas computationformula is shown as follows
FWHM = radicFWHM 1199092 + FWHM 1199102 (24)
The experiment for the binary image restoration withoutreference images is carried out by the algorithm proposed inthis paper The binary images for experiments were taken bythe 12m AO telescope from Chinese Academy of Sciencesin Yunnan observatory on December 3 2006 The size forimaging CCD is 320 times 240 pixels array the main parametersof imaging system are as follows the atmospheric coherentlength 119903
0= 13 cm telescope aperture 119863 = 103 all
imaging observation band is 07sim09 center wavelength 120582 =072 the CCD pixel size is 67 120583m and optical imagingfocal length 119897 = 20 Recovery experiments based on RLIBD algorithm Wiener-IBD algorithm and the proposedalgorithm in this paper are compared the frame selectiontechnology is taken in this paper 40 frames are picked from100 frames of degradation images as blind convolution imagesto be processed
Figure 3 is the comparison results ofmultiframedegradedimages and restoration images based on three kinds ofalgorithms Figures 3(a) 3(b) 3(c) and 3(d) are for theobservation of the binary image (to save space we show onlyfour frames the rest is omitted) Figure 3(e) is the estimationof the point spread function (PSF) based on the algorithmin this paper Figure 3(f) is the restored image based on themethod proposed in this paper the variance of wavefrontphase error 120579
119905(sdot) is 604 times 10minus3 ℎ
0= 903 times 10
minus3 and1205902
119899= 132 times 10
minus5 so that the regularization parameter inthis experiment is 120572 = 115 the parameter of regularizationfunction is 120585 = 134 extrinsic iterations for 50 times the inner
Abstract and Applied Analysis 9
loop of PSF for 4 times and image loop for 2 times namelyThe total times of iteration are 300 FWHM = 617 pixels andthe restoration results are very close to the diffraction limit of12m AO telescope Figure 3(g) is the restoration result basedonRL-IBD algorithm FWHM=562 pixels Figure 3(h) is therestoration result based onWiener-IBD algorithm FWHM=476 pixels Figure 4 is the sketch of energy for the observationimage and three kinds of restored images based on threerestoration algorithms With the comparison it shows thatthe proposed algorithm in this paper is better than Wiener-IBD algorithm and RL-IBD algorithm and needs less time ofiterations
33 Restoration Experiments about Stellar AO Images Alsothe restoration experiment of stellar images was carriedout using the algorithm in this paper the stellar imagesfor experiment were taken by 12m AO telescope from theChinese Academy of Sciences in Yunnan observatory onDecember 1 2006 collection The parameters of the opticalimaging system are the same as that of binary images andthe method of parameters selection for experiments in thispaper is nearly the same Figure 5 is the comparison ofmultiframe stellar observation images and restoration resultsbased on the three algorithms Figures 5(a) 5(b) 5(c) and5(d) are the observation of stellar images (to save space weshow only 4) Figure 5(e) is for the Wiener-IBD restoredimage Figure 5(f) is the RL-IBD restored image Figure 5(g)is the estimation of the point spread function (PSF) basedon algorithm in this paper the iterative initial value ℎ
0=
956 times 10minus3 PSF iterative for six times Figure 5(h) is the
restored image based on algorithm in this paper Figure 5(i)is the energy of the observed image for stellar Figure 5(j) isthe energy of restoration image based on the algorithm in thispaper Figure 5(k) is the graph of comparison results for threerestoration algorithms
4 Conclusions
In this paper according to the characteristic of adaptiveoptics image we establish a degradation image model formultiframe exposure image series and the mathematicalmodel of the point spread function (PSF) for adaptive opticalsystem after closed loop correction is derived the PSF modelbased on phase error and changing over time is solved andwe use the power spectral density of image and constrainsupport domain method for AO image denoising processingFinally this paper applied the actual AO imaging systemparameters as prior knowledge combined with regularizationtechnique to improve the EM blind deconvolution algo-rithm Experimental results show that the established blinddeconvolution algorithm based on improved EM methodcan effectively eliminate the atmospheric turbulence andeliminate the noise pollution brought by the AO closed-loopcorrection which improves the resolution of the adaptiveoptical image The image process of AO observed imagesshows that the restoration quality is better than the samekind of IBD algorithm It also shows that the restoration ofbinary images using the algorithmof this paper has carried on
the identification of PSF and completed the AO image afterprocessing the results of actual AO image restoration haveimportant application value
Conflict of Interests
The authors declare that there is no conflict of interestsregarding the publication of this paper
Acknowledgments
This project is supported by the Scientific and TechnologicalResearch Project of Jinlin province Department of Edu-cation China (no 2013145 and no 201363) and the basisstudy project is supported by Jinlin Provincial Science ampTechnology Department China (no 201105055)
References
[1] J Wenhan ldquoAdaptive optical technologyrdquo Chinese Journal ofNature vol 28 pp 7ndash13 2005
[2] G R Ayers and J C Dainty ldquoIterative blind deconvolutionmethod and its applicationsrdquo Optics Letters vol 13 no 7 pp547ndash549 1988
[3] L Dayu H Lifa and M Quanquan ldquoA high resolution liquidcrystal adaptive optics systemrdquo Acta Photonica Sinica vol 37pp 506ndash508 2008
[4] C Rao F Shen andW Jiang ldquoAnalysis of closed-loopwavefrontresidual error of adaptive optical system using the method ofpower spectrumrdquo Acta Optica Sinica vol 20 no 1 pp 68ndash732000
[5] B Chen The theory and algorithms of adaptive optics imagerestoration [PhD thesis] Information Engineering UniversityZhengzhou China 2008
[6] T Yu R Changhui and W Kai ldquoPost-processing of adaptiveoptics images based on frame selection and multi-frame blinddeconvolutionrdquo inAdaptive Optics Systems vol 7015 of Proceed-ings of SPIE Marseille France June 2008
[7] T J Schulz ldquoNonlinear models and methods for space-objectimaging through atmospheric turbulencerdquo in Proceedings ofthe IEEE International Conference on Image Processing vol 9Lausanne Switzerland 1996
[8] L Yaning G Lei and L Huihui ldquoTotal variation basedband-limited sheralets transform for image denoisingrdquo ActaPhotonica Sinica vol 242 pp 1430ndash1435 2013
[9] J Zhao and J Yang ldquoRemote sensing image denoising algorithmbased on fusion theory using cycle spinning contourlet trans-form and total variation minimizationrdquo Acta Photonica Sinicavol 39 no 9 pp 1658ndash1665 2010
[10] C L Matson and M C Roggemann ldquoNoise reduction inadaptive-optics imagery with the use of support constraintsrdquoApplied Optics vol 34 no 5 pp 767ndash780 1995
[11] DWTyler andC LMatson ldquoReduction of nonstationary noisein telescope imagery using a support constraintrdquoOptics Expressvol 1 no 11 pp 347ndash354 1997
[12] L Zhang J YangW Su et al ldquoResearch on blind deconvolutionalgorithm of multiframe turbulence-degraded imagesrdquo Journalof Information and Computational Science vol 10 no 12 pp3625ndash3634 2013
10 Abstract and Applied Analysis
[13] A P Dempster N M Laird and D B Rubin ldquoMaximumlikelihood from incomplete data via the EM algorithmrdquo Journalof the Royal Statistical Society B vol 39 no 1 pp 1ndash38 1977
[14] L M Mugnier J Conan T Fusco and V Michau ldquoJoint maxi-mum A posteriori estimation of object and PSF for turbulence-degraded imagesrdquo in Bayesian Inference for Inverse Problemsvol 3459 of Proceedings of SPIE pp 50ndash61 San Diego CalifUSA July 1998
[15] H Hanyu Research on Image Restoration Algorithms in ImagingDetection System HuazhongUniversity of Science and Technol-ogy Wuhan China 2004
[16] F Tsumuraya ldquoIterative blind deconvolution method usingLucyrsquos algorithmrdquo Astronomy amp Astrophysics vol 282 no 2 pp699ndash708 1994
[17] S Zhang J Li Y Yang and Z Zhang ldquoBlur identification ofturbulence-degraded IR imagesrdquoOptics and Precision Engineer-ing vol 21 no 2 pp 514ndash521 2013
[18] B Chen C Cheng S GuoG Pu andZGeng ldquoUnsymmetricalmulti-limit iterative blind deconvolution algorithm for adaptiveoptics image restorationrdquoHigh Power Laser and Particle Beamsvol 23 no 2 pp 313ndash318 2011
Submit your manuscripts athttpwwwhindawicom
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
MathematicsJournal of
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Mathematical Problems in Engineering
Hindawi Publishing Corporationhttpwwwhindawicom
Differential EquationsInternational Journal of
Volume 2014
Applied MathematicsJournal of
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Probability and StatisticsHindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Journal of
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Mathematical PhysicsAdvances in
Complex AnalysisJournal of
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
OptimizationJournal of
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
CombinatoricsHindawi Publishing Corporationhttpwwwhindawicom Volume 2014
International Journal of
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Operations ResearchAdvances in
Journal of
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Function Spaces
Abstract and Applied AnalysisHindawi Publishing Corporationhttpwwwhindawicom Volume 2014
International Journal of Mathematics and Mathematical Sciences
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
The Scientific World JournalHindawi Publishing Corporation httpwwwhindawicom Volume 2014
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Algebra
Discrete Dynamics in Nature and Society
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Decision SciencesAdvances in
Discrete MathematicsJournal of
Hindawi Publishing Corporationhttpwwwhindawicom
Volume 2014 Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Stochastic AnalysisInternational Journal of
2 Abstract and Applied Analysis
Atmospheric turbulence
DM
WFS
Controller
Separating slide
Corrected wavefront
Scientific camera
Flat wavefrontDeformed wavefront
Telescopersquos pupil
Figure 1 Works of adaptive optical system for imaging compensa-tion
AO image and researching the denoise processingmethod ofAO image Finally it is given a deconvolution method usingmultiframe AO images by combining AO imaging systemparameters with the regularization technique on the basis ofexpectation maximization algorithm Section 3 verifies thevalidity of the algorithm in this paper by using simulate AOimages actual observation binary AO images and stellar AOimages Section 4 concludes the paper
2 Restoration Algorithm ofAdaptive Optics Images
Adaptive optics system can measure control and correct theerror of optical wavefront in real-time Figure 1 shows theprinciple diagram of an adaptive optical system for imagingcompensationThe system includes wavefront sensor (abbre-viated as WFS) the deformable mirror (abbreviated as DM)high speed processor for wavefront and other parts Thesystem can measure the distortion of optical wavefront inreal-time which is caused by the influence of atmosphericturbulence and convert it to corresponding controlling signaladding to the wavefront correction element which cancompensate distortion of wavefront in real-time and theoptical telescope would be able to get close to target imagingfor diffraction AO images are corrected by AO system butits compensation or correction is not sufficient in order toget clearer target images AO images restoration algorithm isresearched
21 The Imaging Process of Adaptive Optics System Con-sidering the normal optical imaging model first of all theproper adaptive optics imagingmodel is set up which lays the
foundation of restoring high-quality adaptive optical imagesThe acquisition process of adaptive optics image model canconsist of a degradation function and an additive noise[5] the imaging process of AO in spatial frequency can beexpressed as follows
119892 (119909) = (119891 lowast ℎ) (119909) + 119899 (119909)
= intΩ
119891 (119910) ℎ (119909 minus 119910) 119889119910 + 119899 (119909) 119909 119910 isin Ω
(1)
where 119891(119909) is the original image for energy concentrationℎ(119909) is the point spread function (abbreviated as PSF) orimpulse response for image system 119892(119909) is output observa-tion images for the AO system 119899(119909) is the additive noise andΩ is the image area
In application we use multiple frames of short exposureimages 119892
119898119872
119898=1for one target to restore the target image 119891
119898 is the number of frames The model of multiframe shortexposure image with degradation can be written as
119892119898 (119909) = 119891 (119909) lowast ℎ119898 (119909) + 119899 (119909) 1 lt 119898 le 119872 (2)
where 119892119898(119909) and ℎ
119898(119909) represent the 119898th frame of short
exposure AO image with degradation for the target andits corresponding PSF respectively This paper adopts theselection techniques of frames referred to in [6] as the imagepreprocessing which picks up the images with better qualityfrom image sequence of short exposure images in order toimprove the solution of blind convolution and convergenceof iteration
22 Model of Point Spread Function on AO Images In adap-tive optics system the residual error of wavefront correctionmainly consists of incomplete correction error by turbulenceand noise of closed-loop system
When AO system works in closed-loop the transfererror function is the transfer function of wavefront phaseperturbation caused by atmospheric turbulence and thenoise of system introduced bywavefront sensor is closed-looptransfer function [4] Theoretically the mathematical modelof the PSF after closed-loop correction of adaptive opticssystem is shown as follows
ℎ (119910) =
1003816100381610038161003816100381610038161003816int 119875(119906)119890
minus119895sdot(2120587120582119897)119910sdot119906119889119906
1003816100381610038161003816100381610038161003816
2
(3)
In the formula
119875 (119906) = 1 within the lens aperture0 others
(4)
is the pupil function of telescope 119906 is the coordinates for thepupil plane 120582 is the central wavelength of imaging beam 119897 isthe focal length of the optical system 119910 is the coordinates oftwo-dimensional focal plane
Due to the error of focal length and optical deviationin optical system the pupil function should be rectified asfollows
119901 (119906) = 119875 (119906) 119890119895120579(119906) (5)
Abstract and Applied Analysis 3
where 120579(sdot) is the wavefront phase error caused by the error offocal length or optical deviation
So the mathematical model of PSF can be written as
ℎ (119910 120579) =
1003816100381610038161003816100381610038161003816int 119901(119906)119890
minus119895sdot(2120587120582119897)sdot119910sdot119906119889119906
1003816100381610038161003816100381610038161003816
2
=
1003816100381610038161003816100381610038161003816int 119875(119906)119890
119895120579(119906)119890minus119895sdot(2120587120582119897)sdot119910sdot119906
119889119906
1003816100381610038161003816100381610038161003816
2
(6)
The wavefront phase error is caused by the focal lengtherror or optical deviation due to the influence of atmosphericturbulence the PSFmodel changes with the focal length erroror optical deviation over time as follows
ℎ (119910 120579119905) =
1003816100381610038161003816100381610038161003816int 119875(119906)119890
119895120579119905(119906)119890minus119895sdot(2120587120582119897)sdot119910sdot119906
119889119906
1003816100381610038161003816100381610038161003816
2
(7)
In the formula 120579119905(sdot) is the phase error caused by atmospheric
turbulence changing over timeIn terms of AO system the exposure time of observing
images is shorter than turbulent fluctuation If the intensity ofthe target is kept unchanged during exposure time themodelfor series images is
119892 (119910 120579119905119898 119891) = int ℎ (119910 minus 119909 120579
119905119898) 119891 (119909) 119889119909 (8)
In the formula 120579119905119898
is the phase error of turbulence for the119898th frame short exposure AO image
23 The Denoising Processing of AO Images In order toreduce the noise of AO images Chinese researchers putforward themethod to suppress noise based on total variation[8 9] This paper proposes a method to suppress noise that isbased on the power spectral density (abbreviated as PSD) ofimage and the supporting domain field of constraints imagesReference [10] analyzed the method to minimize the noisein symmetrical support domain field of AO images Theanalysis for this method is based on the variance ratio of realimages In this paper the assessment about denoise resultusing noise changed function The denoising process of AOimage is actual a kind of image noise corrected process inFourier spectral domain If theAO image has the symmetricalsupport field its noise mainly consist by CCD read-in noiseand AO noise and then noise power spectral density can beresearched Therefore suppress noise for observation images119892(119909) and 119892
119888(119909) is constrained by field of support domain
119892119888 (119909) = 119892 (119909)119908 (119909) (9)
In the formula 119908(119909) is weighting function in support fieldthe image spectrum 119866
119888(119906) is
119866119888 (119906) = int119866 (V)119882 (119906 minus V) 119889119906 (10)
where 119906 is the spatial frequency of image119866(119906) = 119878(119906)lowast119873(119906)119878(119906) is the signal frequency spectrum function 119873(119906) is thenoise spectrum function and bandwidth function 119882(119906) =119866(119906) lowast 120575 In the limited support domain field119882(119906) is verynarrow which nearly equals the 120575 times of the width of 119866(119906)
According to the definition of J M Conan the PSD ofan image is the Fourier transformation of its covariancedefining the power spectrum of noise and power spectrumof distortion image [11]
PSD(119906)119899119899 = ⟨10038161003816100381610038161003816Re 119866 (119906) minus Re 119866 (119906)10038161003816100381610038161003816
2
⟩
PSD(119906)119891119891 = ⟨10038161003816100381610038161003816Im 119866 (119906) minus Im 119866 (119906)10038161003816100381610038161003816
2
⟩
(11)
In the formula 119866(119906) is the Fourier transformation of image119892(119909) 119866(119906) is the mean value of sample 119866(119906)
Regard the circle as target supporting domain field itsdiameter is from 8 pixels to 185 pixels The definition of thechanging image noise function 119865
119899(119866) is
119865119899 (119866) =
int (PSD(119906)119888119899119899minus PSD(119906)119888
119899119899) 119889119906
intPSD(119906)119888119899119899119889119906
(12)
In the formula PSD(119906)119888119899119899is the power spectral density of the
image with constraints of support domain field and PSD(119906)119888119899119899
is the power spectral density of the image without constraintsof support domain field
24 Improved Expectation Maximization Algorithm Refer-ence [12] deduced the point spread function of multiframedegraded images with turbulence and the iteration formulaof target image based on maximum likelihood criterion butusing thismethod leads to the larger solution space Howeverwe put forward the technique of regularization to modify themaximum likelihood estimation algorithm introduce con-straints in accordance with AO image in physical reality andadopt the improved EM algorithm onmultiframe iteration ofAO images to acquire ideal 119891 and ℎ
119896
241 EM Algorithm by Combining Regularization TechniqueEM algorithm is put forward by Dempster et al [13] forparameters of maximum likelihood estimate of the iterativealgorithm which is iteratively going on two basic stepscalculation steps E (Expectation) and M (Maximization) Inthis paper a new regularizationmethod combined with priorknowledge is applied to the EMalgorithm realizing amethodof multiframe AO image restoration
Based on prior knowledge of target image and PSF jointprobability distribution function defining the maximizationfunction will get target image estimation 119891 and PSF estima-tion ℎ that is
[119891 ℎ] = max119891ℎ
(119901 (119892 | 119891 ℎ 119886)) = max119891ℎ
(exp(minus119869 (119891 ℎ))
119911)
(13)
In the formula 119911 is the normalization factor The costfunction is
[119891 ℎ] = max119891ℎ
(119901 (119892 | 119891 ℎ 119886)) = max119891ℎ
(exp(minus119869 (119891 ℎ))
119911)
(14)
4 Abstract and Applied Analysis
So the problem is converted to minimize cost function119869(119891 ℎ) that is [119891 ℎ] = min
119891ℎ(119869(119891 ℎ)) The cost function
of AO images and the optimization model of its parameterestimation are shown as follows
119869 (119891 ℎ) = 119869119899(119892 | 119891 ℎ 119886) + 119869
ℎ (ℎ | 119886) + 119869119891 (119891 | 119886)
st 119869119899(119892 | 119891 ℎ 119886) = sum
119894119895
1
1205902119899
[119892 (119894 119895) minus (ℎ lowast 119891) (119894 119895)]2
119869ℎ (ℎ | 119886) =
120579119905
2sum
119906
10038161003816100381610038161003816(119906) minus 119867(119906)
10038161003816100381610038161003816
2
PSDℎ (119906)
119869119891(119891 | 119886) = 120572sum
119894119895
Φ(119903 (119891 120574 (119894 119895)))
(15)
Symbol explanation the first item represents the noise costfunction the second item represents the constraints costfunction of PSF and the third item is the cost functionswith target edge constraint 119886 is prior information 1205902
119899is the
variance of mixed noise 120579119905is the wavefront phase error
the power spectral density of the point spread function isPSD(119906)
ℎ= ⟨|(119906) minus 119867(119906)|
2
⟩ (119906) is the Fourier transfor-mation of ℎ(119909) 119867(119906) is the Fourier transform of averagePSF the estimation of PSD
ℎ(119906) can be estimated through the
AO system closed-loop control data and 120572 and 120574(nabla119909) areregularization adjustment parameters using the target edgeconstraint theory of Mugnier isotropic [14] as
Φ(119901) = 119901 minus ln (1 + 119901)
119901 (119891 120574 (nabla119909)) =
10038171003817100381710038171003817nabla119891 (119894 119895)
10038171003817100381710038171003817
120574 (nabla119909)
10038171003817100381710038171003817nabla119891 (119894 119895)
10038171003817100381710038171003817= ((nabla119891
119909(119894 119895))
2+ (nabla119891
119910(119894 119895))
2
)
12
(16)
where 120572 is a constant coefficient 120574(sdot) is a regularized functionthat has variable regularization coefficient associated with thegradient of each point 120574(nabla119909) = expminus|nabla119909|
221205852
is based on [15]0 lt 120574(nabla119909) le 1 parameter 120585 is the controlled smoothingcoefficient for selection and nabla119909 is the gradient amplitudein four directions for the point 119909(119894 119895) (where 119894 119895 are pixelcoordinates and 119909 is gray value of pixel)
This paper will apply all the prior information on theestimation of target and PSF in order to get the optimalestimation of target and the point spread function (PSF)Combine formulas (2) (8) and (15) to establish the costfunction for joint deconvolutionwithmultiframeAO images
119869multi (119891 ℎ) = 119869119899 (119892119898 | 119891 ℎ119898 119886) + 119869ℎ (ℎ119898 | 119886) + 119869119891 (119891 | 119886)
(17)
where119898 is the number of frames
242 Implement of Image Restoration Algorithm This paperis combining with the basic idea of the EM algorithm [13]through limited times of cycle calculations find the mostsuitable PSF estimation on the purpose of restoring target
image drastically modify the cost function (17) and make itcorrespond to full data [15] if 119892
119896(119910 | 119909) is full data set of 119896
times observation results 119892119896(119910) of observed image 119892
119896(119910) and
the expectation of cost function are as follows
119864 (119892119896(119910)) = 119864(sum
119909
119892119896(119910 | 119909))
= sum
119909
ℎ119896(119910 minus 119909) 119891 (119909)
119864 (119869multi (119891 ℎ)) = 119864(119869119899 (sum119909
119892119896| 119891 ℎ119898 119886))
+ 119864 (119869ℎ(ℎ119898| 119886)) + 119864 (119869
119891(119891 | 119886))
(18)
In the iteration process it is supposed that the 119899 minus1 iteration result is known 119864119899(119869multi(119891 ℎ)) represents thecurrent expectation of the cost function In order tominimizethe expectation of the cost function formula (18) can be dif-ferentiated by ℎ119899
119896(119909) and 119891119899(119909) then make the differentiation
equal to zero that is
120597119864119899(119869multi (119891 ℎ))
120597ℎ119899
119896(119909)
= 0120597119864119899(119869multi (119891 ℎ))
120597119891119899 (119909)= 0 (19)
The derivation process in detail is omitted so the finalsolutions are
119891119899+1
119894(119909) = 119891
119899
119894(119909)
+120579119905
120572 lowast Φ (119903 (119891119899
119894(119909) 120574 (119909)))
times sum
119910isin119884
sum
119898
ℎ119899
119894(119910 minus 119909)
times (119892119898minus sum119911isin119883ℎ119899
119894(119910 minus 119911) 119891
119899
119894(119911)
1205902119894
)
(20)
ℎ119899+1
119894(119909) = ℎ
119899
119894(119909)
+120579119905
2sum
119906
10038161003816100381610038161003816119894(119906) minus 119867
119894(119906)10038161003816100381610038161003816
2
119875119878119863ℎ 119894 (119906)
times sum
119910isin119884
119891119899+1
119894(119910 minus 119909)
times (119892119898minus sum119911isin119883ℎ119899
119894(119910 minus 119911) 119891
119899
119894(119911)
1205902119894
)
(21)
In this paper the specific steps of image restorationalgorithm are as follows
Step 1 (initialization operation)
(1) According to the method of Section 23 constrainthe support domain field of observed image and dealwith the noise and so forth regard the result of onetime Wiener filtering as the initial value of targetestimation119891
0
Abstract and Applied Analysis 5
(a) Original image (b) Degraded image frame 1 (c) Degraded image frame 2 (d) Degraded image frame 3
(e) Restored image by Wiener-IBD
(f) Restored image by RL-IBD (g) Restored image by our algo-rithm
Figure 2 Degraded images in experiment and restoration comparison of three algorithms
(a) Degraded image frame 1 (b) Degraded image frame 2 (c) Degraded image frame 3 (d) Degraded image frame 4
(e) Estimated PSF by ouralgorithm
(f) Restored image by our algo-rithm
(g) Restored image by RL-IBDalgorithm
(h) Restored image by Weiner-IBD algorithm
Figure 3 Multiframe original images and restoration comparison of three algorithms
(2) According to the method of Section 22 for the initialPSF estimation ℎ
0 normalized processing should be
taken at same time
(3) Set theMax iteration of extrinsic cycles
(4) SetMax count of target and PSF estimation
(5) Set the inner loop counter variable h count by PSFestimation variable f count of target inner cyclecount
Step 2 (the calculation of parameter values)
(1) According to the formulas of Section 23 the variance1205902
119899of mixed noise of each degradation image is
estimated
(2) According to the method and calculation formulasfrom Section 22 calculate the power spectral densityPSDℎ(119906) of PSF and estimate the wavefront phase
error 120579119905
6 Abstract and Applied Analysis
0
50
100
150
200
0
50
100
150
0
50
100
150
Y (pixel)X (pixel)
Gra
y
(a) Energy spectrum of observed image in Figure 3(a)
0
50
100
150
0
50
100
150
200
250
0
50
100
150
Y (pixel) X(pixel)
Gra
y
(b) Energy spectrum of our algorithm restored image in Figure 3(f)
0
50
100
150
200
250
0
50
100
150
0
50
100
150
Y (pixel)X (pixel)
Gra
y
(c) Energy spectrum of RL-IBD algorithm restored image in Figure 3(g)
0
100
100
150
00
50
50
100
150
200
250
2040
6080
Y (pixel)
X(pixel)
Gra
y
120
(d) Energy spectrum of Wiener-IBD algorithm restored image inFigure 3(h)
Figure 4 Energy spectrum of observed image and restoration comparison of three algorithms
(3) According to the scheme and formulas ofSection 241 estimate the regularization parameters120572 and 120574(nabla119909) of target edge
Step 3 (iterate calculation 119894 = 0 1 119872119886119909 119894119905119890119903119886119905119894119900119899)
(1) The inner loop count variable of PSF h count = 0(2) The iteration process of PSF estimation 119895 = 0 1
119872119886119909 119888119900119906119899119905
(a) Complete the PSF estimation ℎ(119894) with the con-jugate gradient method using optimized for-mula (20)
(b) Consider h count++ j++(c) Judge the value of the loop variable 119895 if 119895 lt119872119886119909 119888119900119906119899119905 continue otherwise turn to (3)
(3) The inner loop counter variable of target estimation119891 119888119900119906119899119905 = 0
(4) The iteration process of target estimation 119896 =
0 1 119872119886119909 119888119900119906119899119905
(a) The conjugate gradient method was used tooptimize formula (21) to complete target imageestimation 119891(119894)
(b) consider f count++ 119896 + +(c) judge loop variable 119896 if 119896 lt 119872119886119909 119888119900119906119899119905 con-
tinue otherwise turn to (5)
(5) Judge whether the extrinsic loop is over if it is 119894 gt119898119886119909 119894119905119890119903119886119905119894119900119899 then turn to Step 4
(6) 119894 = 119894 + 1 return to (1) and continue to loop
Step 4 (the algorithm is in the end) The cycle does not stopuntil the algorithm converges and the output target imageestimation is 119891 that is over
Abstract and Applied Analysis 7
(a) Degraded image frame 1 (b) Degraded image frame 2 (c) Degraded image frame 3 (d) Degraded image frame 4
(e) Restored image by Weiner-IBD algorithm
(f) Restored image by RL-IBDalgorithm
(g) Estimated PSF by ouralgorithm
(h) Restored image by our algo-rithm
0
0
2
4
6
8
100
100
200200
300
400
150
250
times104
50
0
Y (pixel)X
(pixel)
Gra
y
(i) Energy spectrum of observed image in Figure 5(a)
0
0
0
20
40
60
80
100
50
100
150
2
4
6
8
times104
Y (pixel)
X (pixel)
Gra
y
(j) Energy spectrum of our algorithm restored image in Figure 5(h)
01
02
03
04
05
06
07
08
Inte
nsity
Our algorithm
Wiener-IBDRL-IBD
(k) A line plot taken access the horizontal axis of image restored byWeiner-IBD RL-IBD and our algorithm
Figure 5 The multiframe observed stellar images and restoration comparison of three algorithms
8 Abstract and Applied Analysis
3 Results and Analysis of AO ImageRestoration Experiments
31 The Restoration Experiment on Simulate AO Images Thealgorithm proposed in this paper is used to do the simulationexperiment on degradation spot images the evaluation ofimage restoration quality is judged by using rootmean squareerror (RMSE) and improved ΔSNR Consider
RMSE = radic1
11987311198732sum
119894119895
(119891 (119894 119895) minus 119891 (119894 119895))2
ΔSNR = 100 log10
1003817100381710038171003817119892 minus 11989110038171003817100381710038172
10038171003817100381710038171003817119891 minus 119891
10038171003817100381710038171003817
2
(22)
The evaluation of estimation accuracy is defined as fol-lows
120576estimate =
radicsum119894119895isin119878ℎ
10038161003816100381610038161003816ℎ(119894 119895) minus ℎ(119894 119895)
10038161003816100381610038161003816
2
radicsum119894119895isin119878ℎ
1003816100381610038161003816ℎ(119894 119895)10038161003816100381610038162
(23)
In the formula 1198731 and 1198732 represent the total number ofpixels on 119909-axis and 119910-axis on the image respectively (119894 119895)represents a pixel point on image and 119891(119894 119895) represents thegray value of the point (119894 119895) on ideal image
In the simulation experiments of this paper the originalimage is infrared simulation imaging with spots for multipletargets with long distance 10 frames of simulation degradedimage series are generated by image degradation softwarefrom the key optical laboratory of Chinese Academy ofSciences and then add Gaussian white noise and make thesignal-to-noise ratio SNR of image for 20 dB Set parametersthe same as adaptive optical telescope of 12m on observatoryin YunnanThemain parameters of telescope imaging systemare the atmospheric coherence length 119903
0= 13 cm diameter
of telescope 119863 = 108m comprehensive focal length ofoptical system 119897 = 2242m the wavelength in center 120582 =700 nm and the pixel size of CCD is 74 120583m Experiments onIBD algorithm based on RL iterative [16] (RL-IBD) iterativeblind restoration algorithm based on theWiener filtering [17](Wiener-IBD) and the proposed algorithm are compared
Figure 2 is the compared results of the original imagesimulated multiframe degradation images and restorationimages based on three kinds of algorithm the image size is217 times 218 Figure 2(a) is the original target image and tenframes turbulence degraded image sequences are generatedby using image degradation simulation software as shownin Figures 2(b) 2(c) and 2(d) (to save space we show onlythree frames the rest is omitted) Figure 2(e) is the Wiener-IBD restored image RMSE = 1635 Figure 2(f) is the RL-IBD restored image the number of iterations is 800 timesRMSE = 1031 Figure 2(g) is restored image used methodin this paper the variance of wavefront phase error 120579
119905(sdot) is
614 times 10minus3 ℎ0= 826 times 10
minus3 and 1205902119899= 11 times 10
minus5 that theregularization parameters in the experiments are 120572 = 125the parameter of regularization function is 120585 = 134 extrinsiciterations for 100 times inner loop of the PSF for 4 image
Table 1 Comparison of ΔSNR and 120576estimate values among threealgorithms
Algorithmname ΔSNR (dB) 120576estimate Iteration Time-consuming
Wiener-IBD 1982 362 800 20645 sRL-IBD 3046 449 800 17236 sOur algorithm 4164 510 600 10923 s
cycles for 2 namely the total times of cycling is 600 RMSE =1005 Comparing with ΔSNR and precision from Figures2(e) 2(f) and 2(g) are shown in Table 1 Comparison resultsshow that the proposed algorithm in this paper is better thanWiener-IBD algorithm and RL-IBD algorithm and needs lesstime of iterations
32 Restoration Experiments about Binary AO Images With-out the reference of ideal image using the objective qualityevaluation with no reference image the Full Width HalfMaximum (abbreviated as FWHM) [18] is used as objectiveevaluation for image restoration in this paper FWHM refersto imaging peak pixels is two times of half peak includingthe FWHM on 119909 direction and 119910 direction for the betterevaluation in astronomical observation areas computationformula is shown as follows
FWHM = radicFWHM 1199092 + FWHM 1199102 (24)
The experiment for the binary image restoration withoutreference images is carried out by the algorithm proposed inthis paper The binary images for experiments were taken bythe 12m AO telescope from Chinese Academy of Sciencesin Yunnan observatory on December 3 2006 The size forimaging CCD is 320 times 240 pixels array the main parametersof imaging system are as follows the atmospheric coherentlength 119903
0= 13 cm telescope aperture 119863 = 103 all
imaging observation band is 07sim09 center wavelength 120582 =072 the CCD pixel size is 67 120583m and optical imagingfocal length 119897 = 20 Recovery experiments based on RLIBD algorithm Wiener-IBD algorithm and the proposedalgorithm in this paper are compared the frame selectiontechnology is taken in this paper 40 frames are picked from100 frames of degradation images as blind convolution imagesto be processed
Figure 3 is the comparison results ofmultiframedegradedimages and restoration images based on three kinds ofalgorithms Figures 3(a) 3(b) 3(c) and 3(d) are for theobservation of the binary image (to save space we show onlyfour frames the rest is omitted) Figure 3(e) is the estimationof the point spread function (PSF) based on the algorithmin this paper Figure 3(f) is the restored image based on themethod proposed in this paper the variance of wavefrontphase error 120579
119905(sdot) is 604 times 10minus3 ℎ
0= 903 times 10
minus3 and1205902
119899= 132 times 10
minus5 so that the regularization parameter inthis experiment is 120572 = 115 the parameter of regularizationfunction is 120585 = 134 extrinsic iterations for 50 times the inner
Abstract and Applied Analysis 9
loop of PSF for 4 times and image loop for 2 times namelyThe total times of iteration are 300 FWHM = 617 pixels andthe restoration results are very close to the diffraction limit of12m AO telescope Figure 3(g) is the restoration result basedonRL-IBD algorithm FWHM=562 pixels Figure 3(h) is therestoration result based onWiener-IBD algorithm FWHM=476 pixels Figure 4 is the sketch of energy for the observationimage and three kinds of restored images based on threerestoration algorithms With the comparison it shows thatthe proposed algorithm in this paper is better than Wiener-IBD algorithm and RL-IBD algorithm and needs less time ofiterations
33 Restoration Experiments about Stellar AO Images Alsothe restoration experiment of stellar images was carriedout using the algorithm in this paper the stellar imagesfor experiment were taken by 12m AO telescope from theChinese Academy of Sciences in Yunnan observatory onDecember 1 2006 collection The parameters of the opticalimaging system are the same as that of binary images andthe method of parameters selection for experiments in thispaper is nearly the same Figure 5 is the comparison ofmultiframe stellar observation images and restoration resultsbased on the three algorithms Figures 5(a) 5(b) 5(c) and5(d) are the observation of stellar images (to save space weshow only 4) Figure 5(e) is for the Wiener-IBD restoredimage Figure 5(f) is the RL-IBD restored image Figure 5(g)is the estimation of the point spread function (PSF) basedon algorithm in this paper the iterative initial value ℎ
0=
956 times 10minus3 PSF iterative for six times Figure 5(h) is the
restored image based on algorithm in this paper Figure 5(i)is the energy of the observed image for stellar Figure 5(j) isthe energy of restoration image based on the algorithm in thispaper Figure 5(k) is the graph of comparison results for threerestoration algorithms
4 Conclusions
In this paper according to the characteristic of adaptiveoptics image we establish a degradation image model formultiframe exposure image series and the mathematicalmodel of the point spread function (PSF) for adaptive opticalsystem after closed loop correction is derived the PSF modelbased on phase error and changing over time is solved andwe use the power spectral density of image and constrainsupport domain method for AO image denoising processingFinally this paper applied the actual AO imaging systemparameters as prior knowledge combined with regularizationtechnique to improve the EM blind deconvolution algo-rithm Experimental results show that the established blinddeconvolution algorithm based on improved EM methodcan effectively eliminate the atmospheric turbulence andeliminate the noise pollution brought by the AO closed-loopcorrection which improves the resolution of the adaptiveoptical image The image process of AO observed imagesshows that the restoration quality is better than the samekind of IBD algorithm It also shows that the restoration ofbinary images using the algorithmof this paper has carried on
the identification of PSF and completed the AO image afterprocessing the results of actual AO image restoration haveimportant application value
Conflict of Interests
The authors declare that there is no conflict of interestsregarding the publication of this paper
Acknowledgments
This project is supported by the Scientific and TechnologicalResearch Project of Jinlin province Department of Edu-cation China (no 2013145 and no 201363) and the basisstudy project is supported by Jinlin Provincial Science ampTechnology Department China (no 201105055)
References
[1] J Wenhan ldquoAdaptive optical technologyrdquo Chinese Journal ofNature vol 28 pp 7ndash13 2005
[2] G R Ayers and J C Dainty ldquoIterative blind deconvolutionmethod and its applicationsrdquo Optics Letters vol 13 no 7 pp547ndash549 1988
[3] L Dayu H Lifa and M Quanquan ldquoA high resolution liquidcrystal adaptive optics systemrdquo Acta Photonica Sinica vol 37pp 506ndash508 2008
[4] C Rao F Shen andW Jiang ldquoAnalysis of closed-loopwavefrontresidual error of adaptive optical system using the method ofpower spectrumrdquo Acta Optica Sinica vol 20 no 1 pp 68ndash732000
[5] B Chen The theory and algorithms of adaptive optics imagerestoration [PhD thesis] Information Engineering UniversityZhengzhou China 2008
[6] T Yu R Changhui and W Kai ldquoPost-processing of adaptiveoptics images based on frame selection and multi-frame blinddeconvolutionrdquo inAdaptive Optics Systems vol 7015 of Proceed-ings of SPIE Marseille France June 2008
[7] T J Schulz ldquoNonlinear models and methods for space-objectimaging through atmospheric turbulencerdquo in Proceedings ofthe IEEE International Conference on Image Processing vol 9Lausanne Switzerland 1996
[8] L Yaning G Lei and L Huihui ldquoTotal variation basedband-limited sheralets transform for image denoisingrdquo ActaPhotonica Sinica vol 242 pp 1430ndash1435 2013
[9] J Zhao and J Yang ldquoRemote sensing image denoising algorithmbased on fusion theory using cycle spinning contourlet trans-form and total variation minimizationrdquo Acta Photonica Sinicavol 39 no 9 pp 1658ndash1665 2010
[10] C L Matson and M C Roggemann ldquoNoise reduction inadaptive-optics imagery with the use of support constraintsrdquoApplied Optics vol 34 no 5 pp 767ndash780 1995
[11] DWTyler andC LMatson ldquoReduction of nonstationary noisein telescope imagery using a support constraintrdquoOptics Expressvol 1 no 11 pp 347ndash354 1997
[12] L Zhang J YangW Su et al ldquoResearch on blind deconvolutionalgorithm of multiframe turbulence-degraded imagesrdquo Journalof Information and Computational Science vol 10 no 12 pp3625ndash3634 2013
10 Abstract and Applied Analysis
[13] A P Dempster N M Laird and D B Rubin ldquoMaximumlikelihood from incomplete data via the EM algorithmrdquo Journalof the Royal Statistical Society B vol 39 no 1 pp 1ndash38 1977
[14] L M Mugnier J Conan T Fusco and V Michau ldquoJoint maxi-mum A posteriori estimation of object and PSF for turbulence-degraded imagesrdquo in Bayesian Inference for Inverse Problemsvol 3459 of Proceedings of SPIE pp 50ndash61 San Diego CalifUSA July 1998
[15] H Hanyu Research on Image Restoration Algorithms in ImagingDetection System HuazhongUniversity of Science and Technol-ogy Wuhan China 2004
[16] F Tsumuraya ldquoIterative blind deconvolution method usingLucyrsquos algorithmrdquo Astronomy amp Astrophysics vol 282 no 2 pp699ndash708 1994
[17] S Zhang J Li Y Yang and Z Zhang ldquoBlur identification ofturbulence-degraded IR imagesrdquoOptics and Precision Engineer-ing vol 21 no 2 pp 514ndash521 2013
[18] B Chen C Cheng S GuoG Pu andZGeng ldquoUnsymmetricalmulti-limit iterative blind deconvolution algorithm for adaptiveoptics image restorationrdquoHigh Power Laser and Particle Beamsvol 23 no 2 pp 313ndash318 2011
Submit your manuscripts athttpwwwhindawicom
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
MathematicsJournal of
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Mathematical Problems in Engineering
Hindawi Publishing Corporationhttpwwwhindawicom
Differential EquationsInternational Journal of
Volume 2014
Applied MathematicsJournal of
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Probability and StatisticsHindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Journal of
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Mathematical PhysicsAdvances in
Complex AnalysisJournal of
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
OptimizationJournal of
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
CombinatoricsHindawi Publishing Corporationhttpwwwhindawicom Volume 2014
International Journal of
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Operations ResearchAdvances in
Journal of
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Function Spaces
Abstract and Applied AnalysisHindawi Publishing Corporationhttpwwwhindawicom Volume 2014
International Journal of Mathematics and Mathematical Sciences
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
The Scientific World JournalHindawi Publishing Corporation httpwwwhindawicom Volume 2014
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Algebra
Discrete Dynamics in Nature and Society
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Decision SciencesAdvances in
Discrete MathematicsJournal of
Hindawi Publishing Corporationhttpwwwhindawicom
Volume 2014 Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Stochastic AnalysisInternational Journal of
Abstract and Applied Analysis 3
where 120579(sdot) is the wavefront phase error caused by the error offocal length or optical deviation
So the mathematical model of PSF can be written as
ℎ (119910 120579) =
1003816100381610038161003816100381610038161003816int 119901(119906)119890
minus119895sdot(2120587120582119897)sdot119910sdot119906119889119906
1003816100381610038161003816100381610038161003816
2
=
1003816100381610038161003816100381610038161003816int 119875(119906)119890
119895120579(119906)119890minus119895sdot(2120587120582119897)sdot119910sdot119906
119889119906
1003816100381610038161003816100381610038161003816
2
(6)
The wavefront phase error is caused by the focal lengtherror or optical deviation due to the influence of atmosphericturbulence the PSFmodel changes with the focal length erroror optical deviation over time as follows
ℎ (119910 120579119905) =
1003816100381610038161003816100381610038161003816int 119875(119906)119890
119895120579119905(119906)119890minus119895sdot(2120587120582119897)sdot119910sdot119906
119889119906
1003816100381610038161003816100381610038161003816
2
(7)
In the formula 120579119905(sdot) is the phase error caused by atmospheric
turbulence changing over timeIn terms of AO system the exposure time of observing
images is shorter than turbulent fluctuation If the intensity ofthe target is kept unchanged during exposure time themodelfor series images is
119892 (119910 120579119905119898 119891) = int ℎ (119910 minus 119909 120579
119905119898) 119891 (119909) 119889119909 (8)
In the formula 120579119905119898
is the phase error of turbulence for the119898th frame short exposure AO image
23 The Denoising Processing of AO Images In order toreduce the noise of AO images Chinese researchers putforward themethod to suppress noise based on total variation[8 9] This paper proposes a method to suppress noise that isbased on the power spectral density (abbreviated as PSD) ofimage and the supporting domain field of constraints imagesReference [10] analyzed the method to minimize the noisein symmetrical support domain field of AO images Theanalysis for this method is based on the variance ratio of realimages In this paper the assessment about denoise resultusing noise changed function The denoising process of AOimage is actual a kind of image noise corrected process inFourier spectral domain If theAO image has the symmetricalsupport field its noise mainly consist by CCD read-in noiseand AO noise and then noise power spectral density can beresearched Therefore suppress noise for observation images119892(119909) and 119892
119888(119909) is constrained by field of support domain
119892119888 (119909) = 119892 (119909)119908 (119909) (9)
In the formula 119908(119909) is weighting function in support fieldthe image spectrum 119866
119888(119906) is
119866119888 (119906) = int119866 (V)119882 (119906 minus V) 119889119906 (10)
where 119906 is the spatial frequency of image119866(119906) = 119878(119906)lowast119873(119906)119878(119906) is the signal frequency spectrum function 119873(119906) is thenoise spectrum function and bandwidth function 119882(119906) =119866(119906) lowast 120575 In the limited support domain field119882(119906) is verynarrow which nearly equals the 120575 times of the width of 119866(119906)
According to the definition of J M Conan the PSD ofan image is the Fourier transformation of its covariancedefining the power spectrum of noise and power spectrumof distortion image [11]
PSD(119906)119899119899 = ⟨10038161003816100381610038161003816Re 119866 (119906) minus Re 119866 (119906)10038161003816100381610038161003816
2
⟩
PSD(119906)119891119891 = ⟨10038161003816100381610038161003816Im 119866 (119906) minus Im 119866 (119906)10038161003816100381610038161003816
2
⟩
(11)
In the formula 119866(119906) is the Fourier transformation of image119892(119909) 119866(119906) is the mean value of sample 119866(119906)
Regard the circle as target supporting domain field itsdiameter is from 8 pixels to 185 pixels The definition of thechanging image noise function 119865
119899(119866) is
119865119899 (119866) =
int (PSD(119906)119888119899119899minus PSD(119906)119888
119899119899) 119889119906
intPSD(119906)119888119899119899119889119906
(12)
In the formula PSD(119906)119888119899119899is the power spectral density of the
image with constraints of support domain field and PSD(119906)119888119899119899
is the power spectral density of the image without constraintsof support domain field
24 Improved Expectation Maximization Algorithm Refer-ence [12] deduced the point spread function of multiframedegraded images with turbulence and the iteration formulaof target image based on maximum likelihood criterion butusing thismethod leads to the larger solution space Howeverwe put forward the technique of regularization to modify themaximum likelihood estimation algorithm introduce con-straints in accordance with AO image in physical reality andadopt the improved EM algorithm onmultiframe iteration ofAO images to acquire ideal 119891 and ℎ
119896
241 EM Algorithm by Combining Regularization TechniqueEM algorithm is put forward by Dempster et al [13] forparameters of maximum likelihood estimate of the iterativealgorithm which is iteratively going on two basic stepscalculation steps E (Expectation) and M (Maximization) Inthis paper a new regularizationmethod combined with priorknowledge is applied to the EMalgorithm realizing amethodof multiframe AO image restoration
Based on prior knowledge of target image and PSF jointprobability distribution function defining the maximizationfunction will get target image estimation 119891 and PSF estima-tion ℎ that is
[119891 ℎ] = max119891ℎ
(119901 (119892 | 119891 ℎ 119886)) = max119891ℎ
(exp(minus119869 (119891 ℎ))
119911)
(13)
In the formula 119911 is the normalization factor The costfunction is
[119891 ℎ] = max119891ℎ
(119901 (119892 | 119891 ℎ 119886)) = max119891ℎ
(exp(minus119869 (119891 ℎ))
119911)
(14)
4 Abstract and Applied Analysis
So the problem is converted to minimize cost function119869(119891 ℎ) that is [119891 ℎ] = min
119891ℎ(119869(119891 ℎ)) The cost function
of AO images and the optimization model of its parameterestimation are shown as follows
119869 (119891 ℎ) = 119869119899(119892 | 119891 ℎ 119886) + 119869
ℎ (ℎ | 119886) + 119869119891 (119891 | 119886)
st 119869119899(119892 | 119891 ℎ 119886) = sum
119894119895
1
1205902119899
[119892 (119894 119895) minus (ℎ lowast 119891) (119894 119895)]2
119869ℎ (ℎ | 119886) =
120579119905
2sum
119906
10038161003816100381610038161003816(119906) minus 119867(119906)
10038161003816100381610038161003816
2
PSDℎ (119906)
119869119891(119891 | 119886) = 120572sum
119894119895
Φ(119903 (119891 120574 (119894 119895)))
(15)
Symbol explanation the first item represents the noise costfunction the second item represents the constraints costfunction of PSF and the third item is the cost functionswith target edge constraint 119886 is prior information 1205902
119899is the
variance of mixed noise 120579119905is the wavefront phase error
the power spectral density of the point spread function isPSD(119906)
ℎ= ⟨|(119906) minus 119867(119906)|
2
⟩ (119906) is the Fourier transfor-mation of ℎ(119909) 119867(119906) is the Fourier transform of averagePSF the estimation of PSD
ℎ(119906) can be estimated through the
AO system closed-loop control data and 120572 and 120574(nabla119909) areregularization adjustment parameters using the target edgeconstraint theory of Mugnier isotropic [14] as
Φ(119901) = 119901 minus ln (1 + 119901)
119901 (119891 120574 (nabla119909)) =
10038171003817100381710038171003817nabla119891 (119894 119895)
10038171003817100381710038171003817
120574 (nabla119909)
10038171003817100381710038171003817nabla119891 (119894 119895)
10038171003817100381710038171003817= ((nabla119891
119909(119894 119895))
2+ (nabla119891
119910(119894 119895))
2
)
12
(16)
where 120572 is a constant coefficient 120574(sdot) is a regularized functionthat has variable regularization coefficient associated with thegradient of each point 120574(nabla119909) = expminus|nabla119909|
221205852
is based on [15]0 lt 120574(nabla119909) le 1 parameter 120585 is the controlled smoothingcoefficient for selection and nabla119909 is the gradient amplitudein four directions for the point 119909(119894 119895) (where 119894 119895 are pixelcoordinates and 119909 is gray value of pixel)
This paper will apply all the prior information on theestimation of target and PSF in order to get the optimalestimation of target and the point spread function (PSF)Combine formulas (2) (8) and (15) to establish the costfunction for joint deconvolutionwithmultiframeAO images
119869multi (119891 ℎ) = 119869119899 (119892119898 | 119891 ℎ119898 119886) + 119869ℎ (ℎ119898 | 119886) + 119869119891 (119891 | 119886)
(17)
where119898 is the number of frames
242 Implement of Image Restoration Algorithm This paperis combining with the basic idea of the EM algorithm [13]through limited times of cycle calculations find the mostsuitable PSF estimation on the purpose of restoring target
image drastically modify the cost function (17) and make itcorrespond to full data [15] if 119892
119896(119910 | 119909) is full data set of 119896
times observation results 119892119896(119910) of observed image 119892
119896(119910) and
the expectation of cost function are as follows
119864 (119892119896(119910)) = 119864(sum
119909
119892119896(119910 | 119909))
= sum
119909
ℎ119896(119910 minus 119909) 119891 (119909)
119864 (119869multi (119891 ℎ)) = 119864(119869119899 (sum119909
119892119896| 119891 ℎ119898 119886))
+ 119864 (119869ℎ(ℎ119898| 119886)) + 119864 (119869
119891(119891 | 119886))
(18)
In the iteration process it is supposed that the 119899 minus1 iteration result is known 119864119899(119869multi(119891 ℎ)) represents thecurrent expectation of the cost function In order tominimizethe expectation of the cost function formula (18) can be dif-ferentiated by ℎ119899
119896(119909) and 119891119899(119909) then make the differentiation
equal to zero that is
120597119864119899(119869multi (119891 ℎ))
120597ℎ119899
119896(119909)
= 0120597119864119899(119869multi (119891 ℎ))
120597119891119899 (119909)= 0 (19)
The derivation process in detail is omitted so the finalsolutions are
119891119899+1
119894(119909) = 119891
119899
119894(119909)
+120579119905
120572 lowast Φ (119903 (119891119899
119894(119909) 120574 (119909)))
times sum
119910isin119884
sum
119898
ℎ119899
119894(119910 minus 119909)
times (119892119898minus sum119911isin119883ℎ119899
119894(119910 minus 119911) 119891
119899
119894(119911)
1205902119894
)
(20)
ℎ119899+1
119894(119909) = ℎ
119899
119894(119909)
+120579119905
2sum
119906
10038161003816100381610038161003816119894(119906) minus 119867
119894(119906)10038161003816100381610038161003816
2
119875119878119863ℎ 119894 (119906)
times sum
119910isin119884
119891119899+1
119894(119910 minus 119909)
times (119892119898minus sum119911isin119883ℎ119899
119894(119910 minus 119911) 119891
119899
119894(119911)
1205902119894
)
(21)
In this paper the specific steps of image restorationalgorithm are as follows
Step 1 (initialization operation)
(1) According to the method of Section 23 constrainthe support domain field of observed image and dealwith the noise and so forth regard the result of onetime Wiener filtering as the initial value of targetestimation119891
0
Abstract and Applied Analysis 5
(a) Original image (b) Degraded image frame 1 (c) Degraded image frame 2 (d) Degraded image frame 3
(e) Restored image by Wiener-IBD
(f) Restored image by RL-IBD (g) Restored image by our algo-rithm
Figure 2 Degraded images in experiment and restoration comparison of three algorithms
(a) Degraded image frame 1 (b) Degraded image frame 2 (c) Degraded image frame 3 (d) Degraded image frame 4
(e) Estimated PSF by ouralgorithm
(f) Restored image by our algo-rithm
(g) Restored image by RL-IBDalgorithm
(h) Restored image by Weiner-IBD algorithm
Figure 3 Multiframe original images and restoration comparison of three algorithms
(2) According to the method of Section 22 for the initialPSF estimation ℎ
0 normalized processing should be
taken at same time
(3) Set theMax iteration of extrinsic cycles
(4) SetMax count of target and PSF estimation
(5) Set the inner loop counter variable h count by PSFestimation variable f count of target inner cyclecount
Step 2 (the calculation of parameter values)
(1) According to the formulas of Section 23 the variance1205902
119899of mixed noise of each degradation image is
estimated
(2) According to the method and calculation formulasfrom Section 22 calculate the power spectral densityPSDℎ(119906) of PSF and estimate the wavefront phase
error 120579119905
6 Abstract and Applied Analysis
0
50
100
150
200
0
50
100
150
0
50
100
150
Y (pixel)X (pixel)
Gra
y
(a) Energy spectrum of observed image in Figure 3(a)
0
50
100
150
0
50
100
150
200
250
0
50
100
150
Y (pixel) X(pixel)
Gra
y
(b) Energy spectrum of our algorithm restored image in Figure 3(f)
0
50
100
150
200
250
0
50
100
150
0
50
100
150
Y (pixel)X (pixel)
Gra
y
(c) Energy spectrum of RL-IBD algorithm restored image in Figure 3(g)
0
100
100
150
00
50
50
100
150
200
250
2040
6080
Y (pixel)
X(pixel)
Gra
y
120
(d) Energy spectrum of Wiener-IBD algorithm restored image inFigure 3(h)
Figure 4 Energy spectrum of observed image and restoration comparison of three algorithms
(3) According to the scheme and formulas ofSection 241 estimate the regularization parameters120572 and 120574(nabla119909) of target edge
Step 3 (iterate calculation 119894 = 0 1 119872119886119909 119894119905119890119903119886119905119894119900119899)
(1) The inner loop count variable of PSF h count = 0(2) The iteration process of PSF estimation 119895 = 0 1
119872119886119909 119888119900119906119899119905
(a) Complete the PSF estimation ℎ(119894) with the con-jugate gradient method using optimized for-mula (20)
(b) Consider h count++ j++(c) Judge the value of the loop variable 119895 if 119895 lt119872119886119909 119888119900119906119899119905 continue otherwise turn to (3)
(3) The inner loop counter variable of target estimation119891 119888119900119906119899119905 = 0
(4) The iteration process of target estimation 119896 =
0 1 119872119886119909 119888119900119906119899119905
(a) The conjugate gradient method was used tooptimize formula (21) to complete target imageestimation 119891(119894)
(b) consider f count++ 119896 + +(c) judge loop variable 119896 if 119896 lt 119872119886119909 119888119900119906119899119905 con-
tinue otherwise turn to (5)
(5) Judge whether the extrinsic loop is over if it is 119894 gt119898119886119909 119894119905119890119903119886119905119894119900119899 then turn to Step 4
(6) 119894 = 119894 + 1 return to (1) and continue to loop
Step 4 (the algorithm is in the end) The cycle does not stopuntil the algorithm converges and the output target imageestimation is 119891 that is over
Abstract and Applied Analysis 7
(a) Degraded image frame 1 (b) Degraded image frame 2 (c) Degraded image frame 3 (d) Degraded image frame 4
(e) Restored image by Weiner-IBD algorithm
(f) Restored image by RL-IBDalgorithm
(g) Estimated PSF by ouralgorithm
(h) Restored image by our algo-rithm
0
0
2
4
6
8
100
100
200200
300
400
150
250
times104
50
0
Y (pixel)X
(pixel)
Gra
y
(i) Energy spectrum of observed image in Figure 5(a)
0
0
0
20
40
60
80
100
50
100
150
2
4
6
8
times104
Y (pixel)
X (pixel)
Gra
y
(j) Energy spectrum of our algorithm restored image in Figure 5(h)
01
02
03
04
05
06
07
08
Inte
nsity
Our algorithm
Wiener-IBDRL-IBD
(k) A line plot taken access the horizontal axis of image restored byWeiner-IBD RL-IBD and our algorithm
Figure 5 The multiframe observed stellar images and restoration comparison of three algorithms
8 Abstract and Applied Analysis
3 Results and Analysis of AO ImageRestoration Experiments
31 The Restoration Experiment on Simulate AO Images Thealgorithm proposed in this paper is used to do the simulationexperiment on degradation spot images the evaluation ofimage restoration quality is judged by using rootmean squareerror (RMSE) and improved ΔSNR Consider
RMSE = radic1
11987311198732sum
119894119895
(119891 (119894 119895) minus 119891 (119894 119895))2
ΔSNR = 100 log10
1003817100381710038171003817119892 minus 11989110038171003817100381710038172
10038171003817100381710038171003817119891 minus 119891
10038171003817100381710038171003817
2
(22)
The evaluation of estimation accuracy is defined as fol-lows
120576estimate =
radicsum119894119895isin119878ℎ
10038161003816100381610038161003816ℎ(119894 119895) minus ℎ(119894 119895)
10038161003816100381610038161003816
2
radicsum119894119895isin119878ℎ
1003816100381610038161003816ℎ(119894 119895)10038161003816100381610038162
(23)
In the formula 1198731 and 1198732 represent the total number ofpixels on 119909-axis and 119910-axis on the image respectively (119894 119895)represents a pixel point on image and 119891(119894 119895) represents thegray value of the point (119894 119895) on ideal image
In the simulation experiments of this paper the originalimage is infrared simulation imaging with spots for multipletargets with long distance 10 frames of simulation degradedimage series are generated by image degradation softwarefrom the key optical laboratory of Chinese Academy ofSciences and then add Gaussian white noise and make thesignal-to-noise ratio SNR of image for 20 dB Set parametersthe same as adaptive optical telescope of 12m on observatoryin YunnanThemain parameters of telescope imaging systemare the atmospheric coherence length 119903
0= 13 cm diameter
of telescope 119863 = 108m comprehensive focal length ofoptical system 119897 = 2242m the wavelength in center 120582 =700 nm and the pixel size of CCD is 74 120583m Experiments onIBD algorithm based on RL iterative [16] (RL-IBD) iterativeblind restoration algorithm based on theWiener filtering [17](Wiener-IBD) and the proposed algorithm are compared
Figure 2 is the compared results of the original imagesimulated multiframe degradation images and restorationimages based on three kinds of algorithm the image size is217 times 218 Figure 2(a) is the original target image and tenframes turbulence degraded image sequences are generatedby using image degradation simulation software as shownin Figures 2(b) 2(c) and 2(d) (to save space we show onlythree frames the rest is omitted) Figure 2(e) is the Wiener-IBD restored image RMSE = 1635 Figure 2(f) is the RL-IBD restored image the number of iterations is 800 timesRMSE = 1031 Figure 2(g) is restored image used methodin this paper the variance of wavefront phase error 120579
119905(sdot) is
614 times 10minus3 ℎ0= 826 times 10
minus3 and 1205902119899= 11 times 10
minus5 that theregularization parameters in the experiments are 120572 = 125the parameter of regularization function is 120585 = 134 extrinsiciterations for 100 times inner loop of the PSF for 4 image
Table 1 Comparison of ΔSNR and 120576estimate values among threealgorithms
Algorithmname ΔSNR (dB) 120576estimate Iteration Time-consuming
Wiener-IBD 1982 362 800 20645 sRL-IBD 3046 449 800 17236 sOur algorithm 4164 510 600 10923 s
cycles for 2 namely the total times of cycling is 600 RMSE =1005 Comparing with ΔSNR and precision from Figures2(e) 2(f) and 2(g) are shown in Table 1 Comparison resultsshow that the proposed algorithm in this paper is better thanWiener-IBD algorithm and RL-IBD algorithm and needs lesstime of iterations
32 Restoration Experiments about Binary AO Images With-out the reference of ideal image using the objective qualityevaluation with no reference image the Full Width HalfMaximum (abbreviated as FWHM) [18] is used as objectiveevaluation for image restoration in this paper FWHM refersto imaging peak pixels is two times of half peak includingthe FWHM on 119909 direction and 119910 direction for the betterevaluation in astronomical observation areas computationformula is shown as follows
FWHM = radicFWHM 1199092 + FWHM 1199102 (24)
The experiment for the binary image restoration withoutreference images is carried out by the algorithm proposed inthis paper The binary images for experiments were taken bythe 12m AO telescope from Chinese Academy of Sciencesin Yunnan observatory on December 3 2006 The size forimaging CCD is 320 times 240 pixels array the main parametersof imaging system are as follows the atmospheric coherentlength 119903
0= 13 cm telescope aperture 119863 = 103 all
imaging observation band is 07sim09 center wavelength 120582 =072 the CCD pixel size is 67 120583m and optical imagingfocal length 119897 = 20 Recovery experiments based on RLIBD algorithm Wiener-IBD algorithm and the proposedalgorithm in this paper are compared the frame selectiontechnology is taken in this paper 40 frames are picked from100 frames of degradation images as blind convolution imagesto be processed
Figure 3 is the comparison results ofmultiframedegradedimages and restoration images based on three kinds ofalgorithms Figures 3(a) 3(b) 3(c) and 3(d) are for theobservation of the binary image (to save space we show onlyfour frames the rest is omitted) Figure 3(e) is the estimationof the point spread function (PSF) based on the algorithmin this paper Figure 3(f) is the restored image based on themethod proposed in this paper the variance of wavefrontphase error 120579
119905(sdot) is 604 times 10minus3 ℎ
0= 903 times 10
minus3 and1205902
119899= 132 times 10
minus5 so that the regularization parameter inthis experiment is 120572 = 115 the parameter of regularizationfunction is 120585 = 134 extrinsic iterations for 50 times the inner
Abstract and Applied Analysis 9
loop of PSF for 4 times and image loop for 2 times namelyThe total times of iteration are 300 FWHM = 617 pixels andthe restoration results are very close to the diffraction limit of12m AO telescope Figure 3(g) is the restoration result basedonRL-IBD algorithm FWHM=562 pixels Figure 3(h) is therestoration result based onWiener-IBD algorithm FWHM=476 pixels Figure 4 is the sketch of energy for the observationimage and three kinds of restored images based on threerestoration algorithms With the comparison it shows thatthe proposed algorithm in this paper is better than Wiener-IBD algorithm and RL-IBD algorithm and needs less time ofiterations
33 Restoration Experiments about Stellar AO Images Alsothe restoration experiment of stellar images was carriedout using the algorithm in this paper the stellar imagesfor experiment were taken by 12m AO telescope from theChinese Academy of Sciences in Yunnan observatory onDecember 1 2006 collection The parameters of the opticalimaging system are the same as that of binary images andthe method of parameters selection for experiments in thispaper is nearly the same Figure 5 is the comparison ofmultiframe stellar observation images and restoration resultsbased on the three algorithms Figures 5(a) 5(b) 5(c) and5(d) are the observation of stellar images (to save space weshow only 4) Figure 5(e) is for the Wiener-IBD restoredimage Figure 5(f) is the RL-IBD restored image Figure 5(g)is the estimation of the point spread function (PSF) basedon algorithm in this paper the iterative initial value ℎ
0=
956 times 10minus3 PSF iterative for six times Figure 5(h) is the
restored image based on algorithm in this paper Figure 5(i)is the energy of the observed image for stellar Figure 5(j) isthe energy of restoration image based on the algorithm in thispaper Figure 5(k) is the graph of comparison results for threerestoration algorithms
4 Conclusions
In this paper according to the characteristic of adaptiveoptics image we establish a degradation image model formultiframe exposure image series and the mathematicalmodel of the point spread function (PSF) for adaptive opticalsystem after closed loop correction is derived the PSF modelbased on phase error and changing over time is solved andwe use the power spectral density of image and constrainsupport domain method for AO image denoising processingFinally this paper applied the actual AO imaging systemparameters as prior knowledge combined with regularizationtechnique to improve the EM blind deconvolution algo-rithm Experimental results show that the established blinddeconvolution algorithm based on improved EM methodcan effectively eliminate the atmospheric turbulence andeliminate the noise pollution brought by the AO closed-loopcorrection which improves the resolution of the adaptiveoptical image The image process of AO observed imagesshows that the restoration quality is better than the samekind of IBD algorithm It also shows that the restoration ofbinary images using the algorithmof this paper has carried on
the identification of PSF and completed the AO image afterprocessing the results of actual AO image restoration haveimportant application value
Conflict of Interests
The authors declare that there is no conflict of interestsregarding the publication of this paper
Acknowledgments
This project is supported by the Scientific and TechnologicalResearch Project of Jinlin province Department of Edu-cation China (no 2013145 and no 201363) and the basisstudy project is supported by Jinlin Provincial Science ampTechnology Department China (no 201105055)
References
[1] J Wenhan ldquoAdaptive optical technologyrdquo Chinese Journal ofNature vol 28 pp 7ndash13 2005
[2] G R Ayers and J C Dainty ldquoIterative blind deconvolutionmethod and its applicationsrdquo Optics Letters vol 13 no 7 pp547ndash549 1988
[3] L Dayu H Lifa and M Quanquan ldquoA high resolution liquidcrystal adaptive optics systemrdquo Acta Photonica Sinica vol 37pp 506ndash508 2008
[4] C Rao F Shen andW Jiang ldquoAnalysis of closed-loopwavefrontresidual error of adaptive optical system using the method ofpower spectrumrdquo Acta Optica Sinica vol 20 no 1 pp 68ndash732000
[5] B Chen The theory and algorithms of adaptive optics imagerestoration [PhD thesis] Information Engineering UniversityZhengzhou China 2008
[6] T Yu R Changhui and W Kai ldquoPost-processing of adaptiveoptics images based on frame selection and multi-frame blinddeconvolutionrdquo inAdaptive Optics Systems vol 7015 of Proceed-ings of SPIE Marseille France June 2008
[7] T J Schulz ldquoNonlinear models and methods for space-objectimaging through atmospheric turbulencerdquo in Proceedings ofthe IEEE International Conference on Image Processing vol 9Lausanne Switzerland 1996
[8] L Yaning G Lei and L Huihui ldquoTotal variation basedband-limited sheralets transform for image denoisingrdquo ActaPhotonica Sinica vol 242 pp 1430ndash1435 2013
[9] J Zhao and J Yang ldquoRemote sensing image denoising algorithmbased on fusion theory using cycle spinning contourlet trans-form and total variation minimizationrdquo Acta Photonica Sinicavol 39 no 9 pp 1658ndash1665 2010
[10] C L Matson and M C Roggemann ldquoNoise reduction inadaptive-optics imagery with the use of support constraintsrdquoApplied Optics vol 34 no 5 pp 767ndash780 1995
[11] DWTyler andC LMatson ldquoReduction of nonstationary noisein telescope imagery using a support constraintrdquoOptics Expressvol 1 no 11 pp 347ndash354 1997
[12] L Zhang J YangW Su et al ldquoResearch on blind deconvolutionalgorithm of multiframe turbulence-degraded imagesrdquo Journalof Information and Computational Science vol 10 no 12 pp3625ndash3634 2013
10 Abstract and Applied Analysis
[13] A P Dempster N M Laird and D B Rubin ldquoMaximumlikelihood from incomplete data via the EM algorithmrdquo Journalof the Royal Statistical Society B vol 39 no 1 pp 1ndash38 1977
[14] L M Mugnier J Conan T Fusco and V Michau ldquoJoint maxi-mum A posteriori estimation of object and PSF for turbulence-degraded imagesrdquo in Bayesian Inference for Inverse Problemsvol 3459 of Proceedings of SPIE pp 50ndash61 San Diego CalifUSA July 1998
[15] H Hanyu Research on Image Restoration Algorithms in ImagingDetection System HuazhongUniversity of Science and Technol-ogy Wuhan China 2004
[16] F Tsumuraya ldquoIterative blind deconvolution method usingLucyrsquos algorithmrdquo Astronomy amp Astrophysics vol 282 no 2 pp699ndash708 1994
[17] S Zhang J Li Y Yang and Z Zhang ldquoBlur identification ofturbulence-degraded IR imagesrdquoOptics and Precision Engineer-ing vol 21 no 2 pp 514ndash521 2013
[18] B Chen C Cheng S GuoG Pu andZGeng ldquoUnsymmetricalmulti-limit iterative blind deconvolution algorithm for adaptiveoptics image restorationrdquoHigh Power Laser and Particle Beamsvol 23 no 2 pp 313ndash318 2011
Submit your manuscripts athttpwwwhindawicom
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
MathematicsJournal of
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Mathematical Problems in Engineering
Hindawi Publishing Corporationhttpwwwhindawicom
Differential EquationsInternational Journal of
Volume 2014
Applied MathematicsJournal of
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Probability and StatisticsHindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Journal of
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Mathematical PhysicsAdvances in
Complex AnalysisJournal of
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
OptimizationJournal of
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
CombinatoricsHindawi Publishing Corporationhttpwwwhindawicom Volume 2014
International Journal of
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Operations ResearchAdvances in
Journal of
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Function Spaces
Abstract and Applied AnalysisHindawi Publishing Corporationhttpwwwhindawicom Volume 2014
International Journal of Mathematics and Mathematical Sciences
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
The Scientific World JournalHindawi Publishing Corporation httpwwwhindawicom Volume 2014
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Algebra
Discrete Dynamics in Nature and Society
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Decision SciencesAdvances in
Discrete MathematicsJournal of
Hindawi Publishing Corporationhttpwwwhindawicom
Volume 2014 Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Stochastic AnalysisInternational Journal of
4 Abstract and Applied Analysis
So the problem is converted to minimize cost function119869(119891 ℎ) that is [119891 ℎ] = min
119891ℎ(119869(119891 ℎ)) The cost function
of AO images and the optimization model of its parameterestimation are shown as follows
119869 (119891 ℎ) = 119869119899(119892 | 119891 ℎ 119886) + 119869
ℎ (ℎ | 119886) + 119869119891 (119891 | 119886)
st 119869119899(119892 | 119891 ℎ 119886) = sum
119894119895
1
1205902119899
[119892 (119894 119895) minus (ℎ lowast 119891) (119894 119895)]2
119869ℎ (ℎ | 119886) =
120579119905
2sum
119906
10038161003816100381610038161003816(119906) minus 119867(119906)
10038161003816100381610038161003816
2
PSDℎ (119906)
119869119891(119891 | 119886) = 120572sum
119894119895
Φ(119903 (119891 120574 (119894 119895)))
(15)
Symbol explanation the first item represents the noise costfunction the second item represents the constraints costfunction of PSF and the third item is the cost functionswith target edge constraint 119886 is prior information 1205902
119899is the
variance of mixed noise 120579119905is the wavefront phase error
the power spectral density of the point spread function isPSD(119906)
ℎ= ⟨|(119906) minus 119867(119906)|
2
⟩ (119906) is the Fourier transfor-mation of ℎ(119909) 119867(119906) is the Fourier transform of averagePSF the estimation of PSD
ℎ(119906) can be estimated through the
AO system closed-loop control data and 120572 and 120574(nabla119909) areregularization adjustment parameters using the target edgeconstraint theory of Mugnier isotropic [14] as
Φ(119901) = 119901 minus ln (1 + 119901)
119901 (119891 120574 (nabla119909)) =
10038171003817100381710038171003817nabla119891 (119894 119895)
10038171003817100381710038171003817
120574 (nabla119909)
10038171003817100381710038171003817nabla119891 (119894 119895)
10038171003817100381710038171003817= ((nabla119891
119909(119894 119895))
2+ (nabla119891
119910(119894 119895))
2
)
12
(16)
where 120572 is a constant coefficient 120574(sdot) is a regularized functionthat has variable regularization coefficient associated with thegradient of each point 120574(nabla119909) = expminus|nabla119909|
221205852
is based on [15]0 lt 120574(nabla119909) le 1 parameter 120585 is the controlled smoothingcoefficient for selection and nabla119909 is the gradient amplitudein four directions for the point 119909(119894 119895) (where 119894 119895 are pixelcoordinates and 119909 is gray value of pixel)
This paper will apply all the prior information on theestimation of target and PSF in order to get the optimalestimation of target and the point spread function (PSF)Combine formulas (2) (8) and (15) to establish the costfunction for joint deconvolutionwithmultiframeAO images
119869multi (119891 ℎ) = 119869119899 (119892119898 | 119891 ℎ119898 119886) + 119869ℎ (ℎ119898 | 119886) + 119869119891 (119891 | 119886)
(17)
where119898 is the number of frames
242 Implement of Image Restoration Algorithm This paperis combining with the basic idea of the EM algorithm [13]through limited times of cycle calculations find the mostsuitable PSF estimation on the purpose of restoring target
image drastically modify the cost function (17) and make itcorrespond to full data [15] if 119892
119896(119910 | 119909) is full data set of 119896
times observation results 119892119896(119910) of observed image 119892
119896(119910) and
the expectation of cost function are as follows
119864 (119892119896(119910)) = 119864(sum
119909
119892119896(119910 | 119909))
= sum
119909
ℎ119896(119910 minus 119909) 119891 (119909)
119864 (119869multi (119891 ℎ)) = 119864(119869119899 (sum119909
119892119896| 119891 ℎ119898 119886))
+ 119864 (119869ℎ(ℎ119898| 119886)) + 119864 (119869
119891(119891 | 119886))
(18)
In the iteration process it is supposed that the 119899 minus1 iteration result is known 119864119899(119869multi(119891 ℎ)) represents thecurrent expectation of the cost function In order tominimizethe expectation of the cost function formula (18) can be dif-ferentiated by ℎ119899
119896(119909) and 119891119899(119909) then make the differentiation
equal to zero that is
120597119864119899(119869multi (119891 ℎ))
120597ℎ119899
119896(119909)
= 0120597119864119899(119869multi (119891 ℎ))
120597119891119899 (119909)= 0 (19)
The derivation process in detail is omitted so the finalsolutions are
119891119899+1
119894(119909) = 119891
119899
119894(119909)
+120579119905
120572 lowast Φ (119903 (119891119899
119894(119909) 120574 (119909)))
times sum
119910isin119884
sum
119898
ℎ119899
119894(119910 minus 119909)
times (119892119898minus sum119911isin119883ℎ119899
119894(119910 minus 119911) 119891
119899
119894(119911)
1205902119894
)
(20)
ℎ119899+1
119894(119909) = ℎ
119899
119894(119909)
+120579119905
2sum
119906
10038161003816100381610038161003816119894(119906) minus 119867
119894(119906)10038161003816100381610038161003816
2
119875119878119863ℎ 119894 (119906)
times sum
119910isin119884
119891119899+1
119894(119910 minus 119909)
times (119892119898minus sum119911isin119883ℎ119899
119894(119910 minus 119911) 119891
119899
119894(119911)
1205902119894
)
(21)
In this paper the specific steps of image restorationalgorithm are as follows
Step 1 (initialization operation)
(1) According to the method of Section 23 constrainthe support domain field of observed image and dealwith the noise and so forth regard the result of onetime Wiener filtering as the initial value of targetestimation119891
0
Abstract and Applied Analysis 5
(a) Original image (b) Degraded image frame 1 (c) Degraded image frame 2 (d) Degraded image frame 3
(e) Restored image by Wiener-IBD
(f) Restored image by RL-IBD (g) Restored image by our algo-rithm
Figure 2 Degraded images in experiment and restoration comparison of three algorithms
(a) Degraded image frame 1 (b) Degraded image frame 2 (c) Degraded image frame 3 (d) Degraded image frame 4
(e) Estimated PSF by ouralgorithm
(f) Restored image by our algo-rithm
(g) Restored image by RL-IBDalgorithm
(h) Restored image by Weiner-IBD algorithm
Figure 3 Multiframe original images and restoration comparison of three algorithms
(2) According to the method of Section 22 for the initialPSF estimation ℎ
0 normalized processing should be
taken at same time
(3) Set theMax iteration of extrinsic cycles
(4) SetMax count of target and PSF estimation
(5) Set the inner loop counter variable h count by PSFestimation variable f count of target inner cyclecount
Step 2 (the calculation of parameter values)
(1) According to the formulas of Section 23 the variance1205902
119899of mixed noise of each degradation image is
estimated
(2) According to the method and calculation formulasfrom Section 22 calculate the power spectral densityPSDℎ(119906) of PSF and estimate the wavefront phase
error 120579119905
6 Abstract and Applied Analysis
0
50
100
150
200
0
50
100
150
0
50
100
150
Y (pixel)X (pixel)
Gra
y
(a) Energy spectrum of observed image in Figure 3(a)
0
50
100
150
0
50
100
150
200
250
0
50
100
150
Y (pixel) X(pixel)
Gra
y
(b) Energy spectrum of our algorithm restored image in Figure 3(f)
0
50
100
150
200
250
0
50
100
150
0
50
100
150
Y (pixel)X (pixel)
Gra
y
(c) Energy spectrum of RL-IBD algorithm restored image in Figure 3(g)
0
100
100
150
00
50
50
100
150
200
250
2040
6080
Y (pixel)
X(pixel)
Gra
y
120
(d) Energy spectrum of Wiener-IBD algorithm restored image inFigure 3(h)
Figure 4 Energy spectrum of observed image and restoration comparison of three algorithms
(3) According to the scheme and formulas ofSection 241 estimate the regularization parameters120572 and 120574(nabla119909) of target edge
Step 3 (iterate calculation 119894 = 0 1 119872119886119909 119894119905119890119903119886119905119894119900119899)
(1) The inner loop count variable of PSF h count = 0(2) The iteration process of PSF estimation 119895 = 0 1
119872119886119909 119888119900119906119899119905
(a) Complete the PSF estimation ℎ(119894) with the con-jugate gradient method using optimized for-mula (20)
(b) Consider h count++ j++(c) Judge the value of the loop variable 119895 if 119895 lt119872119886119909 119888119900119906119899119905 continue otherwise turn to (3)
(3) The inner loop counter variable of target estimation119891 119888119900119906119899119905 = 0
(4) The iteration process of target estimation 119896 =
0 1 119872119886119909 119888119900119906119899119905
(a) The conjugate gradient method was used tooptimize formula (21) to complete target imageestimation 119891(119894)
(b) consider f count++ 119896 + +(c) judge loop variable 119896 if 119896 lt 119872119886119909 119888119900119906119899119905 con-
tinue otherwise turn to (5)
(5) Judge whether the extrinsic loop is over if it is 119894 gt119898119886119909 119894119905119890119903119886119905119894119900119899 then turn to Step 4
(6) 119894 = 119894 + 1 return to (1) and continue to loop
Step 4 (the algorithm is in the end) The cycle does not stopuntil the algorithm converges and the output target imageestimation is 119891 that is over
Abstract and Applied Analysis 7
(a) Degraded image frame 1 (b) Degraded image frame 2 (c) Degraded image frame 3 (d) Degraded image frame 4
(e) Restored image by Weiner-IBD algorithm
(f) Restored image by RL-IBDalgorithm
(g) Estimated PSF by ouralgorithm
(h) Restored image by our algo-rithm
0
0
2
4
6
8
100
100
200200
300
400
150
250
times104
50
0
Y (pixel)X
(pixel)
Gra
y
(i) Energy spectrum of observed image in Figure 5(a)
0
0
0
20
40
60
80
100
50
100
150
2
4
6
8
times104
Y (pixel)
X (pixel)
Gra
y
(j) Energy spectrum of our algorithm restored image in Figure 5(h)
01
02
03
04
05
06
07
08
Inte
nsity
Our algorithm
Wiener-IBDRL-IBD
(k) A line plot taken access the horizontal axis of image restored byWeiner-IBD RL-IBD and our algorithm
Figure 5 The multiframe observed stellar images and restoration comparison of three algorithms
8 Abstract and Applied Analysis
3 Results and Analysis of AO ImageRestoration Experiments
31 The Restoration Experiment on Simulate AO Images Thealgorithm proposed in this paper is used to do the simulationexperiment on degradation spot images the evaluation ofimage restoration quality is judged by using rootmean squareerror (RMSE) and improved ΔSNR Consider
RMSE = radic1
11987311198732sum
119894119895
(119891 (119894 119895) minus 119891 (119894 119895))2
ΔSNR = 100 log10
1003817100381710038171003817119892 minus 11989110038171003817100381710038172
10038171003817100381710038171003817119891 minus 119891
10038171003817100381710038171003817
2
(22)
The evaluation of estimation accuracy is defined as fol-lows
120576estimate =
radicsum119894119895isin119878ℎ
10038161003816100381610038161003816ℎ(119894 119895) minus ℎ(119894 119895)
10038161003816100381610038161003816
2
radicsum119894119895isin119878ℎ
1003816100381610038161003816ℎ(119894 119895)10038161003816100381610038162
(23)
In the formula 1198731 and 1198732 represent the total number ofpixels on 119909-axis and 119910-axis on the image respectively (119894 119895)represents a pixel point on image and 119891(119894 119895) represents thegray value of the point (119894 119895) on ideal image
In the simulation experiments of this paper the originalimage is infrared simulation imaging with spots for multipletargets with long distance 10 frames of simulation degradedimage series are generated by image degradation softwarefrom the key optical laboratory of Chinese Academy ofSciences and then add Gaussian white noise and make thesignal-to-noise ratio SNR of image for 20 dB Set parametersthe same as adaptive optical telescope of 12m on observatoryin YunnanThemain parameters of telescope imaging systemare the atmospheric coherence length 119903
0= 13 cm diameter
of telescope 119863 = 108m comprehensive focal length ofoptical system 119897 = 2242m the wavelength in center 120582 =700 nm and the pixel size of CCD is 74 120583m Experiments onIBD algorithm based on RL iterative [16] (RL-IBD) iterativeblind restoration algorithm based on theWiener filtering [17](Wiener-IBD) and the proposed algorithm are compared
Figure 2 is the compared results of the original imagesimulated multiframe degradation images and restorationimages based on three kinds of algorithm the image size is217 times 218 Figure 2(a) is the original target image and tenframes turbulence degraded image sequences are generatedby using image degradation simulation software as shownin Figures 2(b) 2(c) and 2(d) (to save space we show onlythree frames the rest is omitted) Figure 2(e) is the Wiener-IBD restored image RMSE = 1635 Figure 2(f) is the RL-IBD restored image the number of iterations is 800 timesRMSE = 1031 Figure 2(g) is restored image used methodin this paper the variance of wavefront phase error 120579
119905(sdot) is
614 times 10minus3 ℎ0= 826 times 10
minus3 and 1205902119899= 11 times 10
minus5 that theregularization parameters in the experiments are 120572 = 125the parameter of regularization function is 120585 = 134 extrinsiciterations for 100 times inner loop of the PSF for 4 image
Table 1 Comparison of ΔSNR and 120576estimate values among threealgorithms
Algorithmname ΔSNR (dB) 120576estimate Iteration Time-consuming
Wiener-IBD 1982 362 800 20645 sRL-IBD 3046 449 800 17236 sOur algorithm 4164 510 600 10923 s
cycles for 2 namely the total times of cycling is 600 RMSE =1005 Comparing with ΔSNR and precision from Figures2(e) 2(f) and 2(g) are shown in Table 1 Comparison resultsshow that the proposed algorithm in this paper is better thanWiener-IBD algorithm and RL-IBD algorithm and needs lesstime of iterations
32 Restoration Experiments about Binary AO Images With-out the reference of ideal image using the objective qualityevaluation with no reference image the Full Width HalfMaximum (abbreviated as FWHM) [18] is used as objectiveevaluation for image restoration in this paper FWHM refersto imaging peak pixels is two times of half peak includingthe FWHM on 119909 direction and 119910 direction for the betterevaluation in astronomical observation areas computationformula is shown as follows
FWHM = radicFWHM 1199092 + FWHM 1199102 (24)
The experiment for the binary image restoration withoutreference images is carried out by the algorithm proposed inthis paper The binary images for experiments were taken bythe 12m AO telescope from Chinese Academy of Sciencesin Yunnan observatory on December 3 2006 The size forimaging CCD is 320 times 240 pixels array the main parametersof imaging system are as follows the atmospheric coherentlength 119903
0= 13 cm telescope aperture 119863 = 103 all
imaging observation band is 07sim09 center wavelength 120582 =072 the CCD pixel size is 67 120583m and optical imagingfocal length 119897 = 20 Recovery experiments based on RLIBD algorithm Wiener-IBD algorithm and the proposedalgorithm in this paper are compared the frame selectiontechnology is taken in this paper 40 frames are picked from100 frames of degradation images as blind convolution imagesto be processed
Figure 3 is the comparison results ofmultiframedegradedimages and restoration images based on three kinds ofalgorithms Figures 3(a) 3(b) 3(c) and 3(d) are for theobservation of the binary image (to save space we show onlyfour frames the rest is omitted) Figure 3(e) is the estimationof the point spread function (PSF) based on the algorithmin this paper Figure 3(f) is the restored image based on themethod proposed in this paper the variance of wavefrontphase error 120579
119905(sdot) is 604 times 10minus3 ℎ
0= 903 times 10
minus3 and1205902
119899= 132 times 10
minus5 so that the regularization parameter inthis experiment is 120572 = 115 the parameter of regularizationfunction is 120585 = 134 extrinsic iterations for 50 times the inner
Abstract and Applied Analysis 9
loop of PSF for 4 times and image loop for 2 times namelyThe total times of iteration are 300 FWHM = 617 pixels andthe restoration results are very close to the diffraction limit of12m AO telescope Figure 3(g) is the restoration result basedonRL-IBD algorithm FWHM=562 pixels Figure 3(h) is therestoration result based onWiener-IBD algorithm FWHM=476 pixels Figure 4 is the sketch of energy for the observationimage and three kinds of restored images based on threerestoration algorithms With the comparison it shows thatthe proposed algorithm in this paper is better than Wiener-IBD algorithm and RL-IBD algorithm and needs less time ofiterations
33 Restoration Experiments about Stellar AO Images Alsothe restoration experiment of stellar images was carriedout using the algorithm in this paper the stellar imagesfor experiment were taken by 12m AO telescope from theChinese Academy of Sciences in Yunnan observatory onDecember 1 2006 collection The parameters of the opticalimaging system are the same as that of binary images andthe method of parameters selection for experiments in thispaper is nearly the same Figure 5 is the comparison ofmultiframe stellar observation images and restoration resultsbased on the three algorithms Figures 5(a) 5(b) 5(c) and5(d) are the observation of stellar images (to save space weshow only 4) Figure 5(e) is for the Wiener-IBD restoredimage Figure 5(f) is the RL-IBD restored image Figure 5(g)is the estimation of the point spread function (PSF) basedon algorithm in this paper the iterative initial value ℎ
0=
956 times 10minus3 PSF iterative for six times Figure 5(h) is the
restored image based on algorithm in this paper Figure 5(i)is the energy of the observed image for stellar Figure 5(j) isthe energy of restoration image based on the algorithm in thispaper Figure 5(k) is the graph of comparison results for threerestoration algorithms
4 Conclusions
In this paper according to the characteristic of adaptiveoptics image we establish a degradation image model formultiframe exposure image series and the mathematicalmodel of the point spread function (PSF) for adaptive opticalsystem after closed loop correction is derived the PSF modelbased on phase error and changing over time is solved andwe use the power spectral density of image and constrainsupport domain method for AO image denoising processingFinally this paper applied the actual AO imaging systemparameters as prior knowledge combined with regularizationtechnique to improve the EM blind deconvolution algo-rithm Experimental results show that the established blinddeconvolution algorithm based on improved EM methodcan effectively eliminate the atmospheric turbulence andeliminate the noise pollution brought by the AO closed-loopcorrection which improves the resolution of the adaptiveoptical image The image process of AO observed imagesshows that the restoration quality is better than the samekind of IBD algorithm It also shows that the restoration ofbinary images using the algorithmof this paper has carried on
the identification of PSF and completed the AO image afterprocessing the results of actual AO image restoration haveimportant application value
Conflict of Interests
The authors declare that there is no conflict of interestsregarding the publication of this paper
Acknowledgments
This project is supported by the Scientific and TechnologicalResearch Project of Jinlin province Department of Edu-cation China (no 2013145 and no 201363) and the basisstudy project is supported by Jinlin Provincial Science ampTechnology Department China (no 201105055)
References
[1] J Wenhan ldquoAdaptive optical technologyrdquo Chinese Journal ofNature vol 28 pp 7ndash13 2005
[2] G R Ayers and J C Dainty ldquoIterative blind deconvolutionmethod and its applicationsrdquo Optics Letters vol 13 no 7 pp547ndash549 1988
[3] L Dayu H Lifa and M Quanquan ldquoA high resolution liquidcrystal adaptive optics systemrdquo Acta Photonica Sinica vol 37pp 506ndash508 2008
[4] C Rao F Shen andW Jiang ldquoAnalysis of closed-loopwavefrontresidual error of adaptive optical system using the method ofpower spectrumrdquo Acta Optica Sinica vol 20 no 1 pp 68ndash732000
[5] B Chen The theory and algorithms of adaptive optics imagerestoration [PhD thesis] Information Engineering UniversityZhengzhou China 2008
[6] T Yu R Changhui and W Kai ldquoPost-processing of adaptiveoptics images based on frame selection and multi-frame blinddeconvolutionrdquo inAdaptive Optics Systems vol 7015 of Proceed-ings of SPIE Marseille France June 2008
[7] T J Schulz ldquoNonlinear models and methods for space-objectimaging through atmospheric turbulencerdquo in Proceedings ofthe IEEE International Conference on Image Processing vol 9Lausanne Switzerland 1996
[8] L Yaning G Lei and L Huihui ldquoTotal variation basedband-limited sheralets transform for image denoisingrdquo ActaPhotonica Sinica vol 242 pp 1430ndash1435 2013
[9] J Zhao and J Yang ldquoRemote sensing image denoising algorithmbased on fusion theory using cycle spinning contourlet trans-form and total variation minimizationrdquo Acta Photonica Sinicavol 39 no 9 pp 1658ndash1665 2010
[10] C L Matson and M C Roggemann ldquoNoise reduction inadaptive-optics imagery with the use of support constraintsrdquoApplied Optics vol 34 no 5 pp 767ndash780 1995
[11] DWTyler andC LMatson ldquoReduction of nonstationary noisein telescope imagery using a support constraintrdquoOptics Expressvol 1 no 11 pp 347ndash354 1997
[12] L Zhang J YangW Su et al ldquoResearch on blind deconvolutionalgorithm of multiframe turbulence-degraded imagesrdquo Journalof Information and Computational Science vol 10 no 12 pp3625ndash3634 2013
10 Abstract and Applied Analysis
[13] A P Dempster N M Laird and D B Rubin ldquoMaximumlikelihood from incomplete data via the EM algorithmrdquo Journalof the Royal Statistical Society B vol 39 no 1 pp 1ndash38 1977
[14] L M Mugnier J Conan T Fusco and V Michau ldquoJoint maxi-mum A posteriori estimation of object and PSF for turbulence-degraded imagesrdquo in Bayesian Inference for Inverse Problemsvol 3459 of Proceedings of SPIE pp 50ndash61 San Diego CalifUSA July 1998
[15] H Hanyu Research on Image Restoration Algorithms in ImagingDetection System HuazhongUniversity of Science and Technol-ogy Wuhan China 2004
[16] F Tsumuraya ldquoIterative blind deconvolution method usingLucyrsquos algorithmrdquo Astronomy amp Astrophysics vol 282 no 2 pp699ndash708 1994
[17] S Zhang J Li Y Yang and Z Zhang ldquoBlur identification ofturbulence-degraded IR imagesrdquoOptics and Precision Engineer-ing vol 21 no 2 pp 514ndash521 2013
[18] B Chen C Cheng S GuoG Pu andZGeng ldquoUnsymmetricalmulti-limit iterative blind deconvolution algorithm for adaptiveoptics image restorationrdquoHigh Power Laser and Particle Beamsvol 23 no 2 pp 313ndash318 2011
Submit your manuscripts athttpwwwhindawicom
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
MathematicsJournal of
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Mathematical Problems in Engineering
Hindawi Publishing Corporationhttpwwwhindawicom
Differential EquationsInternational Journal of
Volume 2014
Applied MathematicsJournal of
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Probability and StatisticsHindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Journal of
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Mathematical PhysicsAdvances in
Complex AnalysisJournal of
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
OptimizationJournal of
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
CombinatoricsHindawi Publishing Corporationhttpwwwhindawicom Volume 2014
International Journal of
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Operations ResearchAdvances in
Journal of
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Function Spaces
Abstract and Applied AnalysisHindawi Publishing Corporationhttpwwwhindawicom Volume 2014
International Journal of Mathematics and Mathematical Sciences
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
The Scientific World JournalHindawi Publishing Corporation httpwwwhindawicom Volume 2014
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Algebra
Discrete Dynamics in Nature and Society
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Decision SciencesAdvances in
Discrete MathematicsJournal of
Hindawi Publishing Corporationhttpwwwhindawicom
Volume 2014 Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Stochastic AnalysisInternational Journal of
Abstract and Applied Analysis 5
(a) Original image (b) Degraded image frame 1 (c) Degraded image frame 2 (d) Degraded image frame 3
(e) Restored image by Wiener-IBD
(f) Restored image by RL-IBD (g) Restored image by our algo-rithm
Figure 2 Degraded images in experiment and restoration comparison of three algorithms
(a) Degraded image frame 1 (b) Degraded image frame 2 (c) Degraded image frame 3 (d) Degraded image frame 4
(e) Estimated PSF by ouralgorithm
(f) Restored image by our algo-rithm
(g) Restored image by RL-IBDalgorithm
(h) Restored image by Weiner-IBD algorithm
Figure 3 Multiframe original images and restoration comparison of three algorithms
(2) According to the method of Section 22 for the initialPSF estimation ℎ
0 normalized processing should be
taken at same time
(3) Set theMax iteration of extrinsic cycles
(4) SetMax count of target and PSF estimation
(5) Set the inner loop counter variable h count by PSFestimation variable f count of target inner cyclecount
Step 2 (the calculation of parameter values)
(1) According to the formulas of Section 23 the variance1205902
119899of mixed noise of each degradation image is
estimated
(2) According to the method and calculation formulasfrom Section 22 calculate the power spectral densityPSDℎ(119906) of PSF and estimate the wavefront phase
error 120579119905
6 Abstract and Applied Analysis
0
50
100
150
200
0
50
100
150
0
50
100
150
Y (pixel)X (pixel)
Gra
y
(a) Energy spectrum of observed image in Figure 3(a)
0
50
100
150
0
50
100
150
200
250
0
50
100
150
Y (pixel) X(pixel)
Gra
y
(b) Energy spectrum of our algorithm restored image in Figure 3(f)
0
50
100
150
200
250
0
50
100
150
0
50
100
150
Y (pixel)X (pixel)
Gra
y
(c) Energy spectrum of RL-IBD algorithm restored image in Figure 3(g)
0
100
100
150
00
50
50
100
150
200
250
2040
6080
Y (pixel)
X(pixel)
Gra
y
120
(d) Energy spectrum of Wiener-IBD algorithm restored image inFigure 3(h)
Figure 4 Energy spectrum of observed image and restoration comparison of three algorithms
(3) According to the scheme and formulas ofSection 241 estimate the regularization parameters120572 and 120574(nabla119909) of target edge
Step 3 (iterate calculation 119894 = 0 1 119872119886119909 119894119905119890119903119886119905119894119900119899)
(1) The inner loop count variable of PSF h count = 0(2) The iteration process of PSF estimation 119895 = 0 1
119872119886119909 119888119900119906119899119905
(a) Complete the PSF estimation ℎ(119894) with the con-jugate gradient method using optimized for-mula (20)
(b) Consider h count++ j++(c) Judge the value of the loop variable 119895 if 119895 lt119872119886119909 119888119900119906119899119905 continue otherwise turn to (3)
(3) The inner loop counter variable of target estimation119891 119888119900119906119899119905 = 0
(4) The iteration process of target estimation 119896 =
0 1 119872119886119909 119888119900119906119899119905
(a) The conjugate gradient method was used tooptimize formula (21) to complete target imageestimation 119891(119894)
(b) consider f count++ 119896 + +(c) judge loop variable 119896 if 119896 lt 119872119886119909 119888119900119906119899119905 con-
tinue otherwise turn to (5)
(5) Judge whether the extrinsic loop is over if it is 119894 gt119898119886119909 119894119905119890119903119886119905119894119900119899 then turn to Step 4
(6) 119894 = 119894 + 1 return to (1) and continue to loop
Step 4 (the algorithm is in the end) The cycle does not stopuntil the algorithm converges and the output target imageestimation is 119891 that is over
Abstract and Applied Analysis 7
(a) Degraded image frame 1 (b) Degraded image frame 2 (c) Degraded image frame 3 (d) Degraded image frame 4
(e) Restored image by Weiner-IBD algorithm
(f) Restored image by RL-IBDalgorithm
(g) Estimated PSF by ouralgorithm
(h) Restored image by our algo-rithm
0
0
2
4
6
8
100
100
200200
300
400
150
250
times104
50
0
Y (pixel)X
(pixel)
Gra
y
(i) Energy spectrum of observed image in Figure 5(a)
0
0
0
20
40
60
80
100
50
100
150
2
4
6
8
times104
Y (pixel)
X (pixel)
Gra
y
(j) Energy spectrum of our algorithm restored image in Figure 5(h)
01
02
03
04
05
06
07
08
Inte
nsity
Our algorithm
Wiener-IBDRL-IBD
(k) A line plot taken access the horizontal axis of image restored byWeiner-IBD RL-IBD and our algorithm
Figure 5 The multiframe observed stellar images and restoration comparison of three algorithms
8 Abstract and Applied Analysis
3 Results and Analysis of AO ImageRestoration Experiments
31 The Restoration Experiment on Simulate AO Images Thealgorithm proposed in this paper is used to do the simulationexperiment on degradation spot images the evaluation ofimage restoration quality is judged by using rootmean squareerror (RMSE) and improved ΔSNR Consider
RMSE = radic1
11987311198732sum
119894119895
(119891 (119894 119895) minus 119891 (119894 119895))2
ΔSNR = 100 log10
1003817100381710038171003817119892 minus 11989110038171003817100381710038172
10038171003817100381710038171003817119891 minus 119891
10038171003817100381710038171003817
2
(22)
The evaluation of estimation accuracy is defined as fol-lows
120576estimate =
radicsum119894119895isin119878ℎ
10038161003816100381610038161003816ℎ(119894 119895) minus ℎ(119894 119895)
10038161003816100381610038161003816
2
radicsum119894119895isin119878ℎ
1003816100381610038161003816ℎ(119894 119895)10038161003816100381610038162
(23)
In the formula 1198731 and 1198732 represent the total number ofpixels on 119909-axis and 119910-axis on the image respectively (119894 119895)represents a pixel point on image and 119891(119894 119895) represents thegray value of the point (119894 119895) on ideal image
In the simulation experiments of this paper the originalimage is infrared simulation imaging with spots for multipletargets with long distance 10 frames of simulation degradedimage series are generated by image degradation softwarefrom the key optical laboratory of Chinese Academy ofSciences and then add Gaussian white noise and make thesignal-to-noise ratio SNR of image for 20 dB Set parametersthe same as adaptive optical telescope of 12m on observatoryin YunnanThemain parameters of telescope imaging systemare the atmospheric coherence length 119903
0= 13 cm diameter
of telescope 119863 = 108m comprehensive focal length ofoptical system 119897 = 2242m the wavelength in center 120582 =700 nm and the pixel size of CCD is 74 120583m Experiments onIBD algorithm based on RL iterative [16] (RL-IBD) iterativeblind restoration algorithm based on theWiener filtering [17](Wiener-IBD) and the proposed algorithm are compared
Figure 2 is the compared results of the original imagesimulated multiframe degradation images and restorationimages based on three kinds of algorithm the image size is217 times 218 Figure 2(a) is the original target image and tenframes turbulence degraded image sequences are generatedby using image degradation simulation software as shownin Figures 2(b) 2(c) and 2(d) (to save space we show onlythree frames the rest is omitted) Figure 2(e) is the Wiener-IBD restored image RMSE = 1635 Figure 2(f) is the RL-IBD restored image the number of iterations is 800 timesRMSE = 1031 Figure 2(g) is restored image used methodin this paper the variance of wavefront phase error 120579
119905(sdot) is
614 times 10minus3 ℎ0= 826 times 10
minus3 and 1205902119899= 11 times 10
minus5 that theregularization parameters in the experiments are 120572 = 125the parameter of regularization function is 120585 = 134 extrinsiciterations for 100 times inner loop of the PSF for 4 image
Table 1 Comparison of ΔSNR and 120576estimate values among threealgorithms
Algorithmname ΔSNR (dB) 120576estimate Iteration Time-consuming
Wiener-IBD 1982 362 800 20645 sRL-IBD 3046 449 800 17236 sOur algorithm 4164 510 600 10923 s
cycles for 2 namely the total times of cycling is 600 RMSE =1005 Comparing with ΔSNR and precision from Figures2(e) 2(f) and 2(g) are shown in Table 1 Comparison resultsshow that the proposed algorithm in this paper is better thanWiener-IBD algorithm and RL-IBD algorithm and needs lesstime of iterations
32 Restoration Experiments about Binary AO Images With-out the reference of ideal image using the objective qualityevaluation with no reference image the Full Width HalfMaximum (abbreviated as FWHM) [18] is used as objectiveevaluation for image restoration in this paper FWHM refersto imaging peak pixels is two times of half peak includingthe FWHM on 119909 direction and 119910 direction for the betterevaluation in astronomical observation areas computationformula is shown as follows
FWHM = radicFWHM 1199092 + FWHM 1199102 (24)
The experiment for the binary image restoration withoutreference images is carried out by the algorithm proposed inthis paper The binary images for experiments were taken bythe 12m AO telescope from Chinese Academy of Sciencesin Yunnan observatory on December 3 2006 The size forimaging CCD is 320 times 240 pixels array the main parametersof imaging system are as follows the atmospheric coherentlength 119903
0= 13 cm telescope aperture 119863 = 103 all
imaging observation band is 07sim09 center wavelength 120582 =072 the CCD pixel size is 67 120583m and optical imagingfocal length 119897 = 20 Recovery experiments based on RLIBD algorithm Wiener-IBD algorithm and the proposedalgorithm in this paper are compared the frame selectiontechnology is taken in this paper 40 frames are picked from100 frames of degradation images as blind convolution imagesto be processed
Figure 3 is the comparison results ofmultiframedegradedimages and restoration images based on three kinds ofalgorithms Figures 3(a) 3(b) 3(c) and 3(d) are for theobservation of the binary image (to save space we show onlyfour frames the rest is omitted) Figure 3(e) is the estimationof the point spread function (PSF) based on the algorithmin this paper Figure 3(f) is the restored image based on themethod proposed in this paper the variance of wavefrontphase error 120579
119905(sdot) is 604 times 10minus3 ℎ
0= 903 times 10
minus3 and1205902
119899= 132 times 10
minus5 so that the regularization parameter inthis experiment is 120572 = 115 the parameter of regularizationfunction is 120585 = 134 extrinsic iterations for 50 times the inner
Abstract and Applied Analysis 9
loop of PSF for 4 times and image loop for 2 times namelyThe total times of iteration are 300 FWHM = 617 pixels andthe restoration results are very close to the diffraction limit of12m AO telescope Figure 3(g) is the restoration result basedonRL-IBD algorithm FWHM=562 pixels Figure 3(h) is therestoration result based onWiener-IBD algorithm FWHM=476 pixels Figure 4 is the sketch of energy for the observationimage and three kinds of restored images based on threerestoration algorithms With the comparison it shows thatthe proposed algorithm in this paper is better than Wiener-IBD algorithm and RL-IBD algorithm and needs less time ofiterations
33 Restoration Experiments about Stellar AO Images Alsothe restoration experiment of stellar images was carriedout using the algorithm in this paper the stellar imagesfor experiment were taken by 12m AO telescope from theChinese Academy of Sciences in Yunnan observatory onDecember 1 2006 collection The parameters of the opticalimaging system are the same as that of binary images andthe method of parameters selection for experiments in thispaper is nearly the same Figure 5 is the comparison ofmultiframe stellar observation images and restoration resultsbased on the three algorithms Figures 5(a) 5(b) 5(c) and5(d) are the observation of stellar images (to save space weshow only 4) Figure 5(e) is for the Wiener-IBD restoredimage Figure 5(f) is the RL-IBD restored image Figure 5(g)is the estimation of the point spread function (PSF) basedon algorithm in this paper the iterative initial value ℎ
0=
956 times 10minus3 PSF iterative for six times Figure 5(h) is the
restored image based on algorithm in this paper Figure 5(i)is the energy of the observed image for stellar Figure 5(j) isthe energy of restoration image based on the algorithm in thispaper Figure 5(k) is the graph of comparison results for threerestoration algorithms
4 Conclusions
In this paper according to the characteristic of adaptiveoptics image we establish a degradation image model formultiframe exposure image series and the mathematicalmodel of the point spread function (PSF) for adaptive opticalsystem after closed loop correction is derived the PSF modelbased on phase error and changing over time is solved andwe use the power spectral density of image and constrainsupport domain method for AO image denoising processingFinally this paper applied the actual AO imaging systemparameters as prior knowledge combined with regularizationtechnique to improve the EM blind deconvolution algo-rithm Experimental results show that the established blinddeconvolution algorithm based on improved EM methodcan effectively eliminate the atmospheric turbulence andeliminate the noise pollution brought by the AO closed-loopcorrection which improves the resolution of the adaptiveoptical image The image process of AO observed imagesshows that the restoration quality is better than the samekind of IBD algorithm It also shows that the restoration ofbinary images using the algorithmof this paper has carried on
the identification of PSF and completed the AO image afterprocessing the results of actual AO image restoration haveimportant application value
Conflict of Interests
The authors declare that there is no conflict of interestsregarding the publication of this paper
Acknowledgments
This project is supported by the Scientific and TechnologicalResearch Project of Jinlin province Department of Edu-cation China (no 2013145 and no 201363) and the basisstudy project is supported by Jinlin Provincial Science ampTechnology Department China (no 201105055)
References
[1] J Wenhan ldquoAdaptive optical technologyrdquo Chinese Journal ofNature vol 28 pp 7ndash13 2005
[2] G R Ayers and J C Dainty ldquoIterative blind deconvolutionmethod and its applicationsrdquo Optics Letters vol 13 no 7 pp547ndash549 1988
[3] L Dayu H Lifa and M Quanquan ldquoA high resolution liquidcrystal adaptive optics systemrdquo Acta Photonica Sinica vol 37pp 506ndash508 2008
[4] C Rao F Shen andW Jiang ldquoAnalysis of closed-loopwavefrontresidual error of adaptive optical system using the method ofpower spectrumrdquo Acta Optica Sinica vol 20 no 1 pp 68ndash732000
[5] B Chen The theory and algorithms of adaptive optics imagerestoration [PhD thesis] Information Engineering UniversityZhengzhou China 2008
[6] T Yu R Changhui and W Kai ldquoPost-processing of adaptiveoptics images based on frame selection and multi-frame blinddeconvolutionrdquo inAdaptive Optics Systems vol 7015 of Proceed-ings of SPIE Marseille France June 2008
[7] T J Schulz ldquoNonlinear models and methods for space-objectimaging through atmospheric turbulencerdquo in Proceedings ofthe IEEE International Conference on Image Processing vol 9Lausanne Switzerland 1996
[8] L Yaning G Lei and L Huihui ldquoTotal variation basedband-limited sheralets transform for image denoisingrdquo ActaPhotonica Sinica vol 242 pp 1430ndash1435 2013
[9] J Zhao and J Yang ldquoRemote sensing image denoising algorithmbased on fusion theory using cycle spinning contourlet trans-form and total variation minimizationrdquo Acta Photonica Sinicavol 39 no 9 pp 1658ndash1665 2010
[10] C L Matson and M C Roggemann ldquoNoise reduction inadaptive-optics imagery with the use of support constraintsrdquoApplied Optics vol 34 no 5 pp 767ndash780 1995
[11] DWTyler andC LMatson ldquoReduction of nonstationary noisein telescope imagery using a support constraintrdquoOptics Expressvol 1 no 11 pp 347ndash354 1997
[12] L Zhang J YangW Su et al ldquoResearch on blind deconvolutionalgorithm of multiframe turbulence-degraded imagesrdquo Journalof Information and Computational Science vol 10 no 12 pp3625ndash3634 2013
10 Abstract and Applied Analysis
[13] A P Dempster N M Laird and D B Rubin ldquoMaximumlikelihood from incomplete data via the EM algorithmrdquo Journalof the Royal Statistical Society B vol 39 no 1 pp 1ndash38 1977
[14] L M Mugnier J Conan T Fusco and V Michau ldquoJoint maxi-mum A posteriori estimation of object and PSF for turbulence-degraded imagesrdquo in Bayesian Inference for Inverse Problemsvol 3459 of Proceedings of SPIE pp 50ndash61 San Diego CalifUSA July 1998
[15] H Hanyu Research on Image Restoration Algorithms in ImagingDetection System HuazhongUniversity of Science and Technol-ogy Wuhan China 2004
[16] F Tsumuraya ldquoIterative blind deconvolution method usingLucyrsquos algorithmrdquo Astronomy amp Astrophysics vol 282 no 2 pp699ndash708 1994
[17] S Zhang J Li Y Yang and Z Zhang ldquoBlur identification ofturbulence-degraded IR imagesrdquoOptics and Precision Engineer-ing vol 21 no 2 pp 514ndash521 2013
[18] B Chen C Cheng S GuoG Pu andZGeng ldquoUnsymmetricalmulti-limit iterative blind deconvolution algorithm for adaptiveoptics image restorationrdquoHigh Power Laser and Particle Beamsvol 23 no 2 pp 313ndash318 2011
Submit your manuscripts athttpwwwhindawicom
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
MathematicsJournal of
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Mathematical Problems in Engineering
Hindawi Publishing Corporationhttpwwwhindawicom
Differential EquationsInternational Journal of
Volume 2014
Applied MathematicsJournal of
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Probability and StatisticsHindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Journal of
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Mathematical PhysicsAdvances in
Complex AnalysisJournal of
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
OptimizationJournal of
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
CombinatoricsHindawi Publishing Corporationhttpwwwhindawicom Volume 2014
International Journal of
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Operations ResearchAdvances in
Journal of
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Function Spaces
Abstract and Applied AnalysisHindawi Publishing Corporationhttpwwwhindawicom Volume 2014
International Journal of Mathematics and Mathematical Sciences
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
The Scientific World JournalHindawi Publishing Corporation httpwwwhindawicom Volume 2014
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Algebra
Discrete Dynamics in Nature and Society
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Decision SciencesAdvances in
Discrete MathematicsJournal of
Hindawi Publishing Corporationhttpwwwhindawicom
Volume 2014 Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Stochastic AnalysisInternational Journal of
6 Abstract and Applied Analysis
0
50
100
150
200
0
50
100
150
0
50
100
150
Y (pixel)X (pixel)
Gra
y
(a) Energy spectrum of observed image in Figure 3(a)
0
50
100
150
0
50
100
150
200
250
0
50
100
150
Y (pixel) X(pixel)
Gra
y
(b) Energy spectrum of our algorithm restored image in Figure 3(f)
0
50
100
150
200
250
0
50
100
150
0
50
100
150
Y (pixel)X (pixel)
Gra
y
(c) Energy spectrum of RL-IBD algorithm restored image in Figure 3(g)
0
100
100
150
00
50
50
100
150
200
250
2040
6080
Y (pixel)
X(pixel)
Gra
y
120
(d) Energy spectrum of Wiener-IBD algorithm restored image inFigure 3(h)
Figure 4 Energy spectrum of observed image and restoration comparison of three algorithms
(3) According to the scheme and formulas ofSection 241 estimate the regularization parameters120572 and 120574(nabla119909) of target edge
Step 3 (iterate calculation 119894 = 0 1 119872119886119909 119894119905119890119903119886119905119894119900119899)
(1) The inner loop count variable of PSF h count = 0(2) The iteration process of PSF estimation 119895 = 0 1
119872119886119909 119888119900119906119899119905
(a) Complete the PSF estimation ℎ(119894) with the con-jugate gradient method using optimized for-mula (20)
(b) Consider h count++ j++(c) Judge the value of the loop variable 119895 if 119895 lt119872119886119909 119888119900119906119899119905 continue otherwise turn to (3)
(3) The inner loop counter variable of target estimation119891 119888119900119906119899119905 = 0
(4) The iteration process of target estimation 119896 =
0 1 119872119886119909 119888119900119906119899119905
(a) The conjugate gradient method was used tooptimize formula (21) to complete target imageestimation 119891(119894)
(b) consider f count++ 119896 + +(c) judge loop variable 119896 if 119896 lt 119872119886119909 119888119900119906119899119905 con-
tinue otherwise turn to (5)
(5) Judge whether the extrinsic loop is over if it is 119894 gt119898119886119909 119894119905119890119903119886119905119894119900119899 then turn to Step 4
(6) 119894 = 119894 + 1 return to (1) and continue to loop
Step 4 (the algorithm is in the end) The cycle does not stopuntil the algorithm converges and the output target imageestimation is 119891 that is over
Abstract and Applied Analysis 7
(a) Degraded image frame 1 (b) Degraded image frame 2 (c) Degraded image frame 3 (d) Degraded image frame 4
(e) Restored image by Weiner-IBD algorithm
(f) Restored image by RL-IBDalgorithm
(g) Estimated PSF by ouralgorithm
(h) Restored image by our algo-rithm
0
0
2
4
6
8
100
100
200200
300
400
150
250
times104
50
0
Y (pixel)X
(pixel)
Gra
y
(i) Energy spectrum of observed image in Figure 5(a)
0
0
0
20
40
60
80
100
50
100
150
2
4
6
8
times104
Y (pixel)
X (pixel)
Gra
y
(j) Energy spectrum of our algorithm restored image in Figure 5(h)
01
02
03
04
05
06
07
08
Inte
nsity
Our algorithm
Wiener-IBDRL-IBD
(k) A line plot taken access the horizontal axis of image restored byWeiner-IBD RL-IBD and our algorithm
Figure 5 The multiframe observed stellar images and restoration comparison of three algorithms
8 Abstract and Applied Analysis
3 Results and Analysis of AO ImageRestoration Experiments
31 The Restoration Experiment on Simulate AO Images Thealgorithm proposed in this paper is used to do the simulationexperiment on degradation spot images the evaluation ofimage restoration quality is judged by using rootmean squareerror (RMSE) and improved ΔSNR Consider
RMSE = radic1
11987311198732sum
119894119895
(119891 (119894 119895) minus 119891 (119894 119895))2
ΔSNR = 100 log10
1003817100381710038171003817119892 minus 11989110038171003817100381710038172
10038171003817100381710038171003817119891 minus 119891
10038171003817100381710038171003817
2
(22)
The evaluation of estimation accuracy is defined as fol-lows
120576estimate =
radicsum119894119895isin119878ℎ
10038161003816100381610038161003816ℎ(119894 119895) minus ℎ(119894 119895)
10038161003816100381610038161003816
2
radicsum119894119895isin119878ℎ
1003816100381610038161003816ℎ(119894 119895)10038161003816100381610038162
(23)
In the formula 1198731 and 1198732 represent the total number ofpixels on 119909-axis and 119910-axis on the image respectively (119894 119895)represents a pixel point on image and 119891(119894 119895) represents thegray value of the point (119894 119895) on ideal image
In the simulation experiments of this paper the originalimage is infrared simulation imaging with spots for multipletargets with long distance 10 frames of simulation degradedimage series are generated by image degradation softwarefrom the key optical laboratory of Chinese Academy ofSciences and then add Gaussian white noise and make thesignal-to-noise ratio SNR of image for 20 dB Set parametersthe same as adaptive optical telescope of 12m on observatoryin YunnanThemain parameters of telescope imaging systemare the atmospheric coherence length 119903
0= 13 cm diameter
of telescope 119863 = 108m comprehensive focal length ofoptical system 119897 = 2242m the wavelength in center 120582 =700 nm and the pixel size of CCD is 74 120583m Experiments onIBD algorithm based on RL iterative [16] (RL-IBD) iterativeblind restoration algorithm based on theWiener filtering [17](Wiener-IBD) and the proposed algorithm are compared
Figure 2 is the compared results of the original imagesimulated multiframe degradation images and restorationimages based on three kinds of algorithm the image size is217 times 218 Figure 2(a) is the original target image and tenframes turbulence degraded image sequences are generatedby using image degradation simulation software as shownin Figures 2(b) 2(c) and 2(d) (to save space we show onlythree frames the rest is omitted) Figure 2(e) is the Wiener-IBD restored image RMSE = 1635 Figure 2(f) is the RL-IBD restored image the number of iterations is 800 timesRMSE = 1031 Figure 2(g) is restored image used methodin this paper the variance of wavefront phase error 120579
119905(sdot) is
614 times 10minus3 ℎ0= 826 times 10
minus3 and 1205902119899= 11 times 10
minus5 that theregularization parameters in the experiments are 120572 = 125the parameter of regularization function is 120585 = 134 extrinsiciterations for 100 times inner loop of the PSF for 4 image
Table 1 Comparison of ΔSNR and 120576estimate values among threealgorithms
Algorithmname ΔSNR (dB) 120576estimate Iteration Time-consuming
Wiener-IBD 1982 362 800 20645 sRL-IBD 3046 449 800 17236 sOur algorithm 4164 510 600 10923 s
cycles for 2 namely the total times of cycling is 600 RMSE =1005 Comparing with ΔSNR and precision from Figures2(e) 2(f) and 2(g) are shown in Table 1 Comparison resultsshow that the proposed algorithm in this paper is better thanWiener-IBD algorithm and RL-IBD algorithm and needs lesstime of iterations
32 Restoration Experiments about Binary AO Images With-out the reference of ideal image using the objective qualityevaluation with no reference image the Full Width HalfMaximum (abbreviated as FWHM) [18] is used as objectiveevaluation for image restoration in this paper FWHM refersto imaging peak pixels is two times of half peak includingthe FWHM on 119909 direction and 119910 direction for the betterevaluation in astronomical observation areas computationformula is shown as follows
FWHM = radicFWHM 1199092 + FWHM 1199102 (24)
The experiment for the binary image restoration withoutreference images is carried out by the algorithm proposed inthis paper The binary images for experiments were taken bythe 12m AO telescope from Chinese Academy of Sciencesin Yunnan observatory on December 3 2006 The size forimaging CCD is 320 times 240 pixels array the main parametersof imaging system are as follows the atmospheric coherentlength 119903
0= 13 cm telescope aperture 119863 = 103 all
imaging observation band is 07sim09 center wavelength 120582 =072 the CCD pixel size is 67 120583m and optical imagingfocal length 119897 = 20 Recovery experiments based on RLIBD algorithm Wiener-IBD algorithm and the proposedalgorithm in this paper are compared the frame selectiontechnology is taken in this paper 40 frames are picked from100 frames of degradation images as blind convolution imagesto be processed
Figure 3 is the comparison results ofmultiframedegradedimages and restoration images based on three kinds ofalgorithms Figures 3(a) 3(b) 3(c) and 3(d) are for theobservation of the binary image (to save space we show onlyfour frames the rest is omitted) Figure 3(e) is the estimationof the point spread function (PSF) based on the algorithmin this paper Figure 3(f) is the restored image based on themethod proposed in this paper the variance of wavefrontphase error 120579
119905(sdot) is 604 times 10minus3 ℎ
0= 903 times 10
minus3 and1205902
119899= 132 times 10
minus5 so that the regularization parameter inthis experiment is 120572 = 115 the parameter of regularizationfunction is 120585 = 134 extrinsic iterations for 50 times the inner
Abstract and Applied Analysis 9
loop of PSF for 4 times and image loop for 2 times namelyThe total times of iteration are 300 FWHM = 617 pixels andthe restoration results are very close to the diffraction limit of12m AO telescope Figure 3(g) is the restoration result basedonRL-IBD algorithm FWHM=562 pixels Figure 3(h) is therestoration result based onWiener-IBD algorithm FWHM=476 pixels Figure 4 is the sketch of energy for the observationimage and three kinds of restored images based on threerestoration algorithms With the comparison it shows thatthe proposed algorithm in this paper is better than Wiener-IBD algorithm and RL-IBD algorithm and needs less time ofiterations
33 Restoration Experiments about Stellar AO Images Alsothe restoration experiment of stellar images was carriedout using the algorithm in this paper the stellar imagesfor experiment were taken by 12m AO telescope from theChinese Academy of Sciences in Yunnan observatory onDecember 1 2006 collection The parameters of the opticalimaging system are the same as that of binary images andthe method of parameters selection for experiments in thispaper is nearly the same Figure 5 is the comparison ofmultiframe stellar observation images and restoration resultsbased on the three algorithms Figures 5(a) 5(b) 5(c) and5(d) are the observation of stellar images (to save space weshow only 4) Figure 5(e) is for the Wiener-IBD restoredimage Figure 5(f) is the RL-IBD restored image Figure 5(g)is the estimation of the point spread function (PSF) basedon algorithm in this paper the iterative initial value ℎ
0=
956 times 10minus3 PSF iterative for six times Figure 5(h) is the
restored image based on algorithm in this paper Figure 5(i)is the energy of the observed image for stellar Figure 5(j) isthe energy of restoration image based on the algorithm in thispaper Figure 5(k) is the graph of comparison results for threerestoration algorithms
4 Conclusions
In this paper according to the characteristic of adaptiveoptics image we establish a degradation image model formultiframe exposure image series and the mathematicalmodel of the point spread function (PSF) for adaptive opticalsystem after closed loop correction is derived the PSF modelbased on phase error and changing over time is solved andwe use the power spectral density of image and constrainsupport domain method for AO image denoising processingFinally this paper applied the actual AO imaging systemparameters as prior knowledge combined with regularizationtechnique to improve the EM blind deconvolution algo-rithm Experimental results show that the established blinddeconvolution algorithm based on improved EM methodcan effectively eliminate the atmospheric turbulence andeliminate the noise pollution brought by the AO closed-loopcorrection which improves the resolution of the adaptiveoptical image The image process of AO observed imagesshows that the restoration quality is better than the samekind of IBD algorithm It also shows that the restoration ofbinary images using the algorithmof this paper has carried on
the identification of PSF and completed the AO image afterprocessing the results of actual AO image restoration haveimportant application value
Conflict of Interests
The authors declare that there is no conflict of interestsregarding the publication of this paper
Acknowledgments
This project is supported by the Scientific and TechnologicalResearch Project of Jinlin province Department of Edu-cation China (no 2013145 and no 201363) and the basisstudy project is supported by Jinlin Provincial Science ampTechnology Department China (no 201105055)
References
[1] J Wenhan ldquoAdaptive optical technologyrdquo Chinese Journal ofNature vol 28 pp 7ndash13 2005
[2] G R Ayers and J C Dainty ldquoIterative blind deconvolutionmethod and its applicationsrdquo Optics Letters vol 13 no 7 pp547ndash549 1988
[3] L Dayu H Lifa and M Quanquan ldquoA high resolution liquidcrystal adaptive optics systemrdquo Acta Photonica Sinica vol 37pp 506ndash508 2008
[4] C Rao F Shen andW Jiang ldquoAnalysis of closed-loopwavefrontresidual error of adaptive optical system using the method ofpower spectrumrdquo Acta Optica Sinica vol 20 no 1 pp 68ndash732000
[5] B Chen The theory and algorithms of adaptive optics imagerestoration [PhD thesis] Information Engineering UniversityZhengzhou China 2008
[6] T Yu R Changhui and W Kai ldquoPost-processing of adaptiveoptics images based on frame selection and multi-frame blinddeconvolutionrdquo inAdaptive Optics Systems vol 7015 of Proceed-ings of SPIE Marseille France June 2008
[7] T J Schulz ldquoNonlinear models and methods for space-objectimaging through atmospheric turbulencerdquo in Proceedings ofthe IEEE International Conference on Image Processing vol 9Lausanne Switzerland 1996
[8] L Yaning G Lei and L Huihui ldquoTotal variation basedband-limited sheralets transform for image denoisingrdquo ActaPhotonica Sinica vol 242 pp 1430ndash1435 2013
[9] J Zhao and J Yang ldquoRemote sensing image denoising algorithmbased on fusion theory using cycle spinning contourlet trans-form and total variation minimizationrdquo Acta Photonica Sinicavol 39 no 9 pp 1658ndash1665 2010
[10] C L Matson and M C Roggemann ldquoNoise reduction inadaptive-optics imagery with the use of support constraintsrdquoApplied Optics vol 34 no 5 pp 767ndash780 1995
[11] DWTyler andC LMatson ldquoReduction of nonstationary noisein telescope imagery using a support constraintrdquoOptics Expressvol 1 no 11 pp 347ndash354 1997
[12] L Zhang J YangW Su et al ldquoResearch on blind deconvolutionalgorithm of multiframe turbulence-degraded imagesrdquo Journalof Information and Computational Science vol 10 no 12 pp3625ndash3634 2013
10 Abstract and Applied Analysis
[13] A P Dempster N M Laird and D B Rubin ldquoMaximumlikelihood from incomplete data via the EM algorithmrdquo Journalof the Royal Statistical Society B vol 39 no 1 pp 1ndash38 1977
[14] L M Mugnier J Conan T Fusco and V Michau ldquoJoint maxi-mum A posteriori estimation of object and PSF for turbulence-degraded imagesrdquo in Bayesian Inference for Inverse Problemsvol 3459 of Proceedings of SPIE pp 50ndash61 San Diego CalifUSA July 1998
[15] H Hanyu Research on Image Restoration Algorithms in ImagingDetection System HuazhongUniversity of Science and Technol-ogy Wuhan China 2004
[16] F Tsumuraya ldquoIterative blind deconvolution method usingLucyrsquos algorithmrdquo Astronomy amp Astrophysics vol 282 no 2 pp699ndash708 1994
[17] S Zhang J Li Y Yang and Z Zhang ldquoBlur identification ofturbulence-degraded IR imagesrdquoOptics and Precision Engineer-ing vol 21 no 2 pp 514ndash521 2013
[18] B Chen C Cheng S GuoG Pu andZGeng ldquoUnsymmetricalmulti-limit iterative blind deconvolution algorithm for adaptiveoptics image restorationrdquoHigh Power Laser and Particle Beamsvol 23 no 2 pp 313ndash318 2011
Submit your manuscripts athttpwwwhindawicom
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
MathematicsJournal of
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Mathematical Problems in Engineering
Hindawi Publishing Corporationhttpwwwhindawicom
Differential EquationsInternational Journal of
Volume 2014
Applied MathematicsJournal of
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Probability and StatisticsHindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Journal of
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Mathematical PhysicsAdvances in
Complex AnalysisJournal of
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
OptimizationJournal of
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
CombinatoricsHindawi Publishing Corporationhttpwwwhindawicom Volume 2014
International Journal of
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Operations ResearchAdvances in
Journal of
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Function Spaces
Abstract and Applied AnalysisHindawi Publishing Corporationhttpwwwhindawicom Volume 2014
International Journal of Mathematics and Mathematical Sciences
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
The Scientific World JournalHindawi Publishing Corporation httpwwwhindawicom Volume 2014
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Algebra
Discrete Dynamics in Nature and Society
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Decision SciencesAdvances in
Discrete MathematicsJournal of
Hindawi Publishing Corporationhttpwwwhindawicom
Volume 2014 Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Stochastic AnalysisInternational Journal of
Abstract and Applied Analysis 7
(a) Degraded image frame 1 (b) Degraded image frame 2 (c) Degraded image frame 3 (d) Degraded image frame 4
(e) Restored image by Weiner-IBD algorithm
(f) Restored image by RL-IBDalgorithm
(g) Estimated PSF by ouralgorithm
(h) Restored image by our algo-rithm
0
0
2
4
6
8
100
100
200200
300
400
150
250
times104
50
0
Y (pixel)X
(pixel)
Gra
y
(i) Energy spectrum of observed image in Figure 5(a)
0
0
0
20
40
60
80
100
50
100
150
2
4
6
8
times104
Y (pixel)
X (pixel)
Gra
y
(j) Energy spectrum of our algorithm restored image in Figure 5(h)
01
02
03
04
05
06
07
08
Inte
nsity
Our algorithm
Wiener-IBDRL-IBD
(k) A line plot taken access the horizontal axis of image restored byWeiner-IBD RL-IBD and our algorithm
Figure 5 The multiframe observed stellar images and restoration comparison of three algorithms
8 Abstract and Applied Analysis
3 Results and Analysis of AO ImageRestoration Experiments
31 The Restoration Experiment on Simulate AO Images Thealgorithm proposed in this paper is used to do the simulationexperiment on degradation spot images the evaluation ofimage restoration quality is judged by using rootmean squareerror (RMSE) and improved ΔSNR Consider
RMSE = radic1
11987311198732sum
119894119895
(119891 (119894 119895) minus 119891 (119894 119895))2
ΔSNR = 100 log10
1003817100381710038171003817119892 minus 11989110038171003817100381710038172
10038171003817100381710038171003817119891 minus 119891
10038171003817100381710038171003817
2
(22)
The evaluation of estimation accuracy is defined as fol-lows
120576estimate =
radicsum119894119895isin119878ℎ
10038161003816100381610038161003816ℎ(119894 119895) minus ℎ(119894 119895)
10038161003816100381610038161003816
2
radicsum119894119895isin119878ℎ
1003816100381610038161003816ℎ(119894 119895)10038161003816100381610038162
(23)
In the formula 1198731 and 1198732 represent the total number ofpixels on 119909-axis and 119910-axis on the image respectively (119894 119895)represents a pixel point on image and 119891(119894 119895) represents thegray value of the point (119894 119895) on ideal image
In the simulation experiments of this paper the originalimage is infrared simulation imaging with spots for multipletargets with long distance 10 frames of simulation degradedimage series are generated by image degradation softwarefrom the key optical laboratory of Chinese Academy ofSciences and then add Gaussian white noise and make thesignal-to-noise ratio SNR of image for 20 dB Set parametersthe same as adaptive optical telescope of 12m on observatoryin YunnanThemain parameters of telescope imaging systemare the atmospheric coherence length 119903
0= 13 cm diameter
of telescope 119863 = 108m comprehensive focal length ofoptical system 119897 = 2242m the wavelength in center 120582 =700 nm and the pixel size of CCD is 74 120583m Experiments onIBD algorithm based on RL iterative [16] (RL-IBD) iterativeblind restoration algorithm based on theWiener filtering [17](Wiener-IBD) and the proposed algorithm are compared
Figure 2 is the compared results of the original imagesimulated multiframe degradation images and restorationimages based on three kinds of algorithm the image size is217 times 218 Figure 2(a) is the original target image and tenframes turbulence degraded image sequences are generatedby using image degradation simulation software as shownin Figures 2(b) 2(c) and 2(d) (to save space we show onlythree frames the rest is omitted) Figure 2(e) is the Wiener-IBD restored image RMSE = 1635 Figure 2(f) is the RL-IBD restored image the number of iterations is 800 timesRMSE = 1031 Figure 2(g) is restored image used methodin this paper the variance of wavefront phase error 120579
119905(sdot) is
614 times 10minus3 ℎ0= 826 times 10
minus3 and 1205902119899= 11 times 10
minus5 that theregularization parameters in the experiments are 120572 = 125the parameter of regularization function is 120585 = 134 extrinsiciterations for 100 times inner loop of the PSF for 4 image
Table 1 Comparison of ΔSNR and 120576estimate values among threealgorithms
Algorithmname ΔSNR (dB) 120576estimate Iteration Time-consuming
Wiener-IBD 1982 362 800 20645 sRL-IBD 3046 449 800 17236 sOur algorithm 4164 510 600 10923 s
cycles for 2 namely the total times of cycling is 600 RMSE =1005 Comparing with ΔSNR and precision from Figures2(e) 2(f) and 2(g) are shown in Table 1 Comparison resultsshow that the proposed algorithm in this paper is better thanWiener-IBD algorithm and RL-IBD algorithm and needs lesstime of iterations
32 Restoration Experiments about Binary AO Images With-out the reference of ideal image using the objective qualityevaluation with no reference image the Full Width HalfMaximum (abbreviated as FWHM) [18] is used as objectiveevaluation for image restoration in this paper FWHM refersto imaging peak pixels is two times of half peak includingthe FWHM on 119909 direction and 119910 direction for the betterevaluation in astronomical observation areas computationformula is shown as follows
FWHM = radicFWHM 1199092 + FWHM 1199102 (24)
The experiment for the binary image restoration withoutreference images is carried out by the algorithm proposed inthis paper The binary images for experiments were taken bythe 12m AO telescope from Chinese Academy of Sciencesin Yunnan observatory on December 3 2006 The size forimaging CCD is 320 times 240 pixels array the main parametersof imaging system are as follows the atmospheric coherentlength 119903
0= 13 cm telescope aperture 119863 = 103 all
imaging observation band is 07sim09 center wavelength 120582 =072 the CCD pixel size is 67 120583m and optical imagingfocal length 119897 = 20 Recovery experiments based on RLIBD algorithm Wiener-IBD algorithm and the proposedalgorithm in this paper are compared the frame selectiontechnology is taken in this paper 40 frames are picked from100 frames of degradation images as blind convolution imagesto be processed
Figure 3 is the comparison results ofmultiframedegradedimages and restoration images based on three kinds ofalgorithms Figures 3(a) 3(b) 3(c) and 3(d) are for theobservation of the binary image (to save space we show onlyfour frames the rest is omitted) Figure 3(e) is the estimationof the point spread function (PSF) based on the algorithmin this paper Figure 3(f) is the restored image based on themethod proposed in this paper the variance of wavefrontphase error 120579
119905(sdot) is 604 times 10minus3 ℎ
0= 903 times 10
minus3 and1205902
119899= 132 times 10
minus5 so that the regularization parameter inthis experiment is 120572 = 115 the parameter of regularizationfunction is 120585 = 134 extrinsic iterations for 50 times the inner
Abstract and Applied Analysis 9
loop of PSF for 4 times and image loop for 2 times namelyThe total times of iteration are 300 FWHM = 617 pixels andthe restoration results are very close to the diffraction limit of12m AO telescope Figure 3(g) is the restoration result basedonRL-IBD algorithm FWHM=562 pixels Figure 3(h) is therestoration result based onWiener-IBD algorithm FWHM=476 pixels Figure 4 is the sketch of energy for the observationimage and three kinds of restored images based on threerestoration algorithms With the comparison it shows thatthe proposed algorithm in this paper is better than Wiener-IBD algorithm and RL-IBD algorithm and needs less time ofiterations
33 Restoration Experiments about Stellar AO Images Alsothe restoration experiment of stellar images was carriedout using the algorithm in this paper the stellar imagesfor experiment were taken by 12m AO telescope from theChinese Academy of Sciences in Yunnan observatory onDecember 1 2006 collection The parameters of the opticalimaging system are the same as that of binary images andthe method of parameters selection for experiments in thispaper is nearly the same Figure 5 is the comparison ofmultiframe stellar observation images and restoration resultsbased on the three algorithms Figures 5(a) 5(b) 5(c) and5(d) are the observation of stellar images (to save space weshow only 4) Figure 5(e) is for the Wiener-IBD restoredimage Figure 5(f) is the RL-IBD restored image Figure 5(g)is the estimation of the point spread function (PSF) basedon algorithm in this paper the iterative initial value ℎ
0=
956 times 10minus3 PSF iterative for six times Figure 5(h) is the
restored image based on algorithm in this paper Figure 5(i)is the energy of the observed image for stellar Figure 5(j) isthe energy of restoration image based on the algorithm in thispaper Figure 5(k) is the graph of comparison results for threerestoration algorithms
4 Conclusions
In this paper according to the characteristic of adaptiveoptics image we establish a degradation image model formultiframe exposure image series and the mathematicalmodel of the point spread function (PSF) for adaptive opticalsystem after closed loop correction is derived the PSF modelbased on phase error and changing over time is solved andwe use the power spectral density of image and constrainsupport domain method for AO image denoising processingFinally this paper applied the actual AO imaging systemparameters as prior knowledge combined with regularizationtechnique to improve the EM blind deconvolution algo-rithm Experimental results show that the established blinddeconvolution algorithm based on improved EM methodcan effectively eliminate the atmospheric turbulence andeliminate the noise pollution brought by the AO closed-loopcorrection which improves the resolution of the adaptiveoptical image The image process of AO observed imagesshows that the restoration quality is better than the samekind of IBD algorithm It also shows that the restoration ofbinary images using the algorithmof this paper has carried on
the identification of PSF and completed the AO image afterprocessing the results of actual AO image restoration haveimportant application value
Conflict of Interests
The authors declare that there is no conflict of interestsregarding the publication of this paper
Acknowledgments
This project is supported by the Scientific and TechnologicalResearch Project of Jinlin province Department of Edu-cation China (no 2013145 and no 201363) and the basisstudy project is supported by Jinlin Provincial Science ampTechnology Department China (no 201105055)
References
[1] J Wenhan ldquoAdaptive optical technologyrdquo Chinese Journal ofNature vol 28 pp 7ndash13 2005
[2] G R Ayers and J C Dainty ldquoIterative blind deconvolutionmethod and its applicationsrdquo Optics Letters vol 13 no 7 pp547ndash549 1988
[3] L Dayu H Lifa and M Quanquan ldquoA high resolution liquidcrystal adaptive optics systemrdquo Acta Photonica Sinica vol 37pp 506ndash508 2008
[4] C Rao F Shen andW Jiang ldquoAnalysis of closed-loopwavefrontresidual error of adaptive optical system using the method ofpower spectrumrdquo Acta Optica Sinica vol 20 no 1 pp 68ndash732000
[5] B Chen The theory and algorithms of adaptive optics imagerestoration [PhD thesis] Information Engineering UniversityZhengzhou China 2008
[6] T Yu R Changhui and W Kai ldquoPost-processing of adaptiveoptics images based on frame selection and multi-frame blinddeconvolutionrdquo inAdaptive Optics Systems vol 7015 of Proceed-ings of SPIE Marseille France June 2008
[7] T J Schulz ldquoNonlinear models and methods for space-objectimaging through atmospheric turbulencerdquo in Proceedings ofthe IEEE International Conference on Image Processing vol 9Lausanne Switzerland 1996
[8] L Yaning G Lei and L Huihui ldquoTotal variation basedband-limited sheralets transform for image denoisingrdquo ActaPhotonica Sinica vol 242 pp 1430ndash1435 2013
[9] J Zhao and J Yang ldquoRemote sensing image denoising algorithmbased on fusion theory using cycle spinning contourlet trans-form and total variation minimizationrdquo Acta Photonica Sinicavol 39 no 9 pp 1658ndash1665 2010
[10] C L Matson and M C Roggemann ldquoNoise reduction inadaptive-optics imagery with the use of support constraintsrdquoApplied Optics vol 34 no 5 pp 767ndash780 1995
[11] DWTyler andC LMatson ldquoReduction of nonstationary noisein telescope imagery using a support constraintrdquoOptics Expressvol 1 no 11 pp 347ndash354 1997
[12] L Zhang J YangW Su et al ldquoResearch on blind deconvolutionalgorithm of multiframe turbulence-degraded imagesrdquo Journalof Information and Computational Science vol 10 no 12 pp3625ndash3634 2013
10 Abstract and Applied Analysis
[13] A P Dempster N M Laird and D B Rubin ldquoMaximumlikelihood from incomplete data via the EM algorithmrdquo Journalof the Royal Statistical Society B vol 39 no 1 pp 1ndash38 1977
[14] L M Mugnier J Conan T Fusco and V Michau ldquoJoint maxi-mum A posteriori estimation of object and PSF for turbulence-degraded imagesrdquo in Bayesian Inference for Inverse Problemsvol 3459 of Proceedings of SPIE pp 50ndash61 San Diego CalifUSA July 1998
[15] H Hanyu Research on Image Restoration Algorithms in ImagingDetection System HuazhongUniversity of Science and Technol-ogy Wuhan China 2004
[16] F Tsumuraya ldquoIterative blind deconvolution method usingLucyrsquos algorithmrdquo Astronomy amp Astrophysics vol 282 no 2 pp699ndash708 1994
[17] S Zhang J Li Y Yang and Z Zhang ldquoBlur identification ofturbulence-degraded IR imagesrdquoOptics and Precision Engineer-ing vol 21 no 2 pp 514ndash521 2013
[18] B Chen C Cheng S GuoG Pu andZGeng ldquoUnsymmetricalmulti-limit iterative blind deconvolution algorithm for adaptiveoptics image restorationrdquoHigh Power Laser and Particle Beamsvol 23 no 2 pp 313ndash318 2011
Submit your manuscripts athttpwwwhindawicom
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
MathematicsJournal of
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Mathematical Problems in Engineering
Hindawi Publishing Corporationhttpwwwhindawicom
Differential EquationsInternational Journal of
Volume 2014
Applied MathematicsJournal of
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Probability and StatisticsHindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Journal of
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Mathematical PhysicsAdvances in
Complex AnalysisJournal of
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
OptimizationJournal of
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
CombinatoricsHindawi Publishing Corporationhttpwwwhindawicom Volume 2014
International Journal of
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Operations ResearchAdvances in
Journal of
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Function Spaces
Abstract and Applied AnalysisHindawi Publishing Corporationhttpwwwhindawicom Volume 2014
International Journal of Mathematics and Mathematical Sciences
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
The Scientific World JournalHindawi Publishing Corporation httpwwwhindawicom Volume 2014
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Algebra
Discrete Dynamics in Nature and Society
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Decision SciencesAdvances in
Discrete MathematicsJournal of
Hindawi Publishing Corporationhttpwwwhindawicom
Volume 2014 Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Stochastic AnalysisInternational Journal of
8 Abstract and Applied Analysis
3 Results and Analysis of AO ImageRestoration Experiments
31 The Restoration Experiment on Simulate AO Images Thealgorithm proposed in this paper is used to do the simulationexperiment on degradation spot images the evaluation ofimage restoration quality is judged by using rootmean squareerror (RMSE) and improved ΔSNR Consider
RMSE = radic1
11987311198732sum
119894119895
(119891 (119894 119895) minus 119891 (119894 119895))2
ΔSNR = 100 log10
1003817100381710038171003817119892 minus 11989110038171003817100381710038172
10038171003817100381710038171003817119891 minus 119891
10038171003817100381710038171003817
2
(22)
The evaluation of estimation accuracy is defined as fol-lows
120576estimate =
radicsum119894119895isin119878ℎ
10038161003816100381610038161003816ℎ(119894 119895) minus ℎ(119894 119895)
10038161003816100381610038161003816
2
radicsum119894119895isin119878ℎ
1003816100381610038161003816ℎ(119894 119895)10038161003816100381610038162
(23)
In the formula 1198731 and 1198732 represent the total number ofpixels on 119909-axis and 119910-axis on the image respectively (119894 119895)represents a pixel point on image and 119891(119894 119895) represents thegray value of the point (119894 119895) on ideal image
In the simulation experiments of this paper the originalimage is infrared simulation imaging with spots for multipletargets with long distance 10 frames of simulation degradedimage series are generated by image degradation softwarefrom the key optical laboratory of Chinese Academy ofSciences and then add Gaussian white noise and make thesignal-to-noise ratio SNR of image for 20 dB Set parametersthe same as adaptive optical telescope of 12m on observatoryin YunnanThemain parameters of telescope imaging systemare the atmospheric coherence length 119903
0= 13 cm diameter
of telescope 119863 = 108m comprehensive focal length ofoptical system 119897 = 2242m the wavelength in center 120582 =700 nm and the pixel size of CCD is 74 120583m Experiments onIBD algorithm based on RL iterative [16] (RL-IBD) iterativeblind restoration algorithm based on theWiener filtering [17](Wiener-IBD) and the proposed algorithm are compared
Figure 2 is the compared results of the original imagesimulated multiframe degradation images and restorationimages based on three kinds of algorithm the image size is217 times 218 Figure 2(a) is the original target image and tenframes turbulence degraded image sequences are generatedby using image degradation simulation software as shownin Figures 2(b) 2(c) and 2(d) (to save space we show onlythree frames the rest is omitted) Figure 2(e) is the Wiener-IBD restored image RMSE = 1635 Figure 2(f) is the RL-IBD restored image the number of iterations is 800 timesRMSE = 1031 Figure 2(g) is restored image used methodin this paper the variance of wavefront phase error 120579
119905(sdot) is
614 times 10minus3 ℎ0= 826 times 10
minus3 and 1205902119899= 11 times 10
minus5 that theregularization parameters in the experiments are 120572 = 125the parameter of regularization function is 120585 = 134 extrinsiciterations for 100 times inner loop of the PSF for 4 image
Table 1 Comparison of ΔSNR and 120576estimate values among threealgorithms
Algorithmname ΔSNR (dB) 120576estimate Iteration Time-consuming
Wiener-IBD 1982 362 800 20645 sRL-IBD 3046 449 800 17236 sOur algorithm 4164 510 600 10923 s
cycles for 2 namely the total times of cycling is 600 RMSE =1005 Comparing with ΔSNR and precision from Figures2(e) 2(f) and 2(g) are shown in Table 1 Comparison resultsshow that the proposed algorithm in this paper is better thanWiener-IBD algorithm and RL-IBD algorithm and needs lesstime of iterations
32 Restoration Experiments about Binary AO Images With-out the reference of ideal image using the objective qualityevaluation with no reference image the Full Width HalfMaximum (abbreviated as FWHM) [18] is used as objectiveevaluation for image restoration in this paper FWHM refersto imaging peak pixels is two times of half peak includingthe FWHM on 119909 direction and 119910 direction for the betterevaluation in astronomical observation areas computationformula is shown as follows
FWHM = radicFWHM 1199092 + FWHM 1199102 (24)
The experiment for the binary image restoration withoutreference images is carried out by the algorithm proposed inthis paper The binary images for experiments were taken bythe 12m AO telescope from Chinese Academy of Sciencesin Yunnan observatory on December 3 2006 The size forimaging CCD is 320 times 240 pixels array the main parametersof imaging system are as follows the atmospheric coherentlength 119903
0= 13 cm telescope aperture 119863 = 103 all
imaging observation band is 07sim09 center wavelength 120582 =072 the CCD pixel size is 67 120583m and optical imagingfocal length 119897 = 20 Recovery experiments based on RLIBD algorithm Wiener-IBD algorithm and the proposedalgorithm in this paper are compared the frame selectiontechnology is taken in this paper 40 frames are picked from100 frames of degradation images as blind convolution imagesto be processed
Figure 3 is the comparison results ofmultiframedegradedimages and restoration images based on three kinds ofalgorithms Figures 3(a) 3(b) 3(c) and 3(d) are for theobservation of the binary image (to save space we show onlyfour frames the rest is omitted) Figure 3(e) is the estimationof the point spread function (PSF) based on the algorithmin this paper Figure 3(f) is the restored image based on themethod proposed in this paper the variance of wavefrontphase error 120579
119905(sdot) is 604 times 10minus3 ℎ
0= 903 times 10
minus3 and1205902
119899= 132 times 10
minus5 so that the regularization parameter inthis experiment is 120572 = 115 the parameter of regularizationfunction is 120585 = 134 extrinsic iterations for 50 times the inner
Abstract and Applied Analysis 9
loop of PSF for 4 times and image loop for 2 times namelyThe total times of iteration are 300 FWHM = 617 pixels andthe restoration results are very close to the diffraction limit of12m AO telescope Figure 3(g) is the restoration result basedonRL-IBD algorithm FWHM=562 pixels Figure 3(h) is therestoration result based onWiener-IBD algorithm FWHM=476 pixels Figure 4 is the sketch of energy for the observationimage and three kinds of restored images based on threerestoration algorithms With the comparison it shows thatthe proposed algorithm in this paper is better than Wiener-IBD algorithm and RL-IBD algorithm and needs less time ofiterations
33 Restoration Experiments about Stellar AO Images Alsothe restoration experiment of stellar images was carriedout using the algorithm in this paper the stellar imagesfor experiment were taken by 12m AO telescope from theChinese Academy of Sciences in Yunnan observatory onDecember 1 2006 collection The parameters of the opticalimaging system are the same as that of binary images andthe method of parameters selection for experiments in thispaper is nearly the same Figure 5 is the comparison ofmultiframe stellar observation images and restoration resultsbased on the three algorithms Figures 5(a) 5(b) 5(c) and5(d) are the observation of stellar images (to save space weshow only 4) Figure 5(e) is for the Wiener-IBD restoredimage Figure 5(f) is the RL-IBD restored image Figure 5(g)is the estimation of the point spread function (PSF) basedon algorithm in this paper the iterative initial value ℎ
0=
956 times 10minus3 PSF iterative for six times Figure 5(h) is the
restored image based on algorithm in this paper Figure 5(i)is the energy of the observed image for stellar Figure 5(j) isthe energy of restoration image based on the algorithm in thispaper Figure 5(k) is the graph of comparison results for threerestoration algorithms
4 Conclusions
In this paper according to the characteristic of adaptiveoptics image we establish a degradation image model formultiframe exposure image series and the mathematicalmodel of the point spread function (PSF) for adaptive opticalsystem after closed loop correction is derived the PSF modelbased on phase error and changing over time is solved andwe use the power spectral density of image and constrainsupport domain method for AO image denoising processingFinally this paper applied the actual AO imaging systemparameters as prior knowledge combined with regularizationtechnique to improve the EM blind deconvolution algo-rithm Experimental results show that the established blinddeconvolution algorithm based on improved EM methodcan effectively eliminate the atmospheric turbulence andeliminate the noise pollution brought by the AO closed-loopcorrection which improves the resolution of the adaptiveoptical image The image process of AO observed imagesshows that the restoration quality is better than the samekind of IBD algorithm It also shows that the restoration ofbinary images using the algorithmof this paper has carried on
the identification of PSF and completed the AO image afterprocessing the results of actual AO image restoration haveimportant application value
Conflict of Interests
The authors declare that there is no conflict of interestsregarding the publication of this paper
Acknowledgments
This project is supported by the Scientific and TechnologicalResearch Project of Jinlin province Department of Edu-cation China (no 2013145 and no 201363) and the basisstudy project is supported by Jinlin Provincial Science ampTechnology Department China (no 201105055)
References
[1] J Wenhan ldquoAdaptive optical technologyrdquo Chinese Journal ofNature vol 28 pp 7ndash13 2005
[2] G R Ayers and J C Dainty ldquoIterative blind deconvolutionmethod and its applicationsrdquo Optics Letters vol 13 no 7 pp547ndash549 1988
[3] L Dayu H Lifa and M Quanquan ldquoA high resolution liquidcrystal adaptive optics systemrdquo Acta Photonica Sinica vol 37pp 506ndash508 2008
[4] C Rao F Shen andW Jiang ldquoAnalysis of closed-loopwavefrontresidual error of adaptive optical system using the method ofpower spectrumrdquo Acta Optica Sinica vol 20 no 1 pp 68ndash732000
[5] B Chen The theory and algorithms of adaptive optics imagerestoration [PhD thesis] Information Engineering UniversityZhengzhou China 2008
[6] T Yu R Changhui and W Kai ldquoPost-processing of adaptiveoptics images based on frame selection and multi-frame blinddeconvolutionrdquo inAdaptive Optics Systems vol 7015 of Proceed-ings of SPIE Marseille France June 2008
[7] T J Schulz ldquoNonlinear models and methods for space-objectimaging through atmospheric turbulencerdquo in Proceedings ofthe IEEE International Conference on Image Processing vol 9Lausanne Switzerland 1996
[8] L Yaning G Lei and L Huihui ldquoTotal variation basedband-limited sheralets transform for image denoisingrdquo ActaPhotonica Sinica vol 242 pp 1430ndash1435 2013
[9] J Zhao and J Yang ldquoRemote sensing image denoising algorithmbased on fusion theory using cycle spinning contourlet trans-form and total variation minimizationrdquo Acta Photonica Sinicavol 39 no 9 pp 1658ndash1665 2010
[10] C L Matson and M C Roggemann ldquoNoise reduction inadaptive-optics imagery with the use of support constraintsrdquoApplied Optics vol 34 no 5 pp 767ndash780 1995
[11] DWTyler andC LMatson ldquoReduction of nonstationary noisein telescope imagery using a support constraintrdquoOptics Expressvol 1 no 11 pp 347ndash354 1997
[12] L Zhang J YangW Su et al ldquoResearch on blind deconvolutionalgorithm of multiframe turbulence-degraded imagesrdquo Journalof Information and Computational Science vol 10 no 12 pp3625ndash3634 2013
10 Abstract and Applied Analysis
[13] A P Dempster N M Laird and D B Rubin ldquoMaximumlikelihood from incomplete data via the EM algorithmrdquo Journalof the Royal Statistical Society B vol 39 no 1 pp 1ndash38 1977
[14] L M Mugnier J Conan T Fusco and V Michau ldquoJoint maxi-mum A posteriori estimation of object and PSF for turbulence-degraded imagesrdquo in Bayesian Inference for Inverse Problemsvol 3459 of Proceedings of SPIE pp 50ndash61 San Diego CalifUSA July 1998
[15] H Hanyu Research on Image Restoration Algorithms in ImagingDetection System HuazhongUniversity of Science and Technol-ogy Wuhan China 2004
[16] F Tsumuraya ldquoIterative blind deconvolution method usingLucyrsquos algorithmrdquo Astronomy amp Astrophysics vol 282 no 2 pp699ndash708 1994
[17] S Zhang J Li Y Yang and Z Zhang ldquoBlur identification ofturbulence-degraded IR imagesrdquoOptics and Precision Engineer-ing vol 21 no 2 pp 514ndash521 2013
[18] B Chen C Cheng S GuoG Pu andZGeng ldquoUnsymmetricalmulti-limit iterative blind deconvolution algorithm for adaptiveoptics image restorationrdquoHigh Power Laser and Particle Beamsvol 23 no 2 pp 313ndash318 2011
Submit your manuscripts athttpwwwhindawicom
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
MathematicsJournal of
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Mathematical Problems in Engineering
Hindawi Publishing Corporationhttpwwwhindawicom
Differential EquationsInternational Journal of
Volume 2014
Applied MathematicsJournal of
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Probability and StatisticsHindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Journal of
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Mathematical PhysicsAdvances in
Complex AnalysisJournal of
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
OptimizationJournal of
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
CombinatoricsHindawi Publishing Corporationhttpwwwhindawicom Volume 2014
International Journal of
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Operations ResearchAdvances in
Journal of
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Function Spaces
Abstract and Applied AnalysisHindawi Publishing Corporationhttpwwwhindawicom Volume 2014
International Journal of Mathematics and Mathematical Sciences
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
The Scientific World JournalHindawi Publishing Corporation httpwwwhindawicom Volume 2014
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Algebra
Discrete Dynamics in Nature and Society
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Decision SciencesAdvances in
Discrete MathematicsJournal of
Hindawi Publishing Corporationhttpwwwhindawicom
Volume 2014 Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Stochastic AnalysisInternational Journal of
Abstract and Applied Analysis 9
loop of PSF for 4 times and image loop for 2 times namelyThe total times of iteration are 300 FWHM = 617 pixels andthe restoration results are very close to the diffraction limit of12m AO telescope Figure 3(g) is the restoration result basedonRL-IBD algorithm FWHM=562 pixels Figure 3(h) is therestoration result based onWiener-IBD algorithm FWHM=476 pixels Figure 4 is the sketch of energy for the observationimage and three kinds of restored images based on threerestoration algorithms With the comparison it shows thatthe proposed algorithm in this paper is better than Wiener-IBD algorithm and RL-IBD algorithm and needs less time ofiterations
33 Restoration Experiments about Stellar AO Images Alsothe restoration experiment of stellar images was carriedout using the algorithm in this paper the stellar imagesfor experiment were taken by 12m AO telescope from theChinese Academy of Sciences in Yunnan observatory onDecember 1 2006 collection The parameters of the opticalimaging system are the same as that of binary images andthe method of parameters selection for experiments in thispaper is nearly the same Figure 5 is the comparison ofmultiframe stellar observation images and restoration resultsbased on the three algorithms Figures 5(a) 5(b) 5(c) and5(d) are the observation of stellar images (to save space weshow only 4) Figure 5(e) is for the Wiener-IBD restoredimage Figure 5(f) is the RL-IBD restored image Figure 5(g)is the estimation of the point spread function (PSF) basedon algorithm in this paper the iterative initial value ℎ
0=
956 times 10minus3 PSF iterative for six times Figure 5(h) is the
restored image based on algorithm in this paper Figure 5(i)is the energy of the observed image for stellar Figure 5(j) isthe energy of restoration image based on the algorithm in thispaper Figure 5(k) is the graph of comparison results for threerestoration algorithms
4 Conclusions
In this paper according to the characteristic of adaptiveoptics image we establish a degradation image model formultiframe exposure image series and the mathematicalmodel of the point spread function (PSF) for adaptive opticalsystem after closed loop correction is derived the PSF modelbased on phase error and changing over time is solved andwe use the power spectral density of image and constrainsupport domain method for AO image denoising processingFinally this paper applied the actual AO imaging systemparameters as prior knowledge combined with regularizationtechnique to improve the EM blind deconvolution algo-rithm Experimental results show that the established blinddeconvolution algorithm based on improved EM methodcan effectively eliminate the atmospheric turbulence andeliminate the noise pollution brought by the AO closed-loopcorrection which improves the resolution of the adaptiveoptical image The image process of AO observed imagesshows that the restoration quality is better than the samekind of IBD algorithm It also shows that the restoration ofbinary images using the algorithmof this paper has carried on
the identification of PSF and completed the AO image afterprocessing the results of actual AO image restoration haveimportant application value
Conflict of Interests
The authors declare that there is no conflict of interestsregarding the publication of this paper
Acknowledgments
This project is supported by the Scientific and TechnologicalResearch Project of Jinlin province Department of Edu-cation China (no 2013145 and no 201363) and the basisstudy project is supported by Jinlin Provincial Science ampTechnology Department China (no 201105055)
References
[1] J Wenhan ldquoAdaptive optical technologyrdquo Chinese Journal ofNature vol 28 pp 7ndash13 2005
[2] G R Ayers and J C Dainty ldquoIterative blind deconvolutionmethod and its applicationsrdquo Optics Letters vol 13 no 7 pp547ndash549 1988
[3] L Dayu H Lifa and M Quanquan ldquoA high resolution liquidcrystal adaptive optics systemrdquo Acta Photonica Sinica vol 37pp 506ndash508 2008
[4] C Rao F Shen andW Jiang ldquoAnalysis of closed-loopwavefrontresidual error of adaptive optical system using the method ofpower spectrumrdquo Acta Optica Sinica vol 20 no 1 pp 68ndash732000
[5] B Chen The theory and algorithms of adaptive optics imagerestoration [PhD thesis] Information Engineering UniversityZhengzhou China 2008
[6] T Yu R Changhui and W Kai ldquoPost-processing of adaptiveoptics images based on frame selection and multi-frame blinddeconvolutionrdquo inAdaptive Optics Systems vol 7015 of Proceed-ings of SPIE Marseille France June 2008
[7] T J Schulz ldquoNonlinear models and methods for space-objectimaging through atmospheric turbulencerdquo in Proceedings ofthe IEEE International Conference on Image Processing vol 9Lausanne Switzerland 1996
[8] L Yaning G Lei and L Huihui ldquoTotal variation basedband-limited sheralets transform for image denoisingrdquo ActaPhotonica Sinica vol 242 pp 1430ndash1435 2013
[9] J Zhao and J Yang ldquoRemote sensing image denoising algorithmbased on fusion theory using cycle spinning contourlet trans-form and total variation minimizationrdquo Acta Photonica Sinicavol 39 no 9 pp 1658ndash1665 2010
[10] C L Matson and M C Roggemann ldquoNoise reduction inadaptive-optics imagery with the use of support constraintsrdquoApplied Optics vol 34 no 5 pp 767ndash780 1995
[11] DWTyler andC LMatson ldquoReduction of nonstationary noisein telescope imagery using a support constraintrdquoOptics Expressvol 1 no 11 pp 347ndash354 1997
[12] L Zhang J YangW Su et al ldquoResearch on blind deconvolutionalgorithm of multiframe turbulence-degraded imagesrdquo Journalof Information and Computational Science vol 10 no 12 pp3625ndash3634 2013
10 Abstract and Applied Analysis
[13] A P Dempster N M Laird and D B Rubin ldquoMaximumlikelihood from incomplete data via the EM algorithmrdquo Journalof the Royal Statistical Society B vol 39 no 1 pp 1ndash38 1977
[14] L M Mugnier J Conan T Fusco and V Michau ldquoJoint maxi-mum A posteriori estimation of object and PSF for turbulence-degraded imagesrdquo in Bayesian Inference for Inverse Problemsvol 3459 of Proceedings of SPIE pp 50ndash61 San Diego CalifUSA July 1998
[15] H Hanyu Research on Image Restoration Algorithms in ImagingDetection System HuazhongUniversity of Science and Technol-ogy Wuhan China 2004
[16] F Tsumuraya ldquoIterative blind deconvolution method usingLucyrsquos algorithmrdquo Astronomy amp Astrophysics vol 282 no 2 pp699ndash708 1994
[17] S Zhang J Li Y Yang and Z Zhang ldquoBlur identification ofturbulence-degraded IR imagesrdquoOptics and Precision Engineer-ing vol 21 no 2 pp 514ndash521 2013
[18] B Chen C Cheng S GuoG Pu andZGeng ldquoUnsymmetricalmulti-limit iterative blind deconvolution algorithm for adaptiveoptics image restorationrdquoHigh Power Laser and Particle Beamsvol 23 no 2 pp 313ndash318 2011
Submit your manuscripts athttpwwwhindawicom
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
MathematicsJournal of
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Mathematical Problems in Engineering
Hindawi Publishing Corporationhttpwwwhindawicom
Differential EquationsInternational Journal of
Volume 2014
Applied MathematicsJournal of
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Probability and StatisticsHindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Journal of
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Mathematical PhysicsAdvances in
Complex AnalysisJournal of
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
OptimizationJournal of
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
CombinatoricsHindawi Publishing Corporationhttpwwwhindawicom Volume 2014
International Journal of
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Operations ResearchAdvances in
Journal of
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Function Spaces
Abstract and Applied AnalysisHindawi Publishing Corporationhttpwwwhindawicom Volume 2014
International Journal of Mathematics and Mathematical Sciences
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
The Scientific World JournalHindawi Publishing Corporation httpwwwhindawicom Volume 2014
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Algebra
Discrete Dynamics in Nature and Society
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Decision SciencesAdvances in
Discrete MathematicsJournal of
Hindawi Publishing Corporationhttpwwwhindawicom
Volume 2014 Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Stochastic AnalysisInternational Journal of
10 Abstract and Applied Analysis
[13] A P Dempster N M Laird and D B Rubin ldquoMaximumlikelihood from incomplete data via the EM algorithmrdquo Journalof the Royal Statistical Society B vol 39 no 1 pp 1ndash38 1977
[14] L M Mugnier J Conan T Fusco and V Michau ldquoJoint maxi-mum A posteriori estimation of object and PSF for turbulence-degraded imagesrdquo in Bayesian Inference for Inverse Problemsvol 3459 of Proceedings of SPIE pp 50ndash61 San Diego CalifUSA July 1998
[15] H Hanyu Research on Image Restoration Algorithms in ImagingDetection System HuazhongUniversity of Science and Technol-ogy Wuhan China 2004
[16] F Tsumuraya ldquoIterative blind deconvolution method usingLucyrsquos algorithmrdquo Astronomy amp Astrophysics vol 282 no 2 pp699ndash708 1994
[17] S Zhang J Li Y Yang and Z Zhang ldquoBlur identification ofturbulence-degraded IR imagesrdquoOptics and Precision Engineer-ing vol 21 no 2 pp 514ndash521 2013
[18] B Chen C Cheng S GuoG Pu andZGeng ldquoUnsymmetricalmulti-limit iterative blind deconvolution algorithm for adaptiveoptics image restorationrdquoHigh Power Laser and Particle Beamsvol 23 no 2 pp 313ndash318 2011
Submit your manuscripts athttpwwwhindawicom
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
MathematicsJournal of
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Mathematical Problems in Engineering
Hindawi Publishing Corporationhttpwwwhindawicom
Differential EquationsInternational Journal of
Volume 2014
Applied MathematicsJournal of
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Probability and StatisticsHindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Journal of
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Mathematical PhysicsAdvances in
Complex AnalysisJournal of
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
OptimizationJournal of
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
CombinatoricsHindawi Publishing Corporationhttpwwwhindawicom Volume 2014
International Journal of
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Operations ResearchAdvances in
Journal of
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Function Spaces
Abstract and Applied AnalysisHindawi Publishing Corporationhttpwwwhindawicom Volume 2014
International Journal of Mathematics and Mathematical Sciences
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
The Scientific World JournalHindawi Publishing Corporation httpwwwhindawicom Volume 2014
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Algebra
Discrete Dynamics in Nature and Society
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Decision SciencesAdvances in
Discrete MathematicsJournal of
Hindawi Publishing Corporationhttpwwwhindawicom
Volume 2014 Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Stochastic AnalysisInternational Journal of
Submit your manuscripts athttpwwwhindawicom
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
MathematicsJournal of
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Mathematical Problems in Engineering
Hindawi Publishing Corporationhttpwwwhindawicom
Differential EquationsInternational Journal of
Volume 2014
Applied MathematicsJournal of
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Probability and StatisticsHindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Journal of
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Mathematical PhysicsAdvances in
Complex AnalysisJournal of
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
OptimizationJournal of
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
CombinatoricsHindawi Publishing Corporationhttpwwwhindawicom Volume 2014
International Journal of
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Operations ResearchAdvances in
Journal of
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Function Spaces
Abstract and Applied AnalysisHindawi Publishing Corporationhttpwwwhindawicom Volume 2014
International Journal of Mathematics and Mathematical Sciences
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
The Scientific World JournalHindawi Publishing Corporation httpwwwhindawicom Volume 2014
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Algebra
Discrete Dynamics in Nature and Society
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Decision SciencesAdvances in
Discrete MathematicsJournal of
Hindawi Publishing Corporationhttpwwwhindawicom
Volume 2014 Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Stochastic AnalysisInternational Journal of