CN102768756B - Universal recovery method for target detection multi-spectrum images - Google Patents
Universal recovery method for target detection multi-spectrum images Download PDFInfo
- Publication number
- CN102768756B CN102768756B CN201210197707.7A CN201210197707A CN102768756B CN 102768756 B CN102768756 B CN 102768756B CN 201210197707 A CN201210197707 A CN 201210197707A CN 102768756 B CN102768756 B CN 102768756B
- Authority
- CN
- China
- Prior art keywords
- image
- spread function
- point spread
- transition
- equation
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Expired - Fee Related
Links
Landscapes
- Image Processing (AREA)
Abstract
The invention relates to a universal recovery method for target detection multi-spectrum images. The method comprises the following steps that a degradation image outline is detected, a transition region is found out through the image outline, and a clear image of the transition region is predicted; K points are selected on the multi-spectrum degradation image transition region, K corresponding regions are ground on the clear image of the predicted transition region, and a matrix equation system is built; nonnegativity restriction and anisotropic space correlation restriction items are added, and an iteration relation equation of x is obtained through solving; 4) the value of x of the point spread function matrix is M*M; 5) the value hat(x)<(n)> of x when n is changed into (n+1) is obtained through the solving on the iteration equation of x; 6) if ||hat(x)<(n)>-hat(x)<(n-1)>||<epsilon, the calculation is stopped, and the point spread function x is obtained; and 7) the point spread function obtained through solving is recovered by a maximum likelihood estimation non-blind deconvolution method, and a clear image is obtained. The method provided by the invention has the beneficial effects that the image degradation mode does not need to be known, the recovery and defuzzification processing on any spectrum images is realized, the time consumption is low, and the recovery effect is good.
Description
Technical field
The present invention relates to a kind of general restored method of target detection multispectral image.
Background technology
In target detection process, owing to being subject to the impact of the many factors such as turbulent flow, motion, scattering, target detection multispectral image includes the multiple degradation modes such as motion, turbulent flow, scattering, various degradation modes are also that not identical, different degradation modes has different degradation models to the degradation effects degree of multispectral detection image
[1], from multispectral detection image, be to be therefore difficult to distinguish concrete degradation modes and various degeneration factor and influence degree thereof.Current restored method is to find various concrete degradation modes and model, then carries out deconvolution restored image
[2-4], but these methods can not be applicable to target detection, because be difficult to distinguish image degradation pattern from multispectral detection image.In addition, current existing image recovery method utilizes a monoblock region of whole degraded image or degraded image to restore, and target area and background area smooth in degraded image have all participated in calculating
[2-4], the degradation information comprising due to smooth target area and background area is not remarkable, and the information of redundancy can be brought uncertainty to restoring, and the accuracy of result is restored in impact, and has increased computation burden, has incured loss through delay the time, is not suitable for being applied in target detection process.Therefore, study and propose consuming time less, recovery effect is good, and general restored method is to be of practical significance very much and practical value.
The technical scheme of prior art mainly contains following several:
(1) iteration blind deconvolution restored method (IBD)
[5]and improve after all restored methods
[6-8](containing TV restored method
[9]deng).
G. R. Ayers and J. C. Dainty proposed the iteration blind deconvolution method (IBD restored method) based on single frames in 1988
[5], and be applied in the recovery of atmospheric turbulence degraded image, image being carried out to nonnegativity restriction in iteration each time.The people's such as Ayers and Dainty research has excited cosmonautics bound pair to meet the eye on every side the very big interest of deconvoluting.Although this method is existence and stability problem under noise situations, be proved to be very rising.1989, the people such as Davery and Lane proposed similar scheme
[6], but algorithm has further been set priori, and support region is known, obtains the estimation of target and PSF with S filter.IBD algorithm computational complexity is lower, but more responsive to noise, and major defect is to lack reliability.Many scholars have delivered the improvement to IBD rudimentary algorithm, and 1992, Lane proposed a kind of Blind Deconvolution technology of the conjugate gradient that is applied to spot image
[7], this algorithm has better robustness than the IBD of Ayers and Dainty proposition.Its basic structure is identical with IBD algorithm.Main difference is to have used in iterative process Gradient Descent minimization error term, thereby obtains good recovery effect.This algorithm has been considered the existence of noise, but for complex target, there will be some may be corresponding to the solution of total error item local minimum.In order to solve Blind Deconvolution algorithm to the susceptibility of noise and to occur some illusions and separate uncertain problem, Lane and Law proposed the blind restoration method of optimizing based on least square in 1996
[8], making it to minimize to obtain optimum solution by setting up an error term, and all minimize this error term by method of steepest descent in each iterative process, the non-negativity constraint applying and support region restriction etc. all supposes it is accurately known.
Iteration blind deconvolution restored method (IBD) is to Feinup phase retrieval algorithm in essence
[10]popularization, the method for employing is iteration, and the non-negative row of image is limited by priori, can obtain by simple liftering the estimated value of target and PSF in iteration each time.Iteration blind deconvolution method is the in the situation that of point spread function the unknown, need not be for referencial use to guiding, but utilize some rational prioris, if target strength and point spread function values are non-negative, in support region size and frequency field, some known characteristic is carried out estimating target intensity.Its key is the application about the priori of degradation property and image.
Iteration blind deconvolution restored method after iteration blind deconvolution restored method (IBD) and improvement, is all to restore by the mode of view picture degraded image data acquisition iteration, multiplex in the recovery of atmospheric turbulence degraded image.The shortcoming of its existence is:
(1) be only applicable to the recovery of atmospheric turbulence degraded image, there is no versatility
IBD and all improving one's methods thereof (containing TV restoring method etc.) are only applicable in the recovery of atmospheric turbulence degraded image, to the image of other degradation modes, as motion blur image, defocus blurred image, the degraded image that composite factor causes maybe cannot be known degraded image of degradation modes etc., can not restore.If need to use different restored methods to other degradation modes image restorations;
(2) length consuming time
IBD and all improving one's methods thereof (containing TV restoring method etc.) are restored by the information of whole image, and all data have participated in computing, and the form of computing employing iteration, have strengthened operand, cause consuming time very long;
(3) can not be used in the recovery of real image
IBD and all improving one's methods thereof (containing TV restoring method etc.) are often only better to emulation degraded image effect, but bad to actual degraded image recovery effect.
(2) Shan restored method
[11]and Fergus restored method
[12]deng the motion blur image restoration method of visible ray.
For the recovery of motion blur image, Fergus has proposed one and has utilized Bayesian model to carry out statistical picture shade of gray distribution statistics characteristic, thereby the moving image restored method of ambiguous estimation image, Shan has proposed a kind of more effective motion blur image restoration method, and the method is the probability Distribution Model based on gradation of image equally.These two algorithms mainly utilize the prioris such as image border probability distribution, all need size of manual input fuzzy core etc.In actual conditions, the size of fuzzy core, direction is very doubt, such restored method major defect is that the priori of hypothesis is not always used in general image.And this class restored method is only used in the recovery of motion blur image, and to other forms of blurred picture impracticable.The shortcoming of its existence is:
(1) be only applicable to visible ray motion blur image restoration, there is no versatility
The motion blur image restoration method of the visible rays such as entropy restored method and Fergus restored method is only applicable in the recovery of atmospheric turbulence degraded image, to the image of other degradation modes, as Turbulence-degraded Images, defocus blurred image, the degraded image that composite factor causes maybe cannot be known degraded image of degradation modes etc., can not restore.If need to use different restored methods to other degradation modes image restorations.
(2) length consuming time
The motion blur image restoration method of the visible rays such as entropy restored method and Fergus restored method is restored by whole information in certain piece region in whole image or image, all data have participated in computing, and the form of computing employing iteration, strengthen operand, cause consuming time very long.
List of references (as patent/paper/standard)
1. M. R. Banham and A. K. Katsagellos, “Digital image restoration,”
IEEE Signal Process. Mag. 14, 24-41 (1997).
2. M. R. Banham and A. K. Katsagellos, “Digital image restoration,”
IEEE Signal Process. Mag. 14, 24-41 (1997).
3. Y. Yitzhaky, N. S. Kopeika, “Identification of blur parameters from motion blurred images,”
Graphics Models and Image Processing,
59(5), 310-320 (1997).
4. A. A. Sawchuk, “Space variant system analysis of image motion,”
J. Opt. Soc. Am. A 63(9), 1052-1063 (1973).
5. G. R. Ayers and J. C. Dainty, “Iterative blind deconvolution method and its applications,”
Opt. Lett. 13(7), 547-549 (1988).
6. B. L. K. Davey, R. G. Lane, R. H. T. Bates, “Blind deconvolution of noisy complex valued image”, Opt. Comm.
69, 353-356, (1989).
7. R. G. Lane, “Blind deconvolution of speckle image,”
J. Opt. Soc. Am. A 9(9), 1508-1514 (1992).
8. N. F. Law, R. G. Lane, “Blind deconvolution using least squares minmization” Optics Communications,
128, 341-352, (1996)
9. T. F. Chan and C. K. Wong, “Total variation blind deconvolution,”
IEEE Trans. Image Process. 7(3), 370-375 (1998).
10. J. R. Fienup, “Reconstruction of an object from the modulus of its Fourier transforms”, Optics Letters,
3(1), 27-29,( 1978).
11. Q. Shan, J. Jia, and A. Agarwala, “High-quality motion deblurring from a single image,”
ACM Trans. on Graphics.
27(3), article 73 (2008).
12. R. Fergus, B. Singh, A. Hertzmann, S. T. Roweis, and W. T. Freeman, “Removing camera shake from a single photograph,” ACM Trans. Graphics
25(3), 787–794 (2006) 。
Summary of the invention
Technical matters to be solved by this invention is the general restored method that proposes a kind of target detection multispectral image for above-mentioned prior art, do not need to know degradation modes, and recovery speed is fast, good to actual degraded image recovery effect.
The present invention solves the problems of the technologies described above adopted technical scheme: the general restored method of target detection multispectral image, and its step comprises:
1) input degraded image, detects degraded image profile, is found out degraded image zone of transition and is predicted zone of transition picture rich in detail by image outline, estimates to obtain point spread function maximum support territory by width of transition zone simultaneously
m×
m;
2) select from multispectral degraded image zone of transition
kindividual, wherein
k>
m 2, on the zone of transition picture rich in detail of prediction, find
kindividual corresponding region, by
kgroup data construct is about the matrix equation group of point spread function, and its dimension is
k×
m 2;
3) in matrix equation, add non-negativity constraint and anisotropic spatial coherence bound term, and solve and obtain
xiterative relation equation
;
4) make initial value
, wherein point spread function matrix number
xsize be
m×
m;
5) by
xiterative equation solve when obtaining n and becoming n+1
xvalue
;
6) if
, stop calculating, otherwise repeating step 5), until satisfy condition, obtain point spread function
x;
7), by solving the point spread function obtaining, restore and obtain picture rich in detail by the non-blind deconvolution method of maximal possibility estimation.
The present invention select degraded image in comprise a large amount of degradation information point do restoration calculation, not only make recovery time greatly reduce, and ensured to greatest extent to restore accuracy and the real-time of result.This method and IBD and all improving one's methods thereof (containing TV restoring method etc.) difference is as follows: 1, IBD and all improving one's methods thereof (containing TV restoring method etc.) are utilized view picture degraded image solution point spread function; And method of the present invention only utilizes the point that comprises a large amount of degradation information on degraded image to carry out solution point spread function; 2, IBD and all improving one's methods thereof (containing TV restoring method etc.) obtain final true picture by point spread function with alternately iterative true picture; And the first solution point spread function of method of the present invention, then direct solution true picture, without iteration, as Fig. 2.
System of the present invention is moved under normal computer environments, for the recovery to various Spectrum curve degradation images and sharpening, can be used for target detection image restoration, infrared terminal guidance image restoration, daily image sharpening photographing etc.
The present invention with respect to the existing beneficial effect of prior art is:
1) general to all spectrum pictures, can realize any spectrum picture, as infrared image, visible images, millimeter-wave image, Terahertz image etc. can restore and sharpening processing, the blurred picture that any fuzzy form is caused, as motion blur, defocusing blurring, pneumatic fuzzy etc. can recovery, and high-definition processing to multispectral target detection image;
2) the present invention carries out point spread function discretize within the scope of limited excitation region, only demand discrete value, do not need psf model and parameter to estimate, and then complicacy and the diversity of avoidance point spread function form, without set up point spread function mathematical model by degradation modes, thereby without knowing image degradation pattern;
3) the present invention only selects some valid pixel information on degraded image to carry out solution point spread function, and has got rid of the point on noise spot and the flat site on degraded image, has improved the accuracy of result of calculation, has also reduced calculated amount.The present invention has simultaneously avoided alternately interative computation in solution point spread function and picture rich in detail process, further reduced calculated amount, thereby method of the present invention is consuming time few;
4) method of the present invention is good to true picture recovery effect.Existing restored method attempts to find by image degradation pattern the mathematical model of corresponding point spread function, but target imaging can be subject to atmospheric disturbance, and motion defocuses and the interference of other complicated factor, is difficult to distinguish the influence degree of various factors.Point spread function form is ignorant, is beyond expression of words by mathematical model, describes and does not meet reality by single mathematical model, and recovery result is also bad; Method of the present invention is described point spread function by the discrete value within the scope of limited excitation region, has avoided expressing point spread function by single mathematical model, and more realistic, recovery effect is also better.Experiment showed, that the present invention has good recovery effect to real image really;
5) method of the present invention is inputted a frame blurred picture, can obtain picture rich in detail, is simple and easy to use.Method of the present invention can machine/spaceborne chip miniaturization (hardware realization) simultaneously.
Brief description of the drawings
Fig. 1 is the process flow diagram of the general restored method of target detection multispectral image of the present invention;
Fig. 2 is the difference of method of the present invention and IBD restored method;
Fig. 3 (a)-3 (c) is for adopting method of the present invention one width Terahertz degraded image to be realized to the concrete steps procedure chart restoring;
Fig. 4 (a) and 4 (b) are respectively infrared degraded image and adopt the result of method processing of the present invention;
Fig. 5 (a) and 5 (b) are respectively visible ray degraded image and adopt the result of method processing of the present invention.
Embodiment
Below in conjunction with accompanying drawing and example, the present invention is further detailed explanation.
(1) Fig. 3 (a) is a width Terahertz degraded image, and input picture detects degraded image profile, by image outline prediction zone of transition picture rich in detail.Estimate to obtain point spread function maximum support territory by width of transition zone simultaneously
m×
m;
(2) select from multispectral degraded image
k(
k>
m 2) individual point, on the zone of transition picture rich in detail of prediction, find
kindividual corresponding region.By
kabout the matrix equation group of point spread function, (dimension is group data construct
k×
m 2);
(3) in matrix equation, add non-negativity constraint and anisotropic spatial coherence bound term, and solve and obtain
xiterative relation equation
;
(4) make initial value
.Wherein point spread function matrix number
xsize be
m×
m;
(5) by
xiterative equation solve when obtaining n and becoming n+1
xvalue
;
(6) if
, (
for setting arbitrarily small value), stop calculating, otherwise repeating step 5), until satisfy condition.Obtain point spread function
x, be denoted as
; Fig. 3 (b) is the degraded image point spread function x being calculated by step (1)-(6);
(7), by solving the point spread function obtaining, restore and obtain picture rich in detail by the non-blind deconvolution method of maximal possibility estimation.Fig. 3 (c) restores by step (7) picture rich in detail obtaining.
For the validity and the versatility that prove that method of the present invention is restored various spectrum pictures, except accompanying drawing 3 Terahertz images, provide in addition accompanying drawing 4 and accompanying drawing 5.Wherein Fig. 4 (a) and Fig. 4 (b) are respectively that infrared degraded image and infrared image restore result, and Fig. 5 (a) and Fig. 5 (b) are that visible ray degraded image can restore result by light image.Experimental result shows that method of the present invention broken away from the predicament of current image restoration length consuming time, all can recover various Spectrum curve degradation images, do not need priori and spectrum source and degradation modes, input a frame blurred picture and can obtain HD image, recovery effect is better, is expected to obtain application in target detection system.
The present invention is not only confined to above-mentioned embodiment; persons skilled in the art are according to content disclosed by the invention; can adopt other multiple embodiments to implement the present invention; therefore; every employing project organization of the present invention and thinking; do some simple change or designs of change and according to the present invention the hardware design technique of content and method, all fall into the scope of protection of the invention.
Claims (1)
1. the general restored method of target detection multispectral image, its step comprises:
1) input degraded image, detects degraded image profile, is found out degraded image zone of transition and is predicted zone of transition picture rich in detail by image outline, estimates to obtain point spread function maximum support territory by width of transition zone simultaneously
m×
m;
2) select from multispectral degraded image zone of transition
kindividual, wherein
k>
m 2, on the zone of transition picture rich in detail of prediction, find
kindividual corresponding region, by
kgroup data construct is about the matrix equation group of point spread function, and its dimension is
k×
m 2;
3) in matrix equation, add non-negativity constraint and anisotropic spatial coherence bound term, and solve and obtain
xiterative relation equation
;
4) make initial value
, wherein point spread function matrix number
xsize be
m×
m;
5) by
xiterative equation solve when obtaining n and becoming n+1
xvalue
;
6) if
,
for setting arbitrarily small value, stop calculating, otherwise repeating step 5), until satisfy condition, obtain point spread function
x;
7), by solving the point spread function obtaining, restore and obtain picture rich in detail by the non-blind deconvolution method of maximal possibility estimation.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201210197707.7A CN102768756B (en) | 2012-06-15 | 2012-06-15 | Universal recovery method for target detection multi-spectrum images |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201210197707.7A CN102768756B (en) | 2012-06-15 | 2012-06-15 | Universal recovery method for target detection multi-spectrum images |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN102768756A CN102768756A (en) | 2012-11-07 |
| CN102768756B true CN102768756B (en) | 2014-08-06 |
Family
ID=47096149
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201210197707.7A Expired - Fee Related CN102768756B (en) | 2012-06-15 | 2012-06-15 | Universal recovery method for target detection multi-spectrum images |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN102768756B (en) |
Families Citing this family (6)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN103049889A (en) * | 2012-12-06 | 2013-04-17 | 华中科技大学 | Preprocessing method for strong noise aero-optic effect degraded image |
| CN103093439A (en) * | 2013-01-23 | 2013-05-08 | 岑夏凤 | Image blind restoration method based on multi-scattering degradation model |
| CN105890757B (en) * | 2014-12-15 | 2019-05-17 | 中国科学院国家天文台 | The one-dimensional spectrum of multi fiber spectrum is extracted using Deconvolution Method |
| CN105809629B (en) * | 2014-12-29 | 2018-05-18 | 清华大学 | A kind of point spread function number estimation method and system |
| CN109035139B (en) * | 2018-06-28 | 2022-11-22 | 长光卫星技术股份有限公司 | High-resolution satellite image modulation transfer function compensation method |
| CN116309190B (en) * | 2023-05-17 | 2023-08-15 | 武汉工程大学 | Turbulence degradation image recovery method based on optimal region median prior |
Citations (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US6470097B1 (en) * | 1999-01-22 | 2002-10-22 | Siemens Corporation Research, Inc. | Total variational blind image restoration from image sequences |
| CN101504765A (en) * | 2009-03-20 | 2009-08-12 | 东华大学 | Motion blur image sequence restoration method employing gradient amalgamation technology |
-
2012
- 2012-06-15 CN CN201210197707.7A patent/CN102768756B/en not_active Expired - Fee Related
Patent Citations (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US6470097B1 (en) * | 1999-01-22 | 2002-10-22 | Siemens Corporation Research, Inc. | Total variational blind image restoration from image sequences |
| CN101504765A (en) * | 2009-03-20 | 2009-08-12 | 东华大学 | Motion blur image sequence restoration method employing gradient amalgamation technology |
Non-Patent Citations (3)
| Title |
|---|
| 基于Poisson 模型的湍流退化图像多帧迭代复原算法;洪汉玉等;《宇航学报》;20041130;第25卷(第6期);649-654 * |
| 洪汉玉等.基于Poisson 模型的湍流退化图像多帧迭代复原算法.《宇航学报》.2004,第25卷(第6期), |
| 洪汉玉等.基于各向异性和非线性正则化的湍流退化图像复原.《宇航学报》.2004,第25卷(第01期), * |
Also Published As
| Publication number | Publication date |
|---|---|
| CN102768756A (en) | 2012-11-07 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN102768756B (en) | Universal recovery method for target detection multi-spectrum images | |
| Li et al. | Total variation blind deconvolution employing split Bregman iteration | |
| Mao et al. | Non rigid geometric distortions correction-application to atmospheric turbulence stabilization | |
| Tang et al. | Non-blind image deblurring method by local and nonlocal total variation models | |
| Luan et al. | Fast single image dehazing based on a regression model | |
| Zhu et al. | Stabilizing and deblurring atmospheric turbulence | |
| Dong et al. | A piecewise local regularized Richardson–Lucy algorithm for remote sensing image deconvolution | |
| US20140105515A1 (en) | Stabilizing and Deblurring Atmospheric Turbulence | |
| CN104103050A (en) | Real video recovery method based on local strategies | |
| US10007970B2 (en) | Image up-sampling with relative edge growth rate priors | |
| Gao et al. | Atmospheric turbulence removal using convolutional neural network | |
| CN104282003A (en) | Digital blurred image blind restoration method based on gradient screening | |
| Zhang et al. | Exploiting image local and nonlocal consistency for mixed Gaussian-impulse noise removal | |
| Zhou et al. | A boundary condition based deconvolution framework for image deblurring | |
| Hong et al. | Multi-frame real image restoration based on double loops with alternative maximum likelihood estimation | |
| Xu et al. | Image deblurring with blur kernel estimation in RGB channels | |
| Liu et al. | Large size single image fast defogging and the real time video defogging FPGA architecture | |
| Li et al. | Space–time super-resolution with patch group cuts prior | |
| Li et al. | Image deblurring for satellite imagery using small-support-regularized deconvolution | |
| Wang et al. | Single image haze removal via attention-based transmission estimation and classification fusion network | |
| Cui et al. | Multi-frame motion deblurring using coded exposure imaging with complementary fluttering sequences | |
| Muthukumar et al. | An efficient color image denoising method for Gaussian and impulsive noises with blur removal | |
| Gao et al. | High performance super-resolution reconstruction of multi-frame degraded images with local weighted anisotropy and successive regularization | |
| Roy et al. | Modeling of Haze image as Ill-posed inverse problem & its solution | |
| Su et al. | High spatial resolution image restoration from subpixel-shifted hyperspectral images |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| C06 | Publication | ||
| PB01 | Publication | ||
| C10 | Entry into substantive examination | ||
| SE01 | Entry into force of request for substantive examination | ||
| C14 | Grant of patent or utility model | ||
| GR01 | Patent grant | ||
| CF01 | Termination of patent right due to non-payment of annual fee |
Granted publication date: 20140806 Termination date: 20170615 |
|
| CF01 | Termination of patent right due to non-payment of annual fee |