CN106127700B - One kind staring infrared degraded image adaptive restoration method - Google Patents
One kind staring infrared degraded image adaptive restoration method Download PDFInfo
- Publication number
- CN106127700B CN106127700B CN201610436012.8A CN201610436012A CN106127700B CN 106127700 B CN106127700 B CN 106127700B CN 201610436012 A CN201610436012 A CN 201610436012A CN 106127700 B CN106127700 B CN 106127700B
- Authority
- CN
- China
- Prior art keywords
- image
- coordinate
- regularization
- comentropy
- anisotropic diffusion
- 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.)
- Active
Links
- 238000000034 method Methods 0.000 title claims abstract description 57
- 230000003044 adaptive effect Effects 0.000 title claims abstract description 22
- 238000009792 diffusion process Methods 0.000 claims abstract description 50
- 230000015556 catabolic process Effects 0.000 claims abstract description 43
- 238000006731 degradation reaction Methods 0.000 claims abstract description 43
- 241000287181 Sturnus vulgaris Species 0.000 claims description 9
- 238000010276 construction Methods 0.000 claims description 3
- 230000008859 change Effects 0.000 claims description 2
- 229910017435 S2 In Inorganic materials 0.000 claims 1
- 230000005484 gravity Effects 0.000 abstract description 4
- 238000003706 image smoothing Methods 0.000 abstract description 3
- 238000001914 filtration Methods 0.000 description 14
- 230000008569 process Effects 0.000 description 10
- 230000009466 transformation Effects 0.000 description 8
- 230000007850 degeneration Effects 0.000 description 6
- 230000003412 degenerative effect Effects 0.000 description 6
- 238000010586 diagram Methods 0.000 description 4
- 230000002776 aggregation Effects 0.000 description 3
- 238000004220 aggregation Methods 0.000 description 3
- 238000003491 array Methods 0.000 description 3
- 230000008901 benefit Effects 0.000 description 3
- 238000012512 characterization method Methods 0.000 description 3
- 238000005516 engineering process Methods 0.000 description 3
- 230000002708 enhancing effect Effects 0.000 description 3
- 230000002401 inhibitory effect Effects 0.000 description 3
- 239000011159 matrix material Substances 0.000 description 3
- 230000009467 reduction Effects 0.000 description 3
- 238000004364 calculation method Methods 0.000 description 2
- 230000007423 decrease Effects 0.000 description 2
- 230000000694 effects Effects 0.000 description 2
- 238000005457 optimization Methods 0.000 description 2
- 241001123248 Arma Species 0.000 description 1
- 238000004458 analytical method Methods 0.000 description 1
- 230000005540 biological transmission Effects 0.000 description 1
- OVBPIULPVIDEAO-LBPRGKRZSA-N folic acid Chemical compound C=1N=C2NC(N)=NC(=O)C2=NC=1CNC1=CC=C(C(=O)N[C@@H](CCC(O)=O)C(O)=O)C=C1 OVBPIULPVIDEAO-LBPRGKRZSA-N 0.000 description 1
- 238000003384 imaging method Methods 0.000 description 1
- 230000033001 locomotion Effects 0.000 description 1
- 238000005192 partition Methods 0.000 description 1
- 238000011084 recovery Methods 0.000 description 1
- 230000000007 visual effect Effects 0.000 description 1
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T5/00—Image enhancement or restoration
Landscapes
- Physics & Mathematics (AREA)
- General Physics & Mathematics (AREA)
- Engineering & Computer Science (AREA)
- Theoretical Computer Science (AREA)
- Image Processing (AREA)
Abstract
The present invention relates to one kind to stare infrared degraded image adaptive restoration method, belongs to image real time transfer field.The method constructs image degradation model A；Calculate separately the comentropy of the input picture and the anisotropic diffusion coefficients of input picture；Any coordinate (i is calculated according to the comentropy and anisotropic diffusion coefficients later, j) the regularization coefficient λ (i at place, j), restored image is finally calculated according to input picture I, image degradation model A and regularization coefficient λ (i, j).The characteristics of combining information entropy H of the present invention and anisotropic diffusion coefficients G, the regularization coefficient acquired according to the specific gravity of the two, has multiple dimensioned restorability, realize that regularization ability in image smoothing area is strong, regularization ability weak function in image detail region keeps image restoration more accurate, simultaneously, high exponent arithmetic(al) and labyrinth is not present during image restoration of the present invention, algorithm operation quantity is small, is easy to hardware real-time implementation.
Description
Technical field
The invention belongs to image real time transfer fields, and in particular to one kind stares infrared degraded image adaptive restoration side
Method.
Background technique
In infrared image acquisition, transmission and treatment process, since by atmospheric perturbation, optics into focus is bad, scenery
It is influenced with factors such as the relative motions of imaging device, the image quality decrease caused, image thickens.In order to obtain height
Signal-to-noise ratio, image high-definition need to restore degraded image according to image degradation model.
Image restoration is basic, premise a treatment process in field of image processing, in primary vision processing
Occupy extremely important status, domestic and foreign scholars take much count of the research of this respect.Image restoration is according to known priori knowledge
How much typical image restoration and blind image restoration two major classes can be divided into.The complexity of image restoration depends primarily on priori
The levels of precision of the acquisition of knowledge.Typical image restoration is the point spread function degenerated according to determining priori knowledge computing system
Number restores degraded image, then using anti-methods degenerated such as liftering, least squares filterings such as liftering method, sky
Domain filtering method and algebraic method etc..Blind image restoration refers to the priori knowledge for not needing system degradation or only needs portion
The priori knowledge that subsystem is degenerated, by establishing model to degenerative process (fuzzy and noise), and then from degraded image feature
Estimate true picture, such as zero blade face partition method, ARMA Parameter Estimation Method, priori fuzzy recognition method.Existing infrared image is multiple
Original method has the disadvantage in that (1) most existing infrared image restored methods are only applicable to the input picture of high s/n ratio, with
Input picture signal-to-noise ratio decline, restored image visual effect be deteriorated, unfavorable understanding and analysis with the mankind or machine to image；
(2) most existing infrared image restored methods carry out regularization using same regularization parameter to entire image, do not utilize
Local feature information causes to have lost a large amount of detailed information in recuperation；(3) most existing image recovery method operands
Greatly, it is not easy to hardware real-time implementation.
Summary of the invention
To solve the above-mentioned problems, the present invention provides one kind to stare infrared degraded image adaptive restoration method, method
Simply, strong applicability, work well and be suitble to hardware real-time implementation.
The present invention stares infrared degraded image adaptive restoration method, mainly comprises the steps that
S1, input picture I is obtained；
S2, construction image degradation model A；
S3, the comentropy for calculating the input picture and the anisotropic diffusion coefficients for calculating the input picture；
S4, the regularization coefficient λ at any coordinate (i, j) is calculated according to the comentropy and anisotropic diffusion coefficients
(i, j),
Wherein, H (i, j) is the comentropy at coordinate (i, j), and G (i, j) is the anisotropy parameter system at coordinate (i, j)
Number, HmaxAnd HminMaximum value and minimum value in respectively comentropy H, GmaxAnd GminRespectively indicate anisotropic diffusion coefficients G
In maximum value and minimum value, 0≤a≤1,0≤b≤1, a+b=1；
S5, restored image I is calculated according to input picture I, image degradation model A and regularization coefficient λ (i, j)out；
S6, output restored image Iout。
Preferably, in the step S2, image degradation model A selects gauss low frequency filter, and expression formula is as follows:
Wherein, A (i, j) indicates degradation model A in the coefficient of coordinate position (i, j), and D (i, j) is coordinate (i, j) away from Fu
The distance of leaf transformation origin, σ1Indicate the degree of Gaussian curve extension.
Preferably, in the step S3, comentropy H (i, j) expression formula of neighborhood P × Q centered on coordinate (i, j)
It is as follows:
Wherein, the intensity profile probability at the position p (i, j) indicates coordinate (i, j), I (i, j) and I (i+m, j+n) difference table
Diagram as I the position coordinate (i, j) and coordinate (i+m, j+n) at gray value, 1≤m≤P, 1≤n≤Q.
Preferably, in the step S3, the anisotropic diffusion coefficients of image are based on following equation and seek:
I (i, j, t) is the image pixel value at the position moment t coordinate (i, j),It is gradient operator, div is that divergence is calculated
Son,It is partial gradient value, G () indicates the diffusion coefficient function of the partial gradient value,It is that inclined is asked to image I
Operation is led, here, the function expression of anisotropic diffusion coefficientsIt is as follows:
Wherein, σ is Image neighborhood standard deviation.
Preferably, in the step S4, a and the equal value 0.5 of b.
Preferably, in the step S5, according to input picture I, image degradation model A and regularization coefficient λ (i, j)
Calculate restored image IoutShi Caiyong iterative method.
In above scheme preferably, in the step S5 ,+1 iterative image I of kth is sought using iterative methodk+1Table
It is as follows up to formula:
Wherein, Ik+1(i, j) indicates image Ik+1Gray value at the position coordinate (i, j), Ik(i, j) indicates image Ik?
Gray value at the position coordinate (i, j), U (i, j) indicate regularization value of the regularization factors U at the position coordinate (i, j), A table
Show that image degradation model, * are convolution algorithms,It is related operation, is multiplication operation, restored image Iout=Ik+1。
In any of the above-described scheme preferably, in the step S5, the number of iterations is 5~20.
Compared with prior art, the present invention has following remarkable advantage: (1) particular for Low SNR Infrared Images,
Inhibit effectively restore detailed information while noise；(2) according to the comentropy of image local area and anisotropy parameter system
Number adaptive polo placement regularization parameters, have multiple dimensioned restorability, realize that regularization ability in image smoothing area is strong, and image is thin
It is weak to save area's regularization ability；(3) image degradation model can flexibly be constructed according to practical degenerative process；(4) there is no high-orders to transport
Calculation and labyrinth, algorithm operation quantity is small, is easy to hardware real-time implementation.
Detailed description of the invention
Fig. 1 is the flow chart for the preferred embodiment that the present invention stares infrared degraded image adaptive restoration method.
Specific embodiment
To keep the purposes, technical schemes and advantages of the invention implemented clearer, below in conjunction in the embodiment of the present invention
Attached drawing, technical solution in the embodiment of the present invention is further described in more detail.In the accompanying drawings, identical from beginning to end or class
As label indicate same or similar element or element with the same or similar functions.Described embodiment is the present invention
A part of the embodiment, instead of all the embodiments.The embodiments described below with reference to the accompanying drawings are exemplary, it is intended to use
It is of the invention in explaining, and be not considered as limiting the invention.Based on the embodiments of the present invention, ordinary skill people
Member's every other embodiment obtained without creative efforts, shall fall within the protection scope of the present invention.Under
Face is described in detail the embodiment of the present invention in conjunction with attached drawing.
In the description of the present invention, it is to be understood that, term " center ", " longitudinal direction ", " transverse direction ", "front", "rear",
The orientation or positional relationship of the instructions such as "left", "right", "vertical", "horizontal", "top", "bottom" "inner", "outside" is based on attached drawing institute
The orientation or positional relationship shown, is merely for convenience of description of the present invention and simplification of the description, rather than the dress of indication or suggestion meaning
It sets or element must have a particular orientation, be constructed and operated in a specific orientation, therefore should not be understood as protecting the present invention
The limitation of range.
The present invention stares infrared degraded image adaptive restoration method as shown in Figure 1, mainly comprising the steps that
S1, input picture I is obtained；
S2, construction image degradation model A；
S3, the comentropy for calculating the input picture and the anisotropic diffusion coefficients for calculating the input picture；
S4, the regularization coefficient λ at any coordinate (i, j) is calculated according to the comentropy and anisotropic diffusion coefficients
(i, j),
Wherein, H (i, j) is the comentropy at coordinate (i, j), and G (i, j) is the anisotropy parameter system at coordinate (i, j)
Number, HmaxAnd HminMaximum value and minimum value in respectively comentropy H, GmaxAnd GminRespectively indicate anisotropic diffusion coefficients G
In maximum value and minimum value, 0≤a≤1,0≤b≤1, a+b=1；
S5, restored image I is calculated according to input picture I, image degradation model A and regularization coefficient λ (i, j)out；
S6, output restored image Iout。
Preferably, in the step S2, image degradation model A selects gauss low frequency filter, and expression formula is as follows:
Wherein, A (i, j) indicates degradation model A in the coefficient of coordinate position (i, j), and D (i, j) is coordinate (i, j) away from Fu
The distance of leaf transformation origin, σ1Indicate the degree of Gaussian curve extension.
Illustrate that the present invention stares infrared degraded image adaptive restoration method below with specific example.
Embodiment 1,
Thermal infrared imager focal plane arrays (FPA) size is 640 × 512, and working frame frequency is 50 frame per second.Image processing platform is adopted
With DSP+FPGA framework, stares infrared degraded image adaptive restoration method and realized in dsp processor, what satisfaction was handled in real time
Demand.
It is understood that in step sl, dsp processor input picture is 16 bit digital images, picture size is 640
× 512, it is described below, the difference round numbers of i in the coordinate (i, j) of image, j, and 1≤i≤512,1≤j≤640.
In the present embodiment, image is carried out restoring processing it is generally necessary to be carried out according to certain image degradation model, one
The degenerative process of image can be modeled as one and acted on original image f (x, y) by a simple general image degradation model
Degeneration system H, the synergy of exercising result and a Gaussian noise n (x, y) causes to have produced degraded image g (x, y).
An approximation of original image f (x, y) can be obtained from given degraded image g (x, y) according to above-mentioned degeneration system H.
The method of removal Gaussian noise has a histogram transformation, low-pass filtering, high-pass filtering, liftering, Wiener filtering, median filtering etc.,
In the present embodiment step S2, image degradation model A selects gauss low frequency filter, and expression formula is as follows:
Wherein, A (i, j) indicates degradation model A in the coefficient of coordinate position (i, j), and D (i, j) is coordinate (i, j) away from Fu
The distance of leaf transformation origin, σ1Indicating the degree of Gaussian curve extension, e is natural constant, about 2.71828.
Image degradation model selects gauss low frequency filter, and setting filter size is 5 × 5, standard deviation 0.6, then
Electric-wave filter matrix expression is as shown in table 1:
15 × 5 gauss low frequency filter of table
| 0 | 0.0004 | 0.0017 | 0.0004 | 0 |
| 0.0004 | 0.0274 | 0.1099 | 0.0274 | 0.0004 |
| 0.0017 | 0.1099 | 0.4407 | 0.1099 | 0.0017 |
| 0.0004 | 0.0274 | 0.1099 | 0.0274 | 0.0004 |
| 0 | 0.0004 | 0.0017 | 0.0004 | 0 |
In step S3, the comentropy and Anisotropic diffusion coefficient of image are calculated separately, it is to be understood that above-mentioned technology
There are diversity for method, are illustrated below with a certain specific equation.
In the step S3, comentropy H (i, j) expression formula of neighborhood P × Q centered on coordinate (i, j) is as follows:
Wherein, the intensity profile probability at the position p (i, j) indicates coordinate (i, j), I (i, j) and I (i+m, j+n) difference table
Diagram as I the position coordinate (i, j) and coordinate (i+m, j+n) at gray value, 1≤m≤P, 1≤n≤Q.
It selects 7*7 neighborhood to calculate comentropy below to be illustrated, i.e. P=7, Q=7, with IinIt is replaced as input picture upper
The image I of formula is stated, then the expression formula after substituting is as follows:
Wherein, 1≤m≤7,1≤n≤7,1≤i≤512,1≤j≤640.
The anisotropic diffusion coefficients G of image is based on following equation and seeks:
I (i, j, t) is the image pixel value at the position moment t coordinate (i, j),It is gradient operator, div is that divergence is calculated
Son,It is partial gradient value, G () indicates the diffusion coefficient function of the partial gradient value,It is that inclined is asked to image I
Operation is led, here, the function expression of anisotropic diffusion coefficientsIt is as follows:
Wherein, σ is Image neighborhood standard deviation, is 0.6, and neighborhood is 5 × 5.
It is understood that any pixel value is there are corresponding comentropy and respectively for described 640 × 512 image
Item anisotropic diffusion coefficient is indicated with H (i, j) and G (i, j) distribution, and therefore, for comentropy, there are maximum value Hmax
With minimum value Hmin, similarly, for anisotropic diffusion coefficients G, there are maximum value GmaxWith minimum value Gmin。
Regularization will mainly solve ill-conditioning problem in restored method.Regularization coefficient λ can gauge signal it is odd
The detailed information of signal can be preferably kept while anisotropic.When regularization coefficient λ increases, high-frequency noise can be effectively suppressed,
However, image detail holding capacity is deteriorated；When regularization coefficient λ reduces, the enhancing of image detail holding capacity.Therefore, according to figure
It is that restoration algorithm is successfully crucial as content-adaptive calculates regularization coefficient.The aggregation of comentropy H characterization image intensity profile
Characteristic, image is more uniform, and comentropy is bigger.Anisotropic diffusion coefficients G characterizes the smoothness of image, and image is more uniform, respectively
Anisotropic diffusion coefficient is smaller.
In order to utmostly remove the singular value generated in recuperation, and the detailed information of image is kept, knot of the present invention
The characteristics of closing comentropy H and anisotropic diffusion coefficients G is expanded in the step S4 according to the comentropy and anisotropy
The regularization coefficient λ (i, j) at any coordinate (i, j) of coefficient calculating is dissipated,
Wherein, a, b are constant coefficients, the specific gravity accounted for when for reflecting comentropy and anisotropic diffusion coefficients load regularization, 0
≤ a≤1,0≤b≤1, a+b=1.In the present embodiment, to guarantee versatility and optimization process effect, a and the equal value 0.5 of b, structure
The regularization coefficient built makes edge detail information while inhibiting the high-frequency noise of smooth region (regularization coefficient λ increase)
(regularization coefficient λ reduction) is positively maintained.
In step S5, restored image is calculated according to input picture I, image degradation model A and regularization coefficient λ (i, j)
Iout, in following formula, restored image Iout=Ik+1, the image I=I of inputk, formula is as follows:
In order to make restored image closer to real image, above-mentioned formula can also be used as iterative formula and carry out using iteration
Number is generally chosen for 5~20 times, i.e., is original image as k=1, and kth+1 time gray value of image later successively exists
It is calculated in the gray value of image of kth time, wherein Ik+1(i, j) indicates image Ik+1Gray scale at the position coordinate (i, j)
Value, Ik(i, j) indicates image IkGray value at the position coordinate (i, j), U (i, j) indicate regularization factors U at coordinate (i, j)
Regularization value at position selects Tikhonov model to calculate.A indicates image degradation model, and * is convolution algorithm,It is related
Operation, 1≤i≤512,1≤j≤640, k=5 in the present embodiment, is multiplication operation, restored image Iout=I6。
Embodiment 2,
Thermal infrared imager focal plane arrays (FPA) size is 320 × 256, and working frame frequency is 50 frame per second.Image processing platform is adopted
With DSP+FPGA framework, stares infrared degraded image adaptive restoration method and realized in dsp processor, what satisfaction was handled in real time
Demand.
It is understood that in step sl, dsp processor input picture is 16 bit digital images, picture size is 320
× 256, it is described below, the difference round numbers of i in the coordinate (i, j) of image, j, and 1≤i≤256,1≤j≤320.
In the present embodiment, image is carried out restoring processing it is generally necessary to be carried out according to certain image degradation model, one
The degenerative process of image can be modeled as one and acted on original image f (x, y) by a simple general image degradation model
Degeneration system H, the synergy of exercising result and a Gaussian noise n (x, y) causes to have produced degraded image g (x, y).
An approximation of original image f (x, y) can be obtained from given degraded image g (x, y) according to above-mentioned degeneration system H.
The method of removal Gaussian noise has a histogram transformation, low-pass filtering, high-pass filtering, liftering, Wiener filtering, median filtering etc.,
In the present embodiment step S2, image degradation model A selects gauss low frequency filter, and expression formula is as follows:
Wherein, A (i, j) indicates degradation model A in the coefficient of coordinate position (i, j), and D (i, j) is coordinate (i, j) away from Fu
The distance of leaf transformation origin, σ1Indicating the degree of Gaussian curve extension, e is natural constant, about 2.71828.
Image degradation model selects gauss low frequency filter, and setting filter size is 5 × 5, standard deviation 0.6, then
Electric-wave filter matrix expression is as shown in table 1:
15 × 5 gauss low frequency filter of table
| 0 | 0.0004 | 0.0017 | 0.0004 | 0 |
| 0.0004 | 0.0274 | 0.1099 | 0.0274 | 0.0004 |
| 0.0017 | 0.1099 | 0.4407 | 0.1099 | 0.0017 |
| 0.0004 | 0.0274 | 0.1099 | 0.0274 | 0.0004 |
| 0 | 0.0004 | 0.0017 | 0.0004 | 0 |
In step S3, the comentropy and Anisotropic diffusion coefficient of image are calculated separately, it is to be understood that above-mentioned technology
There are diversity for method, are illustrated below with a certain specific equation.
In the step S3, comentropy H (i, j) expression formula of neighborhood P × Q centered on coordinate (i, j) is as follows:
Wherein, the intensity profile probability at the position p (i, j) indicates coordinate (i, j), I (i, j) and I (i+m, j+n) difference table
Diagram as I the position coordinate (i, j) and coordinate (i+m, j+n) at gray value, 1≤m≤P, 1≤n≤Q.
It selects 10*10 neighborhood to calculate comentropy below to be illustrated, i.e. P=10, Q=10, with IinAs input picture generation
For the image I of above-mentioned formula, then the expression formula after substituting is as follows:
Wherein, 1≤m≤10,1≤n≤10,1≤i≤256,1≤j≤320.
The anisotropic diffusion coefficients G of image is based on following equation and seeks:
I (i, j, t) is the image pixel value at the position moment t coordinate (i, j),It is gradient operator, div is that divergence is calculated
Son,It is partial gradient value, G () indicates the diffusion coefficient function of the partial gradient value,It is that inclined is asked to image I
Operation is led, here, the function expression of anisotropic diffusion coefficientsIt is as follows:
Wherein, σ is Image neighborhood standard deviation, is 0.6, and neighborhood is 5 × 5.
It is understood that any pixel value is there are corresponding comentropy and respectively for described 320 × 256 image
Item anisotropic diffusion coefficient is indicated with H (i, j) and G (i, j) distribution, and therefore, for comentropy, there are maximum value Hmax
With minimum value Hmin, similarly, for anisotropic diffusion coefficients G, there are maximum value GmaxWith minimum value Gmin。
Regularization will mainly solve ill-conditioning problem in restored method.Regularization coefficient λ can gauge signal it is odd
The detailed information of signal can be preferably kept while anisotropic.When regularization coefficient λ increases, high-frequency noise can be effectively suppressed,
However, image detail holding capacity is deteriorated；When regularization coefficient λ reduces, the enhancing of image detail holding capacity.Therefore, according to figure
It is that restoration algorithm is successfully crucial as content-adaptive calculates regularization coefficient.The aggregation of comentropy H characterization image intensity profile
Characteristic, image is more uniform, and comentropy is bigger.Anisotropic diffusion coefficients G characterizes the smoothness of image, and image is more uniform, respectively
Anisotropic diffusion coefficient is smaller.
In order to utmostly remove the singular value generated in recuperation, and the detailed information of image is kept, knot of the present invention
The characteristics of closing comentropy H and anisotropic diffusion coefficients G is expanded in the step S4 according to the comentropy and anisotropy
The regularization coefficient λ (i, j) at any coordinate (i, j) of coefficient calculating is dissipated,
Wherein, a, b are constant coefficients, the specific gravity accounted for when for reflecting comentropy and anisotropic diffusion coefficients load regularization, 0
≤ a≤1,0≤b≤1, a+b=1.In the present embodiment, to guarantee versatility and optimization process effect, a and the equal value 0.5 of b, structure
The regularization coefficient built makes edge detail information while inhibiting the high-frequency noise of smooth region (regularization coefficient λ increase)
(regularization coefficient λ reduction) is positively maintained.
In step S5, restored image is calculated according to input picture I, image degradation model A and regularization coefficient λ (i, j)
Iout, in following formula, restored image Iout=Ik+1, the image I=I of inputk, formula is as follows:
Above-mentioned formula can also be used as iterative formula carry out using, i.e., be original image as k=1, kth later+
1 gray value of image is successively calculated in the gray value of image of kth time, wherein Ik+1(i, j) indicates image Ik+1It is sitting
Mark the gray value at the position (i, j), Ik(i, j) indicates image IkGray value at the position coordinate (i, j), U (i, j) are indicated just
Then change regularization value of the factor U at the position coordinate (i, j), Tikhonov model is selected to calculate.A indicates image degradation model, *
It is convolution algorithm,It is related operation, 1≤i≤256,1≤j≤320, k=10 in the present embodiment is multiplication operation, restores
Image Iout=I11。
Embodiment 3,
Thermal infrared imager focal plane arrays (FPA) size is 640 × 512, and working frame frequency is 50 frame per second.Image processing platform is adopted
With DSP+FPGA framework, stares infrared degraded image adaptive restoration method and realized in dsp processor, what satisfaction was handled in real time
Demand.
It is understood that in step sl, dsp processor input picture is 16 bit digital images, picture size is 640
× 512, it is described below, the difference round numbers of i in the coordinate (i, j) of image, j, and 1≤i≤512,1≤j≤640.
In the present embodiment, image is carried out restoring processing it is generally necessary to be carried out according to certain image degradation model, one
The degenerative process of image can be modeled as one and acted on original image f (x, y) by a simple general image degradation model
Degeneration system H, the synergy of exercising result and a Gaussian noise n (x, y) causes to have produced degraded image g (x, y).
An approximation of original image f (x, y) can be obtained from given degraded image g (x, y) according to above-mentioned degeneration system H.
The method of removal Gaussian noise has a histogram transformation, low-pass filtering, high-pass filtering, liftering, Wiener filtering, median filtering etc.,
In the present embodiment step S2, image degradation model A selects gauss low frequency filter, and expression formula is as follows:
Wherein, A (i, j) indicates degradation model A in the coefficient of coordinate position (i, j), and D (i, j) is coordinate (i, j) away from Fu
The distance of leaf transformation origin, σ1Indicating the degree of Gaussian curve extension, e is natural constant, about 2.71828.
Image degradation model selects gauss low frequency filter, and setting filter size is 5 × 5, standard deviation 0.6, then
Electric-wave filter matrix expression is as shown in table 1:
15 × 5 gauss low frequency filter of table
In step S3, the comentropy and Anisotropic diffusion coefficient of image are calculated separately, it is to be understood that above-mentioned technology
There are diversity for method, are illustrated below with a certain specific equation.
In the step S3, comentropy H (i, j) expression formula of neighborhood P × Q centered on coordinate (i, j) is as follows:
Wherein, the intensity profile probability at the position p (i, j) indicates coordinate (i, j), I (i, j) and I (i+m, j+n) difference table
Diagram as I the position coordinate (i, j) and coordinate (i+m, j+n) at gray value, 1≤m≤P, 1≤n≤Q.
It selects 7*7 neighborhood to calculate comentropy below to be illustrated, i.e. P=7, Q=7, with IinIt is replaced as input picture upper
The image I of formula is stated, then the expression formula after substituting is as follows:
Wherein, 1≤m≤7,1≤n≤7,1≤i≤512,1≤j≤640.
The anisotropic diffusion coefficients G of image is based on following equation and seeks:
I (i, j, t) is the image pixel value at the position moment t coordinate (i, j),It is gradient operator, div is that divergence is calculated
Son,It is partial gradient value, G () indicates the diffusion coefficient function of the partial gradient value,It is that inclined is asked to image I
Operation is led, here, the function expression of anisotropic diffusion coefficientsIt is as follows:
Wherein, σ is Image neighborhood standard deviation, is 0.6, and neighborhood is 5 × 5.
It is understood that any pixel value is there are corresponding comentropy and respectively for described 640 × 512 image
Item anisotropic diffusion coefficient is indicated with H (i, j) and G (i, j) distribution, and therefore, for comentropy, there are maximum value Hmax
With minimum value Hmin, similarly, for anisotropic diffusion coefficients G, there are maximum value GmaxWith minimum value Gmin。
Regularization will mainly solve ill-conditioning problem in restored method.Regularization coefficient λ can gauge signal it is odd
The detailed information of signal can be preferably kept while anisotropic.When regularization coefficient λ increases, high-frequency noise can be effectively suppressed,
However, image detail holding capacity is deteriorated；When regularization coefficient λ reduces, the enhancing of image detail holding capacity.Therefore, according to figure
It is that restoration algorithm is successfully crucial as content-adaptive calculates regularization coefficient.The aggregation of comentropy H characterization image intensity profile
Characteristic, image is more uniform, and comentropy is bigger.Anisotropic diffusion coefficients G characterizes the smoothness of image, and image is more uniform, respectively
Anisotropic diffusion coefficient is smaller.
In order to utmostly remove the singular value generated in recuperation, and the detailed information of image is kept, knot of the present invention
The characteristics of closing comentropy H and anisotropic diffusion coefficients G is expanded in the step S4 according to the comentropy and anisotropy
The regularization coefficient λ (i, j) at any coordinate (i, j) of coefficient calculating is dissipated,
Wherein, a, b are constant coefficients, the specific gravity accounted for when for reflecting comentropy and anisotropic diffusion coefficients load regularization, 0
≤ a≤1,0≤b≤1, a+b=1.It is unlike the embodiments above, according to the ratio of comentropy and anisotropic diffusion coefficients
Weight, being set as a is 0.5, b 0.6, and the regularization coefficient of building is in the high frequency for inhibiting smooth region (regularization coefficient λ increase)
It is positively maintained edge detail information (regularization coefficient λ reduction) while noise.
In step S5, restored image is calculated according to input picture I, image degradation model A and regularization coefficient λ (i, j)
Iout, in following formula, restored image Iout=Ik+1, the image I=I of inputk, formula is as follows:
In order to make restored image closer to real image, above-mentioned formula can also be used as iterative formula and carry out using iteration
Number is generally chosen for 5~20 times, i.e., is original image as k=1, and kth+1 time gray value of image later successively exists
It is calculated in the gray value of image of kth time, wherein Ik+1(i, j) indicates image Ik+1Gray scale at the position coordinate (i, j)
Value, Ik(i, j) indicates image IkGray value at the position coordinate (i, j), U (i, j) indicate regularization factors U at coordinate (i, j)
Regularization value at position selects Tikhonov model to calculate.A indicates image degradation model, and * is convolution algorithm,It is related
Operation, 1≤i≤512,1≤j≤640, k=20 in the present embodiment, is multiplication operation, restored image Iout=I21。
Compared with prior art, the present invention having following remarkable advantage: (1) particular for Low SNR Infrared Images,
Inhibit effectively restore detailed information while noise；(2) according to the comentropy of image local area and anisotropy parameter system
Number adaptive polo placement regularization parameters, have multiple dimensioned restorability, realize that regularization ability in image smoothing area is strong, and image is thin
It is weak to save area's regularization ability；(3) image degradation model can flexibly be constructed according to practical degenerative process；(4) there is no high-orders to transport
Calculation and labyrinth, algorithm operation quantity is small, is easy to hardware real-time implementation.
It is last it is noted that above embodiments invention is explained in detail, the ordinary skill of this field
Personnel are it is understood that it is still possible to modify the technical solutions described in the foregoing embodiments, or to part
Technical characteristic is equivalently replaced；And these are modified or replaceed, it does not separate the essence of the corresponding technical solution, and the present invention is each
The spirit and scope of embodiment technical solution.
Claims (8)
1. one kind stares infrared degraded image adaptive restoration method characterized by comprising
S1, input picture I is obtained；
S2, construction image degradation model A；
S3, the comentropy for calculating the input picture and the anisotropic diffusion coefficients for calculating the input picture；
S4, the regularization coefficient λ (i, j) at any coordinate (i, j) is calculated according to the comentropy and anisotropic diffusion coefficients,
Wherein, H (i, j) is the comentropy at coordinate (i, j), and G (i, j) is the anisotropic diffusion coefficients at coordinate (i, j),
HmaxAnd HminMaximum value and minimum value in respectively comentropy H, GmaxAnd GminIt respectively indicates in anisotropic diffusion coefficients G
Maximum value and minimum value, 0≤a≤1,0≤b≤1, a+b=1；
S5, restored image I is calculated according to input picture I, image degradation model A and regularization coefficient λ (i, j)out；
S6, output restored image Iout。
2. staring infrared degraded image adaptive restoration method as described in claim 1, it is characterised in that: the step S2
In, image degradation model A selects gauss low frequency filter, and expression formula is as follows:
Wherein, A (i, j) indicates degradation model A in the coefficient of coordinate position (i, j), and D (i, j) is that coordinate (i, j) becomes away from Fourier
Change the distance of origin, σ1Indicate the degree of Gaussian curve extension.
3. staring infrared degraded image adaptive restoration method as described in claim 1, it is characterised in that: the step S3
In, comentropy H (i, j) expression formula of neighborhood P × Q centered on coordinate (i, j) is as follows:
Wherein, the intensity profile probability at the position p (i, j) indicates coordinate (i, j), I (i, j) and I (i+m, j+n) respectively indicate figure
As I the position coordinate (i, j) and coordinate (i+m, j+n) at gray value, 1≤m≤P, 1≤n≤Q.
4. staring infrared degraded image adaptive restoration method as described in claim 1, it is characterised in that: the step S3
In, the anisotropic diffusion coefficients of image are based on following equation and seek:
I (i, j, t) is the image pixel value at the position moment t coordinate (i, j),It is gradient operator, div is divergence operator,
It is partial gradient value, G () indicates the diffusion coefficient function of the partial gradient value,It is that derivative operation is asked to image I,
Here, the function expression of anisotropic diffusion coefficientsIt is as follows:
Wherein, σ is Image neighborhood standard deviation.
5. staring infrared degraded image adaptive restoration method as described in claim 1, it is characterised in that: the step S4
In, a and the equal value 0.5 of b.
6. staring infrared degraded image adaptive restoration method as described in claim 1, it is characterised in that: the step S5
In, restored image I is calculated according to input picture I, image degradation model A and regularization coefficient λ (i, j)outShi Caiyong iteration
Method.
7. staring infrared degraded image adaptive restoration method as claimed in claim 6, it is characterised in that: the step S5
In ,+1 iterative image I of kth is sought using iterative methodk+1Expression formula it is as follows:
Wherein, Ik+1(i, j) indicates image Ik+1Gray value at the position coordinate (i, j), Ik(i, j) indicates image IkIn coordinate
Gray value at the position (i, j), U (i, j) indicate regularization value of the regularization factors U at the position coordinate (i, j), and A indicates figure
As degradation model, * is convolution algorithm,It is related operation, is multiplication operation, restored image Iout=Ik+1。
8. as claimed in claims 6 or 7 stare infrared degraded image adaptive restoration method, it is characterised in that: the step
In S5, the number of iterations is 5~20.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201610436012.8A CN106127700B (en) | 2016-06-17 | 2016-06-17 | One kind staring infrared degraded image adaptive restoration method |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201610436012.8A CN106127700B (en) | 2016-06-17 | 2016-06-17 | One kind staring infrared degraded image adaptive restoration method |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN106127700A CN106127700A (en) | 2016-11-16 |
| CN106127700B true CN106127700B (en) | 2019-06-11 |
Family
ID=57470861
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201610436012.8A Active CN106127700B (en) | 2016-06-17 | 2016-06-17 | One kind staring infrared degraded image adaptive restoration method |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN106127700B (en) |
Families Citing this family (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN106993158A (en) * | 2017-04-07 | 2017-07-28 | 湖北大学 | A kind of active infrared night-viewing DAS (Driver Assistant System) based on image restoration |
| CN109903244A (en) * | 2019-02-21 | 2019-06-18 | 北京遥感设备研究所 | A kind of real-time infrared image restored method |
Citations (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN102063716A (en) * | 2011-01-13 | 2011-05-18 | 耿则勋 | Multiframe iteration blind deconvolution image restoration method based on anisotropic constraint |
| CN103606130A (en) * | 2013-10-22 | 2014-02-26 | 中国电子科技集团公司第二十八研究所 | Infrared degraded image adaptive restoration method |
| CN105005975A (en) * | 2015-07-08 | 2015-10-28 | 南京信息工程大学 | Image de-noising method based on anisotropic diffusion of image entropy and PCNN |
-
2016
- 2016-06-17 CN CN201610436012.8A patent/CN106127700B/en active Active
Patent Citations (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN102063716A (en) * | 2011-01-13 | 2011-05-18 | 耿则勋 | Multiframe iteration blind deconvolution image restoration method based on anisotropic constraint |
| CN103606130A (en) * | 2013-10-22 | 2014-02-26 | 中国电子科技集团公司第二十八研究所 | Infrared degraded image adaptive restoration method |
| CN105005975A (en) * | 2015-07-08 | 2015-10-28 | 南京信息工程大学 | Image de-noising method based on anisotropic diffusion of image entropy and PCNN |
Non-Patent Citations (1)
| Title |
|---|
| 自适应光学图像序列配准与复原算法研究;宋向;《中国优秀硕士学位论文全文数据库 信息科技辑》;20120715(第7期);第3.1.2,5.1,5.2节 |
Also Published As
| Publication number | Publication date |
|---|---|
| CN106127700A (en) | 2016-11-16 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN101980282B (en) | Infrared image dynamic detail enhancement method | |
| CN107369134A (en) | A kind of image recovery method of blurred picture | |
| CN106709877B (en) | A kind of image deblurring method based on multi-parameter canonical Optimized model | |
| CN109544487A (en) | A kind of infrared image enhancing method based on convolutional neural networks | |
| CN107016642A (en) | For to there is image of making an uproar to carry out the method for resolution ratio up-regulation and for there is image of making an uproar to carry out the device of resolution ratio up-regulation | |
| CN103369209A (en) | Video noise reduction device and video noise reduction method | |
| CN110070539A (en) | Image quality evaluating method based on comentropy | |
| CN102819827A (en) | Self-adaption moment matching stripe noise removing method based on gray-level segmentation | |
| CN110060219A (en) | One kind being based on low-rank approximately true figure noise-reduction method | |
| CN106204502B (en) | Based on mixing rank L0Regularization fuzzy core estimation method | |
| CN109671035A (en) | A kind of infrared image enhancing method based on histogram | |
| CN113810611B (en) | Data simulation method and device for event camera | |
| CN106204504B (en) | Enhancement method of low-illumination image based on dark channel prior and tone mapping | |
| WO2019010932A1 (en) | Image region selection method and system favorable for fuzzy kernel estimation | |
| CN106127700B (en) | One kind staring infrared degraded image adaptive restoration method | |
| CN102810202A (en) | Image multistep residual feedback iterative filtering method based on fractional order difference weighting | |
| CN110351453A (en) | A kind of computer video data processing method | |
| WO2017088391A1 (en) | Method and apparatus for video denoising and detail enhancement | |
| CN107292844B (en) | Total variation regularization variation stochastic resonance self-adaptive dark image filtering enhancement method | |
| CN105913391A (en) | Defogging method based on shape variable morphological reconstruction | |
| CN103606130A (en) | Infrared degraded image adaptive restoration method | |
| CN115660994B (en) | Image enhancement method based on regional least square estimation | |
| CN114066786A (en) | Infrared and visible light image fusion method based on sparsity and filter | |
| Cui et al. | A modified Richardson–Lucy algorithm for single image with adaptive reference maps | |
| CN112488920B (en) | Image regularization super-resolution reconstruction method based on Gaussian-like fuzzy core |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| C06 | Publication | ||
| PB01 | Publication | ||
| C10 | Entry into substantive examination | ||
| SE01 | Entry into force of request for substantive examination | ||
| TA01 | Transfer of patent application right |
Effective date of registration: 20181107 Address after: 210000 the 5 building of Tianan Digital City, 36 Yongfeng Avenue, Qinhuai District, Nanjing, Jiangsu. Applicant after: Nanjing Lesi Electronic Equipment Co., Ltd. Address before: 210014 1 East Street, alfalfa garden, Bai Xia District, Nanjing, Jiangsu Applicant before: China Electronic Technology Corporation (Group) 28 Institute |
|
| TA01 | Transfer of patent application right | ||
| GR01 | Patent grant | ||
| GR01 | Patent grant |