A kind of image adaptive non-uniform correction method based on scene
Technical field
The present invention relates to the technical field of optical remote sensing imaging, is related specifically to a kind of image adaptive based on scene
Non-uniform correction method.
Background technology
Infrared focal plane detector is the core component of existing infrared imaging or detection system, be widely used in military and
Civil area, is to ensure space flight and aviation, national defense and military, survey of territorial resources, precision agriculture, environmental monitoring, Atmospheric Survey, extreme
The key technology of the field such as hazard forecasting high speed development.With the raising of focus planardetector technological level, infrared focus plane
Scale has been extended to pixels up to a million.But the restriction due to existing manufacturing technology level and material, causes infrared focus plane battle array
Row output amplitude is simultaneously differed, i.e., infrared focal plane array is responded between each pixel when extraneous same homogeneous radiation field is input into
The inconsistency of output, commonly referred to as this inconsistency noise are heterogeneity noise, be embodied in space and make an uproar on image
Sound or fixed pattern noise, have had a strong impact on the image quality of system, reduce the identification application of image definition and image, pole
Big degree limit infrared focal plane array image-forming systematic difference and development.
Nonuniformity correction mainly exactly solves the heterogeneity noise problem caused by explorer response.Non- in prior art
Homogeneity correction is mainly using scaling method and two big class of scene method.Although the Non-uniformity Correction Algorithm complexity based on calibration
Low, be easy to Project Realization, but this kind of algorithm is easily affected by external environment, and pixel response parameter can be over time
Drift about, and calibrate produce nonuniformity correction coefficient cannot adaptive system parameter drift phenomenon, so adopting mostly at present
With the asymmetric correction method based on scene.It is right to be from based on the parameter renewal of the asymmetric correction method of scene
The estimation of scene, it can be good at tracking parameter drift, strong adaptability.But traditional scene method is mainly based upon two-point method pair
Image carries out nonuniformity correction, finds bright dark homogeneous area in the picture by artificial interpretation, and the region needs to cover entirely
The pixel (or selecting several fritter homogeneous areas to cover whole space dimension) of space dimension, more, uniform for ground object detail
The little and scattered complex scene in region is not simultaneously applied to.Especially, it is impossible to meet the fast automatic correction of current large nuber of images.
Therefore, the demand of the quick correction of the large nuber of images of current complex scene to be met, it is necessary to work out a kind of quick
Effective asymmetric correction method, and image quality and the definition of image is can guarantee that, can analyze for successive image and answer
With laying the foundation.
Content of the invention
Based on the presence of the problems referred to above, the present invention proposes a kind of image adaptive non-uniform correction method based on scene,
It is capable of the nonuniformity correction of the large nuber of images of the various complexity atural object scenes of self adaptation, and can solve the problem that each spectral coverage of image is uniform
The inconsistent situation of characteristic, improves image quality and the definition of image, is that successive image analysis and application lay the foundation.
For this purpose, the present invention is employed the following technical solutions:
A kind of image adaptive non-uniform correction method based on scene, by automatic by different pixels to the DN value of image
Sequence, determines at least one relatively dark homogeneous area and bright homogeneous area respectively, is calculated the school of each pixel according to algorithm
Positive coefficient, completes the correction of an image, automatically into the correction of next image, until completing the non-homogeneous school of all images
Just, following steps are specifically included:
1) include image of the pixel number for K to one, the Z DN value that each pixel is obtained is arranged automatically using computer
Sequence, and a range of intermediate value is intercepted for effective DN value;
2) numerical value for taking minimum in the effective DN value of each pixel consists of at least one relatively dark homogeneous area (K × D
)x, maximum numerical value consists of at least one relatively bright homogeneous area (K × L)y;
3) assembly average for calculating each dark homogeneous area is P1x, in region, the average DN value of each pixel is respectively
Q1x(i), i=1,2 ..., K;The assembly average for calculating each bright homogeneous area is P2y, in region, each pixel is average
DN value is respectively Q2y(i), i=1,2 ..., K;
4) build following linear equation, obtain gain correction factor a (i) of each pixel response, i=1,2 ..., K and
Offset correction factor b (i), i=1,2 ..., K:
5) to real response value DN (i, j) at each pixel of the image, i=1,2 ..., K;J=1,2 ..., Z are carried out
Nonuniformity correction, response DN (i, the j)=a (i) after being corrected × DN (i, j)+b (i), i=1,2 ..., K;J=1,
2 ..., Z;
6) F is introduced as the evaluating of Nonuniformity Correction result, and F=S is set for threshold value, as F≤S, corrected
Become;Work as F>During S, then further corrected, until completing the correction of the view data.
7), after completing the correction of this image, automatically into next image, repeat the above steps, until completing all images
Nonuniformity correction.
Preferably, the F=max (fi),And
Setting F=0.5% is threshold value, and when F≤0.5%, correction is completed;Work as F>When 0.5%, then further corrected, until complete
Become the correction of the view data.
Preferably, include the step of the further correction with gain correction factor a (i), i=1,2 ..., K and partially
Shifting amount correction factor b (i), i=1,2, .., K are initial value, fixingConstant, edge respectivelyWithBoth direction minimizes objective optimization object function fi, try to achieve
All nonuniformity correction coefficientsAccording still further to above-mentioned step 5) method complete the picture number
According to correction.
Preferably, number x of the dark homogeneous area of the composition is 1-10, institute in each dark homogeneous area K × D described
DN value number D of each pixel for taking is 30-60;Number y of the bright homogeneous area of the composition is 1-10, and described each is bright
DN value number L of each pixel taken in homogeneous area K × L is 30-60.
Preferably, effective DN value is the 70%-90% for rearranging rear DN value intermediate value of intercepting.
The present invention adopts above technical scheme, by the DN value of image by different pixel auto-sequencings, determine respectively to
A few relatively dark homogeneous area and bright homogeneous area, are calculated the correction coefficient of each pixel, complete image according to algorithm
Nonuniformity correction.It is big that method of the present invention is suitable for complex scene, dark target, the non-homogeneous serious and correction difficulty such as infrared
Large nuber of images nonuniformity correction, significantly simplify complexity and the cost of correction, bearing calibration be based on scene, but break through
The restriction of scene itself, with good versatility and adaptivity, without the need for artificial interpretation, correction work is full-automatic, meter
Calculate speed fast, calibration result is good.
Description of the drawings
Fig. 1 is non-uniform correction method flowchart in the present invention.
Fig. 2 is the short-wave infrared image obtained in onboard flight test in embodiment.
Wherein, (a1-1) is image of the scenario A at wavelength 1000nm;(a2-1) for scenario A at wavelength 1600nm
Image;(b1-1) it is image of the scenario B at wavelength 1000nm;(b2-1) it is image of the scenario B at wavelength 1600nm;
Fig. 3 is to carry out the short-wave infrared image after nonuniformity correction through method in the present invention in embodiment.
Wherein, (a1-2) is image of the scenario A at wavelength 1000nm after correction;(a2-2) for scenario A after correction in ripple
Image at long 1600nm;(b1-2) it is image of the scenario B at wavelength 1000nm after correction;(b2-2) it is scenario B after correction
Image at wavelength 1600nm;
Specific embodiment
In order that objects, features and advantages of the present invention are more clear, below in conjunction with drawings and Examples, to the present invention
Specific embodiment make more detailed description, in the following description, elaborate a lot of concrete details in order to fill
The understanding present invention for dividing, but the present invention can be implemented with a lot of other modes for being different from description.Therefore, the present invention is not received
The restriction being embodied as of following discloses.
A kind of image adaptive non-uniform correction method based on scene, as shown in figure 1, by the DN value of image by not
With pixel auto-sequencing, at least one relatively dark homogeneous area and bright homogeneous area is determined respectively, be calculated according to algorithm every
The correction coefficient of individual pixel, completes the correction of an image, automatically into the correction of next image, until completing all images
Nonuniformity correction.Specifically include following steps
Embodiment one
1) include image of the pixel number for K to one, the Z DN value that each pixel is obtained is arranged automatically using computer
Sequence, and a range of intermediate value is intercepted for effective DN value;
2) D numerical value for taking minimum in the effective DN value of each pixel constitutes one relatively dark homogeneous area (K × D)1, most
L big numerical value constitutes one relatively bright homogeneous area (K × L)1;
3) assembly average for calculating dark homogeneous area is P11, in region, the average DN value of each pixel is respectively Q11
(i), i=1,2 ..., K;The assembly average for calculating each bright homogeneous area is P21, the average DN of each pixel in region
Value is respectively Q21(i), i=1,2 ..., K;
4) build following linear equation, obtain gain correction factor a (i) of each pixel response, i=1,2 ..., K and
Offset correction factor b (i), i=1,2 ..., K:
5) to real response value DN (i, j) at each pixel of the image, i=1,2 ..., K;J=1,2 ..., Z are carried out
Nonuniformity correction, the response after being corrected
6) F is introduced as the evaluating of Nonuniformity Correction result, and F=S is set for threshold value, as F≤S, corrected
Become;Work as F>During S, then further corrected, until completing the correction of the view data.
7), after completing the correction of this image, automatically into next image, repeat the above steps, until completing all images
Nonuniformity correction.
Preferably, the F=max (fi),And arrange
F=0.5% is threshold value, and when F≤0.5%, correction is completed;Work as F>When 0.5%, then further corrected, until completing this
The correction of view data.
Preferably, include the step of the further correction with gain correction factor a (i), i=1,2 ..., K and partially
Shifting amount correction factor b (i), i=1,2 ..., K is initial value, fixingWithConstant, edge respectivelyWithBoth direction minimizes objective optimization object function fi, try to achieve institute
There is nonuniformity correction coefficientAgain by above-mentioned step 5) method complete the school of the view data
Just.
Preferably, DN value number D of each pixel for being taken in each dark homogeneous area K × D described is 30-60;Institute
DN value number L for stating each pixel taken in each bright homogeneous area K × L is 30-60.
Preferably, effective DN value is the 70%-90% for rearranging rear DN value intermediate value of intercepting.
Embodiment two
1) include image of the pixel number for K to one, the Z DN value that each pixel is obtained is arranged automatically using computer
Sequence, and a range of intermediate value is intercepted for effective DN value;
2) D numerical value for taking minimum in the effective DN value of each pixel constitutes two relatively dark homogeneous area (K × D)1、(K
×D)2L maximum numerical value constitutes one relatively bright homogeneous area (K × L)1、(K×L)2;
3) assembly average for calculating each dark homogeneous area is P11、P12, the average DN value of each pixel point in region
Wei not Q11(i), i=1,2 ..., K, Q12(i), i=1,2 ..., K;The assembly average for calculating each bright homogeneous area is
P21、P22, in region, the average DN value of each pixel is respectively Q21(i), i=1,2 ..., K, Q22(i), i=1,2 ..., K;
4) build following linear equation, obtain gain correction factor a (i) of each pixel response, i=1,2 ..., K and
Offset correction factor b (i), i=1,2 ..., K:
5) to real response value DN (i, j) at each pixel of the image, i=1,2 ..., K;J=1,2 ..., Z are carried out
Nonuniformity correction, the response after being corrected
6) F is introduced as the evaluating of Nonuniformity Correction result, and F=S is set for threshold value, as F≤S, corrected
Become;Work as F>During S, then further corrected, until completing the correction of the view data.
7), after completing the correction of this image, automatically into next image, repeat the above steps, until completing all images
Nonuniformity correction.
Preferably, the F=max (fi),And arrange
F=0.5% is threshold value, and when F≤0.5%, correction is completed;Work as F>When 0.5%, then further corrected, until completing this
The correction of view data.
Preferably, described include the step of further corrected with gain correction factor a (i), i=1,2 ..., K
With offset correction factor b (i), i=1,2 ..., K is initial value, fixingConstant, edge respectivelyBoth direction minimizes objective optimization object function fi, try to achieve
All nonuniformity correction coefficientsAccording still further to above-mentioned step 5) method complete the view data
Correction.
Preferably, DN value number D of each pixel for being taken in each dark homogeneous area K × D described is 30-60;Institute
DN value number L for stating each pixel taken in each bright homogeneous area K × L is 30-60.
Preferably, effective DN value is the 70%-90% for rearranging rear DN value intermediate value of intercepting.
Below by taking the short-wave infrared image obtained in onboard flight experiment as an example, scenario A and scenario B are in wavelength 1000nm
With the image obtained at 1600nm, as shown in Fig. 2 pressing method of the present invention respectively, nonuniformity correction is carried out.The image
Pixel number is 320, and another space dimension is pushed away by 1000 sweeps row and constitute, and effective spectrum channel 181 covers 900nm to 1700nm
Spectral region.
The detailed process of nonuniformity correction is as follows:
1) to one including the image that pixel number is 320,1000 automatically each pixel obtained using computerized algorithm
Individual DN value is rearranged according to order from small to large, and intercepts a range of intermediate value for effective DN value;
2) minimum 40 numerical value in the effective DN value of each pixel are taken respectively dark homogeneous area 320 × 40 is consisted of, maximum
40 numerical value consist of bright homogeneous area 320 × 40;
3) assembly average for calculating dark homogeneous area is P1, in region, the average DN value of each pixel is respectively Q1(i),i
=1,2 ..., 320;The assembly average for calculating bright homogeneous area is P2, in region, the average DN value of each pixel is respectively Q2
(i), i=1,2 ..., 320;
4) following linear equation is built, obtains gain correction factor a (i) i of each pixel response ,=1,2 ..., 320
With offset correction factor b (i), i=1,2 ..., 320:
5) to real response value DN (i, j) at each pixel of the image, i=1,2 ..., 320;J=1,2 ...,
1000 carry out nonuniformity correction, the response after being corrected
6) F is introduced as the evaluating of Nonuniformity Correction result, F=max (fi),And F=0.5% is set for threshold value, when F≤0.5%, correct
Complete;Work as F>When 0.5%, then further corrected, until completing the correction of a view data.Wherein, enter traveling one
The step of correction of step includes with gain correction factor a (i), i=1,2 ..., 320 and offset correction factor b (i), i=1,
2 ..., 320 is initial value, and fixing a (160) and b (160) are constant, respectively along i=159,158 ..., 1 and i=161,
162 ..., 320 both directions minimize objective optimization object function fi, try to achieve all nonuniformity correction coefficientsAccording still further to procedures described above 5) complete the correction of the data.
7), after completing the correction of this image, automatically into next spectral coverage, repeat the above steps, own until completing image
The nonuniformity correction of spectral coverage.
Short-wave infrared image after method nonuniformity correction of the present invention, as shown in Figure 3, it can be seen that image
Heterogeneity obtained obvious improvement, Banded improvement phenomenon is also eliminated, thus using method of the present invention to short
Ripple infrared image calibration result is notable, can substantially improve the image quality of image, when particularly comprising weak signal scenes such as water,
Calibration result is more notable.
Presently preferred embodiments of the present invention is the foregoing is only, not in order to limit the present invention, all essences in the present invention
Any modification, equivalent and improvement that is made within god and principle etc., should be included within the scope of the present invention.