CN104065853A - Infrared camera crosstalk eliminating method - Google Patents

Abstract

An infrared camera crosstalk eliminating method is a method for utilizing image information of adjacent infrared spectrum bands to analyze and position crosstalk areas of images, performing characteristic construction on different pixel elements in the crosstalk areas of the infrared images and finishing crosstalk elimination according to constructed characteristics. The method comprises the steps of 1 obtaining initial data and relevant initialization operation; 2 finishing coarse positioning and fine positioning of the crosstalk areas of the images through a crosstalk positioning module; 3 obtaining a characteristic assembly for crosstalk elimination through an establishment module of a crosstalk model; 4 using a characteristic construction result to perform crosstalk elimination on an embodiment through a crosstalk elimination module and outputting a result. The infrared camera crosstalk eliminating method has the advantages of having fidelity, namely processing the crosstalk areas of the images in a targeted mode and effectively keeping truth of the original data and being low in cost and simple in operation compared with a hardware design means.

Description

A kind of infrared camera crosstalk eliminating method

Technical field

The present invention relates to a kind of infrared camera crosstalk eliminating method, specifically a kind of technology of eliminating infrared remote sensing picture crosstalk, belongs to infrared remote sensing image processing field.

Background technology

Along with spectral investigation deeply and the development of technology, require the broadband detection channels of transducer to develop to narrow wave band or segmentation wavelength band technique.Since 1970, China has developed polytype general and general aviation scanner, and its service band comprises that ultraviolet, visible ray are to infrared.In the air remote sensing application in the fields such as resource, environment, ocean, disaster, obtain a large amount of useful object spectrum information.Wherein, thermal infrared remote sensing has great importance to studying global energy conversion and sustainable development.Especially in field of ecology, by ground measured data and remotely-sensed data, by the parsing of infrared band, the quantification that inverting can be carried out variety of issue.Infrared remote sensing technology set the last word of the science such as space, electronics, optics, computer, biology and ground, be the important component part in modern high technology field.Meanwhile, it for human knowledge territory, exploit natural resources, monitoring of environmental, research disaster and analyze Global climate change etc. new approach is provided.

Early stage Thermal Remote Sensing Image is all single band images of broad band.Because image spatial resolution is lower, more application is in aspects such as geothermal resources investigation, survey for the purpose of locating hydrogeological resources and the volcanoes and earthquakes forecasts in large region.Along with the development of high light spectrum image-forming spectral technique, thermal infrared is many-and Gao spectrum scanner starts to drop into remote sensing application, utilize multichannel sensor, the electromagnetic wave of ground object radiation is divided into several narrower spectral coverage bands (or wave spectrum) and carries out synchronous scanning, obtain the image feature of the different spectral coverage of same atural object, thereby obtain a large amount of information.

But, the Performance and quality of Optical remote satellite to a great extent by they payload---optical sensor is determined.Generally, in IRMSS, different spectral coverage remote-sensing detector is arranged in same substrate, due to reasons such as detector reading circuits, may cause adjacent spectral coverage imaging time, produce and crosstalk each other, this crosstalking can affect image quality, affect follow-up remotely-sensed data analysis, utilize situation.This form of expression of crosstalking is that adjacent spectral coverage (is assumed to be f ₁spectral coverage and f ₂spectral coverage) when point view field imaging, at f ₂when spectral coverage detector receives image formation, f ₁spectral coverage does not now have signal to enter its visual field, but in this spectral coverage output image, relevant position also has signal and occurs.Equally, at f ₁when spectral coverage detector receives image formation, f ₂spectral coverage correspondence position also can be subject to similar impact.Observe the image that imaging obtains, crosstalk and mainly occur in the overlapping region of adjacent spectral coverage light and shade, but not entire image existence is crosstalked.

At present, be mainly divided into two classes about the removing method of crosstalk phenomenon: the method for the improvement to detector itself and employing Digital Image Processing is carried out reprocessing to image.The former cost and power consumption will increase widely, and mostly the latter is the non-universality method proposing towards special object.Temporarily there is no the special cross-interference issue with solving infrared camera of a kind of effective method.

Summary of the invention

The problem existing for above-mentioned prior art and and eliminate the requirement that adjacent infrared camera is crosstalked, the invention provides a kind of infrared camera crosstalk eliminate digital image processing method, specifically be to utilize the image information of adjacent infrared spectral coverage to eliminate cross talk effects each other, thereby reach the object that obtains clear remote sensing images.The form of expression that this method is crosstalked for infrared camera, analysis crosstalk occur principle, utilize Mathematical Morphology Method and device parameter accurately to locate the region of crosstalking and existing, for the feature of image of crosstalk zone, set up adaptive cross-talk models, and adopt the mode of learning, and ask for the parameter of cross-talk models, finally adopt the means elimination that image recovers to crosstalk.This solution is only processed for crosstalk zone, has well kept the authenticity of initial data, when effectively removal is crosstalked, has improved the visual effect of image.Than the solution route of hardware designs, this scheme has advantages of simply effectively, cost is low.

To achieve these goals, the technical solution used in the present invention is: a kind of infrared camera crosstalk eliminating method, mainly comprises: cross-talk models and elimination three modules such as crosstalk are crosstalked, set up in location.The location module of crosstalking comprises crosstalk zone coarse positioning and two steps of fine positioning, and the pixel that accurately crosstalked in location avoids the view data authenticity of being crosstalked destroyed.Setting up cross-talk models module is that the feature differing according to crosstalk zone diverse location degree of crosstalk completes.Train with the data of adjacent spectral coverage correspondence position by the view data of extracting in crosstalk zone, thus the parameter of the elimination model that obtains crosstalking.The elimination module of crosstalking utilizes cross-talk models to calculate the concrete crossfire value of certain pixel, then it is cut down.

Method flow involved in the present invention comprises the following steps: (1) obtains primary data and relevant initialization operation; (2) the crosstalk zone coarse positioning based on Mathematical Morphology Method; (3) utilize the location parameter of the adjacent spectral coverage detector of infrared camera to carry out crosstalk zone fine positioning; (4) set up cross-talk models; 5), by the view data substitution cross-talk models of crosstalk zone, obtain the concrete numerical value of model parameter by training; (6) travel through whole crosstalk zone, eliminate and crosstalk and Output rusults.

Below the each step of the method flow process is elaborated.

Step 1, obtains primary data and relevant initialization operation;

If initial data is not common picture format, will complete the conversion of data format in this step, the data after conversion are read in treatment system, in order to subsequent treatment; Suppose signal f ₁and f ₂be respectively pending adjacent spectral coverage infrared remote sensing image, image array line number is m, and columns is n, and claims that atural object brighter with respect to adjacent spectral coverage in image is target;

Step 2, the crosstalk zone coarse positioning based on Mathematical Morphology Method;

Employing Mathematical Morphology Gradient comes the edge of target in positioning image, given input signal f _i, i=1,2 and structural element s, the morphology inward flange of image:

g _1i＝f _i-(f _i⊙s),i＝1,2 (1)

Wherein, (f _i⊙ s) represents image f _imorphological erosion operation;

In image, straddle the Morphological Gradient on object boundary:

g _2i＝(f _i⊕s) _i-(f _i⊙s),i＝1,2 (2)

Wherein, (f _i⊕ s) _irepresent image f _imorphological dilation;

To g _1iadopt Otsu threshold to carry out binaryzation, just can obtain the inner edge marginal zone of target, be denoted as b _1i, i=1,2; Similarly, to g _2iadopt Otsu threshold to carry out binaryzation, just can obtain straddling the marginal belt on object boundary, be denoted as b _2i, i=1,2;

Because the marginal belt of trying to achieve has certain width, so obtain crosstalk zone edge more accurately by the mode that marginal belt and the inner edge marginal zone of target subtract each other, obtain image f ₁and f ₂the border, left and right of crosstalk zone be

R ₁＝b ₂₂-b ₁₁(3)

R ₂＝b ₂₁-b ₁₂(4)

It should be noted that, what in fact R_1 and R_2 comprised is the border, left and right of several crosstalk zone.Count from left to right, odd number time occur numerical value 1 formation the left margin of crosstalk zone, even number occur numerical value 1 formation the right margin of crosstalk zone; Adjacent border, left and right envelope a crosstalk zone.

Step 3, utilizes the location parameter of the adjacent spectral coverage detector of infrared camera to carry out fine positioning;

Through the processing of step 2, the basic fixed position of crosstalk zone position; For more accurate, we utilize detector to improve positioning precision at suprabasil location parameter; Because adjacent spectral coverage detector shows as the coordinate horizontal direction dislocation p pixel of same atural object in two spectral coverage images in suprabasil dislocation distance map in image, do not consider the dislocation of vertical direction; Without loss of generality, suppose that same object point is at f ₁in image coordinate be (i, j), at f ₂in coordinate be (i, j+p); Utilizing location parameter to carry out pinpoint principle is atural object that amplitude the is larger width (the maximum horizontal number of pixels that this atural object is crossed over) in image while being more than or equal to p pixel, will cause that width is the crosstalk zone of p at adjacent spectral coverage; Otherwise atural object width is much, crosstalk zone width is much; So corresponding target width and the p of every crosstalk zone that step 2 is located compares one by one, determine the border set R ' of final crosstalk zone ₁and R ' ₂;

Step 4, sets up cross-talk models;

Because same atural object is in the infrared lower imaging of different spectral coverage, radiation intensity is different, and without loss of generality, we suppose that same atural object is at f ₁in gray value be less than at f ₂in gray value, f ₂at f ₁in cause crosstalk and will here come into question, otherwise analogize.The feature being raise by the crosstalk physical cause producing and the pixel amplitude of crosstalking, we are to f ₂in single pixel to f ₁middle pixel produces the situation of crosstalking and carries out modeling, specific as follows:

We think to crosstalk and present linear attenuation characteristic, and, along with coordinate y value increases, the degree of crosstalk that single pixel causes reduces gradually, until be zero.We only consider the cross talk effects (because detector is by line scanning imaging) of horizontal direction, so f ₂in the pixel of (i, j) position to f ₁in (i, j) position pixel produce crosstalk for

△x _i,j＝ax _i,j+b (5)

Wherein, a and b are f ₂right linear effect coefficient;

Influence value to (i, j) position horizontal direction later pixel point is:

△x _i,j+k＝c ₁△x _i,j+c ₂,c ₂≤0,k＝1,2,…,K (6)

Wherein, △ x _i,j>=0, x _i,jfor f ₂spectral coverage is not subject to f ₁the gradation of image value of spectral coverage cross talk effects; c ₁for attenuation coefficient, c ₂for rate of decay, K causes the pixel maximum number of crosstalking;

From another angle, f ₁the pixel of (i, the j) position in crosstalk zone is subject to f ₂before middle correspondence position and horizontal direction, N pixel is caused crosstalks, therefore f ₁in crosstalk zone, (i, j) position pixel is subject to total crosstalking:

${\mathrm{\ΔX}}_{i,j}={\mathrm{ax}}_{i,j}+b+\underset{n=1}{\overset{N}{\mathrm{\Σ}}}(c_1{\mathrm{\Δx}}_{i,j-n}+c_2)---\left(7\right)$

Wherein, △ x _{i, j-n}represent f ₂in the crossfire value that causes of the pixel of (i, j-n) position, △ X _i,jf ₁in the total value of crosstalking that is subject to of (i, j) position pixel, x _i,jfor f ₂in the actual value of (i, j) position, N causes the maximum pixel number of crosstalking in (i, j) position;

Thus, image f ₁in the grey scale pixel value eliminated after crosstalking of (i, j) position be:

${f_1}^{\mathrm{new}}(i,j)=f_1(i,j)-{\mathrm{\ΔX}}_{i,j}---\left(8\right)$

Wherein, f ₁(i, j) is f ₁in the measured value of (i, j) position pixel, for the gray value of the elimination of (i, j) position after crosstalking;

Step 5, by the view data substitution cross-talk models of crosstalk zone, obtains the concrete numerical value of model parameter by training;

Step 4 has completed the foundation of cross-talk models, obtain model formation accurately, must obtain a, b, c ₁, c ₂four coefficients, and the situation that the value of K and N is crosstalked with corresponding crosstalk zone is relevant, value adaptively.By f ₁view data in crosstalk zone and f ₂the image data extraction of middle correspondence position out, as the input of model training, learns to obtain coefficient a, b, c by multisample ₁, c ₂, and the maximum occurrences of default N is 5, i.e. f ₁in certain pixel be subject to f ₂in be less than or equal to crosstalking that 5 pixels cause jointly;

Step 6, travels through whole crosstalk zone, eliminates and crosstalks and Output rusults.

Utilize formula (8) to f ₁each pixel of crosstalk zone is eliminated and is crosstalked, and Output rusults.Similar with step 4 and step 5, to f ₂carry out relevant treatment, set up definite f ₂the elimination model of crosstalking, then to f ₂in crosstalk to eliminate and obtain f ₂high-quality image.In order to make treatment effect better, capable of regulating model parameter.

The invention has the beneficial effects as follows: utilize the linear attenuation characteristic of crosstalking, set up adaptive cross-talk models, only process for crosstalk zone, when elimination is crosstalked, effectively kept the authenticity of initial data, and implementation procedure is simple.

Brief description of the drawings

Fig. 1 is infrared camera crosstalk eliminating method flow chart.

Fig. 2 is example crosstalk zone fine positioning effect two width sectional drawings.

Fig. 3 is and example f a) ₁corresponding original image; B) be with example in f ₂corresponding original image; C) be f ₁the result images of crosstalking after eliminating; D) left side is f ₁the sectional drawing of original image, right side is f ₁result image corresponding part sectional drawing.

Embodiment

Below in conjunction with example, application process of the present invention is described further.

The present invention is based on digital image processing method and proposed a kind of infrared multispectral remote sensing picture crosstalk removing method, mainly comprise: the location of crosstalking, the foundation of cross-talk models and three modules of elimination of crosstalking.

The first, obtain primary data and relevant initialization operation.

This example employing in October, 2012 infrared camera carries out at Space City (Beijing) the outdoor scene view data that outdoor scene imaging test obtains.Real data spectral coverage f ₁wavelength be 1.55 ± 0.02 μ m～1.75 ± 0.02 μ m, spectral coverage f ₂wavelength be 2.08 ± 0.03 μ m～2.35 ± 0.03 μ m.The initial data obtaining is carried out to the geometric correction of vertical direction, be then converted into .GIF form and be read in treatment system, image array size is 1024 × 3012.Without loss of generality, we discuss f ₂to f ₁what produce crosstalks.

The second, the crosstalk zone coarse positioning based on Mathematical Morphology Method;

Computation of morphology inward flange:

g _1i(x,y)＝f _i(x,y)-(f _i⊙s)(x,y),i＝1,2 (9)

x＝1,2,...,1024；y＝1,2,...,3012.

Calculating straddles the Morphological Gradient on object boundary:

g _2i(x,y)＝(f _i⊕s)(x,y)-(f _i⊙s)(x,y),i＝1,2 (10)

x＝1,2,...,1024；y＝1,2,...,3012.

Adopt Otsu threshold to entire image g _1iand g _2icarry out binaryzation, obtain image b _1iand b _2i, i=1,2.

Computed image f ₁the border, left and right of crosstalk zone is

R ₁＝b ₂₂-b ₁₁(11)

In this example, image R ₁in left margin and the image f of each crosstalk zone ₁the real border of middle target is consistent, and error occurs in right margin.This will carry out precision improvement by step 3.

The 3rd, utilize the location parameter of adjacent spectral coverage detector in infrared multiple spectrum scanner to carry out crosstalk zone fine positioning;

For example, statistical picture R ₁in the number of pixels n of every two 1 midfeathers while occurring of every row, and and f ₁, f ₂in image, atural object horizontal direction staggered pixels number k carries out size relatively.If n<k, extends to the right margin of this crosstalk zone to the right with left margin and differs k+5 pixel (by foregoing hypothesis, single pixel is subject at most the impact of front 5 pixels); Otherwise, do not change the border of crosstalk zone.To R ₁judge line by line, upgrade, obtain final crosstalk zone boundary image R ' ₁.

The 4th, the view data of extraction crosstalk zone is crosstalked and is eliminated the parameter training of model, determines the elimination model of crosstalking;

By R ' ₁every row in every two 1 occur between pixel all assignment be 1.With f ₁image array dot product, just can extract the view data of all crosstalk zone, and deposits data set s in ₁.With f ₂image array dot product, just can extract and cause the view data of crosstalking, and deposits data set s in ₂.By s ₁and s ₂import in cross-talk models and train, set rate of decay c ₂=-2 obtain coefficient a=0.06, b=6, c ₁=2.8.

The 5th, according to the crosstalking in degradation model removal of images of crosstalking of determining, Output rusults.

Can be obtained by foregoing, at f ₁middle coordinate position is that total crossfire value that the pixel of (i, j) is subject to is

${\mathrm{\&Delta;X}}_{i,j}=0.06x_{i,j}+6+\underset{n=1}{\overset{N}{\mathrm{\&Sigma;}}}(2.8{\mathrm{\&Delta;x}}_{i,j-n}-2)---\left(12\right)$

Wherein x _i,jrepresent f ₂in the pixel amplitude of (i, j) position; △ x _{i, j-n}represent f ₂in (i, j-n) position pixel at f ₂in (i, the j-n) size of crosstalking that causes.

So the each pixel (i, j) in traversal crosstalk zone, calculates

${f_1}^{\mathrm{new}}(i,j)=f_1(i,j)-{\mathrm{\&Delta;X}}_{i,j}---\left(12\right)$

The image being just eliminated after crosstalking

With same step to f ₂process, can be eliminated after crosstalking will with write as image .GIF form, output is preserved.

Claims (1)

1. an infrared camera crosstalk eliminating method, it is characterized in that: by location crosstalk zone, the data of extracting in crosstalk zone are used for training, set up adaptive cross-talk models, elimination is crosstalked and is obtained image clearly, have location and crosstalk, set up cross-talk models and elimination three functional modules of crosstalking, its step is as follows:

Step 1, obtains primary data and relevant initialization operation;

If initial data is not common picture format, will complete the conversion of data format in this step, the data after conversion are read in treatment system, in order to subsequent treatment; Suppose signal f ₁and f ₂be respectively pending adjacent spectral coverage infrared remote sensing image, image array line number is m, and columns is n, and claims that atural object brighter with respect to adjacent spectral coverage in image is target;

Step 2, the crosstalk zone coarse positioning based on Mathematical Morphology Method;

Employing Mathematical Morphology Gradient comes the edge of target in positioning image, given input signal f _i, i=1,2 and structural element s, the morphology inward flange of image:

g _1i＝f _i-(f _i⊙s),i＝1,2 (1)

Wherein, (f _i⊙ s) represents image f _imorphological erosion operation;

In image, straddle the Morphological Gradient on object boundary:

g _2i＝(f _i⊕s)-(f _i⊙s),i＝1,2 (2)

Wherein, (f _i⊕ s) represents image f _imorphological dilation;

To g _1iadopt Otsu threshold to carry out binaryzation, just can obtain the inner edge marginal zone of target, be denoted as b _1i, i=1,2; Similarly, to g _2iadopt Otsu threshold to carry out binaryzation, just can obtain straddling the marginal belt on object boundary, be denoted as b _2i, i=1,2;

Because the marginal belt of trying to achieve has certain width, so obtain crosstalk zone edge more accurately by the mode that marginal belt and the inner edge marginal zone of target subtract each other, obtain image f ₁and f ₂the border, left and right of crosstalk zone be

R ₁＝b ₂₂-b ₁₁(3)

R ₂＝b ₂₁-b ₁₂(4)

Step 3, utilizes the location parameter of the adjacent spectral coverage detector of infrared camera to carry out fine positioning;

Through the processing of step 2, the basic fixed position of crosstalk zone position; For more accurate, we utilize detector to improve positioning precision at suprabasil location parameter; Because adjacent spectral coverage detector shows as the coordinate horizontal direction dislocation p pixel of same atural object in two spectral coverage images in suprabasil dislocation distance map in image, do not consider the dislocation of vertical direction; Without loss of generality, suppose that same object point is at f ₁in image coordinate be (i, j), at f ₂in coordinate be (i, j+p); Utilizing location parameter to carry out pinpoint principle is atural object that amplitude the is larger width (the maximum horizontal number of pixels that this atural object is crossed over) in image while being more than or equal to p pixel, will cause that width is the crosstalk zone of p at adjacent spectral coverage; Otherwise atural object width is much, crosstalk zone width is much; So corresponding target width and the p of every crosstalk zone that step 2 is located compares one by one, determine the border set R ' of final crosstalk zone ₁and R ' ₂;

Step 4, sets up cross-talk models;

Suppose that same atural object is at f ₁in gray value be less than at f ₂in gray value, f ₂in the pixel of (i, j) position to f ₁in (i, j) position pixel produce crosstalk for

△x _i,j＝ax _i,j+b (5)

Wherein, a and b are f ₂right linear effect coefficient;

Influence value to (i, j) position horizontal direction later pixel point is:

△x _i,j+k＝c ₁△x _i,j+c ₂,c ₂≤0,k＝1,2,…,K (6)

Wherein, △ x _i,j>=0, x _i,jfor f ₂spectral coverage is not subject to f ₁the gradation of image value of spectral coverage cross talk effects; c ₁for attenuation coefficient, c ₂for rate of decay, K causes the pixel maximum number of crosstalking;

From another angle, f ₁the pixel of (i, the j) position in crosstalk zone is subject to f ₂before middle correspondence position and horizontal direction, N pixel is caused crosstalks, therefore f ₁in crosstalk zone, (i, j) position pixel is subject to total crosstalking:

Wherein, △ x _{i, j-n}represent f ₂in the crossfire value that causes of the pixel of (i, j-n) position, △ X _i,jf ₁in the total value of crosstalking that is subject to of (i, j) position pixel, x _i,jfor f ₂in the actual value of (i, j) position, N causes the maximum pixel number of crosstalking in (i, j) position;

Thus, image f ₁in the grey scale pixel value eliminated after crosstalking of (i, j) position be:

Wherein, f ₁(i, j) is f ₁in the measured value of (i, j) position pixel, for the gray value of the elimination of (i, j) position after crosstalking;

Step 5, by the view data substitution cross-talk models of crosstalk zone, obtains the concrete numerical value of model parameter by training;

By f ₁view data in crosstalk zone and f ₂the image data extraction of middle correspondence position out, as the input of model training, learns to obtain coefficient a, b, c by multisample ₁, c ₂, and the maximum occurrences of default N, i.e. f ₁in certain pixel be subject to f ₂in be less than or equal to crosstalking that N pixel cause jointly;

Step 6, travels through whole crosstalk zone, eliminates and crosstalks and Output rusults.