CN105069757A - Bidirectional iteration bilateral filtering method for asphalt images obtained through UAV-borne infrared imaging device - Google Patents

Abstract

The invention discloses a bidirectional iteration bilateral filtering method for asphalt images obtained through an UAV-borne infrared imaging device. The method includes a first step of taking original infrared asphalt images by utilizing an unmanned aerial vehicle (UAV), a second step of switching original infrared asphalt images in RGB color space to a Lab color space, a third step of expanding the image boundaries by using the boundary pixels of the infrared asphalt images in the Lab color space obtained in the second step, a forth step of conducting lateral filtering on the images obtained in the third step, a fifth step of conducting vertical filtering on the images obtained in the fourth step, and a sixth step of switching images in the Lab color space processed in the step 5 to the RGB color space and outputting filtered infrared asphalt images. Lateral and vertical bi-directional iteration bilateral filtering means is employed, infrared asphalt image noises can be rapidly eliminated, and the edge detail can be retained. The infrared asphalt image filtering effect is satisfactory, and edge information of images can be retained.

Description

The bidirectional iteration bilateral filtering method of UAV system infrared acquisition pitch image

Technical field

The invention belongs to technical field of image processing, be specifically related to the bidirectional iteration bilateral filtering method of a kind of UAV system infrared acquisition pitch image.

Background technology

In recent years, along with the development of unmanned air vehicle technique, unmanned plane is widely used in productive life, as the erection of line of electric force, and freight transportation, and environment detecting.Unmanned plane relies on it light, and cheap advantage steps into all trades and professions.Because bituminous mixture laying temperature is too high, general device is difficult to take the infrared pitch image obtaining better angle usually, to this, sets up infrared imaging device carry out image acquisition aloft by unmanned plane, can obtain easy-to-handle image in economic security ground.

At present, what infrared thermal imaging detection technique was widely used in bituminous pavement builds in engineering, in bituminous pavement process of deployment, obtains the bituminous pavement image of the condition of high temperature, and according to the temperature variation of infrared pitch image Obtaining Accurate asphalt, road pavement construction quality is very important.But in practice of construction, due to the vibration of image acquisition, the factor impacts such as ambient temperature, often make image there is much noise.These noises can make image degradation, and show as image blurring, feature is flooded, and obtain second-rate image, and it is unfavorable to produce the analysis of image, are difficult to the temperature accurately obtaining bituminous pavement thus.

Traditional filtering algorithm, as although mean filter, medium filtering, Wiener filtering can have good filter effect for different noises, but these only can not retain the high-frequency information of image effectively based on the wave filter of spatial information, so that a large amount of loss in detail of filtered soft edge.

For the problems referred to above, 1998, C.Tomasi and R.Manduchi proposed a kind of non-iterative simple strategy, is called bilateral filtering.Here it is so-called traditional bilateral filtering.It not only takes the strategy of traditional filtering method, considers spatial positional information, also adds the impact of codomain similarity, and therefore, bilateral filtering not only can removal of images noise but also remain image edge information.Although bilateral filtering has above-mentioned advantage, but there is inefficiency in it, the shortcoming that filter effect is not good, in order to boostfiltering effect, mask radius should be made enough large, but, when mask radius is excessive, its computing time is unacceptable, so bilateral filtering is difficult to accomplish real-time process, does not meet the application of some occasion.

Summary of the invention

The object of the present invention is to provide the bidirectional iteration bilateral filtering method of a kind of UAV system infrared acquisition pitch image, to overcome the defect that above-mentioned prior art exists, the present invention adopts the iteration bilateral filtering of horizontal stroke, vertical both direction, infrared pitch picture noise can be eliminated fast and keep edge details, both the filter effect of infrared pitch image had been met, maintain again the marginal information of image preferably, significantly reduce the complexity of traditional bilateral filtering simultaneously.

For achieving the above object, the present invention adopts following technical scheme:

The bidirectional iteration bilateral filtering method of UAV system infrared acquisition pitch image, comprises the following steps:

Step 1: utilize unmanned plane to take and obtain original infrared pitch image;

Step 2: the original infrared pitch image of RGB color space step 1 photographed is transformed into Lab color space;

Step 3: the boundary pixel expanded images border of the infrared pitch image of the Lab color space utilizing step 2 to obtain;

Step 4: the horizontal filtering process of image that step 3 is obtained;

Step 5: the longitudinal filtering process of the image after step 4 is processed;

Step 6: the image of the Lab color space after step 5 being processed is transformed into RGB color space, the infrared pitch image after output filtering.

Further, the original infrared pitch image size in step 1 is 544*486.

Further, in step 2, the original infrared pitch image of RGB color space is transformed into Lab color space, original infrared pitch image is L, a, b tri-components by R, G, B color transition, wherein L represents the brightness of image, a represents the scope from redness to green, and b represents the scope from yellow to blueness.

Further, in step 3 by the infrared pitch image of Lab color space according to expansion radius size be 5, with specular way expanded images borderline region.

Further, the horizontal filtering process of image that following formula obtains step 3 is adopted in step 4:

$\left\{\begin{array}{c}y{\left(L\right)}_{i,j}=\left(1-R_{i,j-1}\·c\right)\·x{\left(L\right)}_{i,j}+R_{i,j-1}\·c\·y{\left(L\right)}_{i,j-1}\\ y{\left(a\right)}_{i,j}=\left(1-R_{i,j-1}\·c\right)\·x{\left(a\right)}_{i,j}+R_{i,j-1}\·c\·y{\left(a\right)}_{i,j-1}\\ y{\left(b\right)}_{i,j}=\left(1-R_{i,j-1}\·c\right)\·x{\left(b\right)}_{i,j}+R_{i,j-1}\·c\·y{\left(b\right)}_{i,j-1}\end{array}\right.$

Wherein, R _{i, j-1}for horizontal neighbor coefficient of similarity, c is iteration coefficient, x (L) _i,j, x (a) _i,jwith x (b) _i,jbe respectively the pixel value that step 3 obtains L, a, b tri-components of image, y (L) _i,j, y (a) _i,jwith y (b) _i,jbe respectively the pixel value of L, a, b tri-components after transverse gradients filtering process, the pixel value of L, a, b tri-components be adjacent is y (L) _{i, j-1}, y (a) _{i, j-1}with y (b) _{i, j-1};

$R_{i,j-1}=\exp (-\frac{{\mathrm{dif}}_L^2+{\mathrm{dif}}_a^2+{\mathrm{dif}}_b^2}{2{\mathrm{\σ}}^2})$

Wherein, σ is Gauss's variance, dif _lfor the difference of brightness L between the horizontal neighbor in Lab color space, dif _afor the difference of a value between horizontal neighbor, dif _bfor the difference of b value between horizontal neighbor, R _{i, j-1}for pixels across coefficient of similarity.

Further, the longitudinal filtering process of image after adopting following formula to process step 4 in step 5:

$\left\{\begin{array}{c}{y}^{\′}{\left(L\right)}_{i,j}=\left(1-R_{i-1,j}\·c\right)\·y{\left(L\right)}_{i,j}+R_{i-1,j}\·c\·y^{\′}{\left(L\right)}_{i-1,j}\\ y^{\′}{\left(a\right)}_{i,j}=\left(1-R_{i-1,j}\·c\right)\·y{\left(a\right)}_{i,j}+R_{i-1,j}\·c\·y^{\′}{\left(a\right)}_{i-1,j}\\ y^{\′}{\left(b\right)}_{i,j}=\left(1-R_{i-1,j}\·c\right)\·y{\left(b\right)}_{i,j}+R_{i-1,j}\·c\·y^{\′}{\left(b\right)}_{i-1,j}\end{array}\right.$

Wherein, R _{i-1, j}for longitudinal neighbor coefficient of similarity, c is iteration coefficient, y (L) _i,j, y (a) _i,jwith y (b) _i,jbe respectively the pixel value that the filtering of step 4 ordinate iteratuin obtains L, a, b tri-components of image, y ' (L) _i,j, y ' (a) _i,jwith y ' (b) _i,jbe respectively the pixel value of L, a, b tri-components after longitudinal gradient filtering process, y ' (L) _{i-1, j}, y ' (a) _{i-1, j}with y ' (b) _{i-1, j}for the pixel value of three components be adjacent;

$R_{i-1,j}=\exp (-\frac{{\mathrm{dif}}_L^{\′2}+{\mathrm{dif}}_a^{\′2}+{\mathrm{dif}}_b^{\′2}}{2{\mathrm{\σ}}^2})$

Wherein, σ is Gauss's variance, dif ' _lfor the difference of brightness L between the longitudinal neighbor in Lab color space, dif ' _afor the difference of a value between longitudinal neighbor, dif ' _bfor the difference of b value between longitudinal neighbor, R _{i-1, j}for longitudinal pixel similarity coefficient.

Compared with prior art, the present invention has following useful technique effect:

After the present invention adopts the iteration bilateral filtering process of horizontal stroke, vertical both direction, the filter effect of bidirectional iteration bilateral filtering algorithm is obviously better than mean filter and its edge keeps effect better, on the other hand, relative to traditional bilateral filtering, processing speed of the present invention improves tens times, and along with the increase of image, the advantage of processing speed of the present invention can be more obvious.So the present invention can eliminate infrared pitch picture noise fast and keep edge details, both meet the filter effect of infrared pitch image, maintained again the marginal information of image preferably, significantly reduced the complexity of traditional bilateral filtering simultaneously.Solve general filtering method and lose the problem of image edge information, also solve long problem of traditional bilateral filtering processing time simultaneously.

Further, the present invention introduces coefficient of similarity between pixel on the basis of interative computation formula, make it can be regarded as bilateral filtering that mask radius is 1, after the iteration bilateral filtering process of horizontal, the vertical both direction of employing like this, the filter effect of bidirectional iteration bilateral filtering algorithm is obviously better than mean filter and its edge keeps effect better.

Accompanying drawing explanation

Fig. 1 is schematic flow sheet of the present invention;

Fig. 2 is that the present invention of the present invention and other filtering method are to infrared pitch image denoising effect comparison diagram; Wherein, a () is original Noise image, b () is mean filter image, c () is traditional bilateral filtering image, (d) ordinate iteratuin bilateral filtering image, (e) abscissa iteration bilateral filtering image, (f) bidirectional iteration bilateral filtering image.

Embodiment

Below in conjunction with accompanying drawing, the present invention is described in further detail:

See Fig. 1, the bidirectional iteration bilateral filtering method of UAV system infrared acquisition pitch image, improve according to existing monodimensional iterative formula, introduce the thought of bilateral filtering, codomain coefficient of similarity is introduced iterative formula, at Lab color space, this formula is utilized to carry out the filtering process of both direction to image.

Step 1: obtain initial infrared pitch image.

Utilize UAV flight's infrared thermal imaging equipment, carry out image acquisition to the asphalt paving road surface in work progress, the original infrared pitch image collected is with much noise, and pending original infrared pitch image size is 544*486.

Step 2: the original infrared pitch image of RGB color space is transformed into Lab color space, image is L, a, b tri-components by R, G, B color transition, wherein L represents the brightness of image, a represents the scope from redness to green, b represents the scope from yellow to blueness, because what Lab color model described is the display mode of color, is a kind of with device-independent color model, therefore carries out at Lab color space the impact that process can reduce equipment.

Step 3: the boundary pixel expanded images border of the infrared pitch image of the Lab color space utilizing step 2 to obtain.

In order to prevent the impact of image boundary, the infrared pitch image of the Lab color space utilizing the Padarray function in MATLAB step 2 to be obtained is 5 according to expansion radius size, with specular way expanded images borderline region.

Step 4: ordinate iteratuin bilateral filtering process.

Iteration two-sided filter can be similar to regards that mask radius is the two-sided filter of 0.5 as, and ordinate iteratuin two-sided filter then effectively can detect longitudinal edge, and effectively carries out landscape images filtering.

The basic thought of ordinate iteratuin bilateral filtering is, asks for horizontal neighbor gray scale difference value, can be judged to be edge pixel point when adjacent Pixel gray difference is larger, contrary to this difference is less, be then non-edge point.

The image slices vegetarian refreshments that horizontal traversal step 3 obtains, carries out filtering process to image.To formula basic parameter assignment, a=0.87, σ=0.8 in this example, calculate the difference between the horizontal neighbor in Lab color space, formula is as follows:

$\left\{\begin{array}{c}{\mathrm{dif}}_L=x{\left(L\right)}_{i,j}-x{\left(L\right)}_{i,j-1}\\ {\mathrm{dif}}_a=x{\left(a\right)}_{i,j}-x{\left(a\right)}_{i,j-1}\\ {\mathrm{dif}}_b=x{\left(b\right)}_{i,j}-x{\left(b\right)}_{i,j-1}\end{array}\right.$

In formula, x (L) _i,jwith x (L) _{i, j-1}be respectively the brightness L component value of horizontal adjacent two pixels in Lab color space, x (a) _i,jwith x (a) _{i, j-1}be respectively a component value of horizontal adjacent two pixels in Lab color space, x (b) _i,jwith x (b) _{i, j-1}be respectively the b component value of horizontal adjacent two pixels in Lab color space, dif _lfor the difference of the horizontal neighbor brightness in Lab color space L, dif _afor the difference of horizontal neighbor a value, dif _bfor the difference of horizontal neighbor b value.

In order to transverse gradients formula is introduced in the impact of Difference of Adjacent Pixels, at the three-channel Difference of Adjacent Pixels of Lab that this calculates in conjunction with Gauss formula and above formula, calculate pixels across coefficient of similarity R _{i, j-1},=formula is as follows:

$R_{i,j-1}=\exp (-\frac{{{\mathrm{dif}}_L}^2+{{\mathrm{dif}}_a}^2+{{\mathrm{dif}}_b}^2}{2{\mathrm{\σ}}^2})$

In formula, dif _lfor the difference of brightness L between the horizontal neighbor in Lab color space, dif _afor the difference of a value between horizontal neighbor, dif _bfor the difference of b value between horizontal neighbor, R _{i, j-1}for pixels across coefficient of similarity.

By horizontal neighbor coefficient of similarity R _{i, j-1}substitute into iterative formula and obtain ordinate iteratuin bilateral filtering formula, when gray scale difference value is larger when between neighbor, judge that this pixel is edge pixel point, horizontal neighbor coefficient of similarity R _{i, j-1}less, little on the impact of this edge pixel point according to neighbor pixel in the known now filtering of ordinate iteratuin bilateral filtering formula, can reach the object of preserving edge pixel, ordinate iteratuin bilateral filtering formula is as follows, according to following iteration bilateral filtering formula, obtain horizontal filtering image;

$\left\{\begin{array}{c}y{\left(L\right)}_{i,j}=\left(1-R_{i,j-1}\·c\right)\·x{\left(L\right)}_{i,j}+R_{i,j-1}\·c\·y{\left(L\right)}_{i,j-1}\\ y{\left(a\right)}_{i,j}=\left(1-R_{i,j-1}\·c\right)\·x{\left(a\right)}_{i,j}+R_{i,j-1}\·c\·y{\left(a\right)}_{i,j-1}\\ y{\left(b\right)}_{i,j}=\left(1-R_{i,j-1}\·c\right)\·x{\left(b\right)}_{i,j}+R_{i,j-1}\·c\·y{\left(b\right)}_{i,j-1}\end{array}\right.$

Wherein, R _{i, j-1}for horizontal neighbor coefficient of similarity, c is iteration coefficient, x (L) _i,j, x (a) _i,jwith x (b) _i,jbe respectively the pixel value that step 3 obtains L, a, b tri-components of image, y (L) _i,j, y (a) _i,jwith y (b) _i,jbe respectively the pixel value of L, a, b tri-components after transverse gradients filtering process, be respectively y (L) with the pixel value of its laterally adjacent L, a, b tri-components _{i, j-1}, y (a) _{i, j-1}with y (b) _{i, j-1}.

Image after the bilateral process of single ordinate iteratuin is as Fig. 2 (e), although edge keeps better, when iteration coefficient c is larger, image there will be pixel lateral excursion phenomenon.Be unfavorable for that image information is extracted.

Step 5: abscissa iteration bilateral filtering process.

Image after step 4 processes occurs grey scale pixel value lateral excursion phenomenon and filter effect is general, owing to being only ordinate iteratuin bilateral filtering, rim detection for transverse direction is unsatisfactory, therefore carry out abscissa iteration bilateral filtering in this step, not only strengthen to the detection of transverse edge and also image better through secondary filtering effect.

First calculate the longitudinal neighbor gray scale difference value through the filtered image of step 4, determine whether edge pixel point according to this difference size.

Image slices vegetarian refreshments after longitudinal traversal step 4 processes, carries out longitudinal filtering process to the horizontal filtered image of step 4.Identical with step 4 to formula basic parameter assignment, calculate the difference between the longitudinal neighbor in Lab color space, formula is as follows:

$\left\{\begin{array}{c}{\mathrm{dif}}_L^{\′}=y{\left(L\right)}_{i,j}-y{\left(L\right)}_{i-1,j}\\ {\mathrm{dif}}_a^{\′}=y{\left(a\right)}_{i,j}-y{\left(a\right)}_{i-1,j}\\ {\mathrm{dif}}_b^{\′}=y{\left(b\right)}_{i,j}-y{\left(b\right)}_{i-1,j}\end{array}\right.$

In formula, y (L) _i,jwith y (L) _{i-1, j}be respectively the brightness L component value of longitudinal adjacent two pixels in Lab color space, y (a) _i,jwith y (a) _{i-1, j}be respectively a component value of longitudinal adjacent two pixels in Lab color space, y (b) _i,jwith y (b) _{i-1, j}be respectively Lab color space longitudinally adjacent two pixel b component values, dif _lfor the difference of the longitudinal neighbor brightness in Lab color space L, dif _afor the difference of mutually longitudinal neighbor a value, dif _bfor the difference of longitudinal neighbor b value.

In order to the impact of longitudinal neighbor gray scale difference value is reflected to Filtering Formula, be incorporated herein longitudinal pixel similarity coefficients R _{i-1, j}, work as coefficients R _{i-1, j}time less, longitudinal neighbor gray scale difference value is larger.Work as coefficients R _{i-1, j}time larger, longitudinal neighbor gray scale difference value is less.Known, longitudinal pixel similarity coefficients R _{i-1, j}longitudinal neighbor grey scale change can be reflected well.According to the three-component difference of L, a, b, calculate longitudinal pixel similarity coefficients R _{i-1, j}, formula is as follows:

$R_{i-1,j}=\exp (-\frac{{\mathrm{dif}}_L^{\′2}+{\mathrm{dif}}_a^{\′2}+{\mathrm{dif}}_b^{\′2}}{2{\mathrm{\σ}}^2})$

In formula, dif ' _lfor the difference of brightness L between the longitudinal neighbor in Lab color space, dif ' _afor the difference of a value between longitudinal neighbor, dif ' _bfor the difference of b value between longitudinal neighbor, R _{i-1, j}for longitudinal pixel similarity coefficient.

By longitudinal pixel similarity coefficients R _{i-1, j}introduce iterative formula, work as coefficients R _{i-1, j}time less, filtered pixel is edge pixel point, and neighbor pixel is little on its impact, reaches the effect that edge keeps.Work as coefficients R _{i-1, j}time larger, filtered pixel is non-edge pixels point, and neighbor pixel is large on its impact, and formula reaches the effect of filtering.Abscissa iteration bilateral filtering formula is as follows, according to following iteration bilateral filtering formula, obtains longitudinal filtering image.

$\left\{\begin{array}{c}{y}^{\′}{\left(L\right)}_{i,j}=\left(1-R_{i-1,j}\·c\right)\·y{\left(L\right)}_{i,j}+R_{i-1,j}\·c\·y^{\′}{\left(L\right)}_{i-1,j}\\ y^{\′}{\left(a\right)}_{i,j}=\left(1-R_{i-1,j}\·c\right)\·y{\left(a\right)}_{i,j}+R_{i-1,j}\·c\·y^{\′}{\left(a\right)}_{i-1,j}\\ y^{\′}{\left(b\right)}_{i,j}=\left(1-R_{i-1,j}\·c\right)\·y{\left(b\right)}_{i,j}+R_{i-1,j}\·c\·y^{\′}{\left(b\right)}_{i-1,j}\end{array}\right.$

In formula, R _{i-1, j}for longitudinal neighbor coefficient of similarity, c is iteration coefficient, y (L) _i,j, y (a) _i,jwith y (b) _i,jbe respectively the pixel value that the filtering of step 4 ordinate iteratuin obtains L, a, b tri-components of image, y ' (L) _i,j, y ' (a) _i,jwith y ' (b) _i,jbe respectively the pixel value of L, a, b tri-components after longitudinal gradient filtering process, y ' (L) _{i-1, j}, y ' (a) _{i-1, j}with y ' (b) _{i-1, j}be respectively and y ' (L) _i,j, y ' (a) _i,jwith y ' (b) _i,jthe pixel value of three longitudinally adjacent components.

Image after the process of single abscissa iteration bilateral filtering is as Fig. 2 (f), and problem is similar with single ordinate iteratuin filtering, occurs pixel vertical misalignment.

Step 6: the image of Lab color space is transformed into RGB color space, exports the infrared pitch image after the process of bidirectional iteration bilateral filtering.

By Fig. 2 (b) adopt radius be 7 filter window, can find out, mean filter soft edge, loss in detail, filter effect is poor, and bilateral filtering is as Fig. 2 (d), same employing radius is the filter window of 7, and not only filter effect better also maintains the edge details of image.Adopt filtering method effect of the present invention as Fig. 2 (f), filter effect is close with traditional bilateral filtering, filter effect or edge keep all there is good effect, and from table 1, be directed to the coloured image of 544*486, the tradition bilateral filtering processing time is 52.872 seconds, and adopt algorithm process of the present invention only to use 2.926 seconds, bidirectional iteration bilateral filtering of the present invention has conspicuousness to promote relative to traditional bilateral filtering processing speed.

Several filtering method processing time contrast of table 1

In sum, filtering method of the present invention is for infrared pitch image denoising, and not only denoising effect is good, and edge keeps better, and processing speed has very large lifting, has great significance to processing also Obtaining Accurate image information in real time.

Claims (6)

1. the bidirectional iteration bilateral filtering method of UAV system infrared acquisition pitch image, is characterized in that, comprise the following steps:

Step 1: utilize unmanned plane to take and obtain original infrared pitch image;

Step 2: the original infrared pitch image of RGB color space step 1 photographed is transformed into Lab color space;

Step 3: the boundary pixel expanded images border of the infrared pitch image of the Lab color space utilizing step 2 to obtain;

Step 4: the horizontal filtering process of image that step 3 is obtained;

Step 5: the longitudinal filtering process of the image after step 4 is processed;

Step 6: the image of the Lab color space after step 5 being processed is transformed into RGB color space, the infrared pitch image after output filtering.

2. the bidirectional iteration bilateral filtering method of UAV system according to claim 1 infrared acquisition pitch image, is characterized in that, the original infrared pitch image size in step 1 is 544*486.

3. the bidirectional iteration bilateral filtering method of UAV system according to claim 1 infrared acquisition pitch image, it is characterized in that, in step 2, the original infrared pitch image of RGB color space is transformed into Lab color space, original infrared pitch image is L, a, b tri-components by R, G, B color transition, wherein L represents the brightness of image, a represents the scope from redness to green, and b represents the scope from yellow to blueness.

4. the bidirectional iteration bilateral filtering method of UAV system according to claim 1 infrared acquisition pitch image, it is characterized in that, in step 3 by the infrared pitch image of Lab color space according to expansion radius size be 5, with specular way expanded images borderline region.

5. the bidirectional iteration bilateral filtering method of UAV system according to claim 1 infrared acquisition pitch image, is characterized in that, adopts the horizontal filtering process of image that following formula obtains step 3 in step 4:

$\left\{\begin{array}{c}y{\left(L\right)}_{i,j}=\left(1-R_{i,j-1}\·c\right)\·x{\left(L\right)}_{i,j}+R_{i,j-1}\·c\·y{\left(L\right)}_{i,j-1}\\ y{\left(a\right)}_{i,j}=\left(1-R_{i,j-1}\·c\right)\·x{\left(a\right)}_{i,j}+R_{i,j-1}\·c\·y{\left(a\right)}_{i,j-1}\\ y{\left(b\right)}_{i,j}=\left(1-R_{i,j-1}\·c\right)\·x{\left(b\right)}_{i,j}+R_{i,j-1}\·c\·y{\left(b\right)}_{i,j-1}\end{array}\right.$

Wherein, R _{i, j-1}for horizontal neighbor coefficient of similarity, c is iteration coefficient, x (L) _i,j, x (a) _i,jwith x (b) _i,jbe respectively the pixel value that step 3 obtains L, a, b tri-components of image, y (L) _i,j, y (a) _i,jwith y (b) _i,jbe respectively the pixel value of L, a, b tri-components after transverse gradients filtering process, the pixel value of L, a, b tri-components be adjacent is y (L) _{i, j-1}, y (a) _{i, j-1}with y (b) _{i, j-1};

$R_{i,j-1}=\exp (-\frac{{\mathrm{dif}}_L^2+{\mathrm{dif}}_a^2+{\mathrm{dif}}_b^2}{2{\mathrm{\σ}}^2})$

Wherein, σ is Gauss's variance, dif _lfor the difference of brightness L between the horizontal neighbor in Lab color space, dif _afor the difference of a value between horizontal neighbor, dif _bfor the difference of b value between horizontal neighbor, R _{i, j-1}for pixels across coefficient of similarity.

6. the bidirectional iteration bilateral filtering method of UAV system according to claim 1 infrared acquisition pitch image, is characterized in that, the longitudinal filtering process of image after adopting following formula to process step 4 in step 5:

$\left\{\begin{array}{c}{y}^{\′}{\left(L\right)}_{i,j}=\left(1-R_{i-1,j}\·c\right)\·y{\left(L\right)}_{i,j}+R_{i-1,j}\·c\·y^{\′}{\left(L\right)}_{i-1,j}\\ y^{\′}{\left(a\right)}_{i,j}=\left(1-R_{i-1,j}\·c\right)\·y{\left(a\right)}_{i,j}+R_{i-1,j}\·c\·y^{\′}{\left(a\right)}_{i-1,j}\\ y^{\′}{\left(b\right)}_{i,j}=\left(1-R_{i-1,j}\·c\right)\·y{\left(b\right)}_{i,j}+R_{i-1,j}\·c\·y^{\′}{\left(b\right)}_{i-1,j}\end{array}\right.$

Wherein, R _{i-1, j}for longitudinal neighbor coefficient of similarity, c is iteration coefficient, y (L) _i,j, y (a) _i,jwith y (b) _i,jbe respectively the pixel value that the filtering of step 4 ordinate iteratuin obtains L, a, b tri-components of image, y ' (L) _i,j, y ' (a) _i,jwith y ' (b) _i,jbe respectively the pixel value of L, a, b tri-components after longitudinal gradient filtering process, y ' (L) _{i-1, j}, y ' (a) _{i-1, j}with y ' (b) _{i-1, j}for the pixel value of three components be adjacent;

$R_{i-1,j}=\exp (-\frac{{\mathrm{dif}}_L^{\′2}+{\mathrm{dif}}_a^{\′2}+{\mathrm{dif}}_b^{\′2}}{2{\mathrm{\σ}}^2})$

Wherein, σ is Gauss's variance, dif ' _lfor the difference of brightness L between the longitudinal neighbor in Lab color space, dif ' _afor the difference of a value between longitudinal neighbor, dif ' _bfor the difference of b value between longitudinal neighbor, R _{i-1, j}for longitudinal pixel similarity coefficient.