CN101783012A - Automatic image defogging method based on dark primary colour - Google Patents

Abstract

The invention discloses an automatic image defogging method based on dark primary color, which is used for solving the problem of information loss because the traditional defogging method highlights details by improving foggy day image contrast. The method provided by the invention comprises: A. calculating the dark primary color of the primary fog images and relevant atmosphere light value; B. according to the luminance component image of the original fog image, calculating a transmission image reflecting local fog concentration in an atmospheric scattering model; and C. determining the defogged primary image according to the fog image, the transmission image and the atmosphere light value in the atmospheric scattering model. Because of being built on the basis of a physical model, the invention can process various fog images in a self-adaption mode; and defogged images have favorable edge details and ideal contrast, and the clarifying effect is superior to the traditional defogging method based on image enhancement.

Description

A kind of automated graphics defogging method capable based on dark primary

Technical field

The invention belongs to the Image Information Processing field, be specifically related to automated graphics defogging method capable based on dark primary.

Background technology

It is problem highly significant in the computer vision field that the mist image sharpening is arranged, Misty Image is carried out the visual effect that mist elimination can improve image, for example most life outdoor videos work system, as video monitoring, topographic(al) reconnaissance, automatic driving etc., all need the clear characteristics of image that extracts exactly, but under the greasy weather situation, because the visibility of scene reduces, feature such as target contrast and color is attenuated in the image, system can't operate as normal, therefore need eliminate the influence of fog to scene image in image.

At present, the disposal route for Misty Image mainly is divided into two classes: based on the Misty Image Enhancement Method of Flame Image Process with based on the Misty Image restored method of physical model.The method of figure image intensifying can improve contrast effectively, and outstanding details is improved visual effect, but may be caused certain loss for the information of outshot.What wherein mainly adopt at the image enchancing method of Misty Image is the histogram equalization method.This method generally strengthens entire image, and do not consider at the different depth of field zone of image and adopt different enhancing strategies, thereby undesirable to the enhancing effect of some Misty Image.Partial histogram equalization method (AHE), though can solve the problem of the different depth of field, the calculated amount of this method is very big, and is impracticable.

Image restoration is the physical process that the research Misty Image is degenerated, and sets up degradation model, and the inverting degenerative process so that obtain without the original image that disturbs degeneration or the optimal estimation value of original image, thereby is improved the Misty Image quality.By contrast, this method is with strong points, and the result's nature that obtains does not generally have information loss, can obtain comparatively desirable mist elimination effect.The atmospheric scattering model that some scholars utilize McCartney (McCartney) to propose, by the greasy weather scene modeling being solved the mist elimination problem of Misty Image, wherein noticeable by a kind of single image defogging method capable of propositions recently such as He Kaiming based on dark primary.Introduction for this method, can reference papers " Single Image Haze Removal Using Dark Channel Prior (based on the single image defogging method capable of dark primary priori) " (being stated from IEEE Conference on ComputerVision and Pattern Recognition (CVPR), in June, 2009).The core of this method just is that scratching the figure method by soft obtains improved Misty Image propagation figure, but this step operation cost is very big and need manual shift, so be difficult to satisfy the realtime graphic processing demands to the conversion scene.

Under this background, study and a kind ofly can improve arithmetic speed, can handle adaptively again variously has the defogging method capable of mist image to seem particularly important.

Summary of the invention

Technical matters to be solved by this invention provides a kind of automated graphics defogging method capable based on dark primary, improve person in charge's visual effect that the mist image is arranged, not only do not need artificial participation, and can reduce calculation cost significantly, when obtaining preferably the sharpening effect, improve the speed of sharpening significantly.

The present invention solves the problems of the technologies described above the technical scheme that is adopted to be:

A kind of automated graphics defogging method capable based on dark primary is characterized in that this method may further comprise the steps:

A, the dark primary image of asking for original mist image and atmosphere light value;

B, the luminance component figure by original mist image ask for the propagation figure of the local fog concentration of reflection in the atmospheric scattering model;

C, basis have the restored image after mist image, propagation figure and atmosphere light value are asked for mist elimination.This method all can obtain mist elimination effect preferably for gray level image and coloured image.

$J(x,y)=\frac{I(x,y)-A}{\max (t(x,y),t_0)}+A;$

The expression formula of asking for restored image is:

X wherein, y is the coordinate figure of each pixel of image, (x y) is the image after restoring to J, and (x y) be original mist image to I, and (x y) schemes for propagating t, and A is the atmosphere light value, t ₀Value be 0.1.

Described steps A is:

When original mist image is gray level image, choose template and on this gray level image, carry out minimum value filtering, obtain the dark primary image; For gray level image, the pixel value of every bit is brightness value;

The value J of described each pixel of dark primary image ^Dark(x, y) determine by following formula:

$J^{\mathrm{dark}}(x,y)=\underset{(x^{\′},y^{\′})\∈\mathrm{\Ω}(x,y)}{\min }\left(J(x^{\′},y^{\′})\right);$

Wherein, Ω (x, y) being is x with the coordinate, the pixel of y is the center, carry out the template zone of minimum value filtering, x ', y ' they are the coordinate figure of each pixel in the template zone, J (x ', y ') each regional area for being divided by the filtering template in original mist image, if the image size of original mist image is 600*400, then template size is 15*15, and the filtering template of the image that other is big or small is chosen by following formula and determined:

Wherein, M, N are respectively the length and width size of original mist image, and the filtering template size is N _m* N _m, in the formula

Be the operation that rounds up, the value of every bit is called as the dark primary image value in the dark primary image;

Again with these dark primary image values according to the rank order of successively decreasing, determine that numerical values recited is point residing position in the dark primary image of preceding 0.1%, then the maximum brightness value in the pairing original mist image-region in these positions is atmosphere light value A;

When original mist image is coloured image, three Color Channels of R, G, B of original mist image is chosen the filtering template respectively carry out minimum value filtering, the minimum value of three image corresponding pixel points of gained after the filtering pixel value as dark primary image corresponding point; Then the value of each pixel of dark primary image of described coloured image is determined by following formula:

$J^{\mathrm{dark}}(x,y)=\underset{c\∈\{r,g,b\}}{\min }\left(\underset{(x^{\′},y^{\′})\∈\mathrm{\Ω}(x,y)}{\min }\left(J^c(x^{\′},y^{\′})\right)\right);$

Wherein, J ^cColor Channel for original image J; For coloured image, be the YCbCr color space with original mist image from the RGB color space conversion earlier then, in the YCbCr color space, monochrome information is represented with single component Y;

Again with these dark primary image values according to the rank order of successively decreasing, determine that numerical values recited is point residing position in the dark primary image of preceding 0.1%, then the maximum brightness value in the pairing original mist image-region in these positions is atmosphere light value A.

Step B comprises:

The luminance picture of B1, the original mist image of acquisition: if original mist image is a gray level image, then the pixel value of every bit is brightness value; Original mist image is luminance picture; If original mist image is a coloured image, be the YCbCr color space with original mist image from the RGB color space conversion earlier then; In the YCbCr color space, monochrome information represents that with single component Y chromatic information is stored with two color difference components Cb and Cr, just the luminance component Y after separating can be extracted thus, obtain luminance picture Y (x, y);

B2, described luminance picture is carried out multiple dimensioned Retinex conversion, can obtain the new luminance picture R that described luminance component edge of image details strengthens, contrast is improved thus _M(x, y);

B3, ask for the inverse luminance picture of described new luminance picture; Inverse luminance picture I _Inv(x, y) determine as follows:

I _Inv(x, y)=C-R _M(x, y), C adjusts parameter for propagating figure in the formula; If original mist image is a gray level image, then the span of C is 0.8～1.2; If original mist image is a coloured image, then the span of C is 1～1.4;

B4, described inverse luminance picture is carried out medium filtering, filtered image is the propagation figure of atmospheric scattering model; Described medium filtering process is, any value in the image, substitutes with the intermediate value of each value in this vertex neighborhood, and described intermediate value be with that middle element value after each value sorts in this vertex neighborhood.

Described step B2 comprises:

$R_M(x,y)=\underset{n=1}{\overset{N_1}{\mathrm{\Σ}}}{\mathrm{\ω}}_n(\log Y(x,y)-\log [F_n(x,y)*Y(x,y)\left]\right)$

Described luminance component image is carried out multiple dimensioned Retinex conversion, and the mathematical form of this processing procedure is as follows:

Wherein, Y (x, y) expression one width of cloth size is the luminance component image of M * N, x=0 wherein, 1,2 ..., M-1 and y=0,1,2 ..., N-1; R _M(x y) is the output image that adopts after multiple dimensioned Retinex transfer pair luminance component image carries out conversion process; This output image R _M(x, size y) and luminance component image Y (x, y) size is identical, is M * N, x=0 wherein, 1,2 ..., M-1 and y=0,1,2 ..., N-1; Promptly can directly obtain the new luminance picture of a width of cloth through after the multiple dimensioned Retinex conversion; N ₁For around function yardstick number, value is 3, promptly selects 3 different yardsticks, is respectively large scale, mesoscale and small scale, so that the contrast of the luminance picture of trying to achieve is enhanced, and when obtaining compressing, dynamic range can keep the key colour of luminance picture; ω _nBe weights, and it is bigger than the value of other 2 yardsticks to give the weight of large scale when choosing under the weights sum that guarantees each yardstick be 1 prerequisite corresponding to each yardstick; F _n(x y) is corresponding weights ω _nN around function, have:

$F_n(x,y)=K_n{e}^{-\frac{x^2+y^2}{C_n^2}};$

Wherein, C _nBe the value (n=1,2,3) of n yardstick, the principle of yardstick value is: the value C of small scale ₁Be 1%～5% of image size, the value C of mesoscale ₂Be 10%～15% of image size, the value C of large scale ₃Be 30%～50% of image size; K _nBe normalized factor.K _nValue should make ∫ ∫ F _n(x, y) dxdy=1.

The principle of yardstick value is: when yardstick be the image size 1%～5% the time, Retinex result can obtain extraordinary image border details, can select it as small scale C ₁Value; When yardstick be the image size 10%～15% the time, Retinex result can obtain image border details and color simultaneously, can select it as mesoscale C ₂Value; When yardstick be the image size 30%～50% the time, Retinex result obtains the color of comparison balance, can select it as large scale C ₃Value.Therefore, C _nThe value of large, medium and small each yardstick can be according to the difference of image size and difference.

The value of the template size that described medium filtering is selected is pressed following formula and is determined:

Wherein, M, N are respectively the length and width size of image, and the template size of medium filtering is Z _m* Z _m,

Be the operation that rounds up.

Retinex (abbreviation of retina " Retina " and cerebral cortex " the Cortex ") color how be of putting forward of Edwin Land (Ai Erwenlande) to regulate to perceive object about the human visual system and the model of brightness.Be different from traditional algorithm for image enhancement, can only strengthen a certain category feature of image as linearity, nonlinear transformation, image sharpening etc., dynamic range as compressed image, or the enhancing edge of image etc., Retinex can be in the gray scale dynamic range compression, the edge strengthens and constant color three aspects reach balance, thereby can adaptively strengthen various dissimilar images.The ultimate principle of Retinex is that piece image is divided into luminance picture and reflected image two parts, then by reducing luminance picture reaches the enhancing image to the influence of reflected image purpose.Because Retinex is based upon on the experiment basis, there is not unified mathematical model, therefore multiple different Retinex algorithm has appearred, as multiple dimensioned Retnixe (MSR) algorithm etc.Although these classic algorithm seem different, they are closely similar in fact, all smoothly extract luminance picture by original image is carried out certain Gauss, and make the luminance picture of extraction accurate as far as possible by the calculating of complexity.

Beneficial effect of the present invention:

The advantage that the detection method that the present invention relates to has is as follows:

The inventive method travelling speed is fast, and is effective, can not only solve many depth of field problem, and do not need artificial participation, has good versatility.Choose the Misty Image that three width of cloth sizes are respectively 600*400,204*209,835*557, adopt Matlab 6.5 at Pentium (R) D, 3.00Ghz, the contrast that experimentizes of the PC of 2GB internal memory, Fig. 3, Fig. 6, Fig. 9 have shown the inventive method and traditional histogram equalization method and up-to-date defogging method capable, at CVPR 08 ', He Kaiming is at CVPR 09 ' as Tan, and the effect of Tarel institute's extracting method on ICCV 09 ' relatively.On visual effect, traditional histogram equalization method makes scenery nearby cross enhancing, and scenery at a distance becomes fuzzyyer.Recently the pseudo-shadow of halation can appear in the image scene target behind the Tan method mist elimination of Ti Chuing, and the Tarel method then can't be removed the fog at the little details of target place.By contrast, result's nature that the He Shi method obtains has been obtained comparatively desirable mist elimination effect.And the inventive method is based upon on the basis of atmospheric scattering physical model, and mist elimination effect and He Shi method are about the same.All obtained enhancing in various degree with scenery at a distance nearby, details is clearly more demarcated, and contrast is compared with original mist image all remarkable enhancing, has also obtained experiment effect preferably in the recovery of image information, and the pseudo-shadow of halation do not occur.

In addition, the inventive method has improved the speed of algorithm significantly owing to the propagation figure method of estimation that has adopted based on luminance component.Experimental data by table 1 is as can be seen: the inventive method obviously is better than histogram equalization method, Tarel method in the mist elimination effect, and close with He Shi method effect the time, speed be the He Shi method 2-3 doubly, and far away faster than the Tan method.Statistics in the following table all obtains under the Matlab environment, and programming realizes that will significantly reduce working time, thereby realized automatism, the real-time of image mist elimination if the inventive method adopts C++.

The inventive method can be widely used in the safe driving backup system or autonomous driving car in the future of video monitoring, topographic(al) reconnaissance and existing vehicle, aircraft, ship.

Table 1 algorithm comparison sheet working time

Picture number (dpi)	Histogram equalization (s)	??Tan(s)	He Shi (s)	??Tarel(s)	The present invention (s)
						Fig. 3 (600*400) Fig. 6	??4.8750??0.9210	??184.4260??38.7254	??13.4180??2.3780	??2.1250??0.4690	??5.1570??0.9530
(204*209) Fig. 9 (835*557)	??10.2350	??249.2972	??18.1532	??6.8600	??9.7660

Description of drawings

Fig. 1 is based on the process flow diagram of the automated graphics defogging method capable of dark primary in the embodiment of the invention;

Fig. 2 propagates the process flow diagram of image for obtaining the atmospheric scattering model in the embodiment of the invention;

Fig. 3 is the gray scale Misty Image of embodiment 1 and the treatment effect of each defogging method capable; (a is original mist image, and b is the histogram equalization design sketch, and c is a Tarel method design sketch, and d is the inventive method design sketch)

Fig. 4 is the dark primary image of embodiment 1;

Fig. 5 is that the atmospheric scattering model of embodiment 1 is propagated figure;

Fig. 6 is the colored Misty Image of embodiment 2 and the treatment effect of each defogging method capable; (a is original mist image, and b is the histogram equalization design sketch, c Wei Heshi method design sketch, d is the inventive method design sketch)

Fig. 7 is the dark primary image of embodiment 2;

Fig. 8 is that the atmospheric scattering model of embodiment 2 is propagated figure;

Fig. 9 is the colored Misty Image of embodiment 3 and the treatment effect of each defogging method capable; (a is original mist image, and b is the histogram equalization design sketch, and c is a tan method design sketch, and d is the inventive method design sketch)

Figure 10 is the dark primary image of embodiment 3;

Figure 11 is that the atmospheric scattering model of embodiment 3 is propagated figure.

Embodiment

The invention will be further described below in conjunction with the drawings and specific embodiments.

Embodiment 1:

Present embodiment is at gray level image, and according to shown in Figure 1, its mist elimination process is undertaken by following three steps:

1, asks for the dark primary image and the atmosphere light value of original misty grey degree image

The value of each pixel of dark primary image of described gray level image is determined by following formula:

$J^{\mathrm{dark}}(x,y)=\underset{(x^{\′},y^{\′})\∈\mathrm{\Ω}(x,y)}{\min }\left(J(x^{\′},y^{\′})\right)$

Wherein, and Ω (x, y) being is x with the coordinate, and the pixel of y is the center, carries out the template zone of minimum value filtering, and x ', y ' they are the coordinate figure of each pixel in the template zone.Because Fig. 3 (a) size is 600*400, then the template size of minimum value filtering is 15*15, and the dark primary image that obtains this gray level image thus as shown in Figure 4.The value of every bit is called as the dark primary image value in this dark primary image, again with these dark primary image values according to the rank order of successively decreasing, determine that numerical values recited is point residing position in the dark primary image of preceding 0.1%, then the maximum brightness value in the pairing original mist image-region in these positions is the value of atmosphere light A.It is 255 that the A value of present embodiment is tried to achieve.

2, ask for atmospheric scattering model propagation figure

Realize that this idiographic flow of handling process is please referring to Fig. 2.May further comprise the steps:

At first, extract the luminance component figure of original mist image, because of former figure is a gray level image, then the pixel value of its every bit is brightness value.

Then, gray level image is carried out multiple dimensioned Retinex (MSR) conversion, the mathematical form of this processing procedure

$R_M(x,y)=\underset{n=1}{\overset{N}{\mathrm{\Σ}}}{\mathrm{\ω}}_n(\log Y(x,y)-\log [F_n(x,y)*Y(x,y)\left]\right)---\left(2\right)$

As follows:

Wherein, R _M(x is to adopt MSR that the luminance component image is carried out output after the conversion process y), and N is the yardstick number around function, generally chooses large, medium and small 3 yardsticks.ω _nBe the weights corresponding to each yardstick, (x y) is luminance picture, F to Y _n(x y) is corresponding weights ω _nN around function, choose Gauss's form.For present embodiment its around function scale size (image size 5%), 36000 (the image size 15%), 120000 that be taken as 12000 respectively (image size 50%), and choosing of the weighted value of each yardstick should guarantee and value is to follow the principle that makes image that the color of comparison balance be arranged under 1 the prerequisite, chooses numerical value 0.05,0.05,0.9 respectively for present embodiment.

Secondly, propagation figure is adjusted the pixel of every bit in the new luminance graph after parameters C deducts the Retinex conversion

Value obtains the pixel value of the inverse luminance graph corresponding point of this luminance graph.This asks for process and determines in the following manner:

I _inv(x，y)＝C-R _M(x，y)????????(3)

Wherein, the span of C is 0.8～1.2, and the value of C is 0.89 in the present embodiment.

At last, described inverse luminance graph is carried out the medium filtering that template size is 13*13, the atmospheric scattering model propagation figure that can obtain reflecting local fog concentration as shown in Figure 5.

3, determine restored image behind the mist elimination

According to the expression formula of atmospheric scattering model, described propagation figure and original mist gray-scale map are directly carried out the relational algebra computing of image, obtain restored image behind this gray level image mist elimination shown in Fig. 3 (d);

The described expression formula of asking for restored image is:

$J(x,y)=\frac{I(x,y)-A}{\max (t(x,y),t_0)}+A---\left(4\right)$

Wherein, (x y) is original mist image to I, and (x, y) for propagating figure, A is the atmosphere light value to t, t ₀Value be 0.1.

In addition, for coloured image, get final product according to following examples processing.

Embodiment 2:

Coloured image Fig. 6 (a) (size is 204*209) is carried out mist elimination to be handled.On R, G, three Color Channels of B, adopt size to carry out minimum value filtering respectively Fig. 6 (a) earlier for the template of 7*7, with the minimum value of three image corresponding pixel points of gained after the filtering pixel value as dark primary image corresponding point, the dark primary image that obtains Fig. 6 (a) as shown in Figure 7.And be 206 by the value that this dark primary image is tried to achieve the atmosphere light A of Fig. 6 (a).

Then, original mist image transitions to the YCbCr color space, is extracted its luminance component image, this luminance graph is carried out the Retinex conversion according to formula (2), carry out the inverse conversion according to formula (3) again, wherein the span of C is 1～1.4, and the present embodiment value is 1.08.Then conversion gained image being carried out template size is that the medium filtering of 6*6 can obtain the propagation figure that characterized by monochrome information as shown in Figure 8.

At last, with the propagation figure t that characterizes by luminance component that tries to achieve (x, y), (x y) and in the atmosphere light value A substitution image restoration procedure expression (4) that obtains, can try to achieve the mist elimination image shown in Fig. 6 (d) to original mist image I.

Embodiment 3:

Coloured image Fig. 9 (a) (size is 835*557) is carried out mist elimination to be handled.On R, G, three Color Channels of B, adopt size to carry out minimum value filtering respectively Fig. 9 (a) earlier for the template of 21*21, with the minimum value of three image corresponding pixel points of gained after the filtering pixel value as dark primary image corresponding point, the dark primary image that obtains Fig. 9 (a) as shown in figure 10.And be 224 by the value that this dark primary image is tried to achieve the atmosphere light A of Fig. 9 (a).

Then, original mist image transitions to the YCbCr color space, is extracted its luminance component image, this luminance graph is carried out the Retinex conversion according to formula (2), carry out the inverse conversion according to formula (3) again, wherein the span of C is 1～1.4, and the present embodiment value is 1.08.Then conversion gained image being carried out template size is that the medium filtering of 18*18 can obtain the propagation figure that characterized by monochrome information as shown in figure 11.

At last, with the propagation figure t that characterizes by luminance component that tries to achieve (x, y), (x y) and in the atmosphere light value A substitution image restoration procedure expression (4) that obtains, can try to achieve the mist elimination image shown in Fig. 9 (d) to original mist image I.

Need to prove, more than disclosed only be instantiation of the present invention, according to thought provided by the invention, those skilled in the art can think and variation, all should fall within the scope of protection of the present invention.

Claims (5)

1. automated graphics defogging method capable based on dark primary is characterized in that this method may further comprise the steps:

A, the dark primary image of asking for original mist image and atmosphere light value;

B, the luminance component figure by original mist image ask for the propagation figure of the local fog concentration of reflection in the atmospheric scattering model;

C, basis have the restored image after mist image, propagation figure and atmosphere light value are asked for mist elimination.This method all can obtain mist elimination effect preferably for gray level image and coloured image.

$J(x,y)=\frac{I(x,y)-A}{\max (t(x,y),t_0)}+A$

The expression formula of asking for restored image is:

X wherein, y is the coordinate figure of each pixel of image, (x y) is the image after restoring to J, and (x y) be original mist image to I, and (x y) schemes for propagating t, and A is the atmosphere light value, t ₀Value be 0.1.

2. the automated graphics defogging method capable based on dark primary according to claim 1 is characterized in that steps A is:

When original mist image is gray level image, choose template and on this gray level image, carry out minimum value filtering, obtain the dark primary image; For gray level image, the pixel value of every bit is brightness value;

The value J of described each pixel of dark primary image ^Dark(x, y) determine by following formula:

$J^{\mathrm{dark}}(x,y)=\underset{(x^{\′},y^{\′})\∈\mathrm{\Ω}(x,y)}{\min }\left(J(x^{\′},y^{\′})\right);$

Wherein, Ω (x, y) being is x with the coordinate, the pixel of y is the center, carry out the template zone of minimum value filtering, x ', y ' they are the coordinate figure of each pixel in the template zone, J (x ', y ') each regional area for being divided by the filtering template in original mist image, if the image size of original mist image is 600*400, then template size is 15*15, and the filtering template of the image that other is big or small is chosen by following formula and determined:

Wherein, M, N are respectively the length and width size of original mist image, and the filtering template size is N _m* N _m, in the formula

Be the operation that rounds up, the value of every bit is called as the dark primary image value in the dark primary image;

Again with these dark primary image values according to the rank order of successively decreasing, determine that numerical values recited is point residing position in the dark primary image of preceding 0.1%, then the maximum brightness value in the pairing original mist image-region in these positions is atmosphere light value A;

When original mist image is coloured image, three Color Channels of R, G, B of original mist image is chosen the filtering template respectively carry out minimum value filtering, the minimum value of three image corresponding pixel points of gained after the filtering pixel value as dark primary image corresponding point; Then the value of each pixel of dark primary image of described coloured image is determined by following formula:

$J^{\mathrm{dark}}(x,y)=\underset{c\∈\{r,g,b\}}{\min }\left(\underset{(x^{\′},y^{\′})\∈\mathrm{\Ω}(x,y)}{\min }\left(J^c(x^{\′},y^{\′})\right)\right);$

Wherein, J ^cColor Channel for original image J; For coloured image, be the YCbCr color space with original mist image from the RGB color space conversion earlier then, in the YCbCr color space, monochrome information is represented with single component Y;

Again with these dark primary image values according to the rank order of successively decreasing, determine that numerical values recited is point residing position in the dark primary image of preceding 0.1%, then the maximum brightness value in the pairing original mist image-region in these positions is atmosphere light value A.

3. the automated graphics defogging method capable based on dark primary according to claim 1 is characterized in that step B comprises:

The luminance picture of B1, the original mist image of acquisition: if original mist image is a gray level image, then the pixel value of every bit is brightness value; Original mist image is luminance picture; If original mist image is a coloured image, be the YCbCr color space with original mist image from the RGB color space conversion earlier then; In the YCbCr color space, monochrome information represents that with single component Y chromatic information is stored with two color difference components Cb and Cr, just the luminance component Y after separating can be extracted thus, obtain luminance picture Y (x, y);

B2, described luminance picture is carried out multiple dimensioned Retinex conversion, can obtain the new luminance picture R that described luminance component edge of image details strengthens, contrast is improved thus _M(x, y);

B3, ask for the inverse luminance picture of described new luminance picture; Inverse luminance picture I _Inv(x, y) determine as follows:

I _Inv(x, y)=C-R _M(x, y), C adjusts parameter for propagating figure in the formula; If original mist image is a gray level image, then the span of C is 0.8～1.2; If original mist image is a coloured image, then the span of C is 1～1.4;

B4, described inverse luminance picture is carried out medium filtering, filtered image is the propagation figure of atmospheric scattering model; Described medium filtering process is, any value in the image, substitutes with the intermediate value of each value in this vertex neighborhood, and described intermediate value be with that middle element value after each value sorts in this vertex neighborhood.

4. the automated graphics defogging method capable based on dark primary as claimed in claim 3 is characterized in that described step B2 comprises:

$R_M(x,y)=\underset{n=1}{\overset{N_1}{\mathrm{\Σ}}}{\mathrm{\ω}}_n(\log Y(x,y)-\log [F_n(x,y)*Y(x,y)\left]\right)$

Described luminance component image is carried out multiple dimensioned Retinex conversion, and the mathematical form of this processing procedure is as follows:

Wherein, Y (x, y) expression one width of cloth size is the luminance component image of M * N, x=0 wherein, 1,2 ..., M-1 and y=0,1,2 ..., N-1; R _M(x y) is the output image that adopts after multiple dimensioned Retinex transfer pair luminance component image carries out conversion process; This output image R _M(x, size y) and luminance component image Y (x, y) size is identical, is M * N, x=0 wherein, 1,2 ..., M-1 and y=0,1,2 ..., N-1; Promptly can directly obtain the new luminance picture of a width of cloth through after the multiple dimensioned Retinex conversion; N ₁For around function yardstick number, value is 3, promptly selects 3 different yardsticks, is respectively large scale, mesoscale and small scale, so that the contrast of the luminance picture of trying to achieve is enhanced, and when obtaining compressing, dynamic range can keep the key colour of luminance picture; ω _nBe weights, and it is bigger than the value of other 2 yardsticks to give the weight of large scale when choosing under the weights sum that guarantees each yardstick be 1 prerequisite corresponding to each yardstick; F _n(x y) is corresponding weights ω _nN around function, have:

$F_n(x,y)=K_n{e}^{-\frac{x^2+y^2}{C_n^2}};$

Wherein, C _nBe the value (n=1,2,3) of n yardstick, the principle of yardstick value is: the value C of small scale ₁Be 1%～5% of image size, the value C of mesoscale ₂Be 10%～15% of image size, the value C of large scale ₃Be 30%～50% of image size; K _nBe normalized factor.

5. as claim 3 or 4 described automated graphics defogging method capables, it is characterized in that the value of the template size that described medium filtering is selected is pressed following formula and determined based on dark primary:

The medium filtering template length of side

Wherein, M, N are respectively the length and width size of image, and the template size of medium filtering is Z _m* Z _m,

Be the operation that rounds up.

Images