CN104200445A - Image defogging method with optimal contrast ratio and minimal information loss - Google Patents

Abstract

The invention relates to an image defogging method with optimal contrast ratio and minimal information loss. The image defogging method is characterized in that through the adoption of functions including the Gauss mixed model, the quadtree, the maximal contrast ratio and the minimal information loss, the image defogging method is realized. The image defogging method comprises the following steps: firstly, based on the Gauss mixed model and the expected value maximal algorithm, images are segmented into two categories, namely, a sky region and a non-sky region; secondly, the quadtree iteration method is adopted in the sky region of the images to estimate the atmospheric illumination intensity of an atmospheric scattering model; then, the grid partition method is adopted to partition the non-sky region of the images and the functions of the maximal contrast ratio and the minimal information loss are adopted to calculate the spreading rate of an atmospheric lighting model of each grid unit; based on the constant coefficient and the optimal spreading rate of the non-sky region, the spreading rate of the sky region is estimated; finally, the recovery images of the sky region and the non-sky region are merged and output.

Description

The image defogging method capable of a kind of optimum contrast and minimum information loss

Technical field

The present invention relates to Computer Image Processing, particularly the image defogging method capable of a kind of optimum contrast and minimum loss of information.

Background technology

Since 2012,74, whole nation emphasis monitoring city nearly half has been subject to serious haze and has polluted, haze shrouds the overhead at us, cover sight line, image color dimness, contrast step-down that the imaging system such as camera and video monitoring is caught, the serious degradation of picture quality, has directly affected the visual effect of image, have a strong impact on their range of application, need to carry out mist elimination processing to the image that is subject to haze pollution in a lot of application scenarios.

Atmospheric medium is mainly made up of air molecule, steam and gasoloid, and in air, contained particulate is the principal element that haze forms, and is also that picture quality produces the basic reason of degenerating.Under haze weather, the light of body surface reflection can be subject to the impact of airborne suspended particle in the process that arrives imaging device, thereby makes equipment cannot obtain clear picture.Particulate has scattering process to light, and scattering loss makes " transmitted light " strength retrogression, has caused the contrast of image to decline.

The research of image mist elimination can be traced back to the image mist elimination of the people such as L.Bissonnette in 1992 for mist and rainy day gas the earliest.Image mist elimination technology has experienced the development of two more than ten years, has obtained larger progress, constantly has new thought and new method to produce and for Practical Project, mainly contains based on model with based on non-model both direction.

The low-light level that method consideration image based on non-model presents and the feature of low contrast, using conventional image enchancing method as basic processing means.The reason that method based on non-model does not need analysis image to degenerate, only by specifically giving prominence to interested partial information in image, decays unwanted information or remove simultaneously.Mainly comprise spatial domain and frequency field two class methods, typical method comprises algorithm of histogram equalization, bent wave conversion, Homomorphic Filtering Algorithm, based on atmosphere modulation transfer function, wavelet method and Retinex algorithm etc.Defogging method capable based on non-model just strengthens the contrast of image, do not consider Misty Image Blur technique and causes for Degradation, do not consider that haze concentration is directly proportional to the target depth of field, in fact the visual effect of just having improved to a certain extent image is not in fact real image mist elimination.

Defogging method capable based on model, by analysis image causes for Degradation, to atmospheric scattering modeling, is realized the recovery of image.Method based on model is divided three classes: (1) recovery based on partial differential equation; (2) recovery based on depth relationship; (3) recovery based on prior imformation.

Image defogging method capable based on partial differential equation is relatively applicable to the color sharpness of image and the occasion that contrast has higher requirements.By atmospheric scattering model, set up the energy optimization model of outdoor image global defogging and local defogging, derive and comprise accordingly the partial differential equation of image gradient and the scene depth of field.But the method has a weak point to be that the desired gradual modification atmospheric scattering of obtaining of picture depth information coefficient all needs the interactive operation by user.

Method based on depth relationship utilizes depth relationship figure to carry out mist elimination processing to image.Calculate the depth map of background image by gathering scene image corresponding under different weather condition, then obtain the degree of depth of foreground target object in conjunction with relevant heuristic information.Although the mist elimination effect of these methods is comparatively satisfactory, it need to be too harsh by the requirement of reference picture, is difficult in actual applications realize.

Based on the restored method of prior imformation, need multiple image or more supplementarys.According to whether known two classes that are divided into of depth information of scene.One class is the known method of hypothesis depth information of scene, by restoring scene contrast, use a simple Gaussian function to predict the light path in scene, but the method needs radar installations to obtain scene depth.Another kind of is to extract scene depth by supplementary.Utilize two-value scattering model, the different angles such as the polarization characteristic of different scattered lights and interactive depth of field estimation are extracted depth information of scene.But the method for polarized light can only be applied to the weak mist of atmospheric scattering degree, and is unsuitable for foggy weather.Some method needs to use the mutual of the image of same scene under different weather state or user, is difficult to meet the practical application request to conversion scene.

According to the mist elimination algorithm of depth information, develop into the degree that realizes single image mist elimination.Tan has proposed the single image defogging method capable of the local contrast of a kind of expansion with mist image, but has color supersaturation, causes halation artifact phenomenon.Fattal has proposed a kind of method of analyzing based on independent component, be incoherent by supposition transmitance and surface projection at regional area, the reflectivity of estimation scenery, infers transmitance when scene light is propagated in air, finally realize the recovery of scene, but the method is only suitable for mist image.He has proposed the classical single image mist elimination based on dark primary priori theory, but the method is not suitable for the object brightness image similar with atmosphere light.

Existing these methods are relatively applicable to special image, have advantage separately.But part defogging method capable shows as contrast and excessively stretches, or can not correctly estimate that the depth of field causes removing the heavier image of haze, or only consider contrast to maximize and do not lost the important information of former figure, or the sky part color of image is twisted.

Summary of the invention

For solving the problems of the technologies described above, the present invention proposes the image defogging method capable of a kind of optimum contrast and minimum loss of information.

For achieving the above object, the present invention proposes a kind of image defogging method capable based on optimum contrast and minimum loss of information, it is characterized in that the method utilization utilizes gauss hybrid models, Si Fenshu, maximizes the functions such as contrast and minimum information loss and has realized image defogging method capable.The concrete steps of the method comprise:

Step 1, sets up the haze iconic model based on McCartney atmospheric scattering model;

Step 2, utilizes gauss hybrid models and expectation maximization algorithm to segment the image into day dummy section and non-day dummy section two classes;

Step 3, the atmosphere intensity of illumination of the method estimated image sky dummy section based on four points of tree iteration;

Step 4, the method piecemeal of dividing based on grid, the spreading rate of each grid cell of the non-day dummy section of image based on optimum contrast and the estimation of minimum information loss criterion;

Step 5, the normal scale-up factor method of the average spreading rate of non-day dummy section of employing and image is estimated the spreading rate of day dummy section;

Step 6, according to atmospheric scattering model, utilizes the model parameter of estimating, merges the Recovery image of output sky and non-day dummy section.

In described step 2, the concrete steps of utilizing gauss hybrid models and expectation maximization algorithm to segment the image into day dummy section and non-day dummy section two classes comprise:

Step 21): adopting FCM Algorithms (FCM) is 2 clusters to haze image I initial division, and initializes weights, the parameters such as average and variance;

Step 22): calculate the posterior probability that each pixel belongs to respectively 2 models, and upgrade the weight of each model according to posterior probability, average and covariance;

Step 23): by the weight after upgrading, average and covariance are upgraded posterior probability, again each pixel is divided in the class of posterior probability maximum, and the log-likelihood function of computed image.

If (2-4) log-likelihood function convergence, stops iteration, otherwise proceeds to step 22).

(2-5 utilizes bayesian criterion pixel to be divided into the model of posterior probability maximum.In 2 class results after cutting apart, selecting the larger class of average in two models is sky image, and another part is non-space image.

The invention has the advantages that and proposed a kind of image defogging method capable based on optimum contrast and minimum loss of information, segment the image into day dummy section and non-day dummy section by gauss hybrid models and expectation maximization algorithm, and zones of different adopts different atmospheric scattering illumination model method for parameter estimation, there is better accuracy and speed faster.Based on the atmosphere intensity of illumination of method estimated image sky dummy section of four points of tree iteration, there is better locality and the advantage such as speed faster.Spreading rate for non-day dummy section is estimated, the method piecemeal that adopts grid to divide, and model parameter has better locality.Based on the non-sky regional spread of the image rate of optimum contrast and the estimation of minimum information loss criterion, considering the maximized while of contrast, it is minimum that information loss is also wanted, and keeps edge and the grain details of former figure as far as possible.For the spreading rate estimation of sky dummy section, the constant coefficient rule of three of the average spreading rate of employing and non-day dummy section, avoids the color of sky part to be twisted or supersaturation, also can realize the natural transition with non-day dummy section.

Brief description of the drawings

Below in conjunction with the drawings and specific embodiments, the present invention is done further and illustrated, above-mentioned and/or otherwise advantage of the present invention will become apparent.

Fig. 1 is process flow diagram of the present invention.

Embodiment

This method is divided into cutting apart of sky and non-day dummy section, maximizes and minimized non-day dummy section mist elimination of information loss based on contrast, estimates day three parts of dummy section mist elimination based on normal scale-up factor method, and concrete workflow as shown in Figure 1.

Step 1, sets up the haze iconic model based on McCartney atmospheric scattering model

Suppose that the image irradiation model under haze condition is, haze image equals the imaging after by haze shield of atmosphere intensity of illumination that the imaging of original object image after haze transmission add infinite point and superposes

I _c(p)＝t(p)J _c(p)+(1-t(p))A _c(1)

Wherein, J _cand I (p) _c(p) represent respectively the pixel p of original image and observed image, c ∈ r, g, b) represent three kinds of Color Channels of red, green, blue.A _cbe illustrated in along the intensity of illumination of observer's direction of visual lines infinite point, be conventionally assumed to be global constant, irrelevant with local location p.T (p) ∈ [0,1] represents the spreading rate along light, is to be determined by the distance of camera and scene point, has reflected the ability of light penetration mist.Be worth greatlyr, show the light from scene surface reflection, the quantity that penetrating fog arrives observer visual field is more, and t (p) is inversely proportional to scene depth, can be calculated by following formula:

t(p)＝e ^-ρd(p)(2)

Wherein ρ is total scattering coefficient, has characterized the particle of unit volume to the scattering power of incident ray, and its value is larger, shows that the degree of incident ray generation scattering is more serious, is conventionally assumed to be 1.D (p) represents the depth of field of camera to pixel p.

Step 2, utilizes gauss hybrid models and expectation maximization algorithm to segment the image into day dummy section and non-day dummy section two classes

Atmosphere intensity of illumination A represents by the sky color of maximum bright spot in image conventionally, because the weather such as haze can cause the sky environment division of image brighter, and background is all darker with respect to the brightness of sky.Therefore, we are brighter sky part and other target and background two parts by the content of gauss hybrid models matching haze image, blending constituent.Then, utilize expectation maximization algorithm to solve the parameter of gauss hybrid models, and be partitioned into sky image with bayesian criterion.

If the N of haze image I pixel { p ₁, p ₂..., p _nindependent same distribution, wherein pixel p _ithe red, green and blue color value that (1≤i≤N) is corresponding is respectively I _r(p _i), I _g(p _i) and I _b(p _i), i.e. (p _i=(I _r(p _i), I _g(p _i), I _b(p _i)) ^t), haze gradation of image information is mixed and is formed by two darker and brighter Gaussian density function, pixel p _imixing probability density function P be expressed as:

$P\left(p_i\right)=\underset{r=1}{\overset{2}{\mathrm{\Σ}}}{\mathrm{\α}}_r{f}_r(p,{\mathrm{\θ}}_r)---\left(3\right)$

Wherein α _rrepresent the weight of r density component in hybrid density, meet α _r>=0 and the parameter θ of r density component _r={ μ _r, σ _r, μ _r, σ _rrepresent respectively average and variance, f _r(p _i, θ _r) be r Gaussian density function component.

The parameter step of estimating mixture model based on maximum expected value (Expectation Maximum) is as follows:

(2-1) adopting FCM Algorithms (FCM) is 2 clusters to haze image I initial division, and initializes weights, the parameters such as average and variance.

(2-2) calculating pixel point p _ibelong to respectively the posterior probability of model r (r=1,2) and upgrade the weight of each model r, average and covariance according to posterior probability:

(2-3) use these weights, average and covariance are upgraded posterior probability, again p _ibe divided in the class of posterior probability maximum, and the log-likelihood function of computed image.

If (2-4) log-likelihood function convergence, stops iteration, otherwise proceeds to (2-2).

(2-5) utilize bayesian criterion pixel p _ibe divided into the model of posterior probability maximum.In 2 class results after cutting apart, select average μ in two models _ra larger class is sky image.

Step 3, the image sky regional atmospheric intensity of illumination based on four points of tree iteration is estimated:

Sky image after cutting apart is carried out four points of tree location of iteration and estimates atmosphere intensity of illumination.Sky image is evenly divided into four according to the position at 1/2nd places of height and width to sky image, then calculate the draw brightness value maximum region of each piecemeal as the piecemeal of dividing next time, iteration is divided, and finishes until piecemeal is less than the threshold value (general value 5 X 5 or 7 X 7) of appointment.Using the mean value of last piecemeal color as atmosphere intensity of illumination.

Step 4, the method piecemeal of dividing based on grid, the spreading rate of each grid cell of the non-day dummy section of image based on optimum contrast and the estimation of minimum information loss criterion

The spreading rate of supposing day dummy section is t ₁, the spreading rate in non-space region is t ₂.And suppose that the depth of field has local similarity, image is carried out to grid division according to 16 × 16 sizes, suppose that the transfer rate of each grid is identical.According to the formula of haze iconic model (1), in the time that spreading rate is fixing, original image J _c(p) non-day dummy section estimated with following formula:

$J_c\left(p\right)=\frac{1}{t_2}(I_c\left(p\right)-A_c)+A_c---\left(7\right)$

Recover original image and only depend on spreading rate t ₂(p).Because the contrast of haze image is lower, we will estimate spreading rate according to maximizing the contrast of each grid and the criterion of minimum loss of information.

The each Color Channel J of original image _c(p) meet and be more than or equal to 0 and be less than or equal to 255 condition, according to formula (7), can obtain

$0\≤\frac{1}{t_2}(I_c\left(p\right)-A_c)+A_c\≤255---\left(8\right)$

Solution formula (8), spreading rate t ₂should meet following formula:

$t_2\≥\max \left\{\underset{c\∈\{r,g,b\}}{\min }\underset{p\∈B}{\min }\right\{\frac{I_c\left(p\right)-A_c}{-A_c}\},\underset{c\∈\{r,g,b\}}{\max }\underset{p\∈B}{\max }\{\frac{I_c\left(p\right)-A_c}{255-A_c}\left\}\right\}---\left(9\right)$

(4-1) the sky regional spread rate based on maximizing contrast is estimated

In computing grid region, each point is contrast with the quadratic sum of the difference of the average of this place grid

$C_{\mathrm{MSE}}=\underset{i=1}{\overset{N_B}{\mathrm{\Σ}}}\frac{{(I_c\left(p\right)-{\stackrel{\‾}{I}}_c\left(p\right))}^2}{{t_2}^2{N}_B}(p\∈B)---\left(10\right)$

Wherein for I _c(p) average of place grid B, N _bfor the pixel number of grid B.Known according to formula (10), spreading rate t ₂be inversely proportional to contrast.Therefore, spreading rate t ₂meet under the constraint of formula (9), in order to maximize contrast, the spreading rate of non-day dummy section is got minimum value:

$t_2=\max \left\{\underset{c\∈\{r,g,b\}}{\min }\underset{p\∈B}{\min }\right\{\frac{I_c\left(p\right)-A_c}{-A_c}\},\underset{c\∈\{r,g,b\}}{\max }\underset{p\∈B}{\max }\{\frac{I_c\left(p\right)-A_c}{255-A_c}\left\}\right\}---\left(11\right)$

(4-2) the sky regional spread rate based on information loss minimum is optimized

Optimum contrast can cause gray-scale value (0, the α of the least part of image _c) and the gray-scale value (β of largest portion _c, 255) (c ∈ r, g, b)) and corresponding information dropout, α _cand β _crepresent observed image I _cthe underflow in the region of blocking and overflow parameter value.Underflow parameter alpha _ccorresponding original image J _cgray-scale value 0, overflow parameter beta _ccorresponding original image J _cgray-scale value 255, respectively substitution formula (7), can obtain:

α _c＝(1-t ₂)A _c (12)

β _c＝255t ₂+(1-t ₂)A _c (13)

Calculate original image J (p) maximum-contrast change after underflow and overflow region area as information loss function.Calculate based on histogrammic information loss function, can obtain:

$E_{\mathrm{loss}}=\underset{c\∈\{r,g,b\}}{\mathrm{\Σ}}\{\underset{i=0}{\overset{{\mathrm{\α}}_c}{\mathrm{\Σ}}}{\left(\frac{i-A_c}{t_2}\right)+A_c)}^2{h}_c\left(i\right)+\underset{i={\mathrm{\β}}_c}{\overset{255}{\mathrm{\Σ}}}{(\frac{i-A_c}{t_2}+A_c-255)}^2{h}_c\left(i\right)\}---\left(14\right)$

Wherein h _c(i) be the histogram of pixel i at Color Channel c, A _cfor the atmosphere intensity of illumination of Color Channel c.In order to meet the requirement that maximizes contrast and minimum losses function simultaneously, according to Lagrange's multiplier function, be converted into the minimum problem of lower surface function.

E(t ₂,λ)＝-C _MSE+λE _loss (15)

λ is the relative importance weights of controlling contrast and information loss.T to formula (15) respectively ₂changes persuing is led with λ, sets up system of equations

$\left\{\begin{array}{c}\frac{\∂E(t_2,\mathrm{\λ})}{{\∂t}_2}=0\\ \frac{\∂E(t_2,\mathrm{\λ})}{\∂\mathrm{\λ}}=0\end{array}\right.---\left(16\right)$

The t of solution formula (16) ₂for the optimum spreading rate of non-day dummy section.

Step 5, the average spreading rate based on constant coefficient and non-day dummy section of image is estimated the spreading rate of day dummy section

Because the sky dummy section in image obviously wants large with respect to the depth of field of non-day dummy section, its corresponding spreading rate should be less.A lot of defogging method capables remove to estimate the spreading rate of day dummy section by non-sky regional spread rate method of estimation, cause noise and the color distortion of day dummy section.According to formula (2), some p that can day dummy section ₁spreading rate t ₁point p with non-space region ₂spreading rate t ₂scale-up factor:

$\frac{t_1}{t_2}=\frac{e^{-d\left(p_1\right)}}{e^{-d\left(p_2\right)}}=e^{d\left(p_2\right)-d\left(p_1\right)}=b---\left(17\right)$

Point p ₁depth of field d (p ₁), some p ₂depth of field d (p ₂), the difference of both distances is d (p ₂)-d (p ₁).The same day, the difference of dummy section and non-sky area pixel point depth of field distance was 1,2,5,10, and 15,20 o'clock, corresponding spreading rate scale-up factor b was as shown in table 1.Can know according to table 1, when the difference of distance is larger, coefficient is less.Under haze weather condition, general visibility is smaller, and day dummy section and the non-sky area pixel point depth of field are relatively little, and coefficient is relatively large.

The range difference of table 1 sky and non-sky pixel and spreading rate ratio

d(p 2)-d(p 1)	b
		-1	0.36787944117144
-2	0.13533528323661
		-5	0.00673794699909
-10	0.00004539992976
		-15	0.00000030590232
-20	0.00000000206115

The spreading rate of supposing day dummy section in image is all identical, is multiplied by the spreading rate of less relative coefficient calculating day dummy section by the spreading rate average of the non-day all grids of dummy section:

$t_1=b\stackrel{\‾}{t_2}---\left(18\right)$

Step 6, according to atmospheric scattering model, utilizes the model parameter of estimating, merges the Recovery image of output sky and non-day dummy section.

The optimum spreading rate t of the non-day dummy section solving according to formula (16) ₂atmosphere intensity of illumination A with the 3rd step estimation _cbring formula (1) into, non-day dummy section of Recovery image.According to the optimum spreading rate t of the sky dummy section of formula (18) ₁atmosphere intensity of illumination A with the 3rd step estimation _c, bring formula (1) into, the sky dummy section of Recovery image.Again two parts content is merged to piece image output.

Innovative point of the present invention comprises:

(1) sky district inclusion the priori such as haze degree and weather condition when image acquisition, and the depth of field of the two differs larger, the present invention is by gauss hybrid models and expect that peaked method is divided into day dummy section and non-day dummy section haze image, adopts different tactful Recovery images to have better specific aim and adaptability to the region after cutting apart.

(2) problem of haze image maximum is that picture contrast is lower, change contrast and can cause information loss, therefore the present invention adopts the method for balance contrast and information loss, both the overall picture quality of the contrast of image can be improved, the local detail such as texture and edge of image can also be retained.

It dummy section and the existing difference of non-day dummy section are also related, and difference is that the different grids of day dummy section have the close depth of field, and contact is all under identical external condition, to obtain.The present invention utilizes the average spreading rate of non-day dummy section, is multiplied by different hazes scale parameter all over the world, does not need the different grids of sky dummy section to estimate respectively, has realized again the self-adaptation of different images according to the average spreading rate of non-day dummy section.

The invention provides the image defogging method capable of a kind of optimum contrast and minimum information loss; method and the approach of this technical scheme of specific implementation are a lot; the above is only the preferred embodiment of the present invention; should be understood that; for those skilled in the art; under the premise without departing from the principles of the invention, can also make some improvements and modifications, these improvements and modifications also should be considered as protection scope of the present invention.In the present embodiment not clear and definite each ingredient all available prior art realized.

Claims (4)

1. an image defogging method capable for optimum contrast and minimum information loss, is characterized in that, comprises the steps:

Step 1, sets up the haze iconic model based on McCartney atmospheric scattering model;

Step 2, utilizes gauss hybrid models and expectation maximization algorithm to segment the image into day dummy section and non-day dummy section two classes;

Step 3, the atmosphere intensity of illumination of the method estimated image sky dummy section based on four points of tree iteration;

Step 4, the method piecemeal of dividing based on grid, the spreading rate of each grid cell of the non-day dummy section of image based on optimum contrast and the estimation of minimum information loss criterion;

Step 5, the normal scale-up factor method of the average spreading rate of non-day dummy section of employing and image is estimated the spreading rate of day dummy section;

Step 6, according to atmospheric scattering model, utilizes the model parameter of estimating, merges the Recovery image of output sky and non-day dummy section.

2. the image defogging method capable of a kind of optimum contrast according to claim 1 and minimum information loss, is characterized in that, the image irradiation model under haze condition is:

I _c(p)＝t(p)J _c(p)+(1-t(p))A _c，

Wherein, J _cand I (p) _c(p) represent respectively the pixel p of original image and observed image, c ∈ r, g, b) represent three kinds of Color Channels of red, green, blue; A _cbe illustrated in along the atmosphere intensity of illumination of observer's direction of visual lines infinite point, t (p) ∈ [0,1] represents the spreading rate along light, and t (p) is inversely proportional to scene depth, and computing formula is:

t(p)＝e ^-ρd(p)，

Wherein ρ is total scattering coefficient, and d (p) represents the depth of field of camera to pixel p.

3. the image defogging method capable of a kind of optimum contrast according to claim 1 and minimum information loss, is characterized in that, in described step 2,

If the N of haze image I pixel { p ₁, p ₂..., p _nindependent same distribution, wherein pixel p _icorresponding red, green and blue color value is respectively I _r(p _i), I _g(p _i) and I _b(p _i), 1≤i≤N, i.e. (p _i=(I _r(p _i), I _g(p _i), I _b(p _i)) ^t); Haze gradation of image information is mixed and is formed by two darker and brighter Gaussian density function, pixel p _imixing probability density function P be expressed as:

$P\left(p_i\right)=\underset{r=1}{\overset{2}{\mathrm{\Σ}}}{\mathrm{\α}}_r{f}_r(p_i,{\mathrm{\θ}}_r),$

Wherein α _rrepresent the weight of r density component in hybrid density, meet α _r>=0 and θ _r={ μ _r, σ _rr density component parameters, wherein μ _r, σ _rrepresent respectively average and variance, f _r(p _i, θ _r) be r Gaussian density function component.

Comprise the steps:

Step 21: adopting FCM Algorithms is two clusters to haze image I initial division, and initializes weights, average and variance;

Step 22: calculate the posterior probability that each pixel belongs to respectively two models, and upgrade the weight of each model according to posterior probability, average and covariance;

Step 23: by the weight after upgrading, average and covariance are upgraded posterior probability, is divided into each pixel in the class of posterior probability maximum again, and the log-likelihood function of computed image.

Step 24: if log-likelihood function convergence stops iteration and carry out step 25, otherwise returns to step 22;

Step 25: utilize bayesian criterion pixel to be divided into the model of posterior probability maximum, in two class results after cutting apart, selecting the larger class of average in two models is sky image, and another part is non-space image.

4. the image defogging method capable of a kind of optimum contrast according to claim 1 and minimum information loss, is characterized in that, in step 3, the sky image after cutting apart is carried out four points of tree location of iteration and estimates atmosphere intensity of illumination; Sky image is evenly divided into four according to the position at 1/2nd places of height and width to sky image, then calculate the draw brightness value maximum region of each piecemeal as the piecemeal of dividing next time, iteration is divided, and finishes until piecemeal is less than the threshold value of appointment; Using the mean value of last piecemeal color as atmosphere intensity of illumination.