CN104657954A - Human vision similarity constraint based image dehazing method, system and equipment - Google Patents

Abstract

The invention discloses a human vision similarity constraint based image dehazing method, system and equipment. The method comprises steps as follows: S1, color minimum channel data of a to-be-processed hazy image are computed; S2, the hazy image is subjected to two-channel joint non-local similarity constraint federated filtering twice respectively to obtain atmospheric scattering data; S3, the hazy image is reconstructed according to the obtained atmospheric scattering data so as to obtain an initialized hazy image; S4, the initialized hazy image is subjected to three-channel joint similarity constraint federated filtering respectively to obtain the image after dehazing. According to the human vision similarity constraint based image dehazing method, system and equipment, the atmospheric scattering data can be estimated accurately, the dehazing image is reconstructed accurately, the accuracy is high, the noise is low, and the method, the system and the equipment can be widely applied in the field of digital image processing.

Description

Based on the image defogging method capable of human eye vision similarity constraint, system and equipment

Technical field

The present invention relates to digital image processing field, particularly relate to a kind of image defogging method capable based on human eye vision similarity constraint, system and equipment.

Background technology

In the shooting situation having mist, image capture device is owing to being subject to the scattering process of aerosol, cause the contrast of the digital picture photographed and color larger decay occurs and degrades, thus affect the validity and reliability of later image/video monitoring, graphical analysis.There is a variety of image defogging method capable in prior art, carry out mist elimination process at present mainly adopt dark primary priori defogging method capable to realize to image, the method is by obtaining a large amount of statistical law obtained without mist image viewing.Dark primary priori defogging method capable is succinctly effective, to various types of mist elimination effect that can acquire a certain degree containing mist image.But; dark primary priori defogging method capable can not directly act on view picture natural image; owing to usually can undergo mutation in the edge of scenery in the natural image Scene degree of depth; after causing adopting dark primary priori defogging method capable to carry out mist elimination process; obvious halo effect can be produced in edge; the precision of mist elimination is low, and after may causing mist elimination, the noise of image rises.

Summary of the invention

In order to solve above-mentioned technical matters, the object of this invention is to provide the image defogging method capable based on human eye vision similarity constraint, another object of the present invention is to provide a kind of image mist elimination system based on human eye vision similarity constraint, and another object of the present invention is to provide a kind of image mist elimination equipment based on human eye vision similarity constraint.

The technical solution adopted for the present invention to solve the technical problems is:

Based on the image defogging method capable of human eye vision similarity constraint, comprising:

S1, calculate the pending color smallest passage data having mist image;

S2, to the non local similarity constraint Federated filter of associating having mist image to carry out 2 two passages respectively, and then obtain atmospheric scattering data;

The atmospheric scattering data reconstruction mist elimination image that S3, utilization obtain also obtains initialization mist elimination image;

S4, three-channel associating similarity constraint Federated filter is carried out to initialization mist elimination image, obtain the image after mist elimination.

Further, described step S1, it is specially:

The pending color smallest passage data having mist image are calculated according to following formula:

M _min(l)＝min(P _R(l),P _Y(l),P _B(l))

In above formula, l is natural number, represents the position of pixel, M _minl () represents the color smallest passage data of pixel l, P _r(l), P _y(l), P _bl () represents pending respectively has mist image M at red, yellow, blue three color components of pixel l.

Further, described step S2, comprising:

S21, carry out the non local similarity constraint Federated filter of associating of 2 two passages according to following formula respectively to there being mist image and obtaining two filtering images:

$\left\{\begin{array}{c}C1=\mathrm{SDL}(M,P_y,N,{\mathrm{\β}}_1)\\ C2=\mathrm{SDL}(M,P_y,N,{\mathrm{\β}}_2)\end{array}\right.$

In above formula, C1, C2 represent two filtering images of acquisition respectively, and M represents pending has mist image, P _yindicate the luminance components of mist image M, N is natural number, represents the number of each pixel being carried out to the pixel participated in required for Federated filter, β ₁, β ₂all represent filtering strength;

S22, according to obtain two filtering images calculate the error image D meeting visual characteristics of human eyes:

D(l)＝|C1(l)-C2(l)|

In above formula, l is natural number, and represent the position of pixel, D (l) represents the value of error image D at pixel l, and C1 (l), C2 (l) represent the value of two filtering images at pixel l of acquisition respectively;

S23, calculate the initial estimate obtaining atmospheric scattering light according to following formula:

$\mathrm{\φ}\left(l\right)=C1\left(l\right)-\frac{C1\left(l\right)}{0.42\·\mathrm{\θ}}\·D\left(l\right)$

In above formula, φ (l) represents the initial estimate of atmospheric scattering light at pixel l, and θ represents the average brightness value of C1 (l);

S24, revise rear acquisition atmospheric scattering data according to the initial estimate of following formula to atmospheric scattering light:

ψ(l)＝max(min(0.91·φ(l),C1(l)),0)

In above formula, ψ (l) represents the atmospheric scattering data at pixel l.

Further, in described step S21, N=40, β ₁=0.01, β ₂=0.07.

Further, in described step S21, the formula that the non local similarity constraint Federated filter of associating carrying out two passages adopts is SDL (im1, im2, num, β), wherein im1, im2 are two measure-alike single channel image, num is for natural number and represent the number of each pixel being carried out to the pixel participated in required for Federated filter, and β represents filtering strength, and the filtering image finally obtained is:

i(k)＝γ ₁(k)·im1(k)+γ ₂(k)

In above formula, k is natural number, and represent the position of pixel, i (k) represents the value of filtering image at pixel k, and im1 (k) represents the value of the pixel k of single channel image im1, γ ₁k () represents the linear coefficient of the filtering of the pixel k of single channel image im1, γ ₂k () represents the biased coefficient of the filtering of the pixel k of single channel image im1, γ ₁(k) and γ ₂k () is obtained by following formulae discovery:

$\left\{\begin{array}{c}{\mathrm{\γ}}_1\left(k\right)=\frac{0.004\·{\mathrm{\Σ}}_{l\∈{\mathrm{SK}}_k}{\mathrm{im}1}_{\min }{\left(l\right)}^2-\mathrm{\μ}{\left(k\right)}^2}{\mathrm{\σ}\left(k\right)+\mathrm{\β}}\\ {\mathrm{\γ}}_2\left(k\right)=\mathrm{\μ}\left(k\right)-{\mathrm{\γ}}_1\left(k\right)\·\mathrm{\μ}\left(k\right)\end{array}\right.$

In above formula, im1 _minl () represents the color smallest passage data of the pixel l of single channel image im1, SK _krepresent the set of the individual similar pixel of the closest num of pixel k, σ (k) represents the standard variance of the similar pixel of the pixel k of single channel image im1, μ (k) represents the similar pixel average of the pixel k of single channel image im1, and σ (k) and μ (k) is obtained by following formulae discovery:

$\left\{\begin{array}{c}\mathrm{\σ}\left(k\right)=0.04\·\underset{l\∈{\mathrm{SK}}_k}{\mathrm{\Σ}}|{\mathrm{im}1}_{\min }\left(l\right)-\mathrm{\μ}\left(k\right)|\\ \mathrm{\μ}\left(k\right)=0.04\·\underset{l\∈{\mathrm{SK}}_k}{\mathrm{\Σ}}{\mathrm{im}1}_{\min }\left(l\right)\end{array}\right.$

Further, the S set K of the individual similar pixel of the closest num of pixel k _kbe the set of combining the minimum num of a similarity pixel with k, the computing formula of described associating similarity is as follows:

$S(k1,k2)=\underset{l=1}{\overset{25}{\mathrm{\Σ}}}|{\mathrm{blk}}_{\mathrm{im}1}^{k1}\left(l\right)-{\mathrm{blk}}_{\mathrm{im}1}^{k2}\left(l\right)|+\mathrm{\λ}\underset{l=1}{\overset{25}{\mathrm{\Σ}}}|{\mathrm{blk}}_{\mathrm{im}2}^{k1}\left(l\right)-{\mathrm{blk}}_{\mathrm{im}2}^{k2}\left(l\right)|$

In above formula, k1, k2 are natural number, all represent the position of pixel, and S (k1, k2) represents pixel k1, k2 associating similarity on single channel image im1, im2, represent the value of the pixel l on the image block of 5 × 5 on single channel image im1 centered by pixel k1, represent the value of the pixel l on the image block of 5 × 5 on single channel image im1 centered by pixel k2, represent the value of the pixel l on the image block of 5 × 5 on single channel image im2 centered by pixel k1, represent the value of the pixel l on the image block of 5 × 5 on single channel image im2 centered by pixel k2, λ represents scale-up factor.

Further, described step S3, comprising:

S31, search and obtain pending have the brightness of mist image front 1.2% the location of pixels set of pixel;

S32, according to the location of pixels set obtained, search the overall brightness Γ of brightness maximum value as mist light having in mist image;

S33, calculate air ature of coal spreading rate according to following formula:

$\mathrm{\ρ}\left(l\right)=1-0.943\·\frac{\mathrm{\ψ}\left(l\right)}{\mathrm{\Γ}}$

In above formula, ρ (l) represents the air ature of coal spreading rate at pixel l, and ψ (l) represents the atmospheric scattering data at pixel l;

S34, rebuild mist elimination image obtain initialization mist elimination image according to following formula:

$\left\{\begin{array}{c}{P}_R^o\left(l\right)=\frac{P_R\left(l\right)-\mathrm{\Γ}}{\max (\mathrm{\ρ}\left(l\right),\mathrm{0,11})}+\mathrm{\Γ}\\ P_G^o\left(l\right)=\frac{P_G\left(l\right)-\mathrm{\Γ}}{\max (\mathrm{\ρ}\left(l\right),\mathrm{0,11})}+\mathrm{\Γ}\\ P_B^o\left(l\right)=\frac{P_B\left(l\right)-\mathrm{\Γ}}{\max (\mathrm{\ρ}\left(l\right),\mathrm{0,11})}+\mathrm{\Γ}\end{array}\right.$

In above formula, represent red, green, blue three color components of initialization mist elimination image respectively, P _r(l), P _y(l), P _bl () represents pending respectively has mist image M at red, yellow, blue three color components of pixel l.

Further, described step S4, it is specially:

According to following formula, three-channel associating similarity constraint Federated filter is carried out to initialization mist elimination image, obtains the image after mist elimination:

$\left\{\begin{array}{c}{I}_R=\mathrm{JDN}(P_R^o,P_G^o,P_B^o,\mathrm{49,0.002})\\ I_G=\mathrm{JDN}(P_G^o,P_R^o,P_B^o,\mathrm{49,0.002})\\ I_B=\mathrm{JDN}(P_B^o,P_G^o,P_R^o,\mathrm{49,0.002})\end{array}\right.$

In above formula, I _r, I _g, I _brepresent red, green, blue three color components of the image after mist elimination respectively, represent red, green, blue three single channel image of initialization mist elimination image respectively.

Further, carrying out the formula that three-channel associating similarity constraint Federated filter adopts is JDN (im1, im2, im3, num, β), wherein im1, im2, im3 are three measure-alike single channel image, num is for natural number and represent the number of each pixel being carried out to the pixel participated in required for Federated filter, and β represents filtering strength, and the filtering image finally obtained is:

O(k)＝η ₁(k)·im1(k)+η ₂(k)

In above formula, k is natural number, and represent the position of pixel, O (k) represents the value of filtering image at pixel k, and im1 (k) represents the value of the pixel k of single channel image im1, η ₁k () represents the linear coefficient of the filtering of the pixel k of single channel image im1, η ₂k () represents the biased coefficient of the filtering of the pixel k of single channel image im1, η ₁(k) and η ₂k () is obtained by following formulae discovery:

In above formula, im1 _minl () represents the color smallest passage data of the pixel l of single channel image im1, SK _krepresent the set of the individual similar pixel of the closest num of pixel k, represent the standard variance of the similar pixel of the pixel k of single channel image im1, represent the similar pixel average of the pixel k of single channel image im1, with obtained by following formulae discovery:

Further, the S set K of the individual similar pixel of the closest num of pixel k _kbe the set of combining the minimum num of a similarity pixel with k, the computing formula of described associating similarity is as follows:

$\begin{array}{c}{S}^{\′}(k1,k2)=\underset{m=1}{\overset{25}{\mathrm{\Σ}}}|{\mathrm{blk}}_{\mathrm{im}1}^{k1}\left(m\right)-{\mathrm{blk}}_{\mathrm{im}1}^{k2}\left(m\right)|+\mathrm{\λ}\underset{m=1}{\overset{25}{\mathrm{\Σ}}}|{\mathrm{blk}}_{\mathrm{im}2}^{k1}\left(m\right)-{\mathrm{blk}}_{\mathrm{im}2}^{k2}\left(m\right)|\\ +\mathrm{\λ}\underset{m=1}{\overset{25}{\mathrm{\Σ}}}|{\mathrm{blk}}_{\mathrm{im}3}^{k1}\left(m\right)-{\mathrm{blk}}_{\mathrm{im}3}^{k2}\left(m\right)|\end{array}$

In above formula, k1, k2 are natural number, all represent the position of pixel, S'(k1, k2) represent pixel k1, k2 associating similarity on single channel image im1, im2, im3, represent the value of the pixel l on the image block of 5 × 5 on single channel image im1 centered by pixel k1, represent the value of the pixel l on the image block of 5 × 5 on single channel image im1 centered by pixel k2, represent the value of the pixel l on the image block of 5 × 5 on single channel image im2 centered by pixel k1, represent the value of the pixel l on the image block of 5 × 5 on single channel image im2 centered by pixel k2, represent the value of the pixel l on the image block of 5 × 5 on single channel image im3 centered by pixel k1, represent the value of the pixel l on the image block of 5 × 5 on single channel image im3 centered by pixel k2, λ represents scale-up factor.

Another technical scheme that the present invention adopts for its technical matters of solution is:

Based on the image mist elimination system of human eye vision similarity constraint, comprising:

First processing module, for calculating the pending color smallest passage data having mist image;

Second processing module, for the non local similarity constraint Federated filter of associating having mist image to carry out 2 two passages respectively, and then obtains atmospheric scattering data;

3rd processing module, for utilizing the atmospheric scattering data reconstruction mist elimination image of acquisition and obtaining initialization mist elimination image;

4th processing module, for carrying out three-channel associating similarity constraint Federated filter to initialization mist elimination image, obtains the image after mist elimination.

The another technical scheme that the present invention adopts for its technical matters of solution is:

Based on the image mist elimination equipment of human eye vision similarity constraint, the processor comprising camera and be connected with camera, described camera is for gathering image, and described processor is used for carrying out mist elimination process to the image that camera gathers, and described processor comprises:

First processing module, for calculating the pending color smallest passage data having mist image;

Second processing module, for the non local similarity constraint Federated filter of associating having mist image to carry out 2 two passages respectively, and then obtains atmospheric scattering data;

3rd processing module, for utilizing the atmospheric scattering data reconstruction mist elimination image of acquisition and obtaining initialization mist elimination image;

4th processing module, for carrying out three-channel associating similarity constraint Federated filter to initialization mist elimination image, obtains the image after mist elimination.

The invention has the beneficial effects as follows: the image defogging method capable based on human eye vision similarity constraint of the present invention, comprising: S1, calculate the pending color smallest passage data having mist image; S2, to the non local similarity constraint Federated filter of associating having mist image to carry out 2 two passages respectively, and then obtain atmospheric scattering data; The atmospheric scattering data reconstruction mist elimination image that S3, utilization obtain also obtains initialization mist elimination image; S4, three-channel associating similarity constraint Federated filter is carried out to initialization mist elimination image, obtain the image after mist elimination.This method accurately can estimate atmospheric scattering light data, thus rebuilds mist elimination image exactly, and precision is high and noise is low.

Another beneficial effect of the present invention is: based on the image mist elimination system of human eye vision similarity constraint, comprising: the first processing module, for calculating the pending color smallest passage data having mist image; Second processing module, for the non local similarity constraint Federated filter of associating having mist image to carry out 2 two passages respectively, and then obtains atmospheric scattering data; 3rd processing module, for utilizing the atmospheric scattering data reconstruction mist elimination image of acquisition and obtaining initialization mist elimination image; 4th processing module, for carrying out three-channel associating similarity constraint Federated filter to initialization mist elimination image, obtains the image after mist elimination.Native system accurately can estimate atmospheric scattering light data, thus rebuilds mist elimination image exactly, and precision is high and noise is low.

Another beneficial effect of the present invention is: based on the image mist elimination equipment of human eye vision similarity constraint, the processor comprising camera and be connected with camera, described camera is for gathering image, described processor is used for carrying out mist elimination process to the image that camera gathers, described processor comprises: the first processing module, for calculating the pending color smallest passage data having mist image; Second processing module, for the non local similarity constraint Federated filter of associating having mist image to carry out 2 two passages respectively, and then obtains atmospheric scattering data; 3rd processing module, for utilizing the atmospheric scattering data reconstruction mist elimination image of acquisition and obtaining initialization mist elimination image; 4th processing module, for carrying out three-channel associating similarity constraint Federated filter to initialization mist elimination image, obtains the image after mist elimination.This equipment accurately can estimate atmospheric scattering light data, thus rebuilds mist elimination image exactly, provides precision high and the low output image of noise.

Accompanying drawing explanation

Below in conjunction with drawings and Examples, the invention will be further described.

Fig. 1 is the process flow diagram of the image mist elimination based on human eye vision similarity constraint of the present invention.

Embodiment

With reference to Fig. 1, the invention provides a kind of image defogging method capable based on human eye vision similarity constraint, comprising:

S1, calculate the pending color smallest passage data having mist image;

S2, to the non local similarity constraint Federated filter of associating having mist image to carry out 2 two passages respectively, and then obtain atmospheric scattering data;

The atmospheric scattering data reconstruction mist elimination image that S3, utilization obtain also obtains initialization mist elimination image;

S4, three-channel associating similarity constraint Federated filter is carried out to initialization mist elimination image, obtain the image after mist elimination.

Be further used as preferred embodiment, described step S1, it is specially:

The pending color smallest passage data having mist image are calculated according to following formula:

M _min(l)＝min(P _R(l)，P _Y(l)，P _B(l))

In above formula, l is natural number, represents the position of pixel, M _minl () represents the color smallest passage data of pixel l, P _r(l), P _y(l), P _bl () represents pending respectively has mist image M at red, yellow, blue three color components of pixel l.

Be further used as preferred embodiment, described step S2, comprising:

S21, carry out the non local similarity constraint Federated filter of associating of 2 two passages according to following formula respectively to there being mist image and obtaining two filtering images:

$\left\{\begin{array}{c}C1=\mathrm{SDL}(M,P_y,N,{\mathrm{\β}}_1)\\ C2=\mathrm{SDL}(M,P_y,N,{\mathrm{\β}}_2)\end{array}\right.$

In above formula, C1, C2 represent two filtering images of acquisition respectively, and M represents pending has mist image, P _yindicate the luminance components of mist image M, N is natural number, represents the number of each pixel being carried out to the pixel participated in required for Federated filter, β ₁, β ₂all represent filtering strength;

S22, according to obtain two filtering images calculate the error image D meeting visual characteristics of human eyes:

D(l)＝|C1(l)-C2(l)|

In above formula, l is natural number, and represent the position of pixel, D (l) represents the value of error image D at pixel l, and C1 (l), C2 (l) represent the value of two filtering images at pixel l of acquisition respectively;

S23, calculate the initial estimate obtaining atmospheric scattering light according to following formula:

$\mathrm{\φ}\left(l\right)=C1\left(l\right)-\frac{C1\left(l\right)}{0.42\·\mathrm{\θ}}\·D\left(l\right)$

In above formula, φ (l) represents the initial estimate of atmospheric scattering light at pixel l, and θ represents the average brightness value of C1 (l);

S24, revise rear acquisition atmospheric scattering data according to the initial estimate of following formula to atmospheric scattering light:

ψ(l)＝max(min(0.91·φ(l),C1(l)),0)

In above formula, ψ (l) represents the atmospheric scattering data at pixel l.

Be further used as preferred embodiment, in described step S21, N=40, β ₁=0.01, β ₂=0.07.

Be further used as preferred embodiment, in described step S21, the formula that the non local similarity constraint Federated filter of associating carrying out two passages adopts is SDL (im1, im2, num, β), wherein im1, im2 are two measure-alike single channel image, num is for natural number and represent the number of each pixel being carried out to the pixel participated in required for Federated filter, and β represents filtering strength, and the filtering image finally obtained is:

i(k)＝γ ₁(k)·im1(k)+γ ₂(k)

In above formula, k is natural number, and represent the position of pixel, i (k) represents the value of filtering image at pixel k, and im1 (k) represents the value of the pixel k of single channel image im1, γ ₁k () represents the linear coefficient of the filtering of the pixel k of single channel image im1, γ ₂k () represents the biased coefficient of the filtering of the pixel k of single channel image im1, γ ₁(k) and γ ₂k () is obtained by following formulae discovery:

$\left\{\begin{array}{c}{\mathrm{\γ}}_1\left(k\right)=\frac{0.004\·{\mathrm{\Σ}}_{l\∈{\mathrm{SK}}_k}{\mathrm{im}1}_{\min }{\left(l\right)}^2-\mathrm{\μ}{\left(k\right)}^2}{\mathrm{\σ}\left(k\right)+\mathrm{\β}}\\ {\mathrm{\γ}}_2\left(k\right)=\mathrm{\μ}\left(k\right)-{\mathrm{\γ}}_1\left(k\right)\·\mathrm{\μ}\left(k\right)\end{array}\right.$

In above formula, im1 _minl () represents the color smallest passage data of the pixel l of single channel image im1, SK _krepresent the set of the individual similar pixel of the closest num of pixel k, σ (k) represents the standard variance of the similar pixel of the pixel k of single channel image im1, μ (k) represents the similar pixel average of the pixel k of single channel image im1, and σ (k) and μ (k) is obtained by following formulae discovery:

$\left\{\begin{array}{c}\mathrm{\σ}\left(k\right)=0.04\·\underset{l\∈{\mathrm{SK}}_k}{\mathrm{\Σ}}|{\mathrm{im}1}_{\min }\left(l\right)-\mathrm{\μ}\left(k\right)|\\ \mathrm{\μ}\left(k\right)=0.04\·\underset{l\∈{\mathrm{SK}}_k}{\mathrm{\Σ}}{\mathrm{im}1}_{\min }\left(l\right)\end{array}\right.$

Be further used as preferred embodiment, the S set K of the individual similar pixel of closest num of pixel k _kbe the set of combining the minimum num of a similarity pixel with k, the computing formula of described associating similarity is as follows:

$S(k1,k2)=\underset{l=1}{\overset{25}{\mathrm{\Σ}}}|{\mathrm{blk}}_{\mathrm{im}1}^{k1}\left(l\right)-{\mathrm{blk}}_{\mathrm{im}1}^{k2}\left(l\right)|+\mathrm{\λ}\underset{l=1}{\overset{25}{\mathrm{\Σ}}}|{\mathrm{blk}}_{\mathrm{im}2}^{k1}\left(l\right)-{\mathrm{blk}}_{\mathrm{im}2}^{k2}\left(l\right)|$

In above formula, k1, k2 are natural number, all represent the position of pixel, and S (k1, k2) represents pixel k1, k2 associating similarity on single channel image im1, im2, represent the value of the pixel l on the image block of 5 × 5 on single channel image im1 centered by pixel k1, represent the value of the pixel l on the image block of 5 × 5 on single channel image im1 centered by pixel k2, represent the value of the pixel l on the image block of 5 × 5 on single channel image im2 centered by pixel k1, represent the value of the pixel l on the image block of 5 × 5 on single channel image im2 centered by pixel k2, λ represents scale-up factor.

Be further used as preferred embodiment, described step S3, comprising:

S31, search and obtain pending have the brightness of mist image front 1.2% the location of pixels set of pixel;

S32, according to the location of pixels set obtained, search the overall brightness Γ of brightness maximum value as mist light having in mist image;

S33, calculate air ature of coal spreading rate according to following formula:

$\mathrm{\ρ}\left(l\right)=1-0.943\·\frac{\mathrm{\ψ}\left(l\right)}{\mathrm{\Γ}}$

In above formula, ρ (l) represents the air ature of coal spreading rate at pixel l, and ψ (l) represents the atmospheric scattering data at pixel l;

S34, rebuild mist elimination image obtain initialization mist elimination image according to following formula:

$\left\{\begin{array}{c}{P}_R^o\left(l\right)=\frac{P_R\left(l\right)-\mathrm{\Γ}}{\max (\mathrm{\ρ}\left(l\right),\mathrm{0,11})}+\mathrm{\Γ}\\ P_G^o\left(l\right)=\frac{P_G\left(l\right)-\mathrm{\Γ}}{\max (\mathrm{\ρ}\left(l\right),\mathrm{0,11})}+\mathrm{\Γ}\\ P_B^o\left(l\right)=\frac{P_B\left(l\right)-\mathrm{\Γ}}{\max (\mathrm{\ρ}\left(l\right),\mathrm{0,11})}+\mathrm{\Γ}\end{array}\right.$

In above formula, represent red, green, blue three color components of initialization mist elimination image respectively, P _r(l), P _y(l), P _bl () represents pending respectively has mist image M at red, yellow, blue three color components of pixel l.

Be further used as preferred embodiment, described step S4, it is specially:

According to following formula, three-channel associating similarity constraint Federated filter is carried out to initialization mist elimination image, obtains the image after mist elimination:

$\left\{\begin{array}{c}{I}_R=\mathrm{JDN}(P_R^o,P_G^o,P_B^o,\mathrm{49,0.002})\\ I_G=\mathrm{JDN}(P_G^o,P_R^o,P_B^o,\mathrm{49,0.002})\\ I_B=\mathrm{JDN}(P_B^o,P_G^o,P_R^o,\mathrm{49,0.002})\end{array}\right.$

In above formula, I _r, I _g, I _brepresent red, green, blue three color components of the image after mist elimination respectively, represent red, green, blue three single channel image of initialization mist elimination image respectively.

Be further used as preferred embodiment, carrying out the formula that three-channel associating similarity constraint Federated filter adopts is JDN (im1, im2, im3, num, β), wherein im1, im2, im3 are three measure-alike single channel image, num is for natural number and represent the number of each pixel being carried out to the pixel participated in required for Federated filter, and β represents filtering strength, and the filtering image finally obtained is:

O(k)＝η ₁(k)·im1(k)+η ₂(k)

In above formula, k is natural number, and represent the position of pixel, O (k) represents the value of filtering image at pixel k, and im1 (k) represents the value of the pixel k of single channel image im1, η ₁k () represents the linear coefficient of the filtering of the pixel k of single channel image im1, η ₂k () represents the biased coefficient of the filtering of the pixel k of single channel image im1, η ₁(k) and η ₂k () is obtained by following formulae discovery:

In above formula, represent the color smallest passage data of the pixel l of single channel image im1, SK _krepresent the set of the individual similar pixel of the closest num of pixel k, represent the standard variance of the similar pixel of the pixel k of single channel image im1, represent the similar pixel average of the pixel k of single channel image im1, with obtained by following formulae discovery:

Be further used as preferred embodiment, the S set K of the individual similar pixel of closest num of pixel k _kbe the set of combining the minimum num of a similarity pixel with k, the computing formula of described associating similarity is as follows:

$\begin{array}{c}{S}^{\′}(k1,k2)=\underset{m=1}{\overset{25}{\mathrm{\Σ}}}|{\mathrm{blk}}_{\mathrm{im}1}^{k1}\left(m\right)-{\mathrm{blk}}_{\mathrm{im}1}^{k2}\left(m\right)|+\mathrm{\λ}\underset{m=1}{\overset{25}{\mathrm{\Σ}}}|{\mathrm{blk}}_{\mathrm{im}2}^{k1}\left(m\right)-{\mathrm{blk}}_{\mathrm{im}2}^{k2}\left(m\right)|\\ +\mathrm{\λ}\underset{m=1}{\overset{25}{\mathrm{\Σ}}}|{\mathrm{blk}}_{\mathrm{im}3}^{k1}\left(m\right)-{\mathrm{blk}}_{\mathrm{im}3}^{k2}\left(m\right)|\end{array}$

In above formula, k1, k2 are natural number, all represent the position of pixel, S'(k1, k2) represent pixel k1, k2 associating similarity on single channel image im1, im2, im3, represent the value of the pixel l on the image block of 5 × 5 on single channel image im1 centered by pixel k1, represent the value of the pixel l on the image block of 5 × 5 on single channel image im1 centered by pixel k2, represent the value of the pixel l on the image block of 5 × 5 on single channel image im2 centered by pixel k1, represent the value of the pixel l on the image block of 5 × 5 on single channel image im2 centered by pixel k2, represent the value of the pixel l on the image block of 5 × 5 on single channel image im3 centered by pixel k1, represent the value of the pixel l on the image block of 5 × 5 on single channel image im3 centered by pixel k2, λ represents scale-up factor.

Below in conjunction with specific embodiment, the invention will be further described.

Embodiment one

With reference to Fig. 1, a kind of image defogging method capable based on human eye vision similarity constraint, comprising:

S1, calculate pending color smallest passage data having mist image according to following formula:

M _min(l)＝min(P _R(l),P _Y(l),P _B(l))

In above formula, l is natural number, represents the position of pixel, M _minl () represents the color smallest passage data of pixel l, P _r(l), P _y(l), P _bl () represents pending respectively has mist image M at red, yellow, blue three color components of pixel l.

S2, to the non local similarity constraint Federated filter of associating having mist image to carry out 2 two passages respectively, and then obtain atmospheric scattering data.

Define the non local similarity constraint Federated filter of associating of two passages: SDL (im1, im2, num, β), wherein im1, im2 are two measure-alike single channel image, num is for natural number and represent the number of each pixel being carried out to the pixel participated in required for Federated filter, and β represents filtering strength.

In single channel image im1, im2 that 2 width yardsticks are identical, num the similar pixel point finding it is combined for each pixel k.

The neighborhood window of definition pixel k 5 × 5 rectangular windows that to be Blk (k) be centered by pixel k, then for any single channel image im, can obtain the image block centered by pixel k wherein (l=1,2 ... 25).

The associating similarity of any two pixel k1 and k2 in single channel image im1 and im2 is:

$S(k1,k2)=\underset{l=1}{\overset{25}{\mathrm{\Σ}}}|{\mathrm{blk}}_{\mathrm{im}1}^{k1}\left(l\right)-{\mathrm{blk}}_{\mathrm{im}1}^{k2}\left(l\right)|+\mathrm{\λ}\underset{l=1}{\overset{25}{\mathrm{\Σ}}}|{\mathrm{blk}}_{\mathrm{im}2}^{k1}\left(l\right)-{\mathrm{blk}}_{\mathrm{im}2}^{k2}\left(l\right)|$

In above formula, k1, k2 are natural number, all represent the position of pixel, and S (k1, k2) represents pixel k1, k2 associating similarity on single channel image im1, im2, represent the value of the pixel l on the image block of 5 × 5 on single channel image im1 centered by pixel k1, represent the value of the pixel l on the image block of 5 × 5 on single channel image im1 centered by pixel k2, represent the value of the pixel l on the image block of 5 × 5 on single channel image im2 centered by pixel k1, represent the value of the pixel l on the image block of 5 × 5 on single channel image im2 centered by pixel k2, λ represents scale-up factor, and wherein λ is preferably 0.25.

Calculate the S (k1, k2) obtained less, the associating similarity of k1, k2 two pixels on im1, im2 is more similar.With whole image for hunting zone, the individual similar pixel SK of closest num can be found for each pixel k _k={ k ₁, k ₂..., k _num.

Be calculated as follows filtering parameter successively:

For each pixel im1 (k) of single channel image im1, calculate its similar pixel average:

$\mathrm{\μ}\left(k\right)=0.04\·\underset{l\∈{\mathrm{SK}}_k}{\mathrm{\Σ}}{\mathrm{im}1}_{\min }\left(l\right)$

For each pixel im1 (k) of single channel image im1, calculate the standard variance of its similar pixel:

$\mathrm{\σ}\left(k\right)=0.04\·\underset{l\∈{\mathrm{SK}}_k}{\mathrm{\Σ}}|{\mathrm{im}1}_{\min }\left(l\right)-\mathrm{\μ}\left(k\right)|$

For each pixel im1 (k) of single channel image im1, calculate the linear coefficient of its filtering:

${\mathrm{\γ}}_1\left(k\right)=\frac{0.004\·{\mathrm{\Σ}}_{l\∈{\mathrm{SK}}_k}{\mathrm{im}1}_{\min }{\left(l\right)}^2-\mathrm{\μ}{\left(k\right)}^2}{\mathrm{\σ}\left(k\right)+\mathrm{\β}}$

For each pixel im1 (k) of single channel image im1, calculate the biased coefficient of its filtering:

γ ₂(k)＝μ(k)-γ ₁(k)·μ(k)

The most each pixel of the similarity constraint image filtering SDL (im1, im2, num, β) of image im1 participates in following calculating, obtains the filtering image exported:

i(k)＝γ ₁(k)·im1(k)+γ ₂(k)

Based on above-mentioned discussion, step S2 specifically comprises:

S21, carry out the non local similarity constraint Federated filter of associating of 2 two passages according to following formula respectively to there being mist image and obtaining two filtering images:

$\left\{\begin{array}{c}C1=\mathrm{SDL}(M,P_y,N,{\mathrm{\β}}_1)\\ C2=\mathrm{SDL}(M,P_y,N,{\mathrm{\β}}_2)\end{array}\right.$

In above formula, C1, C2 represent two filtering images of acquisition respectively, and M represents pending has mist image, P _yindicate the luminance components of mist image M, N is natural number, represents the number of each pixel being carried out to the pixel participated in required for Federated filter, β ₁, β ₂all represent filtering strength; Preferably, in the present embodiment, N=40, β ₁=0.01, β ₂=0.07.

S22, according to obtain two filtering images calculate the error image D meeting visual characteristics of human eyes:

D(l)＝|C1(l)-C2(l)|

In above formula, l is natural number, and represent the position of pixel, D (l) represents the value of error image D at pixel l, and C1 (l), C2 (l) represent the value of two filtering images at pixel l of acquisition respectively;

S23, calculate the initial estimate obtaining atmospheric scattering light according to following formula:

$\mathrm{\φ}\left(l\right)=C1\left(l\right)-\frac{C1\left(l\right)}{0.42\·\mathrm{\θ}}\·D\left(l\right)$

In above formula, φ (l) represents the initial estimate of atmospheric scattering light at pixel l, and θ represents the average brightness value of C1 (l);

S24, revise rear acquisition atmospheric scattering data according to the initial estimate of following formula to atmospheric scattering light:

ψ(l)＝max(min(0.91·φ(l),C1(l)),0)

In above formula, ψ (l) represents the atmospheric scattering data at pixel l.

The atmospheric scattering data reconstruction mist elimination image that S3, utilization obtain also obtains initialization mist elimination image, comprising:

S31, search and obtain pending have the brightness of mist image front 1.2% the location of pixels set of pixel;

S32, according to the location of pixels set obtained, search the overall brightness Γ of brightness maximum value as mist light having in mist image;

S33, calculate air ature of coal spreading rate according to following formula:

$\mathrm{\ρ}\left(l\right)=1-0.943\·\frac{\mathrm{\ψ}\left(l\right)}{\mathrm{\Γ}}$

In above formula, ρ (l) represents the air ature of coal spreading rate at pixel l, and ψ (l) represents the atmospheric scattering data at pixel l;

S34, rebuild mist elimination image obtain initialization mist elimination image according to following formula:

$\left\{\begin{array}{c}{P}_R^o\left(l\right)=\frac{P_R\left(l\right)-\mathrm{\Γ}}{\max (\mathrm{\ρ}\left(l\right),\mathrm{0,11})}+\mathrm{\Γ}\\ P_G^o\left(l\right)=\frac{P_G\left(l\right)-\mathrm{\Γ}}{\max (\mathrm{\ρ}\left(l\right),\mathrm{0,11})}+\mathrm{\Γ}\\ P_B^o\left(l\right)=\frac{P_B\left(l\right)-\mathrm{\Γ}}{\max (\mathrm{\ρ}\left(l\right),\mathrm{0,11})}+\mathrm{\Γ}\end{array}\right.$

In above formula, represent red, green, blue three color components of initialization mist elimination image respectively, P _r(l), P _y(l), P _bl () represents pending respectively has mist image M at red, yellow, blue three color components of pixel l.

S4, three-channel associating similarity constraint Federated filter is carried out to initialization mist elimination image, obtain the image after mist elimination.

Define three-channel associating similarity constraint Federated filter: JDN (im1, im2, im3, num, β), wherein, im1, im2, im3 are three measure-alike single channel image, im1 is the color channel image needing denoising having mist image, and im2, im3 are the denoising joint channel images needing associating consideration.

Noise-removed filtering process is as follows:

In single channel image im1, im2, im3 that 3 width yardsticks are identical, num the similar pixel point finding it is combined for each pixel k.

Defining the associating similarity of any two pixel k1 and k2 in 3 single channel image im1, im2, im3 is:

$\begin{array}{c}{S}^{\′}(k1,k2)=\underset{m=1}{\overset{25}{\mathrm{\Σ}}}|{\mathrm{blk}}_{\mathrm{im}1}^{k1}\left(m\right)-{\mathrm{blk}}_{\mathrm{im}1}^{k2}\left(m\right)|+\mathrm{\λ}\underset{m=1}{\overset{25}{\mathrm{\Σ}}}|{\mathrm{blk}}_{\mathrm{im}2}^{k1}\left(m\right)-{\mathrm{blk}}_{\mathrm{im}2}^{k2}\left(m\right)|\\ +\mathrm{\λ}\underset{m=1}{\overset{25}{\mathrm{\Σ}}}|{\mathrm{blk}}_{\mathrm{im}3}^{k1}\left(m\right)-{\mathrm{blk}}_{\mathrm{im}3}^{k2}\left(m\right)|\end{array}$

In above formula, k1, k2 are natural number, all represent the position of pixel, S'(k1, k2) represent pixel k1, k2 associating similarity on single channel image im1, im2, im3, represent the value of the pixel l on the image block of 5 × 5 on single channel image im1 centered by pixel k1, represent the value of the pixel l on the image block of 5 × 5 on single channel image im1 centered by pixel k2, represent the value of the pixel l on the image block of 5 × 5 on single channel image im2 centered by pixel k1, represent the value of the pixel l on the image block of 5 × 5 on single channel image im2 centered by pixel k2, represent the value of the pixel l on the image block of 5 × 5 on single channel image im3 centered by pixel k1, represent the value of the pixel l on the image block of 5 × 5 on single channel image im3 centered by pixel k2, λ represents scale-up factor.In the present embodiment, λ is preferably 0.4.Calculate the S'(k1, the k2 that obtain) less, the associating similarity of k1, k2 two pixels on im1, im2, im3 is more similar.With whole image for hunting zone, the individual similar pixel SK of closest num can be found for each pixel k _k={ k ₁, k ₂..., k _num.

Be calculated as follows filtering parameter successively:

For each pixel im1 (k) of single channel image im1, calculate its similar pixel average:

$\∂\left(k\right)=0.04\·\underset{l\∈{\mathrm{SK}}_k}{\mathrm{\Σ}}{\mathrm{im}1}_{\min }\left(l\right)$

For each pixel im1 (k) of single channel image im1, calculate the standard variance of its similar pixel:

For each pixel im1 (k) of single channel image im1, calculate the linear coefficient of its filtering:

For each pixel im1 (k) of single channel image im1, calculate the biased coefficient of its filtering:

${\mathrm{\η}}_2\left(k\right)=\∂\left(k\right)-{\mathrm{\η}}_1\left(k\right)\·\∂\left(k\right)$

The Output rusults of the similarity constraint image filtering JDN (im1, im2, im3, num, β) of image im1 is:

O(k)＝η ₁(k)·im1(k)+η ₂(k)

Based on the above discussion to three-channel associating similarity constraint Federated filter, step S4 is specially:

According to following formula, three-channel associating similarity constraint Federated filter is carried out to initialization mist elimination image, obtains the image after mist elimination:

$\left\{\begin{array}{c}{I}_R=\mathrm{JDN}(P_R^o,P_G^o,P_B^o,\mathrm{49,0.002})\\ I_G=\mathrm{JDN}(P_G^o,P_R^o,P_B^o,\mathrm{49,0.002})\\ I_B=\mathrm{JDN}(P_B^o,P_G^o,P_R^o,\mathrm{49,0.002})\end{array}\right.$

In above formula, I _r, I _g, I _brepresent red, green, blue three color components of the image after mist elimination respectively, represent red, green, blue three single channel image of initialization mist elimination image respectively.

This method utilizes the non local similarity Federated filter of the associating of two passages to calculate and obtains atmospheric scattering data, the estimated accuracy of atmospheric scattering data can be improved, and obtain by the non local similarity constraint Federated filter of associating adopting different parameters to carry out two passages the error image meeting visual characteristics of human eyes, the precision of atmospheric scattering data can be improved further, finally adopt three-channel associating similarity constraint Federated filter to rebuild mist elimination image, efficient recovery mist elimination image can be had, the noise of image after further reduction mist elimination.Therefore, this method accurately can estimate atmospheric scattering light data, thus rebuilds mist elimination image exactly, and precision is high and noise is low.

Embodiment two

The present embodiment is and embodiment one floppy disk system one to one:

Based on the image mist elimination system of human eye vision similarity constraint, comprising:

First processing module, for calculating the pending color smallest passage data having mist image;

Second processing module, for the non local similarity constraint Federated filter of associating having mist image to carry out 2 two passages respectively, and then obtains atmospheric scattering data;

3rd processing module, for utilizing the atmospheric scattering data reconstruction mist elimination image of acquisition and obtaining initialization mist elimination image;

4th processing module, for carrying out three-channel associating similarity constraint Federated filter to initialization mist elimination image, obtains the image after mist elimination.

Native system utilizes the non local similarity Federated filter of the associating of two passages to calculate and obtains atmospheric scattering data, the estimated accuracy of atmospheric scattering data can be improved, and obtain by the non local similarity constraint Federated filter of associating adopting different parameters to carry out two passages the error image meeting visual characteristics of human eyes, the precision of atmospheric scattering data can be improved further, finally adopt three-channel associating similarity constraint Federated filter to rebuild mist elimination image, efficient recovery mist elimination image can be had, the noise of image after further reduction mist elimination.Therefore, native system accurately can estimate atmospheric scattering light data, thus rebuilds mist elimination image exactly, and precision is high and noise is low.

Embodiment three

The image mist elimination equipment based on human eye vision similarity constraint of the image defogging method capable of Application Example one, the processor comprising camera and be connected with camera, described camera is for gathering image, described processor is used for carrying out mist elimination process to the image that camera gathers, and described processor comprises:

First processing module, for calculating the pending color smallest passage data having mist image;

Second processing module, for the non local similarity constraint Federated filter of associating having mist image to carry out 2 two passages respectively, and then obtains atmospheric scattering data;

3rd processing module, for utilizing the atmospheric scattering data reconstruction mist elimination image of acquisition and obtaining initialization mist elimination image;

4th processing module, for carrying out three-channel associating similarity constraint Federated filter to initialization mist elimination image, obtains the image after mist elimination.

This equipment is after collection image, utilize the non local similarity Federated filter of the associating of two passages to calculate and obtain atmospheric scattering data, the estimated accuracy of atmospheric scattering data can be improved, and obtain by the non local similarity constraint Federated filter of associating adopting different parameters to carry out two passages the error image meeting visual characteristics of human eyes, the precision of atmospheric scattering data can be improved further, finally adopt three-channel associating similarity constraint Federated filter to rebuild mist elimination image, efficient recovery mist elimination image can be had, the noise of image after further reduction mist elimination.Therefore, this equipment accurately can estimate atmospheric scattering light data, thus rebuilds mist elimination image exactly, provides precision high and the low output image of noise.

More than that better enforcement of the present invention is illustrated, but the invention is not limited to described embodiment, those of ordinary skill in the art also can make all equivalent variations or replacement under the prerequisite without prejudice to spirit of the present invention, and these equivalent modification or replacement are all included in the application's claim limited range.

Claims (12)

1., based on the image defogging method capable of human eye vision similarity constraint, it is characterized in that, comprising:

S1, calculate the pending color smallest passage data having mist image;

S2, to the non local similarity constraint Federated filter of associating having mist image to carry out 2 two passages respectively, and then obtain atmospheric scattering data;

The atmospheric scattering data reconstruction mist elimination image that S3, utilization obtain also obtains initialization mist elimination image;

S4, three-channel associating similarity constraint Federated filter is carried out to initialization mist elimination image, obtain the image after mist elimination.

2. the image defogging method capable based on human eye vision similarity constraint according to claim 1, it is characterized in that, described step S1, it is specially:

The pending color smallest passage data having mist image are calculated according to following formula:

M _min(l)＝min(P _R(l),P _Y(l),P _B(l))

In above formula, l is natural number, represents the position of pixel, M _minl () represents the color smallest passage data of pixel l, P _r(l), P _y(l), P _bl () represents pending respectively has mist image M at red, yellow, blue three color components of pixel l.

3. the image defogging method capable based on human eye vision similarity constraint according to claim 1, it is characterized in that, described step S2, comprising:

S21, carry out the non local similarity constraint Federated filter of associating of 2 two passages according to following formula respectively to there being mist image and obtaining two filtering images:

In above formula, C1, C2 represent two filtering images of acquisition respectively, and M represents pending has mist image, P _yindicate the luminance components of mist image M, N is natural number, represents the number of each pixel being carried out to the pixel participated in required for Federated filter, β ₁, β ₂all represent filtering strength;

S22, according to obtain two filtering images calculate the error image D meeting visual characteristics of human eyes:

D(l)＝|C1(l)-C2(l)|

In above formula, l is natural number, and represent the position of pixel, D (l) represents the value of error image D at pixel l, and C1 (l), C2 (l) represent the value of two filtering images at pixel l of acquisition respectively;

S23, calculate the initial estimate obtaining atmospheric scattering light according to following formula:

In above formula, φ (l) represents the initial estimate of atmospheric scattering light at pixel l, and θ represents the average brightness value of C1 (l);

S24, revise rear acquisition atmospheric scattering data according to the initial estimate of following formula to atmospheric scattering light:

ψ(l)＝max(min(0.91×φ(l),C1(l)),0)

In above formula, ψ (l) represents the atmospheric scattering data at pixel l.

4. the image defogging method capable based on human eye vision similarity constraint according to claim 3, is characterized in that, in described step S21, and N=40, β ₁=0.01, β ₂=0.07.

5. the image defogging method capable based on human eye vision similarity constraint according to claim 3, it is characterized in that, in described step S21, the formula that the non local similarity constraint Federated filter of associating carrying out two passages adopts is SDL (im1, im2, num, β), wherein im1, im2 are two measure-alike single channel image, num is for natural number and represent the number of each pixel being carried out to the pixel participated in required for Federated filter, β represents filtering strength, and the filtering image finally obtained is:

i(k)＝γ ₁(k)·im1(k)+γ ₂(k)

In above formula, k is natural number, and represent the position of pixel, i (k) represents the value of filtering image at pixel k, and im1 (k) represents the value of the pixel k of single channel image im1, γ ₁k () represents the linear coefficient of the filtering of the pixel k of single channel image im1, γ ₂k () represents the biased coefficient of the filtering of the pixel k of single channel image im1, γ ₁(k) and γ ₂k () is obtained by following formulae discovery:

In above formula, im1 _minl () represents the color smallest passage data of the pixel l of single channel image im1, SK _krepresent the set of the individual similar pixel of the closest num of pixel k, σ (k) represents the standard variance of the similar pixel of the pixel k of single channel image im1, μ (k) represents the similar pixel average of the pixel k of single channel image im1, and σ (k) and μ (k) is obtained by following formulae discovery:

。

6. the image defogging method capable based on human eye vision similarity constraint according to claim 5, is characterized in that, the S set K of the individual similar pixel of closest num of pixel k _kbe the set of combining the minimum num of a similarity pixel with k, the computing formula of described associating similarity is as follows:

In above formula, k1, k2 are natural number, all represent the position of pixel, and S (k1, k2) represents pixel k1, k2 associating similarity on single channel image im1, im2, represent the value of the pixel l on the image block of 5 × 5 on single channel image im1 centered by pixel k1, represent the value of the pixel l on the image block of 5 × 5 on single channel image im1 centered by pixel k2, represent the value of the pixel l on the image block of 5 × 5 on single channel image im2 centered by pixel k1, represent the value of the pixel l on the image block of 5 × 5 on single channel image im2 centered by pixel k2, λ represents scale-up factor.

7. the image defogging method capable based on human eye vision similarity constraint according to claim 1, it is characterized in that, described step S3, comprising:

S31, search and obtain pending have the brightness of mist image front 1.2% the location of pixels set of pixel;

S32, according to the location of pixels set obtained, search the overall brightness Γ of brightness maximum value as mist light having in mist image;

S33, calculate air ature of coal spreading rate according to following formula:

In above formula, ρ (l) represents the air ature of coal spreading rate at pixel l, and ψ (l) represents the atmospheric scattering data at pixel l;

S34, rebuild mist elimination image obtain initialization mist elimination image according to following formula:

In above formula, represent red, green, blue three color components of initialization mist elimination image respectively, represent pending respectively and have mist image M at red, yellow, blue three color components of pixel l.

8. the image defogging method capable based on human eye vision similarity constraint according to claim 1, it is characterized in that, described step S4, it is specially:

According to following formula, three-channel associating similarity constraint Federated filter is carried out to initialization mist elimination image, obtains the image after mist elimination:

In above formula, I _r, I _g, I _brepresent red, green, blue three color components of the image after mist elimination respectively, represent red, green, blue three single channel image of initialization mist elimination image respectively.

9. the image defogging method capable based on human eye vision similarity constraint according to claim 8, it is characterized in that, carrying out the formula that three-channel associating similarity constraint Federated filter adopts is JDN (im1, im2, im3, num, β), wherein im1, im2, im3 are three measure-alike single channel image, and num is for natural number and represent the number of each pixel being carried out to the pixel participated in required for Federated filter, β represents filtering strength, and the filtering image finally obtained is:

O(k)＝η ₁(k)·im1(k)+η ₂(k)

In above formula, k is natural number, and represent the position of pixel, O (k) represents the value of filtering image at pixel k, and im1 (k) represents the value of the pixel k of single channel image im1, η ₁k () represents the linear coefficient of the filtering of the pixel k of single channel image im1, η ₂k () represents the biased coefficient of the filtering of the pixel k of single channel image im1, η ₁(k) and η ₂k () is obtained by following formulae discovery:

In above formula, im1 _minl () represents the color smallest passage data of the pixel l of single channel image im1, SK _krepresent the set of the individual similar pixel of the closest num of pixel k, represent the standard variance of the similar pixel of the pixel k of single channel image im1, represent the similar pixel average of the pixel k of single channel image im1, with obtained by following formulae discovery:

。

10. the image defogging method capable based on human eye vision similarity constraint according to claim 9, is characterized in that, the S set K of the individual similar pixel of closest num of pixel k _kbe the set of combining the minimum num of a similarity pixel with k, the computing formula of described associating similarity is as follows:

In above formula, k1, k2 are natural number, all represent the position of pixel, S'(k1, k2) represent pixel k1, k2 associating similarity on single channel image im1, im2, im3, represent the value of the pixel l on the image block of 5 × 5 on single channel image im1 centered by pixel k1, represent the value of the pixel l on the image block of 5 × 5 on single channel image im1 centered by pixel k2, represent the value of the pixel l on the image block of 5 × 5 on single channel image im2 centered by pixel k1, represent the value of the pixel l on the image block of 5 × 5 on single channel image im2 centered by pixel k2, represent the value of the pixel l on the image block of 5 × 5 on single channel image im3 centered by pixel k1, represent the value of the pixel l on the image block of 5 × 5 on single channel image im3 centered by pixel k2, λ represents scale-up factor.

11., based on the image mist elimination system of human eye vision similarity constraint, is characterized in that, comprising:

First processing module, for calculating the pending color smallest passage data having mist image;

Second processing module, for the non local similarity constraint Federated filter of associating having mist image to carry out 2 two passages respectively, and then obtains atmospheric scattering data;

3rd processing module, for utilizing the atmospheric scattering data reconstruction mist elimination image of acquisition and obtaining initialization mist elimination image;

4th processing module, for carrying out three-channel associating similarity constraint Federated filter to initialization mist elimination image, obtains the image after mist elimination.

12. based on the image mist elimination equipment of human eye vision similarity constraint, it is characterized in that, the processor comprising camera and be connected with camera, described camera is for gathering image, described processor is used for carrying out mist elimination process to the image that camera gathers, and described processor comprises:

First processing module, for calculating the pending color smallest passage data having mist image;

Second processing module, for the non local similarity constraint Federated filter of associating having mist image to carry out 2 two passages respectively, and then obtains atmospheric scattering data;

3rd processing module, for utilizing the atmospheric scattering data reconstruction mist elimination image of acquisition and obtaining initialization mist elimination image;

4th processing module, for carrying out three-channel associating similarity constraint Federated filter to initialization mist elimination image, obtains the image after mist elimination.