CN106960421A - Evening images defogging method based on statistical property and illumination estimate - Google Patents

Abstract

The invention discloses a kind of evening images defogging method based on statistical property and illumination estimate, including：Night foggy image is inverted into obtain reverse image；The local atmosphere light with colour cast of reverse image is calculated, and is optimized by directiveness filtering；The initial transmission of triple channel and the rough estimate transmissivity of triple channel are calculated, the rough estimate transmissivity of triple channel is corrected using the bright passage of reverse image, and obtains optimizing transmissivity by directiveness filtering：To solve carry out again after restored image using atmosphere light of the part with colour cast and optimization transmissivity and invert and obtain mist elimination image of the night with colour cast；And color correction is carried out using local grey word, finally give night mist elimination image.The evening images mist elimination image obtained using the present invention can not only effectively recover brightness, the contrast of image, moreover it is possible to effectively correct the colour cast of evening images, significantly improve visual effect, while retaining more image detail informations, and substantially reduce computational complexity.

Description

Evening images defogging method based on statistical property and illumination estimate

Technical field

The present invention relates to a kind of Computer Image Processing method, more particularly to a kind of evening images defogging method.

Background technology

When there is shooting image under mist environment at night, image overall gray value and contrast can be caused to reduce and lose a large amount of Detailed information, it is difficult to recognize region interested, brought to video monitoring, outdoor target identification and tracking, remotely sensed image etc. Very big difficulty.Therefore, evening images defogging problem is urgently to be resolved hurrily in computer vision application field and digital image processing field.

Existing evening images defogging method is less, mainly there is Pei^[1]What is proposed is changed based on dark primary priori and color Evening images defogging algorithm, Zhang^[2]The defogging algorithm and Li based on new model proposed^[3]Based on relative smooth about Hierachical decomposition defogging algorithm of beam etc..The defogging main frame of these algorithms is still based on dark primary priori, but is due to night The special imaging circumstances of foggy image, dark primary priori is under night environment and does not apply to, therefore these algorithms restore the figure Picture is overall partially dark, and the cross-color of various degrees, and substantially, defogging is incomplete for Halo effect at image light source, And calculate complicated.

[bibliography]

[1]Pei S C,Lee T Y.Nighttime haze removal using color transfer pre- processing and dark channel prior[A].Proceedings of the IEEE International Conference on Image Processing[C].Orlando:IEEE Computer Society Press,2012, 957-960。

[2]Zhang J,Cao Y,Wang Z.Nighttime haze removal based on a new imaging model[A].Proceedings of the IEEE International Conference on Image Processing [C].Paris:IEEE Computer Society Press,2014.4557-4561。

[3]Li Y,Tan R T,Brown S Michael.Nighttime haze remov-al with glow and multiple light colors[C].Proceedings of IEEE International Conference on Computer Vi-sion.Santiago:IEEE Computer Society Press,2015:226-234。

[4]He K,Sun J,Tang X.Single image haze removal using dark channel prior[J].Pattern Analysis and Machine Intelligence,IEEE Transactions on,2011, 33(12):2341-2353。

[5]G.Buchsbaum.A spatial processor model for object colour perception.Journal of the Franklin Institute,1980,310(80):1–26。

[6]Meng Gaofeng,WANG Ying,DUAN Jiangyong,et al.Efficient image dehazing with boundary constraint and contextual regularization[C].IEEE International Con-ference on Computer Vision(ICCV),Sydney,Australia,2013:617- 624。

[7]X.Dong,J.T.Wen,W.X.Li,An efficient and integrated algorithm for video enhancement in challenging lighting conditions,in Proceedings of Institute of Electrical and Electronic Engineers International Conference on Computer Vision and Pattern Recognition,pp.1241-1249,2011。

The content of the invention

Regarding to the issue above, the present invention proposes a kind of evening images defogging method based on statistical property and illumination estimate. First the night foggy image new model with the colour cast factor is set up according to the special imaging circumstances of night foggy image；Then lead to The bright channel histogram distribution of the reverse image of statistics night foggy image and low-light (level) image is crossed, by night foggy image defogging Problem is converted into low-light (level) image enhaucament problem, and utilizes improved He^[4]The local atmosphere light with colour cast of method estimation, simultaneously Transmissivity is corrected by the bright passage of the reverse image of night foggy image thin to retain the more edges of mist elimination image Information is saved, finally by local grey-world^[5]Color correction is carried out to mist elimination image.Evening images defogging side of the present invention Method can not only effectively recover brightness, the contrast of image, moreover it is possible to effectively correct the colour cast of evening images, significantly improve vision effect Really, while retaining more image detail informations, and computational complexity is substantially reduced.

In order to solve the above-mentioned technical problem, a kind of evening images based on statistical property and illumination estimate proposed by the present invention Defogging method, step is as follows：

Step 1, the night foggy image of input are image I (x), and image I (x) is inverted to obtain into reverse image

In formula (1), c ∈ { r, g, b }；

Step 2, calculating reverse imageThe local atmosphere light r with colour cast_L(x)A_L(x), and by directiveness filter into Row optimization：

In formula (2), Ω (x) is pixel x local neighborhood, and Ω (y) is neighborhood y local neighborhood, and GF represents guiding filter Ripple；

Step 3, calculating reverse imageTriple channel initial transmission t_aL(x) with the rough estimate transmissivity of triple channel t_bL(x)：

Step 4, utilize reverse imageBright passage to the rough estimate transmissivity t of triple channel_bL(x) it is corrected, and leads to Directiveness filtering is crossed to optimize：

t_L(x)=GF (e^A_lightt_bL(x)) (6)

In formula (5) and formula (6)：A_light is reverse imageBright passage, t_L(x) it is saturating for the rough estimate to triple channel Penetrate rate t_bL(x) the optimization transmissivity after correcting；

Step 5, the local atmosphere light r with colour cast solved using above-mentioned steps 2_L(x)A_L(x) optimization obtained with step 4 Transmissivity t_L(x) restored image r is solved_L(x)J_L(x)：

In formula (7)：ε is fixed constant, ε=0.1；

Step 6, to restored image r_L(x)J_L(x) progress inverts and obtains mist elimination image J of the night with colour cast again_p(x)：

J_p(x)=255-r (x) J (x) (8)

Step 7, using local grey word to mist elimination image J of the night with colour cast_p(x) color correction is carried out, is obtained Final night mist elimination image J (x),

In formula (9)：ω represents whole visible-range, and λ is optical wavelength, and e (λ) represents the distribution of the light of certain wave band, s (x, Be λ) that certain point is to the reflectivity of a certain wavelength in space, p (λ) represents photobehavior of the camera to certain light, m be [0,1] it Between a constant.

Compared with prior art, the beneficial effects of the invention are as follows：

The present invention carries out defogging based on statistical property to evening images, and night foggy image can be regarded as to low-light (level) image Handled, ask for the local atmosphere light with colour cast using the He improved the method for asking for atmosphere light, and utilize reversion figure The bright passage of picture is corrected to rough estimate transmissivity further to retain the details and monochrome information of image, finally by part Grey world carry out color correction to restored image has the robustness of mist scene to multiple light courcess night to improve algorithm.

Brief description of the drawings

Fig. 1 (a) is the bright channel histogram of the reverse image of night foggy image；

Fig. 1 (b) is the bright channel histogram of the reverse image of low-light (level) image；

Fig. 2 (a) is a width night foggy image Trains；

Fig. 2 (b) is defogging algorithm of the document [2] based on new model to the result after image Trains processing；

Fig. 2 (c) is after document [3] is handled image Trains based on the hierachical decomposition defogging algorithm that relative smooth is constrained As a result；

Fig. 2 (d) is the result that evening images defogging method of the present invention is handled image Trains；

Fig. 3 (a) is a width night foggy image Street；

Fig. 3 (b) is defogging algorithm of the document [2] based on new model to the result after image Street processing；

Fig. 3 (c) is after document [3] is handled image Street based on the hierachical decomposition defogging algorithm that relative smooth is constrained As a result；

Fig. 3 (d) is the result that evening images defogging method of the present invention is handled image Street.

Embodiment

Technical solution of the present invention is described in further detail with specific embodiment below in conjunction with the accompanying drawings, described is specific Only the present invention is explained for embodiment, is not intended to limit the invention.

The mentality of designing of evening images defogging method of the present invention is as follows：

1st, the algorithm general principle in common evening images defogging method is：According to document [2], night foggy image into As model：

I (x)=r (x) (J (x) t (x)+(1-t (x)) A (x)) is 1.

Wherein, I (x) is night foggy image, and J (x) is night mist elimination image, and t (x) is transmissivity, and A (x) is bias light Intensity, r (x) is the colour cast factor.

According to boundary constraint^[6], the initial transmission t of Misty Image can be obtained_a(x) with rough estimate transmissivity t_b(x)：

And then obtain night picture rich in detail：

For low-light (level) image I_L(x), first it is inverted：

R_L(x)=255-I_L(x) ⑤

Afterwards by R_L(x) 4. substitution formula obtains：

To J_L(x) inverted again, obtain last enhancing image.

2nd, led to by contrasting the bright of reverse image of the 50 width night foggy images randomly selected and 50 width low-light (level) images The histogram distribution in road, such as Fig. 1 (a) and Fig. 1 (b) are shown.It was found that the reverse image tool of night foggy image and low-light (level) image There is great similitude, therefore night foggy image defogging problem can be converted into low-light (level) image enhaucament problem, according to document [7], the reverse image of low-light (level) image is considered as foggy image on daytime, similar, the reverse image of night foggy image Foggy image on daytime can approximately be regarded as to be handled, but be due to the presence of night environment colour cast, the reversion of night foggy image Image is still even with uneven illumination, the attribute such as colour cast, and existing defogging algorithm is not still applied to, evening images defogging of the present invention Atmosphere light and colour cast are handled as an entirety in method and utilize improved He methods to estimate the atmosphere light with colour cast, It is serious for night foggy image colour cast, cause triple channel transmission difference larger, it is right in evening images defogging method of the present invention The initial transmission that 2. above-mentioned formula calculates carries out three-channel processing simultaneously in order to retain more image details, passes through reverse image Bright passage rough estimate transmissivity is corrected and optimized using directiveness filtering, finally by local grey world Color correction is carried out to mist elimination image.

Evening images defogging method proposed by the present invention based on statistical property and illumination estimate, is comprised the following steps that：

Step 1, the night foggy image of input are image I (x), and image I (x) is inverted to obtain into reverse image

In formula (1), c ∈ { r, g, b }, x is the position of pixel in image I (x)；

Step 2, calculating reverse imageThe local atmosphere light r with colour cast_L(x)A_L(x), and by directiveness filter into Row optimization：

In formula (2), Ω (x) is pixel x local neighborhood, and Ω (y) is field y local neighborhood, and GF represents guiding filter Ripple；

Step 3, in order that transmissivity solves more accurate, the method that document [6] calculates transmissivity is improved, calculated The initial transmission t of triple channel_aL(x) with the rough estimate transmissivity t of triple channel_bL(x)：

Step 4, in order that transmissivity local smoothing method and keeping good light characteristic, utilize reverse imageIt is bright Rough estimate transmissivity t of the passage to triple channel_bL(x) it is corrected, and is optimized by directiveness filtering：

t_L(x)=GF (e^A_lightt_bL(x)) (6)

In formula (5) and formula (6)：A_light is reverse imageBright passage, t_L(x) it is saturating for the rough estimate to triple channel Penetrate rate t_bL(x) the optimization transmissivity after correcting；

Step 5, the local atmosphere light r with colour cast solved using above-mentioned steps 2_L(x)A_L(x) optimization obtained with step 4 Transmissivity t_L(x) restored image r is solved_L(x)J_L(x)：

In formula (7)：ε is fixed constant, and it is that zero, ε takes 0.1 to prevent denominator.

Step 6, to restored image r_L(x)J_L(x) progress inverts and obtains mist elimination image J of the night with colour cast again_p(x)：

J_p(x)=255-r (x) J (x) (8)

Step 7, in order to improve robustness of the algorithm to night multiple light courcess scene, using the grey proposed in document [5] Word algorithms simultaneously carry out localization improvement to mist elimination image J of the night with colour cast_p(x) color correction is carried out, night is finally given Mist elimination image J (x),

In formula (9)：ω represents whole visible-range, and λ is optical wavelength, and e (λ) represents the distribution of the light of certain wave band, s (x, Be λ) that certain point is to the reflectivity of a certain wavelength in space, p (λ) represents photobehavior of the camera to certain light, m be [0,1] it Between a constant.

In order to verify the validity of evening images defogging method proposed by the present invention, defogging is carried out to night foggy image real Test, and contrasted with related algorithm.Fig. 2 (a) is that width night foggy image a Trains, Fig. 2 (b) is to be carried using document [2] The defogging algorithm based on new model gone out is to the defog effect after image Trains processing, and Fig. 2 (c) is to be proposed using document [3] Based on relative smooth constrain hierachical decomposition defogging algorithm to image Trains processing after defog effect, Fig. 2 (d) be this hair Bright evening images defogging method image Trains is handled after defog effect.Fig. 3 (a) is a width night foggy image Street, Fig. 3 (b) are the defogging algorithm based on new model proposed using document [2] to the defogging after image Street processing As a result, Fig. 3 (c) is the hierachical decomposition defogging algorithm based on relative smooth constraint proposed using document [3] to image Street Defogging result after processing；Fig. 3 (d) is the defogging knot after evening images defogging method of the present invention is handled image Street Really.As can be seen that the night mist elimination image obtained after being handled using evening images defogging method of the present invention and document [2] Zhang The hierachical decomposition defogging algorithm constrained based on relative smooth that the defogging algorithm based on new model and document [3] Li proposed is proposed Compared to brightness of image and contrast balanced can be improved, the halo artifact that source region is brought effectively is removed, and can recover more Image detail information, the colour cast of correction chart picture is visual with more preferable visual effect.

For objective evaluation evening images defogging method of the present invention, the colour cast degree and contrast of night mist elimination image are calculated. As shown in table 1.

The objective indicator comparative result of table 1

Shown by the colour cast level data of table 1, the inventive method can effective correction chart picture colour cast；It can be seen by contrast results Go out, the inventive method can improve the overall contrast of the image after defogging.

Test result indicate that, the evening images defogging method proposed by the present invention based on statistical property and illumination estimate can pin The colour cast of the defect that evening images defogging has, effectively correction evening images is carried out based on dark primary method to tradition, is retained More image details, equilibrium improves the overall brightness and contrast of image, with more preferable sense of vision.

Although above in conjunction with accompanying drawing, invention has been described, and the invention is not limited in above-mentioned specific implementation Mode, above-mentioned embodiment is only schematical, rather than restricted, and one of ordinary skill in the art is at this Under the enlightenment of invention, without deviating from the spirit of the invention, many variations can also be made, these belong to the present invention's Within protection.

Claims (1)

1. a kind of evening images defogging method based on statistical property and illumination estimate, it is characterised in that step is as follows：

Step 1, the night foggy image of input are image I (x), and image I (x) is inverted to obtain into reverse image

$I_{inv}^c\left(x\right)=255-I\left(x\right)---\left(1\right)$

In formula (1), c ∈ { r, g, b }；

Step 2, calculating reverse imageThe local atmosphere light r with colour cast_L(x)A_L(x), and by directiveness filtering carry out excellent Change：

$r_L\left(x\right)A_L\left(x\right)=GF\left(\underset{z\∈\mathrm{\Ω}\left(y\right)}{max}\underset{y\∈\mathrm{\Ω}\left(x\right)}{\underset{c\∈\{r,g,b\}}{min}}\right(I\left(z\right)\left)\right))---\left(2\right)$

In formula (2), Ω (x) is pixel x local neighborhood, and Ω (y) is neighborhood y local neighborhood, and GF represents guiding filtering；

Step 3, calculating reverse imageTriple channel initial transmission t_aL(x) with the rough estimate transmissivity t of triple channel_bL (x)：

$t_{aL}\left(x\right)=\min \left(max\right(\frac{r_L^c\left(x\right)A_L^c\left(x\right)-I_{inv}^c\left(x\right)}{r_L^c\left(x\right)A_L^c\left(x\right)},\frac{r_L^c\left(x\right)A_L^c\left(x\right)-I_{inv}^c\left(x\right)}{r_L^c\left(x\right)A_L^c\left(x\right)-1}),1)---\left(3\right)$

$t_{bL}\left(x\right)=\underset{y\∈\mathrm{\Ω}\left(x\right)}{min}\left(\underset{z\∈\mathrm{\Ω}\left(y\right)}{max}\right(t_{aL}\left(z\right))---\left(4\right)$

Step 4, utilize reverse imageBright passage to the rough estimate transmissivity t of triple channel_bL(x) it is corrected, and by referring to The property led filtering is optimized：

$A\_light=\underset{c\∈\{r,g,b\}}{max}\left(I_{inv}^c\right(x\left)\right)---\left(5\right)$

t_L(x)=GF (e^A_lightt_bL(x)) (6)

In formula (5) and formula (6)：A_light is reverse imageBright passage, t_L(x) it is the rough estimate transmissivity to triple channel t_bL(x) the optimization transmissivity after correcting；

Step 5, the local atmosphere light r with colour cast solved using above-mentioned steps 2_L(x)A_L(x) the optimization transmission obtained with step 4 Rate t_L(x) restored image r is solved_L(x)J_L(x)：

$r_L\left(x\right)J_L\left(x\right)=\frac{I_{inv}^c\left(x\right)-r_L\left(x\right)A_L\left(x\right)}{max(\mathrm{\ϵ},t_L(x))}+r_L\left(x\right)A_L\left(x\right)---\left(7\right)$

In formula (7)：ε is fixed constant, ε=0.1；

Step 6, to restored image r_L(x)J_L(x) progress inverts and obtains mist elimination image J of the night with colour cast again_p(x)：

J_p(x)=255-r (x) J (x) (8)

Step 7, using local grey word to mist elimination image J of the night with colour cast_p(x) color correction is carried out, obtains final Night mist elimination image J (x),

$\frac{{\∫}_{\mathrm{\Ω}}{J}_p\left(x\right)dx}{{\∫}_{\mathrm{\Ω}}dx}=\frac{1}{{\∫}_{\mathrm{\Ω}}dx}{\underset{\mathrm{\ω}}{\∫\∫}}_{\mathrm{\Ω}}e\left(\mathrm{\λ}\right)s(x,\mathrm{\λ})p\left(\mathrm{\λ}\right)d\mathrm{\λ}dx$

$\begin{array}{c}={\underset{\mathrm{\ω}}{\∫}}_{\mathrm{\Ω}}e\left(\mathrm{\λ}\right)p\left(\mathrm{\λ}\right)\left(\frac{{\∫}_{\mathrm{\Ω}}s(x,\mathrm{\λ})dx}{{\∫}_{\mathrm{\Ω}}dx}\right)d\mathrm{\λ}\\ =m{\underset{\mathrm{\ω}}{\∫}}_{\mathrm{\Ω}}e\left(\mathrm{\λ}\right)p\left(\mathrm{\λ}\right)d\mathrm{\λ}={\mathrm{me}}_{\mathrm{\Ω}}\end{array}---\left(9\right)$

In formula (9)：ω represents whole visible-range, and λ is optical wavelength, and e (λ) represents the distribution of the light of certain wave band, and s (x, λ) is Certain point is to the reflectivity of a certain wavelength in space, and p (λ) expression cameras are to the photobehavior of certain light, and m is between [0,1] One constant.