CN106157267A - A kind of image mist elimination absorbance optimization method based on dark channel prior - Google Patents

Abstract

The invention discloses a kind of image mist elimination absorbance optimization method based on dark channel prior, the method comprises the steps: step one, utilizes dark channel prior theory to solve rough absorbance, and fog image is carried out thick defogging, obtain middle mist elimination result；Mist elimination result optimizing absorbance in the middle of step 2, utilization, makes absorbance with close to true transmittance values, then using Steerable filter to become more meticulous the absorbance after optimizing, eliminate blocking effect.The present invention is optimized process to absorbance, makes the absorbance after optimization be more nearly the actual value of scene, improves the colour cast phenomenon that traditional dark channel prior algorithm produces at some special area.The ameliorative way that the present invention proposes is the optimization carried out for absorbance, and method is very simple effectively so that the image detail of recovery is greatly enhanced, and overall brightness improves, and visual effect is more preferable, and after mist elimination, image is more true to nature.

Description

A kind of image mist elimination absorbance optimization method based on dark channel prior

Technical field

The present invention relates to Digital Image Processing and the technical field of image mist elimination, be specifically related to a kind of based on dark channel prior Image mist elimination absorbance optimization method.

Background technology

In recent years, haze comes across each big city of China the most continually, some place even annual most of the time All shroud under haze.2013, year Beijing within 5 days, be not only haze sky；In the January of this year, four hazes are had to shroud 30 Individual provinces and cities.When haze is bigger, visibility reduces, and has severely impacted the definition of clapped picture.

The image shot under haze weather can be degraded by serious due to the impact of atmosphere light, and the substantial amounts of fog that adulterates becomes Point, the overall canescence partially of image, marginal information is less, and detectability is substantially reduced.The especially image of shooting under thick fog weather, Detailed information loss is more, is unfavorable for that image is processed by computer, and visual effect is poor.

Depending on can be by serious interference when the system of computer vision works under haze weather, even cannot be normal Work, such as traffic video monitoring, military surveillance etc..Therefore, the clarification method of research haze weather hypograph has important Realistic meaning and practical value, image mist elimination technology was always a study hotspot of image processing field in recent years.This Bright research is that the recovery to fog-degraded image and details strengthen, and improves dark channel prior defogging method on high etc. special The defect of area failures, to strengthen the effect of mist elimination as much as possible.

At present, much work has been done the domestic and international researchers in image mist elimination field, it is proposed that a lot of effective mist elimination sides Method.Existing mist elimination algorithm substantially can be divided into two big classes: method based on image enhaucament and method based on physical model.

The first kind is method based on image enhaucament, and this kind of method is to use general image processing method to being degraded Image strengthens, and improves the quality of image.Such method is simple, and processing speed is very fast, but generally goes fog effect poor, especially It it is denseer for fog or the image of fog skewness.Article [1] (Gonzalez R C, Woods R E.Digital Image Processing.Reading, MA:Addison-Wesley, 1992.) propose the increasing of histogram equalization of the overall situation Strong method, the method is relatively simple, but treatment effect is undesirable, is likely to result in the loss of image portion information, so that image Distortion.Article [2] (Kim T K, Paik J K, Kang B S.Contrast enhancement system using spatially adaptive histogram equalization with temporal filtering.IEEE Transactions on Consumer Electronics, 1998,44 (1): 82-87) rectangular histogram proposing local in is equal Weighing apparatusization, has good treatment effect, but can cause serious blocking effect.Article [3] (Land E H.The retinex Theory of color vision.Scientific America, 1977,237 (6): 108-128) propose based on color The Retinex algorithm of constancy, this algorithm uses illumination-reflection model to simulate the process that Misty Image is degenerated.By eliminating Irradiate component, solve reflecting component and restore without mist scene.Retinex algorithm is unwise to the change in thick fog region and scene depth Sense, the image after recovery would generally retain more fog composition.

Defogging method based on physical model is to produce principle by research fog, understands image degradation mechanism, construction Reason model, thus it is finally inversed by the scene without mist.Such method has inherently carried out mist elimination to image, is generally of well effect Really, be the most also mist elimination research focus.The mist elimination algorithm of main flow has the calculation that Tarel, He Kaiming, Meng Gaofeng et al. propose at present Method, its physical model relied on is atmospherical scattering model.Article [4] (He K, Sun J, Tang X.Single image haze removal using dark channel prior.IEEE Transactions on Pattern Analysis and Machine Intelligence, 2011,33 (12): 2341-2353) by a large amount of without mist image statistics observation of characteristics, Show the priori rule being named as dark channel prior.The method has extraordinary performance on treatment effect, opens image One frontier of mist elimination.Article [5] (Tarel J P, Hautiere N.Fast visibility restoration from a single color or gray level image.In:Proceedings of the 12th IEEE International Conference on Computer Vision,2009.Kyoto:IEEE,2009.2201-2208) In, it is proposed that a kind of method of Quick demisting, use double medium filtering to replace the mini-value filtering in [4] and Steerable filter, greatly Simplify greatly processing procedure, improve efficiency.But the holding edge filter algorithm that medium filtering has not been, the regional area depth of field Sudden change can produce halo effect.And the parameter in algorithm is more, it is impossible to realize self-adaptative adjustment, needs manually to carry out test and adjust Whole, it is restricted in actual applications.Article [6] (Gaofeng MENG, Ying WANG, Jiangyong DUAN, Shiming XIANG,Chunhong PAN.Efficient image dehazing with boundary constraint and contextual regularization.The IEEE International Conference on Computer Vision (ICCV), 2013, pp.617-624) introduce absorbance boundary limit, use the weight letter that Kirsch operator calculates The depth of field change of number reflection local, and use regularization method to calculate absorbance.This algorithm can be prevented effectively from halation phenomenon, but The colour cast phenomenon that sky areas be there will be, especially mist and without mist image.Article [7] (Cui Bingqi, Xie Zhendong, Li Hong, Method based on dark channel prior image mist elimination is improved, information communication (A), 1673-1131 (2013) 06-0060-02) propose to lead to Cross wavelet transformation and extract low-frequency information and the high-frequency information of image, only low frequency part is carried out dark channel prior mist elimination, then and high Frequently information be reconstructed recovery without the clear picture of mist, this method is poor for the detail recovery effect of image, image detail and Before mist elimination, difference is little, and small wave converting method is time-consuming the highest, although shorter than the soft stingy figure time, but with Steerable filter side It is the highest that method compares then time complexity, therefore has considerable restraint in practical engineering application.

Patent CN105631829A uses the method for reverse image to carry out mist elimination, improves former dark algorithm at night etc. The defect of loss of detail in the case of low-light (level), but this algorithm cannot adapt to the scene mist elimination that illumination is higher, and in fact fog figure The brightness of picture is the highest, and in the case of low-light (level), the recovery effects of mist elimination is poor.

Summary of the invention

Present invention aim at: improving the dark channel prior algorithm defect in special area colour cast, transmittance calculation more accords with Closing the true transmittance values of scene, the brightness of image making recovery is higher, and the more life-like nature of tone, details reinforced effects is more preferably.

The technical solution used in the present invention is: a kind of image mist elimination absorbance optimization method based on dark channel prior, should Method comprises the steps:

Step 1): obtain the fog image of rgb format: based on the physical model having mist image, the image of rgb format is entered Row defogging, has the physical model of mist image to be expressed as:

I (x)=J (x) t (x)+A (1-t (x))

Wherein, I (x) treats the image of mist elimination exactly, and J (x) is intended to the image without mist recovered, and A is global atmosphere light component, T (x) is absorbance；

Step 2): ask for the dark scattergram of fog image；

1) first obtain the minima in each pixel RGB component, be stored in a secondary gray scale identical with original image size In figure, i.e. minima passage；

2) minima passage being carried out mini-value filtering, the radius of filtering is determined by window size；

Dark to ask for formula as follows:

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

In formula, min is for doing minimum operation, and r, g, b are respectively three Color Channels, J^cRepresent each logical of coloured image Road, Ω (x) represents a window centered by pixel X, and the theory of dark primary priori is pointed out: J^dark(x)→0；

Step 3): obtain air light value according to dark figure；

Step 4): calculate absorbance according to dark channel prior principle；

Step 5): fog image is carried out defogging, obtains middle mist elimination result；

Step 6): to step 4) absorbance tried to achieve is optimized；

Step 7): restore without mist image, reconstruct the image without mist according to atmospherical scattering model.

Wherein, step 3) in when calculating atmosphere light value, according to there being the physical model of mist image, want to recover without mist Image, premise is to know air light value A, and the method asking for A value is:

1) 0.1% point that in statistics dark, brightness value is the highest；

2) in original mist image, the pixel corresponding with these points is found；

3) finding in these pixels and have the pixel of maximum brightness value, this point is air light value A.

Wherein, step 4) in calculate absorbance time；By there being the deformation of mist image physical model, and combine dark channel prior Theory, rough absorbance t can be derived₁The expression formula of (x), as follows:

$t_1\left(x\right)=1-\underset{y\∈\mathrm{\Ω}\left(x\right)}{\min }\left(\underset{c\∈(r,g,b)}{\min }\frac{I^c\left(y\right)}{A^c}\right)$

Wherein, min is for doing minimum operation, and r, g, b are three Color Channels respectively.

By introducing the factor between [0,1] in above formula, then it is modified to:

$t_1\left(x\right)=1-\mathrm{\ω}*\underset{y\∈\mathrm{\Ω}\left(x\right)}{\min }\left(\underset{c\∈(r,g,b)}{\min }\frac{I^c\left(y\right)}{A^c}\right)$

Wherein, ω is fog factor, removes degree in order to regulate fog, it is to avoid mist elimination is not enough or excessively mist elimination and cause Image fault phenomenon, the fog factor that goes chosen is 0.95.

Wherein, step 5) in thick mist elimination time, the formula of thick mist elimination is as follows:

$J_1^c\left(x\right)=\frac{I^c\left(x\right)-A^c}{max(0.1,t_1(x))}-A^c,c\∈(r,g,b)$

Wherein, max is for doing maximum operation, in order to avoid effect of noise, absorbance limits a lower limit 0.1, Prevent J (x) from the situation of negative value occurring.

Wherein, step 6) in absorbance when optimizing, concrete optimization method is:

1) intermediate object program J is calculated₁The minima passage figure of (x)

2) to rough absorbance t₁X () is optimized, absorbance t after being optimized₂(x), employing equation below:

$t_2\left(x\right)=\frac{A^{\min }-I^{dark}\left(x\right)}{\max (1,A^{\min }-J_1^{\min }(x))}$

Wherein, max for doing maximum operation, A^minFor the minima of air light value, I^darkX () is dark distribution.

3) utilize Steerable filter that the absorbance after optimizing is carried out process of refinement, eliminate blocking effect, obtain final optimization pass After absorbance t (x).

Wherein, step 7) in when restoring without mist image, air light value A and absorbance t (x) that utilization is previously obtained can roots According to physical model reconstruct without mist image J (x), employing equation below:

$J^c\left(x\right)=\frac{I^c\left(x\right)-A^c}{\max (0.1,t(x))}-A^c,c\∈(r,g,b)$

Wherein, in order to avoid effect of noise, absorbance is limited a lower limit 0.1, prevents J (x) from negative value occurring Situation.

The principle of the present invention is:

Technical scheme is divided into two stages: the first stage is that to utilize dark channel prior theory to solve rough saturating Penetrate rate, and fog image is carried out thick defogging, obtain middle mist elimination result；Second stage is that in the middle of utilizing, mist elimination result is excellent Change absorbance, then use Steerable filter that the absorbance after optimizing is become more meticulous.Specific as follows:

1) fog image is carried out thick mist elimination

The air light value calculated by preceding step and rough absorbance, carry out mist elimination according to physical model to fog image Operation, obtains middle mist elimination result.

2) absorbance optimization

Thick absorbance is optimized, makes absorbance with close to true transmittance values, then utilizing Steerable filter to optimization After absorbance carry out process of refinement, eliminate blocking effect.

The advantage of technical solution of the present invention and good effect be:

(1) the colour cast phenomenon of dark channel prior algorithm is improved

Directly the dark without mist image is distributed during dark channel prior Theoretical Calculation absorbance and is approximately equal to zero, so ask The absorbance taken will be overall less than normal, and the details of subregion can be lost, and the image of recovery often produces colour cast phenomenon, Affect the vision sight of image.The present invention is optimized process to absorbance, makes the absorbance after optimization be more nearly field The actual value of scape, improves the colour cast phenomenon that traditional dark channel prior algorithm produces at some special area.

(2) restored image visual effect is more preferable

The ameliorative way that the present invention proposes is the optimization carried out for absorbance, and method is very simple effectively so that restore Image detail be greatly enhanced, brightness is improved, and improves dark channel prior algorithm under the scene that brightness is relatively low The defect that image information loss is more, simultaneously integral color nature true to nature, visual effect is more preferable.

Accompanying drawing explanation

Fig. 1 is mist elimination algorithm flow chart；

Fig. 2 is comparing result schematic diagram, and wherein, Fig. 2 (a) is artwork schematic diagram, and Fig. 2 (b) is dark channel prior algorithm knot Really schematic diagram, Fig. 2 (c) is result schematic diagram of the present invention.

Detailed description of the invention

Below in conjunction with the accompanying drawings and detailed description of the invention further illustrates the present invention.

A kind of based on dark channel prior the image mist elimination absorbance optimization method that the present invention proposes, performs flow process such as Fig. 1 Shown in:

Step 1): obtain the fog image of rgb format.

Step 2): ask for the dark scattergram of fog image.

Step 3): obtaining air light value according to dark figure, the rule chosen here is: choose dark intermediate value maximum Bright spot in the RGB fog image of 0.1% correspondence is as the air light value of this image.

Step 4): calculate absorbance according to dark channel prior principle.

Step 5): fog image is carried out defogging, obtains middle mist elimination result.

Step 6): to step 4) absorbance tried to achieve is optimized.

Step 7): restore without mist image, reconstruct the image without mist according to atmospherical scattering model.

Specifically comprise the following steps that

1, Misty Image physical model

The present invention carries out defogging based on the physical model having mist image to the image of rgb format, has the thing of mist image Reason model can be expressed as:

I (x)=J (x) t (x)+A (1-t (x))

Wherein, I (x) is exactly the image (treating the image of mist elimination) that we have had, and J (x) is the nothing that we are to be recovered The image of mist, A is global atmosphere light component, and t (x) is absorbance.

2, dark is calculated

Dark channel prior theory is pointed out: in the regional area of most non-skies, and certain some pixel always has at least One Color Channel has the lowest value.In other words, the minima of this area light intensity is a least number, levels off to 0, by it Referred to as dark.It specifically asks for process:

1) first obtain the minima in each pixel RGB component, be stored in a secondary gray scale identical with original image size In figure, i.e. minima passage.

2) minima passage being carried out mini-value filtering, the radius of filtering is determined by window size, imitates through later experiments Fruit is summed up, and window size has large effect to removing fog effect.Here, we choose filter radius is 5 the most suitable.

Dark to ask for formula as follows:

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

In formula, min is for doing minimum operation, and r, g, b are three Color Channels respectively, J^cRepresent each logical of coloured image Road, Ω (x) represents a window centered by pixel X, and the theory of dark channel prior is pointed out: J^dark(x)→0。

3, air light value is calculated

According to there being the physical model of mist image, wanting to recover without mist image, premise is to know air light value A.This The method asking for A value in invention is:

1) 0.1% point that in statistics dark, brightness value is the highest.

2) in original mist image, the pixel corresponding with these points is found,

3) finding in these pixels and have the pixel of maximum brightness value, this point is air light value A.

4, absorbance is calculated

By there being the deformation of mist image physical model, and combine the theory of dark channel prior, can derive rough saturating Penetrate rate t₁The expression formula of (x), as follows:

$t_1\left(x\right)=1-\underset{y\∈\mathrm{\Ω}\left(x\right)}{\min }\left(\underset{c\∈(r,g,b)}{\min }\frac{I^c\left(y\right)}{A^c}\right)$

Wherein, min is for doing minimum operation, and r, g, b are three Color Channels respectively.

In actual life, even fine day white clouds, air there is also some granules, therefore, see object at a distance Or the impact of mist can be felt, it addition, the existence of mist allows the mankind feel the existence of the depth of field, therefore, it is necessary to mist elimination time Waiting and retain a certain degree of mist, this then can be modified to by introducing the factor between [0,1] in above formula:

$t_1\left(x\right)=1-\mathrm{\ω}*\underset{y\∈\mathrm{\Ω}\left(x\right)}{\min }\left(\underset{c\∈(r,g,b)}{\min }\frac{I^c\left(y\right)}{A^c}\right)$

Wherein, ω is fog factor, removes degree in order to regulate fog, it is to avoid mist elimination is not enough or excessively mist elimination and cause Image fault phenomenon, what the present invention chose removes fog factor is 0.95.

5, thick mist elimination

The most several steps calculate air light value and rough absorbance, according to physical model to fog image Carry out thick defogging, prepare for further absorbance optimization.The formula of thick mist elimination is as follows:

$J_1^c\left(x\right)=\frac{I^c\left(x\right)-A^c}{max(0.1,t_1(x))}-A^c,c\∈(r,g,b)$

Wherein, max is for doing maximum operation, and r, g, b are respectively three Color Channels, in order to avoid effect of noise, this Invention limits a lower limit 0.1 to absorbance, prevents J (x) from the situation of negative value occur.

6, absorbance optimization

The absorbance obtained above is rough absorbance, due to when asking for absorbance directly by without mist image Dark is approximately equal to zero, and the result of calculation of such absorbance will be less than normal, if directly selecting it to carry out the recovery without mist image, Process the most very well, also can produce colour cast phenomenon.The present invention uses new optimization method to rough absorbance Being optimized, absorbance can be made to approach the true transmittance values of scene, the image of recovery is more life-like.Concrete optimization method It is:

1) intermediate object program J is calculated₁The minima passage figure of (x)

2) to rough absorbance t₁X () is optimized, absorbance t after being optimized₂(x), employing equation below:

$t_2\left(x\right)=\frac{A^{\min }-I^{dark}\left(x\right)}{\max (1,A^{\min }-J_1^{\min }(x))}$

Wherein, max for doing maximum operation, A^minFor the maximum of air light value, I^dark(x)。

Utilize Steerable filter that the absorbance after optimizing is carried out process of refinement, eliminate blocking effect, after obtaining final optimization pass Absorbance t (x)

7, restore without mist image

Utilize the air light value A being previously obtained and absorbance t (x) can reconstruct without mist image J (x) according to physical model, Employing equation below:

$J^c\left(x\right)=\frac{I^c\left(x\right)-A^c}{\max (0.1,t(x))}-A^c,c\∈(r,g,b)$

Wherein, in order to avoid effect of noise, the present invention limits a lower limit 0.1 to absorbance, prevents J (x) from occurring The situation of negative value.

Result is as shown in Figure 2:

From the comparing result of Fig. 2 it can be seen that the present invention is in the case of image fog is denseer, can preferably restore fog Image, the recovery effects of details relatively dark channel prior algorithm is remarkably reinforced, it is noted that dark algorithm building on hand and road Face produces certain colour cast phenomenon, and the present invention then successfully overcomes this defect, and the integral color of image is true to nature, and brightness is also Improving a lot, visual effect is more preferably.

Claims (6)

1. an image mist elimination absorbance optimization method based on dark channel prior, it is characterised in that: the method includes walking as follows Rapid:

Step 1): obtain the fog image of rgb format: based on the physical model having mist image, the image of rgb format is gone Mist operates, and has the physical model of mist image to be expressed as:

I (x)=J (x) t (x)+A (1-t (x))

Wherein, I (x) treats the image of mist elimination exactly, and J (x) is intended to the image without mist recovered, and A is global atmosphere light component, t (x) For absorbance；

Step 2): ask for the dark scattergram of fog image；

1) first obtain the minima in each pixel RGB component, be stored in a secondary gray-scale map identical with original image size, I.e. minima passage；

2) minima passage being carried out mini-value filtering, the radius of filtering is determined by window size；

Dark to ask for formula as follows:

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

In formula, min is minimum operation, and r, g, b are respectively three passages of coloured image, J^cRepresent each logical of coloured image Road, Ω (x) represents a window centered by pixel X, and the theory of dark channel prior is pointed out: J^dark(x)→0；

Step 3): obtain air light value according to dark figure；

Step 4): calculate absorbance according to dark channel prior principle；

Step 5): fog image is carried out defogging, obtains middle mist elimination result；

Step 6): to step 4) absorbance tried to achieve is optimized；

Step 7): restore without mist image, reconstruct the image without mist according to atmospherical scattering model.

Image mist elimination absorbance optimization method based on dark channel prior the most according to claim 1, it is characterised in that:

Step 3) in when calculating atmosphere light value, according to there being the physical model of mist image, want to recover without mist image, premise Being to know air light value A, the method asking for A value is:

1) 0.1% point that in statistics dark, brightness value is the highest；

2) in original mist image, the pixel corresponding with these points is found；

3) finding in these pixels and have the pixel of maximum brightness value, this point is air light value A.

Image mist elimination absorbance optimization method based on dark channel prior the most according to claim 1, it is characterised in that:

Step 4) in calculate absorbance time；By there being the deformation of mist image physical model, and combine the theory of dark channel prior, can To derive rough absorbance t₁The expression formula of (x), as follows:

$t_1\left(x\right)=1-\underset{y\∈\mathrm{\Ω}\left(x\right)}{\min }\left(\underset{c\∈(r,g,b)}{\min }\frac{I^c\left(y\right)}{A^c}\right)$

Wherein, min is minimum operation, and Ω (x) represents a window centered by pixel X；

By introducing the factor between [0,1] in above formula, then it is modified to:

$t_1\left(x\right)=1-\mathrm{\ω}*\underset{y\∈\mathrm{\Ω}\left(x\right)}{\min }\left(\underset{c\∈(r,g.b)}{\min }\frac{I^c\left(y\right)}{A^c}\right)$

Wherein, ω is fog factor, removes degree in order to regulate fog, it is to avoid mist elimination is not enough or excessively mist elimination and the image that causes Distortion phenomenon, the fog factor that goes chosen is 0.95.

Image mist elimination absorbance optimization method based on dark channel prior the most according to claim 1, it is characterised in that:

Step 5) in thick mist elimination time, the formula of thick mist elimination is as follows:

$J_1^c\left(x\right)=\frac{I^c\left(x\right)-A^c}{\max (0.1,t_1(x))}-A^c,c\∈(r,g,b)$

Wherein, max is maximum operation, in order to avoid effect of noise, absorbance limits a lower limit 0.1, prevents J X there is the situation of negative value in ().

Image mist elimination absorbance optimization method based on dark channel prior the most according to claim 1, it is characterised in that:

Step 6) in absorbance optimize time, concrete optimization method is:

1) intermediate object program J is calculated₁The minima passage figure of (x)

2) to rough absorbance t₁X () is optimized, absorbance t after being optimized₂(x), employing equation below:

$t_2\left(x\right)=\frac{A^{\min }-I^{dark}\left(x\right)}{\max (1,A^{\min }-J_1^{\min }(x))}$

Wherein, max for doing maximum operation, A^minFor the minima of air light value, I^darkX () is dark；

3) utilize Steerable filter that the absorbance after optimizing is carried out process of refinement, eliminate blocking effect, after obtaining final optimization pass Absorbance t (x).

Image mist elimination absorbance optimization method based on dark channel prior the most according to claim 1, it is characterised in that:

Step 7) in when restoring without mist image, the air light value A that is previously obtained of utilization and absorbance t (x) can be according to physics moulds Type reconstruct without mist image J (x), employing equation below:

$J^c\left(x\right)=\frac{I^c\left(x\right)-A^c}{\max (0.1,t(x))}-A^c,c\∈(r,g,b)$

Wherein, in order to avoid effect of noise, absorbance is limited a lower limit 0.1, prevent J (x) from the situation of negative value occurring.