CN102682443B - Rapid defogging algorithm based on polarization image guide - Google Patents

Abstract

The invention discloses a rapid defogging algorithm based on polarization image guide, which comprises the following steps of: filtering an acquired conventional intensity image by using a minimum value for solving a dark primary image; estimating an atmosphere parameter by using the dark primary image and solving a coarse restored image of an original foggy weather image; obtaining a guide image with better details after fusing acquired polarization images with different angles; and fusing a polarization guide image with the coarse restored image to obtain a final fine restored image. According to the invention, the coarse restored image is obtained by using dark channel prior, and the details of the coarse restored image are reinforced by using the polarization guide image. The invention has the advantages of restoration and reinforcement, has no any optimization and complex iteration operations, and is high in computing speed and good in restoring effect.

Description

Based on the Quick demisting algorithm of polarization image guiding

Technical field

The present invention relates to image processing field and optical field and intersect, be specifically related to a kind of Quick demisting algorithm based on polarization image guiding, can be applicable to outdoor vision system.

Background technology

The continuous growth of various application scenarios to computer vision demand, its application is more and more extensive, as the safety detection of sensitive sites, video monitoring system, self-navigation etc.But the outdoor vision system of most is not considered the impact of weather in the time of design, namely can only be applied to sunny weather or indoor environment.Mist is modal weather, and along with the increasing of mankind's activity scope and frequency, in air, dust granules increases, and haze weather frequently occurs in recent years.This causes outdoor vision system afunction.Therefore, Quick demisting is significant to outdoor vision system.

Outdoor vision system is a lot, as some accident of highway is easily sent out section, highway bayonet socket monitoring position, the auxiliary driving supervisory system of automobile, the monitoring of the sensitizing ranges such as residential quarter, campus, commercial square, square, railway station, the key monitoring regions such as generating plant and power transmitting device.Outdoor vision system relates to a lot of industries such as transportation industry, public security industry, education, aviation, digital product, remote sensing image processing, food security supervision.

Misty Image is because contrast and color characteristic are attenuated, for further image is processed as conventional image is cut apart, target identification and tracking etc. have increased difficulty.For this problem, most of vision system need to first carry out image enhancing or image restoration, to obtaining sharpening scene information.

Image enchancing method is not considered the forming process of Misty Image, but interested part in image is strengthened, and conventionally can improve contrast, outstanding details, but the problem such as these class methods often have such as nature of mist elimination color of image, and algorithm adaptability is not strong.And image recovery method is by greasy weather imaging process being set up to degradation model, and expect that this process of inverting is to obtain the best estimate of original image.

Compare Enhancement Method, these class methods are with strong points, relatively nature and do not have information loss of the result obtaining, conventionally can obtain better effect, therefore be the emphasis of people research in recent years, wherein help transcendental method the most noticeable (referring to " Single Image Haze Removal Using Dark Channel Prior ") secretly by propositions such as He Kaiming.The core of the method is to utilize statistical law to show that Misty Image exists dark channel characteristic, and utilizes this characteristic to obtain mist elimination result.In literary composition, author has used soft stingy drawing method to obtain accurate transmissivity, but soft stingy drawing method calculated amount is huge, has therefore limited its application.

Summary of the invention

In order to solve greasy weather transmitting image degenerate problem, the invention provides one Image Restoration Algorithm fast, adopt minimum filters to ask for and help figure secretly; Estimate on this basis atmospheric parameter and ask for coarse palinspastic map; Utilize Polarization Image Fusion to generate the good navigational figure of details, merge and obtain final meticulous restored image with coarse palinspastic map.

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

1, the rapid defogging method based on polarization image guiding, is characterized in that, specifically comprises the following steps:

1) polarization image (having polaroid) of collection one width normal image (without polaroid) and three width different angles;

2) the width normal image collecting and three width polarization images are processed respectively: obtain coarse palinspastic map with normal image, with obtaining polarization guiding figure after Polarization Image Fusion;

A. obtain coarse palinspastic map with normal image, its step comprises:

If original normal image I is cromogram, ask its R, G, the three-channel minimum value of B, obtains a width gray-scale map I ^gray; If original image is gray-scale map, I ^gray=I; Then to I ^grayutilize mini-value filtering to obtain dark primary figure I ^dark, wave filter size is N*N; In general, N more computing time longer, N is less, and to help hypothesis secretly more inaccurate, following the example of of compromise is N=7, all guaranteed in computing time and accuracy.These dark primary image values are arranged by the order of successively decreasing again, determined that numerical values recited is point residing position in dark primary image of front 0.1%, the maximum brightness value in original normal image region corresponding to these positions is atmosphere optical parameter A;

Coarse transmission diagram is (position of x represent pixel point):

$t\left(x\right)=1-0.95*\left(\frac{I^{\mathrm{dark}}}{\max \_A}\right)---\left(1\right)$

Wherein, if original graph is max_A=A of gray-scale map, if original graph is max_A=max (A of cromogram _r, A _g, A _b);

Obtaining coarse palinspastic map is:

$I^{\mathrm{coarse}}\left(x\right)=\frac{I\left(x\right)-A}{\max (t\left(x\right),t_0)}+A---\left(2\right)$

B. obtain a width polarization guiding figure I with three width Polarization Image Fusions ^guide, its step comprises:

B1., when the direction of shaking thoroughly of polarization image and reference direction (0 ° of selected direction) angle is θ, the polarization image of the direction of thoroughly shaking arbitrarily can be expressed as:

$I\left(\mathrm{\θ}\right)=\frac{1}{2}I+\frac{1}{2}Q\cos 2\mathrm{\θ}+\frac{1}{2}U\sin 2\mathrm{\θ}---\left(3\right)$

The Stokes vector that I, Q, U are polarization image, I (θ) represents the light intensity of polarization image emergent light in the time that θ shakes direction thoroughly, and I represents total light intensity, and Q represents the poor of X-direction and Y direction polarization image component, and U represents ± 45 ⁰direction polarization image component poor;

The angle of supposing thoroughly shake direction and the reference direction of the three width polarization images that collect is respectively θ ₁, θ ₂, θ ₃, corresponding polarization image is respectively I (θ ₁), I (θ ₂), I (θ ₃); Solve thus Stokes vector I, Q, the U of polarization image; This is the basis that polarization merges;

B2. utilize stokes vector I, Q, U, try to achieve polarization image degree of polarization:

$\mathrm{DOP}=\frac{\sqrt{Q^2+U^2}}{I},0\≤\mathrm{DOP}\≤1---\left(4\right)$

B3. for the contrast of different target object in further stretching polarization image, introduce a degree of polarization stretching function; Define the stretching function M of DOP _dOPfor:

M _DOP=DOP*log ₂(1+DOP) (5)

Utilize enhancing formula calculating polarization guiding figure to be:

I ^guide=(I-Mean(I))(max(M _DOP)-M _DOP) (6)

3) Design guidance wave filter carries out filtering to coarse palinspastic map, more coarse palinspastic map and polarization navigational figure are merged, and obtains image, i.e. restored image after mist elimination;

A. Design guidance wave filter, utilizes polarization guiding figure to carry out filtering to coarse palinspastic map, obtains palinspastic map I ^dehaze, it thes contents are as follows:

A1. guide the filtering equations of wave filter to be:

$I_i^{\mathrm{dehaze}}=\underset{j}{\mathrm{\Σ}}{W}_{\mathrm{ij}}\left(I^{\mathrm{guide}}\right)I_j^{\mathrm{coarse}}---\left(7\right)$

Wherein, I ^guidea mistake! Do not find Reference source.For navigational figure, I ^coarsefor pending input picture, I ^dehazefor output image;

$W_{\mathrm{ij}}=\frac{1}{{\left|W\right|}^2}\underset{k:(i,j)\∈w_k}{\mathrm{\Σ}}(1+\frac{(I_i^{\mathrm{guide}}-{\mathrm{\μ}}_k)(I_j^{\mathrm{guide}}-{\mathrm{\μ}}_k)}{{\mathrm{\σ}}_k^2+\mathrm{\ϵ}})---\left(8\right)$

i ^guidethe window w that is k at center pixel _kan interior linear transformation, μ _kwith respectively navigational figure I ^guideat window w _kinterior average and variance, | w| is number of pixels in window, and ε prevents regularization parameter;

A2. utilize formula (7) to I ^coarsecarry out filtering, utilize pyramid decomposition method on every one deck to I ^coarseand I ^guidemerge, obtain image, i.e. restored image I after mist elimination ^dehaze;

4) adopt color transfer technology, the restored image of coloured image is carried out to color adjustment (if the restored image of gray level image, this step can be economized), it thes contents are as follows:

A. original greasy weather intensity image I is done to white balance processing, obtain having the image I of better color representation ^wbas with reference to image;

B. by I ^wband I ^dehazebe transformed into respectively YIQ color space;

C. utilize reference picture on corresponding passage, to adjust respectively in average and the standard deviation of Y, I, Q passage, be adjusted the image after color, adjustment formula is as follows:

$\left\{\begin{array}{c}{Y}_{\mathrm{cS}}=\frac{{\mathrm{\σ}}_R^Y}{{\mathrm{\σ}}_S^Y}(Y_S-{\mathrm{\μ}}_S^Y)+{\mathrm{\μ}}_R^Y\\ I_{\mathrm{cS}}=\frac{{\mathrm{\σ}}_R^I}{{\mathrm{\σ}}_S^I}(I_S-{\mathrm{\μ}}_S^I)+{\mathrm{\μ}}_R^I\\ Q_{\mathrm{cS}}=\frac{{\mathrm{\σ}}_R^Q}{{\mathrm{\σ}}_S^Q}(Q_S-{\mathrm{\μ}}_S^Q)+{\mathrm{\μ}}_R^Q\end{array}\right.---\left(9\right)$

Wherein subscript R, S, the subfix of cS represents respectively reference picture, adjusts the restored image before color and adjusts the restored image after color; Finally final restored image is transformed into rgb space from YIQ space, so that show.

Step 2) in steps A 1 and A2 there is no precedence relationship, can carry out simultaneously.If employing multi-thread programming, also can greatly reduce working time.

The principles of science of technical scheme of the present invention institute foundation:

The physical principle that Misty Image is restored: under greasy weather condition, the reason that causes image degradation mainly contains two: the light intensity of first target own is owing to being subject to the absorption of suspended particles in atmosphere and scattering process and energy-producing decay, it can reduce brightness of image conventionally, causes picture contrast to decline; Another is that the ambient lightings such as skylight are subject to the scattering process of atmospheric particles and form parasitic light, and it can make image blurring conventionally, causes not nature of image color.This transmitting procedure can be described as following greasy weather imaging model:

$I(\mathrm{\θ},d)=J(\mathrm{\θ},d)t(\mathrm{\θ},d)+A_{\∞}(1-t(\mathrm{\θ},d))---\left(10\right)$

Wherein, I is illustrated in distance for d, and observation angle is the brightness value of θ place observed object, and J represents the brightness of target itself, A _∞be infinite point atmosphere luminance brightness, t has represented the transfer rate of light intensity, is conventionally called again transmission diagram.

Recovery formula is as follows:

$J(\mathrm{\θ},d)=\frac{I(\mathrm{\θ},d)-A_{\∞}(1-t(\mathrm{\θ},d))}{t(\mathrm{\θ},d)}---\left(11\right)$

Beneficial effect of the present invention is:

The present invention first utilizes dark primary priori to obtain coarse palinspastic map, then strengthens this coarse figure details by polarization navigational figure, and the present invention has the advantage of restoring and strengthening concurrently, and without any optimizing and complicated interative computation, computing velocity is rapid, and recovery effect is good.

Brief description of the drawings

Fig. 1 is algorithm flow chart of the present invention.

Embodiment

As shown in Figure 1, based on the rapid defogging method of polarization image guiding, it is characterized in that, specifically comprise the following steps:

1) polarization image (having polaroid) of collection one width normal image (without polaroid) and three width different angles, the angle of thoroughly shake direction and the reference direction of three width polarization images is respectively 0 °, 30 °, 90 °;

2) the width normal image collecting and three width polarization images are processed respectively: obtain coarse palinspastic map with normal image, with obtaining polarization guiding figure after Polarization Image Fusion;

A. obtain coarse palinspastic map with normal image, its step comprises:

If original normal image I is cromogram, ask its R, G, the three-channel minimum value of B, obtains a width gray-scale map I ^gray; If original image is gray-scale map, I ^gray=I; Then to I ^grayutilize mini-value filtering to obtain dark primary figure I ^dark, wave filter size is 7x7, all guaranteed in computing time and accuracy.These dark primary image values are arranged by the order of successively decreasing again, determined that numerical values recited is point residing position in dark primary image of front 0.1%, the maximum brightness value in original normal image region corresponding to these positions is atmosphere optical parameter A;

Coarse transmission diagram is (position of x represent pixel point):

$t\left(x\right)=1-0.95*\left(\frac{I^{\mathrm{dark}}}{\max \_A}\right)---\left(1\right)$

Wherein, if original graph is max_A=A of gray-scale map, if original graph is max_A=max (A of cromogram _r, A _g, A _b);

Obtaining coarse palinspastic map is:

$I^{\mathrm{coarse}}\left(x\right)=\frac{I\left(x\right)-A}{\max \left(t\left(x\right){,t}_0\right)}+A---\left(2\right)$

B. obtain a width polarization guiding figure I with three width Polarization Image Fusions ^guide, its step comprises:

B1., when the direction of shaking thoroughly of polarization image and reference direction (0 ° of selected direction) angle is θ, the polarization image of the direction of thoroughly shaking arbitrarily can be expressed as:

$I\left(\mathrm{\θ}\right)=\frac{1}{2}I+\frac{1}{2}Q\cos 2\mathrm{\θ}+\frac{1}{2}U\sin 2\mathrm{\θ}---\left(3\right)$

The Stokes vector that I, Q, U are polarization image, I (θ) represents the light intensity of polarization image emergent light in the time that θ shakes direction thoroughly, and I represents total light intensity, and Q represents the poor of X-direction and Y direction polarization image component, and U represents ± 45 ⁰direction polarization image component poor;

The angle of supposing thoroughly shake direction and the reference direction of the three width polarization images that collect is respectively θ ₁, θ ₂, θ ₃, corresponding polarization image is respectively I (θ ₁), I (θ ₂), I (θ ₃); Solve thus Stokes vector I, Q, the U of polarization image; This is the basis that polarization merges;

B2. utilize stokes vector I, Q, U, try to achieve polarization image degree of polarization:

$\mathrm{DOP}=\frac{\sqrt{Q^2+U^2}}{I},0\≤\mathrm{DOP}\≤1---\left(4\right)$

B3. for the contrast of different target object in further stretching polarization image, introduce a degree of polarization stretching function; Define the stretching function M of DOP _dOPfor:

$M_{\mathrm{DOP}}=\mathrm{DOP}*{\log }_2(1+\mathrm{DOP})---\left(5\right)$

Utilize enhancing formula calculating polarization guiding figure to be:

$I^{\mathrm{guide}}=(I-\mathrm{Mean}\left(I\right))(\max \left(M_{\mathrm{DOP}}\right)-M_{\mathrm{DOP}})---\left(6\right)$

3) Design guidance wave filter carries out filtering to coarse palinspastic map, more coarse palinspastic map and polarization navigational figure are merged, and obtains image, i.e. restored image after mist elimination;

A. Design guidance wave filter, utilizes polarization guiding figure to carry out filtering to coarse palinspastic map, obtains palinspastic map I ^dehaze, it thes contents are as follows:

A1. guide the filtering equations of wave filter to be:

$I_i^{\mathrm{dehaze}}=\underset{j}{\mathrm{\Σ}}{W}_{\mathrm{ij}}\left(I^{\mathrm{guide}}\right)I_j^{\mathrm{coarse}}---\left(7\right)$

Wherein, I ^guidea mistake! Do not find Reference source.For navigational figure, I ^coarsefor pending input picture, I ^dehazefor output image;

$W_{\mathrm{ij}}=\frac{1}{{\left|w\right|}^2}\underset{k:(i,j)\∈w_k}{\mathrm{\Σ}}(1+\frac{(I_i^{\mathrm{guide}}-{\mathrm{\μ}}_k)(I_j^{\mathrm{guide}}-{\mathrm{\μ}}_k)}{{\mathrm{\σ}}_k^2+\mathrm{\ϵ}})---\left(8\right)$

i ^guidethe window w that is k at center pixel _kan interior linear transformation, θ _kwith respectively navigational figure I ^guideat window w _kinterior average and variance, | w| is number of pixels in window, and ε prevents regularization parameter;

A2. utilize formula (7) to I ^coarsecarry out filtering, utilize pyramid decomposition method on every one deck to I ^coarseand I ^guidemerge, obtain image, i.e. restored image I after mist elimination ^dehaze;

4) adopt color transfer technology, the restored image of coloured image is carried out to color adjustment (if the restored image of gray level image, this step can be economized), it thes contents are as follows:

A. original greasy weather intensity image I is done to white balance processing, obtain having the image I of better color representation ^wbas with reference to image;

B. by I ^wband I ^dehazebe transformed into respectively YIQ color space;

C. utilize reference picture on corresponding passage, to adjust respectively in average and the standard deviation of Y, I, Q passage, be adjusted the image after color, adjustment formula is as follows:

$\left\{\begin{array}{c}{Y}_{\mathrm{cS}}=\frac{{\mathrm{\σ}}_R^Y}{{\mathrm{\σ}}_S^Y}(Y_S-{\mathrm{\μ}}_S^Y)+{\mathrm{\μ}}_R^Y\\ I_{\mathrm{cS}}=\frac{{\mathrm{\σ}}_R^I}{{\mathrm{\σ}}_S^I}(I_S-{\mathrm{\μ}}_S^I)+{\mathrm{\μ}}_R^I\\ Q_{\mathrm{cS}}=\frac{{\mathrm{\σ}}_R^Q}{{\mathrm{\σ}}_S^Q}(Q_S-{\mathrm{\μ}}_S^Q)+{\mathrm{\μ}}_R^Q\end{array}\right.---\left(9\right)$

Wherein subscript R, S, the subfix of cS represents respectively reference picture, adjusts the restored image before color and adjusts the restored image after color; Finally final restored image is transformed into rgb space from YIQ space, so that show.

Claims (1)

1. the rapid defogging method based on polarization image guiding, is characterized in that, specifically comprises the following steps:

1) polarization image of collection one width normal image and three width different angles;

2) the width normal image collecting and three width polarization images are processed respectively: obtain coarse palinspastic map with normal image, with obtaining polarization guiding figure after Polarization Image Fusion;

A. obtain coarse palinspastic map with normal image, its step comprises:

If original normal image I is cromogram, ask its R, G, the three-channel minimum value of B, obtains a width gray-scale map I ^gray; If original normal image I is gray-scale map, I ^gray=I; Then to I ^grayutilize mini-value filtering to obtain dark primary figure I ^dark, wave filter size is N*N; These dark primary image values are arranged by the order of successively decreasing again, determined that numerical values recited is point residing position in dark primary image of front 0.1%, the maximum brightness value in original normal image region corresponding to these positions is atmosphere optical parameter A;

Coarse transmission diagram is:

$t\left(x\right)=1-0.95*\left(\frac{I^{\mathrm{dark}}}{\max \_A}\right)---\left(1\right)$

Wherein, the position of x represent pixel point, if original normal image I is max_A=A of gray-scale map, if original normal image I is max_A=max (A of cromogram _r, A _g, A _b);

Obtaining coarse palinspastic map is:

$I^{\mathrm{coarse}}\left(x\right)=\frac{I\left(x\right)-A}{\max (t\left(x\right),t_0)}+A---\left(2\right)$

B. obtain a width polarization guiding figure I with three width Polarization Image Fusions ^guide, its step comprises:

B1., when the direction of shaking thoroughly of polarization image and reference direction angle are θ, the polarization image of the direction of thoroughly shaking is arbitrarily expressed as:

$I\left(\mathrm{\θ}\right)=\frac{1}{2}{I}_0+\frac{1}{2}Q\cos 2\mathrm{\θ}+\frac{1}{2}U\sin 2\mathrm{\θ}---\left(3\right)$

I ₀, Q, the U Stokes vector that is polarization image, I (θ) represents the light intensity of polarization image emergent light in the time that θ shakes direction thoroughly, I ₀represent total light intensity, Q represents the poor of X-direction and Y direction polarization image component, represent ± 45 ° of direction polarization image components of U poor;

The angle of supposing thoroughly shake direction and the reference direction of the three width polarization images that collect is respectively θ ₁, θ ₂, θ ₃, corresponding polarization image is respectively I (θ ₁), I (θ ₂), I (θ ₃); Solve thus the Stokes vector I of polarization image ₀, Q, U;

B2. utilize stokes vector I ₀, Q, U, try to achieve polarization image degree of polarization:

$\mathrm{DOP}=\sqrt{\frac{Q^2+U^2}{I_0}},0\≤\mathrm{DOP}\≤1---\left(4\right)$

B3. for the contrast of different target object in further stretching polarization image, introduce a degree of polarization stretching function; Define the stretching function M of DOP _dOPfor:

M _DOP=DOP*log ₂（1+DOP） (5)

Utilize enhancing formula calculating polarization guiding figure to be:

I ^guide=(I-Mean(I))(max(M _DOP)-M _DOP) (6)

3) Design guidance wave filter carries out filtering to coarse palinspastic map, more coarse palinspastic map and polarization navigational figure are merged, and obtains image, i.e. restored image after mist elimination;

C. Design guidance wave filter, utilizes polarization guiding figure to carry out filtering to coarse palinspastic map, obtains palinspastic map I ^dehaze, it thes contents are as follows:

C1. guide the filtering equations of wave filter to be:

$I_i^{\mathrm{dehaze}}=\underset{j}{\mathrm{\Σ}}{W}_{\mathrm{ij}}\left(I^{\mathrm{guide}}\right)I_j^{\mathrm{coarse}}---\left(7\right)$

Wherein, I ^guidefor navigational figure, I ^coarsefor pending input picture, I ^dehazefor output image;

$W_{\mathrm{ij}}=\frac{1}{{\left|w\right|}^2}\underset{k:(i,j)\∈w_k}{\mathrm{\Σ}}(1+\frac{(I_i^{\mathrm{guide}}-{\mathrm{\μ}}_k)(I_j^{\mathrm{guide}}-{\mathrm{\μ}}_k)}{{\mathrm{\σ}}_k^2+\mathrm{\ϵ}})---\left(8\right)$

i ^guidethe window w that is k at center pixel _kan interior linear transformation, μ _kwith respectively navigational figure I ^guideat window w _kinterior average and variance, | w| is number of pixels in window, and ε prevents → 0 regularization parameter;

C2. utilize formula (7) to I ^coarsecarry out filtering, utilize pyramid decomposition method on every one deck to I ^coarseand I ^guidemerge, obtain image, i.e. restored image I after mist elimination ^dehaze;

4) adopt color transfer technology, the restored image of coloured image is carried out to color adjustment, if the restored image of gray level image, this step is omitted; It thes contents are as follows:

D. original normal image I is done to white balance processing, obtain having the image I of better color representation ^wbas with reference to image;

E. by I ^wband I ^dehazebe transformed into respectively YIQ color space;

F. utilize reference picture at Y, I _c, Q passage average and standard deviation on corresponding passage, adjust respectively, be adjusted the image after color, adjustment formula is as follows:

$\left\{\begin{array}{c}{Y}_{\mathrm{cS}}=\frac{{\mathrm{\σ}}_R^Y}{{\mathrm{\σ}}_S^Y}(Y_S-{\mathrm{\μ}}_S^Y)+{\mathrm{\μ}}_R^Y\\ I_{\mathrm{cS}}=\frac{{\mathrm{\σ}}_R^I}{{\mathrm{\σ}}_S^I}(I_S-{\mathrm{\μ}}_S^I)+{\mathrm{\μ}}_R^I\\ Q_{\mathrm{cS}}=\frac{{\mathrm{\σ}}_R^Q}{{\mathrm{\σ}}_S^Q}(Q_S-{\mathrm{\μ}}_S^Q)+{\mathrm{\μ}}_R^Q\end{array}\right.---\left(9\right)$

Wherein subscript R, S, the subfix of cS represents respectively reference picture, adjusts the restored image before color and adjusts the restored image after color; Finally final restored image is transformed into rgb space from YIQ space, so that show.