CN105225210A - A kind of self-adapting histogram based on dark strengthens defogging method capable - Google Patents

Abstract

The invention discloses a kind of self-adapting histogram based on dark and strengthen defogging method capable, it adopts dark model to ask for the dark figure of mist image, is some squares by dark diagram root, calculates the weight factor of each square; Then, adopt the implementation method of CLAHE, add up the blocked histogram of each square; Piecemeal restriction contrast histogram is calculated according to blocked histogram; Based on the histogram equalization grey scale mapping relation table that contrast histogram calculation square is corresponding, calculate contrast stretching grey scale mapping; Exploitation right repeated factor, grey scale mapping relation table and contrast stretching grey scale mapping, the grey scale mapping relation table that Computational block is final; The grey scale mapping table of four squares utilizing it to close on, the center pixel of every block adopts original grey scale mapping relation, and other pixel is obtained by the grey scale mapping interpolation of four blocks, obtains the image after demist; Finally, image exports.Dark model and traditional CLAHE algorithm for image enhancement combine by the present invention, have both effectively used the depth of view information in the greasy weather, and avoid again complicated transmission plot and estimate, process single image, mist elimination is effective, real-time.

Description

A kind of self-adapting histogram based on dark strengthens defogging method capable

Technical field

The present invention relates to digital image processing field, is more particularly a kind of rapid image mist elimination disposal route.

Background technology

In existing supervisory system, image is gathered under severe weather conditions, due to a large amount of haze particles suspended in air, absorption, scattering process can be produced to light, thus the image quality decrease causing camera acquisition to arrive, there is degradation phenomenon under image blurring, contrast, have a strong impact on the observing effect of operating personnel.Therefore the method and apparatus of many Penetrating Fogs is had to be applied in the middle of supervisory system at present.

This wherein roughly can be divided into two classes for the method for mist image procossing: a class is based on traditional algorithm for image enhancement, more representative method as restriction contrast self-adapting histogram equilibrium (CLAHE), these class methods do not consider the forming process that mist image is concrete, identical to the enhancing degree in depth of field region different in image.This kind of algorithm computation process is comparatively simple, and travelling speed is fast, is suitable for computing machine parallel practice.But do not consider the Formation rule of mist, be not suitable for all scenes particularly there is the greasy weather scene of the larger depth of field, there will be obvious color distortion.Another kind of, being the defogging method capable based on greasy weather model, is to there being mist image to carry out once recovering without mist image with the inverse process of imaging.This method is with strong points, obtains result nature, does not generally have information loss, can obtain good mist elimination effect.More representational is defogging method capable based on dark, and the atmospherical scattering model that this method proposes based on McCartney, by estimating atmospheric parameter and transmissivity, then recovers without mist image according to imaging model.But this class methods calculated amount is all very large, processes the time that a sub-picture needs at substantial, is difficult to requirement of real time, limit the widespread use of this algorithm at engineering field.

Summary of the invention

The technical problem to be solved in the present invention is existing mist image processing techniques or demist weak effect, or calculated amount is excessive, and restriction uses.

For solving the problems of the technologies described above, the technical solution used in the present invention is: a kind of self-adapting histogram based on dark strengthens defogging method capable, comprises the following steps: step one: read and have mist image I (x, y), based on dark model, ask the dark figure I of I (x, y) ^dark(x, y); Step 2: will I be schemed ^dark(x, y) is divided into the square Ω of several N*N anyhow _i,j, according to dark figure I ^dark(x, y) Computational block Ω _i,jweight factor C _i,j; Step 3: the implementation method adopting CLAHE, to there being mist image I (x, y) to process respectively, statistics Ω _i,jblocked histogram Hist _i,j, namely pixel corresponding in each pixel region in piecemeal is arranged, pixel identical for pixel value is come same row, then each row pixel is according to pixels worth the order arrangement formation histogram increased progressively; Step 4: according to Hist _i,jcalculate piecemeal restriction contrast histogram respectively utilize predefined threshold value to carry out cutting histogram, and the part these cropped is distributed to other parts histogrammic uniformly; Step 5: based on histogram computational block Ω _i,jcorresponding histogram equalization grey scale mapping relation table calculate Ω simultaneously _i,jcorresponding contrast stretching grey scale mapping step 6: utilize the weight factor C that step 2 obtains _i,j, the grey scale mapping relation table that obtains of step 5 with computational block Ω _i,jfinal grey scale mapping relation table Map _i,j: step 7: the grey scale mapping table Map of four squares utilizing it to close on _i,j, Map _{i+1, j}, Map _{i, j+1}, Map _{i+1, j+1}, adopt bilinear interpolation algorithm to calculate, calculate the grey scale mapping value that each pixel is corresponding one by one: the center pixel of every block adopts original grey scale mapping relation, and other pixel is obtained by the grey scale mapping interpolation of four blocks, obtains the image after demist; Step 8: image exports.

The present invention is based on dark model and calculate piecemeal depth of field weight factor, having carried out having the piecemeal restriction contrast histogram equalization of weight and contrast stretching to strengthen to there being mist image.Both under having ensure that the thick fog situation that the depth of field is larger, can by restriction contrast histogram equalization, effectively details in reduction mist, in turn ensure that and to promote in the color rendition of prospect mist image and contrast.The picture quality obtained after strengthening is suitable with dark algorithm.Compared to dark algorithm, blocked histogram is adopted to calculate and bilinear interpolation method of reducing, greatly reduce the number of times that each pixel-map relation calculates, merely add the calculated amount of some bilinear interpolations, and can parallel calculating method be adopted, the hardware such as most suitable FPGA carry out through engineering approaches realization.

Further, the detailed process of step one is: read and have mist image I (x, y), obtain the minimum value I in each pixel RGB component in image _g(x, y), stored in the gray-scale map that a width is identical with original image size, then carry out mini-value filtering to this gray-scale map, the radius of filtering is determined by window size, obtains the dark figure I of I (x, y) ^dark(x, y), wherein c (x, y) represents a window centered by pixel x, y: $I^{dark}(x,y)=\underset{x^{\′},y^{\′}\∈c(x,y)}{\min }\left(I_g\right(x,y\left)\right).$

Further, the detailed process of step 2 is: obtain relevant block Ω respectively _i,jinterior dark I ^darkthe average of (x, y) and maximal value then this average is normalized, using the weight factor C of the value after normalization as relevant block _i,j: theoretical according to dark, weight factor C _i,jcan be used for weighing Ω roughly _i,jdepth of field relation.

Further, the histogram equalization grey scale mapping table of step 5 to ask for formula as follows: map (k) represents the transfer function of kth gray level, and Hist (j)/N represents the pixel quantity summation of 0 ~ j gray level and the ratio of sum of all pixels N.

Further, the contrast stretching grey scale mapping table of step 5 implementation procedure is: carry out contrast stretching to histogram, and this contrast stretching is the conversion realizing dynamic range by calculating piecewise linear transform functional form, and formula is:

$Map\left(k\right)=\left\{\begin{array}{cc}\frac{y_1}{x_1}k& k<x_1\\ \frac{y_2-y_1}{x_2-x_1}(k-x_1)+y_1& x_1\&le;k\&le;x_2\\ \frac{255-y_2}{255-x_2}(k-x_2)+y_2& k>x_2\end{array}\right.$

Wherein x ₁, x ₂determine the tonal range needing conversion, y ₁, y ₂determine the slope of linear transformation.This contrast stretching is the conversion realizing dynamic range by calculating piecewise linear transform functional form, to reach the effect strengthening picture contrast.

Advantage of the present invention is: dark model and traditional CLAHE algorithm for image enhancement are combined, both effectively used the depth of view information in the greasy weather, avoids again complicated transmission plot and estimates.Process single image, mist elimination is effective, real-time.

Accompanying drawing explanation

Fig. 1 is defogging method capable process flow diagram of the present invention.

Fig. 2 histogram cutting amplitude limit schematic diagram.

Fig. 3 is piecewise linear function schematic diagram.

Fig. 4 is interpolation schematic diagram.

Embodiment

As shown in Figure 1, the inventive method comprises the following steps:

1, reading has mist image I (x, y), based on dark model, asks the dark figure I of I (x, y) ^dark(x, y).Detailed process is: read and have mist image I (x, y), obtain the minimum value I in each pixel RGB component in image _g(x, y), stored in the gray-scale map that a width is identical with original image size, then carry out mini-value filtering to this gray-scale map, the radius of filtering is determined by window size, obtains the dark figure I of I (x, y) ^dark(x, y), wherein c (x, y) represents a window centered by pixel x, y.

$I^{dark}(x,y)=\underset{x^{\&prime;},y^{\&prime;}\&Element;c(x,y)}{\min }\left(I_g\right(x,y\left)\right)$

2, by I ^dark(x, y) is divided into the square Ω of several N*N anyhow _i,jcarry out processing (N generally gets 8), according to dark figure I ^dark(x, y) Computational block Ω _i,jweight factor C _i,j, theoretical according to dark, this factor can be used for weighing Ω roughly _i,jdepth of field relation.

Detailed process is: obtain relevant block Ω respectively _i,jinterior dark I ^darkthe average of (x, y) and maximal value then this average is normalized, using the weight factor C of the value after normalization as relevant block _i,j.

$C_{ij}=\frac{I_{i,j}^{dark\_ave}}{I_{i,j}^{dark\_ave\_max}}$

3, the implementation method having mist image I (x, y) to use for reference CLAHE is processed respectively simultaneously, first add up Ω _i,jblocked histogram Hist _i,j; Specific practice is: arrange pixel corresponding in each pixel region in piecemeal, pixel identical for pixel value is come same row, then each row pixel is according to pixels worth the order arrangement formation histogram increased progressively.

4, according to Hist _i,jcalculate piecemeal restriction contrast histogram respectively computing method identical with CLAHE algorithm: utilize predefined threshold value to carry out cutting histogram to reach the object of restriction enlargement range, and the part these cropped is distributed to other parts histogrammic uniformly.

As shown in Figure 2, be contrast cutting clipping processes.The i.e. histogram of histogram formation after cutoff frequency also redistribution of input picture.

5, Computational block Ω _i,jcorresponding grey scale mapping relation table with histogram equalization grey scale mapping table implementation procedure is histogram equalization process, represent the transfer function of kth gray level.Hist (j)/N represents the pixel quantity summation of 0 ~ j gray level and the ratio of sum of all pixels N, because Map (k) is normalized numerical value, be converted to the color value of 0 ~ 255, needs to be multiplied by 255 again.

Contrast stretching grey scale mapping table implementation procedure is: carry out contrast stretching to histogram, and this contrast stretching is the conversion realizing dynamic range by calculating piecewise linear transform functional form, to reach the effect strengthening picture contrast.As shown in Figure 3, be piecewise linear function number curve.Formula is:

$Map\left(k\right)=\left\{\begin{array}{cc}\frac{y_1}{x_1}k& k<x_1\\ \frac{y_2-y_1}{x_2-x_1}(k-x_1)+y_1& x_1\&le;k\&le;x_2\\ \frac{255-y_2}{255-x_2}(k-x_2)+y_2& k>x_2\end{array}\right.$

Wherein x ₁, x ₂determine the tonal range needing conversion, y ₁, y ₂determine the slope of linear transformation.

6, exploitation right repeated factor C _i,jwith grey scale mapping relation table with computational block Ω _i,jfinal grey scale mapping relation table Map _i,j.: CLAHE is better compared with the enhancing effect of large regions to the depth of field in Misty Image that color distortion to a certain degree is then existed to prospect by the reason of two mapping table weightings in this step; And it is not traditional contrast stretching strengthens effect better to foreground image, good compared with the thick fog regional effect of large regions to the depth of field.Because C _i,jcan guestimate Ω _i,jdepth of field relation, so utilize C _i,jright with be weighted average, both eliminated fog preferably, and reduced again and maintain forecolor.

${\mathrm{Map}}_{i,j}=C_{i,j}\&CenterDot;{\mathrm{Map}}_{i,j}^S+(1-C_{i,j})\&CenterDot;{\mathrm{Map}}_{i,j}^C$

7, the grey scale mapping value that every pixel is corresponding is calculated one by one: the grey scale mapping table of four squares utilizing it to close on adopts bilinear interpolation algorithm to calculate.The center pixel of every block adopts original grey scale mapping relation, and other pixel is obtained by the grey scale mapping interpolation of four blocks.

Be illustrated in figure 4 Interpolation Process.Being arranged in figure figure notation is that the pixel of 3 parts adopts four grey scale mapping tables to carry out bilinear interpolation, (figure notation the is 2) part being positioned at edge then adopts left and right or two neighbouring grey scale mapping tables to carry out linear interpolation, and angle point block place (figure notation is 1) directly uses the grey scale mapping relation of this block.

Claims (5)

1. the self-adapting histogram based on dark strengthens a defogging method capable, it is characterized in that comprising the following steps:

Step one: read and have mist image I (x, y), based on dark model, ask the dark figure I of I (x, y) ^dark(x, y);

Step 2: will I be schemed ^dark(x, y) is divided into the square Ω of several N*N anyhow _i,j, according to dark figure I ^dark(x, y) Computational block Ω _i,jweight factor C _i,j;

Step 3: the implementation method adopting CLAHE, to there being mist image I (x, y) to process respectively, statistics Ω _i,jblocked histogram Hist _i,j, namely pixel corresponding in each pixel region in piecemeal is arranged, pixel identical for pixel value is come same row, then each row pixel is according to pixels worth the order arrangement formation histogram increased progressively;

Step 4: according to Hist _i,jcalculate piecemeal restriction contrast histogram respectively utilize predefined threshold value to carry out cutting histogram, and the part these cropped is distributed to other parts histogrammic uniformly;

Step 5: based on histogram computational block Ω _i,jcorresponding histogram equalization grey scale mapping relation table calculate Ω simultaneously _i,jcorresponding contrast stretching grey scale mapping

Step 6: utilize the weight factor C that step 2 obtains _i,j, the grey scale mapping relation table that obtains of step 5 with computational block Ω _i,jfinal grey scale mapping relation table Map _i,j:

Step 7: the grey scale mapping table Map of four squares utilizing it to close on _i,j, Map _{i+1, j}, Map _{i, j+1}, Map _{i+1, j+1}, adopt bilinear interpolation algorithm to calculate, calculate the grey scale mapping value that each pixel is corresponding one by one: the center pixel of every block adopts original grey scale mapping relation, and other pixel is obtained by the grey scale mapping interpolation of four blocks, obtains the image after demist;

Step 8: image exports.

2. a kind of self-adapting histogram based on dark according to claim 1 strengthens defogging method capable, it is characterized in that the detailed process of step one is: read and have mist image I (x, y), obtain the minimum value I in each pixel RGB component in image _g(x, y), stored in the gray-scale map that a width is identical with original image size, then carry out mini-value filtering to this gray-scale map, the radius of filtering is determined by window size, obtains the dark figure I of I (x, y) ^dark(x, y), wherein c (x, y) represents a window centered by pixel x, y: $I^{dark}(x,y)=\underset{x^{\&prime;},y^{\&prime;}\&Element;c(x,y)}{\min }\left(I_g\right(x,y\left)\right).$

3. a kind of self-adapting histogram based on dark according to claim 1 strengthens defogging method capable, it is characterized in that the detailed process of step 2 is: obtain relevant block Ω respectively _i,jinterior dark I ^darkthe average of (x, y) and maximal value then this average is normalized, using the weight factor C of the value after normalization as relevant block _i,j:

4. a kind of self-adapting histogram based on dark according to claim 1 strengthens defogging method capable, it is characterized in that the histogram equalization grey scale mapping table of step 5 to ask for formula as follows:

$Map\left(k\right)=\underset{j=0}{\overset{k}{\&Sigma;}}Hist\left(j\right)/N$

Map (k) represents the transfer function of kth gray level, and Hist (j)/N represents the pixel quantity summation of 0 ~ j gray level and the ratio of sum of all pixels N.

5. a kind of self-adapting histogram based on dark according to claim 1 strengthens defogging method capable, it is characterized in that the contrast stretching grey scale mapping table of step 5 implementation procedure is:

Carry out contrast stretching to histogram, this contrast stretching is the conversion realizing dynamic range by calculating piecewise linear transform functional form, and formula is:

$Map\left(k\right)=\left\{\begin{array}{cc}\frac{y_1}{x_1}k& k<x_1\\ \frac{y_2-y_1}{x_2-x_1}(k-x_1)+y_1& x_1\&le;k\&le;x_2\\ \frac{255-y_2}{255-x_2}(k-x_2)+y_2& k>x_2\end{array}\right.$

Wherein x ₁, x ₂determine the tonal range needing conversion, y ₁, y ₂determine the slope of linear transformation.