CN107424132B

CN107424132B - Optimization method for rapid image defogging

Info

Publication number: CN107424132B
Application number: CN201710610832.9A
Authority: CN
Inventors: 朱红; 柳青林
Original assignee: Xidian University
Current assignee: Xidian University
Priority date: 2017-07-25
Filing date: 2017-07-25
Publication date: 2020-07-07
Anticipated expiration: 2037-07-25
Also published as: CN107424132A

Abstract

The invention discloses an optimization method for rapid image defoggingThe method mainly comprises the following steps: acquiring an original RGB image, carrying out dark channel defogging on the original RGB image to obtain a RGB image subjected to dark channel defogging, and respectively obtaining the gray level image average brightness of the RGB image subjected to dark channel defogging; the RGB image comprises three color channels, each pixel point in the RGB image comprises three color channels, R represents a red channel, G represents a green channel, and B represents a blue channel; obtaining an RGB image with balanced contrast and defogging and the average brightness of the gray level image of the RGB image with balanced contrast and defogging of a dark channel according to the original RGB image, and then calculating the gray level image of the original RGB image and an original RGB image I_sourceThe average brightness of the gray level image is calculated to obtain the brightness unevenness measurement of the original RGB image; and finally, calculating to obtain a synthesized defogged image.

Description

Optimization method for rapid image defogging

Technical Field

The invention relates to the technical field of image defogging, in particular to an optimization method for rapid image defogging, which is suitable for rapidly defogging images or videos shot in haze weather.

Background

The dark channel prior defogging algorithm is a new direction for the rapid development of the image defogging field in the last two years, can well perform defogging operation on a foggy image, and is widely applied to the image defogging field; however, the original dark channel defogging algorithm has a large calculation cost during operations such as dark channel calculation, atmospheric optical parameter calculation, dark channel transmittance refinement and the like, and although some subsequent methods such as guided filtering and the like optimize the performance, the calculation amount is still huge, the model is complex, the parallelism degree is low, and the method is not suitable for running on a low-power-consumption embedded system; compared with a conventional algorithm such as a contrast-limited adaptive image equalization algorithm, the contrast-limited adaptive image equalization algorithm is a contrast enhancement algorithm applied to an image block, and has a certain effect of defogging, but due to the fact that the contrast-limited adaptive image equalization algorithm is a histogram equalization algorithm, supersaturation is easily caused, and the effect is not good enough for scenes with severe light and shade conversion.

Disclosure of Invention

Aiming at the defects in the prior art, the invention aims to provide an image rapid defogging optimization method which is an image defogging method based on a dark channel and contrast histogram equalization limitation and can effectively solve the problems of large calculation amount, complex model, non-robust complex environment and unsuitability for running on low-power consumption equipment in the conventional image defogging algorithm.

In order to achieve the technical purpose, the invention is realized by adopting the following technical scheme.

An optimization method for rapid defogging of an image comprises the following steps:

step 1, obtaining an original RGB image I_sourceFor the original RGB image I_sourceCarrying out dark channel defogging to obtain an RGB image I after dark channel defogging_HazeFreeAnd obtaining an RGB image I after dark channel defogging_HazeFreeAverage brightness L of gray scale image_hazefreemean；

The RGB image comprises three color channels, each pixel point in the RGB image comprises three color channels, R represents a red channel, G represents a green channel, and B represents a blue channel;

step 2, according to the original RGB image I_sourceObtaining an RGB image I after defogging of the contrast-balanced dark channel_HazeFree' and obtaining an RGB image I with the defogged contrast-equalized dark channels_HazeFree' average luminance L of gray-scale image_hazefreemean'；

Step 3, respectively calculating the original RGB image I_sourceGray scale image and original RGB image I_sourceAverage luminance L of the gray-scale image of_sourcemeanFurther calculating to obtain the original RGB image I_sourceThe luminance unevenness measure std;

step 4, according to the RGB image I after the dark channel defogging_HazeFreeRGB image I after defogging of contrast-balanced dark channel_HazeFree', RGB image I after dark channel defogging_HazeFreeAverage brightness L of gray scale image_hazefreemeanRGB image I after defogging of contrast-balanced dark channel_HazeFree' average luminance L of gray-scale image_hazefreemean', original RGB image I_sourceAverage luminance L of the gray-scale image of_sourcemeanAnd an original RGB image I_sourceThe brightness unevenness metric std, and finally calculating to obtain a synthetic defogged image I_HazeFree”。

The invention has the beneficial effects that: compared with the existing algorithm, the method has low computational complexity, can meet the requirement of real-time processing of an embedded system, and can be out of work when the existing algorithm is used under severe conditions, such as overexposure, underexposure and uneven image brightness, but the method can still keep good effect.

Drawings

The present invention will be described in further detail with reference to the accompanying drawings and specific embodiments.

FIG. 1 is a flow chart of an optimization method for fast image defogging according to the present invention;

FIG. 2 is a flow chart of the dark channel defogging method of the present invention;

FIG. 3 is a flow chart of obtaining a RGB image after contrast equalization defogging according to the present invention;

FIG. 4 is a table value diagram of a mapping lookup table of pixels (χ, γ) at the position where the rows are χ and the columns are γ in α th and β th rows of image blocks (α) after bilinear interpolation;

FIG. 5 shows an original RGB image I obtained by the method of the invention_sourceA flow chart of a luminance non-uniformity metric of (1);

FIG. 6 is a defogged image I synthesized using the present invention_HazeFree"is used in the following description.

Detailed Description

Referring to fig. 1, it is a flow chart of an optimization method for fast defogging of an image according to the present invention; the optimization method for the rapid defogging of the image comprises the following steps:

step 1, obtaining an original RGB image I_sourceFor the original RGB image I_sourceCarrying out dark channel defogging to obtain an RGB image I after dark channel defogging_HazeFreeAnd obtaining an RGB image I after dark channel defogging_HazeFreeAverage brightness L of gray scale image_hazefreemean。

Specifically, as shown in fig. 2, an RGB image is obtained, where RGB represents three color channels, and each pixel in the RGB image includes three color channels, where R represents a red channel, G represents a green channel, and B represents a blue channel; selecting three color channels of each pixel point in RGB image8-bit binary data of a track, denoted as original RGB image I_sourceFor the original RGB image I_sourceDark channel processing, i.e. selecting original RGB image I respectively_sourceThe channel brightness value with the minimum brightness in the three color channels of each pixel point is obtained; further obtaining a dark channel image I_darkAnd respectively calculating dark channel image I_darkAverage brightness value L of all the pixels in the middle_darkmeanAnd dark channel image I_darkMaximum luminance L of_darkmaxThe dark channel image I_darkAverage brightness value L of all the pixels in the middle_darkmeanFor dark channel image I_darkDividing the sum of the brightness of all the pixel points by the total number of the pixel points, and obtaining a dark channel image I_darkMaximum luminance L of_darkmaxFor dark channel image I_darkThe brightness of the pixel point with the maximum brightness in all the pixel points.

Then, the dark channel image I is compared point by point_darkAnd dark channel image I_darkAverage brightness value L of all the pixels in the middle_darkmeanCalculating to obtain a transmission image L_trans，L_trans＝Min(I_dark,Min(0.9,1.3×L_darkmean÷255)×I_darkaverage) Min represents a function for comparing point by point to take a minimum value; i is_darkaverageRepresenting dark channel images I_darkThe average value of the brightness of the transversely adjacent 8 pixel points is obtained by the following steps:

dark channel image I_darkIs an N × M-dimensional matrix, i.e. N rows and M columns, N, M are positive integers respectively larger than 0, and any dark channel image I_darkBrightness i of pixel point in middle x row and y column_x,yFor dark channel image I_darkThe average value of the brightness of 8 transversely adjacent pixels (x, y) of the middle x-th row and the y-th column can be regarded as imean_x,yRepresenting dark channel images I_darkThe expression of the average value of the brightness of the transversely adjacent 8 pixel points is as follows:

wherein the subscripts y-3, y-2, or y-1, if any, are less than or equal to 0Subscripts y-3, y-2, or y-1 are set to 1, subscripts y +3, y +2, or y +1 are set to M if more than M, and x ∈ {1,2, …, N }, y ∈ {1,2, …, M }.

Min(I_dark,Min(0.9,1.3×L_darkmean÷255)×I_darkaverage) To regularize the smoothed image after the limiting process, Min (0.9,1.3 × L)_darkmean÷255)×I_darkaverageAnd dark channel image I_darkTo obtain a sum of original RGB image I_sourceThe transfer function measurement images with the same size are marked as transfer images L_trans。

Through L_A＝120+0.5×L_maxCalculating an atmospheric optical parameter estimation value L_A。

By passing

For the original RGB image I_sourceCarrying out dark channel defogging to obtain an RGB image I after dark channel defogging_HazeFreeHere, the original RGB image I may be used if there is a demand for the calculation speed_sourceAnd transmit the image L_transAs an RGB image after dark channel defogging; and rough stretching is carried out for compensation according to the brightness before fusion, so that the defogging quality of a dark channel is not seriously influenced.

Then calculating the RGB image I after the defogging of the dark channel_HazeFreeThe average brightness of the gray level image is calculated by firstly calculating the RGB image I after the defogging of the dark channel_HazeFreeThe process of obtaining the gray scale image is as follows:

dark channel defogged RGB image I_HazeFreeContains N × M pixel points, and the three color channel values of the nth pixel point are recorded as (R)_n,G_n,B_n)，R_nRed channel value, G, representing the nth pixel_nGreen channel value, B, representing the nth pixel_nExpressing the blue channel value of the nth pixel point, and calculating the Gray value Gray of the nth pixel point_n,

Gray_n＝(R_n×299+G_n×587+B_n× 114+ 500)/(1000, making n take 1 toN × N, obtaining the Gray value Gray of the 1 st pixel point respectively₁Gray value Gray of pixel points from N × M_N×MMarked as RGB image I after dark channel defogging_HazeFreeThe gray scale image of (1).

Further obtaining the RGB image I after the defogging of the dark channel by averaging_HazeFreeAverage brightness L of gray scale image_hazefreemean(ii) a The RGB image I after the dark channel defogging_HazeFreeAverage brightness L of gray scale image_hazefreemeanRGB image I after defogging dark channel_HazeFreeThe average value of the brightness of all the pixel points in the gray level image.

Step 2, according to the original RGB image I_sourceObtaining an RGB image I after defogging of the contrast-balanced dark channel_HazeFree' and obtaining an RGB image I with the defogged contrast-equalized dark channels_HazeFree' average luminance L of gray-scale image_hazefreemean'。

Specifically, the RGB image after contrast equalization defogging is calculated as shown in fig. 3; firstly, input original RGB image I_sourceBlocking to obtain N_crop×N_cropImage block, N_cropIs a positive integer greater than 0, N in this embodiment_crop＝10；N_crop×N_cropThe length and the width of each image block in each image block are respectively the same, if the length and the width of each image block are different, the image blocks are cut out through conventional operation after mirror image expansion and defogging processing are carried out on the edges of the corresponding image blocks; each image block is W in length_patchEach image block is H in width_patch(ii) a And the number of the pixel points in each image block is respectively the same and is marked as numpixInTile which is a positive integer greater than 0.

Setting a reduction coefficient as ClipLimit, wherein the ClipLimit is 0.001, and the reduction coefficient is used for controlling the defogging level; using the reduction coefficient as a contrast limiting parameter, wherein each pixel point in each image block is numpixInTile, and then calculating to obtain an original RGB image I_sourceThe reduction amount of numClipLimit (b) of (a),

numBins represents the original RGB image I_sourceAnd numBins is a positive integer greater than 0; original RGB image I in this embodiment_sourceBinary data, which takes values from 0 to 255, so that numBins takes values of 256 here; round represents the rounding function.

(1) Initialization let t ' denote the t ' th image block, t ' ∈ {1,2, …, N_crop×N_crop}，N_crop×N_cropRepresenting the original RGB image I_sourceThe total number of image blocks contained after the partitioning; and the t' th image block corresponds to the original RGB image I_sourceThe image blocks (a, b) in the row a and the column b,

t' has an initial value of 1, and (a, b) has an initial value of (1, 1).

(2) Calculating to obtain histogram vectors of three color channels of the RGB image of the t ' th image block, and averaging the histogram vectors of the three color channels of the RGB image of the t ' th image block to obtain an average histogram vector imgHist of the three color channels of the RGB image of the t ' th image block_t'Then, the limited contrast overflow value totalexpass of the t' th image block is calculated_t'，

totalExcess_t'＝sum(max(imgHist_t'Numlimit, 0)), max is the max-max operation, round is the rounding function, and sum is the sum function.

(3) Calculating to obtain the overflow average quantity avgBinIncr of the t' th image block_t'，

avgBinIncr_t'＝round(totalExcess_t'/numBins)

Then, the histogram average vector imgHist of three color channels of the RGB image of the t' th image block is used_t'The number numpixel of all pixels with the middle color channel brightness value li_liAnd RGB image I_sourceThe reduction numClipLimit is compared, if the brightness value of the color channel is li, the numpixel of all the pixel points is_liIf the color channel brightness value is greater than the numClipLimit, all the images of all the pixel points with the color channel brightness value of li are processedNumber of pixel points numpixel_liCut down to numclippi limit; if the brightness value of the color channel is li, the numpixel of all the pixel points is_liThe upper limit value of the number of the pixel points of the image block smaller than numClipLimit and larger than tth_t'，upperLimit_t'＝numClipLimit-avgBinIncr_t'Then, the number numpixel points of all the pixel points with the color channel brightness value li are set_liIncreasing to numClipLimit, and further obtaining the histogram average vector imgHist of three color channels of the RGB image of the t' th image block_t'The final pixel point number with the middle color channel brightness value li; simultaneously enabling the limited contrast overflow value totalexpass of the t' th image block_t'Subtract (numClipLimit-numpexel)_li) Li ∈ {1,2, …, S ' }, S ' denotes the histogram average vector imgHist of the three color channels of the RGB image of the t ' th image block_t'The initial value of li is 1.

(4) Respectively taking 1 to S 'for li, and repeatedly executing the step (3) to respectively obtain histogram average vectors imgHist of three color channels of the RGB image of the t' -th image block_t'Average histogram vector imgHist of three color channels of RGB image from final pixel point number with middle color channel brightness value of 1 to t' th image block_t'The final number of pixels with the brightness value of the middle color channel being S 'is recorded as the average vector imgHist of histograms of three color channels of the RGB image of the t' th image block_t'The final number of pixels in (1).

(5) The histogram average vector imgHist of three color channels of RGB image of t' th image block_t'The number of final pixel points in (1) is less than upperLimit_t'The sum of the number of pixels of the color channel brightness value and avgbinIncr_t'Simultaneously enabling the limited contrast overflow value totalexpass of the t' th image block_t'Minus avgBinIncr_t'Obtaining the optimized histogram average vector of the three color channels of the RGB image of the t' th image block

(6) If the limited contrast of the t' th image block overflowsOutput totalexpess_t'0, then there is no need to optimize the histogram average vector for the three color channels of the RGB image of the t' th image block

Compensation is carried out, and transposition is carried out (9); if the limited contrast overflow value totalexpass of the t' th image block_t'If not 0, then the optimized histogram average vector of the three color channels of the RGB image of the t' th image block

Performing compensation operation to obtain the final optimized histogram average vector of the three color channels of the RGB image of the t' th image block

The process is as follows:

(7) initialization, namely, enabling k to represent the kth brightness, enabling k ∈ {0,1, …, M }, enabling M to represent the maximum brightness which can be represented by the tth image block, enabling the selected original RGB image to be 8-bit image data in the embodiment, enabling M to be 255, enabling the initial value of k to be 0, and setting the walking number of the tth image block to be stepSize_t'，

stepSize_t'＝max(round(numBins/totalExcess_t'),1)。

(8) If the optimized histogram average vector of three channels of RGB image of t' th image block

If the number of the pixel points corresponding to the kth brightness is less than numClipLimit, the number of the pixel points corresponding to the kth brightness is added with 1, and meanwhile, the limited contrast overflow value totalexpess of the t' th image block is enabled_t' Minus 1.

If the limited contrast overflow value totalexpass of the t' th image block_t'If the number of the color channels is zero, the iteration is stopped, and the optimized histogram average vectors of the three color channels of the RGB image of the t' th image block corresponding to the time when the iteration is stopped are calculated

Final optimized histogram average vector of three color channels of RGB image marked as t' th image block

Then turning to (10); otherwise go to (9).

(9) Let k add stepSize_t'And if k ≦ M returning (8) if k ≦ M>M then returns k to 0 (8).

(10) Final optimized histogram average vector for three color channels of RGB image of t' th image block

Carrying out conventional accumulation summation operation on the color brightness values of all the pixel points to obtain an accumulated distribution function HistSum of the t' th image block_t'。

(11) According to the cumulative distribution function HistSum of the t' th image block_t'Respectively setting the exponential distribution coefficients alpha of the t' th image block_t'，alpha_t'0.4; setting a first distribution coefficient vmax of a t' th image block_t'，vmax_t'1-exp (-alpha); setting a second distribution coefficient val of the t' th image block_t'，val_t'＝(vmax_t'×HistSum_t'NumPixInTile) and when val_t'When the value is more than or equal to 1, the value is regulated to be 1-eps, and the log0 is prevented from occurring; setting a third distribution coefficient temp of the t' th image block_t'，temp_t'＝-1/alpha×log(1-val_t') And then calculating to obtain mapping lookup table mapping of the t' th image block_t'，mapping_t'＝min(temp_t'×M_l,M_l)。

(12) Let t' take 1 to N respectively_crop×N_cropAnd (2) to (11) are repeatedly executed until the mapping lookup table mapping of the 1 st image block is obtained₁To Nth_crop×N_cropMapping lookup table for image blocks

Is recorded as an original RGB image I_sourceN of (A)_crop×N_cropA mapping look-up table.

(13) Let the original RGB image I_sourceN of (A)_crop×N_cropThe q-th image block of the image blocks comprises W_qIndividual pixel point, original RGB image I_sourceN of (A)_crop×N_cropThe q image block in the image blocks corresponds to an original RGB image I_sourceα th row, β th column of image blocks (α), and

and the pixel points at the α th row and β th column image block (α) with the row number of x and the column number of gamma are (x, gamma),

H_patchrepresenting an original RGB image I_sourceMaximum number of lines per image block, W_patchRepresenting an original RGB image I_sourceThe maximum number of columns per image block.

(14) And (3) carrying out bilinear interpolation on pixel points (χ, γ) at the positions of rows χ and columns γ in the α th row and β th column image blocks (α) to obtain mapping lookup table values of the pixel points (χ, γ) at the positions of rows χ and columns γ in the α th row and β th column image blocks (α) after bilinear interpolation.

As shown in fig. 4, after bilinear interpolation is obtained, mapping lookup table values ptmamping of pixel points (χ, γ) at positions where rows are χ and columns are γ in α th and β th image blocks (α) are calculated_(χ,γ)The expression is as follows:

wherein mapping_(α,β)Mapping, a lookup table representing α th and β th lines of image blocks (α) after bilinear interpolation_(α,β-1)Mapping a lookup table representing α th and β -1 th columns of image blocks (α -1) after bilinear interpolation_(α-1,β)Representing the search of the α -1 st line, β th column of image blocks (α -1, β) after bilinear interpolationTABLE, mapping_(α-1,β-1)A lookup table representing the block (α -1, β -1) of the image at line α -1 and column β -1 after bilinear interpolation, H_patchRepresenting an original RGB image I_sourceWidth of each image block, W_patchRepresenting an original RGB image I_sourceThe length of each image block in the image, and the subscript α -1 or β -1 is replaced by 1 when the value is 0.

Mapping lookup table values mapping of pixel points (χ, γ) at positions with row number χ and column number γ in α th and β th rows and columns of image blocks (α) after bilinear interpolation_(χ,γ)The red channel brightness value R of the pixel point (χ, γ) at the position where the line number is χ and the column number is γ in the α th line and β th line image blocks (α) after bilinear interpolation is obtained respectively_(χ,γ)And green channel brightness value G of pixel point (χ, γ) at position where row number is χ and column number is γ in α th row and β th column image block (α) after bilinear interpolation_(χ,γ)And the blue channel brightness value B of a pixel point (chi, gamma) at the position where the number of rows is chi and the number of columns is gamma in the α th row and β th column image block (α) after bilinear interpolation_(χ,γ)The luminance values (R) of the three color channels are recorded as pixel points (χ, γ) at the position where the number of rows is χ and the number of columns is γ in the α th and β th rows and columns of image blocks (α) after bilinear interpolation_(χ,γ),G_(χ,γ),B_(χ,γ))。

(15) According to the luminance values (R) of three color channels of pixel points (x, gamma) at the positions of the α th line and β th column of image block (α) with the number of lines x and the number of columns gamma after bilinear interpolation_(χ,γ),G_(χ,γ),B_(χ,γ)) Calculating to obtain three color channel brightness values I of pixel points (χ, γ) at positions with row number χ and column number γ in α th row and β th column image blocks (α) after defogging_source(χ, γ), whose expression is:

I_source(χ,γ)＝(mapping_(χ,γ)(R_(χ,γ)+1),mapping_(χ,γ)(G_(χ,γ)+1),mapping_(χ,γ)(B_(χ,γ)+1))。

(16) let (X, gamma) take (1,1) to (H) respectively_patch,W_patch) And (14) and (15) are repeatedly executed to respectively obtain the image blocks (α) of α th line and β th column after defogging, wherein the number of lines is 1,Pixel point (1,1) with 1 column number three color channel brightness value I_sourceThe number of rows W in the (1,1) to α th and β th image blocks (α) after defogging_patchThe number of rows is H_patchBrightness value I of three color channels of pixel point (chi, gamma)_source(H_patch,W_patch) And recorded as a defogged α th row and β th column image block (α).

(17) Let (α) take (1,1) to (N) respectively_crop,N_crop) And (14), (15) and (16) are sequentially and repeatedly executed, and the 1 st row and 1 st column image blocks (1,1) to the N th defogged line and column image blocks are respectively obtained_cropLine, N_cropColumn image block (N)_crop,N_crop) And is recorded as an RGB image I after contrast equalization defogging_HazeFree'。

RGB image I after calculating contrast balance defogging_HazeFree' average luminance L of gray-scale image_hazefreemean' in which the contrast is equalized and the RGB image I after defogging_HazeFreeThe calculation process of the gray image of' is:

RGB image I after contrast equalization defogging_HazeFree' includes N × M pixel points, the brightness value of RGB three color channels of the M pixel point is recorded as (R)_m,G_m,B_m) Calculating the gray value of the mth pixel point

Gray_m,Gray_m＝(R_m×299+G_m×587+B_m× 114+ 500)/(1000, and taking M from 1 to N × M to obtain RGB image I with balanced and defogged contrast_HazeFree' Gray Gray value of the 1 st pixel point₁RGB image I after contrast equalization and defogging_HazeFree' the Gray value Gray of the N × M pixel point_N×MAnd is recorded as an RGB image I after contrast equalization defogging_HazeFree' of the present invention.

The RGB image I after the contrast equalization defogging_HazeFree' average luminance L of gray-scale image_hazefreemean' RGB image I after contrast equalization defogging_HazeFree' the average value of the brightness of all the pixels in the gray image.

Step 3, respectively calculatingOriginal RGB image I_sourceGray scale image and original RGB image I_sourceAverage luminance L of the gray-scale image of_sourcemeanFurther calculating to obtain the original RGB image I_sourceThe luminance unevenness measure std.

3.1 in particular, with reference to FIG. 5, the original RGB image I is first calculated_sourceAverage luminance L of the gray-scale image of_sourcemeanAnd the original RGB image I_sourceN_crop×N_cropStandard deviation std of the gray-scale image luminance of the individual image blocks, wherein the original RGB image I_sourceThe calculation process of the gray level image is as follows:

original RGB image I_sourceThe brightness values of the RGB three color channels containing N × M pixel points and the M' th pixel point are recorded as (R)_m',G_m',B_m')，R_m'Red channel value, G, representing the m' th pixel_m'Green channel value, B, representing the m' th pixel_m'And the blue channel value of the m' th pixel point is represented.

3.2 computing the original RGB image I_sourceGray value Gray of middle m' th pixel point_m',

Gray_m'＝(R_m'×299+G_m'×587+B_m'×114+500)÷1000。

3.3 let M' take 1 to N × M respectively, repeat 3.2 to get the original RGB image I_sourceGray value Gray of middle 1 st pixel point₁To the original RGB image I_sourceGray value Gray of middle-Nth × M pixel point_N×MIs recorded as an original RGB image I_sourceGray scale image I_sourcegray。

From the original RGB image I_sourceGray scale image I_sourcegrayTo obtain the original RGB image I_sourceAverage luminance L of the gray-scale image of_sourcemeanThe process is as follows:

the original RGB image I_sourceGray scale image I_sourcegrayDividing the sum of the brightness of all the pixel points by the total number of the pixel points to obtain an original RGB image I_sourceAverage luminance L of the gray-scale image of_sourcemean。

3.4 pairs of original RGB image I_sourceGray scale image I_sourcegrayBlocking to obtain N'_crop×N'_cropGray scale image block, N'_cropIs a positive integer greater than 0, N 'in the embodiment'_crop＝10；N'_crop×N'_cropThe length and width of each image block in each gray image block are respectively the same, if the length and width of each image block are different, mirror image expansion needs to be carried out on the edge of the corresponding image block, and the original RGB image I is processed_sourceGray scale image I_sourcegrayThe gray scale image blocks in the α 'th row and the β' th column are (α ', β'),

the initial value of (α ', β') is (1, 1).

3.5 original RGB image I_sourceGray scale image I_sourcegrayAdding the brightness of all pixel points in the gray image blocks (α ', β') of the α 'th row and β' th column, and dividing the sum by the number of the pixel points in the image blocks (α ', β') to obtain an original RGB image I_sourceGray scale image I_sourcegrayAverage brightness L of the gray image blocks (α ', β') of the α 'th row and β' th column_{(α',β')mean}。

3.6 taking (α ', β ') as (1,1) to (N '_crop,N'_crop) And repeating the execution for 3.5 to obtain the original RGB image I_sourceGray scale image I_sourcegrayAverage brightness L of middle 1 st row and 1 st column gray scale image block (1,1)_(1,1)meanTo the original RGB image I_sourceGray scale image I_sourcegrayMedium to N'_cropLine and N'_cropColumn grayscale image Block (N'_crop,N'_crop) Average brightness of

Is recorded as N'_crop×N'_cropAverage gray-scale luminance of individual gray-scale image blocks

Calculated using statistical methodsTo N'_crop×N'_cropThe average gray scale brightness standard deviation of each gray scale image block is recorded as the original RGB image I_sourceThe luminance unevenness measure std.

Specifically, referring to fig. 6, the calculated dark channel defogged RGB image I_HazeFreeRGB image I after defogging by contrast equalization_HazeFree' As a mix input, calculate the original RGB image I_sourceThe fusion factor Q of (a) is,

q is set between 5 and 70, and if this range is exceeded, is set to the close boundary value of 5 or 70; l is_sourcemeanRepresenting an original RGB image I_sourceStd is the original RGB image I_sourceIs measured by brightness non-uniformity.

Finally, the synthetic defogged image I is obtained through calculation_HazeFree", the expression is:

wherein L is_hazefreemeanRepresenting a dark channel dehazed RGB image I_HazeFreeGray scale image average brightness of, L_hazefreemean' representing RGB image I after contrast equalization defogging_HazeFree' average luminance of gray-scale image, I_HazeFreeRepresenting RGB image after dark channel defogging, I_HazeFree' represents the RGB image after contrast equalization defogging;

performing brightness compensation operation on the RGB image subjected to defogging of the dark channel, and dividing by 8.0 to obtain regularization operation;

representing a dark channel dehazed RGB image I_HazeFreeThe fusion weight of (a) is calculated,

representing contrast-equalized dehazed RGB image I_HazeFree' fusion weight.

Claims

1. An optimization method for rapid defogging of an image is characterized by comprising the following steps:

step 2, according to the original RGB image I_sourceObtaining an RGB image I after defogging of the contrast-balanced dark channel_HazeFree' and obtaining an RGB image I with the defogged contrast-equalized dark channels_HazeFree' average luminance L of gray-scale image_hazefreemean′；

wherein, the original RGB image I_sourceThe luminance unevenness metric std of (1) represents an average gray luminance standard deviation of each row and each column of gray image blocks in the gray image of the original RGB image;

step 4, according to the RGB image I after the dark channel defogging_HazeFreeRGB image I after defogging of contrast-balanced dark channel_HazeFree', RGB image I after dark channel defogging_HazeFreeAverage brightness L of gray scale image_hazefreemeanRGB image I after defogging of contrast-balanced dark channel_HazeFree' average luminance L of gray-scale image_hazefreemean', original RGB image I_sourceAverage luminance L of the gray-scale image of_sourcemeanAnd an original RGB image I_sourceThe brightness unevenness metric std, and finally calculating to obtain a synthetic defogged image I_HazeFree″。

2. The optimization method for rapidly defogging images according to claim 1, wherein the process of the step 1 is as follows:

for the original RGB image I_sourceDark channel processing, i.e. selecting original RGB image I respectively_sourceThe channel brightness value with the minimum brightness in the three color channels of each pixel point is obtained; further obtaining a dark channel image I_darkAnd respectively calculating dark channel image I_darkAverage brightness value L of all the pixels in the middle_darkmeanAnd dark channel image I_darkMaximum luminance L of_darkmaxThe dark channel image I_darkAverage brightness value L of all the pixels in the middle_darkmeanFor dark channel image I_darkDividing the sum of the brightness of all the pixel points by the total number of the pixel points, and obtaining a dark channel image I_darkMaximum luminance L of_darkmaxFor dark channel image I_darkThe brightness of the pixel point with the maximum brightness in all the pixel points;

then, the dark channel image I is compared point by point_darkAnd dark channel image I_darkAverage brightness value L of all the pixels in the middle_darkmeanCalculating to obtain a transmission image L_trans，L_trans＝Min(I_dark，Min(0.9，1.3×L_darkmean÷255)×I_darkaverage) Min represents a function for comparing point by point to take a minimum value; i is_darkaverageRepresenting dark channel images I_darkThe average value of the brightness of the transversely adjacent 8 pixel points is obtained by the following steps:

dark channel image I_darkIs an N × M-dimensional matrix, i.e. N rows and M columns, N, M are positive integers respectively larger than 0, and any dark channel image I_darkBrightness i of pixel point in middle x row and y column_x，yFor dark channel image I_darkAverage value of brightness of 8 transversely adjacent pixel points (x, y) of the middle x-th row and the y-th column is obtained by using imean_x，yRepresenting dark channel images I_darkThe expression of the average value of the brightness of the transversely adjacent 8 pixel points is as follows:

wherein, if the subscript y-3, y-2 or y-1 is less than or equal to 0, the subscript y-3, y-2 or y-1 is set as 1, and if the subscript y +3, y +2 or y +1 is greater than M, the subscript y +3, y +2 or y +1 is set as M, x ∈ {1,2, …, N }, y ∈ {1,2, …, M };

Min(I_dark，Min(0.9，1.3×L_darkmean÷255)×I_darkaverage) To regularize the smoothed image after the limiting process, Min (0.9,1.3 × L)_darkmean÷255)×I_darkaverageAnd dark channel image I_darkTo obtain a sum of original RGB image I_sourceThe transfer function measurement images with the same size are marked as transfer images L_trans；

Through L_A＝120+0.5×L_maxCalculating an atmospheric optical parameter estimation value L_A；

By passing

For the original RGB image I_sourceCarrying out dark channel defogging to obtain an RGB image I after dark channel defogging_HazeFree；

Then calculateDark channel defogged RGB image I_HazeFreeThe average brightness of the gray level image is calculated by firstly calculating the RGB image I after the defogging of the dark channel_HazeFreeThe process of obtaining the gray scale image is as follows:

dark channel defogged RGB image I_HazeFreeContains N × M pixel points, and the three color channel values of the nth pixel point are recorded as (R)_n，G_n，B_n)，R_nRed channel value, G, representing the nth pixel_nGreen channel value, B, representing the nth pixel_nExpressing the blue channel value of the nth pixel point, and calculating the Gray value Gray of the nth pixel point_n，

Gray_n＝(R_n×299+G_n×587+B_n× 114+ 500)/(1000, and making N take 1-N × N to obtain Gray level Gray of the 1 st pixel₁Gray value Gray of pixel points from N × M_N×MMarked as RGB image I after dark channel defogging_HazeFreeThe gray scale image of (1);

3. The method as claimed in claim 2, wherein in step 2, the contrast-equalized dark channel defogged RGB image I_HazeFree' equalizing the dark channel defogged RGB image I with the contrast_HazeFree' average luminance L of gray-scale image_hazefreemean', its substeps are:

firstly, input original RGB image I_sourceBlocking to obtain N_crop×N_cropImage block, N_cropIs a positive integer greater than 0, N_crop×N_cropThe length and width of each image block in each image block are respectively the same, and the length of each image block is W_patchEach image block is H in width_patch(ii) a The number of pixel points in each image block is the same and is marked as numpixInTile which is a positive integer greater than 0;

setting the reduction coefficient as ClipLimit, and calculating to obtain an original RGB image I_sourceThe reduction amount of numClipLimit (b) of (a),

numBins represents the original RGB image I_sourceAnd numBins is a positive integer greater than 0;

t' has an initial value of 1, and (α, b) has an initial value of (1, 1);

(2) calculating to obtain histogram vectors of three color channels of the RGB image of the t ' th image block, and averaging the histogram vectors of the three color channels of the RGB image of the t ' th image block to obtain an average histogram vector imgHist of the three color channels of the RGB image of the t ' th image block_t′Then, the limited contrast overflow value totalexpass of the t' th image block is calculated_t′，

totalExcess_t′＝sum(max(imgHist_t′Numlimit, 0)), max is the maximum operation, round is the rounding function, sum is the summation function;

(3) calculating to obtain the overflow average quantity avgBinIncr of the t' th image block_t′，

avgBinIncr_t′＝round(totalExcess_t′/numBins)

Then will beHistogram average vector imgHist of three color channels of RGB image of t' th image block_t′The number numpixel of all pixels with the middle color channel brightness value li_liAnd RGB image I_sourceThe reduction numClipLimit is compared, if the brightness value of the color channel is li, the numpixel of all the pixel points is_liIf the color channel brightness value is greater than the numClipLimit, the number numpixel of all the pixel points with the color channel brightness value li is numpixel_liCut down to numclippi limit; if the brightness value of the color channel is li, the number numpexeI of all the pixel points is_liThe upper limit value of the number of the pixel points of the image block smaller than numClipLimit and larger than tth_t′，upperLimit_t′＝numClipLimit-avgBinIncr_t′Then, the number numpixel points of all the pixel points with the color channel brightness value li are set_liIncreasing to numClipLimit, and further obtaining the histogram average vector imgHist of three color channels of the RGB image of the t' th image block_t′The final pixel point number with the middle color channel brightness value li; simultaneously enabling the limited contrast overflow value totalexpass of the t' th image block_t′Subtract (numClipLimit-numpexel)_li) Li ∈ {1,2, …, S ' }, S ' denotes the histogram average vector imgHist of the three color channels of the RGB image of the t ' th image block_t′The initial value of li is 1;

(4) respectively taking 1 to S 'for li, and repeatedly executing the step (3) to respectively obtain histogram average vectors imgHist of three color channels of the RGB image of the t' -th image block_t′Average histogram vector imgHist of three color channels of RGB image from final pixel point number with middle color channel brightness value of 1 to t' th image block_t′The final number of pixels with the brightness value of the middle color channel being S 'is recorded as the average vector imgHist of histograms of three color channels of the RGB image of the t' th image block_t′The final number of pixel points in (1);

(5) the histogram average vector imgHist of three color channels of RGB image of t' th image block_t′The number of final pixel points in (1) is less than upperLimit_t′Of the color channel luminance valuesThe number of prime points plus avgBinIncr_t′Simultaneously enabling the limited contrast overflow value totalexpass of the t' th image block_t′Minus avgBinIncr_t′Obtaining the optimized histogram average vector of the three color channels of the RGB image of the t' th image block

(6) If the limited contrast overflow value totalexpass of the t' th image block_t′0, then there is no need to optimize the histogram average vector for the three color channels of the RGB image of the t' th image block

Compensation is carried out, and transposition is carried out (9); if the limited contrast overflow value totalexpass of the t' th image block_t′If not 0, then the optimized histogram average vector of the three color channels of the RGB image of the t' th image block

The process is as follows:

(7) initialization is performed by setting k to represent the k-th luminance, k ∈ {0,1, …, M }, M to represent the maximum luminance that can be represented by the t '-th image block, setting the initial value of k to 0, and setting the number of walks of the t' -th image block to stepSize_t′，

stepSize_t′＝max(round(numBins/totalExcess_t′)，1)；

If the number of the pixel points corresponding to the middle kth brightness is less than the numClipLimit, the kth brightness is orderedAdding 1 to the number of pixel points corresponding to k brightness, and simultaneously enabling the limited contrast overflow value totalexpess of the t' th image block_t′Subtracting 1;

if the limited contrast overflow value totalexpass of the t' th image block_t′If the number of the color channels is zero, the iteration is stopped, and the optimized histogram average vectors of the three color channels of the RGB image of the t' th image block corresponding to the time when the iteration is stopped are calculated

Then turning to (10); otherwise go to (9);

(9) let k add stepSize_t′And if k is less than or equal to M, returning to (8), and if k is more than M, setting k to 0 and returning to (8);

Carrying out conventional accumulation summation operation on the color brightness values of all the pixel points to obtain an accumulated distribution function HistSum of the t' th image block_t′；

(11) According to the cumulative distribution function HistSum of the t' th image block_t′Setting the index distribution coefficient α lpha of the t' th image block_t′，alpha_t′0.4; setting a first distribution coefficient vmax of a t' th image block_t′，vmax_t′1-exp (-alpha); setting a second distribution coefficient val of the t' th image block_t′，val_t′＝(vmax_t′×HistSum_t′(numPixInTile); setting a third distribution coefficient temp of the t' th image block_t′，temp_t′＝-1/alpha×log(1-val_t′) And then calculating to obtain mapping lookup table mapping of the t' th image block_t′，mapping_t′＝min(temp_t′×M_l，M_l)；

Is recorded as an original RGB image I_sourceN of (A)_crop×N_cropA mapping look-up table;

(13) let the original RGB image I_sourceN of (A)_crop×N_cropThe q-th image block of the image blocks comprises W_qIndividual pixel point, original RGB image I_sourceN of (A)_crop×N_cropThe g-th image block in the image blocks corresponds to an original RGB image I_sourceα th row, β th column of image blocks (α), and

for the α th row and β th column image block (α), the pixel point where the row number is χ and the column number is γ is (χ, γ),

H_patchrepresenting an original RGB image I_sourceMaximum number of lines per image block, W_patchRepresenting an original RGB image I_sourceThe maximum number of columns of each image block;

(14) carrying out bilinear interpolation on pixel points (χ, γ) at the positions of rows χ and columns γ in α th and β th image blocks (α) to obtain mapping lookup table values of the pixel points (χ, γ) at the positions of rows χ and columns γ in α th and β th image blocks (α) after bilinear interpolation;

calculating mapping lookup table values ptmapping of pixel points (χ, γ) at positions where the line number is χ and the column number is γ in α th line and β th column image blocks (α) after bilinear interpolation_(χ，γ)The expression is as follows:

wherein mapping_(α，β)Mapping, a lookup table representing α th and β th lines of image blocks (α) after bilinear interpolation_(α，β-1)Mapping a lookup table representing α th and β -1 th columns of image blocks (α -1) after bilinear interpolation_(α-1，β)Mapping a lookup table representing α -1 st and β th lines of image blocks (α -1, β) after bilinear interpolation_{(α-1，β-1)}A lookup table representing the block (α -1, β -1) of the image at line α -1 and column β -1 after bilinear interpolation, H_patchRepresenting an original RGB image I_sourceWidth of each image block, W_patchRepresenting an original RGB image I_sourceThe subscript α -1 or β -1 is replaced by 1 when the value is 0;

respectively obtaining the red channel brightness value R of the pixel point (chi, gamma) at the position where the line number is chi and the column number is gamma in the α th line and β th column image block (α) after bilinear interpolation_(χ，γ)And green channel brightness value G of pixel point (χ, γ) at position where row number is χ and column number is γ in α th row and β th column image block (α) after bilinear interpolation_(χ，γ)And the blue channel brightness value B of a pixel point (chi, gamma) at the position where the number of rows is chi and the number of columns is gamma in the α th row and β th column image block (α) after bilinear interpolation_(χ，γ)The luminance values (R) of the three color channels are recorded as pixel points (χ, γ) at the position where the number of rows is χ and the number of columns is γ in the α th and β th rows and columns of image blocks (α) after bilinear interpolation_(χ，γ)，G_(χ，γ)，B_(χ，γ))；

(15) According to the luminance values (R) of three color channels of pixel points (x, gamma) at the positions of the α th line and β th column of image block (α) with the number of lines x and the number of columns gamma after bilinear interpolation_(χ，γ)，G_(χ，γ)，B_(χ，γ)) Calculating to obtain three color channel brightness values I of pixel points (χ, γ) at positions with row number χ and column number γ in α th row and β th column image blocks (α) after defogging_source(χ, γ), whose expression is:

I_source(χ，γ)＝(ptmappmg_(χ，γ)(R_(χ，γ)+1)，ptmapping_(χ，γ)(G_(χ，γ)+1)，ptmapping_(χ，γ)(B_(χ，γ)+1))；

(16) let (X, gamma) take (1,1) to (H) respectively_patch，W_patch) And (14) and (15) are repeatedly executed to respectively obtain three color channel brightness values I of pixel points (1,1) with the row number of 1 and the column number of 1 in the image blocks (α) of α th row and β th column after defogging_sourceThe number of rows W in the (1,1) to α th and β th image blocks (α) after defogging_patchThe number of rows is H_patchPixel point of (H)_patch，W_patch) Three color channel luminance values I_source(H_patch，W_patch) Marked as a defogged α th row and β th column image block (α);

(17) let (α) take (1,1) to (N) respectively_crop，N_crop) And (14), (15) and (16) are sequentially and repeatedly executed, and the 1 st row and 1 st column image blocks (1,1) to the N th defogged line and column image blocks are respectively obtained_cropLine, N_cropColumn image block (N)_crop，N_crop) And is recorded as an RGB image I after defogging of a contrast-balanced dark channel_HazeFree′；

RGB image I after defogging of contrast-balanced dark channel_HazeFree' average luminance L of gray-scale image_hazefreemean' RGB image I after defogging for contrast-equalized dark channel_HazeFree' the average value of the brightness of all the pixels in the gray image.

4. The method as claimed in claim 3, wherein in step 3, the original RGB image I_sourceAverage luminance L of the gray-scale image of_sourcemeanAnd the original RGB image I_sourceThe luminance nonuniformity metric std of (1) is calculated by:

3.1 computing the original RGB image I_sourceAverage luminance L of the gray-scale image of_hazefreemean' and original RGB image I_sourceN_crop×N_cropStandard deviation std of the gray-scale image luminance of the individual image blocks, wherein the original RGB image I_sourceThe calculation process of the gray level image is as follows:

original RGB image I_sourceThe brightness values of the RGB three color channels containing N × M pixel points and the M' th pixel point are recorded as (R)_m′，G_m′，B_m′)，R_m′Red channel value, G, representing the m' th pixel_m′Green channel value, B, representing the m' th pixel_m′Expressing the blue channel value of the m' th pixel point;

3.2 computing the original RGB image I_sourceGray value Gray of middle m' th pixel point_m′，

Gray_m′＝(R_m′×299+G_m′×587+B_m′×114+500)÷1000；

3.3 let M' take 1 to N × M respectively, repeat 3.2 to get the original RGB image I_sourceGray value Gray of middle 1 st pixel point₁To the original RGB image I_sourceGray value Gray of middle-Nth × M pixel point_N×MIs recorded as an original RGB image I_sourceGray scale image I_sourcegray；

the original RGB image I_sourceGray scale image I_sourcegrayDividing the sum of the brightness of all the pixel points by the total number of the pixel points to obtain an original RGB image I_sourceAverage luminance L of the gray-scale image of_sourcemean；

3.4 pairs of original RGB image I_sourceGray scale image I_sourcegrayBlocking to obtain N'_crop×N′_cropGray scale image block, N'_cropIs a positive integer greater than 0; n'_crop×N′_cropThe length and the width of each image block in the gray-scale image blocks are respectively the same; the original RGB image I_sourceGray scale image I_sourcegrayThe gray scale image blocks in the α 'th row and the β' th column are (α ', β'),

the initial value of (α ', β') is (1, 1);

3.5 original RGB image I_sourceGray scale image I_sourcegrayAdding the brightness of all pixel points in the gray image blocks (α ', β') of the α 'th row and β' th column, and dividing the sum by the number of the pixel points in the image blocks (α ', β') to obtain an original RGB image I_sourceGray scale image I_sourcegrayThe average brightness of the gray image blocks (α ', β') of the α 'th row and β' th column_{(α′，β′)mean}；

3.6 taking (α ', β ') as (1,1) to (N '_crop，N′_crop) And repeating the execution for 3.5 to obtain the original RGB image I_sourceGray scale image I_sourcegrayAverage brightness L of middle 1 st row and 1 st column gray scale image block (1,1)_(1，1)meanTo the original RGB image I_sourceGray scale image I_sourcegrayMedium to N'_cropLine and N'_cropColumn grayscale image Block (N'_crop，N′_crop) Average brightness of

Is recorded as N'_crop×N′_cropAverage gray-scale luminance of individual gray-scale image blocks

Calculated to N 'using statistical methods'_crop×N′_cropThe average gray scale brightness standard deviation of each gray scale image block is recorded as the original RGB image I_sourceThe luminance unevenness measure std.

5. The method as claimed in claim 4, wherein in step 4, the synthesized defogged image isI_HazeFree", the expression is:

wherein Q represents the original RGB image I_sourceFusion factor of (1), L_hazefreemeanRepresenting a dark channel dehazed RGB image I_HazeFreeGray scale image average brightness of, L_hazefreemean' representing RGB image I after contrast-equalized dark channel defogging_HazeFree' average luminance L of gray-scale image_hazefreemean′，I_HazeFreeRepresenting RGB image after dark channel defogging, I_HazeFree' represents the RGB image after defogging of the contrast-equalized dark channel;

the brightness compensation operation for the RGB image after the defogging of the dark channel,

representing contrast-equalized dark channel dehazed RGB image I_HazeFree' a fusion weight; divide by 8.0 is the regularization operation.