CN110349113A - Self-adaptive image defogging method based on dark channel prior improvement - Google Patents
Self-adaptive image defogging method based on dark channel prior improvement Download PDFInfo
- Publication number
- CN110349113A CN110349113A CN201910177949.1A CN201910177949A CN110349113A CN 110349113 A CN110349113 A CN 110349113A CN 201910177949 A CN201910177949 A CN 201910177949A CN 110349113 A CN110349113 A CN 110349113A
- Authority
- CN
- China
- Prior art keywords
- image
- defogging
- value
- dark
- formula
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Granted
Links
- 238000000034 method Methods 0.000 title claims abstract description 168
- 238000012545 processing Methods 0.000 claims abstract description 18
- 238000004364 calculation method Methods 0.000 claims abstract description 16
- 230000015556 catabolic process Effects 0.000 claims abstract description 6
- 238000006731 degradation reaction Methods 0.000 claims abstract description 6
- 230000003044 adaptive effect Effects 0.000 claims description 33
- 230000033228 biological regulation Effects 0.000 claims description 20
- 230000011218 segmentation Effects 0.000 claims description 14
- 239000003595 mist Substances 0.000 claims description 13
- 238000012216 screening Methods 0.000 claims description 13
- 238000004422 calculation algorithm Methods 0.000 claims description 12
- 230000002708 enhancing effect Effects 0.000 claims description 10
- 238000009826 distribution Methods 0.000 claims description 9
- 238000003384 imaging method Methods 0.000 claims description 9
- 230000008859 change Effects 0.000 claims description 7
- 238000013507 mapping Methods 0.000 claims description 6
- 238000006467 substitution reaction Methods 0.000 claims description 6
- 238000004458 analytical method Methods 0.000 claims description 4
- 239000012141 concentrate Substances 0.000 claims description 4
- 238000012935 Averaging Methods 0.000 claims description 3
- 238000000149 argon plasma sintering Methods 0.000 claims description 3
- 239000003086 colorant Substances 0.000 claims description 3
- 238000006073 displacement reaction Methods 0.000 claims description 3
- 238000003709 image segmentation Methods 0.000 claims description 3
- 239000004615 ingredient Substances 0.000 claims description 3
- 238000010200 validation analysis Methods 0.000 claims description 3
- 238000012795 verification Methods 0.000 claims description 3
- 125000001475 halogen functional group Chemical group 0.000 abstract 1
- 230000000694 effects Effects 0.000 description 19
- 238000010586 diagram Methods 0.000 description 8
- 230000008030 elimination Effects 0.000 description 3
- 238000003379 elimination reaction Methods 0.000 description 3
- 238000005516 engineering process Methods 0.000 description 3
- 238000011156 evaluation Methods 0.000 description 3
- 230000007246 mechanism Effects 0.000 description 3
- 238000011084 recovery Methods 0.000 description 3
- 230000005540 biological transmission Effects 0.000 description 2
- 230000007850 degeneration Effects 0.000 description 2
- 239000000203 mixture Substances 0.000 description 2
- 230000008901 benefit Effects 0.000 description 1
- 230000000903 blocking effect Effects 0.000 description 1
- 230000007423 decrease Effects 0.000 description 1
- 230000006866 deterioration Effects 0.000 description 1
- 238000011161 development Methods 0.000 description 1
- 230000018109 developmental process Effects 0.000 description 1
- 238000002474 experimental method Methods 0.000 description 1
- 238000003707 image sharpening Methods 0.000 description 1
- 238000004519 manufacturing process Methods 0.000 description 1
- 238000005259 measurement Methods 0.000 description 1
- 238000012986 modification Methods 0.000 description 1
- 230000004048 modification Effects 0.000 description 1
- 238000012544 monitoring process Methods 0.000 description 1
- 238000005457 optimization Methods 0.000 description 1
- 230000008635 plant growth Effects 0.000 description 1
- 230000008569 process Effects 0.000 description 1
- 230000000750 progressive effect Effects 0.000 description 1
- 230000001737 promoting effect Effects 0.000 description 1
- 238000004445 quantitative analysis Methods 0.000 description 1
- 238000002310 reflectometry Methods 0.000 description 1
- 238000007619 statistical method Methods 0.000 description 1
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T5/00—Image enhancement or restoration
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T5/00—Image enhancement or restoration
- G06T5/40—Image enhancement or restoration using histogram techniques
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T7/00—Image analysis
- G06T7/10—Segmentation; Edge detection
- G06T7/11—Region-based segmentation
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T7/00—Image analysis
- G06T7/10—Segmentation; Edge detection
- G06T7/136—Segmentation; Edge detection involving thresholding
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T7/00—Image analysis
- G06T7/90—Determination of colour characteristics
Landscapes
- Engineering & Computer Science (AREA)
- Physics & Mathematics (AREA)
- General Physics & Mathematics (AREA)
- Theoretical Computer Science (AREA)
- Computer Vision & Pattern Recognition (AREA)
- Image Processing (AREA)
- Facsimile Image Signal Circuits (AREA)
- Image Analysis (AREA)
Abstract
The invention discloses a self-adaptive image defogging method based on dark channel prior improvement, which comprises a self-adaptive image defogging method based on a haze image degradation model and a dark channel prior model; the self-adaptive image defogging method comprises a dark primary color value self-adaptive acquisition method for realizing parameter acquisition or data processing in a self-adaptive mode, an atmospheric light intensity self-adaptive estimation method, a defogging coefficient self-adaptive calculation method and an image color level self-adaptive adjustment method; according to the self-adaptive image defogging method based on the dark channel prior improvement, the defogging parameters are completely obtained according to the characteristics of the haze image, manual parameter setting is not needed, the robustness in the defogging aspect of the haze image with multiple concentrations and multiple scenes is high, the obtained defogging image is large in information amount, high in contrast, clear in image, natural in color and free of halo phenomenon, and the method is obviously superior to a multiscale Retinex enhanced defogging method and a He dark channel prior restoration defogging method.
Description
Technical field
The present invention relates to one kind to be based on the improved adapting to image defogging method of dark primary priori, belongs to the processing of image defogging
Method and technology field.
Background technique
With the rapid development of society being constantly progressive with science and technology, computer vision system has been widely applied to city
The numerous areas such as city's traffic, video monitoring, intelligent navigation, remotely sensed image, military surveillance;However, in recent years, the frequency of haze weather
Numerous appearance has seriously affected the image quality of computer vision system, cause obtain picture contrast decline, cross-color,
It is smudgy, extreme influence is caused with life to people's production;Therefore, image defogging technology is studied, reduces haze to image matter
The influence of amount has important practical value and realistic meaning for promoting computer vision performance.
According to the difference of image defogging mechanism, conventional images defogging method is broadly divided into the image enhancement based on image procossing
Method and image recovery method based on physical model;First kind method directly utilizes image procossing independent of physical model
Algorithm promotes picture contrast, and typical method has: histogram equalization method, Retinex method etc., such method does not consider haze
Image deterioration reason, it is difficult to improve haze image quality from mechanism;Second class method generates model by building haze image,
Inverting haze image degrades process, finally restores fog free images out, typical method has: partial differential equation method, depth fogging method and
Image restoration method based on prior information;In comparison, partial differential equation method and picture depth fogging method usually assume that haze takes
From model is uniformly distributed, haze image is restored using whole unified method, and practical haze is distributed and uneven, causes image
Sharpening treatment effect it is bad, be lost some detailed information;Image restoration method based on prior information is established to reality
On the basis of haze image is for statistical analysis, better haze image recovery effect can be obtained.
By having carried out a large amount of statistical experiments to fog free images and foggy image, discovery fog free images contrast is apparently higher than
Foggy image proposes the defogging method for maximizing local contrast on this basis, and this method passes through local contrast maximum
Change processing achievees the purpose that defogging, but there are big guns and distortion phenomenon for the color of image after handling;In addition, utilizing the propagation of light
With the incoherent priori knowledge of local surfaces reflector segment of scene, scene reflectivity rate is estimated, obtain fog free images, but this method
It is preferable to mist treatment effect premised on obtaining bright-coloured image color, it is difficult to the image restoration for thick fog weather;For
Above-mentioned problem proposes a kind of image defogging method based on dark channel prior theory, using Steerable filter in the prior art
Carry out transmitance optimization instead of soft pick figure, improve defogging processing speed, but after handling image bright area color it is unnatural,
Blocking artifact is obvious;Air light value and transmissivity are estimated using the method that bright, dark combines later, is solved to a certain extent
The cross-color problem of bright area of having determined defogging, but not yet consider the adaptive defogging problem of various concentration haze image;In addition,
It in existing primary colors priori and its improves in defogging algorithm, goes fog coefficient ω mostly to use fixed value for adjust defog effect,
It is less to do adaptive adjustment according to the actual situation;Therefore, in order to solve existing algorithm be difficult to it is pervasive different in concentration, bright dark unknown
More scene haze image defoggings the problems such as, the present invention proposes that a kind of adaptive defogging is calculated on the basis of dark primary priori theoretical
Method, the algorithm defogging parameter are obtained according to haze image unique characteristics completely, are not necessarily to artificial setting parameter, and in more concentration, more
Robustness with higher in terms of scene haze image defogging.
Summary of the invention
To solve the above problems, the invention proposes one kind to be based on the improved adapting to image defogging side of dark primary priori
Method, it is not necessary that defogging parameter is manually set, image clearly, the color of acquisition are naturally, in more concentration, more scene haze images defogging side
Face has better robustness.
It is of the invention based on the improved adapting to image defogging method of dark primary priori, including be based on haze image degeneration mould
The adapting to image defogging method of type and dark primary prior model;The adapting to image defogging method is by using adaptive mode
Realize dark primary value adaptive acquiring method, the atmosphere light intensity adaptive estimation method, defogging system of parameter acquisition or data processing
Number self-adaptive computing method and image tonescale self-adapting regulation method composition;
The dark primary value adaptive acquiring method is to be carried out certainly using quick OSTU algorithm to the bright dark areas of haze image
Segmentation is adapted to, and subregion obtains the dark primary value of bright dark areas;
The atmosphere light intensity adaptive estimation method is to be carried out according to bright area distribution situation to bright dark areas atmosphere light intensity
ART network;
It is described to remove fog coefficient self-adaptive computing method to propose to use gray scale collection by counting haze image histogram feature
Moderate method adaptive polo placement goes fog coefficient;
Described image color range self-adapting regulation method is the color that output image is carried out with color range self-adapting regulation method
Adjustment.
Further, described to be based on haze image degradation model
I (x)=J (x) t (x)+A [1-t (x)], (1)
In formula: I (x) is the haze image intensity that imaging device observes;J (x) is fog free images intensity, i.e. parked figure
Picture;T (x) is transmitance, and reflection light penetrates the ability of haze;A is the atmosphere light intensity of infinite point;J (x) t (x) corresponds to mesh
The part that reflected light enters imaging device after atmospheric scattering is decayed is marked, which exponentially decays with scene depth increase;A
Enter the part of imaging device after [1-t (x)] corresponding atmosphere light scattering, which increases as scene depth increases, can make
It is distorted at Scene Blur and color displacement;
Formula (1) is transplanted and arranged, fog free images intensity is expressed as
It is above it is various in, only I (x) is known terms, and the essence of image defogging is exactly according to above-mentioned model, by transmission
Rate t (x) and atmosphere light intensity A are estimated, clear image J (x) out is restored.
Further, the dark primary prior model is the dark primary that defogging method statistics is restored according to He dark primary priori
Priori rule, performance are exactly to have the presence of dark pixel point in each position of fog free images on the image, these dark pixel points
It is exactly dark primary Jdark, these dark primary point gray values tend to 0, meet
In formula: the RGB channel of subscript c expression image;Y ∈ Ω (x) indicates a window centered on pixel x;
Given that it is known that transmitance t (x) is constant in atmosphere light intensity A, window Ω (x), formula (1) is arranged and is done twice most
Small operation can obtain:
Simultaneous formula (3) and (4), the estimated value that can obtain transmitance are
Consider that fog coefficient ω (0 ω≤1 <) is gone in the dim sense of vision of people's observation distant place scenery in real life, introducing,
Then above formula is modified to
In formula: ω value is smaller, and mist ingredient is more in restored image, and defogging ability is weaker, but be worth it is too small, level off to 0 against
The purpose of defogging;For ω value closer to 1, the ability of defogging is stronger, but color of image can be made supersaturated, lacks the sense of reality;
The estimation method of atmosphere light intensity A is: first obtain image dark primary in the maximum pixel of 0.1% brightness, then these
Pixel corresponds to estimated value of the maximum value of pixel in original image as A;
Atmosphere light intensity A in the transmitance t (x) of estimation and formula (2) can be obtained into fog free images, actually in calculating,
Image fault is caused in order to avoid transmitance is too small, transmitance lower limit t is usually set0, then fog free images be
Further, the dark primary value adaptive acquiring method itself specific steps are as follows: to image pixel carry out
Category division divides obtained the distance between all kinds of maximums by making, determines optimal segmenting threshold;
The first step, it is assumed that image I has L gray level, and i-stage number of pixels is ni, total pixel number isThen
The probability that i grades of pixels occur is pi=ni/N;
Second step divides the image into two class of A and B for the threshold value k of setting, and wherein A class gray scale interval is [0, k], B class
Gray scale interval is [k+1, L-1], then the class interval whole image gray average μ, A gray average μA(k), the class interval B gray average μB
(k) it is respectively as follows:
The probability distribution of two class gray scale intervals meets:
Define inter-class variance when threshold value k is are as follows:
σ2(k)=pA[μ-μA(k)]2+pB[μ-μB(k)]2, (10)
According to maximum between-cluster variance criterion, from 0 to L-1 threshold value k is changed, finding out makes the maximum k value of variance in above formula, i.e.,
For optimal segmenting threshold T;Change a subthreshold k due to every, it is above-mentioned it is various will all be recalculated, to reduce operation time, first
According to haze image histogram low ebb gray feature, several threshold points are filtered out, then substitution is above various, realizes quick OSTU
Threshold segmentation;
Third step, according to gray level i and its pixel number niL relationship is established to [i, ni], it is assumed that curve [i, ni] on it is a certain
Gray level is T at positionj, as its pixel number njWhen meeting following relationship simultaneously, by gray level TjScreening is threshold point:
In formula: Δ is screening window size, takes Δ=3pixels;α is that minimum value limits ratio, takes α=0.1%;
4th step carries out Threshold segmentation to haze image using improved quick OSTU threshold method, is sieved according to its threshold point
It can be seen that threshold point only remains 49 after screening, participating in the threshold value calculated points reduces for choosing and bright dark areas segmentation result
80% or more;Meanwhile bright dark areas by effectively it is separated, facilitate subregion defogging processing;
5th step, according to formula (3), the RGB triple channel minimum value of each pixel is dark primary value, it may be assumed that
If the haze image of various concentration is all made of above formula and solves dark primary value, the mistake of bright area color may result in
Very;This is because the minimum gradation value of bright area is higher with respect to dark areas, if directlying adopt above formula minimum value method seeks dark primary value,
Transmitance can then calculated inaccurate, and triple channel gray average is selected then more to reflect practical feelings as the dark primary value of bright area
Condition;For this purpose, correcting to dark primary minimum value, dark primary value is solved using following formula:
In formula: TcFor c channel segmentation threshold value, c ∈ { R, G, B }.
Further, the atmosphere light intensity adaptive estimation method itself specific steps are as follows: using bright dark subregion
Averaging method carries out atmosphere light intensity ART network, as bright area pixel number accounting Pb cWhen < 10%, illustrate that bright area is less, this
When first take 0.1% bright spot before dark primary image, then find out the gray average conduct of these bright spots corresponding pixel points in original image
Atmosphere light intensity estimated value;As bright area pixel number accounting Pb cWhen >=10%, illustrate that bright area is more, at this time with all bright area pictures
For the gray average of vegetarian refreshments as atmosphere light intensity estimated value, calculation formula is as follows:
In formula: NbFor bright area pixel number.
Further, it is described go fog coefficient self-adaptive computing method itself specific steps are as follows: adaptive using color range
Method of adjustment is to restored image J0Enhancing processing is carried out, by carrying out statistics with histogram to various concentration haze image, finds mist
Haze concentration is bigger, and histogram distribution is more concentrated, and it is [m that most gray values, which concentrate on size,c,Mc] gray scale interval in, utilize
The property proposes a kind of defogging coefficient calculation method based on gray scale concentration degree, meets:
In formula: ωcFor minimum value regulation coefficient, value ωc=0.15;Bound mcAnd McMeet:
In formula: α is bound regulation coefficient, value α=1%;
By the above various corresponding substitution transmitance formula (6), the transmitance of the bright dark areas of haze image can be obtained;It ties again
It closes formula (7), it will be able to the preliminary image J obtained after restoring0。
Further, described image color range self-adapting regulation method itself specific steps are as follows: according to formula (16),
Seek the concentration degree section [m of haze image RGB triple channelc,Mc], mapping table is calculated using following formula:
The mapping table generated using formula (17), to preliminary restored image J0Color range adjustment is carried out, as follows:
J (x)=Map (J0(x)+1)。 (18)
Further, further include validation verification method to adapting to image defogging method, specifically include following step
Rapid: firstly, choosing, forest, river bank, traffic, the city proper and the suburbs isoconcentration of the bright area of sky containing large area be different, bright area face
6 different width haze images of product are experimental subjects, are utilized respectively multiple dimensioned Retinex enhancing defogging method, He dark primary priori
It restores defogging method and adapting to image defogging method and carries out subjective and objective analysis to progress defogging processing, and to experimental result.
The present invention is compared with prior art, of the invention based on the improved adapting to image defogging side of dark primary priori
Method is split the bright dark areas of haze image by quick OSTU method, and subregion obtains bright dark areas dark primary value, and to not
ART network is carried out with region atmosphere light intensity values;According to haze image histogram feature, propose using gray scale concentration degree method come
Fog coefficient is gone in calculating;Enhancing processing is done to image after preliminary restore using color range self-adapting regulation method, to restore mist as far as possible
Original color of haze image;Adapting to image defogging method defogging parameter of the invention is obtained according to haze image unique characteristics completely
It takes, is not necessarily to artificial setting parameter, and the robustness with higher in terms of more concentration, more scene haze image defoggings, obtains
Mist elimination image contain much information, contrast is high, and image clearly, color is naturally, without halation phenomenon, hence it is evident that better than multiple dimensioned
Retinex enhances defogging method and He dark primary priori restores defogging method.
Detailed description of the invention
Fig. 1 is the atmospherical scattering model structural schematic diagram that McCartney is proposed.
Fig. 2 is the flow chart of adapting to image defogging method of the invention.
Fig. 3 is real scene shooting haze image in campus of the invention.
Fig. 4 is threshold point screening of the invention and bright dark areas segmentation result schematic diagram;
Wherein, figure (a) is threshold point screening schematic diagram of the invention, and figure (b) is bright dark areas segmentation result of the invention
Schematic diagram.
Fig. 5 is the defog effect schematic diagram of distinct methods.
Fig. 6 is defog effect objective indicator evaluation situation schematic diagram.
Specific embodiment
As shown in Figure 2 is moved back based on the improved adapting to image defogging method of dark primary priori, including based on haze image
Change the adapting to image defogging method of model and dark primary prior model;The adapting to image defogging method is by using adaptive
Mode realizes that parameter obtains or the dark primary value of data processing adaptively obtains (obtain dark channel image
Adaptively) method, atmosphere light intensity ART network (estimate air light adaptively) method, defogging system
Number adaptive polo placement (calculate dehazing coefficient adaptively) method and image tonescale are adaptively adjusted
Whole (adjust image color level adaptively) method composition;
The dark primary value adaptive acquiring method is to be carried out certainly using quick OSTU algorithm to the bright dark areas of haze image
Segmentation is adapted to, and subregion obtains the dark primary value of bright dark areas;
The atmosphere light intensity adaptive estimation method is to be carried out according to bright area distribution situation to bright dark areas atmosphere light intensity
ART network;
It is described to remove fog coefficient self-adaptive computing method to propose to use gray scale collection by counting haze image histogram feature
Moderate method adaptive polo placement goes fog coefficient;
Described image color range self-adapting regulation method is the color that output image is carried out with color range self-adapting regulation method
Adjustment.
Wherein, haze image degradation model can preferably disclose the degradation mechanism of haze image, according to the model, at
As equipment obtain target image mainly by from target reflecting light attenuation components and ambient diffusing light component two parts constitute,
As shown in Figure 1,
McCartney is simplified, the simplified model for having obtained the degeneration of haze image is
I (x)=J (x) t (x)+A [1-t (x)], (1)
In formula: I (x) is the haze image intensity that imaging device observes;J (x) is fog free images intensity, i.e. parked figure
Picture;T (x) is transmitance, and reflection light penetrates the ability of haze;A is the atmosphere light intensity of infinite point;As shown in Figure 1, J (x) t
(x) target reflecting light in corresponding diagram 1 enters the part of imaging device after atmospheric scattering is decayed, and the part is with scene depth
Increase exponentially decays;In A [1-t (x)] corresponding diagram 1 atmosphere light scattering after enter imaging device part, the part with
Scene depth increases and increases, and will cause Scene Blur and color displacement distortion;
Formula (1) is transplanted and arranged, fog free images intensity is represented by
It is above it is various in, only I (x) is known terms, and the essence of image defogging is exactly according to above-mentioned model, by transmission
Rate t (x) and atmosphere light intensity A are estimated, clear image J (x) out is restored;
The dark primary priori rule of defogging method statistics is restored according to He dark primary priori, shade everywhere may be used in natural scene
See, performance is exactly to have the presence of dark pixel point in each position of fog free images on the image, these dark pixel points are exactly dark
Primary colors Jdark, these dark primary point gray values tend to 0, meet
In formula: the RGB channel of subscript c expression image;Y ∈ Ω (x) indicates a window centered on pixel x;
Given that it is known that transmitance t (x) is constant in atmosphere light intensity A, window Ω (x), formula (1) is arranged and is done twice most
Small operation can obtain:
Simultaneous formula (3) and (4), the estimated value that can obtain transmitance are
Consider that fog coefficient ω (0 ω≤1 <) is gone in the dim sense of vision of people's observation distant place scenery in real life, introducing,
Then above formula is modified to
In formula: ω value is smaller, and mist ingredient is more in restored image, and defogging ability is weaker, but value too small (level off to 0) against
The purpose of defogging;For ω value closer to 1, the ability of defogging is stronger, but color of image can be made supersaturated, lacks the sense of reality;
The estimation method of atmosphere light intensity A is: first obtain image dark primary in the maximum pixel of 0.1% brightness, then these
Pixel corresponds to estimated value of the maximum value of pixel in original image as A;
The transmitance t (x) of estimation and atmosphere light intensity A formula (2) can be obtained into fog free images, actually in calculating, in order to
It avoids transmitance is too small from causing image fault, transmitance lower limit t is usually set0, then fog free images be
The dark primary value adaptive acquiring method itself specific steps are as follows: to image pixel carry out category division,
By making to divide obtained the distance between all kinds of maximums, optimal segmenting threshold is determined;
The first step, it is assumed that image I has L gray level, and i-stage number of pixels is ni, total pixel number isThen
The probability that i-stage pixel occurs is pi=ni/N;
Second step divides the image into two class of A and B for the threshold value k of setting, and wherein A class gray scale interval is [0, k], B class
Gray scale interval is [k+1, L-1], then the class interval whole image gray average μ, A gray average μA(k), the class interval B gray average μB
(k) it is respectively as follows:
The probability distribution of two class gray scale intervals meets:
Define inter-class variance when threshold value k is are as follows:
σ2(k)=pA[μ-μA(k)]2+pB[μ-μB(k)]2, (10)
According to maximum between-cluster variance criterion, from 0 to L-1 threshold value k is changed, finding out makes the maximum k value of variance in above formula, i.e.,
For optimal segmenting threshold T;Change a subthreshold k due to every, it is above-mentioned it is various will all be recalculated, to reduce operation time, first
According to haze image histogram low ebb gray feature, several threshold points are filtered out, then substitution is above various, realizes quick OSTU
Threshold segmentation;
Third step, according to gray level i and its pixel number niL relationship is established to [i, ni], it is assumed that curve [i, ni] on it is a certain
Gray level is T at positionj, as its pixel number njWhen meeting following relationship simultaneously, by gray level TjScreening is threshold point:
In formula: Δ is screening window size, takes Δ=3pixels;α is that minimum value limits ratio, takes α=0.1%;
4th step carries out Threshold segmentation to haze image as shown in Figure 3 using improved quick OSTU threshold method, according to
The screening of its threshold point and bright dark areas segmentation result are as shown in Figure 4, it will thus be seen that threshold point only remains 49 after screening, participates in meter
The threshold value points of calculation reduce 80% or more;Meanwhile bright dark areas by effectively it is separated, facilitate subregion defogging processing;
5th step, according to formula (3), the RGB triple channel minimum value of each pixel is dark primary value, it may be assumed that
If the haze image of various concentration is all made of above formula and solves dark primary value, the mistake of bright area color may result in
Very;This is because the minimum gradation value of bright area is higher with respect to dark areas, if directlying adopt above formula minimum value method seeks dark primary value,
Transmitance can then calculated inaccurate, and triple channel gray average is selected then more to reflect practical feelings as the dark primary value of bright area
Condition;For this purpose, correcting to dark primary minimum value, dark primary value is solved using following formula:
In formula: TcFor c channel segmentation threshold value, c ∈ { R, G, B }.
The atmosphere light intensity adaptive estimation method itself specific steps are as follows: using bright dark subregion averaging method come into
Row atmosphere light intensity ART network, as bright area pixel number accounting Pb cWhen < 10%, illustrate that bright area is less, first takes dark original at this time
0.1% bright spot before chromatic graph picture, the gray average for then finding out these bright spots corresponding pixel points in original image are estimated as atmosphere light intensity
Evaluation;As bright area pixel number accounting Pb cWhen >=10%, illustrate that bright area is more, at this time with the gray scale of all bright area pixels
For mean value as atmosphere light intensity estimated value, calculation formula is as follows:
In formula: NbFor bright area pixel number.
It is described go fog coefficient self-adaptive computing method itself specific steps are as follows: using color range self-adapting regulation method pair
Restored image J0Enhancing processing is carried out, by carrying out statistics with histogram to various concentration haze image, discovery haze concentration is bigger,
Histogram distribution is more concentrated, and it is [m that most gray values, which concentrate on size,c,Mc] gray scale interval in, utilize the property, propose
A kind of defogging coefficient calculation method based on gray scale concentration degree meets:
In formula: ωcFor minimum value regulation coefficient, value ωc=0.15;Bound mcAnd McMeet:
In formula: α is bound regulation coefficient, value α=1%;
By the above various corresponding substitution transmitance formula (6), the transmitance of the bright dark areas of haze image can be obtained;It ties again
It closes formula (7), it will be able to the preliminary image J obtained after restoring0。
Described image color range self-adapting regulation method itself specific steps are as follows: according to formula (16), seek haze figure
As the concentration degree section [m of RGB triple channelc,Mc], mapping table is calculated using following formula:
The mapping table generated using formula (17), to preliminary restored image J0Color range adjustment is carried out, as follows:
J (x)=Map (J0(x)+1)。 (18)
It further include the validation verification method to adapting to image defogging method, specifically includes the following steps: firstly, choosing
Take forest, river bank, traffic, 6 width that the city proper and the suburbs isoconcentration of the bright area of sky containing large area is different, bright area area is different
Haze image is experimental subjects, is utilized respectively multiple dimensioned Retinex enhancing defogging method, He dark primary priori restores defogging method
Subjective and objective analysis, experimental result such as Fig. 5 institute are carried out to progress defogging processing, and to experimental result with adapting to image defogging method
Show, the subjective vision effect of image is analyzed after defogging:
(1) all in all, when not considering sky bright area in image, three kinds of algorithms can effective defogging, improve
Picture quality;But in bright area, bright dark juncture area, multiple dimensioned Retinex enhancing defogging method and He dark primary priori are restored
There are color distortions and halation phenomenon for defogging method, and haze is more serious, and this phenomenon is more obvious;
(2) comparing adapting to image defogging method of the invention and multiple dimensioned Retinex enhances defogging method effect, due to
Retinex algorithm belongs to algorithm for image enhancement, has a distinct increment to picture contrast, and image integrally seems unnatural, exists
Biggish misalignment, and with the increase of haze concentration, defog effect is poorer.
(3) adapting to image defogging method and He dark primary priori of the invention are compared and restores defogging method effect, is being schemed
When sky bright area is few as in, haze is smaller, two kinds of algorithm effects are suitable or even He dark primary priori is restored defogging method and wanted
Slightly it is better than adapting to image defogging method of the invention, such as the defog effect of the first width forest mist scene in Fig. 5;But with
The increase of bright area increased with haze concentration in image, adapting to image defogging method advantage of the invention is gradually obvious, He
Dark primary priori restores defogging method, and bright area has biggish texture and blockiness on high, such as the 5th width and the in Fig. 5
The defog effect of the city proper and the suburbs scene under six width severe hazes;
For objective analysis defog effect, the present invention uses comentropy (Information entropy), mean square deviation
(MSE), Laplace operator (Laplacian) is used as index, carries out quantitative analysis to defogging result;Wherein, comentropy H is pair
The measurement of amount of image information, value is bigger, illustrates that information content is more, definition are as follows:
Meansquaredeviationσ reflects picture contrast, and value is bigger, illustrates that black and white contrast is more obvious, definition are as follows:
In formula: N1And N2Respectively picturedeep and columns, N=N1×N2;μ is image grayscale mean value;
Laplace operator LS is to reflect neighborhood of pixels grey scale change, and value is bigger, and image is more clear, and image outline is got over
Distinctness, definition are as follows:
Defog effect objectively evaluates situation as shown in fig. 6, being analyzed as follows:
All in all, the comentropy, mean square deviation, Laplace operator of adapting to image defogging method of the invention are most of
All enhance defogging method and He dark primary priori than multiple dimensioned Retinex and restore that defogging method is high, illustrates of the invention adaptive
The mist elimination image for answering image defogging method to obtain has more information content, and image is more clear;
Comparing adapting to image defogging method of the invention and multiple dimensioned Retinex enhances defogging method effect, in 6 width
It being evaluated in image, adapting to image defogging method of the invention is suitable with multiple dimensioned Retinex enhancing defogging method effect,
In there are 5 information entropy, 5 mean square deviations and 3 Laplace operator value outlines to be higher than multiple dimensioned Retinex to enhance defogging side
Method illustrates that adapting to image defogging method of the invention is better than multiple dimensioned Retinex enhancing defogging method;
It compares adapting to image defogging method and He dark primary priori of the invention and restores defogging method effect, except for gloomy
Both dense growth of plants and trees mist images Laplace operator value is quite outer, and other indexs of adapting to image defogging method of the invention are above
He dark primary priori restores defogging method, illustrates the validity of adaptive algorithm of the present invention;
Comprehensive subjective evaluation effect, multiple dimensioned Retinex enhancing defogging method and He dark primary priori restore defogging side
The various defogging parameters of method are both needed to subjective determination, and many kinds of parameters is all made of adaptively in adapting to image defogging method of the invention
Mode obtains, and is influenced by the bright dark areas area of image and haze concentration, therefore adapting to image defogging side of the invention
Method has better robustness in terms of more concentration, more scene haze image defoggings.
It is of the invention based on the improved adapting to image defogging method of dark primary priori, by quick OSTU method to haze figure
As bright dark areas is split, subregion obtains bright dark areas dark primary value, and carries out to different zones atmosphere light intensity value adaptive
It should estimate;According to haze image histogram feature, propose to calculate fog coefficient using gray scale concentration degree method;It is adaptive using color range
Method of adjustment is answered to do enhancing processing to image after preliminary restore, to restore original color of haze image as far as possible;It is subjective and objective to comment
Valence the result shows that: the mist elimination image that adapting to image defogging method of the invention obtains contains much information, and contrast is high, and image is clear
Clear, color is naturally, without halation phenomenon, hence it is evident that enhances defogging method better than multiple dimensioned Retinex and the recovery of He dark primary priori is gone
Mist method.
Above-described embodiment is only better embodiment of the invention, therefore all according to structure described in present patent application range
It makes, the equivalent change or modification that feature and principle are done, is included in the scope of the patent application of the present invention.
Claims (8)
1. one kind is based on the improved adapting to image defogging method of dark primary priori, including based on haze image degradation model and secretly
The adapting to image defogging method of primary colors prior model;It is characterized by: the adapting to image defogging method is by using adaptive
Answer mode realize parameter obtain or the dark primary value adaptive acquiring method of data processing, atmosphere light intensity adaptive estimation method,
Fog coefficient self-adaptive computing method and image tonescale self-adapting regulation method is gone to form;
The dark primary value adaptive acquiring method is to be carried out adaptively using quick OSTU algorithm to the bright dark areas of haze image
Segmentation, and subregion obtains the dark primary value of bright dark areas;
The atmosphere light intensity adaptive estimation method is to be carried out to bright dark areas atmosphere light intensity adaptive according to bright area distribution situation
It should estimate;
It is described to remove fog coefficient self-adaptive computing method to propose to use gray scale concentration degree by counting haze image histogram feature
Method adaptive polo placement goes fog coefficient;
Described image color range self-adapting regulation method is the color adjustment that output image is carried out with color range self-adapting regulation method.
2. according to claim 1 be based on the improved adapting to image defogging method of dark primary priori, it is characterised in that: institute
It states and is based on haze image degradation model
I (x)=J (x) t (x)+A [1-t (x)], (1)
In formula: I (x) is the haze image intensity that imaging device observes;J (x) is fog free images intensity, i.e. parked image;t
It (x) is transmitance, reflection light penetrates the ability of haze;A is the atmosphere light intensity of infinite point;The corresponding target reflection of J (x) t (x)
Light enters the part of imaging device after atmospheric scattering is decayed, which exponentially decays with scene depth increase;A[1-t
(x)] enter the part of imaging device after corresponding atmosphere light scattering, which increases as scene depth increases, and will cause field
Scape is fuzzy and color displacement is distorted;
Formula (1) is transplanted and arranged, fog free images intensity is expressed as
It is above it is various in, only I (x) is known terms, and the essence of image defogging is exactly according to above-mentioned model, by transmitance t
(x) estimated with atmosphere light intensity A, restore clear image J (x) out.
3. according to claim 1 be based on the improved adapting to image defogging method of dark primary priori, it is characterised in that: institute
Stating dark primary prior model is the dark primary priori rule that defogging method statistics is restored according to He dark primary priori, shows image
Upper is exactly to have the presence of dark pixel point in each position of fog free images, these dark pixel points are exactly dark primary Jdark, these
Dark primary point gray value tends to 0, meets
In formula: the RGB channel of subscript c expression image;Y ∈ Ω (x) indicates a window centered on pixel x;
Given that it is known that transmitance t (x) is constant in atmosphere light intensity A, window Ω (x), formula (1) is arranged and does minimum fortune twice
It calculates, can obtain:
Simultaneous formula (3) and (4), the estimated value that can obtain transmitance are
Consider real life in people observe distant place scenery the dim sense of vision, introducing go to fog coefficient ω (0 ω≤1 <), then on
Formula is modified to
In formula: ω value is smaller, and mist ingredient is more in restored image, and defogging ability is weaker, but value is too small, levels off to 0 against defogging
Purpose;For ω value closer to 1, the ability of defogging is stronger, but color of image can be made supersaturated, lacks the sense of reality;
The estimation method of atmosphere light intensity A is: first obtaining the maximum pixel of 0.1% brightness in image dark primary, then these pixels
Estimated value of the maximum value as A corresponding to pixel in original image;
Atmosphere light intensity A in the transmitance t (x) of estimation and formula (2) can be obtained into fog free images, actually in calculating, in order to
It avoids transmitance is too small from causing image fault, transmitance lower limit t is usually set0, then fog free images be
4. according to claim 1 be based on the improved adapting to image defogging method of dark primary priori, it is characterised in that: institute
State dark primary value adaptive acquiring method itself specific steps are as follows: to image pixel carry out category division, divided by making
Obtained the distance between all kinds of maximums, determine optimal segmenting threshold;
The first step, it is assumed that image I has L gray level, and i-stage number of pixels is ni, total pixel number isThen i-stage
The probability that pixel occurs is pi=ni/N;
Second step divides the image into two class of A and B for the threshold value k of setting, and wherein A class gray scale interval is [0, k], B class gray scale
Section is [k+1, L-1], then the class interval whole image gray average μ, A gray average μA(k), the class interval B gray average μB(k)
It is respectively as follows:
The probability distribution of two class gray scale intervals meets:
Define inter-class variance when threshold value k is are as follows:
σ2(k)=pA[μ-μA(k)]2+pB[μ-μB(k)]2, (10)
According to maximum between-cluster variance criterion, from 0 to L-1 threshold value k is changed, finding out makes the maximum k value of variance in above formula, as most preferably
Segmentation threshold T;Change a subthreshold k due to every, it is above-mentioned it is various will all be recalculated, to reduce operation time, first according to mist
Haze image histogram low ebb gray feature, filters out several threshold points, and then substitution is above various, realizes quick OSTU threshold value point
It cuts;
Third step, according to gray level i and its pixel number niL relationship is established to [i, ni], it is assumed that curve [i, ni] on a certain position
Place's gray level is Tj, as its pixel number njWhen meeting following relationship simultaneously, by gray level TjScreening is threshold point:
In formula: Δ is screening window size, takes Δ=3pixels;α is that minimum value limits ratio, takes α=0.1%;
4th step, using improved quick OSTU threshold method to haze image carry out Threshold segmentation, according to its threshold point screening and
Bright dark areas segmentation result it can be seen that screening after threshold point only remain 49, participate in calculate threshold value points reduce 80% with
On;Meanwhile bright dark areas by effectively it is separated, facilitate subregion defogging processing;
5th step, according to formula (3), the RGB triple channel minimum value of each pixel is dark primary value, it may be assumed that
If the haze image of various concentration is all made of above formula and solves dark primary value, bright area color distortion may result in;
This is because the minimum gradation value of bright area is higher with respect to dark areas, if directlying adopt above formula minimum value method seeks dark primary value,
Transmitance can be made to calculate inaccurate, and triple channel gray average is selected then more to reflect practical feelings as the dark primary value of bright area
Condition;For this purpose, correcting to dark primary minimum value, dark primary value is solved using following formula:
In formula: TcFor c channel segmentation threshold value, c ∈ { R, G, B }.
5. according to claim 1 be based on the improved adapting to image defogging method of dark primary priori, it is characterised in that: institute
State atmosphere light intensity adaptive estimation method itself specific steps are as follows: carrying out atmosphere light intensity using bright dark subregion averaging method
ART network, as bright area pixel number accounting Pb cWhen < 10%, illustrate that bright area is less, before first taking dark primary image at this time
Then 0.1% bright spot finds out the gray average of these bright spots corresponding pixel points in original image as atmosphere light intensity estimated value;When bright
Area pixel number accounting Pb cWhen >=10%, illustrate that bright area is more, at this time using the gray average of all bright area pixels as
Atmosphere light intensity estimated value, calculation formula are as follows:
In formula: NbFor bright area pixel number.
6. according to claim 1 be based on the improved adapting to image defogging method of dark primary priori, it is characterised in that: institute
State fog coefficient self-adaptive computing method itself specific steps are as follows: using color range self-adapting regulation method to restored image J0
Enhancing processing is carried out, by carrying out statistics with histogram to various concentration haze image, discovery haze concentration is bigger, histogram distribution
It more concentrates, it is [m that most gray values, which concentrate on size,c,Mc] gray scale interval in, using the property, propose a kind of based on ash
The defogging coefficient calculation method of concentration degree is spent, is met:
In formula: ωcFor minimum value regulation coefficient, value ωc=0.15;Bound mcAnd McMeet:
In formula: α is bound regulation coefficient, value α=1%;
By the above various corresponding substitution transmitance formula (6), the transmitance of the bright dark areas of haze image can be obtained;In conjunction with public affairs
Formula (7), it will be able to the preliminary image J obtained after restoring0。
7. according to claim 1 be based on the improved adapting to image defogging method of dark primary priori, it is characterised in that: institute
State image tonescale self-adapting regulation method itself specific steps are as follows: according to formula (16), seek haze image RGB triple channel
Concentration degree section [mc,Mc], mapping table is calculated using following formula:
The mapping table generated using formula (17), to preliminary restored image J0Color range adjustment is carried out, as follows:
J (x)=Map (J0(x)+1)。 (18)
8. according to claim 1 be based on the improved adapting to image defogging method of dark primary priori, it is characterised in that: also
Including the validation verification method to adapting to image defogging method, specifically includes the following steps: firstly, choosing forest, river
Side, the 6 width haze images that traffic, the city proper and the suburbs isoconcentration of the bright area of sky containing large area is different, bright area area is different
For experimental subjects, it is utilized respectively multiple dimensioned Retinex enhancing defogging method, He dark primary priori restores defogging method and adaptive
Image defogging method carries out subjective and objective analysis to progress defogging processing, and to experimental result.
Priority Applications (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201910177949.1A CN110349113B (en) | 2019-03-11 | 2019-03-11 | Adaptive image defogging method based on dark primary color priori improvement |
Applications Claiming Priority (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201910177949.1A CN110349113B (en) | 2019-03-11 | 2019-03-11 | Adaptive image defogging method based on dark primary color priori improvement |
Publications (2)
Publication Number | Publication Date |
---|---|
CN110349113A true CN110349113A (en) | 2019-10-18 |
CN110349113B CN110349113B (en) | 2023-06-27 |
Family
ID=68174217
Family Applications (1)
Application Number | Title | Priority Date | Filing Date |
---|---|---|---|
CN201910177949.1A Active CN110349113B (en) | 2019-03-11 | 2019-03-11 | Adaptive image defogging method based on dark primary color priori improvement |
Country Status (1)
Country | Link |
---|---|
CN (1) | CN110349113B (en) |
Cited By (3)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN112446841A (en) * | 2020-12-14 | 2021-03-05 | 中国科学院长春光学精密机械与物理研究所 | Self-adaptive image recovery method |
CN112785521A (en) * | 2021-01-19 | 2021-05-11 | 澜途集思生态科技集团有限公司 | Remote sensing image processing method under haze condition |
CN117952864A (en) * | 2024-03-20 | 2024-04-30 | 中国科学院西安光学精密机械研究所 | Image defogging method based on region segmentation, storage medium and terminal equipment |
Citations (4)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
US20160078605A1 (en) * | 2014-09-16 | 2016-03-17 | National Taipei University Of Technology | Image restoration method and image processing apparatus using the same |
CN105787904A (en) * | 2016-03-25 | 2016-07-20 | 桂林航天工业学院 | Adaptive global dark channel prior image dehazing method for bright area |
CN107767354A (en) * | 2017-12-08 | 2018-03-06 | 福州大学 | A kind of image defogging algorithm based on dark primary priori |
CN108765336A (en) * | 2018-05-25 | 2018-11-06 | 长安大学 | Image defogging method based on dark bright primary colors priori with auto-adaptive parameter optimization |
-
2019
- 2019-03-11 CN CN201910177949.1A patent/CN110349113B/en active Active
Patent Citations (4)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
US20160078605A1 (en) * | 2014-09-16 | 2016-03-17 | National Taipei University Of Technology | Image restoration method and image processing apparatus using the same |
CN105787904A (en) * | 2016-03-25 | 2016-07-20 | 桂林航天工业学院 | Adaptive global dark channel prior image dehazing method for bright area |
CN107767354A (en) * | 2017-12-08 | 2018-03-06 | 福州大学 | A kind of image defogging algorithm based on dark primary priori |
CN108765336A (en) * | 2018-05-25 | 2018-11-06 | 长安大学 | Image defogging method based on dark bright primary colors priori with auto-adaptive parameter optimization |
Non-Patent Citations (3)
Title |
---|
李云峰: "基于高亮区域自适应处理的监控图像去雾算法", 《计算机应用与软件》 * |
王泽胜: "基于改进暗通道先验的交通图像去雾新方法", 《控制与决策》 * |
邓莉: "针对明亮区域的自适应全局暗原色先验去雾", 《光学精密工程》 * |
Cited By (4)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN112446841A (en) * | 2020-12-14 | 2021-03-05 | 中国科学院长春光学精密机械与物理研究所 | Self-adaptive image recovery method |
CN112446841B (en) * | 2020-12-14 | 2022-05-31 | 中国科学院长春光学精密机械与物理研究所 | Self-adaptive image recovery method |
CN112785521A (en) * | 2021-01-19 | 2021-05-11 | 澜途集思生态科技集团有限公司 | Remote sensing image processing method under haze condition |
CN117952864A (en) * | 2024-03-20 | 2024-04-30 | 中国科学院西安光学精密机械研究所 | Image defogging method based on region segmentation, storage medium and terminal equipment |
Also Published As
Publication number | Publication date |
---|---|
CN110349113B (en) | 2023-06-27 |
Similar Documents
Publication | Publication Date | Title |
---|---|---|
CN106846263B (en) | Based on the image defogging method for merging channel and sky being immunized | |
CN108734670B (en) | Method for restoring single night weak-illumination haze image | |
CN108765342A (en) | A kind of underwater image restoration method based on improvement dark | |
CN109785262A (en) | Image defogging method based on dark channel prior and adaptive histogram equalization | |
CN107767354A (en) | A kind of image defogging algorithm based on dark primary priori | |
CN108596853A (en) | Underwater picture Enhancement Method based on bias light statistical model and transmission map optimization | |
CN111598791B (en) | Image defogging method based on improved dynamic atmospheric scattering coefficient function | |
CN112419162A (en) | Image defogging method and device, electronic equipment and readable storage medium | |
CN106875351A (en) | A kind of defogging method towards large area sky areas image | |
CN107103591A (en) | A kind of single image to the fog method based on image haze concentration sealing | |
CN110349113A (en) | Self-adaptive image defogging method based on dark channel prior improvement | |
CN109255759A (en) | Image defogging method based on sky segmentation and transmissivity adaptive correction | |
CN103455979A (en) | Low illumination level video image enhancement method | |
CN108154492B (en) | A kind of image based on non-local mean filtering goes haze method | |
CN106296618A (en) | A kind of color image defogging method based on Gaussian function weighted histogram regulation | |
CN109949239A (en) | Self-adaptive sharpening method suitable for multi-concentration multi-scene haze image | |
CN111861896A (en) | UUV-oriented underwater image color compensation and recovery method | |
CN104331867B (en) | The method, device and mobile terminal of image defogging | |
CN111563852A (en) | Dark channel prior defogging method based on low-complexity MF | |
CN106447617A (en) | Improved Retinex image defogging method | |
CN104899836A (en) | Foggy image enhancing device and method based on near infrared multispectral imaging | |
CN113822826A (en) | Low-illumination image brightness enhancement method | |
CN115937019A (en) | Non-uniform defogging method combining LSD (local Scale decomposition) quadratic segmentation and deep learning | |
CN109934779B (en) | Defogging method based on guided filtering optimization | |
CN112070691A (en) | Image defogging method based on U-Net |
Legal Events
Date | Code | Title | Description |
---|---|---|---|
PB01 | Publication | ||
PB01 | Publication | ||
SE01 | Entry into force of request for substantive examination | ||
SE01 | Entry into force of request for substantive examination | ||
GR01 | Patent grant | ||
GR01 | Patent grant |