CN103337054A - Two-stage image haze removal method based on single images - Google Patents
Two-stage image haze removal method based on single images Download PDFInfo
- Publication number
- CN103337054A CN103337054A CN2013102408330A CN201310240833A CN103337054A CN 103337054 A CN103337054 A CN 103337054A CN 2013102408330 A CN2013102408330 A CN 2013102408330A CN 201310240833 A CN201310240833 A CN 201310240833A CN 103337054 A CN103337054 A CN 103337054A
- Authority
- CN
- China
- Prior art keywords
- image
- mist
- stage
- color space
- channel
- 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.)
- Pending
Links
Images
Abstract
A two-stage image haze removal method based on single images comprises the following steps: estimating haze concentration according to dark channel prior information; restoring color in each image part according to the estimation result; adopting local gray balance method to enhance contrast of each image part; fusing the brightness channel enhanced image and other channel images to obtain the final image. The two-stage image haze removal method not only obviously enhances the gray contrast of the foreground and the surrounding background in the image, but also remains the texture details in the image, so as to obviously improve the visual effect of the image.
Description
Technical field
The invention belongs to technical field of image processing, be specifically related to a kind of two-stage image defogging method capable based on single image.
Background technology
The image mist elimination is the important content that image is handled and computer vision field is studied, and it is mainly used in fields such as video monitoring, topographic(al) reconnaissance and automatic driving.
The scene depth information of degraded image is an important clue of restoring Misty Image.This kind restored method can be divided into two classes according to scene depth information is whether known.One class is the method for hypothesis scene depth ten-four.This method is proposed the earliest by people such as Oakley.This method of restoring scene contrast based on physical model uses a simple Gaussian function that the light path in the scene is predicted, obtained recovery effect preferably, and do not need weather forecasting information, but the method needs radar installations to obtain scene depth.Another kind of is to carry out the method that scene depth extracts with supplementary.People such as Narasimhan study the method for extracting scene depth information from a plurality of different angles.For example, utilize the two-value scattering model to extract scene information from the colour picture under the different weather condition; Utilize the polarization characteristic of different scattered lights, by the polarized light restoration scenario depth information on the different directions; By determining the discontinuous border of compute depth, extract scene depth the gray level image under two width of cloth different weather conditions; Obtain the scene point degree of depth by interactive depth of field algorithm for estimating and known 3D model, for example the Kopf method namely is to utilize known 3D model to obtain the depth of field, thereby restores Misty Image.On this basis, there is the researcher of China to propose a kind of new image recovery method in recent years again.This method at first to the optical imagery modeling of greasy weather scene, then by means of the reference picture of a fine image and a Misty Image, calculates the depth ratio relation of scene each point.These methods of extracting scene depth combine with the atmospheric scattering model, finally realize the recovery of Misty Image.But the method for above several extraction scene depths also exists certain limitation, can only be applied to the more weak mist of atmospheric scattering degree such as the method for utilizing polarized light, and be unsuitable for foggy weather.Some other method then needs to use the image of same scene under the different weather state or needs the mutual of user, thereby is difficult to satisfy the realtime graphic processing demands to the conversion scene.
In recent years, the image defogging method capable research based on single image causes people's extensive concern.Early stage work is in this respect finished by people such as Tan.They find no mist image with respect to there being the mist image must have higher contrast ratio by statistics, thereby utilize the local contrast that maximizes restored image to reach the purpose of mist elimination, and the color of image after the shortcoming of this method is to restore is usually too saturated.In addition, people such as Fattal are under the local incoherent prerequisite at propagation and the scene objects surface shading light part of hypothesis light, estimate the irradiance of scene, and derive the propagation image thus.Because this method is based on mathematical statistics, and require to have enough colouring informations, thus when handling the image that is faint in color under the thick fog weather, can't obtain believable propagation image, thus the image fault after restoring is bigger.
People such as He Kaiming have proposed the single image mist elimination technology based on dark primary recently.Dark primary priori is from the statistical law to the no mist image data base in open air, and it is based on through observable so ultimate facts--and there is the very low pixel of intensity level of some at least one Color Channel in each regional area of most no mist images in open air.The mist elimination model that utilizes this priori to set up can estimate that the concentration of mist and recovery obtain the image that high-quality removal mist disturbs.But when the brightness of scene target was similar to atmosphere light, this method will lose efficacy.
Summary of the invention
The purpose of this invention is to provide a kind of two-stage defogging method capable towards single image, solved when the brightness of scene target is similar to atmosphere light dark primary priori Problem of Failure.
Technical scheme of the present invention is based on the two-stage image defogging method capable of single image, to repair image at first based on dark primary prior imformation estimation fog concentration, and according to the estimation result; Adopt the local gray level equalization methods to strengthen the contrast of image various piece then, at last luminance channel is strengthened image and obtain final image with other channel image fusions.
The present invention also has following characteristics:
Specifically may further comprise the steps:
Step 1, the input original image utilizes the dark primary transcendental method to estimate surround lighting and transmissivity, the no mist image after obtaining to restore;
Step 2 is transformed into the HIS color space with the no mist image of step 1 gained from rgb color space, extracts luminance component, and brightness and the saturation degree of image are carried out equilibrium treatment, strengthens the contrast of image;
Step 3 returns step 2 gained image transitions to rgb color space, obtains final output image.
In the above-mentioned steps 1, the dark primary transcendental method mainly is based on the statistical law of no mist image; Have at least a Color Channel to have very little color intensity value in the regional area of most of no mist image, this minimum value is tending towards 0, that is:
J
dark(x)→0 (3)
J wherein
cThe Color Channel of presentation video J, Ω (x) are that the center is in the localized mass of x; Utilize this prior imformation, estimate atmosphere light intensity A in the formula (1), calculating transmissivity t (x) is e
-β d (x):
Transmissivity t (x) and atmosphere light A according to formula (1), restore no mist image after determining:
In the above-mentioned steps 2, rgb color space is as follows to the converting expressing formula of HIS color space:
V=max(R,G,B) (8)
The present invention has following beneficial effect:
The present invention has not only obviously strengthened prospect in the image and the intensity contrast of background around it, and has kept the grain details in the image, and compared with prior art, the present invention shows the visual effect that has improved image.
Description of drawings
Fig. 1 is the process flow diagram that the present invention is based on the two-stage image defogging method capable of single image;
Fig. 2 is original greasy weather degraded image M;
Fig. 3 is for utilizing the image after the two-stage image defogging method capable that the present invention is based on single image is handled image M shown in Figure 2;
Fig. 4 is original greasy weather degraded image N;
Fig. 5 is the image that utilizes after existing small wave converting method is handled image N shown in Figure 4.
Fig. 6 is the image that utilizes after existing Retinex method is handled image N shown in Figure 4;
Fig. 7 is the image that utilizes existing dark primary transcendental method that image N shown in Figure 4 is handled;
Fig. 8 is for utilizing the image after the two-stage image defogging method capable that the present invention is based on single image is handled image N shown in Figure 4.
Embodiment
The present invention is described in detail below in conjunction with the drawings and specific embodiments.
The main cause that Misty Image degrades is exactly the scattering process of atmospheric particles.Its scattering process mainly contains two action effects, and the one, the scattering of atmospheric particles during owing to the greasy weather, the target light energy can be decayed before arriving watch-dog, thereby had reduced the intensity of target light energy; In addition, the natural light of sky arrives the monitoring range of watch-dog under the scattering process of greasy weather atmospheric particles, and monitored device is caught, and makes target image be subjected to interference.Just because of the existence of these two action effects, image that the greasy weather claps just can be so unintelligible.These two action effects can be expressed as two physical models: the direct attenuation model of incident light and atmosphere photoimaging model, Narasimhan and Nayar
[1] [2]Deng the people atmospheric scattering model under mist, mist weather condition has been proposed:
I(x)=J(x)e
-βd(x)+A(1-e
-βd(x)) (1)
I (x) is the scene image of input in the formula (1), and J (x) is the scene radiation, and A is the atmosphere light intensity, e
-β d (x)Be transmissivity, wherein β is the atmospheric scattering coefficient, and d (x) is the depth of field.The 1st J (x) e in the formula
-β d (x)Expression incident light attenuation model, because the scattering process of atmospheric particles, the reflected light of a part of body surface loses because of scattering, and the part that is not scattered directly arrives imaging sensor, and the light intensity of arrival is exponential damping along with the distance increase of propagating; The 2nd A (1-e
-β d (x)) expression atmosphere photoimaging model, this is because atmospheric particles causes that to natural scattering of light atmosphere shows the characteristic of light source.The target of image mist elimination recovers J exactly from input picture I, A and e
-β d (x)
The present invention is based on the two-stage image defogging method capable of single image, recover J earlier based on dark primary prior imformation estimation fog concentration, and according to the estimation result, adopt the local gray level equalization methods to strengthen the contrast of image various piece then, as shown in Figure 1, comprise following operation steps:
Step 1, the input original image utilizes the dark primary transcendental method to estimate surround lighting and transmissivity, thus the image after obtaining to restore.The dark primary transcendental method mainly is based on the statistical law of no mist image: have at least a Color Channel to have very little color intensity value in the regional area of most of no mist image, this minimum value is tending towards 0, that is:
J
dark(x)→0 (3)
J wherein
cThe Color Channel of presentation video J, Ω (x) are that the center is in the localized mass of x.Utilize this prior imformation, we can estimate atmosphere light intensity A in the formula (1), and transmissivity t (x) (is e so
-β d (x)) can calculate:
After transmissivity t (x) and atmosphere light A determined, according to formula (1), we just can restore no mist image:
Step 2 to the HIS color space, extracts luminance component with the no mist image transitions of step 1 gained, and brightness and the saturation degree of image are carried out equilibrium treatment, strengthens the contrast of image.Rgb color space is as follows to the converting expressing formula of HIS color space:
V=max(R,G,B) (8)
Step 3 is changed back rgb color space with step 2 calculating gained result and is obtained final output image.
From the contrast of Fig. 2 and Fig. 3 as can be seen, utilize defogging method capable of the present invention that greasy weather degraded image M is handled after, the contrast of image M is obviously promoted, the detailed information in depth of field zone shows, and color of image has also obtained tangible recovery.
Compare with original greasy weather degraded image shown in Figure 4, utilize small wave converting method that image is handled, the brightness of image promotes to some extent, but the part detailed information is still fuzzyyer, is that effect is the poorest in all methods, as shown in Figure 5.Utilize Retinex method mist elimination, color of image obtains certain recovery, but has some halo effects in edge, as shown in Figure 6.Utilize dark primary transcendental method mist elimination, though obtained good effect, but still have the unconspicuous phenomenons of regional area result such as sky, as described in Figure 7.Utilization the present invention is based on the two-stage image defogging method capable mist elimination of single image, compares with other three methods, and the result is the most clear for mist elimination, and has obtained more detailed information, and color fidelity degree is best, as shown in Figure 8.
By contrast as can be seen, dark primary transcendental method pair and the brightness of the scene objects regional area treatment effect similar to atmosphere light is not obvious.The deficiency that the present invention is directed to the dark primary transcendental method is improved, and utilizes the two-stage image defogging method capable that the present invention is based on single image that image is carried out mist elimination, overcome this shortcoming, and image detail information is more outstanding, and color is more natural.
Table 1 has provided above-mentioned several method image shown in Figure 4 has been carried out the quality situation that mist elimination is handled.Main selection standard is poor, average gradient, three objective standards of entropy are contrasted:
1. standard deviation has reflected the discrete situation of gray scale with respect to average as the important indicator of weighing amount of image information, and standard deviation is more big, and grey value profile is overstepping the bounds of propriety looses, and the quantity of information that comprises more levels off to maximum.Formula (9) has provided the definition of standard deviation, and wherein p (g) represents the distribution probability of gray level g, and L is number of greyscale levels,
Be the gray-scale statistical average.
2. average gradient is the sharpness of image, it is the important indicator of weighing image detail contrast ability to express, reflected the speed that image minor detail contrast changes, average gradient is more big, and image level is more many, also just more clear, formula (10) has provided the definition of average gradient, and wherein (i j) is image (i to F, j) gray-scale value, M, N are respectively total line number and total columns of image.
3. entropy is to weigh the important symbol that image information is enriched degree, can contrast details expressive ability between the image by the comparison to image information entropy.Formula (11) has provided the definition of entropy, and wherein p (g) represents the distribution probability of gray level g, and L is number of greyscale levels.
Table 1 several method is to the evaluation of image N mist elimination quality shown in Figure 4
Method | Standard deviation | Average gradient | Entropy (bits/pixel) |
Former figure | 0.22 | 0.02 | 12.81 |
Wavelet transformation | 0.23 | 0.03 | 13.06 |
Retinex | 6.09 | 0.23 | 13.26 |
Dark primary priori | 6.50 | 0.27 | 13.60 |
Defogging method capable of the present invention | 6.80 | 0.30 | 13.80 |
Data from table 1 adopt the image behind the defogging method capable mist elimination of the present invention as can be seen, and its standard deviation, average gradient and entropy all are higher than other several methods, have obviously improved the visual effect of image.
Claims (4)
1. based on the two-stage image defogging method capable of single image, it is characterized in that: repair image various piece color earlier based on dark primary prior imformation estimation fog purulence degree, and according to the estimation result; Adopt the local gray level equalization methods to strengthen the contrast of image various piece then, at last luminance channel is strengthened image and obtain final image with other channel image fusions.
2. the two-stage image defogging method capable based on single image as claimed in claim 1 is characterized in that: specifically may further comprise the steps:
Step 1, the input original image utilizes the dark primary transcendental method to estimate surround lighting and transmissivity, the no mist image after obtaining to restore;
Step 2 is transformed into the HIS color space with the no mist image of step 1 gained from rgb color space, extracts luminance component, and brightness and the saturation degree of image are carried out equilibrium treatment, strengthens the contrast of image;
Step 3 is returned step 2 gained image transitions to rgb color space and is obtained final output image.
3. the two-stage image defogging method capable based on single image as claimed in claim 2, it is characterized in that: in the step 1, the dark primary transcendental method mainly is based on the statistical law of no mist image: have at least a Color Channel to have very little color intensity value in the regional area of most of no mist image, this minimum value is tending towards 0, that is:
J
dark(x)→0 (3)
J wherein
cThe Color Channel of presentation video J, Ω (x) are that the center is in the localized mass of x; Utilize this prior imformation, estimate atmosphere light intensity A in the formula (1), calculating transmissivity t (x) is e
-β d (x):
Transmissivity t (x) and atmosphere light A according to formula (1), restore no mist image after determining:
Priority Applications (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN2013102408330A CN103337054A (en) | 2013-06-17 | 2013-06-17 | Two-stage image haze removal method based on single images |
Applications Claiming Priority (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN2013102408330A CN103337054A (en) | 2013-06-17 | 2013-06-17 | Two-stage image haze removal method based on single images |
Publications (1)
Publication Number | Publication Date |
---|---|
CN103337054A true CN103337054A (en) | 2013-10-02 |
Family
ID=49245204
Family Applications (1)
Application Number | Title | Priority Date | Filing Date |
---|---|---|---|
CN2013102408330A Pending CN103337054A (en) | 2013-06-17 | 2013-06-17 | Two-stage image haze removal method based on single images |
Country Status (1)
Country | Link |
---|---|
CN (1) | CN103337054A (en) |
Cited By (5)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN103985091A (en) * | 2014-04-30 | 2014-08-13 | 西安理工大学 | Single image defogging method based on luminance dark priori method and bilateral filtering |
CN104052967A (en) * | 2014-06-04 | 2014-09-17 | 河海大学 | Intelligent system for acquiring underwater polarization target depth map and method thereof |
CN106462953A (en) * | 2014-06-12 | 2017-02-22 | Eizo株式会社 | Image processing system and computer-readable recording medium |
CN108447034A (en) * | 2018-03-13 | 2018-08-24 | 北京航空航天大学 | A kind of marine Misty Image defogging method decomposed based on illumination |
CN110738624A (en) * | 2019-10-18 | 2020-01-31 | 电子科技大学 | area self-adaptive image defogging system and method |
Citations (2)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
US20110188775A1 (en) * | 2010-02-01 | 2011-08-04 | Microsoft Corporation | Single Image Haze Removal Using Dark Channel Priors |
CN103065127A (en) * | 2012-12-30 | 2013-04-24 | 信帧电子技术(北京)有限公司 | Method and device for recognizing human face in fog day image |
-
2013
- 2013-06-17 CN CN2013102408330A patent/CN103337054A/en active Pending
Patent Citations (2)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
US20110188775A1 (en) * | 2010-02-01 | 2011-08-04 | Microsoft Corporation | Single Image Haze Removal Using Dark Channel Priors |
CN103065127A (en) * | 2012-12-30 | 2013-04-24 | 信帧电子技术(北京)有限公司 | Method and device for recognizing human face in fog day image |
Non-Patent Citations (5)
Title |
---|
GXIAOB: "OPENCV的RGB到HSV颜色空间转换", 《CSDN博客HTTP://BLOG.CSDN.NET/GXIAOB/ARTICLE/DETAILS/8799242》 * |
KAIMING HE, ET AL: "Single Image Haze Removal Using Dark Channel Prior", 《IEEE TRANSACTIONS ON PATTERN ANALYSIS AND MACHINE INTELLIGENCE》 * |
KAIMING HE, ET AL: "Single Image Haze Removal", 《IEEE TRANSACTIONS ON PATTERN ANALYSIS AND MACHINE INTELLIGENCE》 * |
姜冬琴 等: "基于直方图均衡化的彩色图像增强", 《电脑知识与技术》 * |
赵晓丽 等: "基于视觉特性的彩色图像增强算法研究", 《计算机工程与设计》 * |
Cited By (12)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN103985091A (en) * | 2014-04-30 | 2014-08-13 | 西安理工大学 | Single image defogging method based on luminance dark priori method and bilateral filtering |
CN104052967A (en) * | 2014-06-04 | 2014-09-17 | 河海大学 | Intelligent system for acquiring underwater polarization target depth map and method thereof |
CN104052967B (en) * | 2014-06-04 | 2017-04-05 | 河海大学 | Target depth figure is polarized under intelligent water and obtains system and method |
CN106462953A (en) * | 2014-06-12 | 2017-02-22 | Eizo株式会社 | Image processing system and computer-readable recording medium |
US10096092B2 (en) | 2014-06-12 | 2018-10-09 | Eizo Corporation | Image processing system and computer-readable recording medium |
US10102614B2 (en) | 2014-06-12 | 2018-10-16 | Eizo Corporation | Fog removing device and image generating method |
US10157451B2 (en) | 2014-06-12 | 2018-12-18 | Eizo Corporation | Image processing system and computer-readable recording medium |
CN106462953B (en) * | 2014-06-12 | 2019-10-25 | Eizo株式会社 | Image processing system and computer readable recording medium |
CN108447034A (en) * | 2018-03-13 | 2018-08-24 | 北京航空航天大学 | A kind of marine Misty Image defogging method decomposed based on illumination |
CN108447034B (en) * | 2018-03-13 | 2021-08-13 | 北京航空航天大学 | Marine foggy day image defogging method based on illumination decomposition |
CN110738624A (en) * | 2019-10-18 | 2020-01-31 | 电子科技大学 | area self-adaptive image defogging system and method |
CN110738624B (en) * | 2019-10-18 | 2022-02-01 | 电子科技大学 | Area-adaptive image defogging system and method |
Similar Documents
Publication | Publication Date | Title |
---|---|---|
CN103985091A (en) | Single image defogging method based on luminance dark priori method and bilateral filtering | |
Li et al. | Image dehazing using residual-based deep CNN | |
Huang et al. | An advanced single-image visibility restoration algorithm for real-world hazy scenes | |
CN103218778B (en) | The disposal route of a kind of image and video and device | |
CN102831591B (en) | Gaussian filter-based real-time defogging method for single image | |
CN101901473B (en) | Self-adaptive defogging strengthening method of single-frame image | |
CN104794697B (en) | A kind of image defogging method based on dark primary priori | |
CN104318524A (en) | Method, device and system for image enhancement based on YCbCr color space | |
CN104299192B (en) | A kind of single image to the fog method based on atmospheric light scattering physical model | |
CN103337054A (en) | Two-stage image haze removal method based on single images | |
CN107301624A (en) | The convolutional neural networks defogging algorithm pre-processed based on region division and thick fog | |
CN105913390B (en) | A kind of image defogging method and system | |
CN103489166A (en) | Bilateral filter-based single image defogging method | |
CN102768760A (en) | Quick image dehazing method on basis of image textures | |
CN105447825B (en) | Image defogging method and its system | |
CN102436643A (en) | Image defogging method facing to atmospheric scattering proximity effect | |
CN104050637A (en) | Quick image defogging method based on two times of guide filtration | |
CN103077504A (en) | Image haze removal method on basis of self-adaptive illumination calculation | |
Ding et al. | Single-image haze removal using the mean vector L2-norm of RGB image sample window | |
CN105931208A (en) | Physical model-based low-illuminance image enhancement algorithm | |
CN104240192A (en) | Rapid single-image defogging algorithm | |
CN103020921A (en) | Single image defogging method based on local statistical information | |
CN105023246B (en) | A kind of image enchancing method based on contrast and structural similarity | |
CN105913391B (en) | A kind of defogging method can be changed Morphological Reconstruction based on shape | |
CN102254306B (en) | Real-time image defogging method based on image simplified hierachical model |
Legal Events
Date | Code | Title | Description |
---|---|---|---|
C06 | Publication | ||
PB01 | Publication | ||
C10 | Entry into substantive examination | ||
SE01 | Entry into force of request for substantive examination | ||
RJ01 | Rejection of invention patent application after publication | ||
RJ01 | Rejection of invention patent application after publication |
Application publication date: 20131002 |