Frontiers in Computing and Intelligent Systems 
ISSN: 2832-6024 | Vol. 2, No. 3, 2022 

 

22 

A Review of Traditional Dehazing Methods for Foggy 
Images 

Wangming Fang* 

School of Software, Henan Polytechnic University, Jiaozuo 454000, Henan, China 

* Corresponding author: Email: fangwangming@126.com 

 

Abstract: Photographs taken under foggy conditions are prone to low contrast, color deviation, and blurred local details, which 

are detrimental to advanced computer vision tasks such as image recognition, semantic segmentation, and image detection. 

Although deep learning techniques have made significant progress in image dehazing, the effectiveness and role of traditional 

image dehazing methods are still fundamental and have their unique advantages. Based on this, the principles and advantages of 

the existing advanced traditional methods are analyzed from image restoration and enhancement perspectives, and standard 

datasets are introduced. Finally, predictions for the future direction of the traditional image dehazing field are presented. 

Keywords: Atmospheric scattering models; Contrast enhancement; Image dehazing; Image enhancement. 

 

1. Introduction 

Due to the combination of dust, water mist, and other tiny 

particles in the air, a foggy environment is formed with low 

outdoor visibility. Images taken under such conditions largely 

diminish the image's visual effect and adversely affect traffic 

monitoring, gate recognition, and fruit counting realistic 

scene applications. They could be more conducive to image 

recognition, semantic segmentation, image detection, and 

other computer vision tasks. Image dehazing reduces the 

impact of fog on an image as much as possible, and dehazing 

reduces a fogged image to a fog-free image. The removal of 

fog not only improves the image's visual effect but also 

significantly improves the performance of the computer 

vision task. 

There are currently two main types of traditional 

processing methods for image dehazing: one is based on 

image restoration methods for image dehazing. The image 

restoration approach is mainly based on an atmospheric 

scattering model and performs dehazing processes. This 

method is more targeted and achieves natural results with 

little loss of information, but estimating the various 

parameters in the model is a tricky area of research. Another 

type of image-dehazing method is based on image 

enhancement. The image enhancement method removes as 

much image noise as possible, increases the image contrast, 

and improves the image's visual effect. Still, it may cause 

some loss of information for some of the information. 

This paper summarizes and compares image dehazing in 

two traditional directions: image restoration and enhancement. 

It mainly summarizes and reviews the research results in these 

two directions, summarizes their dehazing principles and 

experimental results, gives a brief introduction to commonly 

used image-dehazing datasets, and finally gives a reasonable 

prediction of the future development direction in the field of 

image-dehazing. 

2. Traditional image dehazing methods 
based on image restoration 

The image restoration method is mainly based on the 

backpropagation formula of the atmospheric scattering model, 

the parameters such as atmospheric light value and 

transmittance map in the atmospheric scattering model are 

estimated based on a priori information, and the clear fog-free 

image is restored by substituting the backpropagation formula. 

In general, it has a certain dehazing effect on the image as a 

whole. Still, such methods are based on the model's parameter 

estimation and a priori information and tend to be more 

effective for special scenes with a relatively homogeneous 

environment. In contrast, the dehazing effect could be better 

in common complex natural scenes. 

The atmospheric scattering model is shown in equation (1), 

𝐼(𝑥) = 𝑡(𝑥) ∙ 𝐽(𝑥) + 𝐴 ∙ (1 − 𝑡(𝑥)) (1) 

Where 𝐼(𝑥) is the foggy sky image，𝐽(𝑥) is the clear fog-

free image, A denotes the value of atmospheric light, and 

𝑡(𝑥)denotes the transmittance of the medium, so the goal of 

dehazing based on this is to calculate the value of atmospheric 

light and transmittance based on the existing foggy sky image, 

and thus estimate the fog−free image. When the distribution 

of particles in the atmosphere is spatially uniform, the 

transmittance 𝑡(𝑥)  can be obtained by calculating the 

scattering coefficient 𝛽 and the scene depth 𝑑(𝑥), as shown 

in Equation (2). 

𝑡(𝑥) = 𝑒−𝛽𝑑(𝑥) (2) 

2.1. DCP dehazing method 

In 2009, He et al. [1] proposed the dark channel prior (DCP) 

and found, based on a statistical analysis of data from 

thousands of clear fog-free images, that about 75% of the 

pixels have a value of 0 and 90% of the pixel points have very 

low values. Therefore, He et al. proposed that in local areas 

of daytime fog-free images (non-sky areas), pixel points with 

pixel values tending to 0 exist on at least one of the color 

channels. That is, for an observed image J is that 

𝐽𝑑𝑎𝑟𝑘 = 𝑚𝑖𝑛
𝑦∈Ω(𝑥)

( 𝑚𝑖𝑛
𝑐∈{𝑟,𝑔,𝑏}

(𝐽𝑐(𝑦))) , 𝐽𝑑𝑎𝑟𝑘 → 0 (3) 

Based on this prior it is possible to approximate the 

transmittance map 𝑡(𝑥) and the atmospheric light value A. 

The formula for estimating the transmittance map 𝑡(𝑥) is 

shown in (4), where the 𝜔 takes the value 0.95, and the A 



 

23 

value is estimated by taking the top 0.1% brightest pixels in 

the dark channel map and finding the highest brightness point 

of these pixels in the original foggy sky image value as the 

atmospheric light value A. 

�̃�(𝑥) = 1 − 𝜔 𝑚𝑖𝑛
𝑦∈Ω(𝑥)

(𝑚𝑖𝑛
𝑐

𝐼𝑐(𝑦)

𝐴𝑐 ) (4) 

The dehazing image can be obtained based on the existing 

equation (1), as well as the estimated transmittance map and 

the atmospheric light values. 

J(𝑥) =
I(𝑥) − A

�̃�(𝑥)
+ A (5) 

Although the overall de-fogging effect of the DCP method 

is good, it is less satisfactory for white objects and sky regions. 

And when estimating the window transmittance, the 

Softmatting algorithm used at the beginning could have been 

more efficient. He et al. subsequently proposed an 

improvement by proposing guided filtering instead of the 

Softmatting algorithm, which improved the estimation of the 

transmittance map and effectively improved the algorithm's 

efficiency. 

2.2. CAP dehazing method 

In 2015, Zhu et al. [2] proposed color attenuation prior 

(CAP) by finding that as the size of fog concentration changes, 

the luminance and saturation of foggy sky images also change, 

i.e., the scene saturation value is relatively large in the region 

without fog, and the difference between luminance and 

saturation is slight; the difference between scene luminance 

and saturation is significant in the region with thin fog; and 

the difference between scene luminance and saturation is 

most prominent in the region with thick fog. Thus, based on 

this characteristic that the scene depth has a proportional 

relationship with the difference between luminance and 

saturation, Zhu [2] et al. proposed to use a linear regression 

model to estimate the scene depth 𝑑(𝑥) ， as shown in 

Equation (6). 

𝑑(𝑥) = 𝜃0 + 𝜃1𝑣(𝑥) + 𝜃2𝑠(𝑥) + 𝜀(𝑥) (6) 

where 𝜃0、𝜃1、𝜃2  are unknown parameters, 𝜀(𝑥)  is the 

error term, 𝑣(𝑥)  is the scene brightness, and 𝑠(𝑥)  is the 

scene depth. Zhu et al [2] obtained the values of each 

parameter by supervised learning method for learning training. 

In order to prevent the influence of white objects, a minimum 

operation is required, so the estimated scene depth is finally 

obtained. Therefore, the deformation according to equation (1) 

can be obtained for the dehazed image. 

J(𝑥) =
I(𝑥) − A

𝑚𝑖𝑛{𝑚𝑎𝑥{𝑒−𝛽𝑑(𝑥), 0.1}, 0.9}
+ A (𝟕) 

2.3. IDE dehazing method 

In 2021, Ju et al. [3] proposed that the traditional 

atmospheric scattering model has inherent limitations: since 

the texture features will absorb the atmospheric incident light, 

and the absorption rate of light varies continuously with the 

density of texture features that it will have different lighting 

effects for different scenes, and the traditional atmospheric 

scattering model does not consider this practical point, thus 

leading to poor dehazing effects. Ju et al. [3], on the 

atmospheric scattering model, introduced a new parameter, 

light absorption coefficient 𝛼(𝑥, 𝑦), to establish an enhanced 

atmospheric scattering model (EASM), as shown in Equation 

(8). 

𝐼(𝑥, 𝑦) = 𝐴 ∙ (1 − 𝛼(𝑥, 𝑦)) ∙ 𝜌(𝑥, 𝑦) ∙ 𝑡(𝑥, 𝑦)

+𝐴 ∙ (1 − 𝑡(𝑥, 𝑦)) (8)
 

where the dim effect and light absorption intensity are 

enhanced as the light absorption coefficient 𝛼(𝑥, 𝑦) becomes 

larger when the scene depth 𝑑(𝑥)  decreases, so the light 

absorption coefficient 𝛼(𝑥, 𝑦) is defined as 

𝛼(𝑥, 𝑦) = 1 −
𝑑(𝑥, 𝑦)

𝑚𝑎𝑥(𝑑)
(9) 

Therefore, bringing equation (2) and equation (9) into 

equation (8) yields that 

𝐼(𝑥, 𝑦) = 𝐴 ∙
𝑙𝑛(𝑡(𝑥, 𝑦))

𝑙𝑛(𝑡𝑚𝑖𝑛)
∙ 𝜌(𝑥, 𝑦) ∙ 𝑡(𝑥, 𝑦)

+𝐴 ∙ (1 − 𝑡(𝑥, 𝑦)) (10)

 

Ju et al. [3] proposed the IDE image-dehazing method 

based on the gray world assumption (GWA) and the improved 

atmospheric scattering model, which efficiently implements 

image-dehazing by estimating the values of each parameter 

term through a priori constraints and formula derivation with 

good robustness on foggy sky images of different resolutions 

and better color recovery for dim, foggy sky images. The IDE 

dehazing method takes into account the traditional 

atmospheric scattering. The IDE dehazing method considers 

the limitations of the traditional atmospheric scattering model 

and proposes an enhanced model worthy of further research 

and application. 

3. Traditional image dehazing method 
based on image enhancement 

Image enhancement-based dehazing methods do not 

consider the formation mechanism of fog, and improve the 

visual effect of the image by increasing the image contrast and 

saturation, but will cause a certain loss of detail information 

of the image. At present, the more common methods are 

histogram equalization-based dehazing, Retinex image 

enhancement dehazing and image fusion technology-based 

dehazing, among which the histogram equalization method 

can better improve the image contrast and achieve good visual 

effects, Retinex image enhancement method is more effective 

for low-light foggy day images, image fusion technology 

applied to image dehazing can The image fusion technique 

applied to image dehazing can remove the slight fog while 

retaining the detail information of the image. 

3.1. Histogram equalization dehazing 

Histogram equalization is global histogram equalization. 

Global histogram equalization is a process that changes the 

grayscale histogram of the original image from a relatively 

concentrated portion of grayscale to a uniform distribution of 

grayscale throughout, as shown in Figure 1. Global histogram 

equalization redistributes the pixel values of an image by non-

linearly stretching the image so that the number of pixels in a 

given grayscale is approximately equal. 



 

24 

 
Figure 1. Global Histogram Equalization Process 

Usually, foggy images are whiter than clear images, with 

lower contrast and more concentrated luminance values, 

expressed in the form of single or multiple peaks in the 

histogram and concentrated gray areas in the grayscale 

histogram. After equalization, the grayscale range of the 

foggy image is expanded so that the brightness of each pixel 

is more evenly distributed in the grayscale level, and the 

contrast and clarity of the image are enhanced after global 

histogram equalization, which can achieve a certain dehazing 

effect. The global histogram equalization algorithm usually 

directly processes the overall foggy image, which is suitable 

for the overall brighter or darker image and can effectively 

improve the image's contrast. Still, in most cases, the global 

operation tends to ignore local enhancement, resulting in the 

loss of local information and noticeable visual differences in 

local areas. The actual effect could be better. To solve this 

problem, Kim et al. [4] proposed a histogram equalization 

algorithm for localization, dividing the image into partially or 

non-overlapping local blocks and then performing histogram 

equalization separately to enhance local details. 

3.2. Retinex image enhancement dehazing 

Retinex theory suggests that the color and brightness of a 

point that a person can perceive is not determined by the 

absolute light entering the human eye at that point alone but 

is also influenced by the color and brightness of its 

surroundings. In other words, the image that people can see is 

the result of the combined effect of the reflected component 

of the object and the irradiated component of the light. 

Therefore, dehazing can be achieved by separating the 

irradiated component and the reflected component in the 

foggy image and suppressing the effect of the irradiated 

component. 

Using the single-scale Retinex enhancement method alone 

to enhance foggy sky images has limited effect, and the results 

of dehazing are sensitive to the scale selection of the 

surrounding function. To avoid the drawbacks of the single-

scale Retinex method, Xue et al. [5] proposed a multi-scale 

Retinex method to enhance the visibility of the image. 

Compared with the traditional Retinex enhancement method, 

this method has better results in image enhancement detail 

processing, color fidelity, and image distortion. However, the 

method has limitations regarding causing halos in the dehazed 

images. The method also requires artificial selection of the 

surround function and its corresponding size. Galdran et al. 

[6] provided a mathematical proof of the dual relationship 

between image dehazing and Retinex theory, which 

theoretically connects image dehazing and Retinex 

enhancement through a linear relationship, i.e., after pixel 

intensity, Retinex enhancement is used after inversion, and 

then intensity inversion. This method can be replaced using 

different Retinex enhancement algorithms, and advanced 

dehazing methods can be tried instead. 

3.3. Application of image fusion technique to 

image dehazing 

Image fusion uses image processing methods to fuse 

multiple images to reconstruct an image with better visual 

performance and complete preservation of detailed 

information. Therefore, after fully exploiting the information 

of the original image and obtaining multiple exposure images 

through the original fog map, image fusion techniques can 

achieve specific dehazing effects. 

Galdran et al. [7] first introduced multiple exposure fusion 

based on the Laplace pyramid to image dehazing. Still, the 

Laplace pyramid fusion technique is a basic image fusion 

technique with a limited fusion effect and easy to causes the 

loss of boundary information. Zheng et al. [8] proposed to 

obtain the entropy value of image texture using the grayscale 

difference method after obtaining multiple exposure images 

using gamma correction. The decomposition of the image 

block into three independent components is performed to 

decompose the image block into three independent 

components, then processed separately and fused to obtain the 

final result. Liu et al. [9] proposed a joint contrast 

enhancement and exposure fusion framework that treats 

image dehazing as a local visibility and global contrast 

enhancement problem and then processed separately on three 

independent components. Liu et al. [10] used the gamma 

transform and saturation adjustment to produce a series of 

exposure images, image layer decomposition by bootstrap 

filtering, and then the construction of fusion weights and 

extraction of exposure features, respectively, and finally, the 

two layers of information are fused to obtain a dehazed image. 

The dehazing method based on the image fusion technique 

focuses on the preservation of local details and contrast 

improvement of the image, which can effectively improve the 

performance and robustness of image dehazing. However, the 

fusion-based dehazing method is prone to the problem of 

excessive contrast amplification and color distortion in local 

areas. 

4. Dataset and evaluation index 

In conventional dehazing, the data sources relied on are 

extensive. Almost all foggy sky images under real-world 

conditions can be directly used as image-dehazing processing 

objects, so different researchers collect real foggy sky images 

from different sources. The more commonly used classical 

datasets are the 32 real fog sky images with different 

resolutions and scenes used in the experiments of Zhu et al. 

[2] and the RTTS dataset with more than 4000 real fog sky 

images collected from the web. 

Currently, there are two leading evaluation indicators 

commonly used in the field of traditional image dehazing: 

first, image quality indicators, such as information entropy 

Entropy, BRISQUE, NIQE, BIQI, etc.; second, fog 

concentration evaluation indicators, such as FADE indicator, 

FRFSIM indicator, etc. In addition, the algorithm's 

complexity needs to be explored and analyzed in the image-

dehazing process. Only by continuously optimizing the 

algorithm running time can it be better applied to the actual 



 

25 

production life. 

5. Conclusion 

In this paper, we review traditional advanced dehazing 

methods in image dehazing, classifying and summarizing 

them from both image restoration and image perspectives. 

Image dehazing is a low-level vision task, and restoring a 

foggy image to a fog-free image is beneficial for subsequent 

high-level vision tasks. However, image dehazing faces 

complex and diverse image scenes, the limitations of 

dehazing models are apparent, and the dehazing work is very 

challenging. The traditional image dehazing field can build 

dehazing methods from these perspectives: more reasonable 

dehazing models, dehazing methods with more vital 

generalization ability, and more comprehensive evaluation 

metrics. 

References 

[1] Kaiming He, Jian Sun and Xiaoou Tang, "Single image haze 
removal using dark channel prior," 2009 IEEE Conference on 
Computer Vision and Pattern Recognition, Miami, FL, 2009, 
pp. 1956-1963, doi: 10.1109/CVPR.2009.5206515. 

[2] Q. Zhu, J. Mai and L. Shao, "A Fast Single Image Haze 
Removal Algorithm Using Color Attenuation Prior," in IEEE 
Transactions on Image Processing, vol. 24, no. 11, pp. 3522-
3533, Nov. 2015, doi: 10.1109/TIP.2015.2446191. 

[3] M. Ju, C. Ding, W. Ren, Y. Yang, D. Zhang and Y. J. Guo, 
"IDE: Image Dehazing and Exposure Using an Enhanced 
Atmospheric Scattering Model," in IEEE Transactions on 
Image Processing, vol. 30, pp. 2180-2192, 2021, doi: 
10.1109/TIP.2021.3050643. 

[4] Joung-Youn Kim, Lee-Sup Kim and Seung-Ho Hwang, "An 
advanced contrast enhancement using partially overlapped sub-
block histogram equalization," in IEEE Transactions on 
Circuits and Systems for Video Technology, vol. 11, no. 4, pp. 
475-484, April 2001, doi: 10.1109/76.915354. 

[5] M. Xue, Y. Ji, Z. Yuyan, L. Weiwei and Z. Jiugen, "Video 
Image Dehazing Algorithm Based on Multi-scale Retinex with 
Color Restoration," 2016 International Conference on Smart 
Grid and Electrical Automation (ICSGEA), Zhangjiajie, China, 
2016, pp. 195-200, doi: 10.1109/ICSGEA.2016.42. 

[6] A. Galdran, A. Bria, A. Alvarez-Gila, J. Vazquez-Corral and 
M. Bertalmío, "On the Duality Between Retinex and Image 
Dehazing," 2018 IEEE/CVF Conference on Computer Vision 
and Pattern Recognition, Salt Lake City, UT, USA, 2018, pp. 
8212-8221, doi: 10.1109/CVPR.2018.00857. 

[7] A.Galdran, "Image dehazing by artificial multiple-exposure 
image fusion," Signal Process, 2018 vol. 149, pp. 135-147. 

[8] M. Zheng, G. Qi, Z. Zhu, Y. Li, H. Wei and Y. Liu, "Image 
Dehazing by an Artificial Image Fusion Method Based on 
Adaptive Structure Decomposition," in IEEE Sensors Journal, 
vol. 20, no. 14, pp. 8062-8072, 15 July15, 2020, doi: 
10.1109/JSEN.2020.2981719. 

[9] X. Liu, H. Li and C. Zhu, "Joint Contrast Enhancement and 
Exposure Fusion for Real-World Image Dehazing," in IEEE 
Transactions on Multimedia, vol. 24, pp. 3934-3946, 2022, doi: 
10.1109/TMM.2021.3110483. 

[10] QZ Liu, YQ Luo, K Li, "Single Image Defogging Method 
Based on Image Patch Decomposition and Multi-Exposure 
Image Fusion," Frontiers in neurorobotics, 7 Jul. 2021vol. 15 
700483.