https://doi.org/10.31449/inf.v47i7.4799 Informatica 47 (2023) 51–62 51 
Clarity Method of Low-illumination and Dusty Coal Mine Images 
Based on Improved AMEF 
Chang Su 
1, 2
, Ziqiang Li 
1, 2 *
, Zhongliang Wei 
3
, Naizhong Xu 
4
 and Quan Yuan 
1, 2 
 
1
State Key Laboratory of Mining Response and Disaster Prevention and Control in Deep Coal Mine, Anhui University 
of Science and Technology, Huainan, China 
2
School of Mechanical Engineering, Anhui University of Science and Technology, Huainan, China 
3
School of Computer Science and Engineering, Anhui University of Science and Technology, Huainan, China 
4
School of Mining, Anhui University of Science and Technology, Huainan, China 
E-mail: suchanguser@126.com, 18697534092@163.com, zhlwei@aust.edu.cn, miningearth@126.com, 
yq2331591242@163.com 
*Corresponding Author 
Keywords: artificial multiple-exposure image fusion, S-type function, gradient domain guide filter, homomorphic 
filtering  
Received: April 14, 2023 
The existing image processing methods based on physical models can have a significant impact on 
defogging performance due to inaccurate estimation of the depth of field information. These methods often 
encounter problems such as low brightness, invisible color distortion, and loss of detail when processing 
images with poor lighting conditions, such as those taken in coal mines. To address these issues, this 
paper proposes a new algorithm based on artificial multi-exposure image fusion. The proposed method 
performs global exposure on images with uneven illumination by combining S-type functions and the 
Contrast Limited Adaptive Histogram Equalization (CLAHE) algorithm in the Hue-Saturation-Value 
(HSV) color space. This reduces the spatial dependence of brightness during processing and avoids color 
distortion problems that may arise in the Red-Green-Blue (RGB) color space. To mitigate the issue of 
detail loss, a gradient-domain guided filter is used to preserve fine structures in images, while an improved 
homomorphic filtering algorithm is introduced during the Laplacian pyramid decomposition process to 
reduce image content loss arising from large dark areas. This paper also conducted subjective, objective, 
and computational time comparisons to evaluate performance, providing reliable results regarding speed, 
quality, and reliability in processing hazy images. 
Povzetek: Predlagana metoda umetne združitve več posnetkov omogoča boljše odstranjevanje meglice v 
slikah z neenakomerno osvetlitvijo. Uporablja se algoritem CLAHE v barvnem prostoru HSV in vodeni 
filter za ohranjanje podrobnosti. Rezultati kažejo visoko kakovost in hitrost obdelave zamegljenih slik. 
 
1 Introduction 
The intricate conditions present in coal mines generate 
numerous external environmental factors such as fog, low 
illumination, and glare that negatively impact the quality 
of instrument-captured images. This decline in image 
quality not only jeopardizes the efficiency and 
advancement of automated underground operations but 
also poses significant safety risks for staff. Unfortunately, 
despite ongoing research efforts, effective image dehazing 
methods for these types of images remain lacking. 
In recent years, numerous methods for image 
dehazing and enhancement have been proposed. Shi et al. 
[1] utilized the normalized gamma correction and contrast 
limited adaptive histogram equalization (CLAHE) to 
process the luminance component and enhanced images 
using color correction based on grey world theory. Zhi et 
al. [2, 3] introduced an adaptable S-type function to adjust 
the luminance component of images with uneven  
 
 
 
illumination, which significantly improved image clarity 
and visual effect. Meanwhile, He et al. [4] proposed the 
dark channel prior (DCP) method for foggy image 
processing but its long processing time and low brightness 
led to color distortion. Qin et al. [5] combined the quadtree 
algorithm with DCP for optimized transmission, resulting 
in significant defogging effects without color distortion. 
Similarly, He et al. [6] employed a guided filter (GIF) to 
obtain fine transmission maps but acknowledged its better 
performance in unique environments like low 
illumination. Meng et al. [7] proposed a boundary 
constraint and context regularization method to optimize 
transmission, achieving defogging for foggy images albeit 
with generally darker results. Zhu et al. [8] constructed a 
nonlinear depth-of-field equation using color attenuation 
prior (CAP) to obtain transmission maps that only apply 
to outdoor images with good lighting conditions. Berman 

52 Informatica 47 (2023) 51–62 C. Su et al. 
et al. [9, 10] modeled pixels in hazy images as Red, Green, 
Blue (RGB) color space lines passing through air light 
coordinates termed ‘haze lines’, estimating transmission 
based on pixel position along the line it belongs to. Ehsan 
et al. [11] used dual channel transmission combined with 
DCP and gradient-domain guided filter to refine dual 
transmission maps, while Galdran [12] employed the 
artificial multi-exposure image fusion (AMEF) algorithm 
to remove dust and fog, needing no inversion of physical 
imaging models to restore clear images. Finally, Yang et 
al. [13] estimated atmospheric light in non-uniform 
illumination at night based on superpixel, successfully 
realizing night image dehazing. 
Li et al. [14] and Ullah et al. [15] applied 
convolutional neural network (CNN) models, called Light 
Dehaze and the adversarial neural network based on 
Pix2pix Framework Dehaze (BPFD), respectively, to 
dehaze images and generated corresponding networks 
using datasets. However, these methods have complex 
structures, requiring numerous datasets to improve 
accuracy.  
Currently, many image processing methods are 
designed for outdoor images with good illumination and a 
single environment. However, these methods may not be 
as effective for underground images with low illumination 
and fog. This paper proposes the use of the AMEF method 
for clearing coal mine underground images, which avoids 
the complicated process of obtaining and refining 
atmospheric light. The use of transmittance significantly 
shortens running time and meets the real-time and 
visibility requirements for underground image processing. 
The main objective of this paper is to enhance the 
quality of underground coal mine images while ensuring 
no color distortion or excessive contrast enhancement. In 
summary, the paper's contributions can be distilled into 
three key aspects: 
Firstly, the paper processes the brightness component 
in the HSV color space of the exposure image set using 
CLAHE and a new S-type function to minimize the impact 
of adverse lighting conditions on the image processing. 
This approach compensates for the shortcomings of direct 
operation in RGB color space, such as color distortion and 
excessive contrast enhancement. 
Secondly, a new method for calculating weights is 
proposed, which utilizes the excellent edge preservation 
performance of gradient domain guided filters to preserve 
the fine structure in the image. 
Finally, the paper introduces a modified 
homomorphic filtering algorithm into Laplacian pyramid 
multi-exposure image decomposition to improve the 
quality of blurred images by enhancing detailed content 
through dual-domain transformation. The proposed 
method not only removes fog interference but also 
enhances the contrast of degraded images. 
The rest of the paper is as follows: 
The second section provides an overview of related 
work on atmospheric scattering models and artificial 
multiple-exposure image fusion. The third section 
describes the proposed method and improvements made. 
Section 4 demonstrates the effectiveness of the proposed 
method through experimental processing of actual 
underground coal mine images and comparison with other 
algorithms, with subjective vision and objective 
evaluation as criteria. Finally, Section 5 presents the 
conclusion. 
2 Related works 
2.1 Atmospheric scattering model 
During the process of light scattering, the direction of light 
is subject to deviation due to the scattering effect of solid 
particles and liquid droplets present in the air, resulting in 
a change in the intensity of light. In the subterranean 
environment of coal mines, the influence of dust and fog 
on images is more severe compared to the atmospheric 
environment. Nonetheless, the fundamental principle 
remains the same and can be analyzed by employing the 
atmospheric scattering model, which is the primary 
physical model used to describe fog in images. This model 
is expressed mathematically by Equation (1): 
 I(x)=t(x)J(x)+A(1-t(x)) (1) 
Where I(x) represent the image affected by fog and 
dust, while J(x) is the clear image without fog. The 
transmittance, t(x), is inversely proportional to the depth 
of field, and A is the atmospheric light intensity. Typically, 
prior information is used to infer t(x) and A. By 
transforming Equation (1), an expression for the restored 
image can be obtained. 
 
J(x)=
I(x) -A
t
+A 
(2) 
2.2 Artificial multiple-exposure image 
fusion 
The objective is to establish a spatially varying image 
enhancement technique capable of eliminating the visual 
effects of fog using the Artificial Multiple Exposure 
fusion (AMEF) techniques proposed by Galdran [12]. 
This technique eliminates the need for estimation of 
transmitted and airborne light in Equation (1). To 
accomplish this, the input blur image I(x) with intensity 
values ranging from 0 to 1 is considered. In order to solve 
the image dehazing problem, any solution J(x) must 
contain intensity values lower than I(x). This can be 
achieved by rearranging Equation (1) as follows. 
 t=
A-I(x)
A-J(x)
 (3) 
Since t(x) [0,1], it follows from Equation (3) that A-
I(x) ≤A-J(x), and concludes that J(x) ≤I(x) ∀x. Therefore, 
AMEF can use a simple and effective multiple exposure 
fusion strategy to fuse the information of a group of multi-
exposure versions of the original image I(x), and obtain a 
clear and fog-free image of I(x). 
Artificial multi-exposure image correction was 
carried out by gamma correction [16]. The Gamma 
transform equation is: 
 J(x)=α ⋅I
γ
(x) (4) 
Where α and γ are the gamma correction parameters, 

Clarity Method of Low-illumination and Dusty Coal Mine Images… Informatica 47 (2023) 51–62 53 
it is an actual positive number. In certain cases, the 
dehazing outcomes may turn out excessively dark when 
the input hazy image contains insufficient areas with 
desirable contrast, especially in the case of overexposed 
images. Therefore, the CLAHE [17] method is usually  
 
 
Figure 1: The flow chart of AMEF defogging. 
 
applied to append the enhanced contrast version of the 
original image to the initial set of artificially underexposed 
images. 
If the dimension of the source image E k(x) is m×n, the 
Laplacian pyramid mixture can be expressed as follows: 
 
  J(x)= ∑ us
(m, n)
N
i=1
[∑ P
k
i
(x)W
k
i
(x)
K
k=1
]  (5) 
 
Where us
(m, n)
 is the operator that up-samples any 
given image to E k dimension. In this paper, operator 
P
k
i
=E
k
i
  is defined as the result of the Laplacian pyramid 
construction for each image. W
k
i
 is obtained through 
construct a Gaussian pyramid for the weight maps W
k
 of 
the fog free region in each image. 𝑊 𝑘 is acquired for every 
underexposed image through the multiplication and 
consolidation of both the contrast and saturation maps. 
Loss of contrast and saturation is one of the primary 
visual effects of fog [18, 19]. In the AMEF method, the 
contrast H k(x) at each pixel x in a given source image 
E
k
( x) ={E
k
R
,E
k
G
,E
k
B
} is measured as the absolute value of 
the response to a simple Laplacian filter. The saturation 
S k(x) at each pixel is estimated from the standard deviation 
of the color channel: 
 
H
k
(x)=
∂
2
E
k
∂x
2
(x)+
∂
2
E
k
∂y
2
(x)  (6) 
 
S
k
(x)= ∑ ( E
k
c
( x) -
E
k
R
( x) +E
k
G
( x) +E
k
B
( x)
3
)
c∈{R,G,B}
2
  
(7) 
 
For a detailed explanation of the dark channel prior, 
readers can refer to [12]. The image dehazing process of 
AMEF is shown in Figure 1. 
2.3 Performance comparisons of haze 
removal methods 
As shown in Table 1, and it provides a comparison 
between the state-of-the-art haze removal methods and the 
proposed method.
 
Table 1: Analogy of the state-of-the-art methods and proposed method. 
Methods Information Findings Cons 
Galdran. [12] 
Proposes a method for image 
dehazing using artificial multi 
exposure fusion. 
High efficiency and 
satisfactory result images. 
There is color distortion in the results. 
Nan et al. [20] 
Uses deep learning methods to 
process low light and dusty coal 
mine images. 
The good visual effect and 
it's efficient for all sorts of 
images. 
1. Requires a large number of on-site 
images. 
2. Low operating efficiency. 
3. Complex algorithm structure. 
Zhang et al. [21] 
Provides a method based on 
atmospheric scattering model 
and DCP to enhance mine 
monitoring images. 
Generates satisfactory and 
neat photos at average 
speed. 
1. The result is too dark. 
2. Contains artifacts. 
Wang et al. [22] 
Transform-based low 
illumination mine images haze 
removal. 
Effective noise reduction 
and edge protection. 
1. Inapplicable for massive turbidity 
gradients. 
2. Has oversaturation/undersaturation 
problems. 
Proposed method 
Adopts a multi-scale fusion 
strategy to remove haze from 
coal mine images, and combines 
image enhancement and filtering 
techniques to enhance contrast 
and image details. 
Applications and advantages 
The proposed method is suitable for coal mine images with low 
visibility and dusty influence and is more capable of balancing haze 
removal and visualization requirements compared to other methods. 
It has higher efficiency and simple operation steps. 

54 Informatica 47 (2023) 51–62 C. Su et al. 
3 Proposed methods 
3.1 Image preprocessing based on 
improved homomorphic filtering 
algorithm 
The primary source of illumination is artificial in coal 
mines, and the uneven distribution of illumination can 
result in poor overall image visibility. To address this 
issue, the homomorphic filtering [23] algorithm can be 
applied to underground images to reduce the influence of 
illumination on the image. 
 
 
 
 
Figure 2: The flow chart of homomorphic filtering. 
To convert a nonlinear signal into a linear model, a 
common technique is to apply logarithmic transformation 
(Ln) to the input image f(x, y) and then use Fourier 
transform (FFT) to transform the image from the spatial 
domain to the frequency domain. Next, the transfer 
function H(u, v) of the filter is used to attenuate the low-
frequency components and enhance the high-frequency 
components. The resulting image is then converted back 
to the spatial domain using a fast Fourier transform (FFT
-
1
) after filtering. Finally, the enhanced image g(x, y) is 
obtained through exponential transformation (exp). The 
flowchart for this process is illustrated in Figure 2. 
The foggy image f(x, y) can be mathematically 
represented as the product of its illuminance component 
L(x, y) and reflection component R(x, y) [24]. This 
relationship can be expressed as Equation (8): 
 
  F(x, y)=L(x, y) ⋅R(x, y) (8) 
 
The fundamental principle of homomorphic filtering 
lies in considering the gray value of an image as the 
product of two components: illuminance L(x, y) and 
reflectance R(x, y). By processing the impact of 
illuminance L(x, y) and reflectance R(x, y) on the gray 
image value, the objective of exposing the details of 
shadow areas can be achieved. The flowchart for this 
process is illustrated in Figure 2. 
In order to address the impact of fog and dust in 
underground coal mine images, this paper employs the 
exponential function as the transfer function for a 
homomorphic filter. This approach can effectively filter 
out the influence of fog and dust on the image. The 
conventional exponential transfer function is represented 
by Equation (9). 
 
 T(u, v)=e
(
-D
0
D(u, v)
)
n
 
(9) 
The optimized exponential transfer function is: 
 
 T(u, v)=(r
H
-r
L
)e
(
-cD
0
D(u, v)
2n
)
+r
L
 
(10) 
 
The distance between the frequency (u, v) and the 
center frequency (u 0, v 0) is denoted by D(u, v) in the 
transfer function, where D 0 represents the cutoff 
frequency, r H represents the high-frequency gain, r L 
represents the low-frequency gain, and c is the sharpening 
coefficient. When 0<r L<1 and r H>1, the filter reduces the 
low-frequency and enhances the high-frequency, thereby 
achieving gray dynamic range compression and contrast 
enhancement simultaneously. The image processing effect 
of different parameters is depicted in Figure 3. 
In Figure3, the value of r L is observed to be a crucial 
factor in determining the low-frequency domain, which 
directly affects image contrast and brightness 
enhancement. To cater to the specific environmental 
characteristics of underground coal mine images, this 
paper sets the filter parameters as follows: r L=0.3, r H=1.5, 
D 0=4, and c=3. 
 
Figure 3: Comparison of GIF and GDGIF. 
3.2 Image enhancement based on new S-
type curve enhancement function 
A luminance adjustment is necessary to equalize the 
brightness components of the underground coal mine 
image as it contains both bright and dark areas. The 
adjustment involves enhancing the pixel value of the low-
brightness areas and suppressing the pixel value of the 
high-brightness areas. Therefore, this paper combines an 
S-type curve function [2] with CLAHE algorithm to adjust 
the luminance component. The S-type curve function is 
appropriately revised to adapt to the complex environment 
of low illumination and fog and dust in underground coal 
mine images. The expression of the revised S-type curve 
function is as follows ： 
 
 
 P(x, y)=1/ (1+a√
1-F( x, y)
F( x, y) +τ
) +ε 
(16) 
 
 L(x, y)=max(P(x, y),0) (17) 
 
Among them: 
 

Clarity Method of Low-illumination and Dusty Coal Mine Images… Informatica 47 (2023) 51–62 55 
  ε= − 0.1 |F(x, y) −
1
1+a
2
| (18) 
 
Here, F(x, y) represents the input luminance 
component, a is the illuminance adjustment coefficient, τ 
is constant used to prevent the denominator from being 
zero, usually set to 1E-12, and ε is the fog suppression 
coefficient. To obtain a clearer overall image, the value of 
a is set to 0.9. However, due to the interference of fog and 
dust in the mine environment, the original function may 
not achieve the desired effect. Therefore, a suppression 
coefficient ε is added in this paper. 
 
 
Figure 4: S-type curve function. 
 
When the value of a is set to the different parameters, 
Figure 4 shows the results image of this paper, and the 
difference of function graph between the improved S-type 
function and the original S-type function. 
3.3 Fine contrast based on gradient 
domain guided filtering 
To address the halo artifact issue at the edge of the 
underground coal mine images, caused by the 
conventional guided filter, and taking into account the 
specific features of the coal mine environment, the 
gradient domain guided filter (GDGIF)[25] is used instead 
of the Laplace filter. GDGIF provides a better edge-
preserving performance when processing the underground 
exposure image, which allows obtaining the fine contrast 
function. 
In gradient-domain guided filtering, a filtering 
window centered on pixel k, ω k, is assumed. A local linear 
model is established between the guided image I i and the 
filter output q i, expressed as: 
 
  q
i
=a
k
I
i
+b
k
   
∀i ∈ ω
𝑘 (19) 
 
To determine the linear coefficients a k and b k that 
minimize the difference between the filter output q i and 
the filter input p i in a filtering window, a minimization 
energy equation is defined as follows in gradient domain 
guided filtering. 
 
 E(a
k
,b
k
)= ∑ ((a
k
I
i
+b
k
-p
i
)
2
+
ε
Γ
G(k)
( a
k
-γ
k
)
2
) 
i∈ωk
 
(20) 
 
Among them: 
 
 
 Γ
G(k)
=
1
N
∑
χ
k
+λ
χ
i
+λ
 
N
i=1
 
(21) 
  γ
k
=1-
1
1+e
η( 𝜒 𝑘 -𝜇 χ,∞
)
 (22) 
 
In gradient-domain guided filtering, a minimization 
energy equation is defined to determine the linear 
coefficients a k and b k that minimize the difference between 
the filter output qi and the filter input p i in a filtering 
window. This equation includes several parameters, such 
as ε as the regularization parameter, as the edge sensing 
weight coefficient, which is defined by the local variance 
in the filtering window, λ as a constant, and the size as 
(0.001×L)
2
, where L represents the dynamic range of the 
input image. Additionally, χ k is defined σ
G,1(k)
σ
G,ξ
1(k)
, 
where ξ 1 is the window size of the filter, is the edge 
sensing constraint term, and 𝜇 𝜒 ,∞
 is the mean value of χ i. 
The value of η is calculated as 4/(μ
𝜒 ,∞
-min(χ
i
)). When the 
pixel k is at the edge, γ k is close to 1, and when the pixel k 
is in a flat region, γ k is close to 0. The solution of the 
coefficient can be determined using a linear regression 
equation. 
 
  a
k
=
1
|ω|
∑ I
i
P
i
-μ
k
P
̅
k
+ (
ε
Γ
G(k)
)γ
k
i,k ∈ω
k
σ
k
2
+
ε
Γ
G(k)
 
(23) 
  b
k
=p
k
-a
k
μ
k
 (24) 
   
Where μ k and σ
k
2
 is mean and variance of I in the ω k, 
  
 
Figure 5: Comparison of GIF and GDGIF. 
|ω| is window pixels in the ω k, P
k
= ∑ P
i i∈ω
k
/|ω| is the 
average of P in the ω k. Figure 5 shows the different 
processing effects of the gradient domain guide filter and 
the guide filter when the input image and the guide image 
are the same and the filtering parameters are: r=16, 
ε=0.04. 
In this paper, the contrast function H(x) is composed 
of I
k
γ
(x) initial contrast function of ϕ(x) . For gradient 
domain guided filtering: 
Original image
a=0.1 a=0.5 a=0.9
GIF Original image GDGIF

56 Informatica 47 (2023) 51–62 C. Su et al. 
 
  H(x)=ϕ
k
(x) ⊕G(K, r) (25) 
  
Where ⊕G(k, r) refers to using gradient domain 
guidance filtering on E
k
i
( x), K refers to the guidance 
image, and r refers to the filtering size. In this paper, the 
gray image of the original exposure image is used as K, 
and the filter size is: 
   r= |
1
6
max(h, w)-1| (26) 
Where, ⌈ ⌉ represents the rounding operation, h and 
w represent the height and width of the corresponding 
image. The larger the size of the filter, the more detailed 
information is contained in the contrast component, and 
the operation complexity and time of the filter have no 
relationship with the filter size. 
Filtered image Φ(x) is: 
  Φ(x)=I
max
Ω
(x)-I
min
Ω
(x) (27) 
Where 𝛺 is the size of the given region, 
I
Ω
max(x)= max {I(x)|x ∈Ω}, I
Ω
min (x) =min{I(x)|x ∈Ω}. 
3.4 The specific implementation process of 
proposed algorithm 
The algorithm flow chart is shown in Figure 6. 
 
 
Figure 6: The flow chart of the proposed algorithm. 
 
Firstly, this paper introduces the contrast-enhanced 
image into the artificial expo-sure image set. Unlike the 
original method, which applies the CLAHE algorithm 
directly to the original image, this paper converts the input 
image into HSV color space and uses a combination of 
CLAHE and a new S-type curve function to correct the 
value component in HSV color space and then converts  
the corrected image to RGB color space. 
Next, a group of available exposure images is 
obtained from the multiple exposure image set, and the 
corresponding weight is calculated before artificially 
fusing and defogging. In this paper, the gradient domain 
guided filter is used to obtain the expo-sure image 
contrast, which is combined with the saturation obtained 
from the standard deviation of each color channel to obtain 
the weight. 
Finally, an improved homomorphic filter is 
introduced into the Laplacian pyramid that decomposes 
the available images and is fused step by step with the 
weight map obtained by using the Gaussian pyramid 
decomposition to obtain the final output clear coal mine 
image. 
4 Experimental results 
This section presents an evaluation of the proposed 
method using both subjective visual perception and 
objective evaluation indicators. All the test images were 
obtained from Flickr and [26, 27]. 
All experiments were conducted on a laptop with a 
2.30GHz CPU and 8GB RAM using MATLAB R2017A. 
In order to evaluating the effectiveness of the proposed 
method, several state-of-the-art technologies were used 
for comparison. Figure 7 to Figure 9 depict the 
experimental results. Specifically, Figure 7 illustrates the 
contrast enhancement and equalization of brightness 
components in the HSV color space, compared with the 
enhanced image by the original CLAHE algorithm. 
Meanwhile, Figure 8 and Figure 9 display the processing 
results of the proposed method and other methods such as 
Shin et al. [28], Li et al. [29], Zhu et al. [30], Galdran [12], 
and Ehsan et al. [11]. To ensure a fair comparison, the data 
of the state-of-the-art methods were obtained from the 
public codes of their respective authors. 
4.1 Subjective evaluation 
In Figure 7, the rectangular boxes highlight the hydraulic 
support and coal layer areas that exhibit color distortion 
and contrast over-enhancement issues. It is observed that 
direct application of CLAHE to the original image may 
lead to such problems in certain regions of the image. On 
the other hand, the proposed method employs a con-
version of the original image to the HSV color space for 
adjustment, which can effectively prevent such problems. 
Figure 8 displays the results of various algorithms 
applied to coal mine images. The results of Shin et al. [28] 
Input image Gamma
Correction
Exposure 
images
RGB 
transform to 
HSV
H
S
V
CLAHE
and
S-type function
Corrected V 
Enhanced 
result 
HSV
transform to 
RGB
Multi-exposure 
image set
Homomorp-
filter
Enhnaced 
image
GDGIF
standard 
deviation
Contrast
Saturation
Weight 
maps
Ouput image
Laplacian 
pyramid 
Multi-level images
Multi-level weights
Fusion
Gauss 
pyramid 

Clarity Method of Low-illumination and Dusty Coal Mine Images… Informatica 47 (2023) 51–62 57 
 
Figure 7: Comparison of image enhancement results between proposed method and CLAHE. 
 
Figure 8: Comparison of different haze removal algorithms on coal mine images a-d using state-of-the-art methods 
and the proposed method. Between image a and b is the regions marked in yellow frames are enlarged. (a) input 
image; (b) Galdran. [12]; (c) Shin et al. [28]; (d) Li et al. [29]; (e) Ehsan et al. [11]; (f) Zhu et al. [30]; (g) proposed 
method.
and Ehsan et al. [11] show that the overall brightness of 
the images is lower compared to other algorithms, and the 
details of the images are poorly preserved. Galdran [12] 
and Zhu et al. [30] exhibit different issues such as color 
distortion and excessive contrast enhancement. For 
instance, in the yellow box marked area of images a and 
b, Galdran [12] suffers from color distortion at the 
hydraulic support and grayscale mutation area, while Zhu 
et al. [30] cause serious color and contrast imbalance in 
the entire image c. In contrast, Li et al. [29] and the 
proposed method demonstrate better visual effects, such 
as better brightness levels and clearer human visual 
perception details.  
As shown in Figure 9, this paper not only compares 
the processing effects of different algorithms for close-
range hazy mine images but also processes some 
nighttime images with similar lighting environments as 
the coal mine. For the images with large dark areas and 
fog in the distant view, such as images a and b, Li et al. 
[29] and Ehsan et al. ’s [11] processing results reveal that 
the overall brightness of the image is low. Although the 
fog can be removed, the visual effect of the processed 
image is deteriorating. For images g, h, i, Zhu et al. [30] 
will cause serious color distortion and excessive contrast 
enhancement when processing images whose overall tone 
tends to be consistent due to light and fog. Shin et al. ’s 
[25] dehazed images are still hazy. Galdran [12] and the 
proposed method exhibit more satisfactory results than 
other algorithms. 
4.2 Objective evaluation 
This paper has utilized six commonly used evaluation 
metrics for assessing the quality of haze removal in 
images, namely Peak Signal-to-Noise Ratio (PSNR) [31], 
Structural Similarity (SSIM) [32], Mean Light Intensity 
(MLI), Contrast Index (CI), Entropy (E) and Average 
Gradient (AG). These metrics were chosen to provide a 
com-prehensive evaluation of the performance of different 
haze removal algorithms. 
Figure 10 presents the objective average evaluation 
results of the various algorithms based on the 
aforementioned metrics. In addition, Figure 11 illustrates 
the comparison of the running times and the number of 

58 Informatica 47 (2023) 51–62 C. Su et al. 
processed pixels for each of the six methods used in this paper. 
 
Figure 9: Comparison of different haze removal algorithms on nighttime images e-i using state-of-the-art methods and 
the proposed method. (a) input image; (b) Galdran. [12]; (c) Shin et al. [28]; (d) Li et al. [29]; (e) Ehsan et al. [11]; (f) 
Zhu et al. [30]; (g) proposed method.
According to the results presented in Figure 10, Table 
2 summarizes the performance of six evaluation metrics of 
all experimental images. Galdran [12] and the proposed 
method achieved the highest scores in the SSIM index, 
both exceeding 0.8, which indicates that they can 
effectively retain the content information of the original 
image during the dehazing process. Ehsan et al. [11] 
scored lower than other algorithms in the E, MLI, AG and 
CI indices, suggesting that it may be challenging to meet 
the dehazing and visualization requirements when 
processing such images. Shin et al. [28] showed good 
performance in the E and AG indices, indicating that it can 
effectively retain the structural information of the original 
image. Li et al. [29] ranked relatively high in the E index, 
but its CI and MLI scores were lower, indicating that it has 
good dehazing performance, but the image quality after 
dehazing may be poor. Zhu et al. [30] had the lowest 
scores in both PSNR and SSIM indices, indicating that the 
processed image is severely distorted and may not meet 
visual requirements. The proposed method achieved the 
top ranking in MLI, CI, PSNR, SSIM and E indices, and 
its results in the AG index was only slightly lower than the 
top performer, demonstrating that the proposed method 
can achieve good human visualization performance. 
As shown in Figure 11, the processing time of Shin et 
al. [28] and Ehsan et al. [11] algorithms is significantly 
higher than that of other algorithms. In particular, there is 
an order-of-magnitude difference between Ehsan et al. 
[11] algorithm and the other algorithms. The proposed 
method is second only to Zhu et al. [30] in terms of 
processing efficiency, and its processing time changes 
only slightly with increasing image size, indicating its 
suitability for real-time processing of images. 

Clarity Method of Low-illumination and Dusty Coal Mine Images… Informatica 47 (2023) 51–62 59 
 
Figure 10: Quantitative analysis results (average) of test image.
 
Table 2: Average six metrics values of test hazy images of 
different algorithms (the maximum value is marked in bold). 
 
 
 
Metho
d 
Quantitative metrics 
E MLI CI AG PSNR SSIM 
[24] 7.32 0.35 0.36 0.087 20.21 0.75 
[25] 6.92 0.26 0.23 0.032 20.12 0.71 
[26] 7.12 0.36 0.33 0.065 17.43 0.52 
[12] 7.36 0.39 0.37 0.073 22.62 0.83 
[11] 6.25 0.22 0.18 0.034 17.84 0.65 
Ours 7.49 0.44 0.41 0.083 24.56 0.87 
 
Figure 11: Comparison of processing time of different 
algorithms on haze images with differing numbers of pixels. 
4.3 Discussion 
This article proposed a haze removal method for coal mine 
images, which uses a multi exposure fusion method to 
process the image. Different from previous dehazing 
methods that rely on physical models and may result in 
incorrect estimation of transmittance and atmospheric 
light, the proposed method avoids potential problems with 
over-darkness and artifacts in the image that may arise 
from the presence of white objects or light sources. 
Similarly, the proposed method employed a combination 
of gradient domain guided filtering and improved 
Homomorphic filtering, which effectively preserves 
image details and protects edge information during the 
dehazing process. In terms of efficiency, pyramid 
decomposition is introduced to process hazy images, 
unlike those methods that perform complex iterative 
optimization of model parameters, which ensures the 
image integrity during the processing. At the same time, 
most current haze removal methods for coal mine images 
struggle to balance the requirements of both haze removal 
and image visualization in the challenging low-light and 
dusty environments commonly encountered in coal mine. 
In this study, the proposed method addressed this issue 
through the use of an S-type curve function and CLAHE 
in the HSV color space to enhance image visualization 
while simultaneously removing haze. 
5 Conclusion 
In this paper, AMEF was first introduced, which does not 
directly rely on depth-related physical models but rather 
restores hazy images by fusing the good contrast region of 
multiple exposure images. However, when an input image 
is degraded by multiple physical models, such as 
nighttime haze or water attenuation, AMEF and most 
advanced algorithms cannot solve the color temperature or 
low illuminance problems. An improved method for 
processing fog and dust degradation images in complex 
underground environments based on AMEF was proposed 
Index: AG
0.04
0.08
0.06
0.10
0.02
Index: PSNR Index: SSIM
1.0
0.9
0.8
0.7
0.6
0.5
0.4
27.5
25.0
22.5
20.0
17.5
15.0
Shin et
al[31]
Li et
al[32]
Zhu et
al[33]
Galdran[12] Ehsan et
al[11]
proposed
15.00
17.50
20.00
22.50
25.00
27.50
Index: PSNR
Li et
al[29]
Shin et
al[28]
Zhu et
al[30]
Ehsan et 
al[11]
Galdran
[12]
Proposed
Li et
al[29]
Shin et
al[28]
Zhu et
al[30]
Ehsan et 
al[11]
Galdran
[12]
Proposed Li et
al[29]
Shin et
al[28]
Zhu et
al[30]
Ehsan et 
al[11]
Galdran
[12]
Proposed Shin et
al[31]
Li et
al[32]
Zhu et
al[33]
Galdran[12] Ehsan et
al[11]
proposed
0.4
0.5
0.6
0.7
0.8
0.9
1.0
Index: SSIM
Shin et
al[31]
Li et
al[32]
Zhu et
al[33]
Galdran[12] Ehsan et
al[11]
proposed
0.02
0.04
0.06
0.08
0.10
Index: AG
Index: CI
0.44
0.38
0.32
0.25
0.19
0.13
Index: MLI Index: E
Li et
al[29]
Shin et
al[28]
Zhu et
al[30]
Ehsan et 
al[11]
Galdran
[12]
Proposed
0.50
0.45
0.40
0.35
0.30
0.25
0.20
5.98
7.36
6.90
6.44
7.82
Li et
al[29]
Shin et
al[28]
Zhu et
al[30]
Ehsan et 
al[11]
Galdran
[12]
Proposed Li et
al[29]
Shin et
al[28]
Zhu et
al[30]
Ehsan et 
al[11]
Galdran
[12]
Proposed Shin et
al[31]
Li et
al[32]
Zhu et
al[33]
Galdran[12] Ehsan et
al[11]
proposed
0.13
0.19
0.25
0.32
0.38
0.44
Index: CI
Shin et
al[31]
Li et
al[32]
Zhu et
al[33]
Galdran[12] Ehsan et
al[11]
proposed
0.10
0.15
0.20
0.25
0.30
0.35
0.40
0.45
0.50
0.55
Index: MLI
0.15
Shin et
al[31]
Li et
al[32]
Zhu et
al[33]
Galdran[12]Ehsan
et
al[11]
proposed
5.98
6.44
6.90
7.36
7.82
Index: E
200*200 200*400 400*400 400*600 600*800
0
5
10
15
20
25
30
35
Time/s
Pixel number
 Shin et al[31]
 Li et al[32]
 Zhu et al[33]
 Galdran[12]
 Ehsan et al[11]
 proposed
Pixel number
0.8E+05 1.8E+05 2.8E+05 3.8E+05 4.8E+05
Time/s
30
25
20
15
10
5
0
35

60 Informatica 47 (2023) 51–62 C. Su et al. 
in this paper. Firstly, a new adjustment function was 
introduced into the HSV color space to enhance the 
exposed image for correcting the color distortion. 
Secondly, new weight maps were built through gradient 
domain guided filter to retain the image details as much as 
possible. Lastly, in aspect of contrast enhancement, the 
improved homomorphic filtering algorithm was 
introduced into the Laplace pyramid. According to the 
experimental results, the proposed method was shown to 
be superior to other advanced methods in terms of efficient 
and effective image dehazing. 
There are still some remaining issues worth 
investigating in future work, including a more in-depth 
analysis of model parameters and how to apply the 
proposed method to real-time processing of video images. 
Acknowledgements 
This research was supported by the Independent Research 
Fund of the State Key Laboratory of Mining Response and 
Disaster Prevention and Control in Deep Coal Mines (No. 
SKLMRDPC20ZZ06) and the program in the Youth Elite 
Support Plan in Universities of Anhui Province (No. 
gxyq2020013) and Innovation fund of Anhui University 
of Science and Technology (2021CX2054). 
References 
[1] Shi, Z. H.; Feng, Y. N.; Zhao, M. H.; Zhang, E. H.; He, 
L. F. Normalised gamma transformation-based 
contrast-limited adaptive histogram equalisation with 
colour correction for sand-dust image enhancement. 
Iet Image Process, 2020, 14(4): 747-756. 
https:// doi.org/10.1049/iet-ipr.2019.0992 
[2] Zhi, N.; Mao, S; Li, M. Enhancement algorithm based 
on illumination adjustment for non-uniform 
illuminance video images in coal mine. Journal of 
China Coal Society, 2017, 42(08): 2190-2197. 
https:// doi.org/10.13225/j.cnki.jccs.2017.0048.html 
[3] Zhi, N.; Mao, S; Li, M. An enhancement algorithm for 
coal mine low illumination images based on bi-gamma 
function. Journal of Liaoning Technical 
University(Natural Science) 2018, 37(01): 191-197. 
https://doi.org/10.11956/j.issn.1008-
0562.2018.01.034 
[4] Kaiming H, Jian S and Xiaoou T. Single image haze 
removal using dark channel prior. In 2009 IEEE 
Conference on Computer Vision and Pattern 
Recognition June 20-25; Miami, FL, 2009; pp. 1956-
1963. 
https://doi.org/ 10.1109/CVPR.2009.5206515 
[5] Qin, H; Li, Y; Long, W; Zhao, R. Real-time video 
dehazing using guided filtering and transmissivity 
estimated based on dark channel prior theory. Journal 
of Zhejiang University(Engineering Science) 2018, 
52(07): 1302-1309. 
https://doi.org/10.3785/j.issn.1008-973X.2018.07.010 
 
[6] He, K. M.; Sun, J. A.; Tang, X. O. Guided Image 
Filtering, 11th European Conference on Computer 
Vision, Heraklion, GREECE, Sep 05-11; Heraklion, 
GREECE, 2010; pp. 1-+. 
https://doi.org/10.1007/978-3-642-15549-9_1 
[7] Meng, G.; Wang, Y.; Duan, J.; Xiang, S.; Pan, C. 
Efficient Image Dehazing with Boundary Constraint 
and Contextual Regularization, In Proceedings of the 
IEEE International Conference on Computer Vision 
(ICCV), Sydney, AUSTRALIA, 01-08 December 
2013; pp. 617-624. 
https://doi.org/ 10.1109/iccv.2013.82 
[8] Zhu, Q.; Mai, J.; Shao, L. A Fast Single Image Haze 
Removal Algorithm Using Color Attenuation Prior. 
IEEE Trans Image Process 2015, 24(11): 3522-33. 
https://doi.org/ 10.1109/TIP.2015.2446191. 
[9] Berman, D.; Treibitz, T.; Avidan, S. Non-local 
Image Dehazing. In Proceedings of the 2016 IEEE 
Conference on Computer Vision and Pattern 
Recognition (CVPR), 9 December 2016; pp. 1674-
1682. 
https://doi.org/ 10.1109/cvpr.2016.185. 
[10] Berman, D.; Treibitz, T.; Avidan, S. Air-Light 
Estimation Using Haze-Lines, In Proceedings of the 
IEEE International Conference on Computational 
Photography (ICCP), Stanford Univ, Stanford, CA, 
12-14 May 2017; pp. 115-123. 
https://doi.org/ 10.1109/ICCPHOT.2017.7951489 
[11] Ehsan, S. M.; Imran, M.; Ullah, A.; Elbasi, E. A 
Single Image Dehazing Technique Using the Dual 
Transmission Maps Strategy and Gradient-Domain 
Guided Image Filtering. IEEE Access 2021, 9: 
89055-89063. 
https://doi.org/ 10.1109/access.2021.3090078 
[12] Galdran, A., Image dehazing by artificial multiple-
exposure image fusion. Signal Process 2018, 149: 
135-147. 
https://doi.org/ 10.1016/j.sigpro.2018.03.008 
[13] Yang, M.; Liu, J.; Li, Z. Superpixel-Based Single 
Nighttime Image Haze Removal. IEEE Transactions 
on Multimedia 2018, 20(11): 3008-3018. 
https://doi.org/ 10.1109/tmm.2018.2820327 
[14] Li, S. Y.; Lin, J.; Yang, X.; Ma, J.; Chen, Y. F. 
BPFD-Net: enhanced dehazing model based on 
Pix2pix framework for single image. Machine Vision 
and Applications 2021, 32(6): 1-13. 
https://doi.org/ 10.1007/s00138-021-01248-9 
[15] Ullah, H.; Muhammad, K.; Irfan, M.; Anwar, S.; 
Sajjad, M.; Imran, A. S.; de Albuquerque, V. H. C. 
Light-DehazeNet: A Novel Lightweight CNN 
Architecture for Single Image Dehazing. IEEE Trans 
Image Process 2021, 30: 8968-8982. 
https://doi.org/ 10.1109/TIP.2021.3116790 
[16] Huang, S. C.; Cheng, F. C.; Chiu, Y. S. Efficient 
contrast enhancement using adaptive gamma 
correction with weighting distribution. IEEE Trans 
Image Process 2013, 22(3): 1032-41. 
https://doi.org/ 10.1109/TIP.2012.2226047 
[17] Zuiderveld, K. VIII.5. - Contrast Limited Adaptive 
Histogram Equalization. In Proceedings of the 

Clarity Method of Low-illumination and Dusty Coal Mine Images… Informatica 47 (2023) 51–62 61 
Graphics Gems, Heckbert, P. S., Ed. Academic Press: 
1994; pp. 474-485. 
https://doi.org/10.1016/B978-0-12-336156-
1.50061-6. [18] Xu, G.; Aminu, M. J., An 
Efficient Procedure for Removing Salt and Pepper 
Noise in Images. Informatica 2022, 46(2). 
https://doi.org/ 10.31449/inf.v46i2.3530 
[19] Shi, Z.; Feng, Y.; Zhao, M.; Zhang, E.; He, L. Let 
You See in Sand Dust Weather: A Method Based on 
Halo-Reduced Dark Channel Prior Dehazing for 
Sand-Dust Image Enhancement. IEEE Access 2019, 
7: 116722-116733. 
https://doi.org/ 10.1109/access.2019.2936444 
[20] Nan, Z.; Gong, Y., An Image Enhancement Method 
in Coal Mine Underground Based on Deep Retinex 
Network and Fusion Strategy. In 2021 6th 
International Conference on Image, Vision and 
Computing (ICIVC), 2021; pp 209-214. 
https://doi.org/ 10.1109/icivc52351.2021.9526933. 
[21] Zhang, W.; Zuo, D.; Wang, C.; Sun, B., Research on 
image enhancement algorithm for the monitoring 
system in coal mine hoist. Measurement and Control 
2023. 
https://doi.org/ 10.1177/00202940231173767 
[22] Wang, M.; Tian, Z., Mine image enhancement 
algorithm based on nonsubsampled contourlet 
transform. Journal of China Coal Society, 2020, 
45(9): 3351-3362. 
https://doi.org/ 10.13225/j.cnki.jccs.2019.0798 
[23] Fan, Y.; Zhang, L.; Guo, H.; Hao, H.; Qian, K. Image 
Processing for Laser Imaging Using Adaptive 
Homomorphic Filtering and Total Variation. 
Photonics 2020, 7(2): 30-45. 
https://doi.org/ 10.3390/photonics7020030 
[24] Jobson, D. J.; Rahman, Z.; Woodell, G. A. A 
multiscale retinex for bridging the gap between color 
images and the human observation of scenes. IEEE 
Transactions on Image Processing 1997, 6(7): 965-
976. 
https://doi.org/ 10.1109/83.597272 
[25] Kou, F.; Chen, W.; Wen, C.; Li, Z. Gradient Domain 
Guided Image Filtering. IEEE Trans Image Process 
2015, 24(11): 4528-39. 
https://doi.org/ 10.1109/TIP.2015.2468183. 
[26] Mao, Q.; Wang, Y.; Zhang, X.; Zhao, X.; Zhang, G.; 
Mushayi, K. Clarity method of fog and dust image in 
fully mechanized mining face. Machine Vision and 
Applications 2022, 33(2): 30-45. 
https://doi.org/ 10.1007/s00138-022-01282-1 
[27] Yu, T.; Song, K.; Miao, P.; Yang, G.; Yang, H.; Chen, 
C. Nighttime Single Image Dehazing via Pixel-Wise 
Alpha Blending. IEEE Access 2019, 7: 114619-
114630. 
https://doi.org/ 10.1109/access.2019.2936049 
[28] Shin, J.; Kim, M.; Paik, J.; Lee, S. Radiance–
Reflectance Combined Optimization and Structure-
Guided $\ell _0$-Norm for Single Image Dehazing. 
IEEE Transactions on Multimedia 2020, 22(1): 30-
44. 
https://doi.org/ 10.1109/tmm.2019.2922127 
[29] Li, Z.; Shu, H.; Zheng, C. Multi-Scale Single Image 
Dehazing Using Laplacian and Gaussian Pyramids. 
IEEE Trans Image Process 2021, 30: 9270-9279. 
https://doi.org/ 10.1109/TIP.2021.3123551 
[30] Zhu, Z.; Wei, H.; Hu, G.; Li, Y.; Qi, G.; Mazur, N. 
A Novel Fast Single Image Dehazing Algorithm 
Based on Artificial Multiexposure Image Fusion. 
IEEE Transactions on Instrumentation and 
Measurement 2021, 70, 1-23. 
https://doi.org/ 10.1109/tim.2020.3024335 
[31] Huang, Y.; Niu, B.; Guan, H.; Zhang, S. Enhancing 
Image Watermarking With Adaptive Embedding 
Parameter and PSNR Guarantee. IEEE Transactions 
on Multimedia 2019, 21(10), 2447-2460. 
https://doi.org/ 10.1109/tmm.2019.2907475 
[32] Tang, Y. M.; Ren, F. J.; Pedrycz, W. Fuzzy C-Means 
clustering through SSIM and patch for image 
segmentation. APPLIED SOFT COMPUTING 2020, 
87: 105928-105943. 
https://doi.org/ 10.1016/j.asoc.2019.105928 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 

62 Informatica 47 (2023) 51–62 C. Su et al.