0% found this document useful (0 votes)
70 views36 pages

Fractional Derivative Based Unsharp Masking Approa

Copyright
© © All Rights Reserved
We take content rights seriously. If you suspect this is your content, claim it here.
Available Formats
Download as PDF, TXT or read online on Scribd
0% found this document useful (0 votes)
70 views36 pages

Fractional Derivative Based Unsharp Masking Approa

Copyright
© © All Rights Reserved
We take content rights seriously. If you suspect this is your content, claim it here.
Available Formats
Download as PDF, TXT or read online on Scribd
You are on page 1/ 36

Multimedia Tools and Applications

https://doi.org/10.1007/s11042-020-09795-5

Fractional derivative based Unsharp masking approach


for enhancement of digital images

Kanwarpreet Kaur 1 & Neeru Jindal 1 & Kulbir Singh 1

Received: 20 November 2019 / Revised: 13 August 2020 / Accepted: 2 September 2020

# Springer Science+Business Media, LLC, part of Springer Nature 2020

Abstract
Image visual quality is severely degraded due to various environmental conditions, thus,
leading to the loss in image details. Therefore, an image enhancement approach is
required to improve the visual quality of images. In this paper, Unsharp Masking (UM)
approach based on Riemann-Liouville (RL), Grunwald-Letnikov (GL), and Riesz frac-
tional derivatives is proposed for the image enhancement. The fractional derivatives
based UM approach sharpened the edges of an image while preserving its low and
medium frequency details. Furthermore, the extra parameter of fractional derivative
provides an additional degree of freedom, thus, increasing the effectiveness of the
proposed approach. Extensive simulations carried out on several standard images of
different sizes validated the performance of proposed approach in comparison to the
existing techniques. The capability of the proposed approach is further confirmed by
considering the test images with varying illumination conditions. Moreover, the compar-
ative analysis performed in terms of quantitative measures such as Information Entropy
(IE), Average Gradient (AG), Measure of Enhancement (EME), etc. confirmed that the
proposed UM approach based on Riesz fractional derivative outperforms the existing
state-of-the-art image enhancement techniques. Furthermore, the potential of the pro-
posed approach is validated by considering its application in the medical images.

Keywords Average gradient . Fractional derivative . Information entropy . Measure of


enhancement . Unsharp masking

* Kulbir Singh
[email protected]
Kanwarpreet Kaur
[email protected]

Neeru Jindal
[email protected]

1
Department of Electronics and Communication Engineering, Thapar Institute of Engineering and
Technology, Patiala, Punjab, India

Content courtesy of Springer Nature, terms of use apply. Rights reserved.


Multimedia Tools and Applications

1 Introduction

Image enhancement plays a significant role in various fields of image processing such as image
restoration, image compression, medical imaging, underwater imaging, remote sensing, etc. [29,
32, 40, 43–45]. The image enhancement techniques are primarily used to smooth the irregularities
present in the original image with a minimal change in the image information. Moreover, it leads
to the sharpening of features such as contrast, edges, texture, boundaries thus increasing their
dynamic range for easier detection [32]. Since human perception is extremely sensitive to the
edges present in an image. Therefore, any kind of reduction in high-frequency components leads
to the degradation of visual quality of an image. Hence, it is also necessary to enhance the edges of
an image. Generally, image enhancement techniques such as Unsharp Masking (UM), high boost
filtering, etc. are considered to improve the visual quality of edges in addition to the contrast and
texture of an image. This is also referred to as image sharpening [9]. The image enhancement
techniques are further classified into spatial and transform domains. In the spatial domain,
operations are performed on the pixel itself whereas, in the transform domain, image is trans-
formed into the frequency domain using the Discrete Fourier Transform (DFT), Discrete Cosine
Transform (DCT), Discrete Sine Transform (DST), Fractional Fourier Transform (FrFT), etc. [7,
9, 10, 12, 19, 26]. Both these domains involve UM, Histogram Equalization (HE), high-boost,
homomorphic, and high-frequency emphasis based filtering techniques for image enhancement.
Although several techniques exist for the sharpening of images but UM is extensively used for the
sharpening of edges in an image. Hence, the proposed technique utilized the UM based technique
for image enhancement. In the conventional UM technique, the image is filtered through a high
pass filter. The resultant (filtered) image is then scaled and added to the original image to obtain
the sharpened image [7, 9, 23] as shown in Fig. 1.
In [23], the UM technique is used for the enhancement of medium contrast details. It
utilized the two-directional filters whose coefficients are updated on the basis of Gauss-
Newton scheme. But it results in the noise due to some transient effects. In [1], the Quadratic
Weighted Median Filter (QWM) is used for the enhancement of edges that provides superior
performance in comparison to the Quadratic Volterra (QV) methods. It is mainly focussed on
the enhancement of bright regions. UM is used in combination with the optimization tech-
niques for the enhancement of images in [16, 17]. But, the number of iterations and filter
parameters involved increased its complexity. A three-step approach is suggested in [14] to
sharpen the images. The edge pixels are selected using Horizontal and Vertical Differentiator
(HVD) so that isolated pixels should not be considered as edges. The pixels which are detected
as edge pixels are adjusted adaptively while the non-edge pixels are kept unchanged. This

Input + Output
Image Image

Resultant
Image
High Pass
Filter X

Scaling Factor (α)


Fig. 1 Conventional unsharp masking technique [23]

Content courtesy of Springer Nature, terms of use apply. Rights reserved.


Multimedia Tools and Applications

technique avoided over-sharpening but led to the loss of image information. In [28], Median-
Mean based Sub-Image-Clipped Histogram Equalization (MMSICHE) is used for image
enhancement. However, it can be used only for images with substantial peaks in the histogram.
Moreover, the Gaussian filter is used for image sharpening in [11, 13]. In [13], HE is used in
combination with UM for image sharpening. However, in [11] Gaussian based UM technique
is used in medical images for sharpening of the image edges. Although these techniques
provided better results for contrast enhancement but the Gaussian filter is not able to preserve
the information of an image. Hence, there is a need to preserve the information of an image
along with its enhancement. In [42], blurriness-guided UM (BUM) technique is used to
perform the pixel-wise enhancement that separately processed the base and detail layer.
Although it prevented over-enhancement artifacts but the effectiveness of the BUM approach
is evaluated only on the basis of subjective performance. Moreover, fusion techniques are also
used to enhance the underexposed regions of the image. Such techniques are usually employed
in the weakly illuminated images or night images [43]. Nevertheless, most of these techniques
used high pass filters that led to distortion because of their sensitivity to noise. Moreover, these
integer (nth) order derivative based image enhancement techniques result in the degradation of
the image quality when applied to the smooth areas. Therefore, fractional derivative masks are
used for the enhancement of images as they possess the ability to preserve the low-frequency
features in the smooth areas while retaining the high-frequency features along with enhance-
ment of the medium-frequency texture details [41]. The fractional derivatives such as
Riemann-Liouville (RL), Grunwald-Letnikov (GL), Riesz, etc. are used for the enhancement
of low contrast, medical, foggy, and texture images [3, 5, 6, 8, 15, 24, 25, 31, 34]. RL based
non-linear mask is used for the enhancement of image texture [5]. In [24], RL and GL based
six fractional derivative masks are analyzed for the application of image enhancement. Based
on the error analysis of these fractional masks, the performance of YiFeiPU−2 is considered to
be better in comparison to other masks. An improved GL based fractional derivative mask is
suggested in [6] to enhance the textural details. The enhancement is dependent on the order of
fractional derivative as well as the intensity factor. In [19], matrix Riesz fractional order
differentiator is used in combination with DST for the sharpening of images. However,
Savitzky-Golay differentiator is used to enhance the low contrast and dark images [34]. But,
the problem of over-enhancement and distortion exists in these algorithms. In [25], the Riesz
fractional derivative based mask is used to enhance the images of license plates to aid in easy
detection and recognition. These enhanced images improved the rate of recognition of license
plates. In [8], GL fractional derivative modified with autocorrelation function is used for the
enhancement of texture of images. It is also used to enhance satellite and medical images.
However, the low contrast and non-uniformly illuminated images are enhanced on the basis of
fractional derivative based homomorphic filtering [15]. The fractional differential UM and
Piecewise Gamma Correction (PGC) based mask is used for the enhancement of images. But,
it involved the selection of several parameters [31]. However, the problem of over-illumination
exists in some of these algorithms. Furthermore, most of the existing techniques mainly focus
on contrast and texture enhancement. Still, there is a scope of enhancement of edges while
preserving the image information. Therefore, fractional derivatives based UM technique is
proposed in this paper for the sharpening of image features while preserving its details. The
contributions of this paper are as follows:

& A fractional derivative based Unsharp Masking (UM) approach is proposed for the
enhancement of visual quality of an image.

Content courtesy of Springer Nature, terms of use apply. Rights reserved.


Multimedia Tools and Applications

& Comparative analysis of RL, GL, and Riesz fractional derivative based UM is done in
terms of various performance metrics, such as Information Entropy (IE), Average Gradient
(AG), Structural Similarity Index Measure (SSIM), Measure of Enhancement (EME), etc.
& The effectiveness of the proposed approach is confirmed by performing its comparison
with state-of-the-art existing techniques.
& The efficacy of the proposed approach is further validated by considering the varying
illumination conditions.
& The applicability of the proposed approach is confirmed on the application of Fundus
images.

The paper is organized as follows: Section 2 described the preliminaries of fractional deriv-
atives. Section 3 depicted the proposed UM technique based on the fractional derivative mask.
The performance metrics used for the evaluation of the proposed approach are discussed in
Section 4. Simulation results are discussed in Section 5. The limitations of the proposed
approach are described in Section 6. The conclusion and future scope of the proposed
approach are eventually presented in Section 7.

2 Preliminaries

Fractional order calculus (FOC) is the generalization of integer order calculus. It possesses the
capability to model systems more accurately in comparison to the integer orders.

2.1 Fractional derivatives

The fractional derivatives such as RL, GL, Caputo, and Riesz are generally used in various
applications of image and signal processing. However, the proposed work considered RL, GL,
and Riesz fractional derivative for image enhancement. These fractional derivatives are
defined as follows:

a. Riemann–Liouville (RL): RL derivative for a function z(t) is given as [30]:


 m
μ 1 d t zðτ Þ
D zðtÞ ¼ ∫a dτ ð1Þ
Γðm−μÞ dt ðt−τ Þ1−ðm−μÞ
where, a and t are lower and upper limits of the integration, respectively. μ ∈ R+ (real numbers)
such that m − 1 < μ < m, m is the operation order. Γ(.) is the Euler’s gamma function and
convolution is interpreted at all the instances from τ = 0 to τ = t.

b. Grünwald–Letnikov (GL): The GL based derivative of a function z(t) is given as [30]:

1 ð t−ah Þ Γðμ þ k Þ
μ
D zðt Þ ¼ lim μ ∑ zðt−khÞ ð2Þ
h→0 ΓðμÞh k¼0 Γðk þ 1Þ

 t−a   t−a 
where h is the sampling period, such that h is an integer and k ranges from 0 to h .

Content courtesy of Springer Nature, terms of use apply. Rights reserved.


Multimedia Tools and Applications

c. Riesz: The Riesz derivative of a function z(t) is given as [19, 21]:

 
1 d ∞ zðτ Þ
Dμ zðtÞ ¼ ∫ dτ ð3Þ
2 cosðμπ=2ÞΓð1−μÞ dt −∞ ðt−τ Þμ−1

Here, the approximation of Type-I Riesz fractional derivative is done with step h according to
the second order fractional centered difference model defined as [21]:

Γðμ þ 1Þ ∞ ð−1Þk
Dμ zðtÞ ¼ lim μ ∑ zðt−khÞ ð4Þ
h→0 h k¼−∞ Γðμ=2−k þ 1ÞΓðμ=2 þ k þ 1Þ

2.2 Amplitude and phase response of fractional derivatives

The fractional operators behave as a fractional derivative for μ > 0 and fractional integrators for
μ < 0 whereas an all-pass filter for μ = 0. However, fractional derivatives are considered in this
paper as they enhanced the high-frequency components of signal non-linearly while preserving
the low-frequency components [24]. Therefore, the amplitude and phase response of fractional
derivatives is described on the basis of their Fourier Transform (FT). Initially, the FT of GL
fractional derivative [24] of the square signal z(t) is defined as:

Dμ ½zðtÞ⇔FT ðDμ zÞðωÞ ¼ ðjωÞμ zðωÞ ¼ jωjμ θðωÞzðωÞ ð5Þ

 
jμπ
ðDμ zÞðωÞ ¼ jωjμ :exp sgnðωÞ zðωÞ ð6Þ
2

Similarly, FT of Riesz fractional derivative of the square signal z(t) is defined as:

ðDμ zÞðωÞ ¼ jωjμ zðωÞ ð7Þ

It is observed that the Riesz fractional derivative has zero phase while GL fractional derivative
 
has a constant phase of exp jμπ2 sgnðωÞ that results in the image blurriness distortion [37]. The
amplitude and phase frequency response of fractional derivatives is shown in Fig. 2. It is
observed from Fig. 2a that amplitude decreases with the increase in fractional order for the
frequency range 0 < ω < 1 while for ω > 1, the amplitude rises with an increase in the frequency
when fractional order increases. Figure 2b shows that there is no phase shift for Riesz
fractional derivative whereas, a phase shift occurs in the case of GL in accordance with the
fractional orders.

3 Proposed approach

Although the existing UM techniques enhanced the images but they are not able to sufficiently
preserve the information present in the images. Hence, in order to preserve information details
during the process of image enhancement, fractional derivatives are used in combination with
UM. The proposed technique is intended to achieve edge enhancement in addition to the

Content courtesy of Springer Nature, terms of use apply. Rights reserved.


Multimedia Tools and Applications

(a) (b)
Fig. 2 Amplitude and Phase response of fractional derivatives

improvement in contrast and texture of an image. The framework of the proposed technique is
presented in Fig. 3.
In the proposed approach, a fractional derivative mask is utilized in addition to the
Laplacian mask for the sharpening of image features. Initially, the convolution of input image
z(i, j) is obtained with the discrete Laplacian masks (ML) [7] defined by:

∇2 z ¼ zði þ 1; j þ 1Þ þ zði þ 1; jÞ þ zði þ 1; j−1Þ þ zði; j þ 1Þ þ zði; j−1Þ


þ zði−1; j þ 1Þ þ zði−1; jÞ þ zði−1; j−1Þ−8zði; jÞ ð8Þ

Input Output
Image Image
( , ) ( , )
Laplacian
Mask ( ) X
( , )
( , )
Scaling
Factor ( ) ±
Resultant
Image
Fractional Derivative ( , )
X
Mask ( )
( , ) 2 ( , )

Fractional Scaling
Derivative Discretization of Factor ( )
(µ) Fractional
Derivative

Fig. 3 Framework of the proposed fractional derivative based UM technique

Content courtesy of Springer Nature, terms of use apply. Rights reserved.


Multimedia Tools and Applications

∇2 z ¼ −zði þ 1; j þ 1Þ−zði þ 1; jÞ−zði þ 1; j−1Þ−zði; j þ 1Þ−zði; j−1Þ−zði−1; j þ 1Þ ð9Þ

−zði−1; jÞ−zði−1; j−1Þ þ 8zði; jÞ

Similarly, the convolution of input image z(i, j) is obtained with the fractional derivative mask
(MFD). In the proposed technique, an isotropic mask that is invariant to the increment in the
rotations of 45° [7] is obtained using the discretized mask coefficients of RL, GL, and Riesz
fractional derivatives. Here, the discrete RL fractional derivative is computed numerically by
dividing the interval from “a” to “t” among n equal parts, such that n is very large. The RL
fractional mask coefficients [5, 24] are obtained as follows:

∂μ zði; jÞ zði−k; jÞ n−1  1−μ 1−μ



z ði; jÞ≅ þ ∑ ð k þ 1Þ −2k 1−μ
þ ðk−1Þ
∂iμ Γð2−μÞ Γð2−μÞ k¼1
zði−n; jÞ  
þ ðn−1Þ1−μ −n1−μ þ ð1−μÞn−μ ð10Þ
Γð2−μÞ

∂μ zði; jÞ zði; j−k Þ n−1  


μ zði; jÞ≅ þ ∑ ðk þ 1Þ1−μ −2k 1−μ þ ðk−1Þ1−μ
∂j Γð2−μÞ Γð2−μÞ k¼1
zði; j−nÞ  
þ ðn−1Þ1−μ −n1−μ þ ð1−μÞn−μ ð11Þ
Γð2−μÞ

when k → n = 2m − 1, the size of the mask is (2m + 1) × (2m + 1), n is usually an odd number.
Similarly, the GL fractional mask coefficients [24] are obtained from Eq. (2) as follows:

   
∂μ μ μ2 μ2 μ3
zð i; j Þ≅ þ zð i þ 1; jÞ þ 1− − zði; jÞ
∂iμ 4 8 2 8
      
zði−k; jÞ n−2 Γðk þ 1−μÞ μ μ2 Γðk−μÞ μ2 Γðk−1−μÞ μ2 μ
þ ∑ þ þ 1− þ −
Γð−μÞ k¼1 Γðk þ 2Þ 4 8 Γðk þ 1Þ 4 Γðk Þ 8 4
    
zði−n þ 1; jÞ Γðn−1−μÞ μ 2
Γðn−2−μÞ μ μ 2
þ 1− þ − þ
Γð−μÞ ΓðnÞ 4 Γðk Þ 4 8
 
Γðn−μ−1Þ μ μ2
þ − þ zði−n; jÞ
ΓðnÞΓð−μÞ 4 8
ð12Þ

Content courtesy of Springer Nature, terms of use apply. Rights reserved.


Multimedia Tools and Applications

   
∂μ μ μ2 μ2 μ3
zð i; jÞ≅ þ zði; j þ 1 Þ þ 1− − zði; jÞ
∂jμ 4 8 2 8
      
zði; j−k Þ n−2 Γðk þ 1−μÞ μ μ2 Γðk−μÞ μ2 Γðk−1−μÞ μ2 μ
þ ∑ þ þ 1− þ −
Γð−μÞ k¼1 Γðk þ 2Þ 4 8 Γðk þ 1Þ 4 Γðk Þ 8 4
    
zði; j−n þ 1Þ Γðn−1−μÞ μ 2
Γðn−2−μÞ μ μ 2
þ 1− þ − þ
Γð−μÞ ΓðnÞ 4 Γðk Þ 4 8
 
Γðn−μ−1Þ μ μ2
þ − þ zði; j−nÞ
ΓðnÞΓð−μÞ 4 8
ð13Þ
when k → n = 2m − 1, the size of the mask is (2m + 1) × (2m + 1), n is usually an odd number.
The Riesz fractional mask coefficients are also obtained in a similar manner as RL and GL
mask coefficients from Eq. (4):
 
∂μ 2Γð1 þ μÞ μ μ2
zði; jÞ≅ þ zði þ 1; jÞ
∂iμ ðΓððμ=2Þ þ 1ÞÞ2 4 8
"    #
2Γð1 þ μÞ μ2 2Γð1 þ μÞ μ μ2
þ 1− − þ zði; jÞ þ …
ðΓððμ=2Þ þ 1ÞÞ2 4 Γðμ=2ÞΓððμ=2Þ þ 2Þ 4 8
"  2 
ð−1Þn−1 2Γð1 þ μÞ μ μ
þ − ð14Þ
Γððμ=2Þ−n þ 2ÞΓððμ=2Þ þ nÞ 8 4
 
ð−1Þn 2Γð1 þ μÞ μ2
þ 1−
Γððμ=2Þ−n þ 1ÞΓððμ=2Þ þ n þ 1Þ 4
 #
ð−1Þnþ1 2Γð1 þ μÞ μ μ2
− þ zði−n; jÞ
Γððμ=2Þ þ n þ 2ÞΓððμ=2Þ−nÞ 4 8

 
∂μ 2Γð1 þ μÞ μ μ2
zði; jÞ≅ þ zði; j þ 1Þ
∂ jμ ðΓððμ=2Þ þ 1ÞÞ2 4 8
"    #
2Γð1 þ μÞ μ2 2Γð1 þ μÞ μ μ2
þ 1− − þ zði; jÞ þ …
ðΓððμ=2Þ þ 1ÞÞ2 4 Γðμ=2ÞΓððμ=2Þ þ 2Þ 4 8
"  2 
ð−1Þn−1 2Γð1 þ μÞ μ μ
þ − ð15Þ
Γððμ=2Þ−n þ 2ÞΓððμ=2Þ þ nÞ 8 4
 
ð−1Þn 2Γð1 þ μÞ μ2
þ 1−
Γððμ=2Þ−n þ 1ÞΓððμ=2Þ þ n þ 1Þ 4
nþ1  #
ð−1Þ 2Γð1 þ μÞ μ μ 2
− þ zði; j−nÞ
Γððμ=2Þ þ n þ 2ÞΓððμ=2Þ−nÞ 4 8

Content courtesy of Springer Nature, terms of use apply. Rights reserved.


Multimedia Tools and Applications

The resultant images ZL(i, j) and ZFD(i, j) obtained after convolution with Laplacian and
fractional derivative masks are then scaled by the scaling factors of α1 and α2, respectively.
The scaling factors lie in the range of 0.2–0.7 [32]. The scaling factor α1 is considered to be
less than α2 because the Laplacian mask deemphasizes the slowly varying gray levels while
emphasizing the discontinuities present in the image. Thus, the scaled images are then, added
or subtracted, depending on the center coefficient of the Laplacian mask as in Eq. (8) and Eq.
(9). If the center coefficient of the mask is positive as in Eq. (9), then, scaled images will be
added, otherwise, they will be subtracted [7]. The resultant image thus obtained after scaling is
given as:
Z R ði; jÞ ¼ α1 Z L ði; jÞ  α2 Z FD ði; jÞ ð16Þ

The resultant image ZR(i, j) is added to the original input image to obtain the enhanced image
Z(i, j) given as:
Z ði; jÞ ¼ Z R ði; jÞ þ zði; jÞ ð17Þ

The pseudo-code for the proposed algorithm is given below:

4 Performance metrics

The performance of the proposed approach is evaluated in the terms of performance metrics
such as Information Entropy (IE), Average Gradient (AG), Structural Similarity Index Mea-
sure (SSIM), Measure of Enhancement (EME), Correlation Coefficient (CC), and Gray Level
Co-occurrence Matrix (GLCM) parameters. IE is used to compute the amount of information
present in the sharpened image. It is computed on the basis of Shannon’s information as [6]:

255
E ¼ − ∑ Pi log 2 Pi ð18Þ
i¼0

Content courtesy of Springer Nature, terms of use apply. Rights reserved.


Multimedia Tools and Applications

where E is entropy and Pi is i’s probability in the image. Then, AG is computed to indicate the
clarity of an image. It is determined by [6]:
vffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
u   
u ∂Z k ðx;yÞ 2 þ ∂Z k ðx;yÞ 2
1 P−1 Q−1 t ∂x ∂y
AG ¼ ∑ ∑ ð19Þ
ðP−1ÞðQ−1Þ x¼1 y¼1 2

where,Zk(x, y) is the image’s pixel value at location (x, y). Moreover, the enhancement in an
image is also determined in the terms of SSIM and EME. The SSIM is given by [39]:

  
2μx μy þ K 1 2σxy þ K 2
SSIM ðx; yÞ ¼    ð20Þ
μ2x þ μ2y þ K 1 σ2x þ σ2y þ K 2

where,μx and μy represent the mean intensities, σx and σy represent the contrast while K1 and
K2 represent the constants. It measures the similarity between the enhanced image and the
original image. Furthermore, the enhancement measure EME is determined by dividing an
image Z(x, y) into r1and r2blocks of Wr, s(i, j). It is defined as [22]:

r r   
1 1 2 Z max;r;s
EME ¼ ∑ ∑ 20 ln ð21Þ
r1 r2 s¼1 r¼1 Z min;r;s

where, Zmax; r, s and Zmin; r, s are the maximum and minimum of the image within the block Wr,
s(i, j). To further confirm the efficacy, CC is also computed as [36]:

  
∑Dii¼1 x i −x y i −y
CC ¼ rffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
 2  2ffi ð22Þ
∑i¼1 xi −x ∑i¼1 yi −y
Di Di

where, xi and yi are the values of adjacent pixels while Di refers to the sum of adjacent pixel
pairs along each direction. It determines the relationship between the two images in a precise
manner [33]. Furthermore, the enhancement of the proposed approach is also confirmed from
the GLCM parameters. GLCM is the matrix arrangement that gives various combinations of
pixel values in an image. It shows the relationship between two pixels when they are spatially
related to the distance in different directions. Various GLCMs can be formed by varying the
distance among the pixels in a pair and at different rotation offsets such as 0°, 45°, 90°, 135°
[8]. The parameters computed from the GLCM are contrast, correlation, energy, and homo-
geneity. The contrast shows the intensity that exists between the pixel and its neighbor in the
image. Similar pixel values result in low contrast. Correlation provides the value by which one
pixel is correlated with its neighbor pixels across the image. The decrease in the correlation
indicates more clarity. Energy indicates the uniformity in an image. Homogeneity is defined as
the closeness that exists in the spatial distribution of gray levels over the complete image. The
decrease in the value of energy and homogeneity means the enhancement of an image [8].

Content courtesy of Springer Nature, terms of use apply. Rights reserved.


Multimedia Tools and Applications

Lena Surface Moon Baboon Barbara Pirate


(256×256) (256×256) (256×256) (512×512) (512×512) (512×512)

Bridge Lighthouse Truck Airplane Bark Wall


(512×512) (512×512) (512×512) (512×512) (512×512) (512×512)
Fig. 4 Test images used for simulation

5 Simulation results

The performance of the proposed approach is validated by considering the standard grayscale
images of different sizes from USC-SIPI and LIVE datasets [27, 35] as shown in Fig. 4. The
extensive simulation work is carried out using MATLAB R2016a on Intel® CPU 2.7 GHz
processor with 8 GB RAM. The performance analysis is done in the form of both qualitative
and quantitative measures.

5.1 Effect of various masks in Unsharp masking

In this section, the contribution of Laplacian and fractional derivative masks in the UM
technique is considered. Initially, the edge information obtained with the Laplacian mask
and various fractional derivative masks for Lena and Barbara images are shown in Fig. 5.
It is observed that more information is attained using the GL and Riesz fractional derivative.
This edge information is further used for the enhancement of an image. Moreover, the
contribution of each mask in the UM technique is evaluated in terms of both qualitative and
quantitative measures. Here, the scaling factor of α1 is considered for the Laplacian mask
while the scaling factor of α2 is considered for the fractional derivative masks. The contribu-
tion of each mask in the image enhancement is illustrated in Fig. 6 and Table 1.

Images/ Laplacian RL GL Riesz


Technique

Lena

Barbara

Fig. 5 Edge information obtained using various masks

Content courtesy of Springer Nature, terms of use apply. Rights reserved.


Multimedia Tools and Applications

Images/
Laplacian RL GL Riesz
Technique

Lena

Barbara

Fig. 6 Image enhancement obtained using various masks in UM

It is observed from Fig. 6 that fractional derivative masks also improved the contrast and
texture of the image in addition to sharpening of its edges while the Laplacian mask enhanced
the fine details present in the image. However, more enhancement is achieved in the case of
GL and Riesz fractional derivative mask.

5.2 Performance analysis of proposed approach

In this section, the comparative analysis of the proposed approach is done with respect to the
existing state-of-the-art image enhancement techniques i.e. Conventional UM technique
(CUM) [7], Unsharp Masking Filter Kernel and Gain using Particle Swarm Optimization
(UMFKG) [16], Median-Mean Based Sub-Image-Clipped Histogram Equalization
(MMSICHE) [28], Contrast Enhancement using Exposure Fusion (CEEF) [43], Unsharp
Masking and Histogram Equalization based contrast enhancement (UMHE) [13], Riesz based
UM (RUM) [19], Riesz based enhancement (Riesz) [25], Modified GL (Mod GL) [8], and
Homomorphic Filtering based on fractional derivatives (HF) [15]. After extensive simulations,
the values of scaling factors α1 and α2 in the proposed approach are considered to be 0.2 and
0.21–0.35 respectively. The test images enhanced by the proposed approach with RL, GL, and
Riesz fractional derivatives are illustrated in Fig. 7.
It is observed from Fig. 7 that enhanced images possess sharp edges in addition to contrast
and texture enhancement. This is due to the fact that the proposed approach utilized the
advantages of fractional derivatives as well as the Laplacian mask. Hence, the low-frequency
components are preserved while medium and high-frequency components are enhanced by the
proposed technique, thus, improving its visual quality. Furthermore, the simulation results
obtained for the comparative analysis of various image enhancement techniques for the Lena
and Barbara images are shown in Fig. 8.
It is observed from the zoomed portion of the images that the proposed technique
provides better contrast, texture, and sharp edges in comparison to the existing image
enhancement techniques. The proposed approach is considered for all the fractional orders
varying from 0 to 1. However, it is observed from the qualitative and quantitative
measures that images are more enhanced for the fractional orders ranging from 0.4–0.5,
0.82–0.96, and 0.71–0.95 for RL, GL, and Riesz fractional derivative respectively.
Therefore, this range of fractional orders is considered to further confirm the efficacy of
the proposed approach. IE and AG obtained for various image enhancement techniques are
shown in Tables 2 and 3.

Content courtesy of Springer Nature, terms of use apply. Rights reserved.


Multimedia Tools and Applications

Table 1 Performance metrics for various masks in UM

Images Metrics/Mask IE AG SSIM EME CC Contrast Correlation Energy Homogeneity

Lena Laplacian 7.6068 10.2281 0.9236 11.6497 0.9864 1.0240 0.8086 0.0836 0.7915
RL 7.6383 9.1449 0.9627 8.2607 0.9994 0.8360 0.8714 0.0797 0.8168
GL 7.6678 12.1188 0.8566 11.7118 0.9904 1.2921 0.8065 0.0638 0.7609
Riesz 7.6678 12.1190 0.8566 11.7120 0.9904 1.2922 0.8065 0.0638 0.7609
Barbara Laplacian 7.5979 12.8985 0.9172 12.2666 0.9718 1.8299 0.6486 0.0659 0.7435
RL 7.7197 12.3259 0.9324 9.7718 0.9967 1.5873 0.7681 0.0620 0.7687
GL 7.7525 15.9843 0.8132 12.2506 0.9714 2.3775 0.6473 0.0476 0.7092
Riesz 7.7556 16.3559 0.7946 12.4987 0.9706 2.4431 0.6412 0.0460 0.7015

Bold signifies the optimum parameters

Content courtesy of Springer Nature, terms of use apply. Rights reserved.


Multimedia Tools and Applications

Proposed approach using RL fractional derivative

Lena Surface Moon Baboon Barbara Pirate


(256×256) (256×256) (256×256) (512×512) (512×512) (512×512)

Bridge Lighthouse Truck Airplane Bark Wall


(512×512) (512×512) (512×512) (512×512) (512×512) (512×512)
Proposed approach using GL fractional derivative

Lena Surface Moon Baboon Barbara Pirate


(256×256) (256×256) (256×256) (512×512) (512×512) (512×512)

Bridge Lighthouse Truck Airplane Bark Wall


(512×512) (512×512) (512×512) (512×512) (512×512) (512×512)
Proposed approach using Riesz fractional derivative

Lena Surface Moon Baboon Barbara Pirate


(256×256) (256×256) (256×256) (512×512) (512×512) (512×512)

Bridge Lighthouse Truck Airplane Bark Wall


(512×512) (512×512) (512×512) (512×512) (512×512) (512×512)
Fig. 7 Images enhanced using the proposed approach with different fractional derivatives

It is observed that IE and AG are high for the proposed approach using Riesz fractional
derivative in comparison to the existing techniques because it is able to preserve the informa-
tion details of an image without introducing any blurriness distortion. The performance metrics
for Lena and Bark images are almost comparable for GL and Riesz fractional derivatives.
Tables 4, 5 and 6 show the SSIM, EME, and CC of test images for various image enhancement
techniques.

Content courtesy of Springer Nature, terms of use apply. Rights reserved.


Multimedia Tools and Applications

CUM [7] UMFKG [16] MMSICHE [28] CEEF [43] UMHE [13] RUM [19]

Riesz [25] Mod GL [8] HF [15] RL (µ=0.43) Riesz (µ=0.71) GL (µ=0.83)

CUM [7] UMFKG [16] MMSICHE [28] CEEF [43] UMHE [13] RUM [19]

Riesz [25] Mod GL [8] HF [15] RL (µ=0.43) Riesz (µ=0.95) GL (µ=0.96)


Fig. 8 Enhanced Lena and Barbara images using various image enhancement techniques

It is observed that the proposed approach has minimum SSIM and CC in comparison to
existing state-of-the-art enhancement techniques. This is because the fractional derivative
results in the change in the structure of an image after the enhancement. It is ascertained from
Table 5 that the proposed approach provides maximum EME due to contrast, texture as well as
edge enhancement. Moreover, image enhancement is further confirmed on the basis of GLCM
parameters. Tables 7, 8, 9 and 10 shows the contrast, correlation, energy, and homogeneity
based on GLCM with rotation offset that provided maximum enhancement.
It is noticed from Tables 7, 8, 9 and 10 that Riesz derivative at rotation offset of 45° and
135° provides the best results for all the images. It is inferred that the proposed approach using
Riesz derivative has more contrast and less correlation, energy, and homogeneity, thus
providing a better enhancement in comparison to existing techniques. This is due to the
increase in contrast as well as the texture of images at different rotation offsets.

5.3 Simulation results for images with varying illumination conditions

The enhancement of the proposed approach is further validated by considering the images with
varying illumination conditions. Since, the images are likely to be affected by various
uncontrolled features such as illumination, blurring, etc. during image acquisition that may
affect the further processing of an image. Therefore, an uncontrolled feature of varying
illumination conditions is considered in this paper using the images from VIP Illumination
Saliency Dataset [4, 38] as shown in Fig. 9. Figure 10 shows the simulation results for test
images with different illumination conditions for various techniques.

Content courtesy of Springer Nature, terms of use apply. Rights reserved.


Table 2 IE for the test images of different sizes

Images Original Image CUM UMFKG MMSI-CHE CEEF UMHE RUM Riesz Mod HF Proposed
[7] [16] [28] [43] [13] [19] [25] GL [8] [15]
RL Mask GL Mask Riesz Mask

Lena 7.4429 7.6019 7.7340 7.3971 7.4504 7.4492 7.6627 7.7402 7.5705 7.4673 7.7124 7.7430 7.7430
Surface 6.8554 7.0484 7.0427 6.7430 6.9157 6.9213 7.2030 7.3009 6.8579 6.9134 7.1918 7.3899 7.4055
Moon 6.7093 6.8392 7.1469 6.6048 6.6759 6.7148 7.0614 7.1775 7.0195 6.7723 7.1172 7.3447 7.3796
Baboon 7.3579 7.6305 7.7350 7.2891 7.3474 7.3689 7.6793 7.7636 7.4420 7.4232 7.7583 7.8314 7.8337
Barbara 7.4664 7.6286 7.7476 7.4191 7.4778 7.4696 7.7226 7.7606 7.6494 7.4910 7.7743 7.8121 7.8143
Pirate 7.2367 7.4177 7.5952 7.1981 7.2887 7.2407 7.5193 7.5982 7.3303 7.5151 7.5886 7.6770 7.6838
Bridge 7.7056 7.7936 7.7782 5.7025 7.6385 7.7064 7.6992 7.8044 7.5192 7.6469 7.7444 7.8048 7.8054
Light-house 7.4486 7.6528 7.6413 7.3676 7.4763 7.4751 7.6017 7.6554 7.5545 7.5723 7.4552 7.6805 7.6831
Truck 6.0274 6.6873 6.8817 5.9905 6.5411 6.1638 6.9945 7.0559 6.6018 6.9060 6.7594 7.1443 7.1701
Airplane 4.0045 5.5424 5.5282 3.9918 5.1954 4.1203 5.5734 5.5953 5.5219 5.1235 5.6157 6.0204 6.0482
Bark 7.3250 7.5101 7.5659 7.2642 7.3311 7.4137 7.6060 7.6032 7.3852 7.3935 7.6261 7.6267 7.6268
Wall 6.7990 7.0348 7.1160 6.7255 6.7187 6.9260 7.1011 7.1915 6.9093 6.8684 7.1363 7.2846 7.2974

Bold indicates the maximum value


Multimedia Tools and Applications

Content courtesy of Springer Nature, terms of use apply. Rights reserved.


Table 3 AG for the test images of different sizes
Multimedia Tools and Applications

Images Orig-inal Image CUM UMF-KG MMSI-CHE CEEF UMHE RUM Riesz Mod HF Proposed
[7] [16] [28] [43] [13] [19] [25] GL [15]
[8] RL Mask GL Mask Riesz Mask

Lena 7.1832 10.1148 9.7550 9.7029 7.7851 8.2435 9.2005 11.4619 8.6481 7.3793 10.3526 13.4054 13.4056
Surface 9.8775 13.9778 13.5236 15.6769 10.7423 11.2205 12.9771 16.1438 10.6957 10.1489 14.3920 19.5904 20.0435
Moon 5.6510 8.5587 14.1615 9.9463 6.0477 7.3756 9.6091 10.0408 6.8945 6.0225 9.5849 16.2569 17.2708
Baboon 12.9654 19.8053 21.2802 17.9893 13.9961 16.8518 17.4212 22.5941 15.4728 13.6104 20.3441 26.4310 26.8313
Barbara 8.4663 12.8607 12.9902 11.0631 9.1745 10.9919 11.1771 14.5820 10.8475 9.8650 13.4656 17.9710 18.3447
Pirate 6.1580 8.6520 8.4069 8.8596 6.7007 7.4775 7.9639 9.8343 7.2562 5.9467 9.1319 13.0988 13.5690
Bridge 10.5495 15.3283 13.7734 11.3089 11.6497 13.1917 13.1646 16.9300 12.5553 11.2002 15.1605 18.3029 18.7373
Light-house 9.3854 13.7442 12.5901 11.9738 10.2298 11.7389 11.5873 14.9444 11.3056 10.0300 13.3052 17.3601 17.7846
Truck 5.0927 7.2589 8.7137 9.9063 5.7193 6.8054 7.6722 8.8718 6.1215 6.7042 7.4417 12.1634 12.7377
Airplane 2.3261 3.5418 4.4196 5.5000 2.3878 3.2327 3.3993 4.0282 3.2989 3.1913 3.9212 6.4669 6.8296
Bark 15.1307 20.1894 22.3644 19.5522 15.9436 18.0087 20.5395 22.6908 17.5507 15.9921 21.8454 22.8274 22.8298
Wall 9.6338 13.8782 15.6894 14.3319 10.0062 12.0076 13.7864 15.8666 11.5641 10.1362 14.6374 20.2767 20.8273

Bold indicates the maximum value

Content courtesy of Springer Nature, terms of use apply. Rights reserved.


Table 4 SSIM for the test images of different sizes

Images CUM UMF- MMSI- CEEF UMHE RUM Riesz Mod HF Proposed
[7] KG CHE [43] [13] [19] [25] GL [15]
[16] [28] [8] RL Mask GL Mask Riesz Mask

Lena 0.9522 0.9464 0.9016 0.9653 0.9889 0.9658 0.9040 0.9198 0.9869 0.9337 0.8164 0.8164
Surface 0.9477 0.9536 0.8550 0.9730 0.9897 0.9266 0.8353 0.9300 0.9943 0.9238 0.7830 0.7687
Moon 0.9287 0.6749 0.7852 0.9772 0.9656 0.9332 0.8105 0.8827 0.9925 0.8575 0.5996 0.5680
Baboon 0.9218 0.8797 0.8923 0.9729 0.9646 0.9409 0.8572 0.8845 0.9918 0.8964 0.7447 0.7315
Barbara 0.9454 0.9328 0.9061 0.9649 0.9757 0.9431 0.8923 0.9204 0.9871 0.9131 0.7722 0.7541
Pirate 0.9553 0.8487 0.8434 0.9586 0.9819 0.9606 0.9046 0.9197 0.9919 0.9236 0.7708 0.7499
Bridge 0.9357 0.9597 0.9437 0.9571 0.9723 0.9695 0.8909 0.8981 0.9925 0.9153 0.8396 0.8275
Lighthouse 0.9372 0.9524 0.9225 0.9499 0.9737 0.9752 0.8947 0.9031 0.9937 0.9212 0.7935 0.7777
Truck 0.9510 0.8730 0.7811 0.9552 0.9797 0.8920 0.8379 0.9236 0.9462 0.9254 0.7156 0.6902
Airplane 0.9698 0.8985 0.7183 0.9918 0.9862 0.9661 0.9393 0.9340 0.9544 0.9235 0.7477 0.7181
Bark 0.9442 0.9133 0.9266 0.9811 0.9793 0.9405 0.9077 0.9188 0.9935 0.9181 0.9013 0.9012
Wall 0.9255 0.8711 0.8733 0.9834 0.9659 0.9267 0.8600 0.8899 0.9921 0.8900 0.7192 0.7034

Bold indicates the minimum value


Multimedia Tools and Applications

Content courtesy of Springer Nature, terms of use apply. Rights reserved.


Table 5 EME for the test images of different sizes
Multimedia Tools and Applications

Images Orig- CUM UMF-KG MMSI- CEEF UMHE RUM Riesz Mod HF Proposed
inal Image [7] [16] CHE [43] [13] [19] [25] GL [15]
[28] [8] RL GL Mask RieszMask
Mask

Lena 7.5988 11.7998 15.0852 15.1044 6.3135 9.0491 8.5352 11.8629 8.1587 6.7211 12.6626 15.4038 15.4040
Surface 10.3999 13.6099 15.0332 13.6749 8.8243 11.4365 10.2131 11.5805 11.0271 7.7327 12.1952 16.0958 16.3562
Moon 5.6932 8.4424 15.5824 11.4027 4.9540 7.4565 5.9942 7.0382 7.6288 5.6372 9.6925 14.7492 15.5915
Baboon 12.2243 19.8341 20.8231 20.4840 10.1591 16.5744 13.1132 19.2651 16.7104 9.8709 18.1780 22.1086 22.2685
Barbara 8.1969 13.1465 15.1627 13.6365 7.0782 11.2084 8.6970 12.7210 11.3461 7.1324 11.9893 15.4760 15.7269
Pirate 6.4943 9.5625 17.0639 14.8360 5.5084 8.1253 7.0631 11.1024 8.9307 15.0906 12.4099 16.7491 17.0796
Bridge 12.1201 17.1966 17.5844 13.3870 10.2776 15.9569 13.9294 18.4133 16.2904 9.6357 15.9541 18.9212 19.1939
Light-house 9.6102 14.6467 14.3140 16.1355 8.0082 12.6556 11.0921 16.3762 13.5217 7.5413 13.0294 16.7967 17.0478
Truck 6.6122 9.9246 10.4002 12.6531 5.0606 7.8054 6.6151 9.5920 8.1539 4.8607 7.6330 12.4151 12.8309
Airplane 1.6193 2.5960 3.8532 4.3201 1.5653 2.4025 2.2921 3.4561 2.6254 2.4543 2.5117 4.2976 4.4810
Bark 13.6070 19.6062 20.3739 19.8707 11.4004 17.1136 16.5749 19.7857 16.8317 11.3276 19.4110 19.3108 20.4127
Wall 6.9450 11.2758 14.2894 14.1761 6.3368 9.9690 10.2075 13.8666 10.2836 6.8295 10.6477 14.3951 14.5965

Bold indicates the maximum value

Content courtesy of Springer Nature, terms of use apply. Rights reserved.


Table 6 CC for the test images of different sizes

Images CUM UMFKG MMSI- CEEF UMHE RUM Riesz Mod HF Proposed
[7] [16] CHE [43] [13] [19] [25] GL [15]
[28] [8] RL Mask GL Mask Riesz Mask

Lena 0.9905 0.9983 0.9923 0.9968 0.9961 0.9992 0.9904 0.9876 0.9974 0.9960 0.9840 0.9840
Surface 0.9828 0.9832 0.9754 0.9975 0.9933 0.9957 0.9824 0.9685 0.9983 0.9904 0.9567 0.9529
Moon 0.9860 0.9598 0.9643 0.9963 0.9941 0.9972 0.9869 0.9776 0.9991 0.9901 0.9414 0.9296
Baboon 0.9756 0.9736 0.9886 0.9945 0.9892 0.9989 0.9733 0.9617 0.9978 0.9882 0.9571 0.9542
Barbara 0.9823 0.9906 0.9915 0.9964 0.9920 0.9990 0.9853 0.9749 0.9988 0.9907 0.9671 0.9649
Pirate 0.9934 0.9995 0.9900 0.9965 0.9967 0.9996 0.9948 0.9901 0.9997 0.9972 0.9815 0.9791
Bridge 0.9882 0.9931 0.9837 0.9929 0.9944 0.9977 0.9794 0.9816 0.9992 0.9912 0.9788 0.9767
Lighthouse 0.9874 0.9929 0.9839 0.9932 0.9947 0.9982 0.9768 0.9806 0.9995 0.9920 0.9740 0.9717
Truck 0.9897 0.9942 0.9536 0.9936 0.9956 0.9995 0.9969 0.9802 0.9994 0.9973 0.9596 0.9532
Airplane 0.9910 0.9865 0.9584 0.9947 0.9956 0.9972 0.9802 0.9818 0.9990 0.9958 0.9620 0.9581
Bark 0.9859 0.9823 0.9886 0.9972 0.9947 0.9963 0.9814 0.9839 0.9987 0.9941 0.9813 0.9812
Wall 0.9823 0.9810 0.9801 0.9937 0.9921 0.9940 0.9794 0.9695 0.9986 0.9900 0.9598 0.9546

Bold indicates the minimum value


Multimedia Tools and Applications

Content courtesy of Springer Nature, terms of use apply. Rights reserved.


Table 7 GLCM based contrast measure for the images of different sizes along with its rotation offset
Multimedia Tools and Applications

Images Rot Offset Orig-inal Image CUM UMF MMSI- CEEF UMHE RUM Riesz Mod HF Proposed
[7] -KG CHE [43] [13] [19] [25] GL [15]
[16] [28] [8] RL Mask GL Mask Riesz Mask

Lena 135° 0.5413 1.0631 0.9781 0.8991 0.6287 0.7181 0.8369 1.2992 0.8588 0.7440 1.0645 1.5949 1.5950
Surface 45° 0.6845 1.2671 1.2422 1.5981 0.7858 0.9777 1.0401 1.5343 0.7689 0.7292 1.2984 2.1943 2.2652
Moon 45° 0.3276 0.6079 1.5098 0.7901 0.3597 0.4825 0.4836 0.7312 0.4409 0.4592 0.7122 1.7362 1.9138
Baboon 45° 1.1407 2.6729 3.0223 1.9661 1.3038 1.9616 1.9773 3.4066 1.5810 1.2756 2.7650 4.3480 4.4344
Barbara 135° 0.8133 1.9496 1.9144 1.2692 0.9243 1.4403 1.2961 2.2760 1.4815 0.9084 1.9769 3.1243 3.1970
Pirate 135° 0.3644 0.6775 0.5951 0.6365 0.4255 0.5160 0.5583 0.8202 0.5906 0.3349 0.7111 1.3031 1.3689
Bridge 135° 0.7406 1.5118 1.2581 0.8121 0.8680 1.1561 1.1169 1.8943 1.3027 0.8236 1.2829 2.1507 2.2404
Light-house 45° 0.8853 1.7575 1.5074 1.1274 0.9277 1.3450 1.2531 2.1027 1.4053 0.9934 1.5714 2.3572 2.4248
Truck 45° 0.2842 0.4752 0.7761 0.7093 0.3327 0.3913 0.4806 0.6243 0.4117 0.3907 0.4804 1.1120 1.2003
Airplane 135° 0.1638 0.2870 0.5294 0.3170 0.1813 0.2237 0.2694 0.4035 0.2173 0.2689 0.3901 0.6311 0.6598
Bark 135° 1.1531 2.1548 2.4548 1.7160 1.2858 1.6689 2.2591 2.5574 1.8398 1.2436 2.3326 2.5617 2.5624
Wall 135° 0.6500 1.2629 1.5703 1.3023 0.7178 0.9743 1.1832 1.6384 1.0676 0.6937 1.3846 2.5226 2.6377

Bold indicates the maximum value

Content courtesy of Springer Nature, terms of use apply. Rights reserved.


Table 8 GLCM based correlation measure for the images of different sizes with rotation offset

Images Rot Offset Orig-inal Image CUM UMF MMSI- CEEF UMHE RUM Riesz Mod HF Proposed
[7] -KG CHE [43] [13] [19] [25] GL [15]
[16] [28] [8] RL Mask GL Mask Riesz Mask

Lena 135° 0.8823 0.7954 0.8532 0.8758 0.8582 0.8495 0.8687 0.8064 0.8191 0.8481 0.8442 0.7785 0.7785
Surface 45° 0.6172 0.4660 0.4334 0.6211 0.6006 0.5258 0.6311 0.5620 0.5736 0.6249 0.5445 0.3941 0.3850
Moon 45° 0.7984 0.6782 0.4872 0.8015 0.7834 0.7264 0.8144 0.7737 0.7335 0.8004 0.7471 0.5174 0.4862
Baboon 45° 0.6853 0.4745 0.4926 0.6988 0.6339 0.5518 0.6559 0.4766 0.5135 0.6814 0.5706 0.4314 0.4237
Barbara 135° 0.8155 0.6337 0.7079 0.8215 0.7983 0.7055 0.8128 0.6795 0.7507 0.8598 0.7218 0.6065 0.5997
Pirate 135° 0.9177 0.8583 0.9113 0.9212 0.8961 0.8871 0.9163 0.8810 0.8587 0.9451 0.8998 0.8274 0.8197
Bridge 135° 0.8738 0.7764 0.8060 0.8709 0.8401 0.8189 0.8402 0.7363 0.7748 0.8715 0.8121 0.7432 0.7340
Light-house 45° 0.8476 0.7358 0.7664 0.8432 0.8118 0.7848 0.8151 0.6927 0.7542 0.8428 0.7919 0.7004 0.6921
Truck 45° 0.8274 0.7425 0.7798 0.8169 0.8090 0.7764 0.8396 0.8192 0.7693 0.8542 0.7985 0.6463 0.6259
Airplane 135° 0.8433 0.7436 0.6986 0.8564 0.8109 0.7896 0.8039 0.6851 0.7734 0.8553 0.8109 0.6953 0.6833
Bark 135° 0.7422 0.6237 0.6144 0.7398 0.7680 0.6712 0.6705 0.6150 0.6576 0.7377 0.6790 0.6175 0.6174
Wall 135° 0.6956 0.5557 0.5474 0.7145 0.6702 0.6057 0.6287 0.5453 0.5457 0.6894 0.6115 0.4838 0.4687

Bold indicates the minimum value


Multimedia Tools and Applications

Content courtesy of Springer Nature, terms of use apply. Rights reserved.


Table 9 GLCM based energy measure for the images of different sizes with rotation offset
Multimedia Tools and Applications

Images Rot Offset Orig-inal Image CUM UMF MMSI- CEEF UMHE RUM Riesz Mod HF Proposed
[7] -KG CHE [43] [13] [19] [25] GL [15]
[16] [28] [8] RL Mask GL Mask Riesz Mask

Lena 135° 0.1089 0.0864 0.0759 0.0874 0.1076 0.1004 0.0827 0.0724 0.0918 0.0979 0.0749 0.0591 0.0591
Surface 45° 0.1637 0.1166 0.1201 0.1321 0.1418 0.1369 0.1019 0.0744 0.1532 0.1553 0.0958 0.0624 0.0599
Moon 45° 0.2230 0.1603 0.0889 0.1336 0.2156 0.1819 0.1403 0.1013 0.1801 0.2018 0.1144 0.0625 0.0575
Baboon 135° 0.0737 0.0467 0.0403 0.0486 0.0748 0.0556 0.0466 0.0382 0.0620 0.0626 0.0407 0.0289 0.0283
Barbara 135° 0.0885 0.0676 0.0575 0.0706 0.0838 0.0751 0.0632 0.0576 0.0639 0.0717 0.0592 0.0434 0.0419
Pirate 135° 0.1199 0.0994 0.0914 0.0832 0.1161 0.1090 0.0851 0.0754 0.1053 0.1130 0.0783 0.0604 0.0587
Bridge 135° 0.0663 0.0455 0.0505 0.0615 0.0623 0.0528 0.0507 0.0403 0.0507 0.0613 0.0446 0.0370 0.0361
Light-house 45° 0.0894 0.0668 0.0708 0.0649 0.0884 0.0752 0.0720 0.0632 0.0751 0.0839 0.0722 0.0535 0.0517
Truck 45° 0.2075 0.1629 0.1323 0.1383 0.1989 0.1765 0.1536 0.1308 0.2173 0.1529 0.1515 0.0950 0.0898
Airplane 135° 0.6188 0.5981 0.4035 0.4452 0.6199 0.6144 0.6468 0.6951 0.7056 0.4727 0.4110 0.3330 0.3251
Bark 135° 0.0734 0.0542 0.0526 0.0521 0.0768 0.0612 0.0530 0.0521 0.0590 0.0740 0.0555 0.0517 0.0517
Wall 135° 0.1695 0.1144 0.0898 0.1191 0.1745 0.1348 0.1058 0.0880 0.1138 0.1603 0.1020 0.0683 0.0658

Bold indicates the minimum value

Content courtesy of Springer Nature, terms of use apply. Rights reserved.


Table 10 GLCM based homogeneity measure for the images of different sizes with rotation offset

Images Rot Offset Orig-inal Image CUM UMF MMSI- CEEF UMHE RUM Riesz Mod HF Proposed
[7] -KG CHE [43] [13] [19] [25] GL [15]
[16] [28] [8] RL Mask GL Mask Riesz Mask

Lena 135° 0.8520 0.7991 0.8039 0.8070 0.8399 0.8319 0.8203 0.7848 0.8647 0.8658 0.7997 0.7425 0.7425
Surface 45° 0.7871 0.7205 0.7245 0.7200 0.7696 0.7509 0.7364 0.6889 0.7780 0.7795 0.7127 0.6305 0.6235
Moon 45° 0.8602 0.7896 0.6621 0.7823 0.8510 0.8166 0.8180 0.7609 0.8173 0.8510 0.7722 0.6393 0.6218
Baboon 45° 0.7303 0.6409 0.6241 0.6728 0.7200 0.6734 0.6730 0.6155 0.6900 0.7124 0.6385 0.5623 0.5566
Barbara 135° 0.8115 0.7494 0.7516 0.7842 0.8035 0.7727 0.7819 0.7431 0.7644 0.8008 0.7565 0.6889 0.6811
Pirate 135° 0.8716 0.8239 0.8356 0.8299 0.8543 0.8458 0.8390 0.8059 0.8230 0.8788 0.8188 0.7473 0.7389
Bridge 135° 0.7731 0.6971 0.7178 0.7594 0.7556 0.7268 0.7316 0.6734 0.6985 0.7637 0.7038 0.6572 0.6506
Light-house 45° 0.7953 0.7284 0.7414 0.7580 0.7860 0.7538 0.7525 0.7064 0.7397 0.7830 0.7563 0.6880 0.6804
Truck 45° 0.8721 0.8167 0.7438 0.7795 0.8559 0.8368 0.8183 0.7306 0.8446 0.8409 0.8203 0.7139 0.7017
Airplane 135° 0.9478 0.9264 0.8912 0.9108 0.9419 0.9363 0.9248 0.9150 0.9403 0.9276 0.8986 0.8416 0.8333
Bark 135° 0.6932 0.6232 0.6166 0.6458 0.6838 0.6513 0.6168 0.6247 0.6218 0.6874 0.6256 0.6161 0.6161
Wall 135° 0.7845 0.7005 0.6746 0.7211 0.7751 0.7338 0.7106 0.6713 0.6961 0.7706 0.7015 0.6094 0.6007

Bold indicates the minimum value


Multimedia Tools and Applications

Content courtesy of Springer Nature, terms of use apply. Rights reserved.


Multimedia Tools and Applications

c1 (480×270) c2 (480×270) c3 (480×270) c4 (480×270)

Statue1 Statue2 Statue3 Statue4 Statue5


(352×240) (352×240) (352×240) (352×240) (352×240)
Fig. 9 Test images of VIP illumination saliency dataset used for simulation

It is observed from Fig. 10 that the images enhanced with the proposed approach provide
more sharp features in comparison to the existing techniques. However, in the c3 image, the
edges are blurred at the portion of image where improper lighting occurs in the case of
proposed approach. But the edges are well preserved by HF [15] because this technique is
devoted to enhance the non-uniformly illuminated images. Similarly, on considering the
images with bad illumination, it is observed that some existing techniques provided better
results than the proposed approach for dark and foggy images because they are intended to
provide the contrast enhancement in case of such images. However, the comparative analysis
based on performance metrics is illustrated in Tables 11 and 12.
It is observed from Tables 11 and 12 that the proposed approach provided better results in
comparison to existing techniques except for the dark and foggy images. However, the SSIM
and CC parameters are minimum for MMSICHE [28] and CEEF [43] in some images because
the images enhanced by these techniques are either distorted or drastically changed due to an
increase in the image contrast. Moreover, it can be seen from Fig. 10 that the test images of the
statue enhanced by MMSICHE [28] are generally distorted. However, the contrast of the

Technique/ CUM UMFKG MMSICHE CEEF UMHE RUM Riesz Mod HF Proposed
Images [7] [16] [28] [43] [13] [19] [25] GL [15]
[8] RL Mask GL Mask Riesz Mask

c1

c2

c3

c4

Statue1

Statue2

Statue3

Statue4

Statue5

Fig. 10 Simulation results for test images of VIP Illumination Saliency Dataset

Content courtesy of Springer Nature, terms of use apply. Rights reserved.


Table 11 Performance parameters for the test images of VIP Illumination Saliency Dataset

Metrics Images Orig-inal Image CUM UMF MMSI- CEEF UMHE RUM Riesz Mod HF Proposed
[7] -KG CHE [43] [13] [19] [25] GL [15]
[16] [28] [8] RL Mask GL Mask Riesz Mask

IE c1 6.8656 6.9567 6.9295 6.7968 6.8872 6.8944 7.2159 7.2214 6.8585 6.8770 7.2482 7.3315 7.3394
c2 7.2828 7.3646 7.3500 7.2077 7.3133 7.3005 7.5184 7.5275 7.2630 7.2872 7.5395 7.5823 7.5855
c3 7.6681 7.7117 7.7065 7.6049 7.7150 7.6833 7.6554 7.6623 7.6374 7.6681 7.7239 7.7446 7.7475
c4 7.6124 7.6730 7.6933 7.5408 7.6077 7.6307 7.6942 7.6958 7.5969 7.6024 7.7188 7.7582 7.7610
Statue1 5.8162 5.9727 6.4304 5.7574 5.9435 5.8816 6.0548 6.3519 5.9865 5.8273 6.2176 6.4570 6.4722
Statue2 3.1571 3.2604 3.4829 2.3821 4.5093 3.1942 3.3845 3.6394 3.2011 3.1598 3.5449 3.7049 3.7248
Statue3 3.5082 3.6021 4.3358 3.4963 4.1593 3.5386 3.6588 3.8472 3.7613 3.5053 3.8172 4.0187 4.0214
Statue4 6.4032 6.5399 7.0501 6.3534 6.5832 6.4629 6.6255 6.8583 6.5584 6.4214 6.8207 7.0531 7.0533
Statue5 6.0852 6.2073 6.4221 6.0225 6.2708 6.1312 6.3053 6.4500 6.2706 6.0888 6.3791 6.6292 6.6327
AG c1 4.8922 6.9662 5.9452 6.9149 5.6086 5.9574 7.9777 8.0361 5.5198 5.5029 7.6023 10.2221 10.5028
c2 5.3054 7.4790 6.6486 7.0140 5.9707 6.4251 8.3608 8.3850 5.6407 5.6409 7.7205 9.7472 9.8397
c3 4.3333 5.9146 4.9712 4.9764 5.2074 5.0897 6.7608 6.6241 5.0197 5.0453 5.9956 7.6539 7.8478
c4 4.5952 6.3044 6.3687 5.5882 5.2710 5.4238 7.3240 7.1430 5.3994 5.2787 6.5958 8.7000 8.9233
Statue1 1.6986 2.3059 3.5525 3.7144 1.9402 1.9876 2.1682 2.8733 2.0797 1.7256 2.5995 3.9158 4.1052
Statue2 0.7439 0.9851 1.3043 1.2684 2.7633 0.8517 0.9653 1.2261 0.8450 0.7923 1.1594 1.6358 1.6987
Statue3 0.4910 0.6485 1.9359 2.5492 0.5459 0.5610 0.6217 0.8022 0.6269 0.4908 0.8032 1.3181 1.3429
Statue4 1.4625 1.9042 3.1576 3.0120 1.5963 1.6691 1.8775 2.3208 1.7540 1.4738 2.2256 3.3128 3.4829
Statue5 1.1803 1.5332 1.8609 2.3210 1.2784 1.3453 1.4705 1.7724 1.4171 1.1844 1.6923 2.5593 2.6040
SSIM c1 – 0.9703 0.9890 0.9055 0.9600 0.9888 0.9223 0.9245 0.9316 0.9962 0.9335 0.8400 0.8264
c2 – 0.9705 0.9844 0.9381 0.9464 0.9888 0.9230 0.9249 0.9292 0.9967 0.9478 0.8763 0.8651
c3 – 0.9748 0.9830 0.9599 0.8928 0.9912 0.9222 0.9291 0.9342 0.9966 0.9569 0.8955 0.8856
c4 – 0.9732 0.9640 0.9467 0.9470 0.9905 0.9198 0.9281 0.9293 0.9961 0.9480 0.8670 0.8549
Statue1 – 0.9855 0.8959 0.7249 0.9585 0.9949 0.9792 0.9033 0.9728 0.9984 0.9479 0.8763 0.8647
Statue2 – 0.9948 0.9777 0.9217 0.1821 0.9983 0.9891 0.9527 0.9917 0.9722 0.9628 0.9289 0.9225
Statue3 – 0.9955 0.8768 0.8010 0.9672 0.9985 0.9864 0.9463 0.9889 0.9993 0.9457 0.9053 0.9032
Statue4 – 0.9893 0.8653 0.8550 0.9605 0.9964 0.9746 0.9331 0.9770 0.9986 0.9467 0.8791 0.8668
Statue5 – 0.9913 0.9726 0.8554 0.9583 0.9971 0.9663 0.9216 0.9823 0.9986 0.9479 0.8930 0.8898
EME c1 5.2500 7.1696 9.0606 9.4122 5.4020 8.2744 6.7490 6.4548 5.1858 5.2428 7.4952 9.5228 9.7008
c2 5.3946 7.7754 9.2444 8.6193 4.9861 7.9521 6.3195 6.5376 5.2425 4.7596 7.8284 9.6635 9.8397
c3 6.7071 9.9952 10.4768 9.2369 5.3447 8.8185 7.0841 7.4965 6.6055 6.6551 9.5123 11.7741 12.0103
c4 5.5139 8.2238 9.4898 7.4810 4.4659 7.0041 5.6553 5.8816 5.4043 5.2520 7.3737 9.4088 9.5803
Multimedia Tools and Applications

Content courtesy of Springer Nature, terms of use apply. Rights reserved.


Table 11 (continued)

Metrics Images Orig-inal Image CUM UMF MMSI- CEEF UMHE RUM Riesz Mod HF Proposed
[7] -KG CHE [43] [13] [19] [25] GL [15]
[16] [28] [8] RL Mask GL Mask Riesz Mask

Statue1 2.3513 3.4529 5.1552 5.2666 1.8818 2.9338 2.6037 2.8908 3.5095 2.4374 3.1115 5.1271 5.3089
Statue2 8.1896 8.4911 12.3055 8.5678 4.6696 7.9181 10.2455 9.9433 10.2878 7.3113 9.3336 14.5363 14.9533
Statue3 0.4874 0.6792 2.6180 2.5298 0.4084 0.5856 0.5433 0.6188 0.7566 0.4988 0.6168 1.1119 1.1363
Statue4 1.7264 2.4268 5.0436 5.1203 1.3976 2.0748 1.8343 2.1279 2.5700 1.7909 3.1907 5.1170 5.1274
Statue5 1.5170 2.0935 2.8835 4.0251 1.2193 1.7955 1.5577 1.7377 2.1783 1.5749 1.8413 2.9153
Multimedia Tools and Applications

2.9639
CC c1 – 0.9893 0.9968 0.9865 0.9963 0.9961 0.9989 0.9928 0.9889 0.9986 0.9956 0.9793 0.9772
c2 – 0.9887 0.9941 0.9909 0.9933 0.9956 0.9958 0.9898 0.9868 0.9988 0.9930 0.9810 0.9796
c3 – 0.9963 0.9994 0.9975 0.9899 0.9979 0.9966 0.9943 0.9957 0.9998 0.9967 0.9888 0.9886
c4 – 0.9948 0.9948 0.9962 0.9937 0.9973 0.9978 0.9946 0.9937 0.9968 0.9975 0.9927 0.9879
Statue1 – 0.9978 0.9960 0.9403 0.9973 0.9993 0.9998 0.9994 0.9966 0.9999 0.9991 0.9772 0.9737
Statue2 – 0.9919 0.9764 0.9924 0.9334 0.9972 0.9989 0.9951 0.9903 0.9989 0.9958 0.9861 0.9798
Statue3 – 0.9974 0.9943 0.9385 0.9971 0.9992 0.9996 0.9988 0.9938 0.9998 0.9987 0.9898 0.9891
Statue4 – 0.9987 0.9969 0.9623 0.9984 0.9996 0.9999 0.9995 0.9977 0.9999 0.9994 0.9945 0.9935
Statue5 – 0.9989 0.9970 0.9596 0.9972 0.9997 0.9919 0.9795 0.9982 0.9999 0.9878 0.9829 0.9826

Bold signifies the optimum parameters

Content courtesy of Springer Nature, terms of use apply. Rights reserved.


Table 12 GLCM parameters for the test images of VIP Illumination Saliency Dataset

Images Rot Offset Orig-inal Image CUM UMF MMS-ICHE CEEF UMHE RUM Riesz Mod HF Proposed
[7] -KG [28] [43] [13] [19] [25] GL [15]
[16] [8] RL Mask GL Mask Riesz Mask

Contrast
c1 135° 0.4514 0.8585 0.6628 0.7588 0.5800 0.6713 1.1055 1.1086 0.5038 0.4998 0.9673 1.5883 1.6408
c2 135° 0.4850 0.9638 0.7996 0.7866 0.6057 0.7418 1.1308 1.1787 0.5479 0.5356 1.0210 1.4387 1.4739
c3 135° 0.3140 0.5818 0.4160 0.3713 0.4106 0.4499 0.7370 0.7289 0.4309 0.3595 0.5917 0.8957 0.9250
c4 135° 0.3519 0.6674 0.6983 0.4637 0.4307 0.5036 0.8513 0.8390 0.4978 0.3940 0.7068 1.1175 1.1479
Statue1 45° 0.0846 0.1486 0.3026 0.2296 0.1008 0.1159 0.1393 0.2231 0.1515 0.0860 0.2067 0.4200 0.4451
Statue2 135° 0.0234 0.0372 0.0567 0.0267 0.1628 0.0314 0.0334 0.0501 0.0285 0.0260 0.0442 0.0774 0.0820
Statue3 135° 0.0301 0.0470 0.1500 0.1570 0.0171 0.0360 0.0278 0.0322 0.0446 0.0289 0.0326 0.0764 0.0794
Statue4 45° 0.0926 0.1384 0.2657 0.2092 0.0929 0.1156 0.1125 0.1797 0.1173 0.0860 0.1516 0.3065 0.3302
Statue5 45° 0.0647 0.0980 0.1431 0.1576 0.0737 0.0819 0.0926 0.1197 0.0987 0.0651 0.1090 0.2324 0.2379
Correlation
c1 135° 0.8701 0.7789 0.8229 0.8569 0.8484 0.8175 0.7989 0.8048 0.7507 0.8590 0.8301 0.7408 0.7344
c2 135° 0.8632 0.7601 0.7914 0.8503 0.8349 0.8039 0.7911 0.7903 0.7369 0.8587 0.8087 0.7435 0.7366
c3 135° 0.9558 0.9213 0.9433 0.9570 0.9382 0.9378 0.9271 0.9283 0.9131 0.9548 0.9387 0.9094 0.9065
c4 135° 0.9344 0.8827 0.8800 0.9346 0.9131 0.9086 0.8933 0.8955 0.8609 0.9326 0.9101 0.8627 0.8591
Statue1 45° 0.9679 0.9449 0.9424 0.9609 0.9591 0.9564 0.9659 0.9593 0.9464 0.9676 0.9553 0.9114 0.9064
Statue2 135° 0.8320 0.7636 0.7072 0.8142 0.8716 0.7882 0.8271 0.8224 0.7639 0.8191 0.8220 0.7371 0.7285
Statue3 135° 0.9419 0.9132 0.8807 0.9451 0.9537 0.9317 0.8988 0.9145 0.8223 0.8419 0.9145 0.8232 0.8180
Statue4 45° 0.9720 0.9584 0.9535 0.9668 0.9655 0.9651 0.9754 0.9664 0.9605 0.9711 0.9705 0.9416 0.9373
Statue5 45° 0.9794 0.9690 0.9536 0.9726 0.9734 0.9740 0.9767 0.9725 0.9644 0.9771 0.9742 0.9462 0.9449
Energy
c1 135° 0.2108 0.1956 0.1979 0.1472 0.1956 0.2025 0.1622 0.1826 0.1548 0.1770 0.1889 0.1487 0.1460
c2 135° 0.1639 0.1475 0.1544 0.1449 0.1633 0.1550 0.1163 0.1160 0.1698 0.1694 0.1204 0.1154 0.1132
c3 135° 0.1097 0.1015 0.1098 0.1043 0.1013 0.1050 0.0826 0.0838 0.1095 0.1102 0.0886 0.0799 0.0776
c4 135° 0.1207 0.1075 0.1038 0.1068 0.1225 0.1140 0.0865 0.0875 0.1127 0.1101 0.0910 0.0860 0.0855
Statue1 45° 0.5028 0.4943 0.2936 0.2894 0.5035 0.4986 0.4249 0.4115 0.4461 0.5021 0.2918 0.2703 0.2679
Statue2 135° 0.9407 0.9377 0.9313 0.9389 0.4182 0.9393 0.9337 0.9247 0.9428 0.9394 0.9286 0.9215 0.9202
Statue3 135° 0.4940 0.4798 0.4815 0.2150 0.6241 0.4895 0.7002 0.6465 0.8418 0.8014 0.6960 0.6818 0.6879
Statue4 45° 0.2314 0.2228 0.1654 0.1508 0.2783 0.2271 0.2064 0.1906 0.2666 0.2709 0.1850 0.1620 0.1507
Statue5 45° 0.2919 0.2846 0.2712 0.1947 0.2790 0.2880 0.2536 0.2680 0.2891 0.2835 0.2139 0.1935 0.1928
i-
Multimedia Tools and Applications

Content courtesy of Springer Nature, terms of use apply. Rights reserved.


Table 12 (continued)

Images Rot Offset Orig-inal Image CUM UMF MMS-ICHE CEEF UMHE RUM Riesz Mod HF Proposed
[7] -KG [28] [43] [13] [19] [25] GL [15]
[16] [8] RL Mask GL Mask Riesz Mask

Homogeneity
c1 135° 0.8982 0.8699 0.8801 0.8727 0.8812 0.8821 0.8588 0.8683 0.8669 0.8879 0.8742 0.8344 0.8298
c2 135° 0.9017 0.8726 0.8820 0.8796 0.8895 0.8857 0.8535 0.8570 0.8794 0.8777 0.8629 0.8277 0.8229
c3 135° 0.9104 0.8870 0.9012 0.9034 0.8971 0.8977 0.8734 0.8769 0.8858 0.8874 0.8880 0.8574 0.8540
c4 135° 0.9064 0.8794 0.8744 0.8902 0.8965 0.8928 0.8654 0.8681 0.8804 0.8828 0.8783 0.8400
Multimedia Tools and Applications

0.8396
Statue1 45° 0.9651 0.9520 0.9112 0.9151 0.9606 0.9580 0.9481 0.9318 0.9426 0.9651 0.9296 0.8919 0.8879
Statue2 135° 0.9930 0.9911 0.9883 0.9918 0.9367 0.9921 0.9913 0.9886 0.9923 0.9928 0.9897 0.9852 0.9846
Statue3 135° 0.9850 0.9769 0.9413 0.9319 0.9915 0.9821 0.9863 0.9847 0.9781 0.9856 0.9845 0.9697 0.9688
Statue4 45° 0.9602 0.9477 0.9050 0.9305 0.9606 0.9542 0.9578 0.9389 0.9534 0.9636 0.9497 0.9152 0.9042
Statue5 45° 0.9728 0.9647 0.9554 0.9495 0.9690 0.9684 0.9641 0.9604 0.9608 0.9725 0.9638 0.9346 0.9334

Bold signifies the optimum parameters

Content courtesy of Springer Nature, terms of use apply. Rights reserved.


Multimedia Tools and Applications

(a) Average IE (b) Average AG

(c) Average SSIM (d) Average EME

(e) Average CC (f) Average Contrast

(g) Average Correlation (h) Average Energy

(i) Average Homogeneity

Fig. 11 Average performance metrics for test images of various datasets

Content courtesy of Springer Nature, terms of use apply. Rights reserved.


Multimedia Tools and Applications

Fig. 12 Fundus images

DR (438×292) Healthy (438×292)

mage is enhanced by CEEF [43]. Thus, it is confirmed that the proposed approach provides
better results in comparison to existing state-of-the-art techniques for most of the images with
varying illumination conditions.

5.4 Comparative analysis

In this section, the comparative analysis of the proposed approach is done on the basis of the
average performance metrics for the test images of various datasets in Fig. 11. Due to space
constraints, the UMFKG, MMSICHE, Proposed RL, Riesz, and GL are represented by
UMKG, MHE, RL (P), Riesz (P), and GL (P).
It is perceived from Fig. 11 that the proposed approach provides better performance in
comparison to existing techniques in terms of the average performance metrics. It is observed
that the proposed approach with Riesz fractional derivative has a maximum average IE, AG,
and EME with an improvement of at least 1.41%, 24.53%, and 6.36% respectively for test
images of various datasets. Similarly, the minimum average improvement in contrast as
compared to existing techniques is 37.38%. However, average SSIM and CC are less for
MMSICHE [28] and CEEF [43] in the case of test images of VIP Illumination Saliency
Dataset. This is due to the distortion in the images enhanced by MMSICHE [28]. However,
contrast improvement is achieved in the case of CEEF [43]. The energy is minimum for
MMSICHE [28] in the case of test images with varying illumination conditions, thus indicat-
ing less uniformity among the pixels of GLCM. It is due to the distortion in the foggy image
that led to the severe change in the uniformity among pixels. However, average Correlation
and Homogeneity are minimum for the Riesz derivative based proposed approach, thus,
providing more enhancement as compared to the existing state-of-the-art techniques. There-
fore, it is perceived that the proposed Riesz fractional derivative approach provides more
enhancement.

CUM [7] UMFKG [16] MMSICHE [28] CEEF [43] UMHE [13] RUM [19]
DR

Riesz [25] Mod GL [8] HF [15] RL (µ=0.43) GL (µ=0.96) Riesz (µ=0.95)

CUM [7] UMFKG [16] MMSICHE [28] CEEF [43] UMHE [13] RUM [19]
Healthy

Riesz [25] Mod GL [8] HF [15] RL (µ=0.44) GL (µ=0.95) Riesz (µ=0.94)


Fig. 13 Simulation results for Fundus images

Content courtesy of Springer Nature, terms of use apply. Rights reserved.


Table 13 Performance metrics obtained by various image enhancement techniques for the Fundus images

Images Rot Off-set Orig-inal Image CUM UMF MMS-ICHE CEEF UMHE RUM Riesz Mod HF Proposed
[7] -KG [28] [43] [13] [19] [25] GL [15]
[16] [8] RL Mask GL Mask Riesz Mask

IE
DR – 5.9964 6.0422 6.7793 6.5410 6.0326 6.0173 6.2890 6.5374 6.0055 6.0386 6.4514 6.7488 6.7796
Healthy – 6.0858 6.0926 6.4383 5.9915 6.0701 6.0896 6.3482 6.4957 6.0879 6.1184 6.4685 6.6176 6.6708
AG
DR – 2.2749 3.2775 7.2209 5.8340 3.0818 2.8302 3.0649 3.9818 2.7557 2.4253 4.0098 7.5322 8.0942
Healthy – 1.4615 2.0004 4.5773 5.2350 2.4930 2.5083 2.6525 3.4866 2.4900 2.1206 3.3986 6.3514 6.8398
SSIM
DR – – 0.9726 0.6739 0.6461 0.8247 0.9888 0.9564 0.8933 0.9588 0.9845 0.8922 0.6600 0.6476
Healthy – – 0.9873 0.8510 0.6571 0.9081 0.9889 0.9621 0.9063 0.9600 0.9837 0.9120 0.7015 0.6672
EME
DR – 1.9373 2.9432 5.3082 5.9299 1.4817 2.2455 2.0578 2.7343 2.7170 4.2354 2.6183 5.5695 5.9947
Healthy – 0.9567 1.3478 3.1367 5.1354 1.1129 1.6867 1.5641 2.0413 2.0277 3.8243 2.5401 4.8824 5.1634
CC
DR – – 0.9977 0.9847 0.8787 0.9855 0.9992 0.9999 0.9992 0.9973 0.9845 0.9991 0.9829 0.9791
Healthy – – 0.9994 0.9936 0.8829 0.9904 0.9994 0.9998 0.9990 0.9982 0.9996 0.9992 0.9891 0.9866
Contrast
DR 45° 0.1175 0.1933 0.6188 0.3959 0.1567 0.1610 0.1708 0.2480 0.1505 0.1347 0.2503 0.6268 0.7006
Healthy 45° 0.0639 0.1002 0.2936 0.4251 0.1311 0.1411 0.1485 0.2095 0.1433 0.1108 0.2020 0.4792 0.5304
Correlation
DR 45° 0.9429 0.9084 0.8701 0.9195 0.9643 0.9228 0.9516 0.9419 0.9239 0.9395 0.9411 0.8617 0.8476
Healthy 45° 0.9750 0.9611 0.9215 0.9001 0.9695 0.9455 0.9656 0.9580 0.9426 0.9605 0.9593 0.9060 0.8967
Energy
DR 45° 0.2662 0.2317 0.1210 0.1818 0.3082 0.2451 0.2087 0.1753 0.2527 0.2542 0.1733 0.1194 0.1128
Healthy 45° 0.3182 0.2952 0.1826 0.1674 0.3100 0.2706 0.2137 0.1835 0.2630 0.2819 0.1844 0.1315 0.1252
Homogeneity
DR 45° 0.9458 0.9167 0.8201 0.8770 0.9426 0.9284 0.9267 0.9019 0.9296 0.9396 0.8995 0.8149 0.8015
Healthy 45° 0.9701 0.9539 0.8905 0.8622 0.9487 0.9359 0.9360 0.9139 0.9322 0.9503 0.9155 0.8379 0.8264

Bold signifies the optimum parameters


Multimedia Tools and Applications

Content courtesy of Springer Nature, terms of use apply. Rights reserved.


Multimedia Tools and Applications

Table 14 Average processing time (in secs) of various algorithms

Techniques CUM UMFKG MMSI- CEEF UMHE RUM Riesz Mod GL HF Proposed
[7] [16] C H E [43] [13] [19] [25] [8] [15]
[28]

Processing 3.0672 38.9263 3.3461 3.3126 3.9271 5.4814 1.1649 4.9278 34.4761 3.6798
Time

5.5 Performance analysis on real application of fundus images

Fundus images are commonly used for examining human eyes. However, they are
generally of low contrast. Therefore, it is required to enhance the images because they
aid in the detection of patients with diseases such as Glaucomatous, Diabetic Retinop-
athy (DR), etc. which may further lead to the loss of vision [18]. Hence, the proposed
approach is used for the enhancement of Fundus images of Healthy individual and DR
patients obtained from High Resolution Fundus (HRF) database [2] as shown in Fig. 12.
The performance analysis of the proposed approach as well as the existing approaches is
illustrated in Fig. 13 and Table 13.
It is perceived that the proposed approach provides more enhancement in comparison to the
existing techniques in both qualitative and quantitative forms. However, SSIM and CC are less
for MMSICHE [28] because the images are distorted in this case as observed from Fig. 13.
However, the visual results and other enhancement parameters show the applicability of
proposed approach to enhance the Fundus images.

5.6 Computational time analysis

The computational time is a significant parameter to evaluate the performance of the proposed
approach. The computational complexity for the addition of two images is Θ(MN) whereas,
the computational complexity for computing the fractional mask is O(MN) +
O(MN (log(MN))2) and convolution of the original image with the mask is O(MN
log (MN)) [20]. Thus, the overall computational complexity of the proposed scheme
is 2Θ(MN) + O(MN) + O(MN (log(MN))2) + 2O(MN log (MN). Furthermore, the average time
elapsed for the execution of various algorithms is illustrated in Table 14.
It is evident that the average processing time for the proposed approach is either less
or comparable to the existing techniques except for Riesz [25] approach. The average
processing time is dependent on the size of the image as well as the complexity of the
algorithm.

6 Limitations

The proposed approach provides image enhancement in most of the images while
preserving the image details. However, the few limitations of the proposed approach
are that edges are not efficiently preserved where improper lighting occurs. In addition to
this, the proposed approach is unable to provide an enhancement in the case of dark and
foggy images.

Content courtesy of Springer Nature, terms of use apply. Rights reserved.


Multimedia Tools and Applications

7 Conclusion

In this paper, RL, GL, and Riesz fractional derivative based UM technique is implemented for the
improvement of image visual quality. The proposed approach improved the contrast, texture, and
edges of the images by utilizing fractional derivatives in addition to the Laplacian mask. The
effectiveness of the proposed approach is also confirmed by considering the test images with
varying illumination conditions. The improvement in average IE for proposed Riesz fractional
derivative is 1.41–12.29% for test images of USC-SIPI and LIVE datasets while 1.63–9.74% for
test images of VIP Illumination Saliency Dataset as compared to the existing techniques. The
contrast in the case of GLCM parameters shows the improvement of at least 37.38% in test
images of USC-SIPI and LIVE datasets whereas 42.65% in the test images of VIP Illumination
Saliency Dataset respectively for the proposed Riesz fractional derivative. The analysis of the
proposed technique based on other GLCM parameters also confirmed its effectiveness. More-
over, the enhancement of Fundus images validated its efficacy in the case of practical applica-
tions. Thus, on the basis of the various performance metrics, it is observed that the proposed Riesz
fractional derivative approach outperforms in comparison to the existing techniques. The future
work involves the use of fractional derivatives for the enhancement of color images.

References

1. Aysal TC, Barner KE (2006) Quadratic weighted median filters for edge enhancement of noisy images.
IEEE Trans Image Process 15(11):3294–3310
2. Budai A, Bock R, Maier A, Hornegger J, Michelson G (2013) Robust Vessel Segmentation in Fundus
Images. Int J Biomed Imag 2013:1–11. https://doi.org/10.1155/2013/154860
3. Chen S, Zhao F (2018) The adaptive fractional order differential model for image enhancement based on
segmentation. Int J Pattern Recognit Artif Intell 32:1854005. https://doi.org/10.1142/S0218001418540058
4. Chwyl B, Chung AG, Li FY, Wong A, Clausi DA (2015) TIGER: a texture-illumination guided energy
response model for illumination robust local saliency. Proc IEEE Int Conf image process (ICIP):1970–1974
5. Gan Z, Yang H (2010) Texture enhancement though multiscale mask based on RL fractional differential.
Proc Int Conf Inf Net Autom (ICINA) 333–337
6. Garg V, Singh K (2012) An improved Grunwald-Letnikov fractional differential mask for image texture
enhancement. Int J Adv Comput Sci Appl 3(3):130–135
7. Gonzalez RC, Woods RE (2008) Digital image processing. Englewood Cliffs, Prentice Hall
8. Hemalatha S, Anouncia SM (2018) GL fractional differential operator modified using auto-correlation
function: texture enhancement in images. Ain Shams Eng J 9(4):1689–1704
9. Jain AK (1989) Fundamentals of digital image processing. Englewood Cliffs, Prentice Hall
10. Jindal N, Singh K (2014) Image and video processing using discrete fractional transforms. SIViP 8(8):
1543–1553
11. Justin J, Anoop BN, Williams J (2019) A modified unsharp masking with adaptive threshold and
objectively defined amount based on saturation constraints. Multimed Tools Appl 78(8):11073–11089
12. Kansal S, Tripathi RK (2020) Adaptive geometric filtering based on average brightness of the image and
discrete cosine transform coefficient adjustment for gray and color image enhancement. Arab J Sci Eng 45:
1655–1668
13. Kansal S, Purwar S, Tripathi RK (2018) Image contrast enhancement using unsharp masking and histogram
equalization. Multimed Tools Appl 77(20):26919–26938
14. Kau LJ, Lee TL (2014) A three-step approach with adaptive additive magnitude selection for the sharpening
of images. The Sci World J 2014:1–15. https://doi.org/10.1155/2014/528696
15. Kaur K, Jindal N, Singh K (2019) Improved homomorphic filtering using fractional derivatives for enhancement
of low contrast and non-uniformly illuminated images. Multimed Tools Appl 78(19):27891–27914
16. Kwok N, Shi H (2014) Design of unsharp masking filter kernel and gain using particle swarm optimization.
Proc Int Cong Image Sign Process 217–222
17. Kwok N, Shi H, Fang G, Ha Q (2013) Adaptive scale adjustment design of unsharp masking filters for
image contrast enhancement. Proc Int Conf Mach Learn Cyber 884–889

Content courtesy of Springer Nature, terms of use apply. Rights reserved.


Multimedia Tools and Applications

18. Lavín-Delgado JE, Solís-Pérez JE, Gómez-Aguilar JF, Escobar-Jiménez RF (2020) A new fractional-order
mask for image edge detection based on Caputo–Fabrizio fractional-order derivative without singular
kernel. Circ Syst Sign Process 39(3):1419–1448
19. Lee SL, Tseng CC (2016) Image sharpening using matrix Riesz fractional order differentiator and discrete
sine transform. Proc IEEE Int Conf Consum Electron 1–2
20. Nandal A, Gamboa-Rosales H, Dhaka A, Celaya-Padilla JM, Galvan-Tejada JI, Galvan-Tejada CE,
Martinez-Ruiz FJ, Guzman-Valdivia C (2018) Image edge detection using fractional calculus with feature
and contrast enhancement. Circ Syst Sign Process 37(9):3946–3972
21. Ortigueira MD (2011) Fractional calculus for scientists and engineers. Springer, Dordrecht
22. Panetta K, Samani A, Agaian S (2014) Choosing the optimal spatial domain measure of enhancement for
mammogram images. Int J Biomed Imag 2014:1–8. https://doi.org/10.1155/2014/937849
23. Polesel A, Ramponi G, Mathews VJ (2000) Image enhancement via adaptive unsharp masking. IEEE Trans
Image Process 9(3):505–510
24. Pu YF, Zhou JL, Yuan X (2010) Fractional differential mask: a fractional differential-based approach for
multiscale texture enhancement. IEEE Trans Image Process 19(2):491–511
25. Raghunandan KS, Shivakumara P, Jalab HA, Ibrahim RW, Kumar GH, Pal U, Lu T (2017) Riesz fractional
based model for enhancing license plate detection and recognition. IEEE Trans Circuits Syst Video Technol
28(9):2276–2288
26. Saxena R, Singh K (2005) Fractional Fourier transform: a novel tool for signal processing. J Indian Inst Sci
85(1):11–26
27. Sheikh HR, Wang Z, Cormack L, Bovik AC (2005) LIVE image quality assessment database release 2
(2005). http://live.ece.utexas.edu/research/quality. Accessed 10 September 2019
28. Singh K, Kapoor R (2014) Image enhancement via median-mean based sub-image-clipped histogram
equalization. Optik 125(17):4646–4651
29. Singh G, Singh K (2017) Improved JPEG anti-forensics with better image visual quality and forensic
undetectability. Forens Sci Int 277:133–147
30. Singh K, Saxena R, Kumar S (2013) Caputo-based fractional derivative in fractional Fourier transform
domain. IEEE J Emerg Sel Topics Circ Syst 3(3):330–337
31. Singh H, Kumar A, Balyan LK, Singh GK (2017) A novel optimally weighted framework of piecewise
gamma corrected fractional order masking for satellite image enhancement. Comput Electr Eng 75:245–261
32. Solomon C, Breckon T (2011) Fundamentals of digital image processing: a practical approach with
examples in MATLAB. Wiley-Blackwell, Canterbury
33. Starovoitov VV, Eldarova EE, Iskakov KT (2020) Comparative analysis of the SSIM index and the Pearson
coefficient as a criterion for image similarity. Eurasian J Math Comput Appl 8(1):76–90
34. Suman S, Jha RK (2017) A new technique for image enhancement using digital fractional-order Savitzky–
Golay differentiator. Multidim Syst Sign Process 28(2):709–733
35. The USC-SIPI Image Database [Online] (2017). Available: http://sipi.usc.edu/database/database.php.
Accessed 20 July 2020
36. Tsafack N, Kengne J, Abd-El-Atty B, Iliyasu AM, Hirota K, Abd El-Latif AA (2020) Design and
implementation of a simple dynamical 4-D chaotic circuit with applications in image encryption. Inf Sci
515:191–217
37. Tseng CC, Lee SL (2014) Design of digital Riesz fractional order differentiator. Signal Process 102:32–45
38. VIP Illumination Saliency Dataset [Online] (2019) Available: https://uwaterloo.ca/vision-image-processing-lab.
Accessed 27 July 2020
39. Wang Z, Bovik AC, Sheikh HR, Simoncelli EP (2004) Image quality assessment: from error visibility to
structural similarity. IEEE Trans Image Process 13(4):600–612
40. Xie L, Wang G, Zhang X, Xiao B, Zhou B, Zhang F (2014) Remote sensing image enhancement based on
wavelet analysis and histogram specification. Proc IEEE Int Conf cloud Comput Intell Syst 55–59
41. Yang Q, Chen D, Zhao T, Chen Y (2016) Fractional calculus in image processing: a review. Fract Calc
Appl Anal 19(5):1222–1249
42. Ye W, Ma KK (2018) Blurriness-guided unsharp masking. IEEE Trans Image Process 27(9):4465–4477
43. Ying Z, Li G, Ren Y, Wang R, Wang W (2017) A new image contrast enhancement algorithm using
exposure fusion framework. Proc Int Conf Comp Anal Images Patt 36–46
44. Yu Q, Liu F, Turner I, Burrage K, Vegh V (2012) The use of a Riesz fractional differential-based approach
for texture enhancement in image processing. ANZIAM J 54:590–607
45. Zhuang P, Ding X (2020) Underwater image enhancement using an edge-preserving filtering Retinex
algorithm. Multimed Tools Appl 79:17257–17277

Publisher’s note Springer Nature remains neutral with regard to jurisdictional claims in published maps and
institutional affiliations.

Content courtesy of Springer Nature, terms of use apply. Rights reserved.


Terms and Conditions
Springer Nature journal content, brought to you courtesy of Springer Nature Customer Service Center
GmbH (“Springer Nature”).
Springer Nature supports a reasonable amount of sharing of research papers by authors, subscribers
and authorised users (“Users”), for small-scale personal, non-commercial use provided that all
copyright, trade and service marks and other proprietary notices are maintained. By accessing,
sharing, receiving or otherwise using the Springer Nature journal content you agree to these terms of
use (“Terms”). For these purposes, Springer Nature considers academic use (by researchers and
students) to be non-commercial.
These Terms are supplementary and will apply in addition to any applicable website terms and
conditions, a relevant site licence or a personal subscription. These Terms will prevail over any
conflict or ambiguity with regards to the relevant terms, a site licence or a personal subscription (to
the extent of the conflict or ambiguity only). For Creative Commons-licensed articles, the terms of
the Creative Commons license used will apply.
We collect and use personal data to provide access to the Springer Nature journal content. We may
also use these personal data internally within ResearchGate and Springer Nature and as agreed share
it, in an anonymised way, for purposes of tracking, analysis and reporting. We will not otherwise
disclose your personal data outside the ResearchGate or the Springer Nature group of companies
unless we have your permission as detailed in the Privacy Policy.
While Users may use the Springer Nature journal content for small scale, personal non-commercial
use, it is important to note that Users may not:

1. use such content for the purpose of providing other users with access on a regular or large scale
basis or as a means to circumvent access control;
2. use such content where to do so would be considered a criminal or statutory offence in any
jurisdiction, or gives rise to civil liability, or is otherwise unlawful;
3. falsely or misleadingly imply or suggest endorsement, approval , sponsorship, or association
unless explicitly agreed to by Springer Nature in writing;
4. use bots or other automated methods to access the content or redirect messages
5. override any security feature or exclusionary protocol; or
6. share the content in order to create substitute for Springer Nature products or services or a
systematic database of Springer Nature journal content.
In line with the restriction against commercial use, Springer Nature does not permit the creation of a
product or service that creates revenue, royalties, rent or income from our content or its inclusion as
part of a paid for service or for other commercial gain. Springer Nature journal content cannot be
used for inter-library loans and librarians may not upload Springer Nature journal content on a large
scale into their, or any other, institutional repository.
These terms of use are reviewed regularly and may be amended at any time. Springer Nature is not
obligated to publish any information or content on this website and may remove it or features or
functionality at our sole discretion, at any time with or without notice. Springer Nature may revoke
this licence to you at any time and remove access to any copies of the Springer Nature journal content
which have been saved.
To the fullest extent permitted by law, Springer Nature makes no warranties, representations or
guarantees to Users, either express or implied with respect to the Springer nature journal content and
all parties disclaim and waive any implied warranties or warranties imposed by law, including
merchantability or fitness for any particular purpose.
Please note that these rights do not automatically extend to content, data or other material published
by Springer Nature that may be licensed from third parties.
If you would like to use or distribute our Springer Nature journal content to a wider audience or on a
regular basis or in any other manner not expressly permitted by these Terms, please contact Springer
Nature at

[email protected]

You might also like