RESEARCH

O&G Forum 2024; 34-3s: 2922-2942

AN OUTSTRETCHED EXPLORATION ON
IMPULSE NOISE REDUCTION FOR TARNISHED
IMAGES
S. Prathiba1, B. Sivagami 2

1Research Scholar, Department of Computer Science & Research Centre, S.T. Hindu College, Nagercoil –
629002, Affiliated to Manonmaniam Sundaranar University, Abishekapatti, Tirunelveli – 627012, Tamil Nadu, India.
2Associate Professor, Department of Computer Science & Application, S.T. Hindu College, Nagercoil – 629002,

Affiliated to Manonmaniam Sundaranar University, Abishekapatti, Tirunelveli – 627012, Tamil Nadu, India.
Corresponding author: [email protected]

Abstract
Image de-noising is one of the most fundamental difficulties in image processing and computer vision, with the goal
of estimating the original image by suppressing noise from a noise-contaminated version of the image. The need
for more accurate pictures is steadily increasing, with the growth in the amount of digital images created every day.
Many approaches for removing salt and pepper noise from various image types have been reported in the literature.
This paper explores many de-noising techniques and investigates on noise reduction by considering the de-noising
field's essential properties. This survey considers fifty papers in order to blow light on many de-noising techniques
which helps young researchers to broaden up their knowledge. Since, for the vast majority of image processing
applications, image de-noising is the principal task.
KEY WORD: Salt and Pepper noise, Image de-noising, Impulse noise reduction, Image enhancement, Noise
removal

I. INTRODUCTION [6]. This paper provides a summary and/or a synthesis of the

The elimination of impulsive noise while maintaining the findings of selected research contributions being published by
integrity of an image is a critical issue in image processing. other researchers.
Images degraded by noise leads to deteriorated visual image
quality. Removal of randomly occurring impulses without II A COMPENDIUM OF IMAGE DENOISING
disrupting edges, corners and other sharp structures is a basic METHODS
signal processing requirement. Several ways for reducing noise This study considers fifty papers from different years starting
have already been proposed by researchers. Each method has its from 1996 to 2021. Also by considering each technique in these
own set of benefits and drawbacks. While considering medical papers, they have been categorized and fall in to groups like
images, noise in the acquisition or transmission is very common. Deep learning and Neural network, Mean based filter, Median
The noise signal can be easily misinterpreted and results a based filter, Fuzzy logic and Miscellaneous. Based on the noise
considerable reduction in the fusion effect. To overcome this ratio considered and also its PSNR values, the efficient and
scenario a variation model for diagnostic medical image fusion favorable noise reduction technique is identified.
and denoising has been developed [1]. De-noising is a technique
for reducing image noise while retaining desirable details A) Deep learning and Neural Network
utilizing prior knowledge of the images [2]. De-noising images H. Kong et al. [7] use a Neural Network Adaptive Filter (NNAF)
with Gaussian and Poisson noise has garnered a great deal of for the removal of impetuous noise in digital images. The NNAF
interest in the image processing field [3]. Discriminative filter is used to eliminate the impulses, and pixel classification
learning-based de-noising methods have gotten a lot of attention is utilized to detect the noisy pixels. It shows better performance
and have been explored extensively due to their strong de- than the traditional median type filters. But the shortcoming of
noising performance and much lower inference time compared this filter is its computational complexity due to its large
to model-based de-noising approaches [4]. Filtering images dynamic window size.
from many channels is difficult both in terms of efficiency and Zhe Zhou [26] presents an Adaptive Detail-Preserving Filter
efficacy. A simple transform-threshold-inverse strategic (ADPF) based on the Cloud Model (CM) to remove impulse
approach could generate hypercompetitive results by training a noise. An uncertainty-based detector is used in this filter to
good global patch basis and a local principal component identify the pixels that have been distorted by impulse noise.
analysis transform in the grouping dimension [5]. The The stumbling block of this method is that the edges may get a
preponderance of contemporary image de-noising techniques is blur if the image has a high noise level and also it can detect
intended to enhance de-noising quality. Thus in terms of the only the fixed-valued impulse noise.
amount of parameters and computational complexity, the Fariborz Taherkhani et al. [37] provide a Radial Basis Functions
framework can be extended to numerous existing approaches to (RBFs) interpolation-based approach for estimating the
enable them attain more competitive de-noising performance intensities of damaged pixels from their neighbors. The

OBSTETRICS & GYNAECOLOGY FORUM 2024 | ISSUE 3s | 2922

RESEARCH
O&G Forum 2024; 34-3s: 2922-2942

advantage of using this algorithm is that it restores images with efficiency while using it for high-resolution images and huge
higher visual quality, smoother edges, and better texture detail. noise level.
The demerit of using this algorithm is that it fails to address Sana Sadeghi et al. [25] present a method for impulse noise
Gaussian and Speckle noises. reduction from images using fuzzy cellular automata. The merit
Minghui Zhang et al. [38] put forth a data-driven algorithm for of this method is the Simplicity, robustness, parallel manner and
impulse noise removal via Iterative Scheme-Inspired Network distribution ability for noise enhancement using fuzzy cellular
(ISIN). The suggested network will not only change the focus automata. The limitation of this approach is that the accuracy in
from online optimization to an upfront offline training phase, detecting noisy pixel is less when testing with images with a
but it will also be applied to all new data using the learnt high noise level.
parameters. U. Sahin et al. [30] put forth an image de-noising algorithm to
Lianghai Jin et.al [42] present an image recovery method based restore digital images corrupted by impulse noise. It is based on
on deep convolutional neural network for impulse noise two-dimensional cellular automata (CA) with the help of fuzzy
removal. In this de-noising framework, there are two deep logic theory. The approach describes a local fuzzy transition ru le
CNNs: a classifier network and a regression network. The merit that assigns the next state value as a central pixel value and
of this method is the better de-noising performance. But the assigns a membership value to the corrupted pixel
pitfall is that the running time of this method is very high, also neighborhood. This filter has the benefit of being consistent and
it has a higher computational complexity stable across a wide range of noise levels. The demerit of this
Guanyu Li et al. [45] provide an approach for investigating the filter is that it loses its efficiency while filtering high-resolution
Densely Connected Network for Impulse Noise Removal images.
(DNINR), a method for learning pixel-distribution properties Yi Wang et al. [33] present an adaptive fuzzy switching
from noisy images that use CNN. The goodness of this method weighted mean filter to remove salt-and-pepper (SAP) noise.
is that it shows better performance on edge preservation and Noise detection and noise elimination are the two stages of the
noise suppression. The pitfall of this method is that this scheme de-noising process. The first step is to provide a more precise
loses its glory when applied to other non-Gaussian noises like mathematical expression for SAP noise. Second, in order to
Poisson noise and Rician noise. detect SAP noise, an enhanced maximum Absolute Luminance
Chun Li et al. [56] divulge an impulse noise removal model Difference (ALD) approach is devised.
(INRM) algorithm based on logarithmic image prior for medical Vikas Singh et al. [35] put forth an adaptive Type-2 fuzzy filter
image. Herein used the split Bregman iterative method to solve for removing salt and pepper noise from the images. The benefit
the objective function. The input used in this model are natural of employing this technique is that the filter keeps important
images and CT and MRI images. The goodness of this algorithm visual data even when there is a lot of noise. The stumbling
is that it is better than some existing classic algorithms for block of using this technique is that the computational time
impulse signal removal. The downfall of this algorithm is that it increases drastically for the images which have a high noise
fails to address the inverse problem such as image patching level.
problems, image segmentation problems, image blending to
noise. C) Mean based filter
XuYan et al. [58] developed Unsupervised Image De-noising Wei-Yu Han et al. [8] use the Minimum-Maximum Exclusive
algorithm based on Generative Adversarial Networks Mean (MMEM) filter, to remove impulse noise from highly
(UIDGAN). The model employs perceptual loss and cycle- corrupted images. This technique is preferable since it removes
consistency loss to ensure consistency of content information high impulse noises while simultaneously preserving image
which is considered it to be its shinning side. The drawback of information. The pitfall of this filter is that it loses its efficiency
this method is that it considers many parameters which in turn when it is applied for other types of images other than grey
increases its complexity and processing time. images.
B. Smolka et al. [11] divulge a method, where a new class of
B) Fuzzy logic filters for noise attenuation is introduced. It is considered to be
Stefan Schulte et al. [15] present an impulse noise reduction the modified and improved version of Vector Median Filter
method called a Novel Fuzzy Impulse Noise Detection Method (VMF) and its relationship with commonly used filtering
(NFIND) for color images. Color information is considered in techniques is also investigated. The root of the mean squared
this paper in order to design an improved impulse noise error (RMSE), peak signal to noise ratio (PSNR) and normalized
detection algorithm that filters just the corrupted pixels while mean square error (NMSE) were used for the comparisons The
maintaining color and edge sharpness. The pitfall of this method goodness of this method is that it has a low computational
is that it fails to reduce α-stable (a mixture of Gaussian and complexity. The flaw of this method is that it works efficiently
impulse noise) efficiently. The use of an additive noise reduction only for a particular application, and less reliable. Also, the
method is not adequately examined in this method. degradation of image quality is possible.
Kenny Kal Vin Toh et al. [22] develop a filter called, the Cluster- X.M. Zhang et al. [18] propose the Adaptive Switching Mean
based Adaptive Fuzzy Switching Median (CAFSM) which (ASM) filter to remove impulse noise. The filter uses
consists of a detail-preserving noise filter and a cascading, easy- conditional morphological noise detection to identify the
to-implement impulse detector. The advantages of the proposed corrupted pixels, and then uses the adaptive mean filter to
CAFSM filter are its capability in handling realistic impulse eliminate the identified impulses. In terms of noise reduction
noise model for real-world applications and the relatively fast and detail retention, this ASM filter surpasses many switching -
runtime. The pitfall of this framework is that it loses its

OBSTETRICS & GYNAECOLOGY FORUM 2024 | ISSUE 3s | 2923

 RESEARCH
O&G Forum 2024; 34-3s: 2922-2942

based filters. The stumbling block of this filter is that, it is technique (NLSP) for restoring heavily distorted images due to
incompatible with high-resolution images. impulse noise by removing only corrupted pixel by the median
Samsad Beagum Sheik Fareed et al. [36] present a mean filter value, or by it neighboring pixel value. The benefit of this filter
for effectively removing salt and pepper noise from images with is that, it prevents image blurring for large window sizes. This
greater noise densities. In this method, the filter works under filter also performs consistently and reliably over a wide range
two stages like Impulse Detection and Restoration (IDR). The of noise levels. This loses its efficiency while using switching
first stage finds the noisy pixels, whereas the second stage window technique which heads to low performance
recovers the noisy pixels that have been identified. The Zhengya Xu et al. [20] present a geometric features-based
advantage of employing this filter is that it consumes less time filtering technique called as the Adaptive Geometric Features
to compute than other adaptive filters. The disadvantage of this Based Filtering Technique (AGFF) along with its restoration
filter is its computational complexity technique which is based on the modified median for the
Qianqian Liu et al. [47] put forth a nonlinear Spline Adaptive removal of impulse noise in corrupted color images. The
Filter based on the Robust Geman-McClure estimator (SAF- goodness of this technique is that it provides a very reliable
RGM).Herein used the steady-state excess mean-square-erro r impulse noise type and ratio discrimination method. The pitfall
(EMSE) ζ to measure the performance of an adaptive filter. Also of this technique is that it is not integrated with other benchmark
cost function based on Geman-McClure is used in this approach. techniques to suppress a mixed Gaussian and impulse noise
The merit of using this filter is that it has a better stable contamination for color images which results in low
performance against impulsive noise. The drawbacks of using performance.
this technique are that it has high time consumption and high Smaïl Akkoul et al. [21] propose an Adaptive Switching Median
computational complexity (ASWM) filter for removing impulse noise from distorted
Mustapha Bouhrara et al. [48] develop an efficient method for images. The benefit of ASWM is that no a priori Threshold is to
noise estimation and reduction in multispectral MR images. be given as in the case of a classical SWM (Switching Median
This filter is a multispectral extension of the nonlocal maximum Filter) filter. Instead, using weighted statistics, the threshold is
likelihood filter (NLML) combining both spatial and spectral calculated locally from image pixel intensity values in a sliding
information. The goodness of this filter is that the Numerical window. The advantage of this filter is that, the psycho visual
and experimental analysis indicated the better performance for results are of high quality. The downside of this filter is that it
estimation of noise SD (Standard Deviation). The performance has a fixed window size, which implies it cannot use the
is limited in spatially heterogeneous regions, such as edges and switching window technique.
small structures, where patch redundancy is relatively poor Wei Wang et al. [24] present the framework of switching median
which mitigates its efficiency. filtering for removing impulse noise from corrupted images. In
this method, the noisy pixels are distinguished by Local Outlier
D) Median based filter Factor incorporating with Boundary Discriminative Noise
Zhou Wang et al. [9] use a Progressive Switching Median (PSM) Detection (LOFBDND) algorithm. The advantage of this
filter to remove the impulsive noise and also retaining the framework is that here the noise detection algorithm minimizes
integrity of the images. The merits of this method are that better the miss detection rate and false detection rate. The drawback of
results are obtained while using PSM filters. The stumbling this framework is that it will not support the huge noise level.
block of this method is that it works only for grayscale images; Iyad F. Jafar et al. [27] put forward a method with efficient
hence it can't support other types of images. Also this filter holds Improvements on the Boundary Discriminative Noise Detection
high computational complexity. (BDND) Filtering Algorithm which is a popular switching
F.J. Gallegos-Funes et al. [10] introduces The Median M-type median filter for the removal of high-density impulse noise. This
K-nearest neighbour (MM-KNN) filter to remove the salt and filter is tweaked by removing the restriction on expanding the
pepper noise from highly corrupted images. The robust point of filtering window and incorporates the spatial information of the
the pixels within the filtering frame is estimated by the filters. pixels in the filtering process.
Xiaoyin Xu et al. [12] present an adaptive two-pass rank order Osama S. Faragallah et al. [32] describe an optimal method for
filter (ATPMF) which undergoes two-pass filtering operations suppressing salt-and-pepper (S&P) noise under the Adaptive
to remove salt and pepper noise in highly corrupted images. The Switching Weighted Median Filter (ASWMF) paradigm. The
merit of this method is that the adaptive process detects ASWMF includes noise detection and noise removal stages. The
irregularities in the spatial distribution of the estimated impulse goodness of this technique is that it provides good performance
noise at the same time the false alarm was also corrected for a wide set of images. The stumbling block of this method is
efficiently. The main demerit is a high time consumption since that this method cannot be supportive of the huge noise level.
the filtration method is done twice in this method. Jiayi Chen et al. [40] put forth an Adaptive Sequentially
Zhonghua Ma et al. [13] use a neighborhood evaluated adaptive Weighted Median Filter (ASWMF) for images corrupted by
vector filter (NEAVF) which utilizes a novel neighborhood impulse noise. The benefit of implementing this ASWMF is that
evaluation process to improve the performance of noise it outperforms state-of-the-art filters when there is impulse
detection and detail preservation. The main detriment of this noise. Furthermore, the computation time is really short. The
method is the usage of a highly sophisticated filter that considers stumbling block of this filter is that it is hard to be applied for
color images as an only input and loses its credibility while real-time de-noising.
considering grayscale images. C. Jaspin Jeba Sheela et al. [44] present an Adaptive Switching
K. S. Srinivasan et al. [16] propose a filter which uses a Modified Decision Based Un-symmetric Trimmed Median
decision-based algorithm and non-linear signal processing Filter (ASMDBUTMF) for noise reduction in grayscale MR

OBSTETRICS & GYNAECOLOGY FORUM 2024 | ISSUE 3s | 2924

RESEARCH
O&G Forum 2024; 34-3s: 2922-2942

Images which are affected by salt and pepper noise. The good pixels from noise-free pixels when their intensity levels are
point of this technique is that it can be used as a preprocessing identical is addressed in this study in two steps by detecting
method for scanning machines for better robustness against the fixed-valued impulse noise.
noisy environment. The drawback of this method is that it Ruixuang Wang et al. [31] provide a single-patch technique for
cannot work efficiently for other types of images except MRI. detecting and removing nonpoint wise Random-Valued Impulse
Golam Muktadir Mukti et al. [57] present a MatLab-based Noise Noise within a generalized joint low-rank and sparse matrix
Removal Technique (MNRT) for removing salt and pepper recovery framework. The merit of this method is that it shows
noise from brain MR image. The goodness of this technique is better performance on non-point wise RVIN. The method's
that this weighted median filter provides high quality images by limitations include that, while most image patches are low-rank
removing salt and pepper noises. The drawback of this after being properly orientated, there are a few patches that do
technique is that it loses its efficiency while working with the not meet the low-rank assumption.
kernel size above three. Qing-Qiang Chen et al. [34] illustrate an effective and adaptive
algorithm called Noise Removal Algorithm (NRA) for
E) Miscellaneous removing pepper and salt noise. The algorithm contains noise-
Igor Aizenberg et al. [14] put forth the impulse-detecting pixel-detection and noise-filtering processes. The advantage of
Boolean functions for detection and elimination of impulsive this method is that it performs better in term of the PSNR (Peak
noises. This can be achieved by using single-pass filtering. The Signal to Noise Ratio). The drawback of this approach is that it
smoothing of edges and destruction of details are prevented, only supports grayscale images; thus, it cannot handle any other
which can be considered as the merit of this method. The pitfall type of image, resulting in poor performance.
of this method is that the priority given for edge preservation. Ganzhao Yuan et al. [39] put forth a model in the field of
Hence, the textures are not preserved regularization-based image processing. A new sparse
Wenbin Luo et al. [17] present an algorithm called Impulse optimization method, called `l0TV-PADMM, solves the (Total
Noise Removal Algorithm (INRA) which can remove impulse Variation) TV-based restoration problem with `l0-norm data
noise from corrupted images while preserving image details. fidelity. The benefit of utilizing this approach is that image de-
Impulse noise detection and impulse noise cancellation are the noising and de-blurring difficulties in the presence of impulsive
two steps followed in this algorithm. The goodness of this noise are better addressed. The stumbling block of this
algorithm is the efficiency, and it requires no previous training. technique is that this technique cannot support efficiently high -
The demerit of this algorithm is that, it fails to support gray - resolution images.
scale images of high noise level. Sonali et al. [41] present a noise removal and contrast
Y. Shih et al. [19] present a convection diffusion equation for enhancement algorithm for fundus image. Integration of filters
processing image de-noising, edge preservation and and Contrast Limited Adaptive Histogram Equalization
compression. In this method a PDE (Partial Differential (CLAHE) technique is applied for solving the issues of de-
Equations) based image restoration method called Convection noising and enhancement of color fundus images. The benefits
diffusion equation is used for image de-noising. The of utilizing this technique include the removal of noise and the
implementation gains its merit by removing noise without using enhancement of contrast in fundus images. The demerit of this
the nonlinear smoothing kernel which needs extra cost in technique is that it works only for the Medical domain and
solving the heat equation. The demerit of this method is that due restricted to fundus images
to the implementation of highly sophisticated method the time Xiaoqin Zhang et al. [43] put forth an Exemplar-based image
consumption and computational complexity is very high. de-noising algorithm (EIDA) which has shown better potential
S. Huang et al. [23] present an image restoration method (IRM) for image restoration. The goodness of using this algorithm is
for removing salt-and-pepper noise. This method concentrates that it shows better potential for image restoration. The pitfall of
on the removal of salt-and-pepper noise, where the noisy pixels this algorithm is that it is not dealt with multiple datasets. Also,
can take only the maximum and minimum values in the dynamic it cannot work efficiently for other types of images.
range. The goodness of this method is that it simplifies noisy Qi Wang et al. [46] describe a Fractional Differential Gradient
pixels detection. The demerit of this method is that there is a (FDG) approach for detecting noise locations in images, as well
possibility that some noise-free pixels may also be considered as an enhanced image de-noising algorithm based on fractional
as noisy pixels. integration. The merit of using this model is that it can remove
Zayed M. Ramadan [28] presents a method for Impulse Noise the noise and preserves the details of image edges in a better
Elimination and Edge Preservation (INEEP). In this paper, two manner. The demerit of this model is that it shows less
impulsive noise models are applied to multiple images with performance with other types of images.
various features, and a wide range of noise densities is explored. Lina Jia et al. [49] develop an image de-noising algorithm,
The benefit of this method is that it surpasses existing state-of- which is based on discriminative weighted nuclear norm
the-art methods in the literature of the image restoration field. minimization (DWNNM) in order to improve LDCT (Low-dose
The pitfall of this method is that there is a possibility of blurring computed tomography) image. This method shows better
of images because of high smoothing operation. performance in noise and artifacts removal, and also in details
Umesh Ghanekar et al. [29] introduce an Impulse Detection and structure preservation. The pitfall of this algorithm is that
Scheme (IDS) that detects all kinds of fixed-valued impulse the parameter is selected in a rough manner hence the de-noised
noise and distinguishes between noisy and noise-free pixels of images fail to achieve a better accuracy.
equal intensity levels. The difficulty of differentiating noisy

OBSTETRICS & GYNAECOLOGY FORUM 2024 | ISSUE 3s | 2925

 RESEARCH
O&G Forum 2024; 34-3s: 2922-2942

III PERLUSTRATION AND DISCUSSION

Table 1: Analysis on de-noising algorithm and its Expansion
PUBLICATION &
METHOD AUTHOR NAME DE-NOISING TECHNIQUES
YEAR
NNAF[7] ELSEVIER,1996 H. KONG et al. Neural Network Adaptive Filter
MMEM[8] IEEE,199 7 Wei-Yu Han et al. Minimum-Max i mu m Exclusive Mean
PSM[9] IEEE, 1999 Zhou Wang et al. Progressive Switching Median
F.J. Gallegos-Funes
MM-KNN[10] IEEE, 2002 M-type K-nearest neighbor
et al.
Modified VMF[11] ELSEVIER,2003 B. Smolka et al. Modified vector median filter
ATPMF[12] IEEE,2004 Xiaoyin Xu et al. Adaptive Two-Pass Rank Order Filter
Neighborhood Evaluated Adaptive Vector
NEAVF[13] ELSEVIER,2005 Zhonghua Ma et al.
Filter
Igor Aizenberg et
TBF[14] IEEE,2006 Threshold Boolean Filtering
al.
Novel Fuzzy Impulse Noise Detection
NFIND[15] IEEE,2007 Stefan Schulte et al.
Method
K. S. Srinivasan et
NLSP[16] IEEE,2007 Non-Linear signal processing technique
al.
INRA[17] ELSEVIER,2007 Wenbin Luo et al. Impulse Noise Removal Algorithm
ASM[18] IEEE,2008 X.M. Zhang et al. Adaptive Switching Mean
PDE[19] ELSEVIER,2009 Y. Shih et al. Partial Differential Equations
Adaptive Geometric Features Based Filtering
AGFF[20] IEEE,2009 Zhengya Xu et al.
Technique
ASWM[21] IEEE,2010 Smaïl Akkoul et al. Adaptive Switching Median
Kenny Kal Vin Toh Cluster-based Adaptive Fuzzy Switching
CAFSM[22] IEEE,2010
et al. Median
IRM[23] IEEE,2010 S. Huang et al. Image Restoration Method
Local Outlier Factor incorporating with
LOFBDND[24] IEEE,2011 Wei Wang et al Boundary Discriminative Noise Detection
Algorithm
FCA[25] ELSEVIER,2012 Sana Sadeghi et al. Fuzzy Cellular Automata
ADPF[26] IEEE,2012 Zhe Zhou Adaptive Detail-Preserving Filter
BDND[27] IEEE,2013 Iyad F. Jafar et al. Boundary Discriminative Noise Detection
Impulse Noise Elimination And Edge
INEEP[28] IEEE,2014 Zayed M. Ramadan
Preservation
Umesh Ghanekar et
IDS[29] ELSEVIER,2014 Impulse Detection Scheme
al.
CA[30] ELSEVIER,2014 U. Sahin et al. Cellular Automata
Ruixuang Wang et
RVIN[31] IEEE,2015 Random-Valued Impulse Noise
al.
Osama S.
ASWMF[32] ELSEVIER,2016 Adaptive Switching Weighted Median Filter
Faragallah et al.
ALD[33] IEEE,2016 Yi Wang et al. Absolute Luminance Difference Method
Qing-Qiang Chen et
NRA[34] IET,2017 Noise Removal Algorithm
al.
MF[35] IEEE,2018 Vikas Singh et al. Membership Function
Samsad Beagum
IDR[36] IET,2018 Impulse Detection and Restoration
Sheik Fareed et al.
Fariborz Taherkhani
RBF[37] IET,2017 Radial Basis Functions
et al.
Minghui Zhang et
ISIN[38] SPRINGER,2018 Iterative Scheme-Inspired Network
al.
Total Variation Proximal Alternating
l0TV-PADMM[39] IEEE,2018 Ganzhao Yuan et al.
Direction Method Of Multipliers
Adaptive Sequentially Weighted Median
ASWMF[40] IEEE,2019 Jiayi Chen et al.
Filter

OBSTETRICS & GYNAECOLOGY FORUM 2024 | ISSUE 3s | 2926

RESEARCH
O&G Forum 2024; 34-3s: 2922-2942

Contrast Limited Adaptive Histogram

CLAHE[41] ELSEVIER,2019 Sonali et al.
Equalization
CNN[42] ELSEVIER,2019 Lianghai Jin et.al Deep Convolutional Neural Network
Xiaoqin Zhang et Exemplar-Based Image De-Noising
EIDA[43] IEEE,2020
al. Algorithm
C. Jaspin Jeba Adaptive Switching Modified Decision Based
ASMDBUTMF[44] SPRINGER,2020
Sheela et al. Un-symmetric Trimmed Median Filter
Densely Connected Network for Impulse
DNINR[45] SPRINGER,2020 Guanyu Li et al.
Noise Removal
FDG[46] ELSEVIER,2020 Qi Wang et al. Fractional Differential Gradient
Spline Adaptive Filter based on the Robust
SAF-RGM[47] IEEE,2020 Qianqian Liu et al.
Geman-McClure estimator
Mustapha Bouhrara
NLML[48] IEEE,2017 Non Local Maximum Likelihood filter
et al.
Discriminative Weighted Nuclear Norm
DWNNM[49] IEEE,2018 Lina Jia et al.
Minimization
RNRM[50] IEEE,2019 Hongli Lv et al. Rician Noise Reduction Method
IMF[51] IEEE,2019 Ugur Erkan et al. Iterative Mean Filter
QWT[52] IEEE,2020 Rashid Ali et al. Quaternion Wavelet Transform
GAN[53] IEEE,2021 Miao Tian et al. Generative Adversarial Networks
Unsupervised
IEEE,2021 Swati Rai et al. Unsupervised Noise Learning Framework
NLF[54]
DOF[55] IEEE,2021 Huaian Chen et al. Demand-Oriented Framework
INRM[56] SPRINGER,2021 Chun Li et al. Impulse noise removal model algorithm
Golam Muktadir
MNRT[57] IJRES,2022 Weighted median filter
Mukti et al
De-noising algorithm based on Generative
UIDGAN [58] JTPES, 2024 XuYan et al.
Adversarial Networks

Table 1 represents the analysis on de-noising techniques that is CAFSM[22], FCA[25], IDR[30], SAP[33], MF[35] falls into
been used since the previous fifteen years. The total number of Fuzzy logic, NMEM[8], VMF[11], ASM[18], IDR[36], SAF -
papers considered for this study is fifty. The table also holds the RGM[47], NLML[48] techniques uses Mean based filter,
information regarding the author, the journal on which it g ets PSM[9], MM-KNN[10], ATPMF[12], NEAVF[13], NLSP[16],
published and also the published year. Also the de-noising AGFF[20], ASWM[21], LOFBDND[24], BDND[27 ],
techniques used in the previous fifteen years has been ASWMF[32], ASMDBUTMF[44] techniques uses Median
segregated and classified in to five groups’ namely Deep based filter, Rest of the other techniques like IDBF[14],
learning and neural network, Fuzzy logic, Mean based filter, INRA[17], PDE[19], IRM[23], INEEP[28], IDS[29],
Median based filter and Miscellaneous, and each technique falls RVIN[31], CA[34], l0TV-PADMM[39], CLAHE[41],
on to any appropriate group. Techniques like NNAF[7], EIDA[43], FDG[46], DWNNM[49] falls into Miscellaneous
ADPF[26], RBF[37], ISIN[38], DNINR[42], INRM[56] falls type.
into Deep learning and Neural network group, NFIND[1 5 ],

Table 2: Analysis on merits, demerits and MSE

S. NOISE RATIO;
METHODOLOGY MERITS DEMERITS
NO PSNR
Works effectively on highly Consumes large computational NOT
1. PSM[9]
corrupted images time MENTIONED
Better quality of image
Fails to support high-
2. MM-KNN[10] processing, both in the visual and 15% ; 25.29
resolution images
the analytical sense
Less reliable and degradation
3. Modified VMF[11] Low computational complexity 11.5% ; 38.074
of image quality is possible
Irregularities in the spatial
4. ATPMF[12] distribution of the estimated higher structural complexity 30% ; 37.522
impulse noise are detected
Loses its credibility while
5. NEAVF[13] Better accuracy in noise detection 20% ; 31.30
considering grayscale images

OBSTETRICS & GYNAECOLOGY FORUM 2024 | ISSUE 3s | 2927

 RESEARCH
O&G Forum 2024; 34-3s: 2922-2942

Smoothing of edges and

6. TBF[14] destruction of details are Textures are not preserved 30% ; 28.61
prevented
Does not introduce blurring Fails to reduce α-stable
7. NFIND[15] 40% ; 35.87
artifacts efficiently
Better performance across a wide Fails to support the switching
8. NLSP[16] 60% ; 36.32
range of noise densities window technique
Fails to remove noises of
9. INRA[17] Requires no previous training highly corrupted gray-scale 20% ; 37.36
images
Better performance in terms of
Not competitive to industrial
10. ASM[18] noise suppression and details 50% ; 33.76
standard
preservation
Removes noise without using the
nonlinear smoothing kernel Large time consumption and
11. PDE[19] 5% ; 27.077
which needs extra cost in solving computational complexity
the heat equation
Not integrated with other
Provides a very reliable impulse
benchmark techniques to
12. AGFF[20] noise type and ratio 20% ; 27.659
suppress a mixed Gaussian and
discrimination method
impulse noise contamination
Cannot support the switching
The psycho visual results are of
13. ASWM[21] window technique which leads 30% ; 32.91
high quality
to low performance
Capable in handling realistic Fails to support high-
14. CAFSM[22] impulse noise model for real- resolution images and huge 50% ; 27.45
world applications noise level
There is a possibility that some
NOT
15. IRM[23] Simplifies noisy pixels detection noise-free pixels can be
MENTIONED
misinterpreted as a noisy one
Minimizes the miss detection rate High computational
16. LOFBDND[24] 40% ; 33.95
and false detection rate complexity
Simplicity, robustness, parallel Eliminating noise from color
17. FCA[25] 60% ; 27.8
manner and distribution ability images is not supported
Edges may get blur if the
18. ADPF[26] Better Detection accuracy 60% ; 25.53
image has a high noise level
Large window size increases
noisy pixels are identified ideally
19. BDND[27] large computational 60% ; 34.45
by the detection step
complexity
The preservation of images with Texture areas are affected due
20. INEEP[28] fine details and edges are to blur, because of high 60% ; 30.86
maintained smoothing operation
Better performance for different since it supports two steps of
21. IDS[29] types of fixed valued impulse activity the time complexity 60% ; 31.8
noise increases
It is consistent and stable across a It does not support the
22. CA[30] 60% ; 30.5
wide range of noise densities switching window technique.
Better performance on nonpoint There is a trade-off between
23. RVIN[31] wise Random-Valued Impulse removing RVIN and 20% ; 30.58
Noise preserving fine texture details
Better performance for a wide set It cannot support images with
24. ASWMF[32] 40% ; 33.76
of images high noise level
It can suppress the noise even at
Texture areas can be affected
25. ALD[33] high noise ratios, and performs 50% ; 29.1793
due to blur
well in maintaining edges

OBSTETRICS & GYNAECOLOGY FORUM 2024 | ISSUE 3s | 2928

RESEARCH
O&G Forum 2024; 34-3s: 2922-2942

The accuracy is checked only

26. NRA[34] Better in term of the PSNR by using a limited quantity of 60% ; 32.89
test images
Filter preserves meaningful
Validated only for grayscale
27. MF[35] image details even at a high noise 50% ; 34.88
images
level
Larger the window size, the
Restoring image details are
28. IDR[36] computational complexity also 60% ; 32.25
maintained
increases
Tuning parameters by trial and
Fails to address Gaussian and
29. RBF[37] error to achieve the best result is 60% ; 31.23
Speckle noises
avoided
Better ability to decrease the Network is not dealt with
30. ISIN[38] noise quickly with the simple multiple datasets. 30% ; 27.93
iterative scheme
Image de-noising and de-blurring Not been developed in C++
31. l0TV-PADMM[39] in the presence of impulse noise hence it provides less speed 50% ; 22.4
are addressed in a better manner
Computational time is It is hard to be applied for real-
32. ASWMF[40] 50% ; 34.4
considerably low time de-noising
It works only for the Medical
Removes noise and enhances
33. CLAHE[41] domain and restricted to 0.2; 35.171
contrast in fundus images
fundus images
Running time of this method is
34. CNN[42] Better de-noising performance 20% ; 33.94
very high
Better potential for image It cannot work efficiently for
35. EIDA[43] 20% ; 35.072
restoration other types of images
It can be used as a preprocessing It lacks in providing the
36. ASMDBUTMF[44] method for scanning machines solution for the removal of 50% ; 33.867
for better robustness random noises
It loses its efficiency when
Better performance on edge
applied to other non-Gaussian
37. DNINR[45] preservation and noise 50% ; 31.08
noises like Poisson noise and
suppression
Rician noise
Remove the noise and preserves This model is evaluated only
38. FDG[46] the details of image edges in a by the minimum quantity of 60% ; 34.02
better manner test images
Better stable performance against high time consumption and NOT
39. SAF-RGM[47]
impulsive noise high computational complexity MENTIONED
Performance is limited in
Better performance for estimation NOT
40. NLML[48] spatially heterogeneous
of noise SD MENTIONED
regions
Parameter is selected in a
Better performance in noise and rough manner hence the de-
41. DWNNM[49] 50% ; 25.1314
artifacts removal noised images fail to achieve a
better accuracy
Better performance in terms of larger computational time
42. RNRM[50] objective metrics and visual which in term increases the 15% ; 30.69
inspection complexity
Works better than the methods Fails to remove the random-
43. IMF[51] 60% ; 32.49
using dynamic adaptive windows valued impulse noise
It is not been tested with
The de-noised images have the
44. QWT[52] different types of filters and 75% ; 34.28
finest visual quality
mixed noises
Better in terms of de-noising
Not tested upon the real
45. GAN[53] level, SSIM (structural similarity 10% ; 34.62
hospital environment
index)

OBSTETRICS & GYNAECOLOGY FORUM 2024 | ISSUE 3s | 2929

 RESEARCH
O&G Forum 2024; 34-3s: 2922-2942

Does not require the clean (de- Fails to address any

Unsupervised
46. noised) images for training the degradation in the multimodal 15% ; 35.185
NLF[54]
model images along with the noise
15% ; 30.69
Better performance in terms of The network needs to be
47. DOF[55] the number of parameters and the retrained when faced with
de-noising quality different demands
Fails to address the inverse
Better than existing classic problem such as image
48. INRM[56] algorithms for impulse signal patching problems, image 40%; 41.21
removal segmentation problems, image
blending to noise
It provides high quality images It loses its efficiency while
75%; 58.93
49. MNRT[57] by removing salt and pepper working with the kernel size
noises above three
It considers many parameters
The consistency of content
50. UIDGAN[50] which in turn increases its 60%; 45.27
information is maintained
complexity

Merits of De-Noising Techniques

Performance
11%
24% Nil Training
13% Accuracy

4% Advanced Techniques
8%
Retaining image details
17%
15% Edge Preservation
8%
Other Benefits
User friendly

Fig. 1: Merits of De-Noising techniques.

De-Merits of De-Noising techniques

Window size and resolution

Performance
16% 4%
23% Works on a particular
9% technique
Less powered technique
7%
9%
14% Complexity
18%
Time Consumption

Real time implementation

Fig. 2: De-Merits of De-noising techniques.

OBSTETRICS & GYNAECOLOGY FORUM 2024 | ISSUE 3s | 2930

RESEARCH
O&G Forum 2024; 34-3s: 2922-2942

Table 2 and Fig (1) & (2) describes the Merit and the short [31],[33],[44], [48],[52], works on a particular technique
coming of the de-noising technique that is been considered in [16],[30],[45],[51],[56], less power techniques [20],[21],
our study. The merits of the implemented noise reduction [25],[34], [37],[38], [39],[46], [49],[52], complexity [6],[7],
techniques include performance [10], [16], [18], [31], [32], [42], [19],[24], [27],[29], [36],[47], time consumption [9],[42],
[47], [48], [49], [50], [55], [56], nil training [54], [17], accuracy [47],[50], real time implementation [18],[40],[50],[53],[55 ],
[13], [53], [41], [34], [20], [21], [24], [26], advanced techniques works on a particular type of images [8],[10], [13],[17],
[51], [39], [22], [9], retaining image details [7], [14], [18], [35], [32],[35], [41],[43], [48]. This table also has the details of noise
[36], [43], [46], edge preservation [14], [28], [33], [45], and a ratio considered and also the achieved psnr value.
user-friendly approach [11], [23], [25], [37], [38], [40]. Some of Fig 3 gives a detailed chart on the various noise reduction
the short comings of these techniques were the window size and techniques that is been used on various corrupted noise level and
resolution [27],[36], performance [11],[14], [15],[20], [26],[28], the PSNR value obtained for each techniques.

OBSTETRICS & GYNAECOLOGY FORUM 2024 | ISSUE 3s | 2931

 RESEARCH
O&G Forum 2024; 34-3s: 2922-2942

PSNR
UIDGAN[58]
MNRT[57]
INRM[56]
DOF[55]
Unsupervised NLF[54]
GAN[53]
QWT[52]
IMF[51]
RNRM[50]
DWNNM[49]
FDG[46]
DNINR[45]
ASMDBUTMF[44]
EIDA[43]
CNN[42]
CLAHE[41]
ASWMF[40]
l0TV-PADMM[39]
ISIN[38]
RBF[37]
IDR[36]
MF[35]
NRA[34]
ALD[33]
ASWMF[32] PSNR
RVIN[31]
CA[30]
IDS[29]
INEEP[28]
BDND[27]
ADPF[26]
FCA[25]
LOFBDND[24]
CAFSM[22]
ASWM[21]
AGFF[20]
PDE[19]
ASM[18]
INRA[17]
NLSP[16]
NFIND[15]
TBF[14]
NEAVF[13]
ATPMF[12]
Modified VMF[11]
MM-KNN[10]

0 10 20 30 40 50 60 70
Fig. 3: Implemented noise reduction technique and the PSNR value obtained.
Table 3: Methodoloy and Types of Dataset
S. CORRUPTED NOISE
METHODOLOGY TYPES OF IMAGES IMAGE SIZE
NO PERCENTAGE
1.Corrupted
1. PSM[9] Pepper 512 X 512 5% to 70%
2.Bridge
2. MM-KNN[10] 1.Lenna 256 X 256 15%

OBSTETRICS & GYNAECOLOGY FORUM 2024 | ISSUE 3s | 2932

RESEARCH
O&G Forum 2024; 34-3s: 2922-2942

500 X 500
1.Lenna
500 X 500
3. Modified VMF[11] 2.Peppers 11.5%
500 X 500
3. Gold hill etc…
1.Lenna
4. ATPMF[12] 625 X 625 20% to 35%
2.Boat
1.Lenna
5. NEAVF[13] 256 X 256
2.Parrot 0.5% to 20%
6. TBF[14] 1.Girl 256 X 256 1% to 30%
1.Lenna 256 X 256
2.Baboon 256 X 256
7. NFIND[15]
3. Parrot 175 X 150 5% to 40%
4. Boat 300 X 300
1. Lenna
8. NLSP[16] 2. Girl
256 X 256 10% to 90%
1.Lenna
2. Bridge
9. INRA[17] 20 %
3. Gold hill 512 X 512
Etc …
1.Pepper 512 X 512
10. ASM[18] 10 % to 70%
2.Bridge
1.License plate Image 198 X 85
11. PDE[19] 2.Spade-Heart-Diamond-Club Image 257 X 257
3.Elaine 257 X 257 5% to 20%
1.Boat
512 X 512
12. AGFF[20] 2.Zoomed portion of a Parrot
1986 X 1986 5% to 50%
Etc…
1. Lenna
13. ASWM[21] 2. Boat Not Mentioned
10 % to 60%
3. Pepper etc…
14. CAFSM[22] 100 Grayscale test images 512 X 512 5% to 50%
15. IRM[23] Grey-scale Lena image 512 X 512 10% to 90%
1.Lenna
2. Gold hill
16. LOFBDND[24] 512 X 512 10% to 90%
3. Boat
4. Bridge
1.Lenna
17. FCA[25] 2.Peppers Not Mentioned 10% to 80%
3.Baboon
1. Lenna
2. Corrupted Bridge
18. ADPF[26]
3. Peppers 512 X 512 10% to 90%
4. Baboon
1. Camera man
19. BDND[27] 2. Peppers Not Mentioned
10% to 90%
3. Boat etc…
1. Bridge
20. INEEP[28] 2. Mammogram Not Mentioned
4% to 60%
3. Compound Eye of Fly.
1. Lenna
21. IDS[29] 2. Peppers
512 X 512 10% to 60%
3. Baboon
1. Lenna 256 X 256
22. CA[30] 10% to 70%
2. Bridge 512 X 512
1.Baboon
23. RVIN[31] 2. Finger
300 X 300 10% to 20%
3. Bridge etc…

OBSTETRICS & GYNAECOLOGY FORUM 2024 | ISSUE 3s | 2933

 RESEARCH
O&G Forum 2024; 34-3s: 2922-2942

24. ASWMF[32] Gray Scale Images 256 X 256 20% to 90%

1. Hill
25. ALD[33]
2. Lenna etc… 512 X 512 90%
1. Lenna
26. NRA[34] 512 X 512 10% to 90%
2. Baboon
1. Lenna
2. Corrupted Bridge
27. MF[35]
3. Peppers 512 X 512 20% to 99%
4. Baboon etc…
28. IDR[36] Standard Grey-Scale Images Not Mentioned 40% to 90%
29. RBF[37] 8-bit standard grey-scale images 512 X 512 10% to 95%
Images from Berkeley Segmentation Not Mentioned
30. ISIN[38] 20% to 40%
Dataset
Gray scale and colored images of a
31. l0TV-PADMM[39] 512 X 512 10 % to 90%
Camera Man, Lenna etc…
BSD68 DATASET
32. ASWMF[40] Not Mentioned 10 % to 90%
Containing Medical Images
Red, Blue and Green Channel of Fundus
33. CLAHE[41]
Images 605 X 700 Up to 20%
400 Images from Berkeley segmentation 5% to 60%
34. CNN[42] 180 X 180
dataset
1. Monarch
35. EIDA[43] 2. Barbara
256 X 256 10% to 30%
3. Monarch etc…
Medical Databases namely cancer
Imaging Archive (TCIA) and real time
36. ASMDBUTMF[44] Not Mentioned Up to 99%
database from Kerala Institute of Medical
Science (KIMS)
1.Foreman
2. Bottom
37. DNINR[45]
3. Pentagon 256 X 256 30% to 80%
4. Pepper etc…
38. FDG[46] Lenna images Not Mentioned Up to 50%
39. SAF-RGM[47] Gaussian signal and colored signal Nil Not Mentioned
T2-weighted (T2W) images of human
40. NLML[48] 200 X 180 Not Mentioned
brain
41. DWNNM[49] Low Dose CT Images Nil Up to 80%
42. RNRM[50] 3D MR data 181 × 217 × 181 1% to 15%
43. IMF[51] Peppers Image 512 X 512 Up to 90%
256 X 256
1.Lenna
512 X 512
44. QWT[52] 2.Corrupted 3.Bridge
1024X 1024
4. Peppers etc… 15% to 75%
Synthetic Data obtained from Brain Web
45. GAN[53] 181 X 217 X 181 1% to 10%
dataset
Unsupervised 64 X 64
46. MRI, CT, and LDCT images 5% to 15%
NLF[54] 512 X 512
256 X 256
47. DOF[55] Berkeley segmentation dataset (BSD500) 15% to 70%
512 X 512
Natural images, CT images and MRI Up to 40%
48. INRM[56] NIL
images.
49. MNRT[57] MRI images 256X256 Up to 75%
A Self-Guided Deep Learning Technique
50. UIDGAN[58] NIL Up to 60%
for MRI Image Noise Reduction”, JTPES

OBSTETRICS & GYNAECOLOGY FORUM 2024 | ISSUE 3s | 2934

RESEARCH
O&G Forum 2024; 34-3s: 2922-2942

UIDGAN[58]
MNRT[57]
INRM[56]
DOF[55]
Unsupervised NLF[54]
GAN[53]
QWT[52]
IMF[51]
RNRM[50]
DWNNM[49]
FDG[46]
DNINR[45]
ASMDBUTMF[44]
EIDA[43]
CNN[42]
CLAHE[41]
ASWMF[40]
l0TV-PADMM[39]
ISIN[38]
RBF[37]
IDR[36]
MF[35]
Methodology

NRA[34]
ALD[33]
ASWMF[32]
RVIN[31] NOISE RATIO
CA[30]
IDS[29]
INEEP[28]
BDND[27]
ADPF[26]
FCA[25]
LOFBDND[24]
CAFSM[22]
ASWM[21]
AGFF[20]
PDE[19]
ASM[18]
INRA[17]
NLSP[16]
NFIND[15]
TBF[14]
NEAVF[13]
ATPMF[12]
Modified VMF[11]
MM-KNN[10]

0 20 40 60 80 100 120
Percentage

Fig 4: Implemented Methodology Vs Percentage of noise ratio

Fig 4 represented the pictorial representation of various methodologies that is been used on various corrupted noise level since
the past few years.

OBSTETRICS & GYNAECOLOGY FORUM 2024 | ISSUE 3s | 2935

 RESEARCH
O&G Forum 2024; 34-3s: 2922-2942

Resolution of Input images

480x480

2%3% 256x256
2%2%
3%
3% 512x512
5%

34% 300x300

1024x1024

625x625

605x700
46%
180x180

1986x1986

Fig. 5: Chart on the various resolutions of input images that is been used on various denoising techniques.

Table 3 and Fig (5) describes the various de-noising techniques 14. CAFSM[22] High
that is been used on different noise level. It also holds the 15. IRM[23] High
information about the types of data sets that is been used as an 16. LOFBDND[24] High
input and the resolution of the various types of input images. It 17. FCA[25] Medium
also describes the details of the Noise ratio percentage. From 18. ADPF[26] High
this chart it is clear that most of the noise reduction techniques 19. BDND[27] High
have considered 512 x 512 resolution as its maximum inputs. 20. INEEP[28] Very High
Apart from input images of different resolution certain 21. IDS[29] Very High
techniques also considered databases like Medical Databases 22. CA[30] Very High
namely cancer Imaging Archive (TCIA) and real time database 23. RVIN[31] Very High
from Kerala Institute of Medical Science (KIMS), BSD68
24. ASWMF[32] Very High
DATASET Containing Medical Images, Images from Berkeley
25. ALD[33] Medium
segmentation dataset, Synthetic Data obtained from Brain Web
26. NRA[34] Very High
dataset.
27. MF[35] Very High
Table 4: Performance Evaluation 28. IDR[36] Very High
29. RBF[37] Excellent
S.No METHODOLOGY PERFORMANCE 30. ISIN[38] High
1. PSM[9] Medium 31. l0TV-PADMM[39] Very High
2. MM-KNN[10] Medium 32. ASWMF[40] Very High
3. Modified VMF[11] Fair 33. CLAHE[41] Very High
4. ATPMF[12] Fair 34. CNN[42] Very High
5. NEAVF[13] Medium 35. EIDA[43] Excellent
6. TBF[14] Fair 36. ASMDBUTMF[44] Excellent
7. NFIND[15] Fair 37. DNINR[45] Excellent
8. NLSP[16] Fair 38. FDG[46] Excellent
9. INRA[17] High 39. SAF-RGM[47] High
10. ASM[18] High 40. NLML[48] Excellent
11. PDE[19] High 41. DWNNM[49] Excellent
12. AGFF[20] High 42. RNRM[50] Excellent
13. ASWM[21] High 43. IMF[51] Excellent

OBSTETRICS & GYNAECOLOGY FORUM 2024 | ISSUE 3s | 2936

RESEARCH
O&G Forum 2024; 34-3s: 2922-2942

44. QWT[52] Excellent

49. MNRT[57] Excellent
45. GAN[53] Very High
46. Unsupervised NLF[54] Excellent 50. UIDGAN[58] Excellent
47. DOF[55] Excellent
48. INRM[56] Excellent

OBSTETRICS & GYNAECOLOGY FORUM 2024 | ISSUE 3s | 2937

 RESEARCH
O&G Forum 2024; 34-3s: 2922-2942

UIDGAN[58]
MNRT[57]
INRM[56]
DOF[55]
Unsupervised NLF[54]
GAN[53]
QWT[52]
IMF[51]
RNRM[50]
DWNNM[49]
FDG[46]
DNINR[45]
ASMDBUTMF[44]
EIDA[43]
CNN[42]
CLAHE[41]
ASWMF[40]
l0TV-PADMM[39]
ISIN[38]
RBF[37]
IDR[36]
MF[35]
NRA[34]
ALD[33] PSNR
ASWMF[32]
RVIN[31] NOISE RATIO IN PERCENTAGE
CA[30]
IDS[29]
INEEP[28]
BDND[27]
ADPF[26]
FCA[25]
LOFBDND[24]
CAFSM[22]
ASWM[21]
AGFF[20]
PDE[19]
ASM[18]
INRA[17]
NLSP[16]
NFIND[15]
TBF[14]
NEAVF[13]
ATPMF[12]
Modified VMF[11]
MM-KNN[10]

0 20 40 60 80
Fig. 6: PSNR Vs. Percentage of noise ratio.

OBSTETRICS & GYNAECOLOGY FORUM 2024 | ISSUE 3s | 2938

RESEARCH
O&G Forum 2024; 34-3s: 2922-2942

UIDGAN[58]
MNRT[57]
INRM[56]
DOF[55]
Unsupervised NLF[54]
GAN[53]
QWT[52]
IMF[51]
RNRM[50]
DWNNM[49]
NLML[48]
SAF-RGM[47]
FDG[46]
DNINR[45]
ASMDBUTMF[44]
EIDA[43]
CNN[42]
CLAHE[41]
ASW MF[40]
l0TV-PADMM[39]
IS IN[38]
RBF[37]
IDR[36]
MF[35]
NRA[34]
ALD[33]
ASW MF[32]
RVIN[31]
CA[30]
IDS[29]
INEEP[28]
BDND[27]
ADPF[26]
FCA [25]
LOFBDND[24]
IRM[23]
CAFSM[22]
ASW M[21]
AGFF[20]
PDE[19]
ASM[18]
INRA[17]
NLSP[16]
NFIND[15]
TBF[14]
NEAVF[13]
ATPMF[12]
Modifi ed VMF[11]
MM-KNN[10]
PSM[9]

0 1 2 3 4 5 6

Performance Evaluation

Fig. 7: Performance Evaluation of various noise reduction techniques on various corrupted noise level which is based on its
corresponding PSNR values.

Table 4 describes the various noise detection and reduction techniques and Fig (7) displays the performance evaluation of
techniques. It also carries the information of its performance each techniques by considering PSNR in to account. From the
evaluation. Herein Fig (6) exhibits the pictorial representation table 4 and Fig (7) it is clear that the recent techniques like
of PSNR values on various noise ratio used in various Demand-Oriented Framework (DOF) [55] and Impulse noise

OBSTETRICS & GYNAECOLOGY FORUM 2024 | ISSUE 3s | 2939

 RESEARCH
O&G Forum 2024; 34-3s: 2922-2942

removal model algorithm (INRM) [56] performs better. The 8. Wei-Yu Han and Ja-Chen Lin, “Minimum-Maximum
weighted median filter technique (MNRT) [57] outperforms all Exclusive Mean (MMEM) filter to remove impulse noise from
other filters and methods and has high PSNR value than the state highly corrupted images”, IEEE electronics letter, vol. 33,
of the art method. issue 2, pp. 124–125, Jan. 1997.
IV CONCLUSION 9. Zhou Wang and David Zhang, “Progressive switching
Image de-noising research continues to be in great demand as median filter for the removal of impulse noise from highly
the complexity and requirements of the process have escalated. corrupted images’’, IEEE transactions on circuits and
This study pours light on the virtues and downsides of multiple systems—i: analog and digital signal processing, vol. 46, issue
image de-noising algorithms that have been developed in the 1, pp. 78–80, Jan. 1999.
past few years. The advent of techniques has recently 10. F.J. Gallegos-Funes, V.I. Ponomaryov, S. Sadovnychiy
supplanted the old local de-noising model, resulting in a new and L. Nino-de-Rivera, “Median M-type K-nearest neighbour
theoretical branch and substantial breakthroughs in image de- (MM-KNN) filter to remove impulse noise from corrupted
noising approaches, such as sparse representation, low-rank, and images”, IEEE electronics letter, vol. 38, issue 15, pp. 786 -787,
CNN (more precisely, deep learning) based methods. The July 2002.
purpose of this study is to provide an overview of the different 11. B. Smolka, R. Lukac, A. Chydzinski, K.N. Plataniotis
de-noising methods. Also this study categorizes each and W. Wojciechowski, “Fast adaptive similarity-based
methodology in to five groups. Hence INRM [56] from Deep impulsive noise reduction filter”, ELSEVIER real-time
learning and Neural network, NFIND [15] from Fuzzy logic, imaging, vol. 9, issue 4, pp. 261-276, Aug. 2003.
VMF [11] from Mean based filter, ATPMF [12] from Median 12. Xiaoyin Xu, Eric L. Miller, Dongbin Chen and Mansoor
based filter and INRA [17] are considered to be the favorable Sarhadi, “Adaptive two-pass rank order filter to remove
method holding high PSNR value. On considering the entire impulse noise in highly corrupted images”, IEEE transaction
techniques in our study recent techniques like Demand -Oriented on image processing, vol. 13, issue 2, pp. 238-247, Feb. 2004.
Framework (DOF) [55] and Impulse noise removal model 13. Zhonghua Ma, Dagan Feng and Hong Ren Wu, “A
algorithm (INRM) [56] performs second best. The weighted neighborhood evaluated adaptive vector filter for suppression
median filter technique (MNRT) [57] performs the best and of impulse noise in color images”, ELSEVIER, real -time
gives high PSNR value. Because different types of noise imaging, vol. 11, issue 5-6, pp. 403–416, Oct-Dec. 2005.
necessitate different de-noising approaches, noise analysis can 14. Igor Aizenber, Constantine Butakoff and Dmitriy Paliy,
aid in the development of novel de-noising schemes. “Impulsive noise removal using threshold boolean filtering
based on the impulse detecting functions” IEEE signal
References processing letter, vol. 12, issue 1, pp. 63-66, Jan. 2006.
1. W. Zhao and H. Lu, "Medical Image Fusion and 15. Stefan Schulte, Samuel Morillas, Valentín Gregori, and
Denoising with Alternating Sequential Filter and Adaptive Etienne E. Kerre, “A new fuzzy color correlated impulse noise
Fractional Order Total Variation," in IEEE Transactions on reduction method”, IEEE transactions on image processing,
Instrumentation and Measurement, vol. 66, no. 9, pp. 2283 - vol. 16, issue 11, pp. 2565-2575, March 2007.
2294, Sept. 2017. 16. K. S. Srinivasan and D. Ebenezer, “A New Fast and
2. M. P. Nguyen and S. Y. Chun, "Bounded Self-Weights Efficient Decision-Based Algorithm for Removal of High-
Estimation Method for Non-Local Means Image De-noising Density Impulse Noises”, IEEE signal processing letters,
Using Minimax Estimators," in IEEE Transactions on Image vol.14, issue.3, pp. 189 – 192, Mar. 2007.
Processing, vol. 26, no. 4, pp. 1637-1649, April 2017 17. Wenbin Luo, “An efficient algorithm for the removal of
3. A. Kumar, M. O. Ahmad and M. N. S. Swamy, "A impulse noise from corrupted Images”, ELSEVIER, vol.61,
Framework for Image Denoising Using First and Second issue 83, pp. 551-555, September 2007.
Order Fractional Overlapping Group Sparsity (HF-OLGS) 18. X.M. Zhang, Z.P. Yin and Y.L. Xiong, “Adaptive
Regularizer," in IEEE Access, vol. 7, pp. 26200-26217, 2019. switching mean filter using
4. A. F. M. S. Uddin, T. Chung and S. Bae, "A Perceptually conditional morphological noise detector”, IEEE electronics
Inspired New Blind Image Denoising Method Using L1 and letters, vol.44, issue 6, pp. 406-407, Mar. 2008.
Perceptual Loss," in IEEE Access, vol. 7, pp. 90538 -90549, 19. Y. Shih, C. Rei, H. Wang, “A novel PDE based image
2019. restoration: Convection diffusion equation for image de-
5. Z. Kong and X. Yang, "Color Image and Multispectral noising”, ELSEVIER ,vol.231, issue 215, pp. 771 – 779, 2009.
Image Denoising Using Block Diagonal Representation," in 20. Zhengya Xu, Hong Ren Wu, Bin Qiu and Xinghuo Yu,
IEEE Transactions on Image Processing, vol. 28, no. 9, pp. “Geometric Features-Based Filtering for Suppression of
4247-4259, Sept. 2019. Impulse Noise in Color Images”, IEEE transactions on image
6. H. Chen, Y. Jin, M. Duan, C. Zhu and E. Chen, "DOF: processing, vol. 18, issue 8, pp. 1742 – 1759, Aug. 2009.
A Demand-Oriented Framework for Image Denoising," in 21. Smaïl Akkoul, Roger Lédée, Remy Leconge and Rachid
IEEE Transactions on Industrial Informatics, vol. 17, no. 8, pp. Harba, “A new adaptive switching median filter”, IEEE signal
5369-5379, Aug. 2021. processing letters, vol. 17, issue 6, pp. 587–590, June 2010.
7. H. Kong and L. Guan, “A neural network adaptive filter 22. Kenny Kal Vin Toh and Nor Ashidi Mat Isa, “Cluster-
for the removal of impulse noise in digital images”, based adaptive fuzzy switching median filter for universal
ELSEVIER, neural networks, vol. 9, issue 3, pp. 373-378, April impulse noise reduction”, IEEE transactions on consumer
1996. electronics, vol. 56, issue 4, pp. 2560–2568, Nov. 2010.

OBSTETRICS & GYNAECOLOGY FORUM 2024 | ISSUE 3s | 2940

RESEARCH
O&G Forum 2024; 34-3s: 2922-2942

23. S. Huang and J. Zhu, “Removal of salt-and-pepper functions interpolation”, IET image processing, vol. 12, issue
noise based on compressed sensing”, IEEE electronics letters, 1, pp. 20–30, Nov. 2017.
vol. 46, issue 17, pp. 1198–1199, Aug. 2010. 38. Minghui Zhang, Yiling Liu, Guanyu Li, Binjie Qin and
24. Wei Wang and Peizhong Lu, “An efficient switching Qiegen Liu, “Advances Iterative scheme inspired network for
median filter based on local outlier factor”, IEEE signal impulse noise removal”, SPRINGER, pattern analysis and
processing letters, vol. 18, issue 10, pp. 551–554, Oct. 2011. applications, vol. 23, issue 1, pp. 135-145, Nov. 2018.
25. Sana Sadeghi, Alireza Rezvanian and Ebrahim 39. Ganzhao Yuan and Bernard Ghanem, “l0TV: a sparse
Kamrani, “An efficient method for impulse noise reduction optimization method for impulse noise image restoration”,
from images using fuzzy cellular automata”, ELSEVIER AEU - IEEE transactions on pattern analysis and machine
international journal of electronics and communications, vol. intelligence, vol. 41, issue 2, pp. 352-364, Dec. 2018.
66, issue 9, pp. 772–779, Jan. 2012. 40. Jiayi Chen, Yinwei Zhan and Huiying Cao, “Adaptive
26. Zhe Zhou, “Cognition and removal of impulse noise sequentially weighted median filter for image highly corrupted
with uncertainty”, IEEE transactions on image processing, vol. by impulse noise”, IEEE access, vol. 7, pp. 158545 –158556,
21, issue 7, pp. 3157–3167, July 2012. Oct. 2019.
27. Iyad F. Jafar, Rami A. AlNa’mneh and Khalid A. 41. Sonali, Sima Sahu, Amit Kumar Singh, S.P. Ghrera and
Darabkh, “Efficient improvements on the BDND filtering Mohamed Elhoseny, “An approach for de-noising and contrast
algorithm for the removal of high density impulse noise”, IEEE enhancement of retinal fundus image using CLAHE”,
transactions on image processing, vol. 22, issue 3, pp. 1223 – ELSEVIER, optics & laser technology, vol. 110, pp. 87-98, Feb.
1232, March 2013 2019.
28. Zayed M. Ramadan, “A new method for impulse noise 42. Lianghai Jin, Wenhua Zhang, Guangzhi Ma and Enmin
elimination and edge preservation”, IEEE Canadian Journal Song, “Learning deep CNNs for impulse noise removal in
of Electrical and Computer Engineering, vol. 37, issue 1, pp. images”, ELSEVIER, journal of visual communication and
2–10, 2014. image representation, vol. 62, pp. 193-205, July 2019.
29. Umesh Ghanekar and Rajoo Pandey, “An intensity 43. Xiaoqin Zhang, Jingjing Zheng, Di Wang and Li Zhao,
independent fixed valued impulse noise detector for image “Exemplar-based denoising: a unified low-rank recovery
restoration”, ELSEVIER, AEU-international journal of framework”, IEEE transactions on circuits and systems for
electronics and communications, vol. 68, issue 3, pp. 210 –215, video technology, vol. 30, issue 8, pp. 2538-2549, Aug. 2020.
March 2014. 44. C. Jaspin Jeba Sheelaa and G. Suganthi, “An efficient
30. U. Sahin and S. Uguz, F. Sahin, “Salt and pepper noise de-noising of impulse noise from MRI using adaptive switching
filtering with fuzzy-cellular automata”, ELSEVIER, computers modified decision based un-symmetric trimmed median filter”,
& electrical engineering, vol. 40, issue 1, pp. 59–69, Jan. 2014. ELSEVIER biomedical signal processing and control, vol. 55,
31. Ruixuang Wang, Markus Pakleppa and Emanuele pp. 1-12, Jan. 2020.
Trucco, “Low-rank prior in single patches for nonpoint wise 45. Guanyu Li, Xiaoling Xu, Minghui Zhang and Qiegen
impulse noise removal”, IEEE transactions on image Liu, “Densely connected network for impulse noise removal”,
processing, vol. 24, issue 5, pp. 1485–1496, May 2015. SPRINGER, pattern analysis and applications, vol. 23, issue 3,
32. Osama S. Faragallah and Hani M. Ibrahem, “Adaptive pp. 1263–1275, Feb. 2020.
switching weighted median filter framework for suppressing 46. Qi Wang, Jing Ma, Siyuan Yu and Liying Tan, “Noise
salt-and-pepper noise”, ELSEVIER, AEU-international detection and image de-noising based on fractional calculus”,
journal of electronics and communications, vol. 70, issue 8, pp. ELSEVIER, chaos, solitons & fractals, vol. 131, Feb. 2020.
1034–1040, Aug. 2016. 47. Qianqian Liu and Yigang He, “Robust geman-mcclure
33. Yi Wang, Jiangyun Wang, Xiao Song and Liang Han, based nonlinear spline adaptive filter against impulsive noise”,
“An efficient adaptive fuzzy switching weighted mean filter for IEEE access, vol. 8, pp. 22571–22580, Feb. 2020.
salt-and-pepper noise removal”, IEEE signal processing 48. Mustapha Bouhrara, J.M.Bonny, B.G.Ashinsky,
letters, vol. 23, issue 11, pp. 1582–1586, Nov. 2016. M.C.Maring and R.G.Spencer, "Noise Estimation and
34. Qing-Qiang Chen, Mao-Hsiung Hung and Fumin Zou, Reduction in Magnetic Resonance Imaging Using a New
“Effective and adaptive algorithm for pepper and salt noise Multispectral Nonlocal Maximum-likelihood Filter", IEEE
removal”, IET image processing, vol. 11, issue 9, pp. 709–716, Transactions on Medical Imaging, vol. 36, issue 1, pp . 181-
Feb. 2017. 193, Jan. 2017.
35. Vikas Singh, Raghav Dev, Narendra K. Dhar, Pooja 49. Lina Jia, Quan Zhang, Yu Shang, Yanling Wang, Yi Liu,
Agrawal and Nishchal K. Verma, “Adaptive type-2 fuzzy Na Wang, Zhiguo Gui, Guanru Yang, “De-noising for Low-
approach for filtering salt and pepper noise in grayscale Dose CT Image by Discriminative Weighted Nuclear Norm
images”, IEEE transactions on fuzzy systems, vol. 26, issue 5, Minimization," IEEE Access, vol. 6, pp. 46179-46193, August
pp. 3170-3176, Oct. 2018. 2018.
36. Samsad Beagum Sheik Fareed and Sheeja Shaik 50. Hongli Lv and R. Wang, "De-noising 3D Magnetic
Khader, “Fast adaptive and selective mean filter for the Resonance Images Based on Low-Rank Tensor Approximation
removal of high-density salt and pepper noise”, IET image With Adaptive Multi-rank Estimation," IEEE Access, vol. 7, pp.
processing, vol. 12, issue 8, pp. 1378–1387, Aug. 2018. 85995-86003, June 2019.
37. Fariborz Taherkhani and Mansour Jamzad, “Restoring 51. Ugur Erkan, D. N. H. Thanh, L. M. Hieu and S.
highly corrupted images by impulse noise using radial basis Engínoğlu, "An Iterative Mean Filter for Image Denoising,"
IEEE Access, vol. 7, pp. 167847-167859, November 2019.

OBSTETRICS & GYNAECOLOGY FORUM 2024 | ISSUE 3s | 2941

 RESEARCH
O&G Forum 2024; 34-3s: 2922-2942

52. Rashid Ali, P. Yunfeng and R. U. Amin, "A Novel 56. Chun Li1, Jian Li, Ze Luo, “An impulse noise removal
Bayesian Patch-Based Approach for Image Denoising ", IEEE model algorithm based on logarithmic image prior for medical
Access, vol. 8, pp. 38985-38994, Feb. 2020. image”, SPRINGER, Signal image and video processing, issue
53. M. Tian and K. Song, "Boosting Magnetic Resonance 15, pp. 1145-1152, Jan. 2021.
Image De-noising With Generative Adversarial Networks", 57. G.M. Mukti, Maniruzzaman M.A. Alahe, A.Sarka,
IEEE Access, vol. 9, pp. 62266-62275, April 2021. “Noise Removal from MRI Brain ImagesUsing
54. Swati Rai, J. S. Bhatt and S. K. Patra, "Augmented MedianFiltering Techniques”, IJRES, vol.10, issue 6, pp. 736 -
Noise Learning Framework for Enhancing Medical Image De- 743, Sep. 2022.
noising", IEEE Access, vol. 9, pp. 117153-117168, August 58. X.Yan, M.X.Xiao, W.Wang, Y. Li, F.Zhang, “A Self-
2021. Guided Deep Learning Technique for MRI Image Noise
55. Huaian Chen, Y. Jin, M. Duan, C. Zhu and E. Chen, Reduction”, JTPES, vol. 4, issue 1, ISSN: 2790-1505, 2024
"DOF: A Demand-Oriented Framework for Image Denoising",
IEEE Transactions on Industrial Informatics, vol. 17, no. 8, pp.
5369-5379, Aug. 2021.

OBSTETRICS & GYNAECOLOGY FORUM 2024 | ISSUE 3s | 2942