A Study on Image Segmentation Method for Image Processing

Abstract

Image processing techniques are essential part of the current computer technologies and that it plays vital role in various applications like medical field, object detection, video surveillance system, computer vision etc. The important process of Image processing is Image Segmentation. Image Segmentation is the process of splitting the images into various tiny parts called segments. Image processing makes to simplify the image representation in order to analyze the images. So many algorithms are developed for segmenting images, based on the certain feature of the pixel. In this paper different algorithms of segmentation can be reviewed, analyzed and finally list out the comparison for all the algorithms. This comparison study is useful for increasing accuracy and performance of segmentation methods in various image processing domains.

Full text

A Study on Image Segmentation Method
for Image Processing
S. PRABUa 1 and J.M. GNANASEKARb
a
Associate Professor Department of Computer Science & Engineering, VI Institute of
Technology Sirunkundram-603108, Chengleput Distirct, TamilNadu, India
b
Professor, Department of Computer Science & Engineering, SriVenkateswara
College of Engineering, Sriperumbadur, TamilNadu, India
Abstract. Image processing techniques are essential part of the current computer
technologies and that it plays vital role in various applications like medical field,
object detection, video surveillance system, computer vision etc. The important
process of Image processing is Image Segmentation. Image Segmentation is the
process of splitting the images into various tiny parts called segments. Image
processing makes to simplify the image representation in order to analyze the
images. So many algorithms are developed for segmenting images, based on the
certain feature of the pixel. In this paper different algorithms of segmentation can
be reviewed, analyzed and finally list out the comparison for all the algorithms.
This comparison study is useful for increasing accuracy and performance of
segmentation methods in various image processing domains.
Keywords. Image Segmentation, Digital Image Processing, K-Means Clustering,
Edge detection, histogram.
1. Introduction
Image is binary representation of visual information by means of pixels. It
contains lot of information to perform some useful operations on different area of
applications like medical field, object detection, video surveillance system, computer
vision, pattern recognition, remote sensing etc. With the help of computer algorithms,
we can manipulate the pixels either enhance the quality of image or extract the useful
information from it. Digital image processing has different stages, in which
segmentation is the essential and challenging part in the operation of image processing.
Image segmentation should be segregating the images into meaningful parts that are
having similar features and behaviors. The purpose of segmentation is, to make the
image representation easily, classified meaningfully, and analyzed properly. Using the
image segmentation, we can localize the objects and identifying its boundaries in an
image. Then assign the labels to each pixel in the entire image. Some pixels having the
same labels means they have common characteristics. This paper analyzes the different
methods of segmentation and their algorithms. First, we analyze the segmentation
carried out by means of pixel properties [1] to analyze the local and global properties,
and to tell how speed up the segmentation process by modifying the existing ACM
model along with local and global properties. Expectation of maximization algorithm
[2] needs more iteration to segment the image. Modification can be based on the
1
S. PRABU, Associate Professor Department of Computer Science & Engineering, VI Institute of
Technology Sirunkundram, India
Email: sprabumkm@gmail.com
Recent Trends in Intensive Computing
M. Rajesh et al. (Eds.)
© 2021 The authors and IOS Press.
This article is published online with Open Access by IOS Press and distributed under the terms
of the Creative Commons Attribution Non-Commercial License 4.0 (CC BY-NC 4.0).
doi:10.3233/APC210223
419

likelihood properties we can achieve the minimum number of calculations. EM based
localization and 3D U-Net and EM based attention mechanism will provide the
optimized segmentation results. Color images can be segmented by RGB Histogram [3]
along with firefly algorithm. Combination of SCT-I and SCT-V algorithm and
SAMFO-TH [17] algorithm solves the multilevel threshold problem of RGB color
image. Color hymnographies [6] can be used to identify the color object detection.
Image restoration algorithm [4] can segment the image based on the sample blocks.
This method utilizes the advantages of both Criminisi algorithm & Watershed image
segmentation algorithm. So many factors are affecting while extracting text
information from the image [5]. Noise is more, when the image can be captured from
live video camera. The Encoder – Decoder framework resolves the issues in
segmentation for retrieving text information. To speed up the text retrieval process,
Instance Segmentation Network technique can be used. This paper analyze the different
methods and compare it to give the clear idea about image segmentation process.
2.Related Work
2.1 Segmentation by means of properties of pixel
Segmentation operation can be also performed by means of global and local
properties of pixels [1]. It states that, the global properties are calculated by mean
values of various pixels and successive edges of objects, and the local properties are
characterized by interaction of successive pixels and the boundary of the image.
Image entropy can be done by improved weighting function that can be adapt the
weight between local term and global term, so the segmentation speed can be improved
significantly.
2.1.1 Thresholding Method
Based on the pixel’s intensity, this method partitions the image, into two
parts, they are intensity of pixel value lower than threshold and intensity of pixel value
greater than threshold. Multi stage thresholding approach is necessary for color image
segmentation in many applications [10]. To segment the color components Red, Green
and Blue, more than two optimal thresholds can be needed. In future a greater number
of thresholds is necessary. But a greater number of thresholds can decrease the
performance of segmentation process. The main drawback of multi-level thresholding
leads long execution time [11]. During the processing of low-quality images, cause the
wrong prediction of threshold values. The improved salp swarm algorithm can
minimize the number of thresholds to ensure effective segmentation process. Modified
grasshopper optimization algorithm [16], metaheuristic algorithm, firefly algorithm,
novel population-based bee foraging algorithm [18] are also used for optimized
threshold values.
2.1.2 Edge detection
An edge is a set of connected pixels that creates the boundary between the
two successive regions. The edge detection is basically a process of segmenting the
image into discontinuity of regions. The perfectness of the image-by-image processing
and computer vision can be depending on identifying meaningful edges. Out of all the
edge detection techniques, ‘Canny Edge Detector’ produces good results than other
edge detection methods. The basic limitations of edge detection are edge connectivity
and edge thickness [7]. Multiple values of Threshold approaches need to solve the
above two issues. To find the optimized thresholding values Genetic algorithm can be
S. Prabu and J.M. Gnanasekar / A Study on Image Segmentation Method for Image Processing420

applied which optimize the coefficient of filter [8]. This kind of optimization improves
the quality of segmenting edges in MR scan images.
2.1.3 Clustering methods
Clustering is the process of image information can be replaced by clusters.
Cluster is the collection of similar attribute data points. Pixels, which have common
attributes like same color, same texture or any related attribute. Main issues in
Clustering are identify the correct inter cluster distance and identify the total number of
Clusters. Robust self-sparse fuzzy clustering algorithm improves the clustering result
by reducing the noisy features. The main drawback of K means clustering method is
image segments are disconnected and disseminated with wide distance. To find the
optimal number of clusters is very hard without knowing the initial parameters. The
unsupervised K-means clustering technique is needed to achieve this [9]. To solve the
initialization problem, UK-means algorithm uses number of points to determine the
initial number of clusters. To avoid overlapping, kernel K-means clustering can be
proposed [13]. In which kernel functions are transformed into feature space, to estimate
the correct number of clusters. The combination of depth and semantic information of
images [20] improve the accurate identification of initial center value.
2.2 Optimized Segmentation
Segmentation operation can be improved effectively by applying small
changes in the existing segmentation algorithm called Expectation of maximization; it
has a greater number of iterations and takes more computing power [2]. Compression
based segmentation methods express the approximation of actual pixel values with
some of the sample pixels, so that the size of the image can be tremendously reduced.
To improve the efficiency of compression, the original image can be pre quantized with
higher bits [15]. But sometimes the predictions may not work properly when using high
bit rates.
2.3 Color Image Segmentation
Segmentation operation can be achieved in a color images by RGB
Histogram [3]. By applying firefly algorithm, the optimal multi-level image
segmentation can be achieved. The two new algorithms called SCT-I and SCT-V for
image and video input data. For each frame SCT-V algorithm locate the target of
interest (TOI) for object tracking and SCT-I algorithm maintain the original color in the
target of interest. SAMFO-TH algorithm [17] based on moth flame optimization.
Histogram is developed by dividing the range of the data into same sized classes. Then
for each class, those data set points which present into the class are calculated. The key
parameter of this method is the selection of the threshold value, which can be computed
manually or automatically by some algorithms. The basic concept of this method is, to
identify the mean or median value, so that pixels of the object are brighter than the
background. By applying color hymnographies to increase the color fidelity and
improve the color object detection [6].
2.4 Segmentation based on the sample blocks
Segmentation operation can be achieved by improved image restoration
algorithm based on the sample blocks [4]. It resolves the image repairing problems
occurred when the image restoration process. Criminisi algorithm and watershed image
algorithm is applied to the large amount of image set. Then identify the matching pixel
S. Prabu and J.M. Gnanasekar / A Study on Image Segmentation Method for Image Processing 421

blocks in image segmentation. So that excessive extension of texture blocks for the
process of restoration of images can be avoided.
2.4.1 Region-based methods
This kind of segmentation groups the pixels that have similar properties. It
segregates the pixel, those having the similar characteristics and dissimilar
characteristics. It compares the properties of neighboring pixels and produces the result.
The main objective is to differentiate the homogeneity of the image. That can be
achieved by improved image restoration algorithm based on the sample blocks [4]. A
new method using region of interest for segmenting images, having less computation
complexity, to preprocess the training sets so that to reduce the redundant information.
For color region segmentation, hue division based selective color transfer algorithm
can be used. In which HSV color model transferred into luminance and saturation
integrity.
2.5 Text-Based Segmentation
Segmentation operation can be achieved only in a text-based images, to
retrieve the necessary text information from the whole image [5]. The information may
be single or multiple line, words, or even one or more characters. This paper proposes
different methodologies at the various segmentation stages. It first justifies the
segmentation process in the text context based on the information retrieval. This paper
also discusses different factors affecting the process of segmentation. The Captured
images from camera have more background noise [12]. The main issues are symbols
having several separate primitives with complex background and distortions from
camera. To extract the text information from natural scenes are also complex process.
The encoder – decoder framework [17] is proposed. It is used by the combination of
attention mechanism and connection time classification. To reduce the more processing
time the new technique called Instance Segmentation Network (ISNet) to detect the text
content by generating prototype masks simultaneously.
2.6 Motion & Interactive Segmentation
Motion Segmentation means pixels are grouping together in a particular
movement of the object. The main objective of this technique is to segment the images
for object that are in moving condition. To make the image analysis, and understanding
the image properties, to analyze the motion sequence is important. First Identify the
background and foreground objects, and then make analysis these objects
independently, this model is commonly called ‘variation Model’ which takes two
successive video frames, evaluate the movement ranges between each frame. The main
aim of this method is to identify the object boundaries with minimal user interventions.
3.Survey Discussion
Various segmentation algorithms are discussed and each method has its own
advantages and disadvantages within a particular context. Some methods need the
modification and improvement. The following table 1 compares the various factors of
different segmentation methods.
S. Prabu and J.M. Gnanasekar / A Study on Image Segmentation Method for Image Processing422

Table 1: Comparison of different segmentation techniques
Method Descripti
on Characteristics Advantages Issues
Threshold
ing
Method
Segmentation
done by value
of pixel
Intensity
Types:
Local & Global
Thresholding
Simple & Effective
Processing.
Single Threshold value
not give accurate
result. More Threshold
values leads more
processing time
Edge
detection
Segmentation
based on
discontinuity of
regions
Steps
Filtering
Enhancement
Detection
Using probability for
finding error rate,
Improving signal to
noise ratio.
Highly sensitive to
noise.
More possible for
inaccuracy.
Clustering
methods
Segmentation
by means of
collection of
similar attribute
data points
They are
unsupervised
algorithms
Applicable for real
time problems
1.Identify the correct
inter cluster distance
2. Identify the total
number of Clusters.
Not suitable for wide
distances.
K-means
Clustering
This method is
used when you
have unlabeled
data (i.e. data
without defined
any category)
It clusters, he
given data into
K-clusters or
parts based on
the K-centroids.
Simple to implement.
scalable for large data
sets
image segments are
disconnected and
disseminated with
wide distance
dependant on initial
values.
Region-
based
methods
Segmentation
based on
similar
properties of
pixels
Partitioning an
image into
homogeneous
regions.
Accurately segment
the regions that have
the same properties we
define. It works well
with respect to noise
Expensive
Computation.
Variation of intensity
will affect the result.
Motion &
Interactive
Segmentat
ion
pixels are
grouping
together in a
particular
movement of
t
he objec
t
Criminisi
algorithm &
Watershed
algorithm
The pixels that have
high intensity variation
can be clearly
classified as moving
objects.
If registration of
background pixels are
not perfect, leads false
prediction of moving
objects. More
computation cost.
Compress
ion-based
methods
approximation
of actual pixel
values with
some of the
sample pixels
Types:
Lossless,
Compression
Lossy
Compression
Best suited in storage
and transmission of
images
More computation
complexity.
Histogram
-based
methods
each region’s
pixels are
having similar
properties, like
intensity, color
values, texture.
Best suited for
image
enhancement
Directly process the
color images
Requires high number
of operations on each
pixel. Computationally
slow. Over
segmentation occurred
since no spatial
information.
4. Conclusion
This paper discusses about various image segmentation techniques and
different image segmentation approaches are illustrated and correlated. All these
methodologies are suitable for many applications in medical field, object detection,
computer vision, surveillance system, computer vision etc. From this study, each
segmentation methods are desirable for specific image types. So, the combination of
multiple segmentation methods is needed to get good result. To improve the accuracy
of the segmentation result, need to apply machine learning techniques. The important
parameters of image segmentation are accuracy, complexity, efficiency and
S. Prabu and J.M. Gnanasekar / A Study on Image Segmentation Method for Image Processing 423

interactivity. No particular methods are comfortable for all the types of images. For this
reason, segmentation of images in various applications has high demand in future.
References
[1] ZhenZhou Wang, “Image segmentation by combining the global and local properties”, Elsevier, Expert
Systems With Applications (2017), Vol-87, PP- 30-40.
[2] Lahouaoui Lalaoui, Tayeb. Mohamadi and Abdelhak Djaalab, “New Method for Image Segmentation”,
Elsevier, Procedia - Social and Behavioral Sciences (2015), Vol-195, PP- 1971–1980.
[3] V. Rajinikanth, and M. S. Couceiro, “RGB Histogram based Color Image Segmentation Using Firefly
Algorithm”, Elsevier, Procedia Computer Science(2015), Vol-46, PP- 1449–1457.
[4] Huang Ying, Li Kai, and Yang Ming, “An Improved Image Inpainting Algorithm based on Image
Segmentation”, Elsevier, Procedia Computer Science(2017), Vol-107, PP- 796–801.
[5] Gupta Mehul, Patel Ankita, Dave Namrata, Goradia Rahul and Saurin Sheth, ”Text-Based Image
Segmentation Methodology”, Elsevier, Procedia Computer Science (2014), Vol-14, PP- 465–472.
[6] Gupta Mehul, Patel Ankita, Dave Namrata, Goradia Rahul and Saurin Sheth, “Text-Based Image
Segmentation Methodology”, Elsevier, Procedia Computer Science (2014), Vol-14, PP- 465–472.
[7] Mamta Mittal, Amit Verma, Iqbaldeep Kaur, Bhavneet Kaur, Meenakshi Sharma, Lalit Mohan Goyal,
Sudipta Roy and Tai-Hoon Kim, “An Efficient Edge Detection Approach to Provide Better Edge
Connectivity for Image Analysis”, IEEE Access (2019), Vol-7, PP- 33240–33255.
[8] Ahmed H. Abdel-Gawad, Lobna A. Said, Dave Namrata, and Ahmed G. Radwan, “Optimized Edge
Detection Technique for Brain Tumor Detection in MR Images”, IEEE Access (2020), Vol-8, PP-
136243–136259.
[9] Kristina P. Sinaga and Miin-Shen Yang, “Unsupervised K-Means Clustering Algorithm”, IEEE Access
(2020), Vol-8, PP- 80716–80727.
[10] Heming Jia Jun Ma and Wenlong Song, “Multilevel Thresholding Segmentation for Color Image Using
Modified Moth-Flame Optimization”, IEEE Access (2019), Vol-7, PP- 44097–44134.
[11] Heming Jia Jun Ma and Wenlong Song, “Multilevel Thresholding Segmentation for Color Image Using
Modified Moth-Flame Optimization”, IEEE Access (2019), Vol-7, PP- 44097–44134.
[12] Ahmed A. Ewees, Mohamed Abd Elaziz, Mohammed A. A. Al-Qaness, Hassan A. Khalil and
Sunghwan Kim,“ Two-Step CNN Framework for Text Line Recognition in Camera-Captured Images”,
IEEE Access (2020), Vol-8, PP- 26304–26315.
[13] Soosan Beheshti, Edward Nidoy, and Faizan Rahman, “K-MACE and Kernel K-MACE
Clustering”,IEEE Access (2020), Vol-8, PP- 17390–17403.
[14] Dan Wang, Guoqing Hu, Qianbo Liu, Chengzhi Lyu, and Md Mojahidul Islam, “Region-Based
Nonparametric Model for Interactive Image Segmentation”, IEEE Access (2019), Vol-7, PP- 111124–
111134.
[15] Jianyu Lin,“A New Perspective on Improving the Lossless Compression Efficiency for Initially
Acquired Images”, IEEE Access (2019), Vol-7, PP- 144895–144906.
[16] Hongnan Liang, Heming Jia, Zhikai Xing, Jun Ma, And Xiaoxu Peng, “Modified Grasshopper
Algorithm-Based Multilevel Thresholding for Color Image Segmentation”, IEEE Access (2019), Vol-7,
PP- 11258–11295.
[17] Heming Jia, Jun Ma, And Wenlong Song, “Multilevel Thresholding Segmentation for Color Image
Using Modified Moth-Flame Optimization”, IEEE Access (2019), Vol-7, PP- 44097–44134.
[18] Zhicheng Zhang And Jianqin Yin, “Bee Foraging Algorithm Based Multi-Level Thresholding For
Image Segmentation”, IEEE Access (2020), Vol-8, PP- 16269–16280.
[19] Zhen Zheng , Bingting Zha, Hailu Yuan, Youshi Xuchen, Yanliang Gao, And He Zhang, “Adaptive
Edge Detection Algorithm Based on Improved Grey Prediction Model”, IEEE Access (2020), Vol-8,
PP- 102165–102176.
[20] Hanxiao Rong, Alex Ramirez-Serrano, Lianwu Guan, And Yanbin Gao, “Image Object Extraction
Based on Semantic Detection and Improved K-Means Algorithm”, IEEE Access (2020), Vol-8, PP-
171129–171139.
S. Prabu and J.M. Gnanasekar / A Study on Image Segmentation Method for Image Processing424