Comparison of Traditional Method with

Watershed Threshold Segmentation Technique

R. VASIM AKRAM, ASSISTANT PROFESSOR, vasim487@gmail.com
K.P. KRISHNA SAGAR, ASSISTANT PROFESSOR, kpksagar7@gmail.com
N.A.V. PRASAD, ASSISTANT PROFESSOR, navpvlsi@gmail.com
Department of ECE, Sri Venkateswara Institute of Technology,
N.H 44, Hampapuram, Rapthadu, Anantapuramu, Andhra Pradesh 515722

Abstract— For many image-gathering algorithms and acceptable vision frameworks, image segmentation is
an important first stage in the inquiry process. There isn't currently a one-size-fits-all method since, as several
authors have pointed out, division ceases when the spectator's aim is satisfied. In this study, we covered the
methods utilised for digital image processing and how photos were categorised. In this study, we analysed the
CT images for noise using K-MEANS CLUSTERING (PIXEL BASED APPROACH) and an Improved
Watershed Segmentation with Various Colour Coding methods. The pictures were then analysed using PSNR
MSA RMSE and MSE. Precision, PSNR, MAE, and RMSE are some of the exhibition limits that reveal
standard model expected error in premium units. Its reach extended from infinity to the horizon.

Keywords— Image Segmentation, Digital Images, K-Means, Watershed Segmentation, CT images.

I. INTRODUCTION

Medical imaging refers to the well-known method of creating highly visible pictures that reveal the interior structures of
the body for scientific and medical research purposes, as well as the function of the tissues themselves. The board and
recognisable signs of turbulence are what this cycle is chasing. This loop compiles data on the normal anatomy and
function of the organs, making it easy to spot deviations. This cycle combines natural and radiological imaging
techniques, including sonography, magnetic resonance imaging (MRI), X-ray and gamma imaging, thermal imaging, and
isotope imaging. Recording information regarding the body's area and capacity is made possible by a wide variety of
technological advancements. When compared to those balances that generate images, those solutions have a lot of
obstacles. For a variety of suggestive objectives, billions of images are taken annually all around the globe. Some of them
make advantage of radiation modifications that are both ionising and nonionizing [1]. Without invasive procedures,
clinical imaging captures images of the inside anatomy of the body. These images were sent by means of fast processors
and as a result of numerical and intelligent conversion of energy to signals [2]. Eventually, such signage will be replaced
with more modern graphics. The many types of tissues inside the body are communicated by those indications [9]. In
every case, the computer graphics play a crucial role. Instruction in clinical imaging makes passing reference to
computer-assisted image processing. Picture taking, capacity, introduction, and communication are just a few of the many
ways and activities that make up this training. An image's potential to suggest a percentage of

attributes, such as the ability to lighten or darken a saw sight. Among the many benefits of digital images are their ease of
storage and communication, rapid quality evaluation, many duplication options that preserve quality, rapid and modest
multiplication, and adaptable control. Copyright infringement, inability to enlarge without sacrificing quality, enormous
memory requirements, and the necessity for a faster CPU for control are some of the problems with digital images [3].

The use of a personal computer to manipulate the high-tech image is known as an image processing operation. Flexibility,
adaptability, data storage, and communication are only a few of the many benefits of this technique. The development of
several photo resizing algorithms has allowed for the efficient preservation of the pictures. This process incorporates
many sets of rules to execute into the images concurrently. There is a wide range of possible dimensions for working with
2D and 3D images. In the 1960s, the procedures for managing images were laid down. Space, medicine, language, and
TV image enhancement were just a few of the many areas that made use of these techniques. The advent of personal
computer systems in the 1970s lowered the price and increased the speed of image processing. Image processing became

faster, more rational, and simpler in the 2000s [4].

II. CLASSIFICATION OF DIGITAL IMAGES

Two main types of photos are used in advanced pictures. The four-dimensional array of frequently inspected
values, or pixels, constitutes a raster image. In most cases, the computer-generated images are faraway shots
with complex colour contrasts. The advanced images have a predetermined objective due to the pixel size.
Because of certain missing data, the resizing cycle degrades the quality of the advanced photos. The excellent
shade concealment of computer photos makes them ideal for use in photography. The picture-taking device
manages the objective. The digital images come in a variety of formats, such as Windows bitmap (BMP), Tag
Interleave Format (TIFF), Paintbrush (PCX), Portable Network Graphics (PNG), and many more [5, 6].

In its accurate representation, a vector is shown by the PC as a coiled and twisted object. Line width,
measurement, and hue are only a few of the many properties of the vector. Since vectors are essentially
adaptable images, their quality remains constant regardless of how many times they are copied. Graphs, line
paintings, and configuration all make appropriate use of the vectors.

Applications of digital image processing

The digital image processing has many applications in the medical field such as:

1. Medicine

In medication, numerous procedures are utilized, for example, segmentation and surface investigation, which are
utilized for disease and other issue recognizable pieces of proof. Picture enrolling and combination techniques are
broadly utilized these days particularly in new modalities, for example, PET-CT and PET-MRI. In the field of
bioinformatics, telemedicine, and the configuration less pressure methods are utilized to convey the picture distantly
[1–4][7].
2. Forensics

The basic strategies utilized in this field are edge location, design coordinating, denoising, security, and biometric
purposes, for example, personality, face, and unique mark documentation. A legal science depends on the data set
data about the people. A legal science coordinates the information (unique mark, eye, photograph, and so on) with
the data set to characterize the individual's personality [2].

3. Medical imaging systems

Clinical imaging frameworks utilize the signs got from the patient to deliver pictures. Clinical imaging frameworks
utilize both ionizing and nonionizing sources.

III. K-MEANS CLUSTERING (PIXEL BASED APPROACH)

The k-implies approach is a simple system in solo characterization. The grouping calculations needn't bother with
preparing information. K-implies bunching is a common system. The k-implies bunching calculation group’s
information by iteratively figuring a mean strength for each class and sectioning the picture by ordering every pixel
in the class with the nearest mean [8].

The calculation includes an extreme partner work as its bunching idea; where the information is grouped into the
extended region allow us to take a picture, Y with M information to be grouped into N territory. From the start, all
center qualities are discretionarily assigned. The jth data, wj is allocated to the closest data cluster, Dkbased on the
least Euclidean distance [8], where j = 1,2,3,4, … , … M. and K = 1,2,3,4, … , … , N. Subsequent to completing
the allocating cycle for all information, the new area of focuses is assessed by:

1
D =
k pk ∑j∈D
k
Wj

Where, pk is the number of partners in the kth group. The strength of each cluster is then determined in the
calculation utilizing:

g(D ) = ∑ (‖W − D ‖)
2

k j k
j∈Dk

After arranging the mid values in a sorted order there are two values are recorded to retrieve the clustering process,
those are the mid least strong point Dq and mid high strong point Dt . In order to obtain the best clustering process
the following condition must be satisfied by the link between Dq andDt.

g(Dq ) ≥𝖺b g(Dt)

In the above condition the constant value 𝖺b = 𝖺0 where 𝖺0 is with the distinctive value 0 < 𝖺0 < 1
3
. The mid least

strong point Dq withdrew all its associates when the above condition is not satisfied. When the above condition

is
accepted then it obtains the Dt associates which contains least strong point i.e. Wj < Dt. For the associates of Dt
with more value than Dt they will remain as Dt associates. Both Dq and Dt will be updated using above
conditions.

Here, least strong point and high strong point Dq and Dt have new associates like pq andpt. The cluster that holds
information with esteem more closely resembling the center worth slopes to get the least strength esteem, as a layout
of the Euclidean distance between the center worth and its partners is little (for example the present circumstance
might be right for Dq ). Consequently, making the center Dq eliminate its associates will prompt dismissing a
significant exhibition of the gathering of information. This could likewise develop group change. Thus, a helpless
segmentation could be framed. Subsequent to completing the moving technique, all middles are rearranged utilizing
and the new area, everything being equal, the estimation of 𝖺b is then redesigned by:

All above-expressed strategies are rehashed until it is fulfilled. To guarantee an improved bunching measure,
another state as characterized by estimation is locked in. The total methodology will be rehashed if estimation isn't
fulfilled.

For every repeat, the 𝖺b and 𝖺c are independently rebuilt as indicated by,

Figure 1: Objects in Cluster 1

 IV. IMPROVED WATERSHED SEGMENTATION WITH VARIOUS COLOUR CODING

In this study, the plan strategy's measures were the Accuracy (Ac), which is a measure of actual
results, to take the arrangement into account as one of the markers anticipated from the suggested
calculation. To measure how many positive outcomes are actually certain, one uses the sensitivity
(Se), also known as the true positive (TP) rate or review, and the precesion (PR), often called positive
predictive value. In light of the chaos framework's foundational principles, the aforementioned
execution limits were established as

Take the picture for watershed picture segmentation.

1. Execute an Active form edge detection segmentation algorithm to discover the edges of the picture.

2. Actualize the grey watershed segmentation algorithm to locate the resultant picture as dim watershed
segmentation

3. Presently with the assistance of markers mark each grey – pixel to a particular tone (say red, green, blue) as
appeared in fig. (5). RGB to YIQ is given by:

4. Y 0.299 0.587 0.114 R

Q 0.596 -0.275 -0.321 G

I 0.212 -0.523 0.311 B

Colours yellow (red + green), cyan (blue + green), fuchsia (red + blue), and white (red + green +
blue) are formed by combining the primary colours of red, green, and blue, as seen in the image on
the left. The three primary colours used in RGB shading—red, green, and blue—are shown in the
image on the right, along with their corresponding pair-wise blends. Finally, the shaded area that is
dark is created by subtracting white from each of the three primary colours. Blending colours I and J
yields all X-axis tones and all Y-axis tones as well. An image with shaded and watershed segments
will be the result of this computation.

V. RESULTS
The ground truth, or actual RBCs, are TPs after they have undergone appropriate quality control.
These denote the agreement between the experts and our technique; true negatives (TNs) are the non-
RBCs (often relics or other blood components) that have been correctly non-market exactly; false
positives (FPs) are the number of non-RBCs that were defectively

verified organically, whereas false negatives (FNs) reduced the number of intact RBCs. Pixels that
went undiscovered throughout the inspection process, indicating areas where the suggested method
failed to detect a red blood cell (RBC) that was recently identified by experts. In reality, experts did
not distinguish the TN; instead, we regarded as TN all the blatant structures in the images that were
not differentiated by expert consensus during ground truth structure.

Calculated using a predetermined procedure to determine the true positive (GN), false negative (BN),
false positive (BP), and true negative (GN). The results were based on 500 images, and the suggested
method was determined to be more accurate. Reliability and Accuracy. The watershed threshold
segmentation approach, which makes use of shading transformation coding, successfully refines
Watershed and Kmean in CT head and neck imaging. In comparison to RMSE and MSA, PSNR was
superior.

TABLE 1: TECHNIQUES WITH PARAMETERS

Technique Accuracy Sensitivity Precision PSNR RMSE MAE
K-mean 87.5 89 97.5 78.4567 0.0413 0.0018
Clustering
Watershed 87.9 76.67 91.9 80.5233 0.0211 0.0021
Segmentation

PSNR computes the peek signal-to-noise ratio in (db) limited by two pictures. The PSNR is utilized for standard
evaluation limited by the genuine picture and remade picture. PSNR is straightforwardly corresponding to the norm
of the recreated picture MAE (Mean absolute error) figures the mean mass of the wrong qualities in the arrangement
of expectations regardless of the bearing. RMSE (Root mean squared error) is a four-square accomplishing rule that
computes the mean mass of the defective qualities.

100

80

60 Accuracy
40 Sensitivity

20 Precision

Figure 1 Segmentation Techniques Comparison with Accuracy, Sensitivity, Presicion.

 100

80

60 PSNR
RMSE
40
MAE
20

Figure 2 Methods Comparison with PSNR, RMSE, and MAE

VI. CONCLUSION

In order to determine image metrics and examine them, CT images were tested with noise and analysed using
PSNR, MSA, RMSE, and MSE. Evidence of normal model expectation error in premium units is shown by
examination of the exhibition borders, which includes measures like exactness, PSNR, MAE, and RMSE. The
potential range for it was infinite to zero. These ratings are in an adverse order, meaning that lower values are
better on traditional methods, which proved that the right approaches were required for progress. Using
shading transformation coding, the watershed threshold division approach successfully refined K-implies in
CT head and neck images. In comparison to RMSE and MSA, PSNR was superior.

REFERENCES
[1]. Abdallah Y. Improvement of sonographic appearance using HATTOP methods. International Journal of Science and Research (IJSR).
2015;4(2): 2425-2430. DOI: http://dx.doi.org/ 10.14738/jbemi.55.5283.
[2]. Abdallah Y. Increasing of edges recognition in cardiac scintigraphy for ischemic patients. Journal of Biomedical Engineering and
Medical Imaging. 2016; 2(6):40-48. DOI: http://dx.doi.org/10.14738/jbemi.26.1697.
[3]. Abdallah Y. Application of Analysis Approach in Noise Estimation, Using Image Processing Program. Germany: Lambert Publishing
Press GmbH & Co. KG; 2011. pp. 123-125.
[4]. Abdallah Y, Yousef R. Augmentation of X-rays images using pixel intensity values adjustments. International Journal of Science and
Research (IJSR). 2015;4(2):2425-2430.
[5]. Abdallah YM. History of medical imaging. Archives of Medicine and Health Sciences. 2017;5:275-278. DOI:
10.4103/amhs.amhs_97_17.
[6]. Abdallah Y. An Introduction to PACS in Radiology Service: Theory and Practice. Germany: LAP LAMBERT Academic Publishing;
2012. ISBN: 978-3846588987.
[7]. Abdallah Y. Increasing of Edges Recognition in Cardiac Scintography for Ischemic Patients. Germany: Lambert Publishing Press
GmbH& Co. KG; 2011. pp. 123-125.
[8]. Prachee Tyagi, Tripty Singh, Ravi Nayar, Shiv Kumar, "Performance comparison and analysis of medical image segmentation
techniques", 2018 IEEE International Conference on Current Trends in Advanced Computing (ICCTAC), 2018.