Journal of Theoretical and Applied Information Technology
31
st
October 2016. Vol.92. No.2
© 2005 - 2016 JATIT & LLS. All rights reserved
.
ISSN: 1992-8645 www.jatit.org E-ISSN: 1817-3195
395
K-NEAREST NEIGHBOR BASED DBSCAN CLUSTERING
ALGORITHM FOR IMAGE SEGMENTATION
SURESH KURUMALLA
1
, P SRINIVASA RAO
2
1
Research Scholar in CSE Department, JNTUK Kakinada
2
Professor, CSE Department, Andhra University, Visakhapatnam, AP, India
E-mail id: kurumallasuresh@gmail.com
ABSTRACT
Clustering is a primary and vital part in data mining. Density based clustering approach is one of the
important technique in data mining. The groups that are designed depending on the density are flexible to
understand and do not restrict itself to the outlines of clusters. DBSCAN Algorithm is one of the density
grounded clustering approach which is employed in this paper. The author addressed two drawbacks of
DBSCAN algorithm i.e. determination of Epsilon value and Minimum number of points and further
proposed a novel efficient DBSCAN algorithm as to overcome this drawback. This proposed approach is
introduced mainly for the applications on images as to segment the images very efficiently depending on
the clustering algorithm. The experimental results of the suggested approached showed that the noise is
highly reduced from the image and segmentations of the images are also improved better compared to the
existing image segmentation approaches.
Keywords: Data Mining, Clustering, Density Based Clustering, DBSCAN, K-Nearest Neighbor, Image
Segmentation.
1. INTRODUCTION
Emerging of current methods for logical
information gathering had ensued in huge scale
accretion of information relating to dissimilar
arenas. Data mining is the encouraging
methodology that arrives at the conclusion in the
domain of computer science where it mine the
vital or beneficial knowledge from enormous
data samples or huge amount of data. It is a stage
in the Knowledge Discovery in Databases
(KDD) procedure comprising of the application
for data investigation and detection of
procedures under adequate computational
efficacy restrictions, generates a specific
enumeration of trends above the data [8]. It
employs sophisticated arithmetical study and
modeling methods to reveal patterns and
interactions concealed in administrative data
samples. Clustering plays a significant part in the
data mining. The approach of identifying
similarities amongst information rendering to the
features obtained in the data and merging
analogous data entities into groups is known as
clustering. It is an unsupervised categorization of
distinguishing set of identical entities in outsized
data samples without having definite clusters
through unambiguous features. Clustering
Methods are advantageous in numerous domains
like excavating skin lesion images, pattern
analysis, machine learning circumstances and
numerous other domains.
Clustering is categorized into five
kinds: Partitioned dependent, Hierarchical
dependent, Density dependent, Model dependent
and Grid dependent. Density dependent
approaches are grounded on an approximation of
the density of information. The universal
perception of these approaches is that the clusters
to construct are designed of a group of points of
greatly density surrounded by a lower density
values. In this Method, most segregating
procedures group entities depending on the
distance amongst entities. Such procedures could
discover random designed groups. The universal
perception is to endure developing the specified
cluster as elongated as the density or number of
entities or data values in the neighborhood beats
certain threshold. These procedures could be
employed to filter the noise or outliers. Density
dependent clustering approach is one of the
crucial procedures for grouping in data mining.
The clusters that are designed depending on the
density are flexible to understand and do not
restrict itself to the outline of clusters.
Almost all of the famous clustering
procedures necessitate input parameters that are
rigid to define however have a substantial effect