International Journal of Engineering Trends and Technology (IJETT) – Volume 36 Number 2- June 2016

Video Segmentation using FCM Algorithm

Priyanka Dhiman#1, Mamta Dhanda*2
1
M.Tech.,CSE Department, JMIT, Radaur, Kurukshetra University, India
2
Assistant Professor, CSE Department, JMIT, Radaur, Kurukshetra University, India

Abstract – Now days, video information plays a

crucial role. Storage of video data has become
significant for wide variety of applications such as in Motion Detection:
Forensic, Medical, Navy and Military, Multimedia
Motion detection is essential in many fields, such as
applications. The miracle happens when there is a
pattern recognition, traffic surveillance, object
appropriate indexing method for the stored video
tracking. At present, the existed approaches of
content and proper retrieval method for accessing
moving target detection are optical flow, background
the stored information from the databases. Video
difference, and time difference (frame difference).
Segmentation is the major issue involved in
Motion detection algorithm operates in a specific
retrieving and storing the video data. In this paper
area to observe the change of image for detecting the
we use SLIC and FCM algorithm for video
dynamic object. However, the change in the
segmentation. First of all SLIC algorithm is used for
environment disturbs the motion detection seriously:
tracking of an object from the video. Tracking is
illumination, noise, shadow and so on. [5]
usually performed in the context of higher-level
applications that need the location and/or shape of
the object in each frame. Then FCM algorithm is
used for the segmentation. The advantage of FCM
technique is that it yield regions which are more
homogeneous also it reduces spurious blobs.
Further it was less sensitive to noise. In this paper
both algorithms are used. The result shows that the
proposed method used for segmentation enhances
the image quality.

Keywords — Clustering, Motion Detection, Object

tracking, SLIC, Video Segmentation.
Fig. 1 Block diagram of the proposed video object
segmentation approach [8]
I. INTRODUCTION
OBJECT TRACKING:
Segmentation can be specified that it is a process of
partitioning data into groups of potential subsets that In order to extract useful information of an object of
share homogeneous characteristics. It has become a interest from a video scene and keeping track of its
technique for semantic content extraction and plays orientation, motion, occlusion etc. is defined as
an important part in pattern recognition, digital object tracking. The goal is to review the state-of-
multimedia processing and computer vision. the-art tracking methods, identify new trends and
classify them into different categories. Object
tracking, in general, is a challenging problem. Due
IMAGE SEGMENTATION: to abrupt motion of object, changing appearance
Image segmentation is a necessary but challenging patterns of the object and the scene, object-to-scene
problem. It is a necessary first step in image and object-to-object occlusions, camera motion and
analysis. Because it is used in high-level image non rigid object structures, there are many
interpretation and understanding such as object difficulties arise in tracking of objects.
recognition, robot vision, geographical imaging and The object tracking can be complex, due to camera
medical imaging. In general, image segmentation is motion, loss of information caused by the projection
a task of partitioning an image into non-overlapped, of the 3D world on a 2D image, variations of target
consistent regions that are uniform with respect to scale, partial occlusions, real-time processing
some characteristics like intensity, color, tone or requirements, clutter, and so on.
texture, and more. There are different techniques for
CLUSTERING:
image segmentation e.g. thresholding, clustering,
classifications, artificial neural networks (ANNs), An unsupervised learning task is clustering, where
region growing, edge detection etc. [3] the pixels are classified in to a finite set of categories

ISSN: 2231-5381 http://www.ijettjournal.org Page 106

 International Journal of Engineering Trends and Technology (IJETT) – Volume 36 Number 2- June 2016

known as clusters or groups. It is an unsupervised

classification because there is no available data to
train the pixels. A similarity criteria is defined
between pixels such as distance, intensity and
colour, and then similar pixels are grouped together
to make clusters. The grouping of pixels is done
based on intra class similarity and enhancing the
inter class similarity. Where intra cluster similarities
is measured between the pixels in a cluster and inter
cluster similarity is measured between the clusters.
Clustering is useful in several exploratory decision
making, pattern analysis, and machine learning
situations, document retrieval, including data
mining, image segmentation, and pattern
classification.[6]
Many schemes of clustering are categorized based
on their special characteristic, such as the hard Fig. 2 Block diagram of Video Segmentation.[12]
clustering scheme and the soft (fuzzy) clustering
scheme. The conventional hard clustering scheme
restricts each point of the data set to entirely just one
cluster. As a consequence, with this approach the II. RELATED REVIEW WORK
result of segmentation is often very crisp, i.e., every
pixel of the image belongs to just one class exactly. At present we found that many researchers had done
In fuzzy (soft) clustering, data elements can belong work towards the video segmentation, object tracking
to more than one cluster. The fuzzy set theory and motion detections and in the meantime, few research
described by a membership function. [3] reports are published recently & tried to solve the
problem associated with them. Many of the journals and
VIDEO SEGMENTATION
research papers published during the above span 2013-
The term video segmentation refers to decomposing 2016 have been studied.
a video data into meaningful elementary parts that
have strong correlation with the actual world Shaoping Xu et. al (2013) In the paper author
contained in the video data. The result of video propose an algorithm i.e novel cluster number
segmentation is a set of segments that collectively adaptive fuzzy c-means image segmentation
cover the real entire video data. The major algorithm (CNAFCM), for naturally grouping the
difference between image signal and video signal is pixels of an image into different similar regions
that a video signal keeping of temporal information, when the cluster number is not known beforehand.
which introduces the object motion and includes To calculate the cluster number as initialization
camera motion concept, therefore Video has both parameter of the following FCM clustering, the
temporal nature and spatial (static) nature. author utilize the Grey Level Co-occurrence Matrix
Segmentation of video can thus be temporal, spatial (GLCM) feature selection at the image block level
or spatio-temporal. In a video, segmenting a frame in instead of at the pixel level.[1]
spatial domain is just like a static image. Segmenting
a sequence of video frames in temporal domain is Meenakshi M. Devikar et. al (2013) Histogram
known as temporal segmentation and shot detection. based skillful fuzzy c-means algorithm is proposed
For video data temporal segmentation is used for in this paper for the segmentation. Given image is
considering boundary of shots may operate both in clustered with this proposed Improved Histogram
uncompressed and compressed domain.Temporal based Spatial FCM algorithm . Robustness against
segmentation can identify dynamic objects since noise is enhanced by using the spatial probability of
moving objects have coherent motion that is distinct the nearby pixel. The medical images are denoised
from the background. Spatial segmentation can with effective denoising algorithm before to
examine object boundaries correctly if underlying segmentation. [3]
objects have a different visual appearance from the
background. [9] Sudhanshu Sinha et. al (2014) In this paper,
simplified mean-shift filter and K-Means clustering
both are used in modeling the background. The most
common models are mixture of Gaussian (MOG),
Kernel Density Estimation (KDE), etc. used for
background estimation. With some of the
aforementioned models comparison of the proposed
approach have been made. [2]

ISSN: 2231-5381 http://www.ijettjournal.org Page 107

 International Journal of Engineering Trends and Technology (IJETT) – Volume 36 Number 2- June 2016

RAJASEKHAR NALABOLU et. al (2014) This III. PROPOSED WORK

paper proposes a motion detection system based on
background subtraction using morphological
processing and fuzzy color histogram. In dynamic SLIC Algorithm:
texture scenes, morphological process and filtering
are used effectively for pixel removal from the Superpixels are becoming popular for work in
background that are unwanted. The background computer vision applications. However, there are
subtraction algorithm use a clustering-based feature, few algorithms that output a compact and desired
called fuzzy color histogram (FCH), having an number of regular, superpixels with a low
ability of extremely attenuating color variations computational overhead. A novel algorithm called
generated by background motion. [4] SLIC (Simple Linear Iterative Clustering) in which
clusters pixels are integrated in the five-dimensional
color and image plane space to effectively generate
A Pugazhenthi et. al (2015) It removes the compact, nearly similar superpixels. The simplicity
limitations associated with Fuzzy C-means of our technique makes it extremely easy to use - a
clustering algorithm such as choosing of number of lone parameter identifies the number of superpixels -
clusters initially by selecting from the co-occurrence and the efficiency of the algorithm makes it very
matrix. As the co-occurrence matrix produced the practical. Furthermore, it is more memory efficient
repetition of adjacent gray levels, the proposed and faster than existing methods.
algorithm choose centroids as the gray levels
correlating to the diagonal elements of the co-
occurrence matrix. To the resultant image some Fuzzy C Means Algorithm:
morphological actions were done. The proposed
algorithm consistently removes the background The clustering algorithm, FCM was introduced by
pixels and the performance of the proposed Dunn in 1973 firstly and later extended by Bezdek in
algorithm is evaluated quantitatively. [6] 1981. For image segmentation in pattern recognition
this algorithm has been used as one of the famous
Deepak Kumar Panda et. al (2016) In this context, clustering techniques. In the FCM, each image pixel
author describe a robust background subtraction has certain membership degree corresponding with
technique with three contributions. First, in the each cluster centroid. These membership degrees
background subtraction algorithm, the use of color have values in the range, showing the strength of the
difference histogram (CDH). This is done by association among the pixel and a particular cluster
evaluating the color difference between a pixel and centroid. The FCM algorithm attempts to division
its neighborhood pixels. The use of CDH reduces the every image pixel into a collection of the fuzzy
number of false errors due to the moving cluster centroids using minimizing the weighted
background and illumination variation. Secondly, summation of squared error objective function
the color difference is fuzzified with a Gaussian (U,C) ,
membership function. A novel fuzzy color
difference histogram (FCDH) is finally proposed by
using fuzzy c-means (FCM) clustering and
performed the CDH. The function of FCM clustering
algorithm in CDH minimize the large dimensionality ………… (1)
of the histogram bins in the evaluation and also
lessens the intensity effect variation generated due to Subject to
the change in illumination of the background or
fake motion. [7]

Radhakrishna Achanta et. al (2010) The author

introduce a novel algorithm that use clusters pixels
in the integrated five-dimensional color and image
plane space to productive originate compact, nearly
uniform superpixels. The simplicity of this technique
makes it extremely easy to use - a lone parameter
describe the number of superpixels and it makes this
algorithm very practical due to the efficiency.
Experiments show that this approach produces ……….(2)
superpixels at a minimum computational cost while In image N is the total number of pixels, is the
achieving quality of segmentation equal to or greater membership degree of ith pixel to jth cluster
than four state-of-the-art methods, as estimated by centroid , m is the exponential weight of
boundary recall and under-segmentation error. [10] membership degree that controls the fuzziness of the

ISSN: 2231-5381 http://www.ijettjournal.org Page 108

 International Journal of Engineering Trends and Technology (IJETT) – Volume 36 Number 2- June 2016

resulting partition, and || || is the

distance between and .

……… (3)

Where 1 ≤ j ≤ K and 1≤ i ≤ N . It should be noted

that if then and set anothers
membership degrees of this pixel to 0.

=
…….. (4) Fig. 3 Graphical Representation of PSNR values
Where i≤j≤K and is the multidimensional feature
vector of ith pixel . Taking that values we make a graph. Graphical
representation shows that we get the better results
using FCM algorithm.
Now calculate the average values of both
segmentation and make a graph of that calculated
IV. RESULT ANALYSIS values and it show clearly that we achieve better
segmentation result.
RESULT: First of all we take the frames from the Avg. PSNR value of Existing segmentation = 3.7604
video. Then using SLIC algorithm, track the object
for the further segmentation. Then clustering and Avg. PSNR value of proposed segmentation =
segmentation is done of that object. Now clustering 4.1318
is done with the help of FCM algorithm. Then
calculate the PSNR values of extended and proposed
segmentation. We see in the table that the PSNR
value of proposed technique is high or better than
existing technique.

TABLE 1
PSNR VALUES OF SEGMENTATION

Sr. No. PSNR value of PSNR value of

Existing Proposed
segmentation segmentation
1 3.8539 4.2073
2 3.7592 4.1447
3 3.7277 4.1160
4 3.7305 4.1374
5 3.7932 4.1520
6 3.7665 4.0985
7 3.7505 4.0647 Fig. 4 Graphical Representation of average PSNR
8 3.8572 4.1691 values
9 3.8050 4.1603
10 3.7614 4.1025 In this graph it is clearly seen that the proposed
technique gives the better result of quality of the
11 3.6915 4.0531
video segmentation.
12 3.7428 4.2603
13 3.6855 4.0862
14 3.5436 3.9356
15 3.9382 4.2907

ISSN: 2231-5381 http://www.ijettjournal.org Page 109

 International Journal of Engineering Trends and Technology (IJETT) – Volume 36 Number 2- June 2016

[7] Deepak Kumar Panda, Sukadev Meher, “Detection of Moving

V. CONCLUSION Objects Using Fuzzy Color Difference Histogram Based
Background Subtraction”, IEEE SIGNAL PROCESSING
LETTERS, VOL. 23, NO. 1, JANUARY 2016.
In digital video processing and computer vision, one
[8] S. Padmakala and Dr. G.S. AnandhaMala, “A Novel Video
of the most challenging and active research areas is Object Segmentation Approach for Noisy Video Sequences
video object segmentation. A significant issue for the towards Effective Video Retrieval”, International Journal of
successful use of many video sequences is video Computer Theory and Engineering, Vol.2, No.6, December, 2010.
object segmentation that accentuates partitioning the
[9] Mada Amarnadh, S. Asif Hussain, M. Janardhana Raju, “A
video frames to semantically meaningful video Review of Video Segmentation Techniques” , Global Journal of
objects and backgrounds. Video object segmentation Advanced Engineering Technologies, Special Issue (CTCNSF-
is a vital operation for content-based video coding, 2014).
multimedia content description, intelligent signal
[10] Radhakrishna Achanta, Appu Shaji, Kevin Smith,” SLIC
processing and more. SLIC is used for object tracking Superpixels”, EPFL Technical Report 149300, June 2010.
and FCM is used for clustering. FCM clustering is a
hard and an unsupervised clustering technique which [11] V. M. Vijay Kannan, “High Accuracy Detection and Tracking
will be applied to image segments to clusters with of Objects” , IOSR Journal of Computer Engineering (IOSRJCE)
,ISSN: 2278-0661, ISBN: 2278-8727Volume 8, Issue 4 (Jan. - Feb.
spectral properties. FCM use the distance between 2013)
pixels and cluster centers in the spectral region to
compute the membership function. In this paper we [12] M. Arfan Jaffar, Bilal Ahmed, Nawazish, Naveed, Ayyaz
calculate the PSNR (peak to signal noise ratio) values Hussain, Anwar M. Mirza , “Color Video Segmentation using
Fuzzy C-Mean Clustering with spatial Information” , Department
of existed segmentation and proposed segmentation. of Computer Science , National University of Computer and
The comparison is based on the PSNR values. The Emerging Sciences, A.K. Brohi Road H-11/4, Islamabad,
Proposed technique gives the high values of PSNR. PAKISTAN , ISSN: 1790-5052, Issue 5, Volume 5, May 2009.
Graphical representation of PSNR values shows the
better quality of segmentation.

References
[1] Shaoping Xu, Lingyan Hu, Xiaohui Yang and Xiaoping Liu,
“A Cluster Number Adaptive Fuzzy c-means Algorithm for Image
Segmentation” , International Journal of Signal Processing, Image
Processing and Pattern Recognition Vol.6, No.5 (2013), pp.191-
204.

[2] Sudhanshu Sinha, Manohar Mareboyana, “Video Segmentation

into Background and Foreground Using Simplified Mean Shift
Filter and K-Means Clustering”, Bowie State University, Bowie,
MD 20715, April 3-5, 2014.

[3] Meenakshi M. Devikar and Mahesh Kumar Jha, “

Segmentation of images using histogram based FCM clustering
algorithm and spatial probability”, Department of
Telecommunication Engineering, CMRIT, Bangalore, India,
International Journal of Advances in Engineering & Technology,
Mar. 2013.

[4] RAJASEKHAR NALABOLU, D. V. S. NAGENDRA

KUMAR, “Object Identification in Digital Videos using Color
Histogram Bins and Fuzzy C-Means Clustering Techniques” ,
Mahathma Gandhi Institute of Technology, Hyderabad, India,
ISSN 2319-8885 Vol.03,Issue.46 December-2014.

[5] Jiamin Ning, Yang Yang, and Fei Zhu, “ Background Modeling
and Fuzzy Clustering for Motion Detection from Video” , School
of Computer Science and Technology, Soochow University,
Suzhou, Jiangsu, China, 215006, JOURNAL OF MULTIMEDIA,
VOL. 8, NO. 5, OCTOBER 2013.

[6] A Pugazhenthi, G.Sreenivasulu, A Indhirani, “Background

Removal by Modified Fuzzy C-Means Clustering Algorithm”,
2015 IEEE International Conference on Engineering and
Technology (ICETECH’15), 20th March 2015, Coimbatore, TN,
India, 978-1-4799-1854 6/15/$31.00 ©2015 IEEE.

ISSN: 2231-5381 http://www.ijettjournal.org Page 110