K-Means Clustering in Moving Objects Extraction with Selective Background

K-Means Clustering in Moving Objects Extraction with
Selective Background
Moch Arief Soeleman1
, Aris Nurhindarto2
, Ricardus Anggi Pramunendar3
Faculty of Computer Science Dian Nuswantoro University
Semarang, Indonesia
arief22208@gmail.com1
, arisnurhindarto@yahoo.com2
, ricardus.anggi@research.dinus.ac.id3
Abstract - We presents a technique for moving objects
extraction. There are several different approaches for
moving object extraction, clustering is one of object
extraction method with a stronger teorical foundation
used in many applications. And need high performance in
many extraction process of moving object. We compare
K-Means and Self-Organizing Map method for extraction
moving objects, for performance measurement of moving
object extraction by applying MSE and PSNR. According
to experimental result that the MSE value of K-Means
is smaller than Self-Organizing Map. It is also that
PSNR of K-Means is higher than Self-Organizing Map
algorithm. The result proves that K-Means is a promising
method to cluster pixels in moving objects extraction.
Keywords : extraction, clustering, moving object.
I. INTRODUCTION
The video application areas automated visual
observation of a person or group, content-based visual
image analysis, video tagging, and a mutual human-
computer are facing important and bold problem which is
moving object extraction. The video segmentation
process must be successfully taken before we move to
next processes such as extraction process on feature,
identification, and the basic cognitive process of
arranging into categories.
The video moving objects extraction process is
aimed to separeted into parts an image sequence into
typically particular areas which is able to enclose
significant labels afterward, in which a set of region is
broken down with the exact same one and similar kind
attributes such as pixel degree, a visual attribute, motion.
Problems of moving objects extraction have been
discussed in many literature, in which according to their
primary approaches is roughly classified into three
categories, they are: dissimilar temporal [1] [2]; motion
optical flow [3] [4] [5]; and background difference.
. The initial mode for analysing the frame sequence in
video is through background model in [6] that compose in
maintaining recent shape of moving objects from the
background element. The background model is useful for
segmenting video streams of the background to
foreground.
The background subtraction is commonly applied to
moving object recognition, which contains in upholding
an update archetype of background and perceiving
moving objects as those that diverge from such an
archetype. In the comparison to other oncoming, for
example optical flow in [8], this oncoming is feasible for
the actual time that applications takes a process to occur
by computation process.
Based on the background subtraction in [9], we
determines to apply selective background in moving
object extraction. In this paper, we assess the
performance of clustering algorithm for extraction in
moving objetcs.
II. RELATED WORK
An efficient background registration technique
algorithm for efficient segmentation moving object have
proposed in [10], this method was applied to structure
consistent background from collected different frame
motion information. This technique separates area by
comparing the existing frame from the structured
background.
A vigorous foreground partitioning algorithm have
presented in [11], this approach is used to put into group
whether it includes in the part of a scene behind objects in
the foreground or employed a several intensity and refine
a distinguishing information which having structure
process in the later.
The consistent foreground segmentation approach have
projected in [12], researchers incorporate temporal image
analysis and recommendation background frame to
overcome the glitch occurs on outdoor daylight sections
which cause adjustment of the intensities on the
background recommendation image of moving object
segmentation. The purpose of using transient image
analysis is to discover the object in every frame whether
it is moving or static that emerged problem in background
model.
Other approaches have been used in [13], the method
combines two video segmentation technique using key-
frame retraction and object-based method which have
International Journal of Computer Science and Information Security (IJCSIS),
Vol. 16, No. 6, June 2018
132 https://sites.google.com/site/ijcsis/
ISSN 1947-5500

Selective Background
Video Sequence Extracted
Clustering Process
Classify Pixels &
Morphology
Object Extraction
Region Oriented Image for
MSE and PSNR Measurement
been contructed for effective and robust based video
segmentation algorithm and statistical clustering.
III. REVIEW OF RELATED THEORY
In this section, we describe of how extraction moving
objects based on clustering technique between K-Means
and Self-Organization Map algorithm are presented.
a. K-Means Algorithm
The k-means algorithm is a hard clustering technique that
divides the objects into k clusters, until each objects are
being clustered to one and only one membership to
minimize the sum of squared distances between each data
point and the empirical mean of the corresponding cluster.
Algorithm 1: K-Means Algorithm
1) Select objects to be k initial centroid randomly
2) Estimate the distance of each centroid of each
object by using space or similarity metric; use
nearest centroid point to define each object at the
cluster.
3) Calculate the latest centroid point.
4) Observe the result. If it turns to be different from
the previous one, then return to step 2
3.2 Self-Organization Map Algorithm.
Self-Organization Map (SOM) is an iterative algorithm in
[16] [17] and one of the widely used algorithm for.
clustering. SOM comprises the competitive and
cooperative stage
IV. CLUSTERING-BASED OF MOVING
OBJECT EXTRACTION
In this section, we describe of how to extract moving
object by using clustering techniques modelling. In each
frame, there are steps which is necessary by to perform
moving object extraction and it is shown in Fig. 1. The
sub steps are described below:
Figure 1. Diagram of Extraction Moving Object
a. Background Subtraction
In case of our background model with selective
background is applied to detect the intensity different of
current and background image, we addopt double
different method also known three different method [18].
In the early stages video files are captured and broken
into digital images based on video frames.
The extraction process is performed on a video
where for each frame in a certain time unit is converted
into digital image form. Digital imagery is generated in
the form of JPEG (Joint Photographic Experts Group).
Next to each pixel in the digital image is converted to a
double type that has a range of values between 0 and 1.
The pixel value 0 for the weak colour component
and the value of 1 means a strong colour component.
Although converted into a double type, but the digital
image is still in the RGB colour domain.
This following step of background subtraction with
selective background:
1. Extract all frames on the video
2. Search for background frames automatically by
calculating the mode values in each frame
Algorithm 2 : Self-Organizing Map algorithm
1. Initialize the learning rate , radius of the
neighbor function and random values for
the initial weight
2. Repeat until α reaches 0
a. For k=1 to n
b. The competitive stage: for all
find the winning neuron
that minimize ‖ ‖
c. The cooperative stage: renew each unit
‖ ‖
d. Lessen the rate of and
Vol. 16, No. 6, June 2018
ISSN 1947-5500

3. Convert Current Frame and Background Frame
to grayscale image
4. Reduce between the two frames
5. Convert the resulting image to binary image
6. Perform morphological operations to eliminate
noise
7. Make image of morphological operation result as
masking to visualize moving object
The result of background subtraction with selective
background can show in figure 2.
(a) Background Frame (b) Frame #697
(c) Moving Object with Background Subtraction
Figure 2. Result of object detected with Background
Subtraction
b. Clustering using K-Means, SOM Algorithm
In this stage, we used two methods for moving object
extraction and tracking which is in paralel process
between SOM and K-means. Two different dataset have
been used in each experiment for moving object
extraction.
c. Morphology
A better extraction result is significant, so it needs
morphology in the performance [19]. In the manipulation
process of image features, mmorphology is applied which
is based on shape [20], using basic operation such as
dilatation, erosion, opening, and closing. Sequential
combination of dilatation and erosion is presented in
opening and closing. In [21], it is stated that the aim of
opening process erosion which is followed by dilatation is
encircling corner from inside the objects to obtain filter
detail and simplified images. Meanwhile, small gaps
within the object are closed by the closing (dilatation
followed by erosion). This paper applies closing to
eliminate the flawed in foreground recognition.
d. ROI Cropping for measurement
This stage is processed to create image ground truth,
where as the human operation is cropping region of
interest image reference in moving object clustering for
comparing the performance of moving object extraction
to calculate the MSE and PSNR.
V. EXPERIMENTAL RESULT
a. Data and Results
Algorithms implemented in the process of moving objects
extraction was having an experimental result aimed for
image sequences. It had been proved in the performance
of the proposed method is tested in a sequence of moving
images in real video. We defined two sequences which
represented significant standard situations for video
surveillance systems. The video processing was applied
on moving objects in which the goal intended to be
attained extracted moving objects in the building. We
utilized Matlab sofware ver. 2017b and RAM on PC with
processor i3-6100, 3.70 GHz, with memory 4.00 GB.
1) Sequence Walk1 : Sequence Walk1 of the database
CAVIAR Project1
was labeled and comprise 611 frames
of 484 x 288 in spatial resolution, acquired at frequency
25 fps. It was an example of difficult sequences, where
the lighting condition was not as clear as previous area
and the moving human tended to cover-up the path.
2) Sequence Walk2 : Sequence Walk2 of the 2nd
database
CAVIAR project was labelled and comprised 700 frame
of 388 x 288 in spatial resolution, attained at frequency
25 fps. We have been assigned to test the method
capability to segmenting more than one moving object.
Finally, we found that K-Means was quite successful in
moving objects extraction.
a. Background Frame b. Frame #590
c. Object Detected d. Object Tracked
Figure 3. Result of moving object extraction using K-
Means (Walk1 dataset)
1
http://homepages.inf.ed.ac.uk/rbf/CAVIAR/
Vol. 16, No. 6, June 2018
ISSN 1947-5500

a. Background Frame b. Frame #697
c. Object Detected d. Object Tracked
Figure 4. Result moving object extracted using K-
Means with Rest_WiggleOnFloor dataset.
b. Performance Evaluation
In the measurement of performance in the process of
moving object extraction, Mean Square Error (MSE) and
Peak Signal to Noise Ratio (PSNR) were applied. Both
measurements are used for calculate the altered quality
that renders of extracted and ground truth of image frame
[22] in which better image segmentation was by having
lower value of MSE and higher value of PSNR [23].
Those values of MSE and PSNR were obtained by the
measurement process using [12] and [23], respectively.
   
1 1
1
, ( , ) ( , )
M N
h j
MSE R Q R h j Q h j
MN  
  (1)
 
 
2
10
max
, 10.log
,
PSNR R Q
MSE R Q
 
   
 
(2)
In which represents ground truth image, represents
extraction frame of size and max is image
maximum achievable pixel value .
Figure 5. MSE of Walk1 Dataset using K-Means, SOM
0
5000
10000
15000
20000
25000
30000
1 3 5 7 9 11 13 15 17 19 21 23 25 27 29 31 33 35 37 39 41 43 45 47 49 51 53 55 57
MSE
Frame
Kmeans SOM
Vol. 16, No. 6, June 2018
ISSN 1947-5500

Figure 6. PSNR of Walk1 Dataset using K-Means, SOM
In two video used, we evaluated the comparison of
object moving extraction. They are, video of human
walking sequence in hall, and two people walking from
same direction. Table 1 and Table 2 showed the MSE
and PSNR of two video using K-Means and Self
Organizing Map algorithm. K-Means produced better
extraction result.
TABLE 1. Average MSE of K-Means and SOM
No. Dataset K-Means SOM
1 Dataset1 13.83 13.97
2 Dataset2 9.7 9.80
The MSE of K-Means was having result that was lower
than the MSE SOM, and the PSNR was higher than the
PSNR of SOM. Fig. 4 and Fig. 5 illustrated the MSE
and PSNR of dataset Walk1, correspondingly.
TABLE 2. Average PSNR of K-Means and SOM
No. Dataset K-Means SOM
1 Dataset1 6,93 6,87
2 Dataset2 8,66 8,66
VI. CONCLUSIONS
We presented study of moving object extraction by
using clustering techniques. Based on the results of
research and experiments that have been done, it can be
concluded that background subtraction techniques with
a selective background to produce a good detection
process. In static environments with indoor locations
where the intensity of the lighting is relatively fixed, the
background used can be manually modelled. However,
in an environment with dynamic conditions, an adaptive
background to environmental conditions is required.
This research can also detect pedestrian objects quite
well only by using selection techniques based on the
size of the object. To improve accuracy, a comparison
technique can be performed with pre-prepared training
data. In addition, based on the results of trials that have
been done, the proposed method
The outcome showed which the achievement of object
moving extraction using K-Means is better than SOM
algorithm. K-Means generated smaller MSE and greater
PSNR opposed to SOM.
References
[1] N. Paragios and R. Deriche, “Geodesic active
contours and level sets for the detection and
tracking of moving objetcs,” IEEE Trans. Pattern
0
2
4
6
8
10
12
14
1 3 5 7 9 11 13 15 17 19 21 23 25 27 29 31 33 35 37 39 41 43 45 47 49 51 53 55 57 59
PSNR
Frame
K-Means SOM
Vol. 16, No. 6, June 2018
ISSN 1947-5500

Analysis and Machine Interface, vol. 22, no. 3, pp.
266-280, 2000.
[2] K. Ng and E. Delp, “Object tracking initialization
using automatic moving object detection,” in Proc
of SPIE/IS&T Conference on Visual Information
Processing and Communication, January 2010.
[3] Hui.Y, Yilin.C, Yanzhuo.M, Donglin.B and
Zhaoyang.L, “Motion feature descriptor based
moving objects segmentation,” High Technology
letters, vol. 18, no. 4, pp. 84-89, 2012.
[4] S. Fejes and L. Davis, “What can projections of
flow fields tell us about the visual motion,” in
ICCV Conference, Bombay, India, 1998.
[5] L. WIxson and M. Hansen, “Detecting salient
motion by accumulating directional-consistent
flow,” in ICCV Conference, Corfu, Greece, 1999.
[6] Bovic, The hand book of image and video
processing, Academic Press, 1998.
[7] K. Srinivasan, K. Porkumaran and G.
Sainarayanan, “Improved background subtraction
techniques for security in video applications,” in
3rd International Conference on Anti-couterfeiting
Security, and Identification in Communication ,
2009.
[8] M. M. A. e. Azeem, “Modified background
subtraction algorithm for motion detection in
surveillance systems,” Journal of American Arabic
Academy for Science and Technology, vol. 1, no. 2,
pp. 112 - 123, 2010.
[9] L. Maddalena and A. Petrosino, “A self-organizing
approach to background subtraction for visual
surveillance applications,” IEEE Transactions on
Image Processing , vol. 17, no. 7, pp. 1168 - 1177,
July; 2008.
[10] S. Y. Chien, S.-Y. Ma and L.-G. Che, “Efficient
moving object segmentation algorithm using
background registration technique,” IEEE Trans.
Circuit and System for Video Technology, vol. 12,
no. 7, 2002.
[11] H. Kim, R. S. I. Kitabara, T. Toriyama and K.
Kogure, “Robust foreground segmentation from
color video sequences using background
subtraction with multiple threshold,” in Proc.
KPJR, 2006.
[12] P. Spagnolo, T. D. Orazio, M. Leo and A. Distante,
“Moving object segmentation by background
subtraction and temporal analysis,” Journal Image
and Vision Computing, vol. 24, pp. 411 - 423,
2006.
[13] L. Liu and G. Fan, “Combined key-frame
extraction object-based video segmentation,” IEEE
Trans. Circuit and Systems for Video Technology,
vol. 15, no. 7, 2005.
[14] V. Bhandari, K. B. R and K. M.M, “Moving object
segmentation using fuzzy c-means clustering affine
parameters,” Computer Network and Intelligent
Computing Communication in Computer and
Information Science, vol. 157, pp. 205-210, 2011.
[15] S. Shambharkar and S. Tirpude, “Fuzzy C-Means
clustering for content based image retrieval
system,” in Proc of International Conference on
Advancements in Information Technology with
workshop of ICBMG, 2011.
[16] T. Kohonen, “Self-organized formation of
topologically correct feature map,” Biological
Cybernetics , vol. 43, no. 1, pp. 59-69, 1982.
[17] T. Kohonen, “Self-organizing map,” Springer
Verlag, 1997.
[18] B. Sugandi, H. S. Kim, J. K. Tan and S. Ishikawa,
“Tracking of moving object by using low
resolution image,” in International Conference on
Innovative Computive, Information and Control,
2007.
[19] K. Mahesh and K. Kuppusamy, “A new hybrid
video segmentation algorithm using fuzzy c-means
clustering,” International Journal of Computer
Science , vol. 9, no. 2, 2012.
[20] C. Nagaraju, S. Nagamani, G. R. Prasad and S.
Sunitha, “Morphological edge detection algorithm
based on multi-structure elements of different
directions,” International Journal of Information
and Communication Technology Research, vol. 1,
no. 1, 2011.
[21] A. Amer, “New binary morphological operations
for efective low-cost boundary detection,”
International Journal of Pattern Recognition and
Artificial Intelligence, vol. 17, no. 2, 2002.
[22] K. Bhoyar and O. Kakde, “Color image
segmentation based on color histogram,”
International Journal of Image Processing (IJIP),
vol. 3, no. 6, pp. 282-293, 2010.
[23] Z. Qu, “Two allgorithms of image segmentation
and measurement method of particle's parameters,”
An International Journal Applied Mathematic and
Information Science, no. 1, pp. 105 - 109, 2012.
Author Profile
M. Arief Soeleman Graduated from Dian Nuswantoro
University in 1999, then continued his Master's Degree in
Informatics Engineering at Dian Nuswantoro University
and graduated in 2004. Graduated his Doctoral Program
at the Electrical Engineering Program in 2016 at Sepuluh
Nopember Institute of Surabaya. Field of Research on
computer vision and image processing
Vol. 16, No. 6, June 2018
ISSN 1947-5500

Aris Nurhindarto Graduated from Dian Nuswantoro
University in 1999, then continued his Master's Degree in
Informatics Engineering at Dian Nuswantoro University
and graduated in 2004. Field of interest Photography
research and information system
Ricardus Anggi Pramunendar, Graduated from Dian
Nuswantoro University in 2009, then continued his
Master of Computer Science program at UTEM Malaysia
and graduated in 2012. He is currently completing a
Doctoral program at Gadjah Mada University majoring in
Electrical Engineering. Research and numerous
publications on computer vision and image processing
Vol. 16, No. 6, June 2018
ISSN 1947-5500

K-Means Clustering in Moving Objects Extraction with Selective Background

More Related Content

What's hot (19)

Similar to K-Means Clustering in Moving Objects Extraction with Selective Background (20)

Recently uploaded (20)

K-Means Clustering in Moving Objects Extraction with Selective Background