Review paper on segmentation methods for multiobject feature extraction

IJRET: International Journal of Research in Engineering and Technology eISSN: 2319-1163 | pISSN: 2321-7308
_______________________________________________________________________________________
Volume: 03 Issue: 12 | Dec-2014, Available @ http://www.ijret.org 64
REVIEW PAPER ON SEGMENTATION METHODS FOR
MULTIOBJECT FEATURE EXTRACTION
Sonali Deshmukh1
, Anjali Yadav2
1
ME (IInd
year), Department of ENTC, Smt Kashibai Navle College of Engineering, Maharashtra, India
2
Asst. Prof., Department of ENTC, Smt Kashibai Navle College of Engineering, Maharashtra, India
Abstract
Feature extraction and representation plays a vital role in multimedia processing. It is still a challenge in computer vision system
to extract ideal features that represents intrinsic characteristics of an image. Multiobject feature extraction system means a
system that can extract features and locations of multiple objects in an image. In this paper we have discuss various methods to
extract location and features of multiple objects and describe a system that can extract locations and features of multiple objects
in an image by implementing an algorithm as hardware logic on a field-programmable gate array-based platform. There are
many multiobject extraction methods which can be use for image segmentation based on motion, color intensity and texture. By
calculating zeroth and first order moments of objects it is possible to obtain locations and sizes of multiple objects in an image.
Keywords: multiobject extraction, image segmentation
--------------------------------------------------------------------***---------------------------------------------------------------------
1. INTRODUCTION
In recent years computer programs are become popular for
processing digitize pictorial information.This area of digital
pictorial information processing involved performing
“local” tasks on picture “neighbourhoods.”As researchers
have shown,many of picture processing transformations that
can be achieved by applying certain tasks independently or
simultaneously to each element of the given image. Many
multiobject extraction methods which can be use for image
segmentation are based on motion, color intensity and
texture. In Image segmentation image is divide into regions
or categories, corresponding to the different objects or parts
of objects. Image segmentation methods such as watershed
transformation, graph-cut methods, clustering methods and
neural network approaches have many recent developments.
Most of these methods are suitable for complicated scenes in
many cases such as heart surgery ,brain surgery or
multimedia. This paper focus on modification of algorithms
use in tracking multiple targets with imaging sensors used
onboard an air airborne platform.
2. LITERATURE SURVEY
2.1 Watershed based Image Segmentation
Paper [3] study the watersheds in edge-weighted graphs and
define the watershed cuts following the intuitive idea. of
drops of water flowing on a topographic surface. Watershed
cuts is used as a fundamental method in many powerful
segmentation procedures. The watershed cut correspond to
idea of drops of water on a topographic surface flows
towards the "nearest" minimum. the uniformity of
watersheds can be defined by catch basins. Then by
applying an equivalence theorem, their optimality can be
proved in terms of spanning forests. After this, linear time
algorithm is use to compute them. This is the most efficient
algorithm in theory as well as in practice. If we consider
gray scale image as a topographic surface the pixel gray
level becomes the elevation of a point, the deeper parts like
valley and the basins are correspond to dark areas of the
topographic surface, whereas the crest lines and higher
points correspond to the light areas. In watershed method set
of points are divide such that they satisfied “drop of water
principle”.
In first step the uniformity of watershed cuts is establish, by
proving that they can be equivalently defined by their
catchment basins or by separating lines through the drop of
water principle. Second step establishes the optimality of
watershed-cuts. the link between minimum spanning forests
and flooding from marker algorithms is shown in [19], by
F. Meyer. Third step consists of a linear-time algorithm
which computes the watershed-cuts of an edge weighted
graph
2.2 Grab-Cut based Image Segmentation
For grab cut method the paper [5], focus on a new technique
for general purpose interactive segmentation of N-
dimensional images in which certain pixels are marked as
“object” or “background” to provide hard constraints for
segmentation whereas soft constraints include both
boundary and region information. The globally optimal
segmentation of the N-dimensional image is found by grab
cuts. The obtained solution satisfying the constraints which
gives the best balance of boundary and region properties
among all segmentations. The topology of our segmentation
consists of several isolated parts for both “object” and
“background” segments. This paper presents some
experimental results in the context of medical image
segmentation and photo/video editing. In segmentation
process “object” or “background provides clues on what the
user intends to segment and then computing a global
optimum among all segmentations for rest of the image.

_______________________________________________________________________________________
Boundary and region properties of the segments are defined
as cost function. These properties can be considered as soft
constraints for segmentation. User can add or remove any
hard constraints and can recomputed efficiently using
globally optimal segmentation. This helps to get any desired
segmentation results quickly via very intuitive interaction.
2.3 Clustering based Image Segmentation
In [6] author present an implementation of Lloyd's k-means
clustering algorithm, which is the filtering algorithm. This
algorithm requires a kd-tree as the only major data structure.
It is easy for implementation. The practical efficiency of the
filtering algorithm is establish in two ways. Firstly by
presenting a data-sensitive analysis of the algorithm's
running time, this shows that the algorithm runs faster as the
separation between clusters increases. Secondly it focus on a
number of empirical studies both on real data sets from
applications in color quantization, data compression, image
segmentation and on synthetically generated data. The
simplest method of image segmentation is thresholding
technique. This technique is based on a clip-level (or a
threshold value) to turn a gray-scale image into a binary
image. The key of this technique is to select the threshold
value (or values when multiple-levels are selected). Several
popular methods are used in industry including the
maximum. entropy method, Otsu's method (maximum
variance), and k-means clustering.
2.4 Artificial Neural Network based Segmentation
Paper [7] describes artificial neural network. It is often
called a neural network which is an artificial representation
of human brain that tries to simulate its learning process. It
is widely use in medical image segmentation. Neural
network constitutes a large number of parallel nodes and
Each of them can perform some basic computing. It is based
on life simulation, especially the human brain's learning
process. By transferring the connections among nodes and
connection weights the learning process can be achieved.
During the image segmentation process neural network can
also reduce the requirements of expert intervention.
Table 1: Comparision of different segmentation methods
Segmentat
ion
technique
Methodology Advantages Disadvantag
es
Watershed
Transfor
Mation
Watersheds are
defined by
catchment
basins which
are separated by
dividing lines.
Then Prove
their optimality
in terms of
minimum
spanning forests
The
proposed
algorithm is
most
efficient in
both
theory and
in
practice
 Mo
re sensitive
to noise
 Not
suitable for
High
Resolution
Graph
–Cut
Here the
detailed
This
technique
multiway
cut problem
technical
description of
basic
combinatorial
optimization
frame work for
segmentation
via s/t graphs
has
great
potential for
solving
problems
in
vision and
graphics
is hard and
thus
guaranteed
global
optimality
can be loose
Clustering K –means
clustering
method is used
Efficiency is
increase
because data
points do
not
vary
through-
out
the
computa-
tion
Algorithm
used is
quite
complex
and slow in
practice
Neural
Network
Approache
s
Neural
networks are
use to perform
classification or
clustering
No need to
write
complicated
programs
 It
requires
Long
training time
 Res
ult may be
affected by
initialization
3. PROPOSED SYSTEM
Most of the video cameras used in general systems are
restricted to some specific video formats. These formats are
suitable for characteristics based on human eye which
means the system processing speed is limited to recognition
speed of human eye .This system describes a model that can
extract locations and features of multiple objects in an image
at high frame rate.
Fig 1.Twenty –five local patterns for HLACs
Here a cell-based multiobject feature extraction algorithm is
implemented on hardware. Hardware consists of 25 higher-
order local autocorrelation HLAC feature of 1024 objects in
an image that can be simultaneously extracted for
multiobject recognition.This algorithm can reduce memory
consumption and computational complexity in labeling

_______________________________________________________________________________________
process for multiobject extraction by exchanging its
computational sequence in labeling process of divided cells
which are obtained after calculating the cells moments of
n×n pixels based on the additivity in calculation of moment.
3.1 Implementation
Fig 2. Configuration of High Speed vision: IDP Express
For camera head, FASTCAM MH4-10K (Photron) is used
as input. There are two FPGA’s in hardware, configuration
programmable read-only memories for the FPGAs, serial-to-
parallel converters for camera inputs, and first-in first-out
(FIFO) memories for data transfer between the FPGAs on a
IDP express board. Here, 8-bit color images can be captured
on its image sensor by using a Bayer color filter which were
transferred at 2000 f/s for 512×512 pixels with digital serial
communication in parallel with six pixels FPGA 1 is use to
control camera I/O and PCI-e bus. FPGA 2 is used for
hardware implementation of user-specified tasks. PCI-E bus
is use to connect IDP express board to PC. 8-bit 512×512
images can be transferred at 4000 f/s for two camera heads
[1].
3.2 Hardware implementation logic
The hardware implementation logic of multiobject feature
extraction can be divided into several modules which
includes binarization module ,a cell-based labeling module
,a cell-based features calculation module, and a data selector
module for FIFO output. Figure above shows the schematic
data flow of the implemented circuit.
The binarization module perform the function of converting
512×512 input images into binary images by scanning from
the upper left to the lower right in units of eight pixels using
X and Y address signals. This module can convert both color
and gray images into binary images in parallel for 8 pixels.
For color image a color conversion submodule converts
RGB images into 8-bit HSV color images ,hue H(r),
saturation S(r), and value V(r), in parallel with 8-pixel data
after RGB conversion. After this HSV images are converted
to a binary image B(r) by specifying a certain vivid and
bright color with four thresholds of 𝜃 𝐻𝐿, 𝜃 𝐻𝐻 , , 𝜃𝑆, 𝜃 𝑉.
When 𝜃 𝐻𝐿 < 𝜃 𝐻𝐻
B(r)=
1, (𝐻 ∈ 𝜃 𝐻𝐿, 𝜃 𝐻𝐻 , 𝑆 ≥ 𝜃𝑆, 𝑉 ≥ 𝜃 𝑉
0, 𝑜𝑡ℎ𝑒𝑟𝑤𝑖𝑠𝑒 (1)
Otherwise
B(r)=
1, (𝐻 ∉ 𝜃 𝐻𝐻 , 𝜃 𝐻𝐿 , 𝑆 ≥ 𝜃𝑆, 𝑉 ≥ 𝜃 𝑉)
For Gray image sensor, an input image I(r) is converted to
binary image B(r) with threshold 𝜃 𝐿 is given below
I(r) =
1, (𝜃 𝐿 ≤ 𝐼(𝑟))
Fig. 3. Schematic data flow for implemented circuit
The cell based feature extraction module performs
calculation of 25 HLACS , Fi(Γab) (i = 0, . . . , 24), and two
first moments, M10(Γab) and M01(Γab), for 4096 cells Γab
of 8×8 pixels by using a two-line cache, 27 neighbor-
processing submodules, and 27 cell-based summation
submodules. Zerothorder moment can be calculated by
sharing the circuit for the zerothorder HLAC in this module.
Cell-Level HLAC Calculation:
For every cell Γab of n×n pixels 25 HLACs 𝐹𝑖( 𝛤𝑎𝑏 ) are
calculated as follows.
𝐹𝑖( 𝛤𝑎𝑏 ) = 𝐵 𝑟 𝐵 𝑟 + 𝑎1 𝐵(𝑟 + 𝑎2)𝑟∈𝛤 𝑎𝑏
(4)

_______________________________________________________________________________________
The cell-based labeling module can use a flag map
submodule, a labeling submodule, and 27 label-domain
summation submodules to calculate 25 HLACs and two
first-order moments,for 1024 labeled regions .
The data selector module select FIFO output for an external
PC with X and Y address signals from the input image or the
27 types of label-domain features of 1024 labeled objects.
This all circuit module can be implemented on FPGA 2 of
the IDP Express board
4. CONCLUSION
In this review paper of feature extraction, the overview of
various segmentation methodologies applied for digital
image processing is explained. It also describe the proposed
system based on cell-based multiobject feature extraction
Algorithm which is implemented on hardware and can
extract sizes, positions, and HLACs of 1024 labeled objects
in an image by dividing a 512×512 image into 64×64 cells,
and it can also perform multiobject extraction of 512×512
images in real time at 2000 f/s.
REFERENCES
[1] Qingyi Gu, Takeshi Takaki, and Idaku Ishii, Fast
FPGA-Based Multiobject Feature Extraction,” IEEE
Transactions On Circuits And Systems For Video
Technology, Vol. 23, No. 1, January 2013.
[2] G. Hamarneh and X. Li, “Watershed segmentation
using prior shape and appearance knowledge,”
Image Vis. Comput., vol. 27, no. 1, pp. 59–68,2009
[3] J. Cousty, G. Bertrand, L. Najman, and M. Couprie,
“Watershed cuts: Minimum spanning forests and the
drop of water principle,” IEEE Trans. Patt. Anal.
Mach. Intell., vol. 31, no. 8, pp. 1362–1374, Aug.
2009.
[4] Rother, V. Kolmogorov, and A. Blake, “GrabCut:
Interactive foreground extraction using iterated
graph cuts,” presented at the ACM SIGGRAPH,
New York, 2004.
[5] Y. Boykov and G. Funka-Lea, “Graph cuts and
efficient N-D image segmentation,” Int. J. Comput.
Vis., vol. 70, no. 2, pp. 109–131, 2006.
[6] T. Kanungo, D. M. Mount, N. S. Netanyahu, C. D.
Piatko, R. Silverman, and A. Y. Wu, “An efficient
k-means clustering algorithm: Analysis and
implementation,” IEEE Trans. Patt. Anal. Mach.
Intell., vol. 24, no. 7 pp. 881–892, Jul. 2002.
[7] H.P. Narkhede Review of Image Segmentation
Techniques International Journal of Science and
Modern Engineering (IJISME) ISSN: 2319-6386,
Volume-1, Issue-8, July 2013
[8] Y. Boykov and M.-P. Jolly, “Interactive graph cuts
for optimal boundary and region segmentation of
objects in N-D images,” in Proc. Int. Comput. Vis.,
2001, pp. 105–112.
BIOGRAPHIES
Ms. Sonali P. Deshmukh, received the Diploma in
Engineering and B.E. degrees in Electronics and
Telecommunication Engineering from Sant Gadge Baba
Amravti University in 2009 and 2012, respectively.
Currently pursuing M.E in VLSI & Embedded Systems
degree from SKNCOE, Pune under Savitribai Phule Pune
University
Mrs. Anjali A. Yadav, currently assistant Professor in
Department of Electronics and Telecommunication
Engineering ,SKNCOE, Pune under Savitribai Phule Pune
University

Review paper on segmentation methods for multiobject feature extraction

Recommended

More Related Content

What's hot (20)

Similar to Review paper on segmentation methods for multiobject feature extraction (20)

More from eSAT Journals (20)

Recently uploaded (20)

Review paper on segmentation methods for multiobject feature extraction