Image Segmentation using Active Contour
with Level Set
Mandeep Kaur
Electronics & Electrical Engineering Department
Lovely Professional University, Punjab, India. 144402
kaurmandeep1190@gmail.com
Abstract – In this paper a novel algorithm for image
segmentation has been proposed based on active
contour model and level set. In this we use the signed
pressure force function using local information of the
image to be segmented. First the level set function is
selectively penalised to be binary n then a Gaussian
kernel is applied for smoothing. Thus, this model can
work with heterogeneous images. In addition, by
taking the advantages of Geodesic active contour
(GAC) and Chan-Vese (C-V) model, the method
could deal with objects even with discrete blur
boundaries and gives exact results in detecting object
boundaries. One of its other advantages over C-V
method is that the cost of re initialisation is also
reduced in this method as we do not need re-
initialisation of the level set function. Experimental
results demonstrate that the proposed model is
effective in segmenting biomedical images and the
images with blur and weak edges.
Keywords: Image Segmentation, Active Contours,
Snakes, Level Sets, GAC, C-V method.
I.INTRODUCTION
The basic purpose of segmentation is to identify
and isolate the region of interest from the given
image. Image segmentation is used to locate
boundaries and objects in image [1] [3][13]. In
other words we can say that it is the process of
assigning a label to every pixel in an image such
that pixels with the same label share certain visual
characteristics. The problem of image segmentation
can be formulated as follows.
Given image I = {pi}, a complete segmentation
problem is to determine connected subset Ri = (Ri
I), such that
⋃ by image forces that pull it toward features such as
Segmentation is based on homogeneity of the
image characteristics such as intensity, colour,
texture, or the combination of all these information
[14].
There are number of different approaches in image
segmentation such as: histogram-based method,
model-based methods, edge-based technique,
region-based methods, watershed technique and
active contour methods.
II.ACTIVE CONTOUR MODEL
We have used the active contour model in oyr
paper for the segmentation process. In this
technique the user suggest an initial contour. This
framework attempts to minimize an energy
associated to the current contour as a sum of an
internal and external energy:
The external energy is supposed to be minimal
when the snake is at the object boundary position.
The internal energy is supposed to be minimal
when the snake has a shape which is supposed to be
relevant considering the shape of the sought object
[1][2].
Different ACM approaches
 Snakes
 Geometric active contours
 Geodesic active contours
A Snakes
Snakes model was introduced by Michael Kass,
Andrew Witkin and Demetri Terzopoulos in 1988
[1], for interactive interpretation of images, in
which user-imposed constraint forces guide the
snake near features of the interesting image. They
stated that a snake is an energy-minimizing spline
guided by external constraint forces and influenced
lines and edges. Snakes are greatly used in
applications like object tracking, shape recognition,
segmentation, edge detection, stereo matching. A
simple elastic snake is thus defined by
 a set of n points
 an internal elastic energy term
 an external edge based energy term
The snakes model try to segment the image based
on the following energy:
(3)
where, ∫ | | | | (4)
Snakes are autonomous and self-adapting in their
search for a minimal energy state. They can be
easily manipulated using external image forces.
They can be made sensitive to image scale by
incorporating Gaussian smoothing in the image
energy function. They can be used to track dynamic
objects in temporal as well as the spatial
dimensions. But some of the drawbacks are that
they can often get stuck in local minima states; this
may be overcome by using simulated annealing
techniques at the expense of longer computation
times. They often overlook minute features in the
process of minimizing the energy over the entire
path of their contours.
B. Geometric active contours
Geometric active contours attempt to segment an
object based on its edges, in a level-set framework
[4]. The initial contour is chosen to include the
object. The contour evolves according to
Where is a function which should
drop to zero at edges. The contour evolution tends
to smooth the contour, if no other information is
available. The contour according to this evolution
will shrink to a point. Hence, a balloon force may
be added
C. Geodesic active contours
The choice of a balloon force is arbitrary. It is not
clear if we actually minimize some functional, and
the global minimizer is not clear either [5]. The
geodesic active contour tries to remedy this by
minimizing the following weighted length
functional:
∫ be a given image. Let C(q):[0,1]
g(l) constitutes an (inverse) edge indicator. For
example,
√| |
The functional state that curves segmenting the
object should try and surround it with a minimal
weighted arc length. This can be given a physical
interpretation: We are looking for the trajectory of
a particle on a map, where the potential energy at
each point is g(I ), and we assume the particle's
trajectory should form a closed simple curve. The
potential energy of the particle is given by
u (c) = - λ g (l)2 (9)
Snakes is the classical model in active contour
method [1], [3]. It gives an efficient framework for
image segmentation, but it cannot change the
topology in the process of segmentation process.
To overcome this drawback, the level set method
was proposed in order to isolate shapes from their
background. Since then, the active contour with
level set methods has been widely applied to image
segmentation in the fields of computer vision and
image processing. GAC is an edge based active
contour model that works well when the objects
and background in segmented image are
heterogeneous, but we cannot get satisfied results
when dealing with object with discrete/ blur
boundaries or noise. Whereas Chen-Vese (C-V)
model is a region-based active contour models,
which instead of using the image gradient, use the
statistical information inside and outside the initial
curve to evolve the contour towards the boundary
of desired object. It give better performance while
compared with the edge-based model such as the
ability to work well with the object having
blur/weak boundary, and less sensitive to the initial
position of the contour. In order to get the
advantages of both GAC and C-V model, some
hybrid model are proposed. Recently, Zhang [12]
proposed the sign pressure force in an active
contour model. This model takes the advantages of
GAC and C-V models and could optionally select
local or global segmentation.
III. REFERENCE MODELS
A.GAC Model
Geodesic Active Contour model is an enhanced
version of the classical snakes in [1]. This model
depends on the image gradient to stop the evolution
of the curve. Thus it surpasses the regions
having blur or weak boundaries. Let Ω be a
bounded open subset of R2 and I: [0,a] x [0,b] R+
R2 be a
parameterized planar curve in Ω. The GAC model
is formulated by minimising the energy term given
below:
∫ (| ( )|)| | 10)
where g is the ESF. The corresponding level set
formulation for GAC is given as:
| |
where is a real constant called balloon force to
control the expanding and velocity, к is the
euclidean curvature of curve C, and g is the edge
based function.
B. C-V Model
Chan and Vese proposed an active contour without
edges that can be seen as a special case of
Munford-Shah problem. This model utilizes the
homogeneity information of the object as a term in
the energy function unlike the GAC model which
relies on the image gradient. In this method the
image is considered to be including two regions i.e
c1 and c2 inside and outside the curve C
respectively. For a given model Ω, the C-V model
is formulated by minimizing the following energy
functional:
is the dirac function. The regularised version is
selected as follows:
( ( ))
{
( )
IV. HYBRID MODEL
In our hybrid model first of all we calculate the
SPF (Signed Pressure Force) function. Its value
always lies in the range [-1,1]. It is used to control
the motion of the contour by modulating the sign of
the pressure force, so that the contour shrinks when
outside and expands when it is inside the region of
interest.

( )
∫ | | | |

∫ | | (17)

where c1 and c2 are defined in Eq.(13) and (14),

(12) respectively.

where c1 and c2 are the average intensities inside The significance of Eq. (17) can be explained as
and outside the contour, respectively. With the follows. Refer to Figure 1, we assume that the
level set method we assume intensities inside and outside the object are
homogeneous. It is intuitive that ( )
{ }
and the equal signs cannot be
obtained simultaneously wherever the contour is.
{ }
Hence, there is
{ }

By minimizing Eq.(6), we get the values for c 1 and ( ) ( )

c2 as:
(18)
∫
∫

∫ ( )
∫ ( )

The corresponding variational level set formulation

for this is:

[ Fig 1
]
Substituting the SPF function in Eq. (17) for the
where are fixed ESF in Eq. (11), the level set formulation of the
parameters, controls the smoothness of zero level proposed model is as follows:
set, increases the propagation velocity and λ1 and
λ2 controls the image data drive force inside and
( )( ( ) )| |
outside the contour respectively. is the gradient | |
operator. H( ) is the heaviside function and
 ( ) results on natural image with both the reference
model and the hybrid model.
In the traditional level set methods, the level set
function is initialized to be an SDF to its interface
in order to prevent it from being too steep or flat
near its interface, and re-initialization is required in
the evolution. Unfortunately, many existing re-
initialization methods have an undesirable side
effect of moving the zero level set away from its
interface. Furthermore, it is difficult to decide when
and how to apply the re-initialization. In addition,
re-initialization is a very expensive operation. To
solve these problems, we propose a novel level set
method, which utilizes a Gaussian filter to (a)
regularize the selective binary level set function
after each iteration. The procedure of penalizing
level set function to be binary is optional according
to the desired property of evolution. If we want
local segmentation property, the procedure is
necessary; otherwise, it is unnecessary.
Since we utilize a Gaussian filter to smooth the
level set function to keep the interface regular, the
(b) (c)
regular term | | | | is unnecessary. In
addition, the term in Eq. (19) can also be
removed, because our model utilizes the statistical
information of regions, which has a larger capture
range and capacity of anti-edge leakage. Finally,
the level set formulation of the proposed model can
be written as follows:

(d) (e)
( ) | |
Fig 2.Segmentation results on a natural image, (a)
The main procedures of the proposed algorithm are original image (b) & (c) results by the reference
summarized as follows: model after 200 and 500 iterations respectively, (d)
& (e) results from the hybrid model after 100 and
1.Initialize the level set function / as 200 iterations.

where > 0 is a constant, is a subset in the

image domain Ω and is the boundary of .
2. Compute using Eqs. (13) and
(14), respectively.
3. Evolve the level set function according to Eq.
(20).
4. Let = 1 if > 0; otherwise, = -1.
This step has the local segmentation property. If we
want to selectively segment the desired objects, this
step is necessary; otherwise, it is unnecessary.
5. Regularize the level set function with a Gaussian
filter, i.e . segmentation. By taking the advantage of the GAC
6. Check whether the evolution of the level set
function has converged. If not, return to step 2.
V. EXPERIMENTS AND RESULTS
We have done experiments with both , the natural
as well as biomedical images. Fig 1 shows the
(a) (b) (c)
Figure 3.(a) original biomedical image (b) &
(c)segmentation results by the reference model and
the hybrid model after 400 iterations.
VI. CONCLUSION
In this paper a hybrid region based model is
proposed with selective local and global image
and C-V model, this method can effectively detect
and segment the regions with weak or blur edges.
The model is able to change the topology as the
evolution flow of model is derived using level set
method. Experiments are performed on both the
natural as well as biomedical images.
 VI. REFERENCES
[1] Michael Kass, Andrew Witkin, and Demetri Terzopoulos
“Snakes: Active contour models”, International journal of
computer vision, 1988.
[2] Chenyang Xu and Jerry L. Prince “Snakes, Shapes, and
Gradient Vector Flow”,IEEE Transaction on Image
Processing,VOL. 7, NO. 3, MARCH 1998
[3] Tony F. Chan and Luminita A. Vese, “Active Contour
Without Edges”, IEEE Transaction on image processing,
VOL.10, NO.2, 2001
[4] Johan Lie, Marius Lysaker and Xue Cheng Tai “ A binary
level set model and some applications to mumford shah
image segmentation”, IEEE Transaction on image
processing, VOL. 15, NO.5, May 2004.
[5] Chunming Li, Chenyang Xu, Changfeng Gui, and Martin
D. Fox “Level Set EvolutionWithout Re-initialization: A
New Variational Formulation”, Computer vision and
pattern recognition, IEEE Computer Society Conference,
2005.
[6] Haiyun Li, Xiang Chen “A algorithm of medical image
segmentation based on active contour model”, 2007
IEEE/ICME International Conference on Complex
Medical Engineering, Capital Medical University School
of Biomedical Engineering, Beijing China.
[7] Chunming Li, Chiu Yen Kao, John C. Gore and Zhaohua
Ding “Minimization of region scalable fitting energy for
image segmentation”, IEEE Transaction on image
processing, VOL.17, NO.10, October 2008.
[8] Thi-Thao Tran, Van-Truong Pham, Yun-Jen Chiu, and
Kuo-Kai Shyu “Active Contour with Selective Local or
Global Segmentation for Intensity Inhomogeneous Image”,
Computer Science and Information Technology (ICCSIT),
2010 3rd IEEE International Conference.
[9] Kaihua Zhang, Lei Zhang, Huihui Song and Wengang
Zhou, “Active contour with selective local or global
segmentation: A new formulation and level set method.”,
Image and vision computing,2010.
[10] Muhammad Hameed Siddiqi, Sungyoung Lee, Young-Koo
Lee “Object Segmentation by Comparison of Active
Contour Snake and Level Set in Biomedical Applications”,
2011 IEEE International Conference on Bioinformatics
and Biomedicine.
[11] Arie Nakhmani and Allen Tannenbaum “Self-Crossing
Detection and Location for Parametric Active Contours”,
IEEE Transactions on Image Processing,VOL. 21, NO. 7,
July 2012.
[12] Satoshi Urata, Hiroshi Yasukawa “Improvement of
contour extraction precision of active contour model with
structuring elements”, Acoustics, Speech and Signal
Processing (ICASSP), 2012 IEEE International
Conference.
[13] F.Samopa, A. Asano.,"Hybrid Image Thresholding Method
using Edge Detection", IJCSNS International Journal of
Computer Science and Network Security, Vol.9 No.4,
PP.292-299, April 2009.
[14] Gonzalez and Woods, "Digital image processing", 2nd
Edition, prentice hall, 2002.
[15] Shuqian He, Jiangqun Ni, Lihua Wu, Hongjian Wei ,
Sixuan Zhao.," Image threshold segmentation method with
2-D histogram based on multi-resolution analysis",
Computer Science & Education, ICCSE, 25-28 July 2009,
PP.753 – 757, Nanning, China
[16] Gang Liu, Robert M. Haralick, "Assignment Problem in
Edge Detection Performance Evaluation," cvpr, vol. 1,
pp.1026, 2000 IEEE Computer Society Conference on
Computer Vision and PatternRecognition(CVPR'00) –
Volume1, 2000.