Local and Global Intensity Information Integrated
Geodesic Model for Image Segmentation
Ping Wang, Kaiqiong Sun and Zhen Chen
Computer Vision Laboratory, Nanchang Hangkong University, Nanchang 330063, China
wangping841@126.com,{kqsun,chenzhen}@nchu.edu.cn
Abstract—In this paper, we propose a novel active contour
model for image segmentation integrating the local and global
intensity information into the geodesic model. A signed pressure
forces function based on the statistical information of gray inten-
sity is used to replace the edge stopping function in the traditional
geodesic model. The minimization of energy of the contour is
implemented by level set method with a binary level set function
avoiding the problem of reinitializing the traditional level set
function to a signed distance function. Experimental results
show the proposed model can obtain improved segmentation
result compared to related methods. It can effectively prevent
the boundary leakage while maintaining robustness against the
initialization position.
Index Terms—image segmentation; geodesic active contour;
level set method
I. INTRODUCTION
Image segmentation is an essential part of the image pro-
cessing and computer vision. In various segmentation methods,
active contour model [1] not only utilizes image information,
but allows for the prior knowledge in the segmentation pro-
cess. Besides, active contour model has feasible algorithm to
support its implementation. The fundamental idea of active
contour model is to construct some constraints to the object
and realize the segmentation process by curve evolution.
According to the property of constraints, the existing active
contour model can be categorized into two types: edge-based
models and region-based models.
One of the most popular edge-based models is the geodesics
active contour (GAC) model [2], which utilizes image gradient
to construct an edge stopping function to stop the contour
evolution on the object boundaries. For the weak boundary
or discrete shape boundary target, contour is difficult to stop
at the border. On the other hand, when the initial contour is
far away from the target boundary, it is difficult for the GAC
model to find the target.
As one of the most typical region-based active contour
models, the active contour model without edge, proposed by
Chan and Vese (CV) [3], utilizes global gray information
within the internal and external contours. Therefore, better
segmentation result can also be obtained for the noise image
and the weak boundary image. However, because of the
gray uniform assumption in CV model, it will produce error
separation results when applied to image with inhomogeneous
gray intensity.
To overcome the gray intensity inhomogeneity of image, Li
[4] introduced a kernel function to define a local binary fitting
(LBF) energy in the region-based active contour model. This
model only takes into account the local gray information. Thus
it is easy to achieve a local minimum and sensitive to the initial
location of active contour.
The three models [2] [3] [4] mentioned above are typical
edge-based model, region-based model and local model, re-
spectively. They make use of different image information to
drive the evolution of curve. On the other hand, some methods
combined the above methods and integrated different image
information to drive the evolution. Zhang [5] proposed an
improved method of GAC model, which utilizes the global
intensity information to construct a signed pressure force
(SPF) function to drive the contour evolution. This method
uses simple iterative speed, but it is also hard to deal with the
image having inhomogeneous gray intensity or fuzzy boundary
images. Along this direction, Zheng [6] proposed an improved
method of GAC model; it utilizes the local gray information
to construct a SPF function. It can overcome inhomogeneous
gray intensity and get faster segmentation speed than LBF.
For only the local gray information is considered, the model
is also sensitive to the initial location of active contour. To
overcome the local minimum problem of the LBF model,
Wang [7] proposed an active contour driven by local and global
intensity fitting energy with application to brain MR image
segmentation.
This paper presents a novel active contour model, which
uses the local and global gray information to construct the SPF
function acting as the stopping function to control the direction
of evolution contour. It can not only overcome inhomogeneous
gray intensity, but also deal with the images having complex
background and weak boundary. Moreover, it is robust to the
initial position.
II.
THE RELATED MODEL
A. The GAC model
Let C(q):[0, 1]→R
2
be a parameterized planar curve in
Ω, the image domain. The GAC model is formulated by
minimizing the following energy functional:
E
GAC
=
1
0
g(|∇I(C(q))|)|C
(q)|dq (1)
where g is a positive, decreasing and regular edge stopping
function.
2012 International Conference on Computer Science and Electronics Engineering
978-0-7695-4647-6/12 $26.00 © 2012 IEEE
DOI 10.1109/ICCSEE.2012.259
129