433
RESEARCH ON EDGE EXTRACTION WITH LEVEL SET METHOD
Zhang Junru a , Jiang Xuezhong a *
a State Key laboratory of Estuarine and Coastal Research, 200062, East China Normal University,ShangHai -
zjunru@gmail.com, xzjiang@sklec.ecnu.edu.cn
Commission IV, WG IV/3
KEY WORDS: Remote Sensing, Image processing, Edge Extraction, Level Set, C-V model
ABSTRACT:
Image edge detection or extraction is always a research hotspot in remote sensing image processing. Up to now, there are many
methods have been used in this area, such as threshold segmentation, edge detection operators, region growing method and so on. But
threshold segmentation cannot meet the accuracy we needed and edge detection operators are always extract all the edges in the
image when sometimes we just need part of them. In this paper, we proposed using level set method to extraction edges that we want.
At first, initial edge was given by Human-Computer Interaction way, and then level set method was used to evolve the initial edge to
make it more accurate. Experiments were carried out in Matlab, and the results show that this method is very useful and high
efficiency.
1. INTRODUCTION
Ostensibly, edge is the junction of different features in remote
sensing image. It usually is the places where image brightness
changes most in the local part of an image. Edge exists
primarily on places between one type of objective and the other,
objectives and backgrounds, region and region, and also is basis
of texture feature extraction, shape feature extraction and image
analysis. Edge extraction is the first step of image analysis and
understanding, occupies a special position in the image
processing and computer vision, and it is the one of the most
important aspects of low-level processing. In the field of remote
sensing image processing, edge detection has a wide range of
applications, many research achievements have been made in
such as road extraction, waterline extraction, cloud detection,
remote sensing image segmentation.
Because of the importance of edge extraction, lots of methods
have been proposed by researchers, they can be divided into a
few major categories: 1. Threshold segmentation(Xie fengying,
Jiang Zhiguo, 2007; Li Jiangtao, Ni Guoqiang, Huang
Guanghua, 2007; R.Medina-Camicer, F.J.Madrid-Cuevas, 2008;
Ety Navon, Ofer Miller, Amir Arerbuch, 2005); 2. Method of
differential operator based on gradient(Phillip A. Mlsna, Jeffrey
J. Rodriguez, 2005; Lijun Ding, Ardeshir Goshtasby, 2001); 3.
Region growing method(Jiangping Fan, Guihua Zeng, Mathurin
Body,Mohand-Said Hacid, 2005; Ye zhou,Jhon Starkey, Lalu,
2004);4. Method of Mathematical Morphology(T.Chen,
Q.H.Wu, R.Rahmani-torkaman, J.Hughes, 2002;); 5. Wavelet
Methods(Dusan Heric,Damjan Zazula, 2007; Ming-Yu,Shih,
Din-Chang Tseng, 2005); 6. Method of Fuzzy
Mathematics(Sakari Murtovaara, Esko Juuso, Raimo Sutinen,
1996; Florence Jacquey, Frédéric Comby, Olivier Strauss,
2008; Liming Hu, H.D. Cheng, MingZhang, 2007); and 7.
Method of neural networks(V.Srinivasan, P.Bhatia, S.h. Ong,
1994; M.Emin Yuksel, 2007). Sometimes, a few of these
methods are combined together to perform edge extraction.
But there are some problems still exist; first. It is hard to
determine thresholds, though many methods are proposed like
neural networks to help to do this, there are still many mistakes
in local area of an image due to the value always determined by
the whole image. Second, sometimes we just want to extract a
few edges like the road, waterline and so on, but many edge
detectors perform on the whole image, so a lot of tasks need to
do after using these edge detectors. Here we proposed a method
to extract edges based on level set method, on one way, we try
to provide a friendly Human-Computer Interactive environment
to perform this task automatically or semi- automatically; on
another way, we hope that after using this method, the
follow-up processing tasks would be very easy.
2. LEVEL SET METHOD
2.1 Fundamental of level set method
Level set method was proposed by Osher and Sethian in 1988
and has been widely used in image segmentation, image
smoothing, motion segmentation, moving target tracking, even
in stereovision and image reparation. As a novel approach of
handling the curve evolution, it implied in a manner to express
the closed planar curves or three-dimensional closed surface, so
as to avoid the evolution process of tracing the curve. Thus
problem of curve evolution is converted to a pure searching for
the numerical answers of partial differential equation.
The main idea of level set method is to embed the propagating
interface as the zero level set of a higher dimensional
hypersurface. By controlling the evolution of the hypersurface,
we can control the evolution of the curve.
Suppose the level set function is 0(x y t , C(p, t) is the zero
level set of C'Xyl , then, we produce an Eulerian
formulation for the motion of the hypersurface propagating
along its normal direction with speed V, where V can be a
function of various arguments, including the curvature, normal
direction, etc. Let C(p, t=0) is the initial curve, ©fxt = 0) is
usually defined by SDF (Signed distance function):
Cxi = C = Hbd(x)