A Modified Image Segmentation Method Using Active Contour Model
Shiping Zhu
1, a
, Ruidong Gao
1, b
1
Department of Measurement Control and Information Technology, School of Instrumentation
Science and Opto-electronics Engineering, Beihang University, Beijing 100191, China
a
spzhu@163.com,
b
rxdj1990@163.com
Keywords: Active contours; Gradient vector flow; Laplace operator; Border leakage; External force
field.
Abstract. Active contours, or snakes, have extensive applications in image segmentation.
Conventional snakes have several drawbacks, such as the initialization contour sensitivity and border
leakage phenomenon. Many new methods have been proposed to address these problems. In this
paper, we present an improved image segmentation method based on snakes. Firstly, we adopt the
multi-step direction method to enlarge the scope of initial contour and obtain more precise edge map.
Then, we decompose the Laplace operator to tangential direction and normal direction, weakening
the border smoothing effect. Finally, two correlational self-adaptive weight functions are added to the
two directions. Thus, the snakes can adaptively adjust the weights of smoothing item and diffusion
item through the local image characteristics. Based on the subjective and objective evaluations, the
proposed method outperforms the state-of-the-art methods and improves the segmentation accuracy.
Introduction
Active contour model, or snake model, was proposed by Kass et al [1], in 1987. Snake model has
been widely applied in the fields of computer vision and image processing [2-6], such as image
segmentation, target tracking, and edge detection. Although the conventional active contour model
has been widely used, it still has its shortcomings [7]. First, the initial contour must be very close to
the interesting image features. Second, border leakage phenomenon is occurred, losing a lot of
important image information. Third, the snake curve is difficult to reach indentation boundaries.
In order to solve these problems, Xu proposed gradient vector flow (GVF) snake model [7], which
is able to expand the capture scope of initial contour and has a certain indentation convergence
capability. The generalized gradient vector flow (GGVF) model proposed by Xu and Prince [8] adds
two weight coefficients changing in the image field based on the original GVF external force field.
Thus, the curve converges rapidly in the flat field and has certain boundary protection effect.
However, the indentation convergence capability has not been greatly improved. NGVF proposed by
Ning [9] decomposes the Laplace operator in the GVF external force field, and only retains normal
component. Thereby NGVF further improves the curve’s indentation capability. Although the
tangential component after decomposition has been added to GVF external force field by later
NBGVF [10], the convergence capability of long and thin indentation has not been significantly
improved. Based on the studies about GGVF and GVF, Qin [11] discovered that no matter GGVF or
GVF can only converge to the indentation with odd pixel width and is of no convergence capability to
the indentation with even pixel width. Therefore, CN-GGVF algorithm was proposed by Qin, with
applying component normalization method in GGVF. CN-GGVF solved the problem that
deformation curve cannot converge to even pixel width, and retained the fast convergence character
in GGVF. However, CN-GGVF has no ideal effect on the protection of weak boundary features of
interesting field and has boundary leakage phenomenon.
2nd International Conference on Electrical, Computer Engineering and Electronics (ICECEE 2015)
© 2015. The authors - Published by Atlantis Press